[GCC][PATCH] Add __artificial__ attribute to Aarch64 NEON intrinsics

Committed on behalf of Tamar Christina <tamar.christina@arm.com>.

gcc/

	* config/aarch64/arm_neon.h: Add gnu_inline and artificial
	attributes to all inlined functions and make them extern.

From-SVN: r240256

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 5b35f9224fcf0b63b4b5c94f57910caa4657aba9..cef1504d9620cadddc7b042fa0c29bd28869e96f 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,8 @@
+2016-09-20  Tamar Christina  <tamar.christina@arm.com>
+
+       * config/aarch64/arm_neon.h: Add gnu_inline and artificial
+       attributes to all inlined functions and make them extern.
+
 2016-09-20  Richard Biener  <rguenther@suse.de>
 
        * debug.h (gcc_debug_hooks): Add filename parameter to early_finish

diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index d0e1fe7e1b113c2d9f2b7966c9b3bd6956e4152f..b4db87b0bc60d22306da810381b62920646aac9b 100644 (file)
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -609,535 +609,619 @@ typedef struct poly16x8x4_t
   })
 
 /* vadd  */
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_f64 (float64x1_t __a, float64x1_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_s64 (int64x1_t __a, int64x1_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_saddlv8qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_saddlv4hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int64x2_t) __builtin_aarch64_saddlv2si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_uaddlv8qi ((int8x8_t) __a,
                                                   (int8x8_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_uaddlv4hi ((int16x4_t) __a,
                                                   (int16x4_t) __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (uint64x2_t) __builtin_aarch64_uaddlv2si ((int32x2_t) __a,
                                                   (int32x2_t) __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_high_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (int16x8_t) __builtin_aarch64_saddl2v16qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_high_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int32x4_t) __builtin_aarch64_saddl2v8hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_high_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int64x2_t) __builtin_aarch64_saddl2v4si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_high_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_uaddl2v16qi ((int8x16_t) __a,
                                                     (int8x16_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_high_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_uaddl2v8hi ((int16x8_t) __a,
                                                    (int16x8_t) __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddl_high_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint64x2_t) __builtin_aarch64_uaddl2v4si ((int32x4_t) __a,
                                                    (int32x4_t) __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_s8 (int16x8_t __a, int8x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_saddwv8qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_s16 (int32x4_t __a, int16x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_saddwv4hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_s32 (int64x2_t __a, int32x2_t __b)
 {
   return (int64x2_t) __builtin_aarch64_saddwv2si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_u8 (uint16x8_t __a, uint8x8_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_uaddwv8qi ((int16x8_t) __a,
                                                   (int8x8_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_u16 (uint32x4_t __a, uint16x4_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_uaddwv4hi ((int32x4_t) __a,
                                                   (int16x4_t) __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_u32 (uint64x2_t __a, uint32x2_t __b)
 {
   return (uint64x2_t) __builtin_aarch64_uaddwv2si ((int64x2_t) __a,
                                                   (int32x2_t) __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_high_s8 (int16x8_t __a, int8x16_t __b)
 {
   return (int16x8_t) __builtin_aarch64_saddw2v16qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_high_s16 (int32x4_t __a, int16x8_t __b)
 {
   return (int32x4_t) __builtin_aarch64_saddw2v8hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_high_s32 (int64x2_t __a, int32x4_t __b)
 {
   return (int64x2_t) __builtin_aarch64_saddw2v4si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_high_u8 (uint16x8_t __a, uint8x16_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_uaddw2v16qi ((int16x8_t) __a,
                                                     (int8x16_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_high_u16 (uint32x4_t __a, uint16x8_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_uaddw2v8hi ((int32x4_t) __a,
                                                    (int16x8_t) __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddw_high_u32 (uint64x2_t __a, uint32x4_t __b)
 {
   return (uint64x2_t) __builtin_aarch64_uaddw2v4si ((int64x2_t) __a,
                                                    (int32x4_t) __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhadd_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (int8x8_t) __builtin_aarch64_shaddv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhadd_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_shaddv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhadd_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_shaddv2si (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhadd_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_uhaddv8qi ((int8x8_t) __a,
                                                  (int8x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhadd_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_uhaddv4hi ((int16x4_t) __a,
                                                   (int16x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhadd_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_uhaddv2si ((int32x2_t) __a,
                                                   (int32x2_t) __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhaddq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (int8x16_t) __builtin_aarch64_shaddv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhaddq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_shaddv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhaddq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_shaddv4si (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (uint8x16_t) __builtin_aarch64_uhaddv16qi ((int8x16_t) __a,
                                                    (int8x16_t) __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_uhaddv8hi ((int16x8_t) __a,
                                                   (int16x8_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_uhaddv4si ((int32x4_t) __a,
                                                   (int32x4_t) __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhadd_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (int8x8_t) __builtin_aarch64_srhaddv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhadd_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_srhaddv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhadd_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_srhaddv2si (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhadd_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_urhaddv8qi ((int8x8_t) __a,
                                                   (int8x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhadd_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_urhaddv4hi ((int16x4_t) __a,
                                                    (int16x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhadd_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_urhaddv2si ((int32x2_t) __a,
                                                    (int32x2_t) __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhaddq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (int8x16_t) __builtin_aarch64_srhaddv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhaddq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_srhaddv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhaddq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_srhaddv4si (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhaddq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (uint8x16_t) __builtin_aarch64_urhaddv16qi ((int8x16_t) __a,
                                                     (int8x16_t) __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhaddq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_urhaddv8hi ((int16x8_t) __a,
                                                    (int16x8_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrhaddq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_urhaddv4si ((int32x4_t) __a,
                                                    (int32x4_t) __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int8x8_t) __builtin_aarch64_addhnv8hi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_addhnv4si (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_addhnv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_addhnv8hi ((int16x8_t) __a,
                                                  (int16x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_addhnv4si ((int32x4_t) __a,
                                                   (int32x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_addhnv2di ((int64x2_t) __a,
                                                   (int64x2_t) __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int8x8_t) __builtin_aarch64_raddhnv8hi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_raddhnv4si (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_raddhnv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_raddhnv8hi ((int16x8_t) __a,
                                                   (int16x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_raddhnv4si ((int32x4_t) __a,
                                                    (int32x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_raddhnv2di ((int64x2_t) __a,
                                                    (int64x2_t) __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_high_s16 (int8x8_t __a, int16x8_t __b, int16x8_t __c)
 {
   return (int8x16_t) __builtin_aarch64_addhn2v8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_high_s32 (int16x4_t __a, int32x4_t __b, int32x4_t __c)
 {
   return (int16x8_t) __builtin_aarch64_addhn2v4si (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_high_s64 (int32x2_t __a, int64x2_t __b, int64x2_t __c)
 {
   return (int32x4_t) __builtin_aarch64_addhn2v2di (__a, __b, __c);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
 {
   return (uint8x16_t) __builtin_aarch64_addhn2v8hi ((int8x8_t) __a,
@@ -1145,7 +1229,8 @@ vaddhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
                                                    (int16x8_t) __c);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
 {
   return (uint16x8_t) __builtin_aarch64_addhn2v4si ((int16x4_t) __a,
@@ -1153,7 +1238,8 @@ vaddhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
                                                    (int32x4_t) __c);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
 {
   return (uint32x4_t) __builtin_aarch64_addhn2v2di ((int32x2_t) __a,
@@ -1161,25 +1247,29 @@ vaddhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
                                                    (int64x2_t) __c);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_high_s16 (int8x8_t __a, int16x8_t __b, int16x8_t __c)
 {
   return (int8x16_t) __builtin_aarch64_raddhn2v8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_high_s32 (int16x4_t __a, int32x4_t __b, int32x4_t __c)
 {
   return (int16x8_t) __builtin_aarch64_raddhn2v4si (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_high_s64 (int32x2_t __a, int64x2_t __b, int64x2_t __c)
 {
   return (int32x4_t) __builtin_aarch64_raddhn2v2di (__a, __b, __c);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
 {
   return (uint8x16_t) __builtin_aarch64_raddhn2v8hi ((int8x8_t) __a,
@@ -1187,7 +1277,8 @@ vraddhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
                                                     (int16x8_t) __c);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
 {
   return (uint16x8_t) __builtin_aarch64_raddhn2v4si ((int16x4_t) __a,
@@ -1195,7 +1286,8 @@ vraddhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
                                                     (int32x4_t) __c);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vraddhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
 {
   return (uint32x4_t) __builtin_aarch64_raddhn2v2di ((int32x2_t) __a,
@@ -1203,1101 +1295,1280 @@ vraddhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
                                                     (int64x2_t) __c);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdiv_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __a / __b;
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdiv_f64 (float64x1_t __a, float64x1_t __b)
 {
   return __a / __b;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdivq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __a / __b;
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdivq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __a / __b;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_f64 (float64x1_t __a, float64x1_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_p8 (poly8x8_t __a, poly8x8_t __b)
 {
   return (poly8x8_t) __builtin_aarch64_pmulv8qi ((int8x8_t) __a,
                                                 (int8x8_t) __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_p8 (poly8x16_t __a, poly8x16_t __b)
 {
   return (poly8x16_t) __builtin_aarch64_pmulv16qi ((int8x16_t) __a,
                                                   (int8x16_t) __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vand_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vand_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vand_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vand_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vand_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vand_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vand_s64 (int64x1_t __a, int64x1_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vand_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vandq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vandq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vandq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vandq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vandq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vandq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vandq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vandq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return __a & __b;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorr_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorr_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorr_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorr_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorr_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorr_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorr_s64 (int64x1_t __a, int64x1_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorr_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorrq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorrq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorrq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorrq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorrq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorrq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorrq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorrq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return __a | __b;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veor_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veor_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veor_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veor_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veor_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veor_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veor_s64 (int64x1_t __a, int64x1_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veor_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veorq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veorq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veorq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veorq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veorq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veorq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veorq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 veorq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return __a ^ __b;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbic_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbic_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbic_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbic_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbic_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbic_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbic_s64 (int64x1_t __a, int64x1_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbic_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbicq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbicq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbicq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbicq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbicq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbicq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbicq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbicq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return __a & ~__b;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorn_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorn_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorn_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorn_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorn_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorn_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorn_s64 (int64x1_t __a, int64x1_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vorn_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vornq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vornq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vornq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vornq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vornq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vornq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vornq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vornq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return __a | ~__b;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_f64 (float64x1_t __a, float64x1_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_s64 (int64x1_t __a, int64x1_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_ssublv8qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_ssublv4hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int64x2_t) __builtin_aarch64_ssublv2si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_usublv8qi ((int8x8_t) __a,
                                                   (int8x8_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_usublv4hi ((int16x4_t) __a,
                                                   (int16x4_t) __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (uint64x2_t) __builtin_aarch64_usublv2si ((int32x2_t) __a,
                                                   (int32x2_t) __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_high_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (int16x8_t) __builtin_aarch64_ssubl2v16qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_high_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int32x4_t) __builtin_aarch64_ssubl2v8hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_high_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int64x2_t) __builtin_aarch64_ssubl2v4si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_high_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_usubl2v16qi ((int8x16_t) __a,
                                                     (int8x16_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_high_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_usubl2v8hi ((int16x8_t) __a,
                                                    (int16x8_t) __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubl_high_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint64x2_t) __builtin_aarch64_usubl2v4si ((int32x4_t) __a,
                                                    (int32x4_t) __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_s8 (int16x8_t __a, int8x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_ssubwv8qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_s16 (int32x4_t __a, int16x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_ssubwv4hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_s32 (int64x2_t __a, int32x2_t __b)
 {
   return (int64x2_t) __builtin_aarch64_ssubwv2si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_u8 (uint16x8_t __a, uint8x8_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_usubwv8qi ((int16x8_t) __a,
                                                   (int8x8_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_u16 (uint32x4_t __a, uint16x4_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_usubwv4hi ((int32x4_t) __a,
                                                   (int16x4_t) __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_u32 (uint64x2_t __a, uint32x2_t __b)
 {
   return (uint64x2_t) __builtin_aarch64_usubwv2si ((int64x2_t) __a,
                                                   (int32x2_t) __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_high_s8 (int16x8_t __a, int8x16_t __b)
 {
   return (int16x8_t) __builtin_aarch64_ssubw2v16qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_high_s16 (int32x4_t __a, int16x8_t __b)
 {
   return (int32x4_t) __builtin_aarch64_ssubw2v8hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_high_s32 (int64x2_t __a, int32x4_t __b)
 {
   return (int64x2_t) __builtin_aarch64_ssubw2v4si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_high_u8 (uint16x8_t __a, uint8x16_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_usubw2v16qi ((int16x8_t) __a,
                                                     (int8x16_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_high_u16 (uint32x4_t __a, uint16x8_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_usubw2v8hi ((int32x4_t) __a,
                                                    (int16x8_t) __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubw_high_u32 (uint64x2_t __a, uint32x4_t __b)
 {
   return (uint64x2_t) __builtin_aarch64_usubw2v4si ((int64x2_t) __a,
                                                    (int32x4_t) __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadd_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (int8x8_t) __builtin_aarch64_sqaddv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadd_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_sqaddv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadd_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_sqaddv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadd_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (int64x1_t) {__builtin_aarch64_sqadddi (__a[0], __b[0])};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadd_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __builtin_aarch64_uqaddv8qi_uuu (__a, __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsub_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (int8x8_t)__builtin_aarch64_shsubv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsub_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_shsubv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsub_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_shsubv2si (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsub_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_uhsubv8qi ((int8x8_t) __a,
                                                  (int8x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsub_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_uhsubv4hi ((int16x4_t) __a,
                                                   (int16x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsub_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_uhsubv2si ((int32x2_t) __a,
                                                   (int32x2_t) __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsubq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (int8x16_t) __builtin_aarch64_shsubv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsubq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_shsubv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsubq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_shsubv4si (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsubq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (uint8x16_t) __builtin_aarch64_uhsubv16qi ((int8x16_t) __a,
                                                    (int8x16_t) __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsubq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_uhsubv8hi ((int16x8_t) __a,
                                                   (int16x8_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vhsubq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_uhsubv4si ((int32x4_t) __a,
                                                   (int32x4_t) __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int8x8_t) __builtin_aarch64_subhnv8hi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_subhnv4si (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_subhnv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_subhnv8hi ((int16x8_t) __a,
                                                  (int16x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_subhnv4si ((int32x4_t) __a,
                                                   (int32x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_subhnv2di ((int64x2_t) __a,
                                                   (int64x2_t) __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int8x8_t) __builtin_aarch64_rsubhnv8hi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_rsubhnv4si (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_rsubhnv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_rsubhnv8hi ((int16x8_t) __a,
                                                   (int16x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_rsubhnv4si ((int32x4_t) __a,
                                                    (int32x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_rsubhnv2di ((int64x2_t) __a,
                                                    (int64x2_t) __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_high_s16 (int8x8_t __a, int16x8_t __b, int16x8_t __c)
 {
   return (int8x16_t) __builtin_aarch64_rsubhn2v8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_high_s32 (int16x4_t __a, int32x4_t __b, int32x4_t __c)
 {
   return (int16x8_t) __builtin_aarch64_rsubhn2v4si (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_high_s64 (int32x2_t __a, int64x2_t __b, int64x2_t __c)
 {
   return (int32x4_t) __builtin_aarch64_rsubhn2v2di (__a, __b, __c);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
 {
   return (uint8x16_t) __builtin_aarch64_rsubhn2v8hi ((int8x8_t) __a,
@@ -2305,7 +2576,8 @@ vrsubhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
                                                     (int16x8_t) __c);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
 {
   return (uint16x8_t) __builtin_aarch64_rsubhn2v4si ((int16x4_t) __a,
@@ -2313,7 +2585,8 @@ vrsubhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
                                                     (int32x4_t) __c);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsubhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
 {
   return (uint32x4_t) __builtin_aarch64_rsubhn2v2di ((int32x2_t) __a,
@@ -2321,25 +2594,29 @@ vrsubhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
                                                     (int64x2_t) __c);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_high_s16 (int8x8_t __a, int16x8_t __b, int16x8_t __c)
 {
   return (int8x16_t) __builtin_aarch64_subhn2v8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_high_s32 (int16x4_t __a, int32x4_t __b, int32x4_t __c)
 {
   return (int16x8_t) __builtin_aarch64_subhn2v4si (__a, __b, __c);;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_high_s64 (int32x2_t __a, int64x2_t __b, int64x2_t __c)
 {
   return (int32x4_t) __builtin_aarch64_subhn2v2di (__a, __b, __c);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
 {
   return (uint8x16_t) __builtin_aarch64_subhn2v8hi ((int8x8_t) __a,
@@ -2347,7 +2624,8 @@ vsubhn_high_u16 (uint8x8_t __a, uint16x8_t __b, uint16x8_t __c)
                                                    (int16x8_t) __c);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
 {
   return (uint16x8_t) __builtin_aarch64_subhn2v4si ((int16x4_t) __a,
@@ -2355,7 +2633,8 @@ vsubhn_high_u32 (uint16x4_t __a, uint32x4_t __b, uint32x4_t __c)
                                                    (int32x4_t) __c);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
 {
   return (uint32x4_t) __builtin_aarch64_subhn2v2di ((int32x2_t) __a,
@@ -2363,373 +2642,435 @@ vsubhn_high_u64 (uint32x2_t __a, uint64x2_t __b, uint64x2_t __c)
                                                    (int64x2_t) __c);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadd_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __builtin_aarch64_uqaddv4hi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadd_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __builtin_aarch64_uqaddv2si_uuu (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadd_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return (uint64x1_t) {__builtin_aarch64_uqadddi_uuu (__a[0], __b[0])};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (int8x16_t) __builtin_aarch64_sqaddv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_sqaddv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_sqaddv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (int64x2_t) __builtin_aarch64_sqaddv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __builtin_aarch64_uqaddv16qi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __builtin_aarch64_uqaddv8hi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __builtin_aarch64_uqaddv4si_uuu (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return __builtin_aarch64_uqaddv2di_uuu (__a, __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsub_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (int8x8_t) __builtin_aarch64_sqsubv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsub_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_sqsubv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsub_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_sqsubv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsub_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (int64x1_t) {__builtin_aarch64_sqsubdi (__a[0], __b[0])};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsub_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return __builtin_aarch64_uqsubv8qi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsub_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return __builtin_aarch64_uqsubv4hi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsub_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return __builtin_aarch64_uqsubv2si_uuu (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsub_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return (uint64x1_t) {__builtin_aarch64_uqsubdi_uuu (__a[0], __b[0])};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (int8x16_t) __builtin_aarch64_sqsubv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_sqsubv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_sqsubv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (int64x2_t) __builtin_aarch64_sqsubv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return __builtin_aarch64_uqsubv16qi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return __builtin_aarch64_uqsubv8hi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return __builtin_aarch64_uqsubv4si_uuu (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return __builtin_aarch64_uqsubv2di_uuu (__a, __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqneg_s8 (int8x8_t __a)
 {
   return (int8x8_t) __builtin_aarch64_sqnegv8qi (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqneg_s16 (int16x4_t __a)
 {
   return (int16x4_t) __builtin_aarch64_sqnegv4hi (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqneg_s32 (int32x2_t __a)
 {
   return (int32x2_t) __builtin_aarch64_sqnegv2si (__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqneg_s64 (int64x1_t __a)
 {
   return (int64x1_t) {__builtin_aarch64_sqnegdi (__a[0])};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqnegq_s8 (int8x16_t __a)
 {
   return (int8x16_t) __builtin_aarch64_sqnegv16qi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqnegq_s16 (int16x8_t __a)
 {
   return (int16x8_t) __builtin_aarch64_sqnegv8hi (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqnegq_s32 (int32x4_t __a)
 {
   return (int32x4_t) __builtin_aarch64_sqnegv4si (__a);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabs_s8 (int8x8_t __a)
 {
   return (int8x8_t) __builtin_aarch64_sqabsv8qi (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabs_s16 (int16x4_t __a)
 {
   return (int16x4_t) __builtin_aarch64_sqabsv4hi (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabs_s32 (int32x2_t __a)
 {
   return (int32x2_t) __builtin_aarch64_sqabsv2si (__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabs_s64 (int64x1_t __a)
 {
   return (int64x1_t) {__builtin_aarch64_sqabsdi (__a[0])};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabsq_s8 (int8x16_t __a)
 {
   return (int8x16_t) __builtin_aarch64_sqabsv16qi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabsq_s16 (int16x8_t __a)
 {
   return (int16x8_t) __builtin_aarch64_sqabsv8hi (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabsq_s32 (int32x4_t __a)
 {
   return (int32x4_t) __builtin_aarch64_sqabsv4si (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulh_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_sqdmulhv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulh_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_sqdmulhv2si (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_sqdmulhv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_sqdmulhv4si (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulh_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_sqrdmulhv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulh_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_sqrdmulhv2si (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_sqrdmulhv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_sqrdmulhv4si (__a, __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_s8 (uint64_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_s16 (uint64_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_s32 (uint64_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_s64 (uint64_t __a)
 {
   return (int64x1_t) {__a};
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_f16 (uint64_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_f32 (uint64_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_u8 (uint64_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_u16 (uint64_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_u32 (uint64_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_u64 (uint64_t __a)
 {
   return (uint64x1_t) {__a};
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_f64 (uint64_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_p8 (uint64_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_p16 (uint64_t __a)
 {
   return (poly16x4_t) __a;
@@ -2737,79 +3078,92 @@ vcreate_p16 (uint64_t __a)
 
 /* vget_lane  */
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_f16 (float16x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_f32 (float32x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_f64 (float64x1_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline poly8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_p8 (poly8x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline poly16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_p16 (poly16x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_s8 (int8x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_s16 (int16x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_s32 (int32x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_s64 (int64x1_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_u8 (uint8x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_u16 (uint16x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_u32 (uint32x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_lane_u64 (uint64x1_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
@@ -2817,79 +3171,92 @@ vget_lane_u64 (uint64x1_t __a, const int __b)
 
 /* vgetq_lane  */
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_f16 (float16x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_f32 (float32x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_f64 (float64x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline poly8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_p8 (poly8x16_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline poly16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_p16 (poly16x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_s8 (int8x16_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_s16 (int16x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_s32 (int32x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_s64 (int64x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_u8 (uint8x16_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_u16 (uint16x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_u32 (uint32x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_u64 (uint64x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
@@ -2897,1873 +3264,2185 @@ vgetq_lane_u64 (uint64x2_t __a, const int __b)
 
 /* vreinterpret  */
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_f16 (float16x4_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_f64 (float64x1_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_s8 (int8x8_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_s16 (int16x4_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_s32 (int32x2_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_s64 (int64x1_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_f32 (float32x2_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_u8 (uint8x8_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_u16 (uint16x4_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_u32 (uint32x2_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_u64 (uint64x1_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p8_p16 (poly16x4_t __a)
 {
   return (poly8x8_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_f64 (float64x2_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_s8 (int8x16_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_s16 (int16x8_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_s32 (int32x4_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_s64 (int64x2_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_f16 (float16x8_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_f32 (float32x4_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_u8 (uint8x16_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_u16 (uint16x8_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_u32 (uint32x4_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_u64 (uint64x2_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p8_p16 (poly16x8_t __a)
 {
   return (poly8x16_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_f16 (float16x4_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_f64 (float64x1_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_s8 (int8x8_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_s16 (int16x4_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_s32 (int32x2_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_s64 (int64x1_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_f32 (float32x2_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_u8 (uint8x8_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_u16 (uint16x4_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_u32 (uint32x2_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_u64 (uint64x1_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_p16_p8 (poly8x8_t __a)
 {
   return (poly16x4_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_f64 (float64x2_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_s8 (int8x16_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_s16 (int16x8_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_s32 (int32x4_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_s64 (int64x2_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_f16 (float16x8_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_f32 (float32x4_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_u8 (uint8x16_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_u16 (uint16x8_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_u32 (uint32x4_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_u64 (uint64x2_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_p16_p8 (poly8x16_t __a)
 {
   return (poly16x8_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_f64 (float64x1_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_s8 (int8x8_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_s16 (int16x4_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_s32 (int32x2_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_s64 (int64x1_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_f32 (float32x2_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_u8 (uint8x8_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_u16 (uint16x4_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_u32 (uint32x2_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_u64 (uint64x1_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_p8 (poly8x8_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f16_p16 (poly16x4_t __a)
 {
   return (float16x4_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_f64 (float64x2_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_s8 (int8x16_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_s16 (int16x8_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_s32 (int32x4_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_s64 (int64x2_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_f32 (float32x4_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_u8 (uint8x16_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_u16 (uint16x8_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_u32 (uint32x4_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_u64 (uint64x2_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_p8 (poly8x16_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f16_p16 (poly16x8_t __a)
 {
   return (float16x8_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_f16 (float16x4_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_f64 (float64x1_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_s8 (int8x8_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_s16 (int16x4_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_s32 (int32x2_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_s64 (int64x1_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_u8 (uint8x8_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_u16 (uint16x4_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_u32 (uint32x2_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_u64 (uint64x1_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_p8 (poly8x8_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f32_p16 (poly16x4_t __a)
 {
   return (float32x2_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_f16 (float16x8_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_f64 (float64x2_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_s8 (int8x16_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_s16 (int16x8_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_s32 (int32x4_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_s64 (int64x2_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_u8 (uint8x16_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_u16 (uint16x8_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_u32 (uint32x4_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_u64 (uint64x2_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_p8 (poly8x16_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f32_p16 (poly16x8_t __a)
 {
   return (float32x4_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_f16 (float16x4_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_f32 (float32x2_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_p8 (poly8x8_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_p16 (poly16x4_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_s8 (int8x8_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_s16 (int16x4_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_s32 (int32x2_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_s64 (int64x1_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_u8 (uint8x8_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_u16 (uint16x4_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_u32 (uint32x2_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x1_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_f64_u64 (uint64x1_t __a)
 {
   return (float64x1_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_f16 (float16x8_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_f32 (float32x4_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_p8 (poly8x16_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_p16 (poly16x8_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_s8 (int8x16_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_s16 (int16x8_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_s32 (int32x4_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_s64 (int64x2_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_u8 (uint8x16_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_u16 (uint16x8_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_u32 (uint32x4_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline float64x2_t __attribute__((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_f64_u64 (uint64x2_t __a)
 {
   return (float64x2_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_f16 (float16x4_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_f64 (float64x1_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_s8 (int8x8_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_s16 (int16x4_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_s32 (int32x2_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_f32 (float32x2_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_u8 (uint8x8_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_u16 (uint16x4_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_u32 (uint32x2_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_u64 (uint64x1_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_p8 (poly8x8_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s64_p16 (poly16x4_t __a)
 {
   return (int64x1_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_f64 (float64x2_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_s8 (int8x16_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_s16 (int16x8_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_s32 (int32x4_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_f16 (float16x8_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_f32 (float32x4_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_u8 (uint8x16_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_u16 (uint16x8_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_u32 (uint32x4_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_u64 (uint64x2_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_p8 (poly8x16_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s64_p16 (poly16x8_t __a)
 {
   return (int64x2_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_f16 (float16x4_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_f64 (float64x1_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_s8 (int8x8_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_s16 (int16x4_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_s32 (int32x2_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_s64 (int64x1_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_f32 (float32x2_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_u8 (uint8x8_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_u16 (uint16x4_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_u32 (uint32x2_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_p8 (poly8x8_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u64_p16 (poly16x4_t __a)
 {
   return (uint64x1_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_f64 (float64x2_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_s8 (int8x16_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_s16 (int16x8_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_s32 (int32x4_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_s64 (int64x2_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_f16 (float16x8_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_f32 (float32x4_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_u8 (uint8x16_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_u16 (uint16x8_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_u32 (uint32x4_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_p8 (poly8x16_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u64_p16 (poly16x8_t __a)
 {
   return (uint64x2_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_f16 (float16x4_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_f64 (float64x1_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_s16 (int16x4_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_s32 (int32x2_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_s64 (int64x1_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_f32 (float32x2_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_u8 (uint8x8_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_u16 (uint16x4_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_u32 (uint32x2_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_u64 (uint64x1_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_p8 (poly8x8_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s8_p16 (poly16x4_t __a)
 {
   return (int8x8_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_f64 (float64x2_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_s16 (int16x8_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_s32 (int32x4_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_s64 (int64x2_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_f16 (float16x8_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_f32 (float32x4_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_u8 (uint8x16_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_u16 (uint16x8_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_u32 (uint32x4_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_u64 (uint64x2_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_p8 (poly8x16_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s8_p16 (poly16x8_t __a)
 {
   return (int8x16_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_f16 (float16x4_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_f64 (float64x1_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_s8 (int8x8_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_s32 (int32x2_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_s64 (int64x1_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_f32 (float32x2_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_u8 (uint8x8_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_u16 (uint16x4_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_u32 (uint32x2_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_u64 (uint64x1_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_p8 (poly8x8_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s16_p16 (poly16x4_t __a)
 {
   return (int16x4_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_f64 (float64x2_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_s8 (int8x16_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_s32 (int32x4_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_s64 (int64x2_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_f16 (float16x8_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_f32 (float32x4_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_u8 (uint8x16_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_u16 (uint16x8_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_u32 (uint32x4_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_u64 (uint64x2_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_p8 (poly8x16_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s16_p16 (poly16x8_t __a)
 {
   return (int16x8_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_f16 (float16x4_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_f64 (float64x1_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_s8 (int8x8_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_s16 (int16x4_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_s64 (int64x1_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_f32 (float32x2_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_u8 (uint8x8_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_u16 (uint16x4_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_u32 (uint32x2_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_u64 (uint64x1_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_p8 (poly8x8_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_s32_p16 (poly16x4_t __a)
 {
   return (int32x2_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_f64 (float64x2_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_s8 (int8x16_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_s16 (int16x8_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_s64 (int64x2_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_f16 (float16x8_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_f32 (float32x4_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_u8 (uint8x16_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_u16 (uint16x8_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_u32 (uint32x4_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_u64 (uint64x2_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_p8 (poly8x16_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_s32_p16 (poly16x8_t __a)
 {
   return (int32x4_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_f16 (float16x4_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_f64 (float64x1_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_s8 (int8x8_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_s16 (int16x4_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_s32 (int32x2_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_s64 (int64x1_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_f32 (float32x2_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_u16 (uint16x4_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_u32 (uint32x2_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_u64 (uint64x1_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_p8 (poly8x8_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u8_p16 (poly16x4_t __a)
 {
   return (uint8x8_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_f64 (float64x2_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_s8 (int8x16_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_s16 (int16x8_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_s32 (int32x4_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_s64 (int64x2_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_f16 (float16x8_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_f32 (float32x4_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_u16 (uint16x8_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_u32 (uint32x4_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_u64 (uint64x2_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_p8 (poly8x16_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u8_p16 (poly16x8_t __a)
 {
   return (uint8x16_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_f16 (float16x4_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_f64 (float64x1_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_s8 (int8x8_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_s16 (int16x4_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_s32 (int32x2_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_s64 (int64x1_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_f32 (float32x2_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_u8 (uint8x8_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_u32 (uint32x2_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_u64 (uint64x1_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_p8 (poly8x8_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u16_p16 (poly16x4_t __a)
 {
   return (uint16x4_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_f64 (float64x2_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_s8 (int8x16_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_s16 (int16x8_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_s32 (int32x4_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_s64 (int64x2_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_f16 (float16x8_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_f32 (float32x4_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_u8 (uint8x16_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_u32 (uint32x4_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_u64 (uint64x2_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_p8 (poly8x16_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u16_p16 (poly16x8_t __a)
 {
   return (uint16x8_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_f16 (float16x4_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_f64 (float64x1_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_s8 (int8x8_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_s16 (int16x4_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_s32 (int32x2_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_s64 (int64x1_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_f32 (float32x2_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_u8 (uint8x8_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_u16 (uint16x4_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_u64 (uint64x1_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_p8 (poly8x8_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpret_u32_p16 (poly16x4_t __a)
 {
   return (uint32x2_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_f64 (float64x2_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_s8 (int8x16_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_s16 (int16x8_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_s32 (int32x4_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_s64 (int64x2_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_f16 (float16x8_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_f32 (float32x4_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_u8 (uint8x16_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_u16 (uint16x8_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_u64 (uint64x2_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_p8 (poly8x16_t __a)
 {
   return (uint32x4_t) __a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vreinterpretq_u32_p16 (poly16x8_t __a)
 {
   return (uint32x4_t) __a;
@@ -4771,79 +5450,92 @@ vreinterpretq_u32_p16 (poly16x8_t __a)
 
 /* vset_lane  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_f16 (float16_t __elem, float16x4_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_f32 (float32_t __elem, float32x2_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_f64 (float64_t __elem, float64x1_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_p8 (poly8_t __elem, poly8x8_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_p16 (poly16_t __elem, poly16x4_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_s8 (int8_t __elem, int8x8_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_s16 (int16_t __elem, int16x4_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_s32 (int32_t __elem, int32x2_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_s64 (int64_t __elem, int64x1_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_u8 (uint8_t __elem, uint8x8_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_u16 (uint16_t __elem, uint16x4_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_u32 (uint32_t __elem, uint32x2_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vset_lane_u64 (uint64_t __elem, uint64x1_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
@@ -4851,79 +5543,92 @@ vset_lane_u64 (uint64_t __elem, uint64x1_t __vec, const int __index)
 
 /* vsetq_lane  */
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_f16 (float16_t __elem, float16x8_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_f32 (float32_t __elem, float32x4_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_f64 (float64_t __elem, float64x2_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_p8 (poly8_t __elem, poly8x16_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_p16 (poly16_t __elem, poly16x8_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_s8 (int8_t __elem, int8x16_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_s16 (int16_t __elem, int16x8_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_s32 (int32_t __elem, int32x4_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_s64 (int64_t __elem, int64x2_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_u8 (uint8_t __elem, uint8x16_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_u16 (uint16_t __elem, uint16x8_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_u32 (uint32_t __elem, uint32x4_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_u64 (uint64_t __elem, uint64x2_t __vec, const int __index)
 {
   return __aarch64_vset_lane_any (__elem, __vec, __index);
@@ -4934,79 +5639,92 @@ vsetq_lane_u64 (uint64_t __elem, uint64x2_t __vec, const int __index)
   uint64x1_t lo = vcreate_u64 (vgetq_lane_u64 (tmp, 0));  \
   return vreinterpret_##__TYPE##_u64 (lo);
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_f16 (float16x8_t __a)
 {
   __GET_LOW (f16);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_f32 (float32x4_t __a)
 {
   __GET_LOW (f32);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_f64 (float64x2_t __a)
 {
   return (float64x1_t) {vgetq_lane_f64 (__a, 0)};
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_p8 (poly8x16_t __a)
 {
   __GET_LOW (p8);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_p16 (poly16x8_t __a)
 {
   __GET_LOW (p16);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_s8 (int8x16_t __a)
 {
   __GET_LOW (s8);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_s16 (int16x8_t __a)
 {
   __GET_LOW (s16);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_s32 (int32x4_t __a)
 {
   __GET_LOW (s32);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_s64 (int64x2_t __a)
 {
   __GET_LOW (s64);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_u8 (uint8x16_t __a)
 {
   __GET_LOW (u8);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_u16 (uint16x8_t __a)
 {
   __GET_LOW (u16);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_u32 (uint32x4_t __a)
 {
   __GET_LOW (u32);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_low_u64 (uint64x2_t __a)
 {
   return vcreate_u64 (vgetq_lane_u64 (__a, 0));
@@ -5019,73 +5737,85 @@ vget_low_u64 (uint64x2_t __a)
   uint64x1_t hi = vcreate_u64 (vgetq_lane_u64 (tmp, 1));       \
   return vreinterpret_##__TYPE##_u64 (hi);
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_f16 (float16x8_t __a)
 {
   __GET_HIGH (f16);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_f32 (float32x4_t __a)
 {
   __GET_HIGH (f32);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_f64 (float64x2_t __a)
 {
   __GET_HIGH (f64);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_p8 (poly8x16_t __a)
 {
   __GET_HIGH (p8);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_p16 (poly16x8_t __a)
 {
   __GET_HIGH (p16);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_s8 (int8x16_t __a)
 {
   __GET_HIGH (s8);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_s16 (int16x8_t __a)
 {
   __GET_HIGH (s16);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_s32 (int32x4_t __a)
 {
   __GET_HIGH (s32);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_s64 (int64x2_t __a)
 {
   __GET_HIGH (s64);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_u8 (uint8x16_t __a)
 {
   __GET_HIGH (u8);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_u16 (uint16x8_t __a)
 {
   __GET_HIGH (u16);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_u32 (uint32x4_t __a)
 {
   __GET_HIGH (u32);
@@ -5093,89 +5823,103 @@ vget_high_u32 (uint32x4_t __a)
 
 #undef __GET_HIGH
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vget_high_u64 (uint64x2_t __a)
 {
   return vcreate_u64 (vgetq_lane_u64 (__a, 1));
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (int8x16_t) __builtin_aarch64_combinev8qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int16x8_t) __builtin_aarch64_combinev4hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int32x4_t) __builtin_aarch64_combinev2si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_s64 (int64x1_t __a, int64x1_t __b)
 {
   return __builtin_aarch64_combinedi (__a[0], __b[0]);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_combinev4hf (__a, __b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_f32 (float32x2_t __a, float32x2_t __b)
 {
   return (float32x4_t) __builtin_aarch64_combinev2sf (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (uint8x16_t) __builtin_aarch64_combinev8qi ((int8x8_t) __a,
                                                     (int8x8_t) __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_combinev4hi ((int16x4_t) __a,
                                                     (int16x4_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_combinev2si ((int32x2_t) __a,
                                                     (int32x2_t) __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return (uint64x2_t) __builtin_aarch64_combinedi (__a[0], __b[0]);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_f64 (float64x1_t __a, float64x1_t __b)
 {
   return __builtin_aarch64_combinedf (__a[0], __b[0]);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_p8 (poly8x8_t __a, poly8x8_t __b)
 {
   return (poly8x16_t) __builtin_aarch64_combinev8qi ((int8x8_t) __a,
                                                     (int8x8_t) __b);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcombine_p16 (poly16x4_t __a, poly16x4_t __b)
 {
   return (poly16x8_t) __builtin_aarch64_combinev4hi ((int16x4_t) __a,
@@ -5184,7 +5928,8 @@ vcombine_p16 (poly16x4_t __a, poly16x4_t __b)
 
 /* Start of temporary inline asm implementations.  */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaba_s8 (int8x8_t a, int8x8_t b, int8x8_t c)
 {
   int8x8_t result;
@@ -5195,7 +5940,8 @@ vaba_s8 (int8x8_t a, int8x8_t b, int8x8_t c)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaba_s16 (int16x4_t a, int16x4_t b, int16x4_t c)
 {
   int16x4_t result;
@@ -5206,7 +5952,8 @@ vaba_s16 (int16x4_t a, int16x4_t b, int16x4_t c)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaba_s32 (int32x2_t a, int32x2_t b, int32x2_t c)
 {
   int32x2_t result;
@@ -5217,7 +5964,8 @@ vaba_s32 (int32x2_t a, int32x2_t b, int32x2_t c)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaba_u8 (uint8x8_t a, uint8x8_t b, uint8x8_t c)
 {
   uint8x8_t result;
@@ -5228,7 +5976,8 @@ vaba_u8 (uint8x8_t a, uint8x8_t b, uint8x8_t c)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaba_u16 (uint16x4_t a, uint16x4_t b, uint16x4_t c)
 {
   uint16x4_t result;
@@ -5239,7 +5988,8 @@ vaba_u16 (uint16x4_t a, uint16x4_t b, uint16x4_t c)
   return result;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaba_u32 (uint32x2_t a, uint32x2_t b, uint32x2_t c)
 {
   uint32x2_t result;
@@ -5250,7 +6000,8 @@ vaba_u32 (uint32x2_t a, uint32x2_t b, uint32x2_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_high_s8 (int16x8_t a, int8x16_t b, int8x16_t c)
 {
   int16x8_t result;
@@ -5261,7 +6012,8 @@ vabal_high_s8 (int16x8_t a, int8x16_t b, int8x16_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_high_s16 (int32x4_t a, int16x8_t b, int16x8_t c)
 {
   int32x4_t result;
@@ -5272,7 +6024,8 @@ vabal_high_s16 (int32x4_t a, int16x8_t b, int16x8_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_high_s32 (int64x2_t a, int32x4_t b, int32x4_t c)
 {
   int64x2_t result;
@@ -5283,7 +6036,8 @@ vabal_high_s32 (int64x2_t a, int32x4_t b, int32x4_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_high_u8 (uint16x8_t a, uint8x16_t b, uint8x16_t c)
 {
   uint16x8_t result;
@@ -5294,7 +6048,8 @@ vabal_high_u8 (uint16x8_t a, uint8x16_t b, uint8x16_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_high_u16 (uint32x4_t a, uint16x8_t b, uint16x8_t c)
 {
   uint32x4_t result;
@@ -5305,7 +6060,8 @@ vabal_high_u16 (uint32x4_t a, uint16x8_t b, uint16x8_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_high_u32 (uint64x2_t a, uint32x4_t b, uint32x4_t c)
 {
   uint64x2_t result;
@@ -5316,7 +6072,8 @@ vabal_high_u32 (uint64x2_t a, uint32x4_t b, uint32x4_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_s8 (int16x8_t a, int8x8_t b, int8x8_t c)
 {
   int16x8_t result;
@@ -5327,7 +6084,8 @@ vabal_s8 (int16x8_t a, int8x8_t b, int8x8_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_s16 (int32x4_t a, int16x4_t b, int16x4_t c)
 {
   int32x4_t result;
@@ -5338,7 +6096,8 @@ vabal_s16 (int32x4_t a, int16x4_t b, int16x4_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_s32 (int64x2_t a, int32x2_t b, int32x2_t c)
 {
   int64x2_t result;
@@ -5349,7 +6108,8 @@ vabal_s32 (int64x2_t a, int32x2_t b, int32x2_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_u8 (uint16x8_t a, uint8x8_t b, uint8x8_t c)
 {
   uint16x8_t result;
@@ -5360,7 +6120,8 @@ vabal_u8 (uint16x8_t a, uint8x8_t b, uint8x8_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_u16 (uint32x4_t a, uint16x4_t b, uint16x4_t c)
 {
   uint32x4_t result;
@@ -5371,7 +6132,8 @@ vabal_u16 (uint32x4_t a, uint16x4_t b, uint16x4_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabal_u32 (uint64x2_t a, uint32x2_t b, uint32x2_t c)
 {
   uint64x2_t result;
@@ -5382,7 +6144,8 @@ vabal_u32 (uint64x2_t a, uint32x2_t b, uint32x2_t c)
   return result;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabaq_s8 (int8x16_t a, int8x16_t b, int8x16_t c)
 {
   int8x16_t result;
@@ -5393,7 +6156,8 @@ vabaq_s8 (int8x16_t a, int8x16_t b, int8x16_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabaq_s16 (int16x8_t a, int16x8_t b, int16x8_t c)
 {
   int16x8_t result;
@@ -5404,7 +6168,8 @@ vabaq_s16 (int16x8_t a, int16x8_t b, int16x8_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabaq_s32 (int32x4_t a, int32x4_t b, int32x4_t c)
 {
   int32x4_t result;
@@ -5415,7 +6180,8 @@ vabaq_s32 (int32x4_t a, int32x4_t b, int32x4_t c)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabaq_u8 (uint8x16_t a, uint8x16_t b, uint8x16_t c)
 {
   uint8x16_t result;
@@ -5426,7 +6192,8 @@ vabaq_u8 (uint8x16_t a, uint8x16_t b, uint8x16_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabaq_u16 (uint16x8_t a, uint16x8_t b, uint16x8_t c)
 {
   uint16x8_t result;
@@ -5437,7 +6204,8 @@ vabaq_u16 (uint16x8_t a, uint16x8_t b, uint16x8_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabaq_u32 (uint32x4_t a, uint32x4_t b, uint32x4_t c)
 {
   uint32x4_t result;
@@ -5448,7 +6216,8 @@ vabaq_u32 (uint32x4_t a, uint32x4_t b, uint32x4_t c)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabd_s8 (int8x8_t a, int8x8_t b)
 {
   int8x8_t result;
@@ -5459,7 +6228,8 @@ vabd_s8 (int8x8_t a, int8x8_t b)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabd_s16 (int16x4_t a, int16x4_t b)
 {
   int16x4_t result;
@@ -5470,7 +6240,8 @@ vabd_s16 (int16x4_t a, int16x4_t b)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabd_s32 (int32x2_t a, int32x2_t b)
 {
   int32x2_t result;
@@ -5481,7 +6252,8 @@ vabd_s32 (int32x2_t a, int32x2_t b)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabd_u8 (uint8x8_t a, uint8x8_t b)
 {
   uint8x8_t result;
@@ -5492,7 +6264,8 @@ vabd_u8 (uint8x8_t a, uint8x8_t b)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabd_u16 (uint16x4_t a, uint16x4_t b)
 {
   uint16x4_t result;
@@ -5503,7 +6276,8 @@ vabd_u16 (uint16x4_t a, uint16x4_t b)
   return result;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabd_u32 (uint32x2_t a, uint32x2_t b)
 {
   uint32x2_t result;
@@ -5514,7 +6288,8 @@ vabd_u32 (uint32x2_t a, uint32x2_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_s8 (int8x16_t a, int8x16_t b)
 {
   int16x8_t result;
@@ -5525,7 +6300,8 @@ vabdl_high_s8 (int8x16_t a, int8x16_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_s16 (int16x8_t a, int16x8_t b)
 {
   int32x4_t result;
@@ -5536,7 +6312,8 @@ vabdl_high_s16 (int16x8_t a, int16x8_t b)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_s32 (int32x4_t a, int32x4_t b)
 {
   int64x2_t result;
@@ -5547,7 +6324,8 @@ vabdl_high_s32 (int32x4_t a, int32x4_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_u8 (uint8x16_t a, uint8x16_t b)
 {
   uint16x8_t result;
@@ -5558,7 +6336,8 @@ vabdl_high_u8 (uint8x16_t a, uint8x16_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_u16 (uint16x8_t a, uint16x8_t b)
 {
   uint32x4_t result;
@@ -5569,7 +6348,8 @@ vabdl_high_u16 (uint16x8_t a, uint16x8_t b)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_u32 (uint32x4_t a, uint32x4_t b)
 {
   uint64x2_t result;
@@ -5580,7 +6360,8 @@ vabdl_high_u32 (uint32x4_t a, uint32x4_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_s8 (int8x8_t a, int8x8_t b)
 {
   int16x8_t result;
@@ -5591,7 +6372,8 @@ vabdl_s8 (int8x8_t a, int8x8_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_s16 (int16x4_t a, int16x4_t b)
 {
   int32x4_t result;
@@ -5602,7 +6384,8 @@ vabdl_s16 (int16x4_t a, int16x4_t b)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_s32 (int32x2_t a, int32x2_t b)
 {
   int64x2_t result;
@@ -5613,7 +6396,8 @@ vabdl_s32 (int32x2_t a, int32x2_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_u8 (uint8x8_t a, uint8x8_t b)
 {
   uint16x8_t result;
@@ -5624,7 +6408,8 @@ vabdl_u8 (uint8x8_t a, uint8x8_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_u16 (uint16x4_t a, uint16x4_t b)
 {
   uint32x4_t result;
@@ -5635,7 +6420,8 @@ vabdl_u16 (uint16x4_t a, uint16x4_t b)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_u32 (uint32x2_t a, uint32x2_t b)
 {
   uint64x2_t result;
@@ -5646,7 +6432,8 @@ vabdl_u32 (uint32x2_t a, uint32x2_t b)
   return result;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdq_s8 (int8x16_t a, int8x16_t b)
 {
   int8x16_t result;
@@ -5657,7 +6444,8 @@ vabdq_s8 (int8x16_t a, int8x16_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdq_s16 (int16x8_t a, int16x8_t b)
 {
   int16x8_t result;
@@ -5668,7 +6456,8 @@ vabdq_s16 (int16x8_t a, int16x8_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdq_s32 (int32x4_t a, int32x4_t b)
 {
   int32x4_t result;
@@ -5679,7 +6468,8 @@ vabdq_s32 (int32x4_t a, int32x4_t b)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdq_u8 (uint8x16_t a, uint8x16_t b)
 {
   uint8x16_t result;
@@ -5690,7 +6480,8 @@ vabdq_u8 (uint8x16_t a, uint8x16_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdq_u16 (uint16x8_t a, uint16x8_t b)
 {
   uint16x8_t result;
@@ -5701,7 +6492,8 @@ vabdq_u16 (uint16x8_t a, uint16x8_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdq_u32 (uint32x4_t a, uint32x4_t b)
 {
   uint32x4_t result;
@@ -5712,7 +6504,8 @@ vabdq_u32 (uint32x4_t a, uint32x4_t b)
   return result;
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlv_s8 (int8x8_t a)
 {
   int16_t result;
@@ -5723,7 +6516,8 @@ vaddlv_s8 (int8x8_t a)
   return result;
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlv_s16 (int16x4_t a)
 {
   int32_t result;
@@ -5734,7 +6528,8 @@ vaddlv_s16 (int16x4_t a)
   return result;
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlv_u8 (uint8x8_t a)
 {
   uint16_t result;
@@ -5745,7 +6540,8 @@ vaddlv_u8 (uint8x8_t a)
   return result;
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlv_u16 (uint16x4_t a)
 {
   uint32_t result;
@@ -5756,7 +6552,8 @@ vaddlv_u16 (uint16x4_t a)
   return result;
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlvq_s8 (int8x16_t a)
 {
   int16_t result;
@@ -5767,7 +6564,8 @@ vaddlvq_s8 (int8x16_t a)
   return result;
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlvq_s16 (int16x8_t a)
 {
   int32_t result;
@@ -5778,7 +6576,8 @@ vaddlvq_s16 (int16x8_t a)
   return result;
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlvq_s32 (int32x4_t a)
 {
   int64_t result;
@@ -5789,7 +6588,8 @@ vaddlvq_s32 (int32x4_t a)
   return result;
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlvq_u8 (uint8x16_t a)
 {
   uint16_t result;
@@ -5800,7 +6600,8 @@ vaddlvq_u8 (uint8x16_t a)
   return result;
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlvq_u16 (uint16x8_t a)
 {
   uint32_t result;
@@ -5811,7 +6612,8 @@ vaddlvq_u16 (uint16x8_t a)
   return result;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlvq_u32 (uint32x4_t a)
 {
   uint64_t result;
@@ -5822,7 +6624,8 @@ vaddlvq_u32 (uint32x4_t a)
   return result;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtx_f32_f64 (float64x2_t a)
 {
   float32x2_t result;
@@ -5833,7 +6636,8 @@ vcvtx_f32_f64 (float64x2_t a)
   return result;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtx_high_f32_f64 (float32x2_t a, float64x2_t b)
 {
   float32x4_t result;
@@ -5844,7 +6648,8 @@ vcvtx_high_f32_f64 (float32x2_t a, float64x2_t b)
   return result;
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtxd_f32_f64 (float64_t a)
 {
   float32_t result;
@@ -5855,7 +6660,8 @@ vcvtxd_f32_f64 (float64_t a)
   return result;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_n_f32 (float32x2_t a, float32x2_t b, float32_t c)
 {
   float32x2_t result;
@@ -5867,7 +6673,8 @@ vmla_n_f32 (float32x2_t a, float32x2_t b, float32_t c)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_n_s16 (int16x4_t a, int16x4_t b, int16_t c)
 {
   int16x4_t result;
@@ -5878,7 +6685,8 @@ vmla_n_s16 (int16x4_t a, int16x4_t b, int16_t c)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_n_s32 (int32x2_t a, int32x2_t b, int32_t c)
 {
   int32x2_t result;
@@ -5889,7 +6697,8 @@ vmla_n_s32 (int32x2_t a, int32x2_t b, int32_t c)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_n_u16 (uint16x4_t a, uint16x4_t b, uint16_t c)
 {
   uint16x4_t result;
@@ -5900,7 +6709,8 @@ vmla_n_u16 (uint16x4_t a, uint16x4_t b, uint16_t c)
   return result;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_n_u32 (uint32x2_t a, uint32x2_t b, uint32_t c)
 {
   uint32x2_t result;
@@ -5911,7 +6721,8 @@ vmla_n_u32 (uint32x2_t a, uint32x2_t b, uint32_t c)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_s8 (int8x8_t a, int8x8_t b, int8x8_t c)
 {
   int8x8_t result;
@@ -5922,7 +6733,8 @@ vmla_s8 (int8x8_t a, int8x8_t b, int8x8_t c)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_s16 (int16x4_t a, int16x4_t b, int16x4_t c)
 {
   int16x4_t result;
@@ -5933,7 +6745,8 @@ vmla_s16 (int16x4_t a, int16x4_t b, int16x4_t c)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_s32 (int32x2_t a, int32x2_t b, int32x2_t c)
 {
   int32x2_t result;
@@ -5944,7 +6757,8 @@ vmla_s32 (int32x2_t a, int32x2_t b, int32x2_t c)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_u8 (uint8x8_t a, uint8x8_t b, uint8x8_t c)
 {
   uint8x8_t result;
@@ -5955,7 +6769,8 @@ vmla_u8 (uint8x8_t a, uint8x8_t b, uint8x8_t c)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_u16 (uint16x4_t a, uint16x4_t b, uint16x4_t c)
 {
   uint16x4_t result;
@@ -5966,7 +6781,8 @@ vmla_u16 (uint16x4_t a, uint16x4_t b, uint16x4_t c)
   return result;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_u32 (uint32x2_t a, uint32x2_t b, uint32x2_t c)
 {
   uint32x2_t result;
@@ -6089,7 +6905,8 @@ vmla_u32 (uint32x2_t a, uint32x2_t b, uint32x2_t c)
        result;                                                          \
      })
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_n_s16 (int32x4_t a, int16x8_t b, int16_t c)
 {
   int32x4_t result;
@@ -6100,7 +6917,8 @@ vmlal_high_n_s16 (int32x4_t a, int16x8_t b, int16_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_n_s32 (int64x2_t a, int32x4_t b, int32_t c)
 {
   int64x2_t result;
@@ -6111,7 +6929,8 @@ vmlal_high_n_s32 (int64x2_t a, int32x4_t b, int32_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_n_u16 (uint32x4_t a, uint16x8_t b, uint16_t c)
 {
   uint32x4_t result;
@@ -6122,7 +6941,8 @@ vmlal_high_n_u16 (uint32x4_t a, uint16x8_t b, uint16_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_n_u32 (uint64x2_t a, uint32x4_t b, uint32_t c)
 {
   uint64x2_t result;
@@ -6133,7 +6953,8 @@ vmlal_high_n_u32 (uint64x2_t a, uint32x4_t b, uint32_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_s8 (int16x8_t a, int8x16_t b, int8x16_t c)
 {
   int16x8_t result;
@@ -6144,7 +6965,8 @@ vmlal_high_s8 (int16x8_t a, int8x16_t b, int8x16_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_s16 (int32x4_t a, int16x8_t b, int16x8_t c)
 {
   int32x4_t result;
@@ -6155,7 +6977,8 @@ vmlal_high_s16 (int32x4_t a, int16x8_t b, int16x8_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_s32 (int64x2_t a, int32x4_t b, int32x4_t c)
 {
   int64x2_t result;
@@ -6166,7 +6989,8 @@ vmlal_high_s32 (int64x2_t a, int32x4_t b, int32x4_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_u8 (uint16x8_t a, uint8x16_t b, uint8x16_t c)
 {
   uint16x8_t result;
@@ -6177,7 +7001,8 @@ vmlal_high_u8 (uint16x8_t a, uint8x16_t b, uint8x16_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_u16 (uint32x4_t a, uint16x8_t b, uint16x8_t c)
 {
   uint32x4_t result;
@@ -6188,7 +7013,8 @@ vmlal_high_u16 (uint32x4_t a, uint16x8_t b, uint16x8_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_high_u32 (uint64x2_t a, uint32x4_t b, uint32x4_t c)
 {
   uint64x2_t result;
@@ -6311,7 +7137,8 @@ vmlal_high_u32 (uint64x2_t a, uint32x4_t b, uint32x4_t c)
        result;                                                          \
      })
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_n_s16 (int32x4_t a, int16x4_t b, int16_t c)
 {
   int32x4_t result;
@@ -6322,7 +7149,8 @@ vmlal_n_s16 (int32x4_t a, int16x4_t b, int16_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_n_s32 (int64x2_t a, int32x2_t b, int32_t c)
 {
   int64x2_t result;
@@ -6333,7 +7161,8 @@ vmlal_n_s32 (int64x2_t a, int32x2_t b, int32_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_n_u16 (uint32x4_t a, uint16x4_t b, uint16_t c)
 {
   uint32x4_t result;
@@ -6344,7 +7173,8 @@ vmlal_n_u16 (uint32x4_t a, uint16x4_t b, uint16_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_n_u32 (uint64x2_t a, uint32x2_t b, uint32_t c)
 {
   uint64x2_t result;
@@ -6355,7 +7185,8 @@ vmlal_n_u32 (uint64x2_t a, uint32x2_t b, uint32_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_s8 (int16x8_t a, int8x8_t b, int8x8_t c)
 {
   int16x8_t result;
@@ -6366,7 +7197,8 @@ vmlal_s8 (int16x8_t a, int8x8_t b, int8x8_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_s16 (int32x4_t a, int16x4_t b, int16x4_t c)
 {
   int32x4_t result;
@@ -6377,7 +7209,8 @@ vmlal_s16 (int32x4_t a, int16x4_t b, int16x4_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_s32 (int64x2_t a, int32x2_t b, int32x2_t c)
 {
   int64x2_t result;
@@ -6388,7 +7221,8 @@ vmlal_s32 (int64x2_t a, int32x2_t b, int32x2_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_u8 (uint16x8_t a, uint8x8_t b, uint8x8_t c)
 {
   uint16x8_t result;
@@ -6399,7 +7233,8 @@ vmlal_u8 (uint16x8_t a, uint8x8_t b, uint8x8_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_u16 (uint32x4_t a, uint16x4_t b, uint16x4_t c)
 {
   uint32x4_t result;
@@ -6410,7 +7245,8 @@ vmlal_u16 (uint32x4_t a, uint16x4_t b, uint16x4_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlal_u32 (uint64x2_t a, uint32x2_t b, uint32x2_t c)
 {
   uint64x2_t result;
@@ -6421,7 +7257,8 @@ vmlal_u32 (uint64x2_t a, uint32x2_t b, uint32x2_t c)
   return result;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_n_f32 (float32x4_t a, float32x4_t b, float32_t c)
 {
   float32x4_t result;
@@ -6433,7 +7270,8 @@ vmlaq_n_f32 (float32x4_t a, float32x4_t b, float32_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_n_s16 (int16x8_t a, int16x8_t b, int16_t c)
 {
   int16x8_t result;
@@ -6444,7 +7282,8 @@ vmlaq_n_s16 (int16x8_t a, int16x8_t b, int16_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_n_s32 (int32x4_t a, int32x4_t b, int32_t c)
 {
   int32x4_t result;
@@ -6455,7 +7294,8 @@ vmlaq_n_s32 (int32x4_t a, int32x4_t b, int32_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_n_u16 (uint16x8_t a, uint16x8_t b, uint16_t c)
 {
   uint16x8_t result;
@@ -6466,7 +7306,8 @@ vmlaq_n_u16 (uint16x8_t a, uint16x8_t b, uint16_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_n_u32 (uint32x4_t a, uint32x4_t b, uint32_t c)
 {
   uint32x4_t result;
@@ -6477,7 +7318,8 @@ vmlaq_n_u32 (uint32x4_t a, uint32x4_t b, uint32_t c)
   return result;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_s8 (int8x16_t a, int8x16_t b, int8x16_t c)
 {
   int8x16_t result;
@@ -6488,7 +7330,8 @@ vmlaq_s8 (int8x16_t a, int8x16_t b, int8x16_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_s16 (int16x8_t a, int16x8_t b, int16x8_t c)
 {
   int16x8_t result;
@@ -6499,7 +7342,8 @@ vmlaq_s16 (int16x8_t a, int16x8_t b, int16x8_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_s32 (int32x4_t a, int32x4_t b, int32x4_t c)
 {
   int32x4_t result;
@@ -6510,7 +7354,8 @@ vmlaq_s32 (int32x4_t a, int32x4_t b, int32x4_t c)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_u8 (uint8x16_t a, uint8x16_t b, uint8x16_t c)
 {
   uint8x16_t result;
@@ -6521,7 +7366,8 @@ vmlaq_u8 (uint8x16_t a, uint8x16_t b, uint8x16_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_u16 (uint16x8_t a, uint16x8_t b, uint16x8_t c)
 {
   uint16x8_t result;
@@ -6532,7 +7378,8 @@ vmlaq_u16 (uint16x8_t a, uint16x8_t b, uint16x8_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_u32 (uint32x4_t a, uint32x4_t b, uint32x4_t c)
 {
   uint32x4_t result;
@@ -6543,7 +7390,8 @@ vmlaq_u32 (uint32x4_t a, uint32x4_t b, uint32x4_t c)
   return result;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_n_f32 (float32x2_t a, float32x2_t b, float32_t c)
 {
   float32x2_t result;
@@ -6555,7 +7403,8 @@ vmls_n_f32 (float32x2_t a, float32x2_t b, float32_t c)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_n_s16 (int16x4_t a, int16x4_t b, int16_t c)
 {
   int16x4_t result;
@@ -6566,7 +7415,8 @@ vmls_n_s16 (int16x4_t a, int16x4_t b, int16_t c)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_n_s32 (int32x2_t a, int32x2_t b, int32_t c)
 {
   int32x2_t result;
@@ -6577,7 +7427,8 @@ vmls_n_s32 (int32x2_t a, int32x2_t b, int32_t c)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_n_u16 (uint16x4_t a, uint16x4_t b, uint16_t c)
 {
   uint16x4_t result;
@@ -6588,7 +7439,8 @@ vmls_n_u16 (uint16x4_t a, uint16x4_t b, uint16_t c)
   return result;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_n_u32 (uint32x2_t a, uint32x2_t b, uint32_t c)
 {
   uint32x2_t result;
@@ -6599,7 +7451,8 @@ vmls_n_u32 (uint32x2_t a, uint32x2_t b, uint32_t c)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_s8 (int8x8_t a, int8x8_t b, int8x8_t c)
 {
   int8x8_t result;
@@ -6610,7 +7463,8 @@ vmls_s8 (int8x8_t a, int8x8_t b, int8x8_t c)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_s16 (int16x4_t a, int16x4_t b, int16x4_t c)
 {
   int16x4_t result;
@@ -6621,7 +7475,8 @@ vmls_s16 (int16x4_t a, int16x4_t b, int16x4_t c)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_s32 (int32x2_t a, int32x2_t b, int32x2_t c)
 {
   int32x2_t result;
@@ -6632,7 +7487,8 @@ vmls_s32 (int32x2_t a, int32x2_t b, int32x2_t c)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_u8 (uint8x8_t a, uint8x8_t b, uint8x8_t c)
 {
   uint8x8_t result;
@@ -6643,7 +7499,8 @@ vmls_u8 (uint8x8_t a, uint8x8_t b, uint8x8_t c)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_u16 (uint16x4_t a, uint16x4_t b, uint16x4_t c)
 {
   uint16x4_t result;
@@ -6654,7 +7511,8 @@ vmls_u16 (uint16x4_t a, uint16x4_t b, uint16x4_t c)
   return result;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_u32 (uint32x2_t a, uint32x2_t b, uint32x2_t c)
 {
   uint32x2_t result;
@@ -6777,7 +7635,8 @@ vmls_u32 (uint32x2_t a, uint32x2_t b, uint32x2_t c)
        result;                                                          \
      })
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_n_s16 (int32x4_t a, int16x8_t b, int16_t c)
 {
   int32x4_t result;
@@ -6788,7 +7647,8 @@ vmlsl_high_n_s16 (int32x4_t a, int16x8_t b, int16_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_n_s32 (int64x2_t a, int32x4_t b, int32_t c)
 {
   int64x2_t result;
@@ -6799,7 +7659,8 @@ vmlsl_high_n_s32 (int64x2_t a, int32x4_t b, int32_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_n_u16 (uint32x4_t a, uint16x8_t b, uint16_t c)
 {
   uint32x4_t result;
@@ -6810,7 +7671,8 @@ vmlsl_high_n_u16 (uint32x4_t a, uint16x8_t b, uint16_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_n_u32 (uint64x2_t a, uint32x4_t b, uint32_t c)
 {
   uint64x2_t result;
@@ -6821,7 +7683,8 @@ vmlsl_high_n_u32 (uint64x2_t a, uint32x4_t b, uint32_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_s8 (int16x8_t a, int8x16_t b, int8x16_t c)
 {
   int16x8_t result;
@@ -6832,7 +7695,8 @@ vmlsl_high_s8 (int16x8_t a, int8x16_t b, int8x16_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_s16 (int32x4_t a, int16x8_t b, int16x8_t c)
 {
   int32x4_t result;
@@ -6843,7 +7707,8 @@ vmlsl_high_s16 (int32x4_t a, int16x8_t b, int16x8_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_s32 (int64x2_t a, int32x4_t b, int32x4_t c)
 {
   int64x2_t result;
@@ -6854,7 +7719,8 @@ vmlsl_high_s32 (int64x2_t a, int32x4_t b, int32x4_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_u8 (uint16x8_t a, uint8x16_t b, uint8x16_t c)
 {
   uint16x8_t result;
@@ -6865,7 +7731,8 @@ vmlsl_high_u8 (uint16x8_t a, uint8x16_t b, uint8x16_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_u16 (uint32x4_t a, uint16x8_t b, uint16x8_t c)
 {
   uint32x4_t result;
@@ -6876,7 +7743,8 @@ vmlsl_high_u16 (uint32x4_t a, uint16x8_t b, uint16x8_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_high_u32 (uint64x2_t a, uint32x4_t b, uint32x4_t c)
 {
   uint64x2_t result;
@@ -6999,7 +7867,8 @@ vmlsl_high_u32 (uint64x2_t a, uint32x4_t b, uint32x4_t c)
        result;                                                          \
      })
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_n_s16 (int32x4_t a, int16x4_t b, int16_t c)
 {
   int32x4_t result;
@@ -7010,7 +7879,8 @@ vmlsl_n_s16 (int32x4_t a, int16x4_t b, int16_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_n_s32 (int64x2_t a, int32x2_t b, int32_t c)
 {
   int64x2_t result;
@@ -7021,7 +7891,8 @@ vmlsl_n_s32 (int64x2_t a, int32x2_t b, int32_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_n_u16 (uint32x4_t a, uint16x4_t b, uint16_t c)
 {
   uint32x4_t result;
@@ -7032,7 +7903,8 @@ vmlsl_n_u16 (uint32x4_t a, uint16x4_t b, uint16_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_n_u32 (uint64x2_t a, uint32x2_t b, uint32_t c)
 {
   uint64x2_t result;
@@ -7043,7 +7915,8 @@ vmlsl_n_u32 (uint64x2_t a, uint32x2_t b, uint32_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_s8 (int16x8_t a, int8x8_t b, int8x8_t c)
 {
   int16x8_t result;
@@ -7054,7 +7927,8 @@ vmlsl_s8 (int16x8_t a, int8x8_t b, int8x8_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_s16 (int32x4_t a, int16x4_t b, int16x4_t c)
 {
   int32x4_t result;
@@ -7065,7 +7939,8 @@ vmlsl_s16 (int32x4_t a, int16x4_t b, int16x4_t c)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_s32 (int64x2_t a, int32x2_t b, int32x2_t c)
 {
   int64x2_t result;
@@ -7076,7 +7951,8 @@ vmlsl_s32 (int64x2_t a, int32x2_t b, int32x2_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_u8 (uint16x8_t a, uint8x8_t b, uint8x8_t c)
 {
   uint16x8_t result;
@@ -7087,7 +7963,8 @@ vmlsl_u8 (uint16x8_t a, uint8x8_t b, uint8x8_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_u16 (uint32x4_t a, uint16x4_t b, uint16x4_t c)
 {
   uint32x4_t result;
@@ -7098,7 +7975,8 @@ vmlsl_u16 (uint32x4_t a, uint16x4_t b, uint16x4_t c)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsl_u32 (uint64x2_t a, uint32x2_t b, uint32x2_t c)
 {
   uint64x2_t result;
@@ -7109,7 +7987,8 @@ vmlsl_u32 (uint64x2_t a, uint32x2_t b, uint32x2_t c)
   return result;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_n_f32 (float32x4_t a, float32x4_t b, float32_t c)
 {
   float32x4_t result;
@@ -7121,7 +8000,8 @@ vmlsq_n_f32 (float32x4_t a, float32x4_t b, float32_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_n_s16 (int16x8_t a, int16x8_t b, int16_t c)
 {
   int16x8_t result;
@@ -7132,7 +8012,8 @@ vmlsq_n_s16 (int16x8_t a, int16x8_t b, int16_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_n_s32 (int32x4_t a, int32x4_t b, int32_t c)
 {
   int32x4_t result;
@@ -7143,7 +8024,8 @@ vmlsq_n_s32 (int32x4_t a, int32x4_t b, int32_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_n_u16 (uint16x8_t a, uint16x8_t b, uint16_t c)
 {
   uint16x8_t result;
@@ -7154,7 +8036,8 @@ vmlsq_n_u16 (uint16x8_t a, uint16x8_t b, uint16_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_n_u32 (uint32x4_t a, uint32x4_t b, uint32_t c)
 {
   uint32x4_t result;
@@ -7165,7 +8048,8 @@ vmlsq_n_u32 (uint32x4_t a, uint32x4_t b, uint32_t c)
   return result;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_s8 (int8x16_t a, int8x16_t b, int8x16_t c)
 {
   int8x16_t result;
@@ -7176,7 +8060,8 @@ vmlsq_s8 (int8x16_t a, int8x16_t b, int8x16_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_s16 (int16x8_t a, int16x8_t b, int16x8_t c)
 {
   int16x8_t result;
@@ -7187,7 +8072,8 @@ vmlsq_s16 (int16x8_t a, int16x8_t b, int16x8_t c)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_s32 (int32x4_t a, int32x4_t b, int32x4_t c)
 {
   int32x4_t result;
@@ -7198,7 +8084,8 @@ vmlsq_s32 (int32x4_t a, int32x4_t b, int32x4_t c)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_u8 (uint8x16_t a, uint8x16_t b, uint8x16_t c)
 {
   uint8x16_t result;
@@ -7209,7 +8096,8 @@ vmlsq_u8 (uint8x16_t a, uint8x16_t b, uint8x16_t c)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_u16 (uint16x8_t a, uint16x8_t b, uint16x8_t c)
 {
   uint16x8_t result;
@@ -7220,7 +8108,8 @@ vmlsq_u16 (uint16x8_t a, uint16x8_t b, uint16x8_t c)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_u32 (uint32x4_t a, uint32x4_t b, uint32x4_t c)
 {
   uint32x4_t result;
@@ -7231,7 +8120,8 @@ vmlsq_u32 (uint32x4_t a, uint32x4_t b, uint32x4_t c)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_high_s8 (int8x16_t a)
 {
   int16x8_t result;
@@ -7242,7 +8132,8 @@ vmovl_high_s8 (int8x16_t a)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_high_s16 (int16x8_t a)
 {
   int32x4_t result;
@@ -7253,7 +8144,8 @@ vmovl_high_s16 (int16x8_t a)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_high_s32 (int32x4_t a)
 {
   int64x2_t result;
@@ -7264,7 +8156,8 @@ vmovl_high_s32 (int32x4_t a)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_high_u8 (uint8x16_t a)
 {
   uint16x8_t result;
@@ -7275,7 +8168,8 @@ vmovl_high_u8 (uint8x16_t a)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_high_u16 (uint16x8_t a)
 {
   uint32x4_t result;
@@ -7286,7 +8180,8 @@ vmovl_high_u16 (uint16x8_t a)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_high_u32 (uint32x4_t a)
 {
   uint64x2_t result;
@@ -7297,7 +8192,8 @@ vmovl_high_u32 (uint32x4_t a)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_s8 (int8x8_t a)
 {
   int16x8_t result;
@@ -7308,7 +8204,8 @@ vmovl_s8 (int8x8_t a)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_s16 (int16x4_t a)
 {
   int32x4_t result;
@@ -7319,7 +8216,8 @@ vmovl_s16 (int16x4_t a)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_s32 (int32x2_t a)
 {
   int64x2_t result;
@@ -7330,7 +8228,8 @@ vmovl_s32 (int32x2_t a)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_u8 (uint8x8_t a)
 {
   uint16x8_t result;
@@ -7341,7 +8240,8 @@ vmovl_u8 (uint8x8_t a)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_u16 (uint16x4_t a)
 {
   uint32x4_t result;
@@ -7352,7 +8252,8 @@ vmovl_u16 (uint16x4_t a)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovl_u32 (uint32x2_t a)
 {
   uint64x2_t result;
@@ -7363,7 +8264,8 @@ vmovl_u32 (uint32x2_t a)
   return result;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_high_s16 (int8x8_t a, int16x8_t b)
 {
   int8x16_t result = vcombine_s8 (a, vcreate_s8 (__AARCH64_UINT64_C (0x0)));
@@ -7374,7 +8276,8 @@ vmovn_high_s16 (int8x8_t a, int16x8_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_high_s32 (int16x4_t a, int32x4_t b)
 {
   int16x8_t result = vcombine_s16 (a, vcreate_s16 (__AARCH64_UINT64_C (0x0)));
@@ -7385,7 +8288,8 @@ vmovn_high_s32 (int16x4_t a, int32x4_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_high_s64 (int32x2_t a, int64x2_t b)
 {
   int32x4_t result = vcombine_s32 (a, vcreate_s32 (__AARCH64_UINT64_C (0x0)));
@@ -7396,7 +8300,8 @@ vmovn_high_s64 (int32x2_t a, int64x2_t b)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_high_u16 (uint8x8_t a, uint16x8_t b)
 {
   uint8x16_t result = vcombine_u8 (a, vcreate_u8 (__AARCH64_UINT64_C (0x0)));
@@ -7407,7 +8312,8 @@ vmovn_high_u16 (uint8x8_t a, uint16x8_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_high_u32 (uint16x4_t a, uint32x4_t b)
 {
   uint16x8_t result = vcombine_u16 (a, vcreate_u16 (__AARCH64_UINT64_C (0x0)));
@@ -7418,7 +8324,8 @@ vmovn_high_u32 (uint16x4_t a, uint32x4_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_high_u64 (uint32x2_t a, uint64x2_t b)
 {
   uint32x4_t result = vcombine_u32 (a, vcreate_u32 (__AARCH64_UINT64_C (0x0)));
@@ -7429,7 +8336,8 @@ vmovn_high_u64 (uint32x2_t a, uint64x2_t b)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_s16 (int16x8_t a)
 {
   int8x8_t result;
@@ -7440,7 +8348,8 @@ vmovn_s16 (int16x8_t a)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_s32 (int32x4_t a)
 {
   int16x4_t result;
@@ -7451,7 +8360,8 @@ vmovn_s32 (int32x4_t a)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_s64 (int64x2_t a)
 {
   int32x2_t result;
@@ -7462,7 +8372,8 @@ vmovn_s64 (int64x2_t a)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_u16 (uint16x8_t a)
 {
   uint8x8_t result;
@@ -7473,7 +8384,8 @@ vmovn_u16 (uint16x8_t a)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_u32 (uint32x4_t a)
 {
   uint16x4_t result;
@@ -7484,7 +8396,8 @@ vmovn_u32 (uint32x4_t a)
   return result;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovn_u64 (uint64x2_t a)
 {
   uint32x2_t result;
@@ -7599,7 +8512,8 @@ vmovn_u64 (uint64x2_t a)
        result;                                                          \
      })
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_n_s16 (int16x8_t a, int16_t b)
 {
   int32x4_t result;
@@ -7610,7 +8524,8 @@ vmull_high_n_s16 (int16x8_t a, int16_t b)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_n_s32 (int32x4_t a, int32_t b)
 {
   int64x2_t result;
@@ -7621,7 +8536,8 @@ vmull_high_n_s32 (int32x4_t a, int32_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_n_u16 (uint16x8_t a, uint16_t b)
 {
   uint32x4_t result;
@@ -7632,7 +8548,8 @@ vmull_high_n_u16 (uint16x8_t a, uint16_t b)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_n_u32 (uint32x4_t a, uint32_t b)
 {
   uint64x2_t result;
@@ -7643,7 +8560,8 @@ vmull_high_n_u32 (uint32x4_t a, uint32_t b)
   return result;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_p8 (poly8x16_t a, poly8x16_t b)
 {
   poly16x8_t result;
@@ -7654,7 +8572,8 @@ vmull_high_p8 (poly8x16_t a, poly8x16_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_s8 (int8x16_t a, int8x16_t b)
 {
   int16x8_t result;
@@ -7665,7 +8584,8 @@ vmull_high_s8 (int8x16_t a, int8x16_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_s16 (int16x8_t a, int16x8_t b)
 {
   int32x4_t result;
@@ -7676,7 +8596,8 @@ vmull_high_s16 (int16x8_t a, int16x8_t b)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_s32 (int32x4_t a, int32x4_t b)
 {
   int64x2_t result;
@@ -7687,7 +8608,8 @@ vmull_high_s32 (int32x4_t a, int32x4_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_u8 (uint8x16_t a, uint8x16_t b)
 {
   uint16x8_t result;
@@ -7698,7 +8620,8 @@ vmull_high_u8 (uint8x16_t a, uint8x16_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_u16 (uint16x8_t a, uint16x8_t b)
 {
   uint32x4_t result;
@@ -7709,7 +8632,8 @@ vmull_high_u16 (uint16x8_t a, uint16x8_t b)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_u32 (uint32x4_t a, uint32x4_t b)
 {
   uint64x2_t result;
@@ -7824,7 +8748,8 @@ vmull_high_u32 (uint32x4_t a, uint32x4_t b)
        result;                                                          \
      })
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_n_s16 (int16x4_t a, int16_t b)
 {
   int32x4_t result;
@@ -7835,7 +8760,8 @@ vmull_n_s16 (int16x4_t a, int16_t b)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_n_s32 (int32x2_t a, int32_t b)
 {
   int64x2_t result;
@@ -7846,7 +8772,8 @@ vmull_n_s32 (int32x2_t a, int32_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_n_u16 (uint16x4_t a, uint16_t b)
 {
   uint32x4_t result;
@@ -7857,7 +8784,8 @@ vmull_n_u16 (uint16x4_t a, uint16_t b)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_n_u32 (uint32x2_t a, uint32_t b)
 {
   uint64x2_t result;
@@ -7868,7 +8796,8 @@ vmull_n_u32 (uint32x2_t a, uint32_t b)
   return result;
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_p8 (poly8x8_t a, poly8x8_t b)
 {
   poly16x8_t result;
@@ -7879,7 +8808,8 @@ vmull_p8 (poly8x8_t a, poly8x8_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_s8 (int8x8_t a, int8x8_t b)
 {
   int16x8_t result;
@@ -7890,7 +8820,8 @@ vmull_s8 (int8x8_t a, int8x8_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_s16 (int16x4_t a, int16x4_t b)
 {
   int32x4_t result;
@@ -7901,7 +8832,8 @@ vmull_s16 (int16x4_t a, int16x4_t b)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_s32 (int32x2_t a, int32x2_t b)
 {
   int64x2_t result;
@@ -7912,7 +8844,8 @@ vmull_s32 (int32x2_t a, int32x2_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_u8 (uint8x8_t a, uint8x8_t b)
 {
   uint16x8_t result;
@@ -7923,7 +8856,8 @@ vmull_u8 (uint8x8_t a, uint8x8_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_u16 (uint16x4_t a, uint16x4_t b)
 {
   uint32x4_t result;
@@ -7934,7 +8868,8 @@ vmull_u16 (uint16x4_t a, uint16x4_t b)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_u32 (uint32x2_t a, uint32x2_t b)
 {
   uint64x2_t result;
@@ -7945,7 +8880,8 @@ vmull_u32 (uint32x2_t a, uint32x2_t b)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadal_s8 (int16x4_t a, int8x8_t b)
 {
   int16x4_t result;
@@ -7956,7 +8892,8 @@ vpadal_s8 (int16x4_t a, int8x8_t b)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadal_s16 (int32x2_t a, int16x4_t b)
 {
   int32x2_t result;
@@ -7967,7 +8904,8 @@ vpadal_s16 (int32x2_t a, int16x4_t b)
   return result;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadal_s32 (int64x1_t a, int32x2_t b)
 {
   int64x1_t result;
@@ -7978,7 +8916,8 @@ vpadal_s32 (int64x1_t a, int32x2_t b)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadal_u8 (uint16x4_t a, uint8x8_t b)
 {
   uint16x4_t result;
@@ -7989,7 +8928,8 @@ vpadal_u8 (uint16x4_t a, uint8x8_t b)
   return result;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadal_u16 (uint32x2_t a, uint16x4_t b)
 {
   uint32x2_t result;
@@ -8000,7 +8940,8 @@ vpadal_u16 (uint32x2_t a, uint16x4_t b)
   return result;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadal_u32 (uint64x1_t a, uint32x2_t b)
 {
   uint64x1_t result;
@@ -8011,7 +8952,8 @@ vpadal_u32 (uint64x1_t a, uint32x2_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadalq_s8 (int16x8_t a, int8x16_t b)
 {
   int16x8_t result;
@@ -8022,7 +8964,8 @@ vpadalq_s8 (int16x8_t a, int8x16_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadalq_s16 (int32x4_t a, int16x8_t b)
 {
   int32x4_t result;
@@ -8033,7 +8976,8 @@ vpadalq_s16 (int32x4_t a, int16x8_t b)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadalq_s32 (int64x2_t a, int32x4_t b)
 {
   int64x2_t result;
@@ -8044,7 +8988,8 @@ vpadalq_s32 (int64x2_t a, int32x4_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadalq_u8 (uint16x8_t a, uint8x16_t b)
 {
   uint16x8_t result;
@@ -8055,7 +9000,8 @@ vpadalq_u8 (uint16x8_t a, uint8x16_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadalq_u16 (uint32x4_t a, uint16x8_t b)
 {
   uint32x4_t result;
@@ -8066,7 +9012,8 @@ vpadalq_u16 (uint32x4_t a, uint16x8_t b)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadalq_u32 (uint64x2_t a, uint32x4_t b)
 {
   uint64x2_t result;
@@ -8077,7 +9024,8 @@ vpadalq_u32 (uint64x2_t a, uint32x4_t b)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddl_s8 (int8x8_t a)
 {
   int16x4_t result;
@@ -8088,7 +9036,8 @@ vpaddl_s8 (int8x8_t a)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddl_s16 (int16x4_t a)
 {
   int32x2_t result;
@@ -8099,7 +9048,8 @@ vpaddl_s16 (int16x4_t a)
   return result;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddl_s32 (int32x2_t a)
 {
   int64x1_t result;
@@ -8110,7 +9060,8 @@ vpaddl_s32 (int32x2_t a)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddl_u8 (uint8x8_t a)
 {
   uint16x4_t result;
@@ -8121,7 +9072,8 @@ vpaddl_u8 (uint8x8_t a)
   return result;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddl_u16 (uint16x4_t a)
 {
   uint32x2_t result;
@@ -8132,7 +9084,8 @@ vpaddl_u16 (uint16x4_t a)
   return result;
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddl_u32 (uint32x2_t a)
 {
   uint64x1_t result;
@@ -8143,7 +9096,8 @@ vpaddl_u32 (uint32x2_t a)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddlq_s8 (int8x16_t a)
 {
   int16x8_t result;
@@ -8154,7 +9108,8 @@ vpaddlq_s8 (int8x16_t a)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddlq_s16 (int16x8_t a)
 {
   int32x4_t result;
@@ -8165,7 +9120,8 @@ vpaddlq_s16 (int16x8_t a)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddlq_s32 (int32x4_t a)
 {
   int64x2_t result;
@@ -8176,7 +9132,8 @@ vpaddlq_s32 (int32x4_t a)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddlq_u8 (uint8x16_t a)
 {
   uint16x8_t result;
@@ -8187,7 +9144,8 @@ vpaddlq_u8 (uint8x16_t a)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddlq_u16 (uint16x8_t a)
 {
   uint32x4_t result;
@@ -8198,7 +9156,8 @@ vpaddlq_u16 (uint16x8_t a)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddlq_u32 (uint32x4_t a)
 {
   uint64x2_t result;
@@ -8209,7 +9168,8 @@ vpaddlq_u32 (uint32x4_t a)
   return result;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_s8 (int8x16_t a, int8x16_t b)
 {
   int8x16_t result;
@@ -8220,7 +9180,8 @@ vpaddq_s8 (int8x16_t a, int8x16_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_s16 (int16x8_t a, int16x8_t b)
 {
   int16x8_t result;
@@ -8231,7 +9192,8 @@ vpaddq_s16 (int16x8_t a, int16x8_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_s32 (int32x4_t a, int32x4_t b)
 {
   int32x4_t result;
@@ -8242,7 +9204,8 @@ vpaddq_s32 (int32x4_t a, int32x4_t b)
   return result;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_s64 (int64x2_t a, int64x2_t b)
 {
   int64x2_t result;
@@ -8253,7 +9216,8 @@ vpaddq_s64 (int64x2_t a, int64x2_t b)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_u8 (uint8x16_t a, uint8x16_t b)
 {
   uint8x16_t result;
@@ -8264,7 +9228,8 @@ vpaddq_u8 (uint8x16_t a, uint8x16_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_u16 (uint16x8_t a, uint16x8_t b)
 {
   uint16x8_t result;
@@ -8275,7 +9240,8 @@ vpaddq_u16 (uint16x8_t a, uint16x8_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_u32 (uint32x4_t a, uint32x4_t b)
 {
   uint32x4_t result;
@@ -8286,7 +9252,8 @@ vpaddq_u32 (uint32x4_t a, uint32x4_t b)
   return result;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_u64 (uint64x2_t a, uint64x2_t b)
 {
   uint64x2_t result;
@@ -8297,7 +9264,8 @@ vpaddq_u64 (uint64x2_t a, uint64x2_t b)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulh_n_s16 (int16x4_t a, int16_t b)
 {
   int16x4_t result;
@@ -8308,7 +9276,8 @@ vqdmulh_n_s16 (int16x4_t a, int16_t b)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulh_n_s32 (int32x2_t a, int32_t b)
 {
   int32x2_t result;
@@ -8319,7 +9288,8 @@ vqdmulh_n_s32 (int32x2_t a, int32_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhq_n_s16 (int16x8_t a, int16_t b)
 {
   int16x8_t result;
@@ -8330,7 +9300,8 @@ vqdmulhq_n_s16 (int16x8_t a, int16_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhq_n_s32 (int32x4_t a, int32_t b)
 {
   int32x4_t result;
@@ -8341,7 +9312,8 @@ vqdmulhq_n_s32 (int32x4_t a, int32_t b)
   return result;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_high_s16 (int8x8_t a, int16x8_t b)
 {
   int8x16_t result = vcombine_s8 (a, vcreate_s8 (__AARCH64_UINT64_C (0x0)));
@@ -8352,7 +9324,8 @@ vqmovn_high_s16 (int8x8_t a, int16x8_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_high_s32 (int16x4_t a, int32x4_t b)
 {
   int16x8_t result = vcombine_s16 (a, vcreate_s16 (__AARCH64_UINT64_C (0x0)));
@@ -8363,7 +9336,8 @@ vqmovn_high_s32 (int16x4_t a, int32x4_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_high_s64 (int32x2_t a, int64x2_t b)
 {
   int32x4_t result = vcombine_s32 (a, vcreate_s32 (__AARCH64_UINT64_C (0x0)));
@@ -8374,7 +9348,8 @@ vqmovn_high_s64 (int32x2_t a, int64x2_t b)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_high_u16 (uint8x8_t a, uint16x8_t b)
 {
   uint8x16_t result = vcombine_u8 (a, vcreate_u8 (__AARCH64_UINT64_C (0x0)));
@@ -8385,7 +9360,8 @@ vqmovn_high_u16 (uint8x8_t a, uint16x8_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_high_u32 (uint16x4_t a, uint32x4_t b)
 {
   uint16x8_t result = vcombine_u16 (a, vcreate_u16 (__AARCH64_UINT64_C (0x0)));
@@ -8396,7 +9372,8 @@ vqmovn_high_u32 (uint16x4_t a, uint32x4_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_high_u64 (uint32x2_t a, uint64x2_t b)
 {
   uint32x4_t result = vcombine_u32 (a, vcreate_u32 (__AARCH64_UINT64_C (0x0)));
@@ -8407,7 +9384,8 @@ vqmovn_high_u64 (uint32x2_t a, uint64x2_t b)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovun_high_s16 (uint8x8_t a, int16x8_t b)
 {
   uint8x16_t result = vcombine_u8 (a, vcreate_u8 (__AARCH64_UINT64_C (0x0)));
@@ -8418,7 +9396,8 @@ vqmovun_high_s16 (uint8x8_t a, int16x8_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovun_high_s32 (uint16x4_t a, int32x4_t b)
 {
   uint16x8_t result = vcombine_u16 (a, vcreate_u16 (__AARCH64_UINT64_C (0x0)));
@@ -8429,7 +9408,8 @@ vqmovun_high_s32 (uint16x4_t a, int32x4_t b)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovun_high_s64 (uint32x2_t a, int64x2_t b)
 {
   uint32x4_t result = vcombine_u32 (a, vcreate_u32 (__AARCH64_UINT64_C (0x0)));
@@ -8440,7 +9420,8 @@ vqmovun_high_s64 (uint32x2_t a, int64x2_t b)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulh_n_s16 (int16x4_t a, int16_t b)
 {
   int16x4_t result;
@@ -8451,7 +9432,8 @@ vqrdmulh_n_s16 (int16x4_t a, int16_t b)
   return result;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulh_n_s32 (int32x2_t a, int32_t b)
 {
   int32x2_t result;
@@ -8462,7 +9444,8 @@ vqrdmulh_n_s32 (int32x2_t a, int32_t b)
   return result;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhq_n_s16 (int16x8_t a, int16_t b)
 {
   int16x8_t result;
@@ -8473,7 +9456,8 @@ vqrdmulhq_n_s16 (int16x8_t a, int16_t b)
   return result;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhq_n_s32 (int32x4_t a, int32_t b)
 {
   int32x4_t result;
@@ -8916,7 +9900,8 @@ vqrdmulhq_n_s32 (int32x4_t a, int32_t b)
        result;                                                          \
      })
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrte_u32 (uint32x2_t a)
 {
   uint32x2_t result;
@@ -8927,7 +9912,8 @@ vrsqrte_u32 (uint32x2_t a)
   return result;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrteq_u32 (uint32x4_t a)
 {
   uint32x4_t result;
@@ -9204,7 +10190,8 @@ vrsqrteq_u32 (uint32x4_t a)
        result;                                                          \
      })
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_p8 (poly8x8_t a, poly8x8_t b)
 {
   uint8x8_t result;
@@ -9215,7 +10202,8 @@ vtst_p8 (poly8x8_t a, poly8x8_t b)
   return result;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_p16 (poly16x4_t a, poly16x4_t b)
 {
   uint16x4_t result;
@@ -9226,7 +10214,8 @@ vtst_p16 (poly16x4_t a, poly16x4_t b)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_p8 (poly8x16_t a, poly8x16_t b)
 {
   uint8x16_t result;
@@ -9237,7 +10226,8 @@ vtstq_p8 (poly8x16_t a, poly8x16_t b)
   return result;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_p16 (poly16x8_t a, poly16x8_t b)
 {
   uint16x8_t result;
@@ -9336,8 +10326,8 @@ __STRUCTN (float, 64, 4)
 
 #define __ST2_LANE_FUNC(intype, largetype, ptrtype, mode,                   \
                        qmode, ptr_mode, funcsuffix, signedtype)             \
-__extension__ static __inline void                                          \
-__attribute__ ((__always_inline__))                                         \
+__extension__ extern __inline void                                          \
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) \
 vst2_lane_ ## funcsuffix (ptrtype *__ptr,                                   \
                          intype __b, const int __c)                         \
 {                                                                           \
@@ -9386,8 +10376,8 @@ __ST2_LANE_FUNC (uint64x1x2_t, uint64x2x2_t, uint64_t, di, v2di, di, u64,
 
 #undef __ST2_LANE_FUNC
 #define __ST2_LANE_FUNC(intype, ptrtype, mode, ptr_mode, funcsuffix)       \
-__extension__ static __inline void                                         \
-__attribute__ ((__always_inline__))                                        \
+__extension__ extern __inline void                                         \
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) \
 vst2q_lane_ ## funcsuffix (ptrtype *__ptr,                                 \
                           intype __b, const int __c)                       \
 {                                                                          \
@@ -9413,8 +10403,8 @@ __ST2_LANE_FUNC (uint64x2x2_t, uint64_t, v2di, di, u64)
 
 #define __ST3_LANE_FUNC(intype, largetype, ptrtype, mode,                   \
                        qmode, ptr_mode, funcsuffix, signedtype)             \
-__extension__ static __inline void                                          \
-__attribute__ ((__always_inline__))                                         \
+__extension__ extern __inline void                                          \
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) \
 vst3_lane_ ## funcsuffix (ptrtype *__ptr,                                   \
                          intype __b, const int __c)                         \
 {                                                                           \
@@ -9468,8 +10458,8 @@ __ST3_LANE_FUNC (uint64x1x3_t, uint64x2x3_t, uint64_t, di, v2di, di, u64,
 
 #undef __ST3_LANE_FUNC
 #define __ST3_LANE_FUNC(intype, ptrtype, mode, ptr_mode, funcsuffix)       \
-__extension__ static __inline void                                         \
-__attribute__ ((__always_inline__))                                        \
+__extension__ extern __inline void                                         \
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) \
 vst3q_lane_ ## funcsuffix (ptrtype *__ptr,                                 \
                           intype __b, const int __c)                       \
 {                                                                          \
@@ -9495,8 +10485,8 @@ __ST3_LANE_FUNC (uint64x2x3_t, uint64_t, v2di, di, u64)
 
 #define __ST4_LANE_FUNC(intype, largetype, ptrtype, mode,                   \
                        qmode, ptr_mode, funcsuffix, signedtype)             \
-__extension__ static __inline void                                          \
-__attribute__ ((__always_inline__))                                         \
+__extension__ extern __inline void                                          \
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) \
 vst4_lane_ ## funcsuffix (ptrtype *__ptr,                                   \
                          intype __b, const int __c)                         \
 {                                                                           \
@@ -9555,8 +10545,8 @@ __ST4_LANE_FUNC (uint64x1x4_t, uint64x2x4_t, uint64_t, di, v2di, di, u64,
 
 #undef __ST4_LANE_FUNC
 #define __ST4_LANE_FUNC(intype, ptrtype, mode, ptr_mode, funcsuffix)       \
-__extension__ static __inline void                                         \
-__attribute__ ((__always_inline__))                                        \
+__extension__ extern __inline void                                         \
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) \
 vst4q_lane_ ## funcsuffix (ptrtype *__ptr,                                 \
                           intype __b, const int __c)                       \
 {                                                                          \
@@ -9580,7 +10570,8 @@ __ST4_LANE_FUNC (uint16x8x4_t, uint16_t, v8hi, hi, u16)
 __ST4_LANE_FUNC (uint32x4x4_t, uint32_t, v4si, si, u32)
 __ST4_LANE_FUNC (uint64x2x4_t, uint64_t, v2di, di, u64)
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlv_s32 (int32x2_t a)
 {
   int64_t result;
@@ -9588,7 +10579,8 @@ vaddlv_s32 (int32x2_t a)
   return result;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddlv_u32 (uint32x2_t a)
 {
   uint64_t result;
@@ -9596,49 +10588,57 @@ vaddlv_u32 (uint32x2_t a)
   return result;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulh_laneq_s16 (int16x4_t __a, int16x8_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_laneqv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulh_laneq_s32 (int32x2_t __a, int32x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_laneqv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhq_laneq_s16 (int16x8_t __a, int16x8_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_laneqv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhq_laneq_s32 (int32x4_t __a, int32x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_laneqv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulh_laneq_s16 (int16x4_t __a, int16x8_t __b, const int __c)
 {
   return  __builtin_aarch64_sqrdmulh_laneqv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulh_laneq_s32 (int32x2_t __a, int32x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_laneqv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhq_laneq_s16 (int16x8_t __a, int16x8_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_laneqv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhq_laneq_s32 (int32x4_t __a, int32x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_laneqv4si (__a, __b, __c);
@@ -9646,7 +10646,8 @@ vqrdmulhq_laneq_s32 (int32x4_t __a, int32x4_t __b, const int __c)
 
 /* Table intrinsics.  */
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl1_p8 (poly8x16_t a, uint8x8_t b)
 {
   poly8x8_t result;
@@ -9657,7 +10658,8 @@ vqtbl1_p8 (poly8x16_t a, uint8x8_t b)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl1_s8 (int8x16_t a, uint8x8_t b)
 {
   int8x8_t result;
@@ -9668,7 +10670,8 @@ vqtbl1_s8 (int8x16_t a, uint8x8_t b)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl1_u8 (uint8x16_t a, uint8x8_t b)
 {
   uint8x8_t result;
@@ -9679,7 +10682,8 @@ vqtbl1_u8 (uint8x16_t a, uint8x8_t b)
   return result;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl1q_p8 (poly8x16_t a, uint8x16_t b)
 {
   poly8x16_t result;
@@ -9690,7 +10694,8 @@ vqtbl1q_p8 (poly8x16_t a, uint8x16_t b)
   return result;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl1q_s8 (int8x16_t a, uint8x16_t b)
 {
   int8x16_t result;
@@ -9701,7 +10706,8 @@ vqtbl1q_s8 (int8x16_t a, uint8x16_t b)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl1q_u8 (uint8x16_t a, uint8x16_t b)
 {
   uint8x16_t result;
@@ -9712,7 +10718,8 @@ vqtbl1q_u8 (uint8x16_t a, uint8x16_t b)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx1_s8 (int8x8_t r, int8x16_t tab, uint8x8_t idx)
 {
   int8x8_t result = r;
@@ -9723,7 +10730,8 @@ vqtbx1_s8 (int8x8_t r, int8x16_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx1_u8 (uint8x8_t r, uint8x16_t tab, uint8x8_t idx)
 {
   uint8x8_t result = r;
@@ -9734,7 +10742,8 @@ vqtbx1_u8 (uint8x8_t r, uint8x16_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx1_p8 (poly8x8_t r, poly8x16_t tab, uint8x8_t idx)
 {
   poly8x8_t result = r;
@@ -9745,7 +10754,8 @@ vqtbx1_p8 (poly8x8_t r, poly8x16_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx1q_s8 (int8x16_t r, int8x16_t tab, uint8x16_t idx)
 {
   int8x16_t result = r;
@@ -9756,7 +10766,8 @@ vqtbx1q_s8 (int8x16_t r, int8x16_t tab, uint8x16_t idx)
   return result;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx1q_u8 (uint8x16_t r, uint8x16_t tab, uint8x16_t idx)
 {
   uint8x16_t result = r;
@@ -9767,7 +10778,8 @@ vqtbx1q_u8 (uint8x16_t r, uint8x16_t tab, uint8x16_t idx)
   return result;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx1q_p8 (poly8x16_t r, poly8x16_t tab, uint8x16_t idx)
 {
   poly8x16_t result = r;
@@ -9780,7 +10792,8 @@ vqtbx1q_p8 (poly8x16_t r, poly8x16_t tab, uint8x16_t idx)
 
 /* V7 legacy table intrinsics.  */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl1_s8 (int8x8_t tab, int8x8_t idx)
 {
   int8x8_t result;
@@ -9792,7 +10805,8 @@ vtbl1_s8 (int8x8_t tab, int8x8_t idx)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl1_u8 (uint8x8_t tab, uint8x8_t idx)
 {
   uint8x8_t result;
@@ -9804,7 +10818,8 @@ vtbl1_u8 (uint8x8_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl1_p8 (poly8x8_t tab, uint8x8_t idx)
 {
   poly8x8_t result;
@@ -9816,7 +10831,8 @@ vtbl1_p8 (poly8x8_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl2_s8 (int8x8x2_t tab, int8x8_t idx)
 {
   int8x8_t result;
@@ -9828,7 +10844,8 @@ vtbl2_s8 (int8x8x2_t tab, int8x8_t idx)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl2_u8 (uint8x8x2_t tab, uint8x8_t idx)
 {
   uint8x8_t result;
@@ -9840,7 +10857,8 @@ vtbl2_u8 (uint8x8x2_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl2_p8 (poly8x8x2_t tab, uint8x8_t idx)
 {
   poly8x8_t result;
@@ -9852,7 +10870,8 @@ vtbl2_p8 (poly8x8x2_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl3_s8 (int8x8x3_t tab, int8x8_t idx)
 {
   int8x8_t result;
@@ -9868,7 +10887,8 @@ vtbl3_s8 (int8x8x3_t tab, int8x8_t idx)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl3_u8 (uint8x8x3_t tab, uint8x8_t idx)
 {
   uint8x8_t result;
@@ -9884,7 +10904,8 @@ vtbl3_u8 (uint8x8x3_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl3_p8 (poly8x8x3_t tab, uint8x8_t idx)
 {
   poly8x8_t result;
@@ -9900,7 +10921,8 @@ vtbl3_p8 (poly8x8x3_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl4_s8 (int8x8x4_t tab, int8x8_t idx)
 {
   int8x8_t result;
@@ -9916,7 +10938,8 @@ vtbl4_s8 (int8x8x4_t tab, int8x8_t idx)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl4_u8 (uint8x8x4_t tab, uint8x8_t idx)
 {
   uint8x8_t result;
@@ -9932,7 +10955,8 @@ vtbl4_u8 (uint8x8x4_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbl4_p8 (poly8x8x4_t tab, uint8x8_t idx)
 {
   poly8x8_t result;
@@ -9948,7 +10972,8 @@ vtbl4_p8 (poly8x8x4_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx2_s8 (int8x8_t r, int8x8x2_t tab, int8x8_t idx)
 {
   int8x8_t result = r;
@@ -9960,7 +10985,8 @@ vtbx2_s8 (int8x8_t r, int8x8x2_t tab, int8x8_t idx)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx2_u8 (uint8x8_t r, uint8x8x2_t tab, uint8x8_t idx)
 {
   uint8x8_t result = r;
@@ -9972,7 +10998,8 @@ vtbx2_u8 (uint8x8_t r, uint8x8x2_t tab, uint8x8_t idx)
   return result;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx2_p8 (poly8x8_t r, poly8x8x2_t tab, uint8x8_t idx)
 {
   poly8x8_t result = r;
@@ -9990,38 +11017,44 @@ vtbx2_p8 (poly8x8_t r, poly8x8x2_t tab, uint8x8_t idx)
 
 /* vabd.  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabds_f32 (float32_t __a, float32_t __b)
 {
   return __builtin_aarch64_fabdsf (__a, __b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdd_f64 (float64_t __a, float64_t __b)
 {
   return __builtin_aarch64_fabddf (__a, __b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabd_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __builtin_aarch64_fabdv2sf (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabd_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (float64x1_t) {vabdd_f64 (vget_lane_f64 (__a, 0),
                                   vget_lane_f64 (__b, 0))};
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_fabdv4sf (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_fabdv2df (__a, __b);
@@ -10029,73 +11062,85 @@ vabdq_f64 (float64x2_t __a, float64x2_t __b)
 
 /* vabs  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabs_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_absv2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabs_f64 (float64x1_t __a)
 {
   return (float64x1_t) {__builtin_fabs (__a[0])};
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabs_s8 (int8x8_t __a)
 {
   return __builtin_aarch64_absv8qi (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabs_s16 (int16x4_t __a)
 {
   return __builtin_aarch64_absv4hi (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabs_s32 (int32x2_t __a)
 {
   return __builtin_aarch64_absv2si (__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabs_s64 (int64x1_t __a)
 {
   return (int64x1_t) {__builtin_aarch64_absdi (__a[0])};
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabsq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_absv4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabsq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_absv2df (__a);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabsq_s8 (int8x16_t __a)
 {
   return __builtin_aarch64_absv16qi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabsq_s16 (int16x8_t __a)
 {
   return __builtin_aarch64_absv8hi (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabsq_s32 (int32x4_t __a)
 {
   return __builtin_aarch64_absv4si (__a);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabsq_s64 (int64x2_t __a)
 {
   return __builtin_aarch64_absv2di (__a);
@@ -10103,13 +11148,15 @@ vabsq_s64 (int64x2_t __a)
 
 /* vadd */
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddd_s64 (int64_t __a, int64_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddd_u64 (uint64_t __a, uint64_t __b)
 {
   return __a + __b;
@@ -10117,103 +11164,120 @@ vaddd_u64 (uint64_t __a, uint64_t __b)
 
 /* vaddv */
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddv_s8 (int8x8_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v8qi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddv_s16 (int16x4_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v4hi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddv_s32 (int32x2_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v2si (__a);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddv_u8 (uint8x8_t __a)
 {
   return (uint8_t) __builtin_aarch64_reduc_plus_scal_v8qi ((int8x8_t) __a);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddv_u16 (uint16x4_t __a)
 {
   return (uint16_t) __builtin_aarch64_reduc_plus_scal_v4hi ((int16x4_t) __a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddv_u32 (uint32x2_t __a)
 {
   return (int32_t) __builtin_aarch64_reduc_plus_scal_v2si ((int32x2_t) __a);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_s8 (int8x16_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v16qi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_s16 (int16x8_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v8hi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_s32 (int32x4_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v4si (__a);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_s64 (int64x2_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v2di (__a);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_u8 (uint8x16_t __a)
 {
   return (uint8_t) __builtin_aarch64_reduc_plus_scal_v16qi ((int8x16_t) __a);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_u16 (uint16x8_t __a)
 {
   return (uint16_t) __builtin_aarch64_reduc_plus_scal_v8hi ((int16x8_t) __a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_u32 (uint32x4_t __a)
 {
   return (uint32_t) __builtin_aarch64_reduc_plus_scal_v4si ((int32x4_t) __a);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_u64 (uint64x2_t __a)
 {
   return (uint64_t) __builtin_aarch64_reduc_plus_scal_v2di ((int64x2_t) __a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddv_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v2sf (__a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v4sf (__a);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddvq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v2df (__a);
@@ -10221,160 +11285,186 @@ vaddvq_f64 (float64x2_t __a)
 
 /* vbsl  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_f16 (uint16x4_t __a, float16x4_t __b, float16x4_t __c)
 {
   return __builtin_aarch64_simd_bslv4hf_suss (__a, __b, __c);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_f32 (uint32x2_t __a, float32x2_t __b, float32x2_t __c)
 {
   return __builtin_aarch64_simd_bslv2sf_suss (__a, __b, __c);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_f64 (uint64x1_t __a, float64x1_t __b, float64x1_t __c)
 {
   return (float64x1_t)
     { __builtin_aarch64_simd_bsldf_suss (__a[0], __b[0], __c[0]) };
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_p8 (uint8x8_t __a, poly8x8_t __b, poly8x8_t __c)
 {
   return __builtin_aarch64_simd_bslv8qi_pupp (__a, __b, __c);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_p16 (uint16x4_t __a, poly16x4_t __b, poly16x4_t __c)
 {
   return __builtin_aarch64_simd_bslv4hi_pupp (__a, __b, __c);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_s8 (uint8x8_t __a, int8x8_t __b, int8x8_t __c)
 {
   return __builtin_aarch64_simd_bslv8qi_suss (__a, __b, __c);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_s16 (uint16x4_t __a, int16x4_t __b, int16x4_t __c)
 {
   return __builtin_aarch64_simd_bslv4hi_suss (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_s32 (uint32x2_t __a, int32x2_t __b, int32x2_t __c)
 {
   return __builtin_aarch64_simd_bslv2si_suss (__a, __b, __c);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_s64 (uint64x1_t __a, int64x1_t __b, int64x1_t __c)
 {
   return (int64x1_t)
       {__builtin_aarch64_simd_bsldi_suss (__a[0], __b[0], __c[0])};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_u8 (uint8x8_t __a, uint8x8_t __b, uint8x8_t __c)
 {
   return __builtin_aarch64_simd_bslv8qi_uuuu (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_u16 (uint16x4_t __a, uint16x4_t __b, uint16x4_t __c)
 {
   return __builtin_aarch64_simd_bslv4hi_uuuu (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_u32 (uint32x2_t __a, uint32x2_t __b, uint32x2_t __c)
 {
   return __builtin_aarch64_simd_bslv2si_uuuu (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbsl_u64 (uint64x1_t __a, uint64x1_t __b, uint64x1_t __c)
 {
   return (uint64x1_t)
       {__builtin_aarch64_simd_bsldi_uuuu (__a[0], __b[0], __c[0])};
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_f16 (uint16x8_t __a, float16x8_t __b, float16x8_t __c)
 {
   return __builtin_aarch64_simd_bslv8hf_suss (__a, __b, __c);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_f32 (uint32x4_t __a, float32x4_t __b, float32x4_t __c)
 {
   return __builtin_aarch64_simd_bslv4sf_suss (__a, __b, __c);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_f64 (uint64x2_t __a, float64x2_t __b, float64x2_t __c)
 {
   return __builtin_aarch64_simd_bslv2df_suss (__a, __b, __c);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_p8 (uint8x16_t __a, poly8x16_t __b, poly8x16_t __c)
 {
   return __builtin_aarch64_simd_bslv16qi_pupp (__a, __b, __c);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_p16 (uint16x8_t __a, poly16x8_t __b, poly16x8_t __c)
 {
   return __builtin_aarch64_simd_bslv8hi_pupp (__a, __b, __c);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_s8 (uint8x16_t __a, int8x16_t __b, int8x16_t __c)
 {
   return __builtin_aarch64_simd_bslv16qi_suss (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_s16 (uint16x8_t __a, int16x8_t __b, int16x8_t __c)
 {
   return __builtin_aarch64_simd_bslv8hi_suss (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_s32 (uint32x4_t __a, int32x4_t __b, int32x4_t __c)
 {
   return __builtin_aarch64_simd_bslv4si_suss (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_s64 (uint64x2_t __a, int64x2_t __b, int64x2_t __c)
 {
   return __builtin_aarch64_simd_bslv2di_suss (__a, __b, __c);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_u8 (uint8x16_t __a, uint8x16_t __b, uint8x16_t __c)
 {
   return __builtin_aarch64_simd_bslv16qi_uuuu (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_u16 (uint16x8_t __a, uint16x8_t __b, uint16x8_t __c)
 {
   return __builtin_aarch64_simd_bslv8hi_uuuu (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_u32 (uint32x4_t __a, uint32x4_t __b, uint32x4_t __c)
 {
   return __builtin_aarch64_simd_bslv4si_uuuu (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vbslq_u64 (uint64x2_t __a, uint64x2_t __b, uint64x2_t __c)
 {
   return __builtin_aarch64_simd_bslv2di_uuuu (__a, __b, __c);
@@ -10384,217 +11474,253 @@ vbslq_u64 (uint64x2_t __a, uint64x2_t __b, uint64x2_t __c)
 #pragma GCC push_options
 #pragma GCC target ("arch=armv8.1-a")
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlah_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
 {
   return __builtin_aarch64_sqrdmlahv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlah_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
 {
   return __builtin_aarch64_sqrdmlahv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
 {
   return __builtin_aarch64_sqrdmlahv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
 {
   return __builtin_aarch64_sqrdmlahv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlsh_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c)
 {
   return __builtin_aarch64_sqrdmlshv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlsh_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c)
 {
   return __builtin_aarch64_sqrdmlshv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c)
 {
   return __builtin_aarch64_sqrdmlshv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c)
 {
   return __builtin_aarch64_sqrdmlshv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlah_laneq_s16 (int16x4_t __a, int16x4_t __b, int16x8_t __c, const int __d)
 {
   return  __builtin_aarch64_sqrdmlah_laneqv4hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlah_laneq_s32 (int32x2_t __a, int32x2_t __b, int32x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_laneqv2si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahq_laneq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_laneqv8hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahq_laneq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_laneqv4si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlsh_laneq_s16 (int16x4_t __a, int16x4_t __b, int16x8_t __c, const int __d)
 {
   return  __builtin_aarch64_sqrdmlsh_laneqv4hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlsh_laneq_s32 (int32x2_t __a, int32x2_t __b, int32x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_laneqv2si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshq_laneq_s16 (int16x8_t __a, int16x8_t __b, int16x8_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_laneqv8hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshq_laneq_s32 (int32x4_t __a, int32x4_t __b, int32x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_laneqv4si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlah_lane_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
 {
   return  __builtin_aarch64_sqrdmlah_lanev4hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlah_lane_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_lanev2si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahq_lane_s16 (int16x8_t __a, int16x8_t __b, int16x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_lanev8hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahq_lane_s32 (int32x4_t __a, int32x4_t __b, int32x2_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_lanev4si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahh_s16 (int16_t __a, int16_t __b, int16_t __c)
 {
   return (int16_t) __builtin_aarch64_sqrdmlahhi (__a, __b, __c);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahh_lane_s16 (int16_t __a, int16_t __b, int16x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_lanehi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahh_laneq_s16 (int16_t __a, int16_t __b, int16x8_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_laneqhi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahs_s32 (int32_t __a, int32_t __b, int32_t __c)
 {
   return (int32_t) __builtin_aarch64_sqrdmlahsi (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahs_lane_s32 (int32_t __a, int32_t __b, int32x2_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_lanesi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlahs_laneq_s32 (int32_t __a, int32_t __b, int32x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlah_laneqsi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlsh_lane_s16 (int16x4_t __a, int16x4_t __b, int16x4_t __c, const int __d)
 {
   return  __builtin_aarch64_sqrdmlsh_lanev4hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlsh_lane_s32 (int32x2_t __a, int32x2_t __b, int32x2_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_lanev2si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshq_lane_s16 (int16x8_t __a, int16x8_t __b, int16x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_lanev8hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshq_lane_s32 (int32x4_t __a, int32x4_t __b, int32x2_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_lanev4si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshh_s16 (int16_t __a, int16_t __b, int16_t __c)
 {
   return (int16_t) __builtin_aarch64_sqrdmlshhi (__a, __b, __c);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshh_lane_s16 (int16_t __a, int16_t __b, int16x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_lanehi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshh_laneq_s16 (int16_t __a, int16_t __b, int16x8_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_laneqhi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshs_s32 (int32_t __a, int32_t __b, int32_t __c)
 {
   return (int32_t) __builtin_aarch64_sqrdmlshsi (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshs_lane_s32 (int32_t __a, int32_t __b, int32x2_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_lanesi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmlshs_laneq_s32 (int32_t __a, int32_t __b, int32x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqrdmlsh_laneqsi (__a, __b, __c, __d);
@@ -10605,25 +11731,29 @@ vqrdmlshs_laneq_s32 (int32_t __a, int32_t __b, int32x4_t __c, const int __d)
 #pragma GCC target ("+nothing+crypto")
 /* vaes  */
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaeseq_u8 (uint8x16_t data, uint8x16_t key)
 {
   return __builtin_aarch64_crypto_aesev16qi_uuu (data, key);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaesdq_u8 (uint8x16_t data, uint8x16_t key)
 {
   return __builtin_aarch64_crypto_aesdv16qi_uuu (data, key);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaesmcq_u8 (uint8x16_t data)
 {
   return __builtin_aarch64_crypto_aesmcv16qi_uu (data);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaesimcq_u8 (uint8x16_t data)
 {
   return __builtin_aarch64_crypto_aesimcv16qi_uu (data);
@@ -10632,37 +11762,43 @@ vaesimcq_u8 (uint8x16_t data)
 
 /* vcage  */
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcage_f64 (float64x1_t __a, float64x1_t __b)
 {
   return vabs_f64 (__a) >= vabs_f64 (__b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcages_f32 (float32_t __a, float32_t __b)
 {
   return __builtin_fabsf (__a) >= __builtin_fabsf (__b) ? -1 : 0;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcage_f32 (float32x2_t __a, float32x2_t __b)
 {
   return vabs_f32 (__a) >= vabs_f32 (__b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcageq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return vabsq_f32 (__a) >= vabsq_f32 (__b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcaged_f64 (float64_t __a, float64_t __b)
 {
   return __builtin_fabs (__a) >= __builtin_fabs (__b) ? -1 : 0;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcageq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return vabsq_f64 (__a) >= vabsq_f64 (__b);
@@ -10670,37 +11806,43 @@ vcageq_f64 (float64x2_t __a, float64x2_t __b)
 
 /* vcagt  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcagts_f32 (float32_t __a, float32_t __b)
 {
   return __builtin_fabsf (__a) > __builtin_fabsf (__b) ? -1 : 0;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcagt_f32 (float32x2_t __a, float32x2_t __b)
 {
   return vabs_f32 (__a) > vabs_f32 (__b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcagt_f64 (float64x1_t __a, float64x1_t __b)
 {
   return vabs_f64 (__a) > vabs_f64 (__b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcagtq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return vabsq_f32 (__a) > vabsq_f32 (__b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcagtd_f64 (float64_t __a, float64_t __b)
 {
   return __builtin_fabs (__a) > __builtin_fabs (__b) ? -1 : 0;
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcagtq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return vabsq_f64 (__a) > vabsq_f64 (__b);
@@ -10708,37 +11850,43 @@ vcagtq_f64 (float64x2_t __a, float64x2_t __b)
 
 /* vcale  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcale_f32 (float32x2_t __a, float32x2_t __b)
 {
   return vabs_f32 (__a) <= vabs_f32 (__b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcale_f64 (float64x1_t __a, float64x1_t __b)
 {
   return vabs_f64 (__a) <= vabs_f64 (__b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcaled_f64 (float64_t __a, float64_t __b)
 {
   return __builtin_fabs (__a) <= __builtin_fabs (__b) ? -1 : 0;
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcales_f32 (float32_t __a, float32_t __b)
 {
   return __builtin_fabsf (__a) <= __builtin_fabsf (__b) ? -1 : 0;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcaleq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return vabsq_f32 (__a) <= vabsq_f32 (__b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcaleq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return vabsq_f64 (__a) <= vabsq_f64 (__b);
@@ -10746,37 +11894,43 @@ vcaleq_f64 (float64x2_t __a, float64x2_t __b)
 
 /* vcalt  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcalt_f32 (float32x2_t __a, float32x2_t __b)
 {
   return vabs_f32 (__a) < vabs_f32 (__b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcalt_f64 (float64x1_t __a, float64x1_t __b)
 {
   return vabs_f64 (__a) < vabs_f64 (__b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcaltd_f64 (float64_t __a, float64_t __b)
 {
   return __builtin_fabs (__a) < __builtin_fabs (__b) ? -1 : 0;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcaltq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return vabsq_f32 (__a) < vabsq_f32 (__b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcaltq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return vabsq_f64 (__a) < vabsq_f64 (__b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcalts_f32 (float32_t __a, float32_t __b)
 {
   return __builtin_fabsf (__a) < __builtin_fabsf (__b) ? -1 : 0;
@@ -10784,133 +11938,155 @@ vcalts_f32 (float32_t __a, float32_t __b)
 
 /* vceq - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_f32 (float32x2_t __a, float32x2_t __b)
 {
   return (uint32x2_t) (__a == __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (uint64x1_t) (__a == __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_p8 (poly8x8_t __a, poly8x8_t __b)
 {
   return (uint8x8_t) (__a == __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (uint8x8_t) (__a == __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (uint16x4_t) (__a == __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (uint32x2_t) (__a == __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) (__a == __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (__a == __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (__a == __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (__a == __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return (__a == __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return (uint32x4_t) (__a == __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return (uint64x2_t) (__a == __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_p8 (poly8x16_t __a, poly8x16_t __b)
 {
   return (uint8x16_t) (__a == __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (uint8x16_t) (__a == __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (uint16x8_t) (__a == __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (uint32x4_t) (__a == __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (uint64x2_t) (__a == __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (__a == __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (__a == __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (__a == __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return (__a == __b);
@@ -10918,25 +12094,29 @@ vceqq_u64 (uint64x2_t __a, uint64x2_t __b)
 
 /* vceq - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqs_f32 (float32_t __a, float32_t __b)
 {
   return __a == __b ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqd_s64 (int64_t __a, int64_t __b)
 {
   return __a == __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqd_u64 (uint64_t __a, uint64_t __b)
 {
   return __a == __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqd_f64 (float64_t __a, float64_t __b)
 {
   return __a == __b ? -1ll : 0ll;
@@ -10944,133 +12124,155 @@ vceqd_f64 (float64_t __a, float64_t __b)
 
 /* vceqz - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_f32 (float32x2_t __a)
 {
   return (uint32x2_t) (__a == 0.0f);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_f64 (float64x1_t __a)
 {
   return (uint64x1_t) (__a == (float64x1_t) {0.0});
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_p8 (poly8x8_t __a)
 {
   return (uint8x8_t) (__a == 0);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_s8 (int8x8_t __a)
 {
   return (uint8x8_t) (__a == 0);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_s16 (int16x4_t __a)
 {
   return (uint16x4_t) (__a == 0);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_s32 (int32x2_t __a)
 {
   return (uint32x2_t) (__a == 0);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_s64 (int64x1_t __a)
 {
   return (uint64x1_t) (__a == __AARCH64_INT64_C (0));
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_u8 (uint8x8_t __a)
 {
   return (__a == 0);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_u16 (uint16x4_t __a)
 {
   return (__a == 0);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_u32 (uint32x2_t __a)
 {
   return (__a == 0);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_u64 (uint64x1_t __a)
 {
   return (__a == __AARCH64_UINT64_C (0));
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_f32 (float32x4_t __a)
 {
   return (uint32x4_t) (__a == 0.0f);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_f64 (float64x2_t __a)
 {
   return (uint64x2_t) (__a == 0.0f);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_p8 (poly8x16_t __a)
 {
   return (uint8x16_t) (__a == 0);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_s8 (int8x16_t __a)
 {
   return (uint8x16_t) (__a == 0);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_s16 (int16x8_t __a)
 {
   return (uint16x8_t) (__a == 0);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_s32 (int32x4_t __a)
 {
   return (uint32x4_t) (__a == 0);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_s64 (int64x2_t __a)
 {
   return (uint64x2_t) (__a == __AARCH64_INT64_C (0));
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_u8 (uint8x16_t __a)
 {
   return (__a == 0);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_u16 (uint16x8_t __a)
 {
   return (__a == 0);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_u32 (uint32x4_t __a)
 {
   return (__a == 0);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_u64 (uint64x2_t __a)
 {
   return (__a == __AARCH64_UINT64_C (0));
@@ -11078,25 +12280,29 @@ vceqzq_u64 (uint64x2_t __a)
 
 /* vceqz - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzs_f32 (float32_t __a)
 {
   return __a == 0.0f ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzd_s64 (int64_t __a)
 {
   return __a == 0 ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzd_u64 (uint64_t __a)
 {
   return __a == 0 ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzd_f64 (float64_t __a)
 {
   return __a == 0.0 ? -1ll : 0ll;
@@ -11104,121 +12310,141 @@ vceqzd_f64 (float64_t __a)
 
 /* vcge - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_f32 (float32x2_t __a, float32x2_t __b)
 {
   return (uint32x2_t) (__a >= __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (uint64x1_t) (__a >= __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (uint8x8_t) (__a >= __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (uint16x4_t) (__a >= __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (uint32x2_t) (__a >= __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) (__a >= __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (__a >= __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (__a >= __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (__a >= __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return (__a >= __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return (uint32x4_t) (__a >= __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return (uint64x2_t) (__a >= __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (uint8x16_t) (__a >= __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (uint16x8_t) (__a >= __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (uint32x4_t) (__a >= __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (uint64x2_t) (__a >= __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (__a >= __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (__a >= __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (__a >= __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return (__a >= __b);
@@ -11226,25 +12452,29 @@ vcgeq_u64 (uint64x2_t __a, uint64x2_t __b)
 
 /* vcge - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcges_f32 (float32_t __a, float32_t __b)
 {
   return __a >= __b ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcged_s64 (int64_t __a, int64_t __b)
 {
   return __a >= __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcged_u64 (uint64_t __a, uint64_t __b)
 {
   return __a >= __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcged_f64 (float64_t __a, float64_t __b)
 {
   return __a >= __b ? -1ll : 0ll;
@@ -11252,73 +12482,85 @@ vcged_f64 (float64_t __a, float64_t __b)
 
 /* vcgez - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgez_f32 (float32x2_t __a)
 {
   return (uint32x2_t) (__a >= 0.0f);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgez_f64 (float64x1_t __a)
 {
   return (uint64x1_t) (__a[0] >= (float64x1_t) {0.0});
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgez_s8 (int8x8_t __a)
 {
   return (uint8x8_t) (__a >= 0);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgez_s16 (int16x4_t __a)
 {
   return (uint16x4_t) (__a >= 0);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgez_s32 (int32x2_t __a)
 {
   return (uint32x2_t) (__a >= 0);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgez_s64 (int64x1_t __a)
 {
   return (uint64x1_t) (__a >= __AARCH64_INT64_C (0));
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezq_f32 (float32x4_t __a)
 {
   return (uint32x4_t) (__a >= 0.0f);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezq_f64 (float64x2_t __a)
 {
   return (uint64x2_t) (__a >= 0.0);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezq_s8 (int8x16_t __a)
 {
   return (uint8x16_t) (__a >= 0);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezq_s16 (int16x8_t __a)
 {
   return (uint16x8_t) (__a >= 0);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezq_s32 (int32x4_t __a)
 {
   return (uint32x4_t) (__a >= 0);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezq_s64 (int64x2_t __a)
 {
   return (uint64x2_t) (__a >= __AARCH64_INT64_C (0));
@@ -11326,19 +12568,22 @@ vcgezq_s64 (int64x2_t __a)
 
 /* vcgez - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezs_f32 (float32_t __a)
 {
   return __a >= 0.0f ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezd_s64 (int64_t __a)
 {
   return __a >= 0 ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezd_f64 (float64_t __a)
 {
   return __a >= 0.0 ? -1ll : 0ll;
@@ -11346,121 +12591,141 @@ vcgezd_f64 (float64_t __a)
 
 /* vcgt - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_f32 (float32x2_t __a, float32x2_t __b)
 {
   return (uint32x2_t) (__a > __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (uint64x1_t) (__a > __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (uint8x8_t) (__a > __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (uint16x4_t) (__a > __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (uint32x2_t) (__a > __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) (__a > __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (__a > __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (__a > __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (__a > __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return (__a > __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return (uint32x4_t) (__a > __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return (uint64x2_t) (__a > __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (uint8x16_t) (__a > __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (uint16x8_t) (__a > __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (uint32x4_t) (__a > __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (uint64x2_t) (__a > __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (__a > __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (__a > __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (__a > __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return (__a > __b);
@@ -11468,25 +12733,29 @@ vcgtq_u64 (uint64x2_t __a, uint64x2_t __b)
 
 /* vcgt - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgts_f32 (float32_t __a, float32_t __b)
 {
   return __a > __b ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtd_s64 (int64_t __a, int64_t __b)
 {
   return __a > __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtd_u64 (uint64_t __a, uint64_t __b)
 {
   return __a > __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtd_f64 (float64_t __a, float64_t __b)
 {
   return __a > __b ? -1ll : 0ll;
@@ -11494,73 +12763,85 @@ vcgtd_f64 (float64_t __a, float64_t __b)
 
 /* vcgtz - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtz_f32 (float32x2_t __a)
 {
   return (uint32x2_t) (__a > 0.0f);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtz_f64 (float64x1_t __a)
 {
   return (uint64x1_t) (__a > (float64x1_t) {0.0});
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtz_s8 (int8x8_t __a)
 {
   return (uint8x8_t) (__a > 0);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtz_s16 (int16x4_t __a)
 {
   return (uint16x4_t) (__a > 0);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtz_s32 (int32x2_t __a)
 {
   return (uint32x2_t) (__a > 0);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtz_s64 (int64x1_t __a)
 {
   return (uint64x1_t) (__a > __AARCH64_INT64_C (0));
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzq_f32 (float32x4_t __a)
 {
   return (uint32x4_t) (__a > 0.0f);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzq_f64 (float64x2_t __a)
 {
     return (uint64x2_t) (__a > 0.0);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzq_s8 (int8x16_t __a)
 {
   return (uint8x16_t) (__a > 0);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzq_s16 (int16x8_t __a)
 {
   return (uint16x8_t) (__a > 0);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzq_s32 (int32x4_t __a)
 {
   return (uint32x4_t) (__a > 0);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzq_s64 (int64x2_t __a)
 {
   return (uint64x2_t) (__a > __AARCH64_INT64_C (0));
@@ -11568,19 +12849,22 @@ vcgtzq_s64 (int64x2_t __a)
 
 /* vcgtz - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzs_f32 (float32_t __a)
 {
   return __a > 0.0f ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzd_s64 (int64_t __a)
 {
   return __a > 0 ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzd_f64 (float64_t __a)
 {
   return __a > 0.0 ? -1ll : 0ll;
@@ -11588,121 +12872,141 @@ vcgtzd_f64 (float64_t __a)
 
 /* vcle - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_f32 (float32x2_t __a, float32x2_t __b)
 {
   return (uint32x2_t) (__a <= __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (uint64x1_t) (__a <= __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (uint8x8_t) (__a <= __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (uint16x4_t) (__a <= __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (uint32x2_t) (__a <= __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) (__a <= __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (__a <= __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (__a <= __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (__a <= __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return (__a <= __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return (uint32x4_t) (__a <= __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return (uint64x2_t) (__a <= __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (uint8x16_t) (__a <= __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (uint16x8_t) (__a <= __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (uint32x4_t) (__a <= __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (uint64x2_t) (__a <= __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (__a <= __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (__a <= __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (__a <= __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return (__a <= __b);
@@ -11710,25 +13014,29 @@ vcleq_u64 (uint64x2_t __a, uint64x2_t __b)
 
 /* vcle - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcles_f32 (float32_t __a, float32_t __b)
 {
   return __a <= __b ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcled_s64 (int64_t __a, int64_t __b)
 {
   return __a <= __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcled_u64 (uint64_t __a, uint64_t __b)
 {
   return __a <= __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcled_f64 (float64_t __a, float64_t __b)
 {
   return __a <= __b ? -1ll : 0ll;
@@ -11736,73 +13044,85 @@ vcled_f64 (float64_t __a, float64_t __b)
 
 /* vclez - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclez_f32 (float32x2_t __a)
 {
   return (uint32x2_t) (__a <= 0.0f);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclez_f64 (float64x1_t __a)
 {
   return (uint64x1_t) (__a <= (float64x1_t) {0.0});
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclez_s8 (int8x8_t __a)
 {
   return (uint8x8_t) (__a <= 0);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclez_s16 (int16x4_t __a)
 {
   return (uint16x4_t) (__a <= 0);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclez_s32 (int32x2_t __a)
 {
   return (uint32x2_t) (__a <= 0);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclez_s64 (int64x1_t __a)
 {
   return (uint64x1_t) (__a <= __AARCH64_INT64_C (0));
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezq_f32 (float32x4_t __a)
 {
   return (uint32x4_t) (__a <= 0.0f);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezq_f64 (float64x2_t __a)
 {
   return (uint64x2_t) (__a <= 0.0);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezq_s8 (int8x16_t __a)
 {
   return (uint8x16_t) (__a <= 0);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezq_s16 (int16x8_t __a)
 {
   return (uint16x8_t) (__a <= 0);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezq_s32 (int32x4_t __a)
 {
   return (uint32x4_t) (__a <= 0);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezq_s64 (int64x2_t __a)
 {
   return (uint64x2_t) (__a <= __AARCH64_INT64_C (0));
@@ -11810,19 +13130,22 @@ vclezq_s64 (int64x2_t __a)
 
 /* vclez - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezs_f32 (float32_t __a)
 {
   return __a <= 0.0f ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezd_s64 (int64_t __a)
 {
   return __a <= 0 ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezd_f64 (float64_t __a)
 {
   return __a <= 0.0 ? -1ll : 0ll;
@@ -11830,121 +13153,141 @@ vclezd_f64 (float64_t __a)
 
 /* vclt - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_f32 (float32x2_t __a, float32x2_t __b)
 {
   return (uint32x2_t) (__a < __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (uint64x1_t) (__a < __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (uint8x8_t) (__a < __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (uint16x4_t) (__a < __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (uint32x2_t) (__a < __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) (__a < __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (__a < __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (__a < __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (__a < __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return (__a < __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return (uint32x4_t) (__a < __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return (uint64x2_t) (__a < __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (uint8x16_t) (__a < __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (uint16x8_t) (__a < __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (uint32x4_t) (__a < __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (uint64x2_t) (__a < __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (__a < __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (__a < __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (__a < __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return (__a < __b);
@@ -11952,25 +13295,29 @@ vcltq_u64 (uint64x2_t __a, uint64x2_t __b)
 
 /* vclt - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclts_f32 (float32_t __a, float32_t __b)
 {
   return __a < __b ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltd_s64 (int64_t __a, int64_t __b)
 {
   return __a < __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltd_u64 (uint64_t __a, uint64_t __b)
 {
   return __a < __b ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltd_f64 (float64_t __a, float64_t __b)
 {
   return __a < __b ? -1ll : 0ll;
@@ -11978,73 +13325,85 @@ vcltd_f64 (float64_t __a, float64_t __b)
 
 /* vcltz - vector.  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltz_f32 (float32x2_t __a)
 {
   return (uint32x2_t) (__a < 0.0f);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltz_f64 (float64x1_t __a)
 {
   return (uint64x1_t) (__a < (float64x1_t) {0.0});
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltz_s8 (int8x8_t __a)
 {
   return (uint8x8_t) (__a < 0);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltz_s16 (int16x4_t __a)
 {
   return (uint16x4_t) (__a < 0);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltz_s32 (int32x2_t __a)
 {
   return (uint32x2_t) (__a < 0);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltz_s64 (int64x1_t __a)
 {
   return (uint64x1_t) (__a < __AARCH64_INT64_C (0));
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzq_f32 (float32x4_t __a)
 {
   return (uint32x4_t) (__a < 0.0f);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzq_f64 (float64x2_t __a)
 {
   return (uint64x2_t) (__a < 0.0);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzq_s8 (int8x16_t __a)
 {
   return (uint8x16_t) (__a < 0);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzq_s16 (int16x8_t __a)
 {
   return (uint16x8_t) (__a < 0);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzq_s32 (int32x4_t __a)
 {
   return (uint32x4_t) (__a < 0);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzq_s64 (int64x2_t __a)
 {
   return (uint64x2_t) (__a < __AARCH64_INT64_C (0));
@@ -12052,19 +13411,22 @@ vcltzq_s64 (int64x2_t __a)
 
 /* vcltz - scalar.  */
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzs_f32 (float32_t __a)
 {
   return __a < 0.0f ? -1 : 0;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzd_s64 (int64_t __a)
 {
   return __a < 0 ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzd_f64 (float64_t __a)
 {
   return __a < 0.0 ? -1ll : 0ll;
@@ -12072,37 +13434,43 @@ vcltzd_f64 (float64_t __a)
 
 /* vcls.  */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcls_s8 (int8x8_t __a)
 {
   return __builtin_aarch64_clrsbv8qi (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcls_s16 (int16x4_t __a)
 {
   return __builtin_aarch64_clrsbv4hi (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcls_s32 (int32x2_t __a)
 {
   return __builtin_aarch64_clrsbv2si (__a);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclsq_s8 (int8x16_t __a)
 {
   return __builtin_aarch64_clrsbv16qi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclsq_s16 (int16x8_t __a)
 {
   return __builtin_aarch64_clrsbv8hi (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclsq_s32 (int32x4_t __a)
 {
   return __builtin_aarch64_clrsbv4si (__a);
@@ -12110,73 +13478,85 @@ vclsq_s32 (int32x4_t __a)
 
 /* vclz.  */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclz_s8 (int8x8_t __a)
 {
   return __builtin_aarch64_clzv8qi (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclz_s16 (int16x4_t __a)
 {
   return __builtin_aarch64_clzv4hi (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclz_s32 (int32x2_t __a)
 {
   return __builtin_aarch64_clzv2si (__a);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclz_u8 (uint8x8_t __a)
 {
   return (uint8x8_t)__builtin_aarch64_clzv8qi ((int8x8_t)__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclz_u16 (uint16x4_t __a)
 {
   return (uint16x4_t)__builtin_aarch64_clzv4hi ((int16x4_t)__a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclz_u32 (uint32x2_t __a)
 {
   return (uint32x2_t)__builtin_aarch64_clzv2si ((int32x2_t)__a);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclzq_s8 (int8x16_t __a)
 {
   return __builtin_aarch64_clzv16qi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclzq_s16 (int16x8_t __a)
 {
   return __builtin_aarch64_clzv8hi (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclzq_s32 (int32x4_t __a)
 {
   return __builtin_aarch64_clzv4si (__a);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclzq_u8 (uint8x16_t __a)
 {
   return (uint8x16_t)__builtin_aarch64_clzv16qi ((int8x16_t)__a);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclzq_u16 (uint16x8_t __a)
 {
   return (uint16x8_t)__builtin_aarch64_clzv8hi ((int16x8_t)__a);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclzq_u32 (uint32x4_t __a)
 {
   return (uint32x4_t)__builtin_aarch64_clzv4si ((int32x4_t)__a);
@@ -12184,37 +13564,43 @@ vclzq_u32 (uint32x4_t __a)
 
 /* vcnt.  */
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcnt_p8 (poly8x8_t __a)
 {
   return (poly8x8_t) __builtin_aarch64_popcountv8qi ((int8x8_t) __a);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcnt_s8 (int8x8_t __a)
 {
   return __builtin_aarch64_popcountv8qi (__a);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcnt_u8 (uint8x8_t __a)
 {
   return (uint8x8_t) __builtin_aarch64_popcountv8qi ((int8x8_t) __a);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcntq_p8 (poly8x16_t __a)
 {
   return (poly8x16_t) __builtin_aarch64_popcountv16qi ((int8x16_t) __a);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcntq_s8 (int8x16_t __a)
 {
   return __builtin_aarch64_popcountv16qi (__a);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcntq_u8 (uint8x16_t __a)
 {
   return (uint8x16_t) __builtin_aarch64_popcountv16qi ((int8x16_t) __a);
@@ -12222,7 +13608,8 @@ vcntq_u8 (uint8x16_t __a)
 
 /* vcopy_lane.  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_f32 (float32x2_t __a, const int __lane1,
                float32x2_t __b, const int __lane2)
 {
@@ -12230,7 +13617,8 @@ vcopy_lane_f32 (float32x2_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_f64 (float64x1_t __a, const int __lane1,
                float64x1_t __b, const int __lane2)
 {
@@ -12238,7 +13626,8 @@ vcopy_lane_f64 (float64x1_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_p8 (poly8x8_t __a, const int __lane1,
               poly8x8_t __b, const int __lane2)
 {
@@ -12246,7 +13635,8 @@ vcopy_lane_p8 (poly8x8_t __a, const int __lane1,
                                 __a, __lane1);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_p16 (poly16x4_t __a, const int __lane1,
                poly16x4_t __b, const int __lane2)
 {
@@ -12254,7 +13644,8 @@ vcopy_lane_p16 (poly16x4_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_s8 (int8x8_t __a, const int __lane1,
               int8x8_t __b, const int __lane2)
 {
@@ -12262,7 +13653,8 @@ vcopy_lane_s8 (int8x8_t __a, const int __lane1,
                                 __a, __lane1);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_s16 (int16x4_t __a, const int __lane1,
                int16x4_t __b, const int __lane2)
 {
@@ -12270,7 +13662,8 @@ vcopy_lane_s16 (int16x4_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_s32 (int32x2_t __a, const int __lane1,
                int32x2_t __b, const int __lane2)
 {
@@ -12278,7 +13671,8 @@ vcopy_lane_s32 (int32x2_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_s64 (int64x1_t __a, const int __lane1,
                int64x1_t __b, const int __lane2)
 {
@@ -12286,7 +13680,8 @@ vcopy_lane_s64 (int64x1_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_u8 (uint8x8_t __a, const int __lane1,
               uint8x8_t __b, const int __lane2)
 {
@@ -12294,7 +13689,8 @@ vcopy_lane_u8 (uint8x8_t __a, const int __lane1,
                                 __a, __lane1);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_u16 (uint16x4_t __a, const int __lane1,
                uint16x4_t __b, const int __lane2)
 {
@@ -12302,7 +13698,8 @@ vcopy_lane_u16 (uint16x4_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_u32 (uint32x2_t __a, const int __lane1,
                uint32x2_t __b, const int __lane2)
 {
@@ -12310,7 +13707,8 @@ vcopy_lane_u32 (uint32x2_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_lane_u64 (uint64x1_t __a, const int __lane1,
                uint64x1_t __b, const int __lane2)
 {
@@ -12320,7 +13718,8 @@ vcopy_lane_u64 (uint64x1_t __a, const int __lane1,
 
 /* vcopy_laneq.  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_f32 (float32x2_t __a, const int __lane1,
                 float32x4_t __b, const int __lane2)
 {
@@ -12328,7 +13727,8 @@ vcopy_laneq_f32 (float32x2_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_f64 (float64x1_t __a, const int __lane1,
                 float64x2_t __b, const int __lane2)
 {
@@ -12336,7 +13736,8 @@ vcopy_laneq_f64 (float64x1_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_p8 (poly8x8_t __a, const int __lane1,
                poly8x16_t __b, const int __lane2)
 {
@@ -12344,7 +13745,8 @@ vcopy_laneq_p8 (poly8x8_t __a, const int __lane1,
                                 __a, __lane1);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_p16 (poly16x4_t __a, const int __lane1,
                 poly16x8_t __b, const int __lane2)
 {
@@ -12352,7 +13754,8 @@ vcopy_laneq_p16 (poly16x4_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_s8 (int8x8_t __a, const int __lane1,
                int8x16_t __b, const int __lane2)
 {
@@ -12360,7 +13763,8 @@ vcopy_laneq_s8 (int8x8_t __a, const int __lane1,
                                 __a, __lane1);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_s16 (int16x4_t __a, const int __lane1,
                 int16x8_t __b, const int __lane2)
 {
@@ -12368,7 +13772,8 @@ vcopy_laneq_s16 (int16x4_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_s32 (int32x2_t __a, const int __lane1,
                 int32x4_t __b, const int __lane2)
 {
@@ -12376,7 +13781,8 @@ vcopy_laneq_s32 (int32x2_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_s64 (int64x1_t __a, const int __lane1,
                 int64x2_t __b, const int __lane2)
 {
@@ -12384,7 +13790,8 @@ vcopy_laneq_s64 (int64x1_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_u8 (uint8x8_t __a, const int __lane1,
                uint8x16_t __b, const int __lane2)
 {
@@ -12392,7 +13799,8 @@ vcopy_laneq_u8 (uint8x8_t __a, const int __lane1,
                                 __a, __lane1);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_u16 (uint16x4_t __a, const int __lane1,
                 uint16x8_t __b, const int __lane2)
 {
@@ -12400,7 +13808,8 @@ vcopy_laneq_u16 (uint16x4_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_u32 (uint32x2_t __a, const int __lane1,
                 uint32x4_t __b, const int __lane2)
 {
@@ -12408,7 +13817,8 @@ vcopy_laneq_u32 (uint32x2_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopy_laneq_u64 (uint64x1_t __a, const int __lane1,
                 uint64x2_t __b, const int __lane2)
 {
@@ -12418,7 +13828,8 @@ vcopy_laneq_u64 (uint64x1_t __a, const int __lane1,
 
 /* vcopyq_lane.  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_f32 (float32x4_t __a, const int __lane1,
                 float32x2_t __b, const int __lane2)
 {
@@ -12426,7 +13837,8 @@ vcopyq_lane_f32 (float32x4_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_f64 (float64x2_t __a, const int __lane1,
                 float64x1_t __b, const int __lane2)
 {
@@ -12434,7 +13846,8 @@ vcopyq_lane_f64 (float64x2_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_p8 (poly8x16_t __a, const int __lane1,
                poly8x8_t __b, const int __lane2)
 {
@@ -12442,7 +13855,8 @@ vcopyq_lane_p8 (poly8x16_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_p16 (poly16x8_t __a, const int __lane1,
                 poly16x4_t __b, const int __lane2)
 {
@@ -12450,7 +13864,8 @@ vcopyq_lane_p16 (poly16x8_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_s8 (int8x16_t __a, const int __lane1,
                int8x8_t __b, const int __lane2)
 {
@@ -12458,7 +13873,8 @@ vcopyq_lane_s8 (int8x16_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_s16 (int16x8_t __a, const int __lane1,
                 int16x4_t __b, const int __lane2)
 {
@@ -12466,7 +13882,8 @@ vcopyq_lane_s16 (int16x8_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_s32 (int32x4_t __a, const int __lane1,
                 int32x2_t __b, const int __lane2)
 {
@@ -12474,7 +13891,8 @@ vcopyq_lane_s32 (int32x4_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_s64 (int64x2_t __a, const int __lane1,
                 int64x1_t __b, const int __lane2)
 {
@@ -12482,7 +13900,8 @@ vcopyq_lane_s64 (int64x2_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_u8 (uint8x16_t __a, const int __lane1,
                uint8x8_t __b, const int __lane2)
 {
@@ -12490,7 +13909,8 @@ vcopyq_lane_u8 (uint8x16_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_u16 (uint16x8_t __a, const int __lane1,
                 uint16x4_t __b, const int __lane2)
 {
@@ -12498,7 +13918,8 @@ vcopyq_lane_u16 (uint16x8_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_u32 (uint32x4_t __a, const int __lane1,
                 uint32x2_t __b, const int __lane2)
 {
@@ -12506,7 +13927,8 @@ vcopyq_lane_u32 (uint32x4_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_lane_u64 (uint64x2_t __a, const int __lane1,
                 uint64x1_t __b, const int __lane2)
 {
@@ -12516,7 +13938,8 @@ vcopyq_lane_u64 (uint64x2_t __a, const int __lane1,
 
 /* vcopyq_laneq.  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_f32 (float32x4_t __a, const int __lane1,
                  float32x4_t __b, const int __lane2)
 {
@@ -12524,7 +13947,8 @@ vcopyq_laneq_f32 (float32x4_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_f64 (float64x2_t __a, const int __lane1,
                  float64x2_t __b, const int __lane2)
 {
@@ -12532,7 +13956,8 @@ vcopyq_laneq_f64 (float64x2_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_p8 (poly8x16_t __a, const int __lane1,
                 poly8x16_t __b, const int __lane2)
 {
@@ -12540,7 +13965,8 @@ vcopyq_laneq_p8 (poly8x16_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_p16 (poly16x8_t __a, const int __lane1,
                  poly16x8_t __b, const int __lane2)
 {
@@ -12548,7 +13974,8 @@ vcopyq_laneq_p16 (poly16x8_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_s8 (int8x16_t __a, const int __lane1,
                 int8x16_t __b, const int __lane2)
 {
@@ -12556,7 +13983,8 @@ vcopyq_laneq_s8 (int8x16_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_s16 (int16x8_t __a, const int __lane1,
                  int16x8_t __b, const int __lane2)
 {
@@ -12564,7 +13992,8 @@ vcopyq_laneq_s16 (int16x8_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_s32 (int32x4_t __a, const int __lane1,
                  int32x4_t __b, const int __lane2)
 {
@@ -12572,7 +14001,8 @@ vcopyq_laneq_s32 (int32x4_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_s64 (int64x2_t __a, const int __lane1,
                  int64x2_t __b, const int __lane2)
 {
@@ -12580,7 +14010,8 @@ vcopyq_laneq_s64 (int64x2_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_u8 (uint8x16_t __a, const int __lane1,
                 uint8x16_t __b, const int __lane2)
 {
@@ -12588,7 +14019,8 @@ vcopyq_laneq_u8 (uint8x16_t __a, const int __lane1,
                                  __a, __lane1);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_u16 (uint16x8_t __a, const int __lane1,
                  uint16x8_t __b, const int __lane2)
 {
@@ -12596,7 +14028,8 @@ vcopyq_laneq_u16 (uint16x8_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_u32 (uint32x4_t __a, const int __lane1,
                  uint32x4_t __b, const int __lane2)
 {
@@ -12604,7 +14037,8 @@ vcopyq_laneq_u32 (uint32x4_t __a, const int __lane1,
                                   __a, __lane1);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcopyq_laneq_u64 (uint64x2_t __a, const int __lane1,
                  uint64x2_t __b, const int __lane2)
 {
@@ -12614,25 +14048,29 @@ vcopyq_laneq_u64 (uint64x2_t __a, const int __lane1,
 
 /* vcvt (double -> float).  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f16_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_float_truncate_lo_v4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_high_f16_f32 (float16x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_float_truncate_hi_v8hf (__a, __b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f32_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_float_truncate_lo_v2sf (__a);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_high_f32_f64 (float32x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_float_truncate_hi_v4sf (__a, __b);
@@ -12640,26 +14078,30 @@ vcvt_high_f32_f64 (float32x2_t __a, float64x2_t __b)
 
 /* vcvt (float -> double).  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f32_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_float_extend_lo_v4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f64_f32 (float32x2_t __a)
 {
 
   return __builtin_aarch64_float_extend_lo_v2df (__a);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_high_f32_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_vec_unpacks_hi_v8hf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_high_f64_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_vec_unpacks_hi_v4sf (__a);
@@ -12667,75 +14109,87 @@ vcvt_high_f64_f32 (float32x4_t __a)
 
 /* vcvt (<u>fixed-point -> float).  */
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtd_n_f64_s64 (int64_t __a, const int __b)
 {
   return __builtin_aarch64_scvtfdi (__a, __b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtd_n_f64_u64 (uint64_t __a, const int __b)
 {
   return __builtin_aarch64_ucvtfdi_sus (__a, __b);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvts_n_f32_s32 (int32_t __a, const int __b)
 {
   return __builtin_aarch64_scvtfsi (__a, __b);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvts_n_f32_u32 (uint32_t __a, const int __b)
 {
   return __builtin_aarch64_ucvtfsi_sus (__a, __b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_f32_s32 (int32x2_t __a, const int __b)
 {
   return __builtin_aarch64_scvtfv2si (__a, __b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_f32_u32 (uint32x2_t __a, const int __b)
 {
   return __builtin_aarch64_ucvtfv2si_sus (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_f64_s64 (int64x1_t __a, const int __b)
 {
   return (float64x1_t)
     { __builtin_aarch64_scvtfdi (vget_lane_s64 (__a, 0), __b) };
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_f64_u64 (uint64x1_t __a, const int __b)
 {
   return (float64x1_t)
     { __builtin_aarch64_ucvtfdi_sus (vget_lane_u64 (__a, 0), __b) };
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_f32_s32 (int32x4_t __a, const int __b)
 {
   return __builtin_aarch64_scvtfv4si (__a, __b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_f32_u32 (uint32x4_t __a, const int __b)
 {
   return __builtin_aarch64_ucvtfv4si_sus (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_f64_s64 (int64x2_t __a, const int __b)
 {
   return __builtin_aarch64_scvtfv2di (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_f64_u64 (uint64x2_t __a, const int __b)
 {
   return __builtin_aarch64_ucvtfv2di_sus (__a, __b);
@@ -12743,75 +14197,87 @@ vcvtq_n_f64_u64 (uint64x2_t __a, const int __b)
 
 /* vcvt (float -> <u>fixed-point).  */
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtd_n_s64_f64 (float64_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzsdf (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtd_n_u64_f64 (float64_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzudf_uss (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvts_n_s32_f32 (float32_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzssf (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvts_n_u32_f32 (float32_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzusf_uss (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_s32_f32 (float32x2_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzsv2sf (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_u32_f32 (float32x2_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzuv2sf_uss (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_s64_f64 (float64x1_t __a, const int __b)
 {
   return (int64x1_t)
     { __builtin_aarch64_fcvtzsdf (vget_lane_f64 (__a, 0), __b) };
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_u64_f64 (float64x1_t __a, const int __b)
 {
   return (uint64x1_t)
     { __builtin_aarch64_fcvtzudf_uss (vget_lane_f64 (__a, 0), __b) };
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_s32_f32 (float32x4_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzsv4sf (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_u32_f32 (float32x4_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzuv4sf_uss (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_s64_f64 (float64x2_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzsv2df (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_u64_f64 (float64x2_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzuv2df_uss (__a, __b);
@@ -12819,73 +14285,85 @@ vcvtq_n_u64_f64 (float64x2_t __a, const int __b)
 
 /* vcvt  (<u>int -> float)  */
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtd_f64_s64 (int64_t __a)
 {
   return (float64_t) __a;
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtd_f64_u64 (uint64_t __a)
 {
   return (float64_t) __a;
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvts_f32_s32 (int32_t __a)
 {
   return (float32_t) __a;
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvts_f32_u32 (uint32_t __a)
 {
   return (float32_t) __a;
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f32_s32 (int32x2_t __a)
 {
   return __builtin_aarch64_floatv2siv2sf (__a);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f32_u32 (uint32x2_t __a)
 {
   return __builtin_aarch64_floatunsv2siv2sf ((int32x2_t) __a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f64_s64 (int64x1_t __a)
 {
   return (float64x1_t) { vget_lane_s64 (__a, 0) };
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f64_u64 (uint64x1_t __a)
 {
   return (float64x1_t) { vget_lane_u64 (__a, 0) };
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_f32_s32 (int32x4_t __a)
 {
   return __builtin_aarch64_floatv4siv4sf (__a);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_f32_u32 (uint32x4_t __a)
 {
   return __builtin_aarch64_floatunsv4siv4sf ((int32x4_t) __a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_f64_s64 (int64x2_t __a)
 {
   return __builtin_aarch64_floatv2div2df (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_f64_u64 (uint64x2_t __a)
 {
   return __builtin_aarch64_floatunsv2div2df ((int64x2_t) __a);
@@ -12893,73 +14371,85 @@ vcvtq_f64_u64 (uint64x2_t __a)
 
 /* vcvt (float -> <u>int)  */
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtd_s64_f64 (float64_t __a)
 {
   return (int64_t) __a;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtd_u64_f64 (float64_t __a)
 {
   return (uint64_t) __a;
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvts_s32_f32 (float32_t __a)
 {
   return (int32_t) __a;
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvts_u32_f32 (float32_t __a)
 {
   return (uint32_t) __a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_s32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lbtruncv2sfv2si (__a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_u32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lbtruncuv2sfv2si_us (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_s32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lbtruncv4sfv4si (__a);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_u32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lbtruncuv4sfv4si_us (__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_s64_f64 (float64x1_t __a)
 {
   return (int64x1_t) {vcvtd_s64_f64 (__a[0])};
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_u64_f64 (float64x1_t __a)
 {
   return (uint64x1_t) {vcvtd_u64_f64 (__a[0])};
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_s64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lbtruncv2dfv2di (__a);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_u64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lbtruncuv2dfv2di_us (__a);
@@ -12967,73 +14457,85 @@ vcvtq_u64_f64 (float64x2_t __a)
 
 /* vcvta  */
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtad_s64_f64 (float64_t __a)
 {
   return __builtin_aarch64_lrounddfdi (__a);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtad_u64_f64 (float64_t __a)
 {
   return __builtin_aarch64_lroundudfdi_us (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtas_s32_f32 (float32_t __a)
 {
   return __builtin_aarch64_lroundsfsi (__a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtas_u32_f32 (float32_t __a)
 {
   return __builtin_aarch64_lroundusfsi_us (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvta_s32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lroundv2sfv2si (__a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvta_u32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lrounduv2sfv2si_us (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtaq_s32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lroundv4sfv4si (__a);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtaq_u32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lrounduv4sfv4si_us (__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvta_s64_f64 (float64x1_t __a)
 {
   return (int64x1_t) {vcvtad_s64_f64 (__a[0])};
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvta_u64_f64 (float64x1_t __a)
 {
   return (uint64x1_t) {vcvtad_u64_f64 (__a[0])};
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtaq_s64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lroundv2dfv2di (__a);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtaq_u64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lrounduv2dfv2di_us (__a);
@@ -13041,73 +14543,85 @@ vcvtaq_u64_f64 (float64x2_t __a)
 
 /* vcvtm  */
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtmd_s64_f64 (float64_t __a)
 {
   return __builtin_llfloor (__a);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtmd_u64_f64 (float64_t __a)
 {
   return __builtin_aarch64_lfloorudfdi_us (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtms_s32_f32 (float32_t __a)
 {
   return __builtin_ifloorf (__a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtms_u32_f32 (float32_t __a)
 {
   return __builtin_aarch64_lfloorusfsi_us (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtm_s32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lfloorv2sfv2si (__a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtm_u32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lflooruv2sfv2si_us (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtmq_s32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lfloorv4sfv4si (__a);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtmq_u32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lflooruv4sfv4si_us (__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtm_s64_f64 (float64x1_t __a)
 {
   return (int64x1_t) {vcvtmd_s64_f64 (__a[0])};
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtm_u64_f64 (float64x1_t __a)
 {
   return (uint64x1_t) {vcvtmd_u64_f64 (__a[0])};
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtmq_s64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lfloorv2dfv2di (__a);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtmq_u64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lflooruv2dfv2di_us (__a);
@@ -13115,73 +14629,85 @@ vcvtmq_u64_f64 (float64x2_t __a)
 
 /* vcvtn  */
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtnd_s64_f64 (float64_t __a)
 {
   return __builtin_aarch64_lfrintndfdi (__a);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtnd_u64_f64 (float64_t __a)
 {
   return __builtin_aarch64_lfrintnudfdi_us (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtns_s32_f32 (float32_t __a)
 {
   return __builtin_aarch64_lfrintnsfsi (__a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtns_u32_f32 (float32_t __a)
 {
   return __builtin_aarch64_lfrintnusfsi_us (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtn_s32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lfrintnv2sfv2si (__a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtn_u32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lfrintnuv2sfv2si_us (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtnq_s32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lfrintnv4sfv4si (__a);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtnq_u32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lfrintnuv4sfv4si_us (__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtn_s64_f64 (float64x1_t __a)
 {
   return (int64x1_t) {vcvtnd_s64_f64 (__a[0])};
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtn_u64_f64 (float64x1_t __a)
 {
   return (uint64x1_t) {vcvtnd_u64_f64 (__a[0])};
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtnq_s64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lfrintnv2dfv2di (__a);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtnq_u64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lfrintnuv2dfv2di_us (__a);
@@ -13189,73 +14715,85 @@ vcvtnq_u64_f64 (float64x2_t __a)
 
 /* vcvtp  */
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtpd_s64_f64 (float64_t __a)
 {
   return __builtin_llceil (__a);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtpd_u64_f64 (float64_t __a)
 {
   return __builtin_aarch64_lceiludfdi_us (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtps_s32_f32 (float32_t __a)
 {
   return __builtin_iceilf (__a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtps_u32_f32 (float32_t __a)
 {
   return __builtin_aarch64_lceilusfsi_us (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtp_s32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lceilv2sfv2si (__a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtp_u32_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_lceiluv2sfv2si_us (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtpq_s32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lceilv4sfv4si (__a);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtpq_u32_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_lceiluv4sfv4si_us (__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtp_s64_f64 (float64x1_t __a)
 {
   return (int64x1_t) {vcvtpd_s64_f64 (__a[0])};
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtp_u64_f64 (float64x1_t __a)
 {
   return (uint64x1_t) {vcvtpd_u64_f64 (__a[0])};
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtpq_s64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lceilv2dfv2di (__a);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtpq_u64_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_lceiluv2dfv2di_us (__a);
@@ -13263,79 +14801,92 @@ vcvtpq_u64_f64 (float64x2_t __a)
 
 /* vdup_n  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_f16 (float16_t __a)
 {
   return (float16x4_t) {__a, __a, __a, __a};
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_f32 (float32_t __a)
 {
   return (float32x2_t) {__a, __a};
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_f64 (float64_t __a)
 {
   return (float64x1_t) {__a};
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_p8 (poly8_t __a)
 {
   return (poly8x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_p16 (poly16_t __a)
 {
   return (poly16x4_t) {__a, __a, __a, __a};
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_s8 (int8_t __a)
 {
   return (int8x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_s16 (int16_t __a)
 {
   return (int16x4_t) {__a, __a, __a, __a};
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_s32 (int32_t __a)
 {
   return (int32x2_t) {__a, __a};
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_s64 (int64_t __a)
 {
   return (int64x1_t) {__a};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_u8 (uint8_t __a)
 {
   return (uint8x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_u16 (uint16_t __a)
 {
   return (uint16x4_t) {__a, __a, __a, __a};
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_u32 (uint32_t __a)
 {
   return (uint32x2_t) {__a, __a};
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_n_u64 (uint64_t __a)
 {
   return (uint64x1_t) {__a};
@@ -13343,82 +14894,95 @@ vdup_n_u64 (uint64_t __a)
 
 /* vdupq_n  */
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_f16 (float16_t __a)
 {
   return (float16x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_f32 (float32_t __a)
 {
   return (float32x4_t) {__a, __a, __a, __a};
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_f64 (float64_t __a)
 {
   return (float64x2_t) {__a, __a};
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_p8 (uint32_t __a)
 {
   return (poly8x16_t) {__a, __a, __a, __a, __a, __a, __a, __a,
                       __a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_p16 (uint32_t __a)
 {
   return (poly16x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_s8 (int32_t __a)
 {
   return (int8x16_t) {__a, __a, __a, __a, __a, __a, __a, __a,
                      __a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_s16 (int32_t __a)
 {
   return (int16x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_s32 (int32_t __a)
 {
   return (int32x4_t) {__a, __a, __a, __a};
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_s64 (int64_t __a)
 {
   return (int64x2_t) {__a, __a};
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_u8 (uint32_t __a)
 {
   return (uint8x16_t) {__a, __a, __a, __a, __a, __a, __a, __a,
                       __a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_u16 (uint32_t __a)
 {
   return (uint16x8_t) {__a, __a, __a, __a, __a, __a, __a, __a};
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_u32 (uint32_t __a)
 {
   return (uint32x4_t) {__a, __a, __a, __a};
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_n_u64 (uint64_t __a)
 {
   return (uint64x2_t) {__a, __a};
@@ -13426,79 +14990,92 @@ vdupq_n_u64 (uint64_t __a)
 
 /* vdup_lane  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_f16 (float16x4_t __a, const int __b)
 {
   return __aarch64_vdup_lane_f16 (__a, __b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_f32 (float32x2_t __a, const int __b)
 {
   return __aarch64_vdup_lane_f32 (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_f64 (float64x1_t __a, const int __b)
 {
   return __aarch64_vdup_lane_f64 (__a, __b);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_p8 (poly8x8_t __a, const int __b)
 {
   return __aarch64_vdup_lane_p8 (__a, __b);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_p16 (poly16x4_t __a, const int __b)
 {
   return __aarch64_vdup_lane_p16 (__a, __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_s8 (int8x8_t __a, const int __b)
 {
   return __aarch64_vdup_lane_s8 (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_s16 (int16x4_t __a, const int __b)
 {
   return __aarch64_vdup_lane_s16 (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_s32 (int32x2_t __a, const int __b)
 {
   return __aarch64_vdup_lane_s32 (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_s64 (int64x1_t __a, const int __b)
 {
   return __aarch64_vdup_lane_s64 (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_u8 (uint8x8_t __a, const int __b)
 {
   return __aarch64_vdup_lane_u8 (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_u16 (uint16x4_t __a, const int __b)
 {
   return __aarch64_vdup_lane_u16 (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_u32 (uint32x2_t __a, const int __b)
 {
   return __aarch64_vdup_lane_u32 (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_lane_u64 (uint64x1_t __a, const int __b)
 {
   return __aarch64_vdup_lane_u64 (__a, __b);
@@ -13506,79 +15083,92 @@ vdup_lane_u64 (uint64x1_t __a, const int __b)
 
 /* vdup_laneq  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_f16 (float16x8_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_f16 (__a, __b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_f32 (float32x4_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_f32 (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_f64 (float64x2_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_f64 (__a, __b);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_p8 (poly8x16_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_p8 (__a, __b);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_p16 (poly16x8_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_p16 (__a, __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_s8 (int8x16_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_s8 (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_s16 (int16x8_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_s16 (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_s32 (int32x4_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_s32 (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_s64 (int64x2_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_s64 (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_u8 (uint8x16_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_u8 (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_u16 (uint16x8_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_u16 (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_u32 (uint32x4_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_u32 (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdup_laneq_u64 (uint64x2_t __a, const int __b)
 {
   return __aarch64_vdup_laneq_u64 (__a, __b);
@@ -13586,79 +15176,92 @@ vdup_laneq_u64 (uint64x2_t __a, const int __b)
 
 /* vdupq_lane  */
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_f16 (float16x4_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_f16 (__a, __b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_f32 (float32x2_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_f32 (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_f64 (float64x1_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_f64 (__a, __b);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_p8 (poly8x8_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_p8 (__a, __b);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_p16 (poly16x4_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_p16 (__a, __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_s8 (int8x8_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_s8 (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_s16 (int16x4_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_s16 (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_s32 (int32x2_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_s32 (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_s64 (int64x1_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_s64 (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_u8 (uint8x8_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_u8 (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_u16 (uint16x4_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_u16 (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_u32 (uint32x2_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_u32 (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_lane_u64 (uint64x1_t __a, const int __b)
 {
   return __aarch64_vdupq_lane_u64 (__a, __b);
@@ -13666,98 +15269,114 @@ vdupq_lane_u64 (uint64x1_t __a, const int __b)
 
 /* vdupq_laneq  */
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_f16 (float16x8_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_f16 (__a, __b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_f32 (float32x4_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_f32 (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_f64 (float64x2_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_f64 (__a, __b);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_p8 (poly8x16_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_p8 (__a, __b);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_p16 (poly16x8_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_p16 (__a, __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_s8 (int8x16_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_s8 (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_s16 (int16x8_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_s16 (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_s32 (int32x4_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_s32 (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_s64 (int64x2_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_s64 (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_u8 (uint8x16_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_u8 (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_u16 (uint16x8_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_u16 (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_u32 (uint32x4_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_u32 (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupq_laneq_u64 (uint64x2_t __a, const int __b)
 {
   return __aarch64_vdupq_laneq_u64 (__a, __b);
 }
 
 /* vdupb_lane  */
-__extension__ static __inline poly8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupb_lane_p8 (poly8x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupb_lane_s8 (int8x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupb_lane_u8 (uint8x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
@@ -13765,25 +15384,29 @@ vdupb_lane_u8 (uint8x8_t __a, const int __b)
 
 /* vduph_lane  */
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vduph_lane_f16 (float16x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline poly16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vduph_lane_p16 (poly16x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vduph_lane_s16 (int16x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vduph_lane_u16 (uint16x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
@@ -13791,40 +15414,46 @@ vduph_lane_u16 (uint16x4_t __a, const int __b)
 
 /* vdups_lane  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdups_lane_f32 (float32x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdups_lane_s32 (int32x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdups_lane_u32 (uint32x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
 /* vdupd_lane  */
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupd_lane_f64 (float64x1_t __a, const int __b)
 {
   __AARCH64_LANE_CHECK (__a, __b);
   return __a[0];
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupd_lane_s64 (int64x1_t __a, const int __b)
 {
   __AARCH64_LANE_CHECK (__a, __b);
   return __a[0];
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupd_lane_u64 (uint64x1_t __a, const int __b)
 {
   __AARCH64_LANE_CHECK (__a, __b);
@@ -13832,19 +15461,22 @@ vdupd_lane_u64 (uint64x1_t __a, const int __b)
 }
 
 /* vdupb_laneq  */
-__extension__ static __inline poly8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupb_laneq_p8 (poly8x16_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupb_laneq_s8 (int8x16_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupb_laneq_u8 (uint8x16_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
@@ -13852,25 +15484,29 @@ vdupb_laneq_u8 (uint8x16_t __a, const int __b)
 
 /* vduph_laneq  */
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vduph_laneq_f16 (float16x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline poly16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vduph_laneq_p16 (poly16x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vduph_laneq_s16 (int16x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vduph_laneq_u16 (uint16x8_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
@@ -13878,38 +15514,44 @@ vduph_laneq_u16 (uint16x8_t __a, const int __b)
 
 /* vdups_laneq  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdups_laneq_f32 (float32x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdups_laneq_s32 (int32x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdups_laneq_u32 (uint32x4_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
 /* vdupd_laneq  */
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupd_laneq_f64 (float64x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupd_laneq_s64 (int64x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdupd_laneq_u64 (uint64x2_t __a, const int __b)
 {
   return __aarch64_vget_lane_any (__a, __b);
@@ -13917,7 +15559,8 @@ vdupd_laneq_u64 (uint64x2_t __a, const int __b)
 
 /* vext  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_f16 (float16x4_t __a, float16x4_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -13930,7 +15573,8 @@ vext_f16 (float16x4_t __a, float16x4_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_f32 (float32x2_t __a, float32x2_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -13941,14 +15585,16 @@ vext_f32 (float32x2_t __a, float32x2_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_f64 (float64x1_t __a, float64x1_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
   /* The only possible index to the assembler instruction returns element 0.  */
   return __a;
 }
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_p8 (poly8x8_t __a, poly8x8_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -13961,7 +15607,8 @@ vext_p8 (poly8x8_t __a, poly8x8_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_p16 (poly16x4_t __a, poly16x4_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -13973,7 +15620,8 @@ vext_p16 (poly16x4_t __a, poly16x4_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_s8 (int8x8_t __a, int8x8_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -13986,7 +15634,8 @@ vext_s8 (int8x8_t __a, int8x8_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_s16 (int16x4_t __a, int16x4_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -13998,7 +15647,8 @@ vext_s16 (int16x4_t __a, int16x4_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_s32 (int32x2_t __a, int32x2_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14009,7 +15659,8 @@ vext_s32 (int32x2_t __a, int32x2_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_s64 (int64x1_t __a, int64x1_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14017,7 +15668,8 @@ vext_s64 (int64x1_t __a, int64x1_t __b, __const int __c)
   return __a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_u8 (uint8x8_t __a, uint8x8_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14030,7 +15682,8 @@ vext_u8 (uint8x8_t __a, uint8x8_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_u16 (uint16x4_t __a, uint16x4_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14042,7 +15695,8 @@ vext_u16 (uint16x4_t __a, uint16x4_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_u32 (uint32x2_t __a, uint32x2_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14053,7 +15707,8 @@ vext_u32 (uint32x2_t __a, uint32x2_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vext_u64 (uint64x1_t __a, uint64x1_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14061,7 +15716,8 @@ vext_u64 (uint64x1_t __a, uint64x1_t __b, __const int __c)
   return __a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_f16 (float16x8_t __a, float16x8_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14077,7 +15733,8 @@ vextq_f16 (float16x8_t __a, float16x8_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_f32 (float32x4_t __a, float32x4_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14089,7 +15746,8 @@ vextq_f32 (float32x4_t __a, float32x4_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_f64 (float64x2_t __a, float64x2_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14100,7 +15758,8 @@ vextq_f64 (float64x2_t __a, float64x2_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_p8 (poly8x16_t __a, poly8x16_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14115,7 +15774,8 @@ vextq_p8 (poly8x16_t __a, poly8x16_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_p16 (poly16x8_t __a, poly16x8_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14128,7 +15788,8 @@ vextq_p16 (poly16x8_t __a, poly16x8_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_s8 (int8x16_t __a, int8x16_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14143,7 +15804,8 @@ vextq_s8 (int8x16_t __a, int8x16_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_s16 (int16x8_t __a, int16x8_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14156,7 +15818,8 @@ vextq_s16 (int16x8_t __a, int16x8_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_s32 (int32x4_t __a, int32x4_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14168,7 +15831,8 @@ vextq_s32 (int32x4_t __a, int32x4_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_s64 (int64x2_t __a, int64x2_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14179,7 +15843,8 @@ vextq_s64 (int64x2_t __a, int64x2_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_u8 (uint8x16_t __a, uint8x16_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14194,7 +15859,8 @@ vextq_u8 (uint8x16_t __a, uint8x16_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_u16 (uint16x8_t __a, uint16x8_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14207,7 +15873,8 @@ vextq_u16 (uint16x8_t __a, uint16x8_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_u32 (uint32x4_t __a, uint32x4_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14219,7 +15886,8 @@ vextq_u32 (uint32x4_t __a, uint32x4_t __b, __const int __c)
 #endif
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vextq_u64 (uint64x2_t __a, uint64x2_t __b, __const int __c)
 {
   __AARCH64_LANE_CHECK (__a, __c);
@@ -14232,49 +15900,57 @@ vextq_u64 (uint64x2_t __a, uint64x2_t __b, __const int __c)
 
 /* vfma  */
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_f64 (float64x1_t __a, float64x1_t __b, float64x1_t __c)
 {
   return (float64x1_t) {__builtin_fma (__b[0], __c[0], __a[0])};
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_f32 (float32x2_t __a, float32x2_t __b, float32x2_t __c)
 {
   return __builtin_aarch64_fmav2sf (__b, __c, __a);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_f32 (float32x4_t __a, float32x4_t __b, float32x4_t __c)
 {
   return __builtin_aarch64_fmav4sf (__b, __c, __a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_f64 (float64x2_t __a, float64x2_t __b, float64x2_t __c)
 {
   return __builtin_aarch64_fmav2df (__b, __c, __a);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_n_f32 (float32x2_t __a, float32x2_t __b, float32_t __c)
 {
   return __builtin_aarch64_fmav2sf (__b, vdup_n_f32 (__c), __a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_n_f64 (float64x1_t __a, float64x1_t __b, float64_t __c)
 {
   return (float64x1_t) {__b[0] * __c + __a[0]};
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_n_f32 (float32x4_t __a, float32x4_t __b, float32_t __c)
 {
   return __builtin_aarch64_fmav4sf (__b, vdupq_n_f32 (__c), __a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_n_f64 (float64x2_t __a, float64x2_t __b, float64_t __c)
 {
   return __builtin_aarch64_fmav2df (__b, vdupq_n_f64 (__c), __a);
@@ -14282,7 +15958,8 @@ vfmaq_n_f64 (float64x2_t __a, float64x2_t __b, float64_t __c)
 
 /* vfma_lane  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_lane_f32 (float32x2_t __a, float32x2_t __b,
               float32x2_t __c, const int __lane)
 {
@@ -14291,21 +15968,24 @@ vfma_lane_f32 (float32x2_t __a, float32x2_t __b,
                                    __a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_lane_f64 (float64x1_t __a, float64x1_t __b,
               float64x1_t __c, const int __lane)
 {
   return (float64x1_t) {__builtin_fma (__b[0], __c[0], __a[0])};
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmad_lane_f64 (float64_t __a, float64_t __b,
                float64x1_t __c, const int __lane)
 {
   return __builtin_fma (__b, __c[0], __a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmas_lane_f32 (float32_t __a, float32_t __b,
                float32x2_t __c, const int __lane)
 {
@@ -14314,7 +15994,8 @@ vfmas_lane_f32 (float32_t __a, float32_t __b,
 
 /* vfma_laneq  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_laneq_f32 (float32x2_t __a, float32x2_t __b,
                float32x4_t __c, const int __lane)
 {
@@ -14323,7 +16004,8 @@ vfma_laneq_f32 (float32x2_t __a, float32x2_t __b,
                                    __a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_laneq_f64 (float64x1_t __a, float64x1_t __b,
                float64x2_t __c, const int __lane)
 {
@@ -14331,14 +16013,16 @@ vfma_laneq_f64 (float64x1_t __a, float64x1_t __b,
   return (float64x1_t) {__builtin_fma (__b[0], __c0, __a[0])};
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmad_laneq_f64 (float64_t __a, float64_t __b,
                 float64x2_t __c, const int __lane)
 {
   return __builtin_fma (__b, __aarch64_vget_lane_any (__c, __lane), __a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmas_laneq_f32 (float32_t __a, float32_t __b,
                 float32x4_t __c, const int __lane)
 {
@@ -14347,7 +16031,8 @@ vfmas_laneq_f32 (float32_t __a, float32_t __b,
 
 /* vfmaq_lane  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_lane_f32 (float32x4_t __a, float32x4_t __b,
                float32x2_t __c, const int __lane)
 {
@@ -14356,7 +16041,8 @@ vfmaq_lane_f32 (float32x4_t __a, float32x4_t __b,
                                    __a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_lane_f64 (float64x2_t __a, float64x2_t __b,
                float64x1_t __c, const int __lane)
 {
@@ -14365,7 +16051,8 @@ vfmaq_lane_f64 (float64x2_t __a, float64x2_t __b,
 
 /* vfmaq_laneq  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_laneq_f32 (float32x4_t __a, float32x4_t __b,
                 float32x4_t __c, const int __lane)
 {
@@ -14374,7 +16061,8 @@ vfmaq_laneq_f32 (float32x4_t __a, float32x4_t __b,
                                    __a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_laneq_f64 (float64x2_t __a, float64x2_t __b,
                 float64x2_t __c, const int __lane)
 {
@@ -14385,49 +16073,57 @@ vfmaq_laneq_f64 (float64x2_t __a, float64x2_t __b,
 
 /* vfms  */
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_f64 (float64x1_t __a, float64x1_t __b, float64x1_t __c)
 {
   return (float64x1_t) {__builtin_fma (-__b[0], __c[0], __a[0])};
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_f32 (float32x2_t __a, float32x2_t __b, float32x2_t __c)
 {
   return __builtin_aarch64_fmav2sf (-__b, __c, __a);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_f32 (float32x4_t __a, float32x4_t __b, float32x4_t __c)
 {
   return __builtin_aarch64_fmav4sf (-__b, __c, __a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_f64 (float64x2_t __a, float64x2_t __b, float64x2_t __c)
 {
   return __builtin_aarch64_fmav2df (-__b, __c, __a);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_n_f32 (float32x2_t __a, float32x2_t __b, float32_t __c)
 {
   return __builtin_aarch64_fmav2sf (-__b, vdup_n_f32 (__c), __a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_n_f64 (float64x1_t __a, float64x1_t __b, float64_t __c)
 {
   return (float64x1_t) {-__b[0] * __c + __a[0]};
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_n_f32 (float32x4_t __a, float32x4_t __b, float32_t __c)
 {
   return __builtin_aarch64_fmav4sf (-__b, vdupq_n_f32 (__c), __a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_n_f64 (float64x2_t __a, float64x2_t __b, float64_t __c)
 {
   return __builtin_aarch64_fmav2df (-__b, vdupq_n_f64 (__c), __a);
@@ -14435,7 +16131,8 @@ vfmsq_n_f64 (float64x2_t __a, float64x2_t __b, float64_t __c)
 
 /* vfms_lane  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_lane_f32 (float32x2_t __a, float32x2_t __b,
               float32x2_t __c, const int __lane)
 {
@@ -14444,21 +16141,24 @@ vfms_lane_f32 (float32x2_t __a, float32x2_t __b,
                                    __a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_lane_f64 (float64x1_t __a, float64x1_t __b,
               float64x1_t __c, const int __lane)
 {
   return (float64x1_t) {__builtin_fma (-__b[0], __c[0], __a[0])};
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsd_lane_f64 (float64_t __a, float64_t __b,
                float64x1_t __c, const int __lane)
 {
   return __builtin_fma (-__b, __c[0], __a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmss_lane_f32 (float32_t __a, float32_t __b,
                float32x2_t __c, const int __lane)
 {
@@ -14467,7 +16167,8 @@ vfmss_lane_f32 (float32_t __a, float32_t __b,
 
 /* vfms_laneq  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_laneq_f32 (float32x2_t __a, float32x2_t __b,
                float32x4_t __c, const int __lane)
 {
@@ -14476,7 +16177,8 @@ vfms_laneq_f32 (float32x2_t __a, float32x2_t __b,
                                    __a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_laneq_f64 (float64x1_t __a, float64x1_t __b,
                float64x2_t __c, const int __lane)
 {
@@ -14484,14 +16186,16 @@ vfms_laneq_f64 (float64x1_t __a, float64x1_t __b,
   return (float64x1_t) {__builtin_fma (-__b[0], __c0, __a[0])};
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsd_laneq_f64 (float64_t __a, float64_t __b,
                 float64x2_t __c, const int __lane)
 {
   return __builtin_fma (-__b, __aarch64_vget_lane_any (__c, __lane), __a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmss_laneq_f32 (float32_t __a, float32_t __b,
                 float32x4_t __c, const int __lane)
 {
@@ -14500,7 +16204,8 @@ vfmss_laneq_f32 (float32_t __a, float32_t __b,
 
 /* vfmsq_lane  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_lane_f32 (float32x4_t __a, float32x4_t __b,
                float32x2_t __c, const int __lane)
 {
@@ -14509,7 +16214,8 @@ vfmsq_lane_f32 (float32x4_t __a, float32x4_t __b,
                                    __a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_lane_f64 (float64x2_t __a, float64x2_t __b,
                float64x1_t __c, const int __lane)
 {
@@ -14518,7 +16224,8 @@ vfmsq_lane_f64 (float64x2_t __a, float64x2_t __b,
 
 /* vfmsq_laneq  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_laneq_f32 (float32x4_t __a, float32x4_t __b,
                 float32x4_t __c, const int __lane)
 {
@@ -14527,7 +16234,8 @@ vfmsq_laneq_f32 (float32x4_t __a, float32x4_t __b,
                                    __a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_laneq_f64 (float64x2_t __a, float64x2_t __b,
                 float64x2_t __c, const int __lane)
 {
@@ -14538,84 +16246,97 @@ vfmsq_laneq_f64 (float64x2_t __a, float64x2_t __b,
 
 /* vld1 */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_f16 (const float16_t *__a)
 {
   return __builtin_aarch64_ld1v4hf (__a);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_f32 (const float32_t *a)
 {
   return __builtin_aarch64_ld1v2sf ((const __builtin_aarch64_simd_sf *) a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_f64 (const float64_t *a)
 {
   return (float64x1_t) {*a};
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_p8 (const poly8_t *a)
 {
   return (poly8x8_t)
     __builtin_aarch64_ld1v8qi ((const __builtin_aarch64_simd_qi *) a);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_p16 (const poly16_t *a)
 {
   return (poly16x4_t)
     __builtin_aarch64_ld1v4hi ((const __builtin_aarch64_simd_hi *) a);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_s8 (const int8_t *a)
 {
   return __builtin_aarch64_ld1v8qi ((const __builtin_aarch64_simd_qi *) a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_s16 (const int16_t *a)
 {
   return __builtin_aarch64_ld1v4hi ((const __builtin_aarch64_simd_hi *) a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_s32 (const int32_t *a)
 {
   return __builtin_aarch64_ld1v2si ((const __builtin_aarch64_simd_si *) a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_s64 (const int64_t *a)
 {
   return (int64x1_t) {*a};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_u8 (const uint8_t *a)
 {
   return (uint8x8_t)
     __builtin_aarch64_ld1v8qi ((const __builtin_aarch64_simd_qi *) a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_u16 (const uint16_t *a)
 {
   return (uint16x4_t)
     __builtin_aarch64_ld1v4hi ((const __builtin_aarch64_simd_hi *) a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_u32 (const uint32_t *a)
 {
   return (uint32x2_t)
     __builtin_aarch64_ld1v2si ((const __builtin_aarch64_simd_si *) a);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_u64 (const uint64_t *a)
 {
   return (uint64x1_t) {*a};
@@ -14623,84 +16344,97 @@ vld1_u64 (const uint64_t *a)
 
 /* vld1q */
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_f16 (const float16_t *__a)
 {
   return __builtin_aarch64_ld1v8hf (__a);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_f32 (const float32_t *a)
 {
   return __builtin_aarch64_ld1v4sf ((const __builtin_aarch64_simd_sf *) a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_f64 (const float64_t *a)
 {
   return __builtin_aarch64_ld1v2df ((const __builtin_aarch64_simd_df *) a);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_p8 (const poly8_t *a)
 {
   return (poly8x16_t)
     __builtin_aarch64_ld1v16qi ((const __builtin_aarch64_simd_qi *) a);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_p16 (const poly16_t *a)
 {
   return (poly16x8_t)
     __builtin_aarch64_ld1v8hi ((const __builtin_aarch64_simd_hi *) a);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_s8 (const int8_t *a)
 {
   return __builtin_aarch64_ld1v16qi ((const __builtin_aarch64_simd_qi *) a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_s16 (const int16_t *a)
 {
   return __builtin_aarch64_ld1v8hi ((const __builtin_aarch64_simd_hi *) a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_s32 (const int32_t *a)
 {
   return __builtin_aarch64_ld1v4si ((const __builtin_aarch64_simd_si *) a);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_s64 (const int64_t *a)
 {
   return __builtin_aarch64_ld1v2di ((const __builtin_aarch64_simd_di *) a);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_u8 (const uint8_t *a)
 {
   return (uint8x16_t)
     __builtin_aarch64_ld1v16qi ((const __builtin_aarch64_simd_qi *) a);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_u16 (const uint16_t *a)
 {
   return (uint16x8_t)
     __builtin_aarch64_ld1v8hi ((const __builtin_aarch64_simd_hi *) a);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_u32 (const uint32_t *a)
 {
   return (uint32x4_t)
     __builtin_aarch64_ld1v4si ((const __builtin_aarch64_simd_si *) a);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_u64 (const uint64_t *a)
 {
   return (uint64x2_t)
@@ -14709,79 +16443,92 @@ vld1q_u64 (const uint64_t *a)
 
 /* vld1_dup  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_f16 (const float16_t* __a)
 {
   return vdup_n_f16 (*__a);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_f32 (const float32_t* __a)
 {
   return vdup_n_f32 (*__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_f64 (const float64_t* __a)
 {
   return vdup_n_f64 (*__a);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_p8 (const poly8_t* __a)
 {
   return vdup_n_p8 (*__a);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_p16 (const poly16_t* __a)
 {
   return vdup_n_p16 (*__a);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_s8 (const int8_t* __a)
 {
   return vdup_n_s8 (*__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_s16 (const int16_t* __a)
 {
   return vdup_n_s16 (*__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_s32 (const int32_t* __a)
 {
   return vdup_n_s32 (*__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_s64 (const int64_t* __a)
 {
   return vdup_n_s64 (*__a);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_u8 (const uint8_t* __a)
 {
   return vdup_n_u8 (*__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_u16 (const uint16_t* __a)
 {
   return vdup_n_u16 (*__a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_u32 (const uint32_t* __a)
 {
   return vdup_n_u32 (*__a);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_dup_u64 (const uint64_t* __a)
 {
   return vdup_n_u64 (*__a);
@@ -14789,79 +16536,92 @@ vld1_dup_u64 (const uint64_t* __a)
 
 /* vld1q_dup  */
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_f16 (const float16_t* __a)
 {
   return vdupq_n_f16 (*__a);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_f32 (const float32_t* __a)
 {
   return vdupq_n_f32 (*__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_f64 (const float64_t* __a)
 {
   return vdupq_n_f64 (*__a);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_p8 (const poly8_t* __a)
 {
   return vdupq_n_p8 (*__a);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_p16 (const poly16_t* __a)
 {
   return vdupq_n_p16 (*__a);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_s8 (const int8_t* __a)
 {
   return vdupq_n_s8 (*__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_s16 (const int16_t* __a)
 {
   return vdupq_n_s16 (*__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_s32 (const int32_t* __a)
 {
   return vdupq_n_s32 (*__a);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_s64 (const int64_t* __a)
 {
   return vdupq_n_s64 (*__a);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_u8 (const uint8_t* __a)
 {
   return vdupq_n_u8 (*__a);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_u16 (const uint16_t* __a)
 {
   return vdupq_n_u16 (*__a);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_u32 (const uint32_t* __a)
 {
   return vdupq_n_u32 (*__a);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_dup_u64 (const uint64_t* __a)
 {
   return vdupq_n_u64 (*__a);
@@ -14869,79 +16629,92 @@ vld1q_dup_u64 (const uint64_t* __a)
 
 /* vld1_lane  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_f16 (const float16_t *__src, float16x4_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_f32 (const float32_t *__src, float32x2_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_f64 (const float64_t *__src, float64x1_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_p8 (const poly8_t *__src, poly8x8_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_p16 (const poly16_t *__src, poly16x4_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_s8 (const int8_t *__src, int8x8_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_s16 (const int16_t *__src, int16x4_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_s32 (const int32_t *__src, int32x2_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_s64 (const int64_t *__src, int64x1_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_u8 (const uint8_t *__src, uint8x8_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_u16 (const uint16_t *__src, uint16x4_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_u32 (const uint32_t *__src, uint32x2_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1_lane_u64 (const uint64_t *__src, uint64x1_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
@@ -14949,79 +16722,92 @@ vld1_lane_u64 (const uint64_t *__src, uint64x1_t __vec, const int __lane)
 
 /* vld1q_lane  */
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_f16 (const float16_t *__src, float16x8_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_f32 (const float32_t *__src, float32x4_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_f64 (const float64_t *__src, float64x2_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_p8 (const poly8_t *__src, poly8x16_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_p16 (const poly16_t *__src, poly16x8_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_s8 (const int8_t *__src, int8x16_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_s16 (const int16_t *__src, int16x8_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_s32 (const int32_t *__src, int32x4_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_s64 (const int64_t *__src, int64x2_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_u8 (const uint8_t *__src, uint8x16_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_u16 (const uint16_t *__src, uint16x8_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_u32 (const uint32_t *__src, uint32x4_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld1q_lane_u64 (const uint64_t *__src, uint64x2_t __vec, const int __lane)
 {
   return __aarch64_vset_lane_any (*__src, __vec, __lane);
@@ -15029,7 +16815,8 @@ vld1q_lane_u64 (const uint64_t *__src, uint64x2_t __vec, const int __lane)
 
 /* vldn */
 
-__extension__ static __inline int64x1x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_s64 (const int64_t * __a)
 {
   int64x1x2_t ret;
@@ -15040,7 +16827,8 @@ vld2_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x1x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_u64 (const uint64_t * __a)
 {
   uint64x1x2_t ret;
@@ -15051,7 +16839,8 @@ vld2_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x1x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_f64 (const float64_t * __a)
 {
   float64x1x2_t ret;
@@ -15062,7 +16851,8 @@ vld2_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_s8 (const int8_t * __a)
 {
   int8x8x2_t ret;
@@ -15073,7 +16863,8 @@ vld2_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_p8 (const poly8_t * __a)
 {
   poly8x8x2_t ret;
@@ -15084,7 +16875,8 @@ vld2_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_s16 (const int16_t * __a)
 {
   int16x4x2_t ret;
@@ -15095,7 +16887,8 @@ vld2_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_p16 (const poly16_t * __a)
 {
   poly16x4x2_t ret;
@@ -15106,7 +16899,8 @@ vld2_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_s32 (const int32_t * __a)
 {
   int32x2x2_t ret;
@@ -15117,7 +16911,8 @@ vld2_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_u8 (const uint8_t * __a)
 {
   uint8x8x2_t ret;
@@ -15128,7 +16923,8 @@ vld2_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_u16 (const uint16_t * __a)
 {
   uint16x4x2_t ret;
@@ -15139,7 +16935,8 @@ vld2_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_u32 (const uint32_t * __a)
 {
   uint32x2x2_t ret;
@@ -15150,7 +16947,8 @@ vld2_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_f16 (const float16_t * __a)
 {
   float16x4x2_t ret;
@@ -15161,7 +16959,8 @@ vld2_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_f32 (const float32_t * __a)
 {
   float32x2x2_t ret;
@@ -15172,7 +16971,8 @@ vld2_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x16x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_s8 (const int8_t * __a)
 {
   int8x16x2_t ret;
@@ -15183,7 +16983,8 @@ vld2q_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x16x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_p8 (const poly8_t * __a)
 {
   poly8x16x2_t ret;
@@ -15194,7 +16995,8 @@ vld2q_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_s16 (const int16_t * __a)
 {
   int16x8x2_t ret;
@@ -15205,7 +17007,8 @@ vld2q_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_p16 (const poly16_t * __a)
 {
   poly16x8x2_t ret;
@@ -15216,7 +17019,8 @@ vld2q_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_s32 (const int32_t * __a)
 {
   int32x4x2_t ret;
@@ -15227,7 +17031,8 @@ vld2q_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_s64 (const int64_t * __a)
 {
   int64x2x2_t ret;
@@ -15238,7 +17043,8 @@ vld2q_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x16x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_u8 (const uint8_t * __a)
 {
   uint8x16x2_t ret;
@@ -15249,7 +17055,8 @@ vld2q_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_u16 (const uint16_t * __a)
 {
   uint16x8x2_t ret;
@@ -15260,7 +17067,8 @@ vld2q_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_u32 (const uint32_t * __a)
 {
   uint32x4x2_t ret;
@@ -15271,7 +17079,8 @@ vld2q_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_u64 (const uint64_t * __a)
 {
   uint64x2x2_t ret;
@@ -15282,7 +17091,8 @@ vld2q_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_f16 (const float16_t * __a)
 {
   float16x8x2_t ret;
@@ -15293,7 +17103,8 @@ vld2q_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_f32 (const float32_t * __a)
 {
   float32x4x2_t ret;
@@ -15304,7 +17115,8 @@ vld2q_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_f64 (const float64_t * __a)
 {
   float64x2x2_t ret;
@@ -15315,7 +17127,8 @@ vld2q_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x1x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_s64 (const int64_t * __a)
 {
   int64x1x3_t ret;
@@ -15327,7 +17140,8 @@ vld3_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x1x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_u64 (const uint64_t * __a)
 {
   uint64x1x3_t ret;
@@ -15339,7 +17153,8 @@ vld3_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x1x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_f64 (const float64_t * __a)
 {
   float64x1x3_t ret;
@@ -15351,7 +17166,8 @@ vld3_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_s8 (const int8_t * __a)
 {
   int8x8x3_t ret;
@@ -15363,7 +17179,8 @@ vld3_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_p8 (const poly8_t * __a)
 {
   poly8x8x3_t ret;
@@ -15375,7 +17192,8 @@ vld3_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_s16 (const int16_t * __a)
 {
   int16x4x3_t ret;
@@ -15387,7 +17205,8 @@ vld3_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_p16 (const poly16_t * __a)
 {
   poly16x4x3_t ret;
@@ -15399,7 +17218,8 @@ vld3_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_s32 (const int32_t * __a)
 {
   int32x2x3_t ret;
@@ -15411,7 +17231,8 @@ vld3_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_u8 (const uint8_t * __a)
 {
   uint8x8x3_t ret;
@@ -15423,7 +17244,8 @@ vld3_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_u16 (const uint16_t * __a)
 {
   uint16x4x3_t ret;
@@ -15435,7 +17257,8 @@ vld3_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_u32 (const uint32_t * __a)
 {
   uint32x2x3_t ret;
@@ -15447,7 +17270,8 @@ vld3_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_f16 (const float16_t * __a)
 {
   float16x4x3_t ret;
@@ -15459,7 +17283,8 @@ vld3_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_f32 (const float32_t * __a)
 {
   float32x2x3_t ret;
@@ -15471,7 +17296,8 @@ vld3_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x16x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_s8 (const int8_t * __a)
 {
   int8x16x3_t ret;
@@ -15483,7 +17309,8 @@ vld3q_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x16x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_p8 (const poly8_t * __a)
 {
   poly8x16x3_t ret;
@@ -15495,7 +17322,8 @@ vld3q_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_s16 (const int16_t * __a)
 {
   int16x8x3_t ret;
@@ -15507,7 +17335,8 @@ vld3q_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_p16 (const poly16_t * __a)
 {
   poly16x8x3_t ret;
@@ -15519,7 +17348,8 @@ vld3q_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_s32 (const int32_t * __a)
 {
   int32x4x3_t ret;
@@ -15531,7 +17361,8 @@ vld3q_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_s64 (const int64_t * __a)
 {
   int64x2x3_t ret;
@@ -15543,7 +17374,8 @@ vld3q_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x16x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_u8 (const uint8_t * __a)
 {
   uint8x16x3_t ret;
@@ -15555,7 +17387,8 @@ vld3q_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_u16 (const uint16_t * __a)
 {
   uint16x8x3_t ret;
@@ -15567,7 +17400,8 @@ vld3q_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_u32 (const uint32_t * __a)
 {
   uint32x4x3_t ret;
@@ -15579,7 +17413,8 @@ vld3q_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_u64 (const uint64_t * __a)
 {
   uint64x2x3_t ret;
@@ -15591,7 +17426,8 @@ vld3q_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_f16 (const float16_t * __a)
 {
   float16x8x3_t ret;
@@ -15603,7 +17439,8 @@ vld3q_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_f32 (const float32_t * __a)
 {
   float32x4x3_t ret;
@@ -15615,7 +17452,8 @@ vld3q_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_f64 (const float64_t * __a)
 {
   float64x2x3_t ret;
@@ -15627,7 +17465,8 @@ vld3q_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x1x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_s64 (const int64_t * __a)
 {
   int64x1x4_t ret;
@@ -15640,7 +17479,8 @@ vld4_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x1x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_u64 (const uint64_t * __a)
 {
   uint64x1x4_t ret;
@@ -15653,7 +17493,8 @@ vld4_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x1x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_f64 (const float64_t * __a)
 {
   float64x1x4_t ret;
@@ -15666,7 +17507,8 @@ vld4_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_s8 (const int8_t * __a)
 {
   int8x8x4_t ret;
@@ -15679,7 +17521,8 @@ vld4_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_p8 (const poly8_t * __a)
 {
   poly8x8x4_t ret;
@@ -15692,7 +17535,8 @@ vld4_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_s16 (const int16_t * __a)
 {
   int16x4x4_t ret;
@@ -15705,7 +17549,8 @@ vld4_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_p16 (const poly16_t * __a)
 {
   poly16x4x4_t ret;
@@ -15718,7 +17563,8 @@ vld4_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_s32 (const int32_t * __a)
 {
   int32x2x4_t ret;
@@ -15731,7 +17577,8 @@ vld4_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_u8 (const uint8_t * __a)
 {
   uint8x8x4_t ret;
@@ -15744,7 +17591,8 @@ vld4_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_u16 (const uint16_t * __a)
 {
   uint16x4x4_t ret;
@@ -15757,7 +17605,8 @@ vld4_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_u32 (const uint32_t * __a)
 {
   uint32x2x4_t ret;
@@ -15770,7 +17619,8 @@ vld4_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_f16 (const float16_t * __a)
 {
   float16x4x4_t ret;
@@ -15783,7 +17633,8 @@ vld4_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_f32 (const float32_t * __a)
 {
   float32x2x4_t ret;
@@ -15796,7 +17647,8 @@ vld4_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_s8 (const int8_t * __a)
 {
   int8x16x4_t ret;
@@ -15809,7 +17661,8 @@ vld4q_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_p8 (const poly8_t * __a)
 {
   poly8x16x4_t ret;
@@ -15822,7 +17675,8 @@ vld4q_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_s16 (const int16_t * __a)
 {
   int16x8x4_t ret;
@@ -15835,7 +17689,8 @@ vld4q_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_p16 (const poly16_t * __a)
 {
   poly16x8x4_t ret;
@@ -15848,7 +17703,8 @@ vld4q_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_s32 (const int32_t * __a)
 {
   int32x4x4_t ret;
@@ -15861,7 +17717,8 @@ vld4q_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_s64 (const int64_t * __a)
 {
   int64x2x4_t ret;
@@ -15874,7 +17731,8 @@ vld4q_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_u8 (const uint8_t * __a)
 {
   uint8x16x4_t ret;
@@ -15887,7 +17745,8 @@ vld4q_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_u16 (const uint16_t * __a)
 {
   uint16x8x4_t ret;
@@ -15900,7 +17759,8 @@ vld4q_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_u32 (const uint32_t * __a)
 {
   uint32x4x4_t ret;
@@ -15913,7 +17773,8 @@ vld4q_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_u64 (const uint64_t * __a)
 {
   uint64x2x4_t ret;
@@ -15926,7 +17787,8 @@ vld4q_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_f16 (const float16_t * __a)
 {
   float16x8x4_t ret;
@@ -15939,7 +17801,8 @@ vld4q_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_f32 (const float32_t * __a)
 {
   float32x4x4_t ret;
@@ -15952,7 +17815,8 @@ vld4q_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_f64 (const float64_t * __a)
 {
   float64x2x4_t ret;
@@ -15967,7 +17831,8 @@ vld4q_f64 (const float64_t * __a)
 
 /* vldn_dup */
 
-__extension__ static __inline int8x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_s8 (const int8_t * __a)
 {
   int8x8x2_t ret;
@@ -15978,7 +17843,8 @@ vld2_dup_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_s16 (const int16_t * __a)
 {
   int16x4x2_t ret;
@@ -15989,7 +17855,8 @@ vld2_dup_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_s32 (const int32_t * __a)
 {
   int32x2x2_t ret;
@@ -16000,7 +17867,8 @@ vld2_dup_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_f16 (const float16_t * __a)
 {
   float16x4x2_t ret;
@@ -16011,7 +17879,8 @@ vld2_dup_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_f32 (const float32_t * __a)
 {
   float32x2x2_t ret;
@@ -16022,7 +17891,8 @@ vld2_dup_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x1x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_f64 (const float64_t * __a)
 {
   float64x1x2_t ret;
@@ -16033,7 +17903,8 @@ vld2_dup_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_u8 (const uint8_t * __a)
 {
   uint8x8x2_t ret;
@@ -16044,7 +17915,8 @@ vld2_dup_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_u16 (const uint16_t * __a)
 {
   uint16x4x2_t ret;
@@ -16055,7 +17927,8 @@ vld2_dup_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_u32 (const uint32_t * __a)
 {
   uint32x2x2_t ret;
@@ -16066,7 +17939,8 @@ vld2_dup_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_p8 (const poly8_t * __a)
 {
   poly8x8x2_t ret;
@@ -16077,7 +17951,8 @@ vld2_dup_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_p16 (const poly16_t * __a)
 {
   poly16x4x2_t ret;
@@ -16088,7 +17963,8 @@ vld2_dup_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x1x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_s64 (const int64_t * __a)
 {
   int64x1x2_t ret;
@@ -16099,7 +17975,8 @@ vld2_dup_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x1x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2_dup_u64 (const uint64_t * __a)
 {
   uint64x1x2_t ret;
@@ -16110,7 +17987,8 @@ vld2_dup_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x16x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_s8 (const int8_t * __a)
 {
   int8x16x2_t ret;
@@ -16121,7 +17999,8 @@ vld2q_dup_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x16x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_p8 (const poly8_t * __a)
 {
   poly8x16x2_t ret;
@@ -16132,7 +18011,8 @@ vld2q_dup_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_s16 (const int16_t * __a)
 {
   int16x8x2_t ret;
@@ -16143,7 +18023,8 @@ vld2q_dup_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_p16 (const poly16_t * __a)
 {
   poly16x8x2_t ret;
@@ -16154,7 +18035,8 @@ vld2q_dup_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_s32 (const int32_t * __a)
 {
   int32x4x2_t ret;
@@ -16165,7 +18047,8 @@ vld2q_dup_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_s64 (const int64_t * __a)
 {
   int64x2x2_t ret;
@@ -16176,7 +18059,8 @@ vld2q_dup_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x16x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_u8 (const uint8_t * __a)
 {
   uint8x16x2_t ret;
@@ -16187,7 +18071,8 @@ vld2q_dup_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_u16 (const uint16_t * __a)
 {
   uint16x8x2_t ret;
@@ -16198,7 +18083,8 @@ vld2q_dup_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_u32 (const uint32_t * __a)
 {
   uint32x4x2_t ret;
@@ -16209,7 +18095,8 @@ vld2q_dup_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_u64 (const uint64_t * __a)
 {
   uint64x2x2_t ret;
@@ -16220,7 +18107,8 @@ vld2q_dup_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_f16 (const float16_t * __a)
 {
   float16x8x2_t ret;
@@ -16231,7 +18119,8 @@ vld2q_dup_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_f32 (const float32_t * __a)
 {
   float32x4x2_t ret;
@@ -16242,7 +18131,8 @@ vld2q_dup_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld2q_dup_f64 (const float64_t * __a)
 {
   float64x2x2_t ret;
@@ -16253,7 +18143,8 @@ vld2q_dup_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x1x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_s64 (const int64_t * __a)
 {
   int64x1x3_t ret;
@@ -16265,7 +18156,8 @@ vld3_dup_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x1x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_u64 (const uint64_t * __a)
 {
   uint64x1x3_t ret;
@@ -16277,7 +18169,8 @@ vld3_dup_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x1x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_f64 (const float64_t * __a)
 {
   float64x1x3_t ret;
@@ -16289,7 +18182,8 @@ vld3_dup_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_s8 (const int8_t * __a)
 {
   int8x8x3_t ret;
@@ -16301,7 +18195,8 @@ vld3_dup_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_p8 (const poly8_t * __a)
 {
   poly8x8x3_t ret;
@@ -16313,7 +18208,8 @@ vld3_dup_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_s16 (const int16_t * __a)
 {
   int16x4x3_t ret;
@@ -16325,7 +18221,8 @@ vld3_dup_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_p16 (const poly16_t * __a)
 {
   poly16x4x3_t ret;
@@ -16337,7 +18234,8 @@ vld3_dup_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_s32 (const int32_t * __a)
 {
   int32x2x3_t ret;
@@ -16349,7 +18247,8 @@ vld3_dup_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_u8 (const uint8_t * __a)
 {
   uint8x8x3_t ret;
@@ -16361,7 +18260,8 @@ vld3_dup_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_u16 (const uint16_t * __a)
 {
   uint16x4x3_t ret;
@@ -16373,7 +18273,8 @@ vld3_dup_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_u32 (const uint32_t * __a)
 {
   uint32x2x3_t ret;
@@ -16385,7 +18286,8 @@ vld3_dup_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_f16 (const float16_t * __a)
 {
   float16x4x3_t ret;
@@ -16397,7 +18299,8 @@ vld3_dup_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3_dup_f32 (const float32_t * __a)
 {
   float32x2x3_t ret;
@@ -16409,7 +18312,8 @@ vld3_dup_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x16x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_s8 (const int8_t * __a)
 {
   int8x16x3_t ret;
@@ -16421,7 +18325,8 @@ vld3q_dup_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x16x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_p8 (const poly8_t * __a)
 {
   poly8x16x3_t ret;
@@ -16433,7 +18338,8 @@ vld3q_dup_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_s16 (const int16_t * __a)
 {
   int16x8x3_t ret;
@@ -16445,7 +18351,8 @@ vld3q_dup_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_p16 (const poly16_t * __a)
 {
   poly16x8x3_t ret;
@@ -16457,7 +18364,8 @@ vld3q_dup_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_s32 (const int32_t * __a)
 {
   int32x4x3_t ret;
@@ -16469,7 +18377,8 @@ vld3q_dup_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_s64 (const int64_t * __a)
 {
   int64x2x3_t ret;
@@ -16481,7 +18390,8 @@ vld3q_dup_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x16x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_u8 (const uint8_t * __a)
 {
   uint8x16x3_t ret;
@@ -16493,7 +18403,8 @@ vld3q_dup_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_u16 (const uint16_t * __a)
 {
   uint16x8x3_t ret;
@@ -16505,7 +18416,8 @@ vld3q_dup_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_u32 (const uint32_t * __a)
 {
   uint32x4x3_t ret;
@@ -16517,7 +18429,8 @@ vld3q_dup_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_u64 (const uint64_t * __a)
 {
   uint64x2x3_t ret;
@@ -16529,7 +18442,8 @@ vld3q_dup_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x8x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_f16 (const float16_t * __a)
 {
   float16x8x3_t ret;
@@ -16541,7 +18455,8 @@ vld3q_dup_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x4x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_f32 (const float32_t * __a)
 {
   float32x4x3_t ret;
@@ -16553,7 +18468,8 @@ vld3q_dup_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x2x3_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2x3_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld3q_dup_f64 (const float64_t * __a)
 {
   float64x2x3_t ret;
@@ -16565,7 +18481,8 @@ vld3q_dup_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x1x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_s64 (const int64_t * __a)
 {
   int64x1x4_t ret;
@@ -16578,7 +18495,8 @@ vld4_dup_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x1x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_u64 (const uint64_t * __a)
 {
   uint64x1x4_t ret;
@@ -16591,7 +18509,8 @@ vld4_dup_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x1x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_f64 (const float64_t * __a)
 {
   float64x1x4_t ret;
@@ -16604,7 +18523,8 @@ vld4_dup_f64 (const float64_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_s8 (const int8_t * __a)
 {
   int8x8x4_t ret;
@@ -16617,7 +18537,8 @@ vld4_dup_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_p8 (const poly8_t * __a)
 {
   poly8x8x4_t ret;
@@ -16630,7 +18551,8 @@ vld4_dup_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_s16 (const int16_t * __a)
 {
   int16x4x4_t ret;
@@ -16643,7 +18565,8 @@ vld4_dup_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_p16 (const poly16_t * __a)
 {
   poly16x4x4_t ret;
@@ -16656,7 +18579,8 @@ vld4_dup_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_s32 (const int32_t * __a)
 {
   int32x2x4_t ret;
@@ -16669,7 +18593,8 @@ vld4_dup_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_u8 (const uint8_t * __a)
 {
   uint8x8x4_t ret;
@@ -16682,7 +18607,8 @@ vld4_dup_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_u16 (const uint16_t * __a)
 {
   uint16x4x4_t ret;
@@ -16695,7 +18621,8 @@ vld4_dup_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_u32 (const uint32_t * __a)
 {
   uint32x2x4_t ret;
@@ -16708,7 +18635,8 @@ vld4_dup_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_f16 (const float16_t * __a)
 {
   float16x4x4_t ret;
@@ -16721,7 +18649,8 @@ vld4_dup_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4_dup_f32 (const float32_t * __a)
 {
   float32x2x4_t ret;
@@ -16734,7 +18663,8 @@ vld4_dup_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int8x16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_s8 (const int8_t * __a)
 {
   int8x16x4_t ret;
@@ -16747,7 +18677,8 @@ vld4q_dup_s8 (const int8_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly8x16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_p8 (const poly8_t * __a)
 {
   poly8x16x4_t ret;
@@ -16760,7 +18691,8 @@ vld4q_dup_p8 (const poly8_t * __a)
   return ret;
 }
 
-__extension__ static __inline int16x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_s16 (const int16_t * __a)
 {
   int16x8x4_t ret;
@@ -16773,7 +18705,8 @@ vld4q_dup_s16 (const int16_t * __a)
   return ret;
 }
 
-__extension__ static __inline poly16x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_p16 (const poly16_t * __a)
 {
   poly16x8x4_t ret;
@@ -16786,7 +18719,8 @@ vld4q_dup_p16 (const poly16_t * __a)
   return ret;
 }
 
-__extension__ static __inline int32x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_s32 (const int32_t * __a)
 {
   int32x4x4_t ret;
@@ -16799,7 +18733,8 @@ vld4q_dup_s32 (const int32_t * __a)
   return ret;
 }
 
-__extension__ static __inline int64x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_s64 (const int64_t * __a)
 {
   int64x2x4_t ret;
@@ -16812,7 +18747,8 @@ vld4q_dup_s64 (const int64_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint8x16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_u8 (const uint8_t * __a)
 {
   uint8x16x4_t ret;
@@ -16825,7 +18761,8 @@ vld4q_dup_u8 (const uint8_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint16x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_u16 (const uint16_t * __a)
 {
   uint16x8x4_t ret;
@@ -16838,7 +18775,8 @@ vld4q_dup_u16 (const uint16_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint32x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_u32 (const uint32_t * __a)
 {
   uint32x4x4_t ret;
@@ -16851,7 +18789,8 @@ vld4q_dup_u32 (const uint32_t * __a)
   return ret;
 }
 
-__extension__ static __inline uint64x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_u64 (const uint64_t * __a)
 {
   uint64x2x4_t ret;
@@ -16864,7 +18803,8 @@ vld4q_dup_u64 (const uint64_t * __a)
   return ret;
 }
 
-__extension__ static __inline float16x8x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_f16 (const float16_t * __a)
 {
   float16x8x4_t ret;
@@ -16877,7 +18817,8 @@ vld4q_dup_f16 (const float16_t * __a)
   return ret;
 }
 
-__extension__ static __inline float32x4x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_f32 (const float32_t * __a)
 {
   float32x4x4_t ret;
@@ -16890,7 +18831,8 @@ vld4q_dup_f32 (const float32_t * __a)
   return ret;
 }
 
-__extension__ static __inline float64x2x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vld4q_dup_f64 (const float64_t * __a)
 {
   float64x2x4_t ret;
@@ -16907,7 +18849,8 @@ vld4q_dup_f64 (const float64_t * __a)
 
 #define __LD2_LANE_FUNC(intype, vectype, largetype, ptrtype, mode,        \
                         qmode, ptrmode, funcsuffix, signedtype)           \
-__extension__ static __inline intype __attribute__ ((__always_inline__))   \
+__extension__ extern __inline intype \
+__attribute__ ((__always_inline__, __gnu_inline__,__artificial__)) \
 vld2_lane_##funcsuffix (const ptrtype * __ptr, intype __b, const int __c)  \
 {                                                                         \
   __builtin_aarch64_simd_oi __o;                                          \
@@ -16961,7 +18904,8 @@ __LD2_LANE_FUNC (uint64x1x2_t, uint64x1_t, uint64x2x2_t, uint64_t, di, v2di, di,
 /* vld2q_lane */
 
 #define __LD2_LANE_FUNC(intype, vtype, ptrtype, mode, ptrmode, funcsuffix) \
-__extension__ static __inline intype __attribute__ ((__always_inline__))   \
+__extension__ extern __inline intype \
+__attribute__ ((__always_inline__, __gnu_inline__,__artificial__)) \
 vld2q_lane_##funcsuffix (const ptrtype * __ptr, intype __b, const int __c) \
 {                                                                         \
   __builtin_aarch64_simd_oi __o;                                          \
@@ -16995,7 +18939,8 @@ __LD2_LANE_FUNC (uint64x2x2_t, uint64x2_t, uint64_t, v2di, di, u64)
 
 #define __LD3_LANE_FUNC(intype, vectype, largetype, ptrtype, mode,        \
                         qmode, ptrmode, funcsuffix, signedtype)           \
-__extension__ static __inline intype __attribute__ ((__always_inline__))   \
+__extension__ extern __inline intype \
+__attribute__ ((__always_inline__, __gnu_inline__,__artificial__)) \
 vld3_lane_##funcsuffix (const ptrtype * __ptr, intype __b, const int __c)  \
 {                                                                         \
   __builtin_aarch64_simd_ci __o;                                          \
@@ -17055,7 +19000,8 @@ __LD3_LANE_FUNC (uint64x1x3_t, uint64x1_t, uint64x2x3_t, uint64_t, di, v2di, di,
 /* vld3q_lane */
 
 #define __LD3_LANE_FUNC(intype, vtype, ptrtype, mode, ptrmode, funcsuffix) \
-__extension__ static __inline intype __attribute__ ((__always_inline__))   \
+__extension__ extern __inline intype \
+__attribute__ ((__always_inline__, __gnu_inline__,__artificial__)) \
 vld3q_lane_##funcsuffix (const ptrtype * __ptr, intype __b, const int __c) \
 {                                                                         \
   __builtin_aarch64_simd_ci __o;                                          \
@@ -17091,7 +19037,8 @@ __LD3_LANE_FUNC (uint64x2x3_t, uint64x2_t, uint64_t, v2di, di, u64)
 
 #define __LD4_LANE_FUNC(intype, vectype, largetype, ptrtype, mode,        \
                         qmode, ptrmode, funcsuffix, signedtype)           \
-__extension__ static __inline intype __attribute__ ((__always_inline__))   \
+__extension__ extern __inline intype \
+__attribute__ ((__always_inline__, __gnu_inline__,__artificial__)) \
 vld4_lane_##funcsuffix (const ptrtype * __ptr, intype __b, const int __c)  \
 {                                                                         \
   __builtin_aarch64_simd_xi __o;                                          \
@@ -17159,7 +19106,8 @@ __LD4_LANE_FUNC (uint64x1x4_t, uint64x1_t, uint64x2x4_t, uint64_t, di, v2di, di,
 /* vld4q_lane */
 
 #define __LD4_LANE_FUNC(intype, vtype, ptrtype, mode, ptrmode, funcsuffix) \
-__extension__ static __inline intype __attribute__ ((__always_inline__))   \
+__extension__ extern __inline intype \
+__attribute__ ((__always_inline__, __gnu_inline__,__artificial__)) \
 vld4q_lane_##funcsuffix (const ptrtype * __ptr, intype __b, const int __c) \
 {                                                                         \
   __builtin_aarch64_simd_xi __o;                                          \
@@ -17195,13 +19143,15 @@ __LD4_LANE_FUNC (uint64x2x4_t, uint64x2_t, uint64_t, v2di, di, u64)
 
 /* vmax */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmax_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __builtin_aarch64_smax_nanv2sf (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmax_f64 (float64x1_t __a, float64x1_t __b)
 {
     return (float64x1_t)
@@ -17209,90 +19159,104 @@ vmax_f64 (float64x1_t __a, float64x1_t __b)
                                      vget_lane_f64 (__b, 0)) };
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmax_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_smaxv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmax_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_smaxv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmax_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_smaxv2si (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmax_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_umaxv8qi ((int8x8_t) __a,
                                                 (int8x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmax_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_umaxv4hi ((int16x4_t) __a,
                                                  (int16x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmax_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_umaxv2si ((int32x2_t) __a,
                                                  (int32x2_t) __b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_smax_nanv4sf (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_smax_nanv2df (__a, __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_smaxv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_smaxv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_smaxv4si (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (uint8x16_t) __builtin_aarch64_umaxv16qi ((int8x16_t) __a,
                                                   (int8x16_t) __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_umaxv8hi ((int16x8_t) __a,
                                                  (int16x8_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_umaxv4si ((int32x4_t) __a,
@@ -17300,109 +19264,127 @@ vmaxq_u32 (uint32x4_t __a, uint32x4_t __b)
 }
 /* vmulx */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __builtin_aarch64_fmulxv2sf (__a, __b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_fmulxv4sf (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (float64x1_t) {__builtin_aarch64_fmulxdf (__a[0], __b[0])};
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_fmulxv2df (__a, __b);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxs_f32 (float32_t __a, float32_t __b)
 {
   return __builtin_aarch64_fmulxsf (__a, __b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxd_f64 (float64_t __a, float64_t __b)
 {
   return __builtin_aarch64_fmulxdf (__a, __b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_lane_f32 (float32x2_t __a, float32x2_t __v, const int __lane)
 {
   return vmulx_f32 (__a, __aarch64_vdup_lane_f32 (__v, __lane));
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_lane_f64 (float64x1_t __a, float64x1_t __v, const int __lane)
 {
   return vmulx_f64 (__a, __aarch64_vdup_lane_f64 (__v, __lane));
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_lane_f32 (float32x4_t __a, float32x2_t __v, const int __lane)
 {
   return vmulxq_f32 (__a, __aarch64_vdupq_lane_f32 (__v, __lane));
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_lane_f64 (float64x2_t __a, float64x1_t __v, const int __lane)
 {
   return vmulxq_f64 (__a, __aarch64_vdupq_lane_f64 (__v, __lane));
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_laneq_f32 (float32x2_t __a, float32x4_t __v, const int __lane)
 {
   return vmulx_f32 (__a, __aarch64_vdup_laneq_f32 (__v, __lane));
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_laneq_f64 (float64x1_t __a, float64x2_t __v, const int __lane)
 {
   return vmulx_f64 (__a, __aarch64_vdup_laneq_f64 (__v, __lane));
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_laneq_f32 (float32x4_t __a, float32x4_t __v, const int __lane)
 {
   return vmulxq_f32 (__a, __aarch64_vdupq_laneq_f32 (__v, __lane));
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_laneq_f64 (float64x2_t __a, float64x2_t __v, const int __lane)
 {
   return vmulxq_f64 (__a, __aarch64_vdupq_laneq_f64 (__v, __lane));
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxs_lane_f32 (float32_t __a, float32x2_t __v, const int __lane)
 {
   return vmulxs_f32 (__a, __aarch64_vget_lane_any (__v, __lane));
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxs_laneq_f32 (float32_t __a, float32x4_t __v, const int __lane)
 {
   return vmulxs_f32 (__a, __aarch64_vget_lane_any (__v, __lane));
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxd_lane_f64 (float64_t __a, float64x1_t __v, const int __lane)
 {
   return vmulxd_f64 (__a, __aarch64_vget_lane_any (__v, __lane));
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxd_laneq_f64 (float64_t __a, float64x2_t __v, const int __lane)
 {
   return vmulxd_f64 (__a, __aarch64_vget_lane_any (__v, __lane));
@@ -17410,109 +19392,126 @@ vmulxd_laneq_f64 (float64_t __a, float64x2_t __v, const int __lane)
 
 /* vpmax  */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmax_s8 (int8x8_t a, int8x8_t b)
 {
   return __builtin_aarch64_smaxpv8qi (a, b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmax_s16 (int16x4_t a, int16x4_t b)
 {
   return __builtin_aarch64_smaxpv4hi (a, b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmax_s32 (int32x2_t a, int32x2_t b)
 {
   return __builtin_aarch64_smaxpv2si (a, b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmax_u8 (uint8x8_t a, uint8x8_t b)
 {
   return (uint8x8_t) __builtin_aarch64_umaxpv8qi ((int8x8_t) a,
                                                  (int8x8_t) b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmax_u16 (uint16x4_t a, uint16x4_t b)
 {
   return (uint16x4_t) __builtin_aarch64_umaxpv4hi ((int16x4_t) a,
                                                   (int16x4_t) b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmax_u32 (uint32x2_t a, uint32x2_t b)
 {
   return (uint32x2_t) __builtin_aarch64_umaxpv2si ((int32x2_t) a,
                                                   (int32x2_t) b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxq_s8 (int8x16_t a, int8x16_t b)
 {
   return __builtin_aarch64_smaxpv16qi (a, b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxq_s16 (int16x8_t a, int16x8_t b)
 {
   return __builtin_aarch64_smaxpv8hi (a, b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxq_s32 (int32x4_t a, int32x4_t b)
 {
   return __builtin_aarch64_smaxpv4si (a, b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxq_u8 (uint8x16_t a, uint8x16_t b)
 {
   return (uint8x16_t) __builtin_aarch64_umaxpv16qi ((int8x16_t) a,
                                                    (int8x16_t) b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxq_u16 (uint16x8_t a, uint16x8_t b)
 {
   return (uint16x8_t) __builtin_aarch64_umaxpv8hi ((int16x8_t) a,
                                                   (int16x8_t) b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxq_u32 (uint32x4_t a, uint32x4_t b)
 {
   return (uint32x4_t) __builtin_aarch64_umaxpv4si ((int32x4_t) a,
                                                   (int32x4_t) b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmax_f32 (float32x2_t a, float32x2_t b)
 {
   return __builtin_aarch64_smax_nanpv2sf (a, b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxq_f32 (float32x4_t a, float32x4_t b)
 {
   return __builtin_aarch64_smax_nanpv4sf (a, b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxq_f64 (float64x2_t a, float64x2_t b)
 {
   return __builtin_aarch64_smax_nanpv2df (a, b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxqd_f64 (float64x2_t a)
 {
   return __builtin_aarch64_reduc_smax_nan_scal_v2df (a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxs_f32 (float32x2_t a)
 {
   return __builtin_aarch64_reduc_smax_nan_scal_v2sf (a);
@@ -17520,31 +19519,36 @@ vpmaxs_f32 (float32x2_t a)
 
 /* vpmaxnm  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxnm_f32 (float32x2_t a, float32x2_t b)
 {
   return __builtin_aarch64_smaxpv2sf (a, b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxnmq_f32 (float32x4_t a, float32x4_t b)
 {
   return __builtin_aarch64_smaxpv4sf (a, b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxnmq_f64 (float64x2_t a, float64x2_t b)
 {
   return __builtin_aarch64_smaxpv2df (a, b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxnmqd_f64 (float64x2_t a)
 {
   return __builtin_aarch64_reduc_smax_scal_v2df (a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxnms_f32 (float32x2_t a)
 {
   return __builtin_aarch64_reduc_smax_scal_v2sf (a);
@@ -17552,109 +19556,126 @@ vpmaxnms_f32 (float32x2_t a)
 
 /* vpmin  */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmin_s8 (int8x8_t a, int8x8_t b)
 {
   return __builtin_aarch64_sminpv8qi (a, b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmin_s16 (int16x4_t a, int16x4_t b)
 {
   return __builtin_aarch64_sminpv4hi (a, b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmin_s32 (int32x2_t a, int32x2_t b)
 {
   return __builtin_aarch64_sminpv2si (a, b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmin_u8 (uint8x8_t a, uint8x8_t b)
 {
   return (uint8x8_t) __builtin_aarch64_uminpv8qi ((int8x8_t) a,
                                                  (int8x8_t) b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmin_u16 (uint16x4_t a, uint16x4_t b)
 {
   return (uint16x4_t) __builtin_aarch64_uminpv4hi ((int16x4_t) a,
                                                   (int16x4_t) b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmin_u32 (uint32x2_t a, uint32x2_t b)
 {
   return (uint32x2_t) __builtin_aarch64_uminpv2si ((int32x2_t) a,
                                                   (int32x2_t) b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminq_s8 (int8x16_t a, int8x16_t b)
 {
   return __builtin_aarch64_sminpv16qi (a, b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminq_s16 (int16x8_t a, int16x8_t b)
 {
   return __builtin_aarch64_sminpv8hi (a, b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminq_s32 (int32x4_t a, int32x4_t b)
 {
   return __builtin_aarch64_sminpv4si (a, b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminq_u8 (uint8x16_t a, uint8x16_t b)
 {
   return (uint8x16_t) __builtin_aarch64_uminpv16qi ((int8x16_t) a,
                                                    (int8x16_t) b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminq_u16 (uint16x8_t a, uint16x8_t b)
 {
   return (uint16x8_t) __builtin_aarch64_uminpv8hi ((int16x8_t) a,
                                                   (int16x8_t) b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminq_u32 (uint32x4_t a, uint32x4_t b)
 {
   return (uint32x4_t) __builtin_aarch64_uminpv4si ((int32x4_t) a,
                                                   (int32x4_t) b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmin_f32 (float32x2_t a, float32x2_t b)
 {
   return __builtin_aarch64_smin_nanpv2sf (a, b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminq_f32 (float32x4_t a, float32x4_t b)
 {
   return __builtin_aarch64_smin_nanpv4sf (a, b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminq_f64 (float64x2_t a, float64x2_t b)
 {
   return __builtin_aarch64_smin_nanpv2df (a, b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminqd_f64 (float64x2_t a)
 {
   return __builtin_aarch64_reduc_smin_nan_scal_v2df (a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmins_f32 (float32x2_t a)
 {
   return __builtin_aarch64_reduc_smin_nan_scal_v2sf (a);
@@ -17662,31 +19683,36 @@ vpmins_f32 (float32x2_t a)
 
 /* vpminnm  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminnm_f32 (float32x2_t a, float32x2_t b)
 {
   return __builtin_aarch64_sminpv2sf (a, b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminnmq_f32 (float32x4_t a, float32x4_t b)
 {
   return __builtin_aarch64_sminpv4sf (a, b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminnmq_f64 (float64x2_t a, float64x2_t b)
 {
   return __builtin_aarch64_sminpv2df (a, b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminnmqd_f64 (float64x2_t a)
 {
   return __builtin_aarch64_reduc_smin_scal_v2df (a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminnms_f32 (float32x2_t a)
 {
   return __builtin_aarch64_reduc_smin_scal_v2sf (a);
@@ -17694,13 +19720,15 @@ vpminnms_f32 (float32x2_t a)
 
 /* vmaxnm  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnm_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __builtin_aarch64_fmaxv2sf (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnm_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (float64x1_t)
@@ -17708,13 +19736,15 @@ vmaxnm_f64 (float64x1_t __a, float64x1_t __b)
                                vget_lane_f64 (__b, 0)) };
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnmq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_fmaxv4sf (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnmq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_fmaxv2df (__a, __b);
@@ -17722,91 +19752,106 @@ vmaxnmq_f64 (float64x2_t __a, float64x2_t __b)
 
 /* vmaxv  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxv_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_reduc_smax_nan_scal_v2sf (__a);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxv_s8 (int8x8_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v8qi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxv_s16 (int16x4_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v4hi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxv_s32 (int32x2_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v2si (__a);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxv_u8 (uint8x8_t __a)
 {
   return __builtin_aarch64_reduc_umax_scal_v8qi_uu (__a);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxv_u16 (uint16x4_t __a)
 {
   return __builtin_aarch64_reduc_umax_scal_v4hi_uu (__a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxv_u32 (uint32x2_t __a)
 {
   return __builtin_aarch64_reduc_umax_scal_v2si_uu (__a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxvq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_reduc_smax_nan_scal_v4sf (__a);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxvq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_reduc_smax_nan_scal_v2df (__a);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxvq_s8 (int8x16_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v16qi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxvq_s16 (int16x8_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v8hi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxvq_s32 (int32x4_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v4si (__a);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxvq_u8 (uint8x16_t __a)
 {
   return __builtin_aarch64_reduc_umax_scal_v16qi_uu (__a);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxvq_u16 (uint16x8_t __a)
 {
   return __builtin_aarch64_reduc_umax_scal_v8hi_uu (__a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxvq_u32 (uint32x4_t __a)
 {
   return __builtin_aarch64_reduc_umax_scal_v4si_uu (__a);
@@ -17814,19 +19859,22 @@ vmaxvq_u32 (uint32x4_t __a)
 
 /* vmaxnmv  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnmv_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v2sf (__a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnmvq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v4sf (__a);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnmvq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v2df (__a);
@@ -17834,13 +19882,15 @@ vmaxnmvq_f64 (float64x2_t __a)
 
 /* vmin  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmin_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __builtin_aarch64_smin_nanv2sf (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmin_f64 (float64x1_t __a, float64x1_t __b)
 {
     return (float64x1_t)
@@ -17848,90 +19898,104 @@ vmin_f64 (float64x1_t __a, float64x1_t __b)
                                          vget_lane_f64 (__b, 0)) };
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmin_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_sminv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmin_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_sminv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmin_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_sminv2si (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmin_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_uminv8qi ((int8x8_t) __a,
                                                 (int8x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmin_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_uminv4hi ((int16x4_t) __a,
                                                  (int16x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmin_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_uminv2si ((int32x2_t) __a,
                                                  (int32x2_t) __b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_smin_nanv4sf (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_smin_nanv2df (__a, __b);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_sminv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_sminv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_sminv4si (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return (uint8x16_t) __builtin_aarch64_uminv16qi ((int8x16_t) __a,
                                                   (int8x16_t) __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return (uint16x8_t) __builtin_aarch64_uminv8hi ((int16x8_t) __a,
                                                  (int16x8_t) __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return (uint32x4_t) __builtin_aarch64_uminv4si ((int32x4_t) __a,
@@ -17940,13 +20004,15 @@ vminq_u32 (uint32x4_t __a, uint32x4_t __b)
 
 /* vminnm  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnm_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __builtin_aarch64_fminv2sf (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnm_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (float64x1_t)
@@ -17954,13 +20020,15 @@ vminnm_f64 (float64x1_t __a, float64x1_t __b)
                                vget_lane_f64 (__b, 0)) };
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnmq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_fminv4sf (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnmq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_fminv2df (__a, __b);
@@ -17968,91 +20036,106 @@ vminnmq_f64 (float64x2_t __a, float64x2_t __b)
 
 /* vminv  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminv_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_reduc_smin_nan_scal_v2sf (__a);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminv_s8 (int8x8_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v8qi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminv_s16 (int16x4_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v4hi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminv_s32 (int32x2_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v2si (__a);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminv_u8 (uint8x8_t __a)
 {
   return __builtin_aarch64_reduc_umin_scal_v8qi_uu (__a);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminv_u16 (uint16x4_t __a)
 {
   return __builtin_aarch64_reduc_umin_scal_v4hi_uu (__a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminv_u32 (uint32x2_t __a)
 {
   return __builtin_aarch64_reduc_umin_scal_v2si_uu (__a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminvq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_reduc_smin_nan_scal_v4sf (__a);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminvq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_reduc_smin_nan_scal_v2df (__a);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminvq_s8 (int8x16_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v16qi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminvq_s16 (int16x8_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v8hi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminvq_s32 (int32x4_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v4si (__a);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminvq_u8 (uint8x16_t __a)
 {
   return __builtin_aarch64_reduc_umin_scal_v16qi_uu (__a);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminvq_u16 (uint16x8_t __a)
 {
   return __builtin_aarch64_reduc_umin_scal_v8hi_uu (__a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminvq_u32 (uint32x4_t __a)
 {
   return __builtin_aarch64_reduc_umin_scal_v4si_uu (__a);
@@ -18060,19 +20143,22 @@ vminvq_u32 (uint32x4_t __a)
 
 /* vminnmv  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnmv_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v2sf (__a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnmvq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v4sf (__a);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnmvq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v2df (__a);
@@ -18080,25 +20166,29 @@ vminnmvq_f64 (float64x2_t __a)
 
 /* vmla */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_f32 (float32x2_t a, float32x2_t b, float32x2_t c)
 {
   return a + b * c;
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_f64 (float64x1_t __a, float64x1_t __b, float64x1_t __c)
 {
   return __a + __b * __c;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_f32 (float32x4_t a, float32x4_t b, float32x4_t c)
 {
   return a + b * c;
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_f64 (float64x2_t a, float64x2_t b, float64x2_t c)
 {
   return a + b * c;
@@ -18106,35 +20196,40 @@ vmlaq_f64 (float64x2_t a, float64x2_t b, float64x2_t c)
 
 /* vmla_lane  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_lane_f32 (float32x2_t __a, float32x2_t __b,
               float32x2_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_lane_s16 (int16x4_t __a, int16x4_t __b,
                int16x4_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_lane_s32 (int32x2_t __a, int32x2_t __b,
                int32x2_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_lane_u16 (uint16x4_t __a, uint16x4_t __b,
                uint16x4_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_lane_u32 (uint32x2_t __a, uint32x2_t __b,
               uint32x2_t __c, const int __lane)
 {
@@ -18143,35 +20238,40 @@ vmla_lane_u32 (uint32x2_t __a, uint32x2_t __b,
 
 /* vmla_laneq  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_laneq_f32 (float32x2_t __a, float32x2_t __b,
                float32x4_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_laneq_s16 (int16x4_t __a, int16x4_t __b,
                int16x8_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_laneq_s32 (int32x2_t __a, int32x2_t __b,
                int32x4_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_laneq_u16 (uint16x4_t __a, uint16x4_t __b,
                uint16x8_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmla_laneq_u32 (uint32x2_t __a, uint32x2_t __b,
                uint32x4_t __c, const int __lane)
 {
@@ -18180,35 +20280,40 @@ vmla_laneq_u32 (uint32x2_t __a, uint32x2_t __b,
 
 /* vmlaq_lane  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_lane_f32 (float32x4_t __a, float32x4_t __b,
                float32x2_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_lane_s16 (int16x8_t __a, int16x8_t __b,
                int16x4_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_lane_s32 (int32x4_t __a, int32x4_t __b,
                int32x2_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_lane_u16 (uint16x8_t __a, uint16x8_t __b,
                uint16x4_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_lane_u32 (uint32x4_t __a, uint32x4_t __b,
                uint32x2_t __c, const int __lane)
 {
@@ -18217,35 +20322,40 @@ vmlaq_lane_u32 (uint32x4_t __a, uint32x4_t __b,
 
   /* vmlaq_laneq  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_laneq_f32 (float32x4_t __a, float32x4_t __b,
                 float32x4_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_laneq_s16 (int16x8_t __a, int16x8_t __b,
                int16x8_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_laneq_s32 (int32x4_t __a, int32x4_t __b,
                int32x4_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_laneq_u16 (uint16x8_t __a, uint16x8_t __b,
                uint16x8_t __c, const int __lane)
 {
   return (__a + (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlaq_laneq_u32 (uint32x4_t __a, uint32x4_t __b,
                uint32x4_t __c, const int __lane)
 {
@@ -18254,25 +20364,29 @@ vmlaq_laneq_u32 (uint32x4_t __a, uint32x4_t __b,
 
 /* vmls  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_f32 (float32x2_t a, float32x2_t b, float32x2_t c)
 {
   return a - b * c;
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_f64 (float64x1_t __a, float64x1_t __b, float64x1_t __c)
 {
   return __a - __b * __c;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_f32 (float32x4_t a, float32x4_t b, float32x4_t c)
 {
   return a - b * c;
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_f64 (float64x2_t a, float64x2_t b, float64x2_t c)
 {
   return a - b * c;
@@ -18280,35 +20394,40 @@ vmlsq_f64 (float64x2_t a, float64x2_t b, float64x2_t c)
 
 /* vmls_lane  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_lane_f32 (float32x2_t __a, float32x2_t __b,
               float32x2_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_lane_s16 (int16x4_t __a, int16x4_t __b,
                int16x4_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_lane_s32 (int32x2_t __a, int32x2_t __b,
                int32x2_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_lane_u16 (uint16x4_t __a, uint16x4_t __b,
                uint16x4_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_lane_u32 (uint32x2_t __a, uint32x2_t __b,
               uint32x2_t __c, const int __lane)
 {
@@ -18317,35 +20436,40 @@ vmls_lane_u32 (uint32x2_t __a, uint32x2_t __b,
 
 /* vmls_laneq  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_laneq_f32 (float32x2_t __a, float32x2_t __b,
               float32x4_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_laneq_s16 (int16x4_t __a, int16x4_t __b,
                int16x8_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_laneq_s32 (int32x2_t __a, int32x2_t __b,
                int32x4_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_laneq_u16 (uint16x4_t __a, uint16x4_t __b,
                uint16x8_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmls_laneq_u32 (uint32x2_t __a, uint32x2_t __b,
                uint32x4_t __c, const int __lane)
 {
@@ -18354,35 +20478,40 @@ vmls_laneq_u32 (uint32x2_t __a, uint32x2_t __b,
 
 /* vmlsq_lane  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_lane_f32 (float32x4_t __a, float32x4_t __b,
                float32x2_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_lane_s16 (int16x8_t __a, int16x8_t __b,
                int16x4_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_lane_s32 (int32x4_t __a, int32x4_t __b,
                int32x2_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_lane_u16 (uint16x8_t __a, uint16x8_t __b,
                uint16x4_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_lane_u32 (uint32x4_t __a, uint32x4_t __b,
                uint32x2_t __c, const int __lane)
 {
@@ -18391,34 +20520,39 @@ vmlsq_lane_u32 (uint32x4_t __a, uint32x4_t __b,
 
   /* vmlsq_laneq  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_laneq_f32 (float32x4_t __a, float32x4_t __b,
                float32x4_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_laneq_s16 (int16x8_t __a, int16x8_t __b,
                int16x8_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_laneq_s32 (int32x4_t __a, int32x4_t __b,
                int32x4_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_laneq_u16 (uint16x8_t __a, uint16x8_t __b,
                uint16x8_t __c, const int __lane)
 {
   return (__a - (__b * __aarch64_vget_lane_any (__c, __lane)));
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmlsq_laneq_u32 (uint32x4_t __a, uint32x4_t __b,
                uint32x4_t __c, const int __lane)
 {
@@ -18427,157 +20561,183 @@ vmlsq_laneq_u32 (uint32x4_t __a, uint32x4_t __b,
 
 /* vmov_n_  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_f16 (float16_t __a)
 {
   return vdup_n_f16 (__a);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_f32 (float32_t __a)
 {
   return vdup_n_f32 (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_f64 (float64_t __a)
 {
   return (float64x1_t) {__a};
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_p8 (poly8_t __a)
 {
   return vdup_n_p8 (__a);
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_p16 (poly16_t __a)
 {
   return vdup_n_p16 (__a);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_s8 (int8_t __a)
 {
   return vdup_n_s8 (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_s16 (int16_t __a)
 {
   return vdup_n_s16 (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_s32 (int32_t __a)
 {
   return vdup_n_s32 (__a);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_s64 (int64_t __a)
 {
   return (int64x1_t) {__a};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_u8 (uint8_t __a)
 {
   return vdup_n_u8 (__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_u16 (uint16_t __a)
 {
     return vdup_n_u16 (__a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_u32 (uint32_t __a)
 {
    return vdup_n_u32 (__a);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmov_n_u64 (uint64_t __a)
 {
   return (uint64x1_t) {__a};
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_f16 (float16_t __a)
 {
   return vdupq_n_f16 (__a);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_f32 (float32_t __a)
 {
   return vdupq_n_f32 (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_f64 (float64_t __a)
 {
   return vdupq_n_f64 (__a);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_p8 (poly8_t __a)
 {
   return vdupq_n_p8 (__a);
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_p16 (poly16_t __a)
 {
   return vdupq_n_p16 (__a);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_s8 (int8_t __a)
 {
   return vdupq_n_s8 (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_s16 (int16_t __a)
 {
   return vdupq_n_s16 (__a);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_s32 (int32_t __a)
 {
   return vdupq_n_s32 (__a);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_s64 (int64_t __a)
 {
   return vdupq_n_s64 (__a);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_u8 (uint8_t __a)
 {
   return vdupq_n_u8 (__a);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_u16 (uint16_t __a)
 {
   return vdupq_n_u16 (__a);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_u32 (uint32_t __a)
 {
   return vdupq_n_u32 (__a);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmovq_n_u64 (uint64_t __a)
 {
   return vdupq_n_u64 (__a);
@@ -18585,37 +20745,43 @@ vmovq_n_u64 (uint64_t __a)
 
 /* vmul_lane  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_lane_f32 (float32x2_t __a, float32x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_lane_f64 (float64x1_t __a, float64x1_t __b, const int __lane)
 {
   return __a * __b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_lane_s16 (int16x4_t __a, int16x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_lane_s32 (int32x2_t __a, int32x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_lane_u16 (uint16x4_t __a, uint16x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_lane_u32 (uint32x2_t __a, uint32x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
@@ -18623,13 +20789,15 @@ vmul_lane_u32 (uint32x2_t __a, uint32x2_t __b, const int __lane)
 
 /* vmuld_lane  */
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmuld_lane_f64 (float64_t __a, float64x1_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmuld_laneq_f64 (float64_t __a, float64x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
@@ -18637,13 +20805,15 @@ vmuld_laneq_f64 (float64_t __a, float64x2_t __b, const int __lane)
 
 /* vmuls_lane  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmuls_lane_f32 (float32_t __a, float32x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmuls_laneq_f32 (float32_t __a, float32x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
@@ -18651,37 +20821,43 @@ vmuls_laneq_f32 (float32_t __a, float32x4_t __b, const int __lane)
 
 /* vmul_laneq  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_laneq_f32 (float32x2_t __a, float32x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_laneq_f64 (float64x1_t __a, float64x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_laneq_s16 (int16x4_t __a, int16x8_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_laneq_s32 (int32x2_t __a, int32x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_laneq_u16 (uint16x4_t __a, uint16x8_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_laneq_u32 (uint32x2_t __a, uint32x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
@@ -18689,7 +20865,8 @@ vmul_laneq_u32 (uint32x2_t __a, uint32x4_t __b, const int __lane)
 
 /* vmul_n  */
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_n_f64  (float64x1_t __a, float64_t __b)
 {
   return (float64x1_t) { vget_lane_f64 (__a, 0) * __b };
@@ -18697,38 +20874,44 @@ vmul_n_f64  (float64x1_t __a, float64_t __b)
 
 /* vmulq_lane  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_lane_f32 (float32x4_t __a, float32x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_lane_f64 (float64x2_t __a, float64x1_t __b, const int __lane)
 {
   __AARCH64_LANE_CHECK (__a, __lane);
   return __a * __b[0];
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_lane_s16 (int16x8_t __a, int16x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_lane_s32 (int32x4_t __a, int32x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_lane_u16 (uint16x8_t __a, uint16x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_lane_u32 (uint32x4_t __a, uint32x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
@@ -18736,37 +20919,43 @@ vmulq_lane_u32 (uint32x4_t __a, uint32x2_t __b, const int __lane)
 
 /* vmulq_laneq  */
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_laneq_f32 (float32x4_t __a, float32x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_laneq_f64 (float64x2_t __a, float64x2_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_laneq_s16 (int16x8_t __a, int16x8_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_laneq_s32 (int32x4_t __a, int32x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_laneq_u16 (uint16x8_t __a, uint16x8_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_laneq_u32 (uint32x4_t __a, uint32x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
@@ -18774,67 +20963,78 @@ vmulq_laneq_u32 (uint32x4_t __a, uint32x4_t __b, const int __lane)
 
 /* vmul_n.  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_n_f32 (float32x2_t __a, float32_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_n_f32 (float32x4_t __a, float32_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_n_f64 (float64x2_t __a, float64_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_n_s16 (int16x4_t __a, int16_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_n_s16 (int16x8_t __a, int16_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_n_s32 (int32x2_t __a, int32_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_n_s32 (int32x4_t __a, int32_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_n_u16 (uint16x4_t __a, uint16_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_n_u16 (uint16x8_t __a, uint16_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_n_u32 (uint32x2_t __a, uint32_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_n_u32 (uint32x4_t __a, uint32_t __b)
 {
   return __a * __b;
@@ -18842,85 +21042,99 @@ vmulq_n_u32 (uint32x4_t __a, uint32_t __b)
 
 /* vmvn  */
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvn_p8 (poly8x8_t __a)
 {
   return (poly8x8_t) ~((int8x8_t) __a);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvn_s8 (int8x8_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvn_s16 (int16x4_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvn_s32 (int32x2_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvn_u8 (uint8x8_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvn_u16 (uint16x4_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvn_u32 (uint32x2_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvnq_p8 (poly8x16_t __a)
 {
   return (poly8x16_t) ~((int8x16_t) __a);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvnq_s8 (int8x16_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvnq_s16 (int16x8_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvnq_s32 (int32x4_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvnq_u8 (uint8x16_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvnq_u16 (uint16x8_t __a)
 {
   return ~__a;
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmvnq_u32 (uint32x4_t __a)
 {
   return ~__a;
@@ -18928,73 +21142,85 @@ vmvnq_u32 (uint32x4_t __a)
 
 /* vneg  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vneg_f32 (float32x2_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vneg_f64 (float64x1_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vneg_s8 (int8x8_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vneg_s16 (int16x4_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vneg_s32 (int32x2_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vneg_s64 (int64x1_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vnegq_f32 (float32x4_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vnegq_f64 (float64x2_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vnegq_s8 (int8x16_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vnegq_s16 (int16x8_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vnegq_s32 (int32x4_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vnegq_s64 (int64x2_t __a)
 {
   return -__a;
@@ -19002,82 +21228,95 @@ vnegq_s64 (int64x2_t __a)
 
 /* vpadd  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadd_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __builtin_aarch64_faddpv2sf (__a, __b);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_faddpv4sf (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_faddpv2df (__a, __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadd_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_addpv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadd_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_addpv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadd_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_addpv2si (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadd_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return (uint8x8_t) __builtin_aarch64_addpv8qi ((int8x8_t) __a,
                                                 (int8x8_t) __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadd_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return (uint16x4_t) __builtin_aarch64_addpv4hi ((int16x4_t) __a,
                                                  (int16x4_t) __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadd_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return (uint32x2_t) __builtin_aarch64_addpv2si ((int32x2_t) __a,
                                                  (int32x2_t) __b);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadds_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v2sf (__a);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddd_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_reduc_plus_scal_v2df (__a);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddd_s64 (int64x2_t __a)
 {
   return __builtin_aarch64_addpdi (__a);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddd_u64 (uint64x2_t __a)
 {
   return __builtin_aarch64_addpdi ((int64x2_t) __a);
@@ -19085,31 +21324,36 @@ vpaddd_u64 (uint64x2_t __a)
 
 /* vqabs */
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabsq_s64 (int64x2_t __a)
 {
   return (int64x2_t) __builtin_aarch64_sqabsv2di (__a);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabsb_s8 (int8_t __a)
 {
   return (int8_t) __builtin_aarch64_sqabsqi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabsh_s16 (int16_t __a)
 {
   return (int16_t) __builtin_aarch64_sqabshi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabss_s32 (int32_t __a)
 {
   return (int32_t) __builtin_aarch64_sqabssi (__a);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqabsd_s64 (int64_t __a)
 {
   return __builtin_aarch64_sqabsdi (__a);
@@ -19117,49 +21361,57 @@ vqabsd_s64 (int64_t __a)
 
 /* vqadd */
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddb_s8 (int8_t __a, int8_t __b)
 {
   return (int8_t) __builtin_aarch64_sqaddqi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddh_s16 (int16_t __a, int16_t __b)
 {
   return (int16_t) __builtin_aarch64_sqaddhi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadds_s32 (int32_t __a, int32_t __b)
 {
   return (int32_t) __builtin_aarch64_sqaddsi (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddd_s64 (int64_t __a, int64_t __b)
 {
   return __builtin_aarch64_sqadddi (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddb_u8 (uint8_t __a, uint8_t __b)
 {
   return (uint8_t) __builtin_aarch64_uqaddqi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddh_u16 (uint16_t __a, uint16_t __b)
 {
   return (uint16_t) __builtin_aarch64_uqaddhi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqadds_u32 (uint32_t __a, uint32_t __b)
 {
   return (uint32_t) __builtin_aarch64_uqaddsi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqaddd_u64 (uint64_t __a, uint64_t __b)
 {
   return __builtin_aarch64_uqadddi_uuu (__a, __b);
@@ -19167,137 +21419,159 @@ vqaddd_u64 (uint64_t __a, uint64_t __b)
 
 /* vqdmlal */
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
 {
   return __builtin_aarch64_sqdmlalv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_high_s16 (int32x4_t __a, int16x8_t __b, int16x8_t __c)
 {
   return __builtin_aarch64_sqdmlal2v8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_high_lane_s16 (int32x4_t __a, int16x8_t __b, int16x4_t __c,
                       int const __d)
 {
   return __builtin_aarch64_sqdmlal2_lanev8hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_high_laneq_s16 (int32x4_t __a, int16x8_t __b, int16x8_t __c,
                        int const __d)
 {
   return __builtin_aarch64_sqdmlal2_laneqv8hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_high_n_s16 (int32x4_t __a, int16x8_t __b, int16_t __c)
 {
   return __builtin_aarch64_sqdmlal2_nv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_lane_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c, int const __d)
 {
   return __builtin_aarch64_sqdmlal_lanev4hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_laneq_s16 (int32x4_t __a, int16x4_t __b, int16x8_t __c, int const __d)
 {
   return __builtin_aarch64_sqdmlal_laneqv4hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_n_s16 (int32x4_t __a, int16x4_t __b, int16_t __c)
 {
   return __builtin_aarch64_sqdmlal_nv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
 {
   return __builtin_aarch64_sqdmlalv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_high_s32 (int64x2_t __a, int32x4_t __b, int32x4_t __c)
 {
   return __builtin_aarch64_sqdmlal2v4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_high_lane_s32 (int64x2_t __a, int32x4_t __b, int32x2_t __c,
                       int const __d)
 {
   return __builtin_aarch64_sqdmlal2_lanev4si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_high_laneq_s32 (int64x2_t __a, int32x4_t __b, int32x4_t __c,
                        int const __d)
 {
   return __builtin_aarch64_sqdmlal2_laneqv4si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_high_n_s32 (int64x2_t __a, int32x4_t __b, int32_t __c)
 {
   return __builtin_aarch64_sqdmlal2_nv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_lane_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c, int const __d)
 {
   return __builtin_aarch64_sqdmlal_lanev2si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_laneq_s32 (int64x2_t __a, int32x2_t __b, int32x4_t __c, int const __d)
 {
   return __builtin_aarch64_sqdmlal_laneqv2si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlal_n_s32 (int64x2_t __a, int32x2_t __b, int32_t __c)
 {
   return __builtin_aarch64_sqdmlal_nv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlalh_s16 (int32_t __a, int16_t __b, int16_t __c)
 {
   return __builtin_aarch64_sqdmlalhi (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlalh_lane_s16 (int32_t __a, int16_t __b, int16x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqdmlal_lanehi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlalh_laneq_s16 (int32_t __a, int16_t __b, int16x8_t __c, const int __d)
 {
   return __builtin_aarch64_sqdmlal_laneqhi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlals_s32 (int64_t __a, int32_t __b, int32_t __c)
 {
   return __builtin_aarch64_sqdmlalsi (__a, __b, __c);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlals_lane_s32 (int64_t __a, int32_t __b, int32x2_t __c, const int __d)
 {
   return __builtin_aarch64_sqdmlal_lanesi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlals_laneq_s32 (int64_t __a, int32_t __b, int32x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqdmlal_laneqsi (__a, __b, __c, __d);
@@ -19305,137 +21579,159 @@ vqdmlals_laneq_s32 (int64_t __a, int32_t __b, int32x4_t __c, const int __d)
 
 /* vqdmlsl */
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c)
 {
   return __builtin_aarch64_sqdmlslv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_high_s16 (int32x4_t __a, int16x8_t __b, int16x8_t __c)
 {
   return __builtin_aarch64_sqdmlsl2v8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_high_lane_s16 (int32x4_t __a, int16x8_t __b, int16x4_t __c,
                       int const __d)
 {
   return __builtin_aarch64_sqdmlsl2_lanev8hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_high_laneq_s16 (int32x4_t __a, int16x8_t __b, int16x8_t __c,
                        int const __d)
 {
   return __builtin_aarch64_sqdmlsl2_laneqv8hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_high_n_s16 (int32x4_t __a, int16x8_t __b, int16_t __c)
 {
   return __builtin_aarch64_sqdmlsl2_nv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_lane_s16 (int32x4_t __a, int16x4_t __b, int16x4_t __c, int const __d)
 {
   return __builtin_aarch64_sqdmlsl_lanev4hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_laneq_s16 (int32x4_t __a, int16x4_t __b, int16x8_t __c, int const __d)
 {
   return __builtin_aarch64_sqdmlsl_laneqv4hi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_n_s16 (int32x4_t __a, int16x4_t __b, int16_t __c)
 {
   return __builtin_aarch64_sqdmlsl_nv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c)
 {
   return __builtin_aarch64_sqdmlslv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_high_s32 (int64x2_t __a, int32x4_t __b, int32x4_t __c)
 {
   return __builtin_aarch64_sqdmlsl2v4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_high_lane_s32 (int64x2_t __a, int32x4_t __b, int32x2_t __c,
                       int const __d)
 {
   return __builtin_aarch64_sqdmlsl2_lanev4si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_high_laneq_s32 (int64x2_t __a, int32x4_t __b, int32x4_t __c,
                        int const __d)
 {
   return __builtin_aarch64_sqdmlsl2_laneqv4si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_high_n_s32 (int64x2_t __a, int32x4_t __b, int32_t __c)
 {
   return __builtin_aarch64_sqdmlsl2_nv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_lane_s32 (int64x2_t __a, int32x2_t __b, int32x2_t __c, int const __d)
 {
   return __builtin_aarch64_sqdmlsl_lanev2si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_laneq_s32 (int64x2_t __a, int32x2_t __b, int32x4_t __c, int const __d)
 {
   return __builtin_aarch64_sqdmlsl_laneqv2si (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsl_n_s32 (int64x2_t __a, int32x2_t __b, int32_t __c)
 {
   return __builtin_aarch64_sqdmlsl_nv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlslh_s16 (int32_t __a, int16_t __b, int16_t __c)
 {
   return __builtin_aarch64_sqdmlslhi (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlslh_lane_s16 (int32_t __a, int16_t __b, int16x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqdmlsl_lanehi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlslh_laneq_s16 (int32_t __a, int16_t __b, int16x8_t __c, const int __d)
 {
   return __builtin_aarch64_sqdmlsl_laneqhi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsls_s32 (int64_t __a, int32_t __b, int32_t __c)
 {
   return __builtin_aarch64_sqdmlslsi (__a, __b, __c);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsls_lane_s32 (int64_t __a, int32_t __b, int32x2_t __c, const int __d)
 {
   return __builtin_aarch64_sqdmlsl_lanesi (__a, __b, __c, __d);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmlsls_laneq_s32 (int64_t __a, int32_t __b, int32x4_t __c, const int __d)
 {
   return __builtin_aarch64_sqdmlsl_laneqsi (__a, __b, __c, __d);
@@ -19443,61 +21739,71 @@ vqdmlsls_laneq_s32 (int64_t __a, int32_t __b, int32x4_t __c, const int __d)
 
 /* vqdmulh */
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulh_lane_s16 (int16x4_t __a, int16x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_lanev4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulh_lane_s32 (int32x2_t __a, int32x2_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_lanev2si (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhq_lane_s16 (int16x8_t __a, int16x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_lanev8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhq_lane_s32 (int32x4_t __a, int32x2_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_lanev4si (__a, __b, __c);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhh_s16 (int16_t __a, int16_t __b)
 {
   return (int16_t) __builtin_aarch64_sqdmulhhi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhh_lane_s16 (int16_t __a, int16x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_lanehi (__a, __b, __c);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhh_laneq_s16 (int16_t __a, int16x8_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_laneqhi (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhs_s32 (int32_t __a, int32_t __b)
 {
   return (int32_t) __builtin_aarch64_sqdmulhsi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhs_lane_s32 (int32_t __a, int32x2_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_lanesi (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulhs_laneq_s32 (int32_t __a, int32x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmulh_laneqsi (__a, __b, __c);
@@ -19505,133 +21811,155 @@ vqdmulhs_laneq_s32 (int32_t __a, int32x4_t __b, const int __c)
 
 /* vqdmull */
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_sqdmullv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_high_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_sqdmull2v8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_high_lane_s16 (int16x8_t __a, int16x4_t __b, int const __c)
 {
   return __builtin_aarch64_sqdmull2_lanev8hi (__a, __b,__c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_high_laneq_s16 (int16x8_t __a, int16x8_t __b, int const __c)
 {
   return __builtin_aarch64_sqdmull2_laneqv8hi (__a, __b,__c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_high_n_s16 (int16x8_t __a, int16_t __b)
 {
   return __builtin_aarch64_sqdmull2_nv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_lane_s16 (int16x4_t __a, int16x4_t __b, int const __c)
 {
   return __builtin_aarch64_sqdmull_lanev4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_laneq_s16 (int16x4_t __a, int16x8_t __b, int const __c)
 {
   return __builtin_aarch64_sqdmull_laneqv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_n_s16 (int16x4_t __a, int16_t __b)
 {
   return __builtin_aarch64_sqdmull_nv4hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_sqdmullv2si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_high_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_sqdmull2v4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_high_lane_s32 (int32x4_t __a, int32x2_t __b, int const __c)
 {
   return __builtin_aarch64_sqdmull2_lanev4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_high_laneq_s32 (int32x4_t __a, int32x4_t __b, int const __c)
 {
   return __builtin_aarch64_sqdmull2_laneqv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_high_n_s32 (int32x4_t __a, int32_t __b)
 {
   return __builtin_aarch64_sqdmull2_nv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_lane_s32 (int32x2_t __a, int32x2_t __b, int const __c)
 {
   return __builtin_aarch64_sqdmull_lanev2si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_laneq_s32 (int32x2_t __a, int32x4_t __b, int const __c)
 {
   return __builtin_aarch64_sqdmull_laneqv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmull_n_s32 (int32x2_t __a, int32_t __b)
 {
   return __builtin_aarch64_sqdmull_nv2si (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmullh_s16 (int16_t __a, int16_t __b)
 {
   return (int32_t) __builtin_aarch64_sqdmullhi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmullh_lane_s16 (int16_t __a, int16x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmull_lanehi (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmullh_laneq_s16 (int16_t __a, int16x8_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmull_laneqhi (__a, __b, __c);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulls_s32 (int32_t __a, int32_t __b)
 {
   return __builtin_aarch64_sqdmullsi (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulls_lane_s32 (int32_t __a, int32x2_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmull_lanesi (__a, __b, __c);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqdmulls_laneq_s32 (int32_t __a, int32x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqdmull_laneqsi (__a, __b, __c);
@@ -19639,73 +21967,85 @@ vqdmulls_laneq_s32 (int32_t __a, int32x4_t __b, const int __c)
 
 /* vqmovn */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_s16 (int16x8_t __a)
 {
   return (int8x8_t) __builtin_aarch64_sqmovnv8hi (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_s32 (int32x4_t __a)
 {
   return (int16x4_t) __builtin_aarch64_sqmovnv4si (__a);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_s64 (int64x2_t __a)
 {
   return (int32x2_t) __builtin_aarch64_sqmovnv2di (__a);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_u16 (uint16x8_t __a)
 {
   return (uint8x8_t) __builtin_aarch64_uqmovnv8hi ((int16x8_t) __a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_u32 (uint32x4_t __a)
 {
   return (uint16x4_t) __builtin_aarch64_uqmovnv4si ((int32x4_t) __a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovn_u64 (uint64x2_t __a)
 {
   return (uint32x2_t) __builtin_aarch64_uqmovnv2di ((int64x2_t) __a);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovnh_s16 (int16_t __a)
 {
   return (int8_t) __builtin_aarch64_sqmovnhi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovns_s32 (int32_t __a)
 {
   return (int16_t) __builtin_aarch64_sqmovnsi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovnd_s64 (int64_t __a)
 {
   return (int32_t) __builtin_aarch64_sqmovndi (__a);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovnh_u16 (uint16_t __a)
 {
   return (uint8_t) __builtin_aarch64_uqmovnhi (__a);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovns_u32 (uint32_t __a)
 {
   return (uint16_t) __builtin_aarch64_uqmovnsi (__a);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovnd_u64 (uint64_t __a)
 {
   return (uint32_t) __builtin_aarch64_uqmovndi (__a);
@@ -19713,37 +22053,43 @@ vqmovnd_u64 (uint64_t __a)
 
 /* vqmovun */
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovun_s16 (int16x8_t __a)
 {
   return (uint8x8_t) __builtin_aarch64_sqmovunv8hi (__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovun_s32 (int32x4_t __a)
 {
   return (uint16x4_t) __builtin_aarch64_sqmovunv4si (__a);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovun_s64 (int64x2_t __a)
 {
   return (uint32x2_t) __builtin_aarch64_sqmovunv2di (__a);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovunh_s16 (int16_t __a)
 {
   return (int8_t) __builtin_aarch64_sqmovunhi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovuns_s32 (int32_t __a)
 {
   return (int16_t) __builtin_aarch64_sqmovunsi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqmovund_s64 (int64_t __a)
 {
   return (int32_t) __builtin_aarch64_sqmovundi (__a);
@@ -19751,31 +22097,36 @@ vqmovund_s64 (int64_t __a)
 
 /* vqneg */
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqnegq_s64 (int64x2_t __a)
 {
   return (int64x2_t) __builtin_aarch64_sqnegv2di (__a);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqnegb_s8 (int8_t __a)
 {
   return (int8_t) __builtin_aarch64_sqnegqi (__a);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqnegh_s16 (int16_t __a)
 {
   return (int16_t) __builtin_aarch64_sqneghi (__a);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqnegs_s32 (int32_t __a)
 {
   return (int32_t) __builtin_aarch64_sqnegsi (__a);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqnegd_s64 (int64_t __a)
 {
   return __builtin_aarch64_sqnegdi (__a);
@@ -19783,61 +22134,71 @@ vqnegd_s64 (int64_t __a)
 
 /* vqrdmulh */
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulh_lane_s16 (int16x4_t __a, int16x4_t __b, const int __c)
 {
   return  __builtin_aarch64_sqrdmulh_lanev4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulh_lane_s32 (int32x2_t __a, int32x2_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_lanev2si (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhq_lane_s16 (int16x8_t __a, int16x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_lanev8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhq_lane_s32 (int32x4_t __a, int32x2_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_lanev4si (__a, __b, __c);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhh_s16 (int16_t __a, int16_t __b)
 {
   return (int16_t) __builtin_aarch64_sqrdmulhhi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhh_lane_s16 (int16_t __a, int16x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_lanehi (__a, __b, __c);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhh_laneq_s16 (int16_t __a, int16x8_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_laneqhi (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhs_s32 (int32_t __a, int32_t __b)
 {
   return (int32_t) __builtin_aarch64_sqrdmulhsi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhs_lane_s32 (int32_t __a, int32x2_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_lanesi (__a, __b, __c);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrdmulhs_laneq_s32 (int32_t __a, int32x4_t __b, const int __c)
 {
   return __builtin_aarch64_sqrdmulh_laneqsi (__a, __b, __c);
@@ -19845,145 +22206,169 @@ vqrdmulhs_laneq_s32 (int32_t __a, int32x4_t __b, const int __c)
 
 /* vqrshl */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshl_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_sqrshlv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshl_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_sqrshlv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshl_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_sqrshlv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshl_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (int64x1_t) {__builtin_aarch64_sqrshldi (__a[0], __b[0])};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshl_u8 (uint8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_uqrshlv8qi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshl_u16 (uint16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_uqrshlv4hi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshl_u32 (uint32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_uqrshlv2si_uus ( __a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshl_u64 (uint64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) {__builtin_aarch64_uqrshldi_uus (__a[0], __b[0])};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_sqrshlv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_sqrshlv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_sqrshlv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __builtin_aarch64_sqrshlv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlq_u8 (uint8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_uqrshlv16qi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlq_u16 (uint16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_uqrshlv8hi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlq_u32 (uint32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_uqrshlv4si_uus ( __a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlq_u64 (uint64x2_t __a, int64x2_t __b)
 {
   return __builtin_aarch64_uqrshlv2di_uus ( __a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlb_s8 (int8_t __a, int8_t __b)
 {
   return __builtin_aarch64_sqrshlqi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlh_s16 (int16_t __a, int16_t __b)
 {
   return __builtin_aarch64_sqrshlhi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshls_s32 (int32_t __a, int32_t __b)
 {
   return __builtin_aarch64_sqrshlsi (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshld_s64 (int64_t __a, int64_t __b)
 {
   return __builtin_aarch64_sqrshldi (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlb_u8 (uint8_t __a, uint8_t __b)
 {
   return __builtin_aarch64_uqrshlqi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshlh_u16 (uint16_t __a, uint16_t __b)
 {
   return __builtin_aarch64_uqrshlhi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshls_u32 (uint32_t __a, uint32_t __b)
 {
   return __builtin_aarch64_uqrshlsi_uus (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshld_u64 (uint64_t __a, uint64_t __b)
 {
   return __builtin_aarch64_uqrshldi_uus (__a, __b);
@@ -19991,73 +22376,85 @@ vqrshld_u64 (uint64_t __a, uint64_t __b)
 
 /* vqrshrn */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrn_n_s16 (int16x8_t __a, const int __b)
 {
   return (int8x8_t) __builtin_aarch64_sqrshrn_nv8hi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrn_n_s32 (int32x4_t __a, const int __b)
 {
   return (int16x4_t) __builtin_aarch64_sqrshrn_nv4si (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrn_n_s64 (int64x2_t __a, const int __b)
 {
   return (int32x2_t) __builtin_aarch64_sqrshrn_nv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrn_n_u16 (uint16x8_t __a, const int __b)
 {
   return __builtin_aarch64_uqrshrn_nv8hi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrn_n_u32 (uint32x4_t __a, const int __b)
 {
   return __builtin_aarch64_uqrshrn_nv4si_uus ( __a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrn_n_u64 (uint64x2_t __a, const int __b)
 {
   return __builtin_aarch64_uqrshrn_nv2di_uus ( __a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrnh_n_s16 (int16_t __a, const int __b)
 {
   return (int8_t) __builtin_aarch64_sqrshrn_nhi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrns_n_s32 (int32_t __a, const int __b)
 {
   return (int16_t) __builtin_aarch64_sqrshrn_nsi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrnd_n_s64 (int64_t __a, const int __b)
 {
   return (int32_t) __builtin_aarch64_sqrshrn_ndi (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrnh_n_u16 (uint16_t __a, const int __b)
 {
   return __builtin_aarch64_uqrshrn_nhi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrns_n_u32 (uint32_t __a, const int __b)
 {
   return __builtin_aarch64_uqrshrn_nsi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrnd_n_u64 (uint64_t __a, const int __b)
 {
   return __builtin_aarch64_uqrshrn_ndi_uus (__a, __b);
@@ -20065,37 +22462,43 @@ vqrshrnd_n_u64 (uint64_t __a, const int __b)
 
 /* vqrshrun */
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrun_n_s16 (int16x8_t __a, const int __b)
 {
   return (uint8x8_t) __builtin_aarch64_sqrshrun_nv8hi (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrun_n_s32 (int32x4_t __a, const int __b)
 {
   return (uint16x4_t) __builtin_aarch64_sqrshrun_nv4si (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrun_n_s64 (int64x2_t __a, const int __b)
 {
   return (uint32x2_t) __builtin_aarch64_sqrshrun_nv2di (__a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrunh_n_s16 (int16_t __a, const int __b)
 {
   return (int8_t) __builtin_aarch64_sqrshrun_nhi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshruns_n_s32 (int32_t __a, const int __b)
 {
   return (int16_t) __builtin_aarch64_sqrshrun_nsi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqrshrund_n_s64 (int64_t __a, const int __b)
 {
   return (int32_t) __builtin_aarch64_sqrshrun_ndi (__a, __b);
@@ -20103,289 +22506,337 @@ vqrshrund_n_s64 (int64_t __a, const int __b)
 
 /* vqshl */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_sqshlv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_sqshlv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_sqshlv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (int64x1_t) {__builtin_aarch64_sqshldi (__a[0], __b[0])};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_u8 (uint8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_uqshlv8qi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_u16 (uint16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_uqshlv4hi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_u32 (uint32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_uqshlv2si_uus ( __a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_u64 (uint64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) {__builtin_aarch64_uqshldi_uus (__a[0], __b[0])};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_sqshlv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_sqshlv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_sqshlv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __builtin_aarch64_sqshlv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_u8 (uint8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_uqshlv16qi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_u16 (uint16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_uqshlv8hi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_u32 (uint32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_uqshlv4si_uus ( __a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_u64 (uint64x2_t __a, int64x2_t __b)
 {
   return __builtin_aarch64_uqshlv2di_uus ( __a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlb_s8 (int8_t __a, int8_t __b)
 {
   return __builtin_aarch64_sqshlqi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlh_s16 (int16_t __a, int16_t __b)
 {
   return __builtin_aarch64_sqshlhi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshls_s32 (int32_t __a, int32_t __b)
 {
   return __builtin_aarch64_sqshlsi (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshld_s64 (int64_t __a, int64_t __b)
 {
   return __builtin_aarch64_sqshldi (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlb_u8 (uint8_t __a, uint8_t __b)
 {
   return __builtin_aarch64_uqshlqi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlh_u16 (uint16_t __a, uint16_t __b)
 {
   return __builtin_aarch64_uqshlhi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshls_u32 (uint32_t __a, uint32_t __b)
 {
   return __builtin_aarch64_uqshlsi_uus (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshld_u64 (uint64_t __a, uint64_t __b)
 {
   return __builtin_aarch64_uqshldi_uus (__a, __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_n_s8 (int8x8_t __a, const int __b)
 {
   return (int8x8_t) __builtin_aarch64_sqshl_nv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_n_s16 (int16x4_t __a, const int __b)
 {
   return (int16x4_t) __builtin_aarch64_sqshl_nv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_n_s32 (int32x2_t __a, const int __b)
 {
   return (int32x2_t) __builtin_aarch64_sqshl_nv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_n_s64 (int64x1_t __a, const int __b)
 {
   return (int64x1_t) {__builtin_aarch64_sqshl_ndi (__a[0], __b)};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_n_u8 (uint8x8_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nv8qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_n_u16 (uint16x4_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nv4hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_n_u32 (uint32x2_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nv2si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshl_n_u64 (uint64x1_t __a, const int __b)
 {
   return (uint64x1_t) {__builtin_aarch64_uqshl_ndi_uus (__a[0], __b)};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_n_s8 (int8x16_t __a, const int __b)
 {
   return (int8x16_t) __builtin_aarch64_sqshl_nv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_n_s16 (int16x8_t __a, const int __b)
 {
   return (int16x8_t) __builtin_aarch64_sqshl_nv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_n_s32 (int32x4_t __a, const int __b)
 {
   return (int32x4_t) __builtin_aarch64_sqshl_nv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_n_s64 (int64x2_t __a, const int __b)
 {
   return (int64x2_t) __builtin_aarch64_sqshl_nv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_n_u8 (uint8x16_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nv16qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_n_u16 (uint16x8_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nv8hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_n_u32 (uint32x4_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nv4si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlq_n_u64 (uint64x2_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nv2di_uus (__a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlb_n_s8 (int8_t __a, const int __b)
 {
   return (int8_t) __builtin_aarch64_sqshl_nqi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlh_n_s16 (int16_t __a, const int __b)
 {
   return (int16_t) __builtin_aarch64_sqshl_nhi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshls_n_s32 (int32_t __a, const int __b)
 {
   return (int32_t) __builtin_aarch64_sqshl_nsi (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshld_n_s64 (int64_t __a, const int __b)
 {
   return __builtin_aarch64_sqshl_ndi (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlb_n_u8 (uint8_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nqi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlh_n_u16 (uint16_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nhi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshls_n_u32 (uint32_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_nsi_uus (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshld_n_u64 (uint64_t __a, const int __b)
 {
   return __builtin_aarch64_uqshl_ndi_uus (__a, __b);
@@ -20393,73 +22844,85 @@ vqshld_n_u64 (uint64_t __a, const int __b)
 
 /* vqshlu */
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlu_n_s8 (int8x8_t __a, const int __b)
 {
   return __builtin_aarch64_sqshlu_nv8qi_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlu_n_s16 (int16x4_t __a, const int __b)
 {
   return __builtin_aarch64_sqshlu_nv4hi_uss (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlu_n_s32 (int32x2_t __a, const int __b)
 {
   return __builtin_aarch64_sqshlu_nv2si_uss (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlu_n_s64 (int64x1_t __a, const int __b)
 {
   return (uint64x1_t) {__builtin_aarch64_sqshlu_ndi_uss (__a[0], __b)};
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshluq_n_s8 (int8x16_t __a, const int __b)
 {
   return __builtin_aarch64_sqshlu_nv16qi_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshluq_n_s16 (int16x8_t __a, const int __b)
 {
   return __builtin_aarch64_sqshlu_nv8hi_uss (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshluq_n_s32 (int32x4_t __a, const int __b)
 {
   return __builtin_aarch64_sqshlu_nv4si_uss (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshluq_n_s64 (int64x2_t __a, const int __b)
 {
   return __builtin_aarch64_sqshlu_nv2di_uss (__a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlub_n_s8 (int8_t __a, const int __b)
 {
   return (int8_t) __builtin_aarch64_sqshlu_nqi_uss (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshluh_n_s16 (int16_t __a, const int __b)
 {
   return (int16_t) __builtin_aarch64_sqshlu_nhi_uss (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlus_n_s32 (int32_t __a, const int __b)
 {
   return (int32_t) __builtin_aarch64_sqshlu_nsi_uss (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshlud_n_s64 (int64_t __a, const int __b)
 {
   return __builtin_aarch64_sqshlu_ndi_uss (__a, __b);
@@ -20467,73 +22930,85 @@ vqshlud_n_s64 (int64_t __a, const int __b)
 
 /* vqshrn */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrn_n_s16 (int16x8_t __a, const int __b)
 {
   return (int8x8_t) __builtin_aarch64_sqshrn_nv8hi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrn_n_s32 (int32x4_t __a, const int __b)
 {
   return (int16x4_t) __builtin_aarch64_sqshrn_nv4si (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrn_n_s64 (int64x2_t __a, const int __b)
 {
   return (int32x2_t) __builtin_aarch64_sqshrn_nv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrn_n_u16 (uint16x8_t __a, const int __b)
 {
   return __builtin_aarch64_uqshrn_nv8hi_uus ( __a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrn_n_u32 (uint32x4_t __a, const int __b)
 {
   return __builtin_aarch64_uqshrn_nv4si_uus ( __a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrn_n_u64 (uint64x2_t __a, const int __b)
 {
   return __builtin_aarch64_uqshrn_nv2di_uus ( __a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrnh_n_s16 (int16_t __a, const int __b)
 {
   return (int8_t) __builtin_aarch64_sqshrn_nhi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrns_n_s32 (int32_t __a, const int __b)
 {
   return (int16_t) __builtin_aarch64_sqshrn_nsi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrnd_n_s64 (int64_t __a, const int __b)
 {
   return (int32_t) __builtin_aarch64_sqshrn_ndi (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrnh_n_u16 (uint16_t __a, const int __b)
 {
   return __builtin_aarch64_uqshrn_nhi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrns_n_u32 (uint32_t __a, const int __b)
 {
   return __builtin_aarch64_uqshrn_nsi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrnd_n_u64 (uint64_t __a, const int __b)
 {
   return __builtin_aarch64_uqshrn_ndi_uus (__a, __b);
@@ -20541,37 +23016,43 @@ vqshrnd_n_u64 (uint64_t __a, const int __b)
 
 /* vqshrun */
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrun_n_s16 (int16x8_t __a, const int __b)
 {
   return (uint8x8_t) __builtin_aarch64_sqshrun_nv8hi (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrun_n_s32 (int32x4_t __a, const int __b)
 {
   return (uint16x4_t) __builtin_aarch64_sqshrun_nv4si (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrun_n_s64 (int64x2_t __a, const int __b)
 {
   return (uint32x2_t) __builtin_aarch64_sqshrun_nv2di (__a, __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrunh_n_s16 (int16_t __a, const int __b)
 {
   return (int8_t) __builtin_aarch64_sqshrun_nhi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshruns_n_s32 (int32_t __a, const int __b)
 {
   return (int16_t) __builtin_aarch64_sqshrun_nsi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqshrund_n_s64 (int64_t __a, const int __b)
 {
   return (int32_t) __builtin_aarch64_sqshrun_ndi (__a, __b);
@@ -20579,49 +23060,57 @@ vqshrund_n_s64 (int64_t __a, const int __b)
 
 /* vqsub */
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubb_s8 (int8_t __a, int8_t __b)
 {
   return (int8_t) __builtin_aarch64_sqsubqi (__a, __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubh_s16 (int16_t __a, int16_t __b)
 {
   return (int16_t) __builtin_aarch64_sqsubhi (__a, __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubs_s32 (int32_t __a, int32_t __b)
 {
   return (int32_t) __builtin_aarch64_sqsubsi (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubd_s64 (int64_t __a, int64_t __b)
 {
   return __builtin_aarch64_sqsubdi (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubb_u8 (uint8_t __a, uint8_t __b)
 {
   return (uint8_t) __builtin_aarch64_uqsubqi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubh_u16 (uint16_t __a, uint16_t __b)
 {
   return (uint16_t) __builtin_aarch64_uqsubhi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubs_u32 (uint32_t __a, uint32_t __b)
 {
   return (uint32_t) __builtin_aarch64_uqsubsi_uuu (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqsubd_u64 (uint64_t __a, uint64_t __b)
 {
   return __builtin_aarch64_uqsubdi_uuu (__a, __b);
@@ -20629,7 +23118,8 @@ vqsubd_u64 (uint64_t __a, uint64_t __b)
 
 /* vqtbl2 */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl2_s8 (int8x16x2_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20638,7 +23128,8 @@ vqtbl2_s8 (int8x16x2_t tab, uint8x8_t idx)
   return __builtin_aarch64_tbl3v8qi (__o, (int8x8_t)idx);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl2_u8 (uint8x16x2_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20647,7 +23138,8 @@ vqtbl2_u8 (uint8x16x2_t tab, uint8x8_t idx)
   return (uint8x8_t)__builtin_aarch64_tbl3v8qi (__o, (int8x8_t)idx);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl2_p8 (poly8x16x2_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20656,7 +23148,8 @@ vqtbl2_p8 (poly8x16x2_t tab, uint8x8_t idx)
   return (poly8x8_t)__builtin_aarch64_tbl3v8qi (__o, (int8x8_t)idx);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl2q_s8 (int8x16x2_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20665,7 +23158,8 @@ vqtbl2q_s8 (int8x16x2_t tab, uint8x16_t idx)
   return __builtin_aarch64_tbl3v16qi (__o, (int8x16_t)idx);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl2q_u8 (uint8x16x2_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20674,7 +23168,8 @@ vqtbl2q_u8 (uint8x16x2_t tab, uint8x16_t idx)
   return (uint8x16_t)__builtin_aarch64_tbl3v16qi (__o, (int8x16_t)idx);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl2q_p8 (poly8x16x2_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20685,7 +23180,8 @@ vqtbl2q_p8 (poly8x16x2_t tab, uint8x16_t idx)
 
 /* vqtbl3 */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl3_s8 (int8x16x3_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20695,7 +23191,8 @@ vqtbl3_s8 (int8x16x3_t tab, uint8x8_t idx)
   return __builtin_aarch64_qtbl3v8qi (__o, (int8x8_t)idx);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl3_u8 (uint8x16x3_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20705,7 +23202,8 @@ vqtbl3_u8 (uint8x16x3_t tab, uint8x8_t idx)
   return (uint8x8_t)__builtin_aarch64_qtbl3v8qi (__o, (int8x8_t)idx);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl3_p8 (poly8x16x3_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20715,7 +23213,8 @@ vqtbl3_p8 (poly8x16x3_t tab, uint8x8_t idx)
   return (poly8x8_t)__builtin_aarch64_qtbl3v8qi (__o, (int8x8_t)idx);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl3q_s8 (int8x16x3_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20725,7 +23224,8 @@ vqtbl3q_s8 (int8x16x3_t tab, uint8x16_t idx)
   return __builtin_aarch64_qtbl3v16qi (__o, (int8x16_t)idx);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl3q_u8 (uint8x16x3_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20735,7 +23235,8 @@ vqtbl3q_u8 (uint8x16x3_t tab, uint8x16_t idx)
   return (uint8x16_t)__builtin_aarch64_qtbl3v16qi (__o, (int8x16_t)idx);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl3q_p8 (poly8x16x3_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20747,7 +23248,8 @@ vqtbl3q_p8 (poly8x16x3_t tab, uint8x16_t idx)
 
 /* vqtbl4 */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl4_s8 (int8x16x4_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20758,7 +23260,8 @@ vqtbl4_s8 (int8x16x4_t tab, uint8x8_t idx)
   return __builtin_aarch64_qtbl4v8qi (__o, (int8x8_t)idx);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl4_u8 (uint8x16x4_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20769,7 +23272,8 @@ vqtbl4_u8 (uint8x16x4_t tab, uint8x8_t idx)
   return (uint8x8_t)__builtin_aarch64_qtbl4v8qi (__o, (int8x8_t)idx);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl4_p8 (poly8x16x4_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20780,7 +23284,8 @@ vqtbl4_p8 (poly8x16x4_t tab, uint8x8_t idx)
   return (poly8x8_t)__builtin_aarch64_qtbl4v8qi (__o, (int8x8_t)idx);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl4q_s8 (int8x16x4_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20791,7 +23296,8 @@ vqtbl4q_s8 (int8x16x4_t tab, uint8x16_t idx)
   return __builtin_aarch64_qtbl4v16qi (__o, (int8x16_t)idx);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl4q_u8 (uint8x16x4_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20802,7 +23308,8 @@ vqtbl4q_u8 (uint8x16x4_t tab, uint8x16_t idx)
   return (uint8x16_t)__builtin_aarch64_qtbl4v16qi (__o, (int8x16_t)idx);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbl4q_p8 (poly8x16x4_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20815,7 +23322,8 @@ vqtbl4q_p8 (poly8x16x4_t tab, uint8x16_t idx)
 
 
 /* vqtbx2 */
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx2_s8 (int8x8_t r, int8x16x2_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20824,7 +23332,8 @@ vqtbx2_s8 (int8x8_t r, int8x16x2_t tab, uint8x8_t idx)
   return __builtin_aarch64_tbx4v8qi (r, __o, (int8x8_t)idx);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx2_u8 (uint8x8_t r, uint8x16x2_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20834,7 +23343,8 @@ vqtbx2_u8 (uint8x8_t r, uint8x16x2_t tab, uint8x8_t idx)
                                                (int8x8_t)idx);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx2_p8 (poly8x8_t r, poly8x16x2_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20844,7 +23354,8 @@ vqtbx2_p8 (poly8x8_t r, poly8x16x2_t tab, uint8x8_t idx)
                                                (int8x8_t)idx);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx2q_s8 (int8x16_t r, int8x16x2_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20853,7 +23364,8 @@ vqtbx2q_s8 (int8x16_t r, int8x16x2_t tab, uint8x16_t idx)
   return __builtin_aarch64_tbx4v16qi (r, __o, (int8x16_t)idx);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx2q_u8 (uint8x16_t r, uint8x16x2_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20863,7 +23375,8 @@ vqtbx2q_u8 (uint8x16_t r, uint8x16x2_t tab, uint8x16_t idx)
                                                  (int8x16_t)idx);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx2q_p8 (poly8x16_t r, poly8x16x2_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_oi __o;
@@ -20874,7 +23387,8 @@ vqtbx2q_p8 (poly8x16_t r, poly8x16x2_t tab, uint8x16_t idx)
 }
 
 /* vqtbx3 */
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx3_s8 (int8x8_t r, int8x16x3_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20884,7 +23398,8 @@ vqtbx3_s8 (int8x8_t r, int8x16x3_t tab, uint8x8_t idx)
   return __builtin_aarch64_qtbx3v8qi (r, __o, (int8x8_t)idx);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx3_u8 (uint8x8_t r, uint8x16x3_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20895,7 +23410,8 @@ vqtbx3_u8 (uint8x8_t r, uint8x16x3_t tab, uint8x8_t idx)
                                                 (int8x8_t)idx);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx3_p8 (poly8x8_t r, poly8x16x3_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20906,7 +23422,8 @@ vqtbx3_p8 (poly8x8_t r, poly8x16x3_t tab, uint8x8_t idx)
                                                 (int8x8_t)idx);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx3q_s8 (int8x16_t r, int8x16x3_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20916,7 +23433,8 @@ vqtbx3q_s8 (int8x16_t r, int8x16x3_t tab, uint8x16_t idx)
   return __builtin_aarch64_qtbx3v16qi (r, __o, (int8x16_t)idx);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx3q_u8 (uint8x16_t r, uint8x16x3_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20927,7 +23445,8 @@ vqtbx3q_u8 (uint8x16_t r, uint8x16x3_t tab, uint8x16_t idx)
                                                   (int8x16_t)idx);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx3q_p8 (poly8x16_t r, poly8x16x3_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_ci __o;
@@ -20940,7 +23459,8 @@ vqtbx3q_p8 (poly8x16_t r, poly8x16x3_t tab, uint8x16_t idx)
 
 /* vqtbx4 */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx4_s8 (int8x8_t r, int8x16x4_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20951,7 +23471,8 @@ vqtbx4_s8 (int8x8_t r, int8x16x4_t tab, uint8x8_t idx)
   return __builtin_aarch64_qtbx4v8qi (r, __o, (int8x8_t)idx);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx4_u8 (uint8x8_t r, uint8x16x4_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20963,7 +23484,8 @@ vqtbx4_u8 (uint8x8_t r, uint8x16x4_t tab, uint8x8_t idx)
                                                 (int8x8_t)idx);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx4_p8 (poly8x8_t r, poly8x16x4_t tab, uint8x8_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20975,7 +23497,8 @@ vqtbx4_p8 (poly8x8_t r, poly8x16x4_t tab, uint8x8_t idx)
                                                 (int8x8_t)idx);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx4q_s8 (int8x16_t r, int8x16x4_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20986,7 +23509,8 @@ vqtbx4q_s8 (int8x16_t r, int8x16x4_t tab, uint8x16_t idx)
   return __builtin_aarch64_qtbx4v16qi (r, __o, (int8x16_t)idx);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx4q_u8 (uint8x16_t r, uint8x16x4_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -20998,7 +23522,8 @@ vqtbx4q_u8 (uint8x16_t r, uint8x16x4_t tab, uint8x16_t idx)
                                                   (int8x16_t)idx);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vqtbx4q_p8 (poly8x16_t r, poly8x16x4_t tab, uint8x16_t idx)
 {
   __builtin_aarch64_simd_xi __o;
@@ -21012,37 +23537,43 @@ vqtbx4q_p8 (poly8x16_t r, poly8x16x4_t tab, uint8x16_t idx)
 
 /* vrbit  */
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrbit_p8 (poly8x8_t __a)
 {
   return (poly8x8_t) __builtin_aarch64_rbitv8qi ((int8x8_t) __a);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrbit_s8 (int8x8_t __a)
 {
   return __builtin_aarch64_rbitv8qi (__a);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrbit_u8 (uint8x8_t __a)
 {
   return (uint8x8_t) __builtin_aarch64_rbitv8qi ((int8x8_t) __a);
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrbitq_p8 (poly8x16_t __a)
 {
   return (poly8x16_t) __builtin_aarch64_rbitv16qi ((int8x16_t)__a);
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrbitq_s8 (int8x16_t __a)
 {
   return __builtin_aarch64_rbitv16qi (__a);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrbitq_u8 (uint8x16_t __a)
 {
   return (uint8x16_t) __builtin_aarch64_rbitv16qi ((int8x16_t) __a);
@@ -21050,49 +23581,57 @@ vrbitq_u8 (uint8x16_t __a)
 
 /* vrecpe  */
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpe_u32 (uint32x2_t __a)
 {
   return (uint32x2_t) __builtin_aarch64_urecpev2si ((int32x2_t) __a);
 }
- 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpeq_u32 (uint32x4_t __a)
 {
   return (uint32x4_t) __builtin_aarch64_urecpev4si ((int32x4_t) __a);
 }
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpes_f32 (float32_t __a)
 {
   return __builtin_aarch64_frecpesf (__a);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecped_f64 (float64_t __a)
 {
   return __builtin_aarch64_frecpedf (__a);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpe_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_frecpev2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpe_f64 (float64x1_t __a)
 {
   return (float64x1_t) { vrecped_f64 (vget_lane_f64 (__a, 0)) };
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpeq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_frecpev4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpeq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_frecpev2df (__a);
@@ -21100,38 +23639,44 @@ vrecpeq_f64 (float64x2_t __a)
 
 /* vrecps  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpss_f32 (float32_t __a, float32_t __b)
 {
   return __builtin_aarch64_frecpssf (__a, __b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpsd_f64 (float64_t __a, float64_t __b)
 {
   return __builtin_aarch64_frecpsdf (__a, __b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecps_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __builtin_aarch64_frecpsv2sf (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecps_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (float64x1_t) { vrecpsd_f64 (vget_lane_f64 (__a, 0),
                                      vget_lane_f64 (__b, 0)) };
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpsq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_frecpsv4sf (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpsq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_frecpsv2df (__a, __b);
@@ -21139,13 +23684,15 @@ vrecpsq_f64 (float64x2_t __a, float64x2_t __b)
 
 /* vrecpx  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpxs_f32 (float32_t __a)
 {
   return __builtin_aarch64_frecpxsf (__a);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpxd_f64 (float64_t __a)
 {
   return __builtin_aarch64_frecpxdf (__a);
@@ -21154,238 +23701,276 @@ vrecpxd_f64 (float64_t __a)
 
 /* vrev  */
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev16_p8 (poly8x8_t a)
 {
   return __builtin_shuffle (a, (uint8x8_t) { 1, 0, 3, 2, 5, 4, 7, 6 });
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev16_s8 (int8x8_t a)
 {
   return __builtin_shuffle (a, (uint8x8_t) { 1, 0, 3, 2, 5, 4, 7, 6 });
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev16_u8 (uint8x8_t a)
 {
   return __builtin_shuffle (a, (uint8x8_t) { 1, 0, 3, 2, 5, 4, 7, 6 });
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev16q_p8 (poly8x16_t a)
 {
   return __builtin_shuffle (a,
       (uint8x16_t) { 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14 });
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev16q_s8 (int8x16_t a)
 {
   return __builtin_shuffle (a,
       (uint8x16_t) { 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14 });
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev16q_u8 (uint8x16_t a)
 {
   return __builtin_shuffle (a,
       (uint8x16_t) { 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14 });
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32_p8 (poly8x8_t a)
 {
   return __builtin_shuffle (a, (uint8x8_t) { 3, 2, 1, 0, 7, 6, 5, 4 });
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32_p16 (poly16x4_t a)
 {
   return __builtin_shuffle (a, (uint16x4_t) { 1, 0, 3, 2 });
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32_s8 (int8x8_t a)
 {
   return __builtin_shuffle (a, (uint8x8_t) { 3, 2, 1, 0, 7, 6, 5, 4 });
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32_s16 (int16x4_t a)
 {
   return __builtin_shuffle (a, (uint16x4_t) { 1, 0, 3, 2 });
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32_u8 (uint8x8_t a)
 {
   return __builtin_shuffle (a, (uint8x8_t) { 3, 2, 1, 0, 7, 6, 5, 4 });
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32_u16 (uint16x4_t a)
 {
   return __builtin_shuffle (a, (uint16x4_t) { 1, 0, 3, 2 });
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32q_p8 (poly8x16_t a)
 {
   return __builtin_shuffle (a,
       (uint8x16_t) { 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12 });
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32q_p16 (poly16x8_t a)
 {
   return __builtin_shuffle (a, (uint16x8_t) { 1, 0, 3, 2, 5, 4, 7, 6 });
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32q_s8 (int8x16_t a)
 {
   return __builtin_shuffle (a,
       (uint8x16_t) { 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12 });
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32q_s16 (int16x8_t a)
 {
   return __builtin_shuffle (a, (uint16x8_t) { 1, 0, 3, 2, 5, 4, 7, 6 });
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32q_u8 (uint8x16_t a)
 {
   return __builtin_shuffle (a,
       (uint8x16_t) { 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12 });
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev32q_u16 (uint16x8_t a)
 {
   return __builtin_shuffle (a, (uint16x8_t) { 1, 0, 3, 2, 5, 4, 7, 6 });
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_f16 (float16x4_t __a)
 {
   return __builtin_shuffle (__a, (uint16x4_t) { 3, 2, 1, 0 });
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_f32 (float32x2_t a)
 {
   return __builtin_shuffle (a, (uint32x2_t) { 1, 0 });
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_p8 (poly8x8_t a)
 {
   return __builtin_shuffle (a, (uint8x8_t) { 7, 6, 5, 4, 3, 2, 1, 0 });
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_p16 (poly16x4_t a)
 {
   return __builtin_shuffle (a, (uint16x4_t) { 3, 2, 1, 0 });
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_s8 (int8x8_t a)
 {
   return __builtin_shuffle (a, (uint8x8_t) { 7, 6, 5, 4, 3, 2, 1, 0 });
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_s16 (int16x4_t a)
 {
   return __builtin_shuffle (a, (uint16x4_t) { 3, 2, 1, 0 });
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_s32 (int32x2_t a)
 {
   return __builtin_shuffle (a, (uint32x2_t) { 1, 0 });
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_u8 (uint8x8_t a)
 {
   return __builtin_shuffle (a, (uint8x8_t) { 7, 6, 5, 4, 3, 2, 1, 0 });
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_u16 (uint16x4_t a)
 {
   return __builtin_shuffle (a, (uint16x4_t) { 3, 2, 1, 0 });
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64_u32 (uint32x2_t a)
 {
   return __builtin_shuffle (a, (uint32x2_t) { 1, 0 });
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_f16 (float16x8_t __a)
 {
   return __builtin_shuffle (__a, (uint16x8_t) { 3, 2, 1, 0, 7, 6, 5, 4 });
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_f32 (float32x4_t a)
 {
   return __builtin_shuffle (a, (uint32x4_t) { 1, 0, 3, 2 });
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_p8 (poly8x16_t a)
 {
   return __builtin_shuffle (a,
       (uint8x16_t) { 7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8 });
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_p16 (poly16x8_t a)
 {
   return __builtin_shuffle (a, (uint16x8_t) { 3, 2, 1, 0, 7, 6, 5, 4 });
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_s8 (int8x16_t a)
 {
   return __builtin_shuffle (a,
       (uint8x16_t) { 7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8 });
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_s16 (int16x8_t a)
 {
   return __builtin_shuffle (a, (uint16x8_t) { 3, 2, 1, 0, 7, 6, 5, 4 });
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_s32 (int32x4_t a)
 {
   return __builtin_shuffle (a, (uint32x4_t) { 1, 0, 3, 2 });
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_u8 (uint8x16_t a)
 {
   return __builtin_shuffle (a,
       (uint8x16_t) { 7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8 });
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_u16 (uint16x8_t a)
 {
   return __builtin_shuffle (a, (uint16x8_t) { 3, 2, 1, 0, 7, 6, 5, 4 });
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrev64q_u32 (uint32x4_t a)
 {
   return __builtin_shuffle (a, (uint32x4_t) { 1, 0, 3, 2 });
@@ -21393,25 +23978,29 @@ vrev64q_u32 (uint32x4_t a)
 
 /* vrnd  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrnd_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_btruncv2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrnd_f64 (float64x1_t __a)
 {
   return vset_lane_f64 (__builtin_trunc (vget_lane_f64 (__a, 0)), __a, 0);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_btruncv4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_btruncv2df (__a);
@@ -21419,25 +24008,29 @@ vrndq_f64 (float64x2_t __a)
 
 /* vrnda  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrnda_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_roundv2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrnda_f64 (float64x1_t __a)
 {
   return vset_lane_f64 (__builtin_round (vget_lane_f64 (__a, 0)), __a, 0);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndaq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_roundv4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndaq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_roundv2df (__a);
@@ -21445,25 +24038,29 @@ vrndaq_f64 (float64x2_t __a)
 
 /* vrndi  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndi_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_nearbyintv2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndi_f64 (float64x1_t __a)
 {
   return vset_lane_f64 (__builtin_nearbyint (vget_lane_f64 (__a, 0)), __a, 0);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndiq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_nearbyintv4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndiq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_nearbyintv2df (__a);
@@ -21471,25 +24068,29 @@ vrndiq_f64 (float64x2_t __a)
 
 /* vrndm  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndm_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_floorv2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndm_f64 (float64x1_t __a)
 {
   return vset_lane_f64 (__builtin_floor (vget_lane_f64 (__a, 0)), __a, 0);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndmq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_floorv4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndmq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_floorv2df (__a);
@@ -21497,25 +24098,29 @@ vrndmq_f64 (float64x2_t __a)
 
 /* vrndn  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndn_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_frintnv2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndn_f64 (float64x1_t __a)
 {
   return (float64x1_t) {__builtin_aarch64_frintndf (__a[0])};
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndnq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_frintnv4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndnq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_frintnv2df (__a);
@@ -21523,25 +24128,29 @@ vrndnq_f64 (float64x2_t __a)
 
 /* vrndp  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndp_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_ceilv2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndp_f64 (float64x1_t __a)
 {
   return vset_lane_f64 (__builtin_ceil (vget_lane_f64 (__a, 0)), __a, 0);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndpq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_ceilv4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndpq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_ceilv2df (__a);
@@ -21549,25 +24158,29 @@ vrndpq_f64 (float64x2_t __a)
 
 /* vrndx  */
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndx_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_rintv2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndx_f64 (float64x1_t __a)
 {
   return vset_lane_f64 (__builtin_rint (vget_lane_f64 (__a, 0)), __a, 0);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndxq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_rintv4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndxq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_rintv2df (__a);
@@ -21575,109 +24188,127 @@ vrndxq_f64 (float64x2_t __a)
 
 /* vrshl */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshl_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (int8x8_t) __builtin_aarch64_srshlv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshl_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (int16x4_t) __builtin_aarch64_srshlv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshl_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (int32x2_t) __builtin_aarch64_srshlv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshl_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (int64x1_t) {__builtin_aarch64_srshldi (__a[0], __b[0])};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshl_u8 (uint8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_urshlv8qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshl_u16 (uint16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_urshlv4hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshl_u32 (uint32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_urshlv2si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshl_u64 (uint64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) {__builtin_aarch64_urshldi_uus (__a[0], __b[0])};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshlq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (int8x16_t) __builtin_aarch64_srshlv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshlq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (int16x8_t) __builtin_aarch64_srshlv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshlq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (int32x4_t) __builtin_aarch64_srshlv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshlq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (int64x2_t) __builtin_aarch64_srshlv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshlq_u8 (uint8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_urshlv16qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshlq_u16 (uint16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_urshlv8hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshlq_u32 (uint32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_urshlv4si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshlq_u64 (uint64x2_t __a, int64x2_t __b)
 {
   return __builtin_aarch64_urshlv2di_uus (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshld_s64 (int64_t __a, int64_t __b)
 {
   return __builtin_aarch64_srshldi (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshld_u64 (uint64_t __a, int64_t __b)
 {
   return __builtin_aarch64_urshldi_uus (__a, __b);
@@ -21685,109 +24316,127 @@ vrshld_u64 (uint64_t __a, int64_t __b)
 
 /* vrshr */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshr_n_s8 (int8x8_t __a, const int __b)
 {
   return (int8x8_t) __builtin_aarch64_srshr_nv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshr_n_s16 (int16x4_t __a, const int __b)
 {
   return (int16x4_t) __builtin_aarch64_srshr_nv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshr_n_s32 (int32x2_t __a, const int __b)
 {
   return (int32x2_t) __builtin_aarch64_srshr_nv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshr_n_s64 (int64x1_t __a, const int __b)
 {
   return (int64x1_t) {__builtin_aarch64_srshr_ndi (__a[0], __b)};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshr_n_u8 (uint8x8_t __a, const int __b)
 {
   return __builtin_aarch64_urshr_nv8qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshr_n_u16 (uint16x4_t __a, const int __b)
 {
   return __builtin_aarch64_urshr_nv4hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshr_n_u32 (uint32x2_t __a, const int __b)
 {
   return __builtin_aarch64_urshr_nv2si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshr_n_u64 (uint64x1_t __a, const int __b)
 {
   return (uint64x1_t) {__builtin_aarch64_urshr_ndi_uus (__a[0], __b)};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrq_n_s8 (int8x16_t __a, const int __b)
 {
   return (int8x16_t) __builtin_aarch64_srshr_nv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrq_n_s16 (int16x8_t __a, const int __b)
 {
   return (int16x8_t) __builtin_aarch64_srshr_nv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrq_n_s32 (int32x4_t __a, const int __b)
 {
   return (int32x4_t) __builtin_aarch64_srshr_nv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrq_n_s64 (int64x2_t __a, const int __b)
 {
   return (int64x2_t) __builtin_aarch64_srshr_nv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrq_n_u8 (uint8x16_t __a, const int __b)
 {
   return __builtin_aarch64_urshr_nv16qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrq_n_u16 (uint16x8_t __a, const int __b)
 {
   return __builtin_aarch64_urshr_nv8hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrq_n_u32 (uint32x4_t __a, const int __b)
 {
   return __builtin_aarch64_urshr_nv4si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrq_n_u64 (uint64x2_t __a, const int __b)
 {
   return __builtin_aarch64_urshr_nv2di_uus (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrd_n_s64 (int64_t __a, const int __b)
 {
   return __builtin_aarch64_srshr_ndi (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrshrd_n_u64 (uint64_t __a, const int __b)
 {
   return __builtin_aarch64_urshr_ndi_uus (__a, __b);
@@ -21795,37 +24444,43 @@ vrshrd_n_u64 (uint64_t __a, const int __b)
 
 /* vrsqrte.  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrtes_f32 (float32_t __a)
 {
   return __builtin_aarch64_rsqrtesf (__a);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrted_f64 (float64_t __a)
 {
   return __builtin_aarch64_rsqrtedf (__a);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrte_f32 (float32x2_t __a)
 {
   return __builtin_aarch64_rsqrtev2sf (__a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrte_f64 (float64x1_t __a)
 {
   return (float64x1_t) {vrsqrted_f64 (vget_lane_f64 (__a, 0))};
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrteq_f32 (float32x4_t __a)
 {
   return __builtin_aarch64_rsqrtev4sf (__a);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrteq_f64 (float64x2_t __a)
 {
   return __builtin_aarch64_rsqrtev2df (__a);
@@ -21833,38 +24488,44 @@ vrsqrteq_f64 (float64x2_t __a)
 
 /* vrsqrts.  */
 
-__extension__ static __inline float32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrtss_f32 (float32_t __a, float32_t __b)
 {
   return __builtin_aarch64_rsqrtssf (__a, __b);
 }
 
-__extension__ static __inline float64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrtsd_f64 (float64_t __a, float64_t __b)
 {
   return __builtin_aarch64_rsqrtsdf (__a, __b);
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrts_f32 (float32x2_t __a, float32x2_t __b)
 {
   return __builtin_aarch64_rsqrtsv2sf (__a, __b);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrts_f64 (float64x1_t __a, float64x1_t __b)
 {
   return (float64x1_t) {vrsqrtsd_f64 (vget_lane_f64 (__a, 0),
                                      vget_lane_f64 (__b, 0))};
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrtsq_f32 (float32x4_t __a, float32x4_t __b)
 {
   return __builtin_aarch64_rsqrtsv4sf (__a, __b);
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrtsq_f64 (float64x2_t __a, float64x2_t __b)
 {
   return __builtin_aarch64_rsqrtsv2df (__a, __b);
@@ -21872,109 +24533,127 @@ vrsqrtsq_f64 (float64x2_t __a, float64x2_t __b)
 
 /* vrsra */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsra_n_s8 (int8x8_t __a, int8x8_t __b, const int __c)
 {
   return (int8x8_t) __builtin_aarch64_srsra_nv8qi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsra_n_s16 (int16x4_t __a, int16x4_t __b, const int __c)
 {
   return (int16x4_t) __builtin_aarch64_srsra_nv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsra_n_s32 (int32x2_t __a, int32x2_t __b, const int __c)
 {
   return (int32x2_t) __builtin_aarch64_srsra_nv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsra_n_s64 (int64x1_t __a, int64x1_t __b, const int __c)
 {
   return (int64x1_t) {__builtin_aarch64_srsra_ndi (__a[0], __b[0], __c)};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsra_n_u8 (uint8x8_t __a, uint8x8_t __b, const int __c)
 {
   return __builtin_aarch64_ursra_nv8qi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsra_n_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
 {
   return __builtin_aarch64_ursra_nv4hi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsra_n_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
 {
   return __builtin_aarch64_ursra_nv2si_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsra_n_u64 (uint64x1_t __a, uint64x1_t __b, const int __c)
 {
   return (uint64x1_t) {__builtin_aarch64_ursra_ndi_uuus (__a[0], __b[0], __c)};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsraq_n_s8 (int8x16_t __a, int8x16_t __b, const int __c)
 {
   return (int8x16_t) __builtin_aarch64_srsra_nv16qi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsraq_n_s16 (int16x8_t __a, int16x8_t __b, const int __c)
 {
   return (int16x8_t) __builtin_aarch64_srsra_nv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsraq_n_s32 (int32x4_t __a, int32x4_t __b, const int __c)
 {
   return (int32x4_t) __builtin_aarch64_srsra_nv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsraq_n_s64 (int64x2_t __a, int64x2_t __b, const int __c)
 {
   return (int64x2_t) __builtin_aarch64_srsra_nv2di (__a, __b, __c);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsraq_n_u8 (uint8x16_t __a, uint8x16_t __b, const int __c)
 {
   return __builtin_aarch64_ursra_nv16qi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsraq_n_u16 (uint16x8_t __a, uint16x8_t __b, const int __c)
 {
   return __builtin_aarch64_ursra_nv8hi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsraq_n_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
 {
   return __builtin_aarch64_ursra_nv4si_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsraq_n_u64 (uint64x2_t __a, uint64x2_t __b, const int __c)
 {
   return __builtin_aarch64_ursra_nv2di_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsrad_n_s64 (int64_t __a, int64_t __b, const int __c)
 {
   return __builtin_aarch64_srsra_ndi (__a, __b, __c);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsrad_n_u64 (uint64_t __a, uint64_t __b, const int __c)
 {
   return __builtin_aarch64_ursra_ndi_uuus (__a, __b, __c);
@@ -21985,74 +24664,86 @@ vrsrad_n_u64 (uint64_t __a, uint64_t __b, const int __c)
 
 /* vsha1  */
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha1cq_u32 (uint32x4_t hash_abcd, uint32_t hash_e, uint32x4_t wk)
 {
   return __builtin_aarch64_crypto_sha1cv4si_uuuu (hash_abcd, hash_e, wk);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha1mq_u32 (uint32x4_t hash_abcd, uint32_t hash_e, uint32x4_t wk)
 {
   return __builtin_aarch64_crypto_sha1mv4si_uuuu (hash_abcd, hash_e, wk);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha1pq_u32 (uint32x4_t hash_abcd, uint32_t hash_e, uint32x4_t wk)
 {
   return __builtin_aarch64_crypto_sha1pv4si_uuuu (hash_abcd, hash_e, wk);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha1h_u32 (uint32_t hash_e)
 {
   return __builtin_aarch64_crypto_sha1hsi_uu (hash_e);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha1su0q_u32 (uint32x4_t w0_3, uint32x4_t w4_7, uint32x4_t w8_11)
 {
   return __builtin_aarch64_crypto_sha1su0v4si_uuuu (w0_3, w4_7, w8_11);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha1su1q_u32 (uint32x4_t tw0_3, uint32x4_t w12_15)
 {
   return __builtin_aarch64_crypto_sha1su1v4si_uuu (tw0_3, w12_15);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha256hq_u32 (uint32x4_t hash_abcd, uint32x4_t hash_efgh, uint32x4_t wk)
 {
   return __builtin_aarch64_crypto_sha256hv4si_uuuu (hash_abcd, hash_efgh, wk);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha256h2q_u32 (uint32x4_t hash_efgh, uint32x4_t hash_abcd, uint32x4_t wk)
 {
   return __builtin_aarch64_crypto_sha256h2v4si_uuuu (hash_efgh, hash_abcd, wk);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha256su0q_u32 (uint32x4_t w0_3, uint32x4_t w4_7)
 {
   return __builtin_aarch64_crypto_sha256su0v4si_uuu (w0_3, w4_7);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsha256su1q_u32 (uint32x4_t tw0_3, uint32x4_t w8_11, uint32x4_t w12_15)
 {
   return __builtin_aarch64_crypto_sha256su1v4si_uuuu (tw0_3, w8_11, w12_15);
 }
 
-__extension__ static __inline poly128_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly128_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_p64 (poly64_t a, poly64_t b)
 {
   return
     __builtin_aarch64_crypto_pmulldi_ppp (a, b);
 }
 
-__extension__ static __inline poly128_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly128_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmull_high_p64 (poly64x2_t a, poly64x2_t b)
 {
   return __builtin_aarch64_crypto_pmullv2di_ppp (a, b);
@@ -22062,289 +24753,337 @@ vmull_high_p64 (poly64x2_t a, poly64x2_t b)
 
 /* vshl */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_n_s8 (int8x8_t __a, const int __b)
 {
   return (int8x8_t) __builtin_aarch64_ashlv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_n_s16 (int16x4_t __a, const int __b)
 {
   return (int16x4_t) __builtin_aarch64_ashlv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_n_s32 (int32x2_t __a, const int __b)
 {
   return (int32x2_t) __builtin_aarch64_ashlv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_n_s64 (int64x1_t __a, const int __b)
 {
   return (int64x1_t) {__builtin_aarch64_ashldi (__a[0], __b)};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_n_u8 (uint8x8_t __a, const int __b)
 {
   return (uint8x8_t) __builtin_aarch64_ashlv8qi ((int8x8_t) __a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_n_u16 (uint16x4_t __a, const int __b)
 {
   return (uint16x4_t) __builtin_aarch64_ashlv4hi ((int16x4_t) __a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_n_u32 (uint32x2_t __a, const int __b)
 {
   return (uint32x2_t) __builtin_aarch64_ashlv2si ((int32x2_t) __a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_n_u64 (uint64x1_t __a, const int __b)
 {
   return (uint64x1_t) {__builtin_aarch64_ashldi ((int64_t) __a[0], __b)};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_n_s8 (int8x16_t __a, const int __b)
 {
   return (int8x16_t) __builtin_aarch64_ashlv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_n_s16 (int16x8_t __a, const int __b)
 {
   return (int16x8_t) __builtin_aarch64_ashlv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_n_s32 (int32x4_t __a, const int __b)
 {
   return (int32x4_t) __builtin_aarch64_ashlv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_n_s64 (int64x2_t __a, const int __b)
 {
   return (int64x2_t) __builtin_aarch64_ashlv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_n_u8 (uint8x16_t __a, const int __b)
 {
   return (uint8x16_t) __builtin_aarch64_ashlv16qi ((int8x16_t) __a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_n_u16 (uint16x8_t __a, const int __b)
 {
   return (uint16x8_t) __builtin_aarch64_ashlv8hi ((int16x8_t) __a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_n_u32 (uint32x4_t __a, const int __b)
 {
   return (uint32x4_t) __builtin_aarch64_ashlv4si ((int32x4_t) __a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_n_u64 (uint64x2_t __a, const int __b)
 {
   return (uint64x2_t) __builtin_aarch64_ashlv2di ((int64x2_t) __a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshld_n_s64 (int64_t __a, const int __b)
 {
   return __builtin_aarch64_ashldi (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshld_n_u64 (uint64_t __a, const int __b)
 {
   return (uint64_t) __builtin_aarch64_ashldi (__a, __b);
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_s8 (int8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_sshlv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_s16 (int16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_sshlv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_s32 (int32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_sshlv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (int64x1_t) {__builtin_aarch64_sshldi (__a[0], __b[0])};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_u8 (uint8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_ushlv8qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_u16 (uint16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_ushlv4hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_u32 (uint32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_ushlv2si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshl_u64 (uint64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) {__builtin_aarch64_ushldi_uus (__a[0], __b[0])};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_sshlv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_sshlv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_sshlv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return __builtin_aarch64_sshlv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_u8 (uint8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_ushlv16qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_u16 (uint16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_ushlv8hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_u32 (uint32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_ushlv4si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshlq_u64 (uint64x2_t __a, int64x2_t __b)
 {
   return __builtin_aarch64_ushlv2di_uus (__a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshld_s64 (int64_t __a, int64_t __b)
 {
   return __builtin_aarch64_sshldi (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshld_u64 (uint64_t __a, uint64_t __b)
 {
   return __builtin_aarch64_ushldi_uus (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_high_n_s8 (int8x16_t __a, const int __b)
 {
   return __builtin_aarch64_sshll2_nv16qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_high_n_s16 (int16x8_t __a, const int __b)
 {
   return __builtin_aarch64_sshll2_nv8hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_high_n_s32 (int32x4_t __a, const int __b)
 {
   return __builtin_aarch64_sshll2_nv4si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_high_n_u8 (uint8x16_t __a, const int __b)
 {
   return (uint16x8_t) __builtin_aarch64_ushll2_nv16qi ((int8x16_t) __a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_high_n_u16 (uint16x8_t __a, const int __b)
 {
   return (uint32x4_t) __builtin_aarch64_ushll2_nv8hi ((int16x8_t) __a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_high_n_u32 (uint32x4_t __a, const int __b)
 {
   return (uint64x2_t) __builtin_aarch64_ushll2_nv4si ((int32x4_t) __a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_n_s8 (int8x8_t __a, const int __b)
 {
   return __builtin_aarch64_sshll_nv8qi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_n_s16 (int16x4_t __a, const int __b)
 {
   return __builtin_aarch64_sshll_nv4hi (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_n_s32 (int32x2_t __a, const int __b)
 {
   return __builtin_aarch64_sshll_nv2si (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_n_u8 (uint8x8_t __a, const int __b)
 {
   return __builtin_aarch64_ushll_nv8qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_n_u16 (uint16x4_t __a, const int __b)
 {
   return __builtin_aarch64_ushll_nv4hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshll_n_u32 (uint32x2_t __a, const int __b)
 {
   return __builtin_aarch64_ushll_nv2si_uus (__a, __b);
@@ -22352,109 +25091,127 @@ vshll_n_u32 (uint32x2_t __a, const int __b)
 
 /* vshr */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshr_n_s8 (int8x8_t __a, const int __b)
 {
   return (int8x8_t) __builtin_aarch64_ashrv8qi (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshr_n_s16 (int16x4_t __a, const int __b)
 {
   return (int16x4_t) __builtin_aarch64_ashrv4hi (__a, __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshr_n_s32 (int32x2_t __a, const int __b)
 {
   return (int32x2_t) __builtin_aarch64_ashrv2si (__a, __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshr_n_s64 (int64x1_t __a, const int __b)
 {
   return (int64x1_t) {__builtin_aarch64_ashr_simddi (__a[0], __b)};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshr_n_u8 (uint8x8_t __a, const int __b)
 {
   return (uint8x8_t) __builtin_aarch64_lshrv8qi ((int8x8_t) __a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshr_n_u16 (uint16x4_t __a, const int __b)
 {
   return (uint16x4_t) __builtin_aarch64_lshrv4hi ((int16x4_t) __a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshr_n_u32 (uint32x2_t __a, const int __b)
 {
   return (uint32x2_t) __builtin_aarch64_lshrv2si ((int32x2_t) __a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshr_n_u64 (uint64x1_t __a, const int __b)
 {
   return (uint64x1_t) {__builtin_aarch64_lshr_simddi_uus ( __a[0], __b)};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrq_n_s8 (int8x16_t __a, const int __b)
 {
   return (int8x16_t) __builtin_aarch64_ashrv16qi (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrq_n_s16 (int16x8_t __a, const int __b)
 {
   return (int16x8_t) __builtin_aarch64_ashrv8hi (__a, __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrq_n_s32 (int32x4_t __a, const int __b)
 {
   return (int32x4_t) __builtin_aarch64_ashrv4si (__a, __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrq_n_s64 (int64x2_t __a, const int __b)
 {
   return (int64x2_t) __builtin_aarch64_ashrv2di (__a, __b);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrq_n_u8 (uint8x16_t __a, const int __b)
 {
   return (uint8x16_t) __builtin_aarch64_lshrv16qi ((int8x16_t) __a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrq_n_u16 (uint16x8_t __a, const int __b)
 {
   return (uint16x8_t) __builtin_aarch64_lshrv8hi ((int16x8_t) __a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrq_n_u32 (uint32x4_t __a, const int __b)
 {
   return (uint32x4_t) __builtin_aarch64_lshrv4si ((int32x4_t) __a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrq_n_u64 (uint64x2_t __a, const int __b)
 {
   return (uint64x2_t) __builtin_aarch64_lshrv2di ((int64x2_t) __a, __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrd_n_s64 (int64_t __a, const int __b)
 {
   return __builtin_aarch64_ashr_simddi (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vshrd_n_u64 (uint64_t __a, const int __b)
 {
   return __builtin_aarch64_lshr_simddi_uus (__a, __b);
@@ -22462,109 +25219,127 @@ vshrd_n_u64 (uint64_t __a, const int __b)
 
 /* vsli */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsli_n_s8 (int8x8_t __a, int8x8_t __b, const int __c)
 {
   return (int8x8_t) __builtin_aarch64_ssli_nv8qi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsli_n_s16 (int16x4_t __a, int16x4_t __b, const int __c)
 {
   return (int16x4_t) __builtin_aarch64_ssli_nv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsli_n_s32 (int32x2_t __a, int32x2_t __b, const int __c)
 {
   return (int32x2_t) __builtin_aarch64_ssli_nv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsli_n_s64 (int64x1_t __a, int64x1_t __b, const int __c)
 {
   return (int64x1_t) {__builtin_aarch64_ssli_ndi (__a[0], __b[0], __c)};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsli_n_u8 (uint8x8_t __a, uint8x8_t __b, const int __c)
 {
   return __builtin_aarch64_usli_nv8qi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsli_n_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
 {
   return __builtin_aarch64_usli_nv4hi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsli_n_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
 {
   return __builtin_aarch64_usli_nv2si_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsli_n_u64 (uint64x1_t __a, uint64x1_t __b, const int __c)
 {
   return (uint64x1_t) {__builtin_aarch64_usli_ndi_uuus (__a[0], __b[0], __c)};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsliq_n_s8 (int8x16_t __a, int8x16_t __b, const int __c)
 {
   return (int8x16_t) __builtin_aarch64_ssli_nv16qi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsliq_n_s16 (int16x8_t __a, int16x8_t __b, const int __c)
 {
   return (int16x8_t) __builtin_aarch64_ssli_nv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsliq_n_s32 (int32x4_t __a, int32x4_t __b, const int __c)
 {
   return (int32x4_t) __builtin_aarch64_ssli_nv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsliq_n_s64 (int64x2_t __a, int64x2_t __b, const int __c)
 {
   return (int64x2_t) __builtin_aarch64_ssli_nv2di (__a, __b, __c);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsliq_n_u8 (uint8x16_t __a, uint8x16_t __b, const int __c)
 {
   return __builtin_aarch64_usli_nv16qi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsliq_n_u16 (uint16x8_t __a, uint16x8_t __b, const int __c)
 {
   return __builtin_aarch64_usli_nv8hi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsliq_n_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
 {
   return __builtin_aarch64_usli_nv4si_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsliq_n_u64 (uint64x2_t __a, uint64x2_t __b, const int __c)
 {
   return __builtin_aarch64_usli_nv2di_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vslid_n_s64 (int64_t __a, int64_t __b, const int __c)
 {
   return __builtin_aarch64_ssli_ndi (__a, __b, __c);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vslid_n_u64 (uint64_t __a, uint64_t __b, const int __c)
 {
   return __builtin_aarch64_usli_ndi_uuus (__a, __b, __c);
@@ -22572,98 +25347,114 @@ vslid_n_u64 (uint64_t __a, uint64_t __b, const int __c)
 
 /* vsqadd */
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqadd_u8 (uint8x8_t __a, int8x8_t __b)
 {
   return __builtin_aarch64_usqaddv8qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqadd_u16 (uint16x4_t __a, int16x4_t __b)
 {
   return __builtin_aarch64_usqaddv4hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqadd_u32 (uint32x2_t __a, int32x2_t __b)
 {
   return __builtin_aarch64_usqaddv2si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqadd_u64 (uint64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) {__builtin_aarch64_usqadddi_uus (__a[0], __b[0])};
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqaddq_u8 (uint8x16_t __a, int8x16_t __b)
 {
   return __builtin_aarch64_usqaddv16qi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqaddq_u16 (uint16x8_t __a, int16x8_t __b)
 {
   return __builtin_aarch64_usqaddv8hi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqaddq_u32 (uint32x4_t __a, int32x4_t __b)
 {
   return __builtin_aarch64_usqaddv4si_uus (__a, __b);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqaddq_u64 (uint64x2_t __a, int64x2_t __b)
 {
   return __builtin_aarch64_usqaddv2di_uus (__a, __b);
 }
 
-__extension__ static __inline uint8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqaddb_u8 (uint8_t __a, int8_t __b)
 {
   return __builtin_aarch64_usqaddqi_uus (__a, __b);
 }
 
-__extension__ static __inline uint16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqaddh_u16 (uint16_t __a, int16_t __b)
 {
   return __builtin_aarch64_usqaddhi_uus (__a, __b);
 }
 
-__extension__ static __inline uint32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqadds_u32 (uint32_t __a, int32_t __b)
 {
   return __builtin_aarch64_usqaddsi_uus (__a, __b);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqaddd_u64 (uint64_t __a, int64_t __b)
 {
   return __builtin_aarch64_usqadddi_uus (__a, __b);
 }
 
 /* vsqrt */
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqrt_f32 (float32x2_t a)
 {
   return __builtin_aarch64_sqrtv2sf (a);
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqrtq_f32 (float32x4_t a)
 {
   return __builtin_aarch64_sqrtv4sf (a);
 }
 
-__extension__ static __inline float64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqrt_f64 (float64x1_t a)
 {
   return (float64x1_t) { __builtin_aarch64_sqrtdf (a[0]) };
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqrtq_f64 (float64x2_t a)
 {
   return __builtin_aarch64_sqrtv2df (a);
@@ -22671,109 +25462,127 @@ vsqrtq_f64 (float64x2_t a)
 
 /* vsra */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsra_n_s8 (int8x8_t __a, int8x8_t __b, const int __c)
 {
   return (int8x8_t) __builtin_aarch64_ssra_nv8qi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsra_n_s16 (int16x4_t __a, int16x4_t __b, const int __c)
 {
   return (int16x4_t) __builtin_aarch64_ssra_nv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsra_n_s32 (int32x2_t __a, int32x2_t __b, const int __c)
 {
   return (int32x2_t) __builtin_aarch64_ssra_nv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsra_n_s64 (int64x1_t __a, int64x1_t __b, const int __c)
 {
   return (int64x1_t) {__builtin_aarch64_ssra_ndi (__a[0], __b[0], __c)};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsra_n_u8 (uint8x8_t __a, uint8x8_t __b, const int __c)
 {
   return __builtin_aarch64_usra_nv8qi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsra_n_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
 {
   return __builtin_aarch64_usra_nv4hi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsra_n_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
 {
   return __builtin_aarch64_usra_nv2si_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsra_n_u64 (uint64x1_t __a, uint64x1_t __b, const int __c)
 {
   return (uint64x1_t) {__builtin_aarch64_usra_ndi_uuus (__a[0], __b[0], __c)};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsraq_n_s8 (int8x16_t __a, int8x16_t __b, const int __c)
 {
   return (int8x16_t) __builtin_aarch64_ssra_nv16qi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsraq_n_s16 (int16x8_t __a, int16x8_t __b, const int __c)
 {
   return (int16x8_t) __builtin_aarch64_ssra_nv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsraq_n_s32 (int32x4_t __a, int32x4_t __b, const int __c)
 {
   return (int32x4_t) __builtin_aarch64_ssra_nv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsraq_n_s64 (int64x2_t __a, int64x2_t __b, const int __c)
 {
   return (int64x2_t) __builtin_aarch64_ssra_nv2di (__a, __b, __c);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsraq_n_u8 (uint8x16_t __a, uint8x16_t __b, const int __c)
 {
   return __builtin_aarch64_usra_nv16qi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsraq_n_u16 (uint16x8_t __a, uint16x8_t __b, const int __c)
 {
   return __builtin_aarch64_usra_nv8hi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsraq_n_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
 {
   return __builtin_aarch64_usra_nv4si_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsraq_n_u64 (uint64x2_t __a, uint64x2_t __b, const int __c)
 {
   return __builtin_aarch64_usra_nv2di_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsrad_n_s64 (int64_t __a, int64_t __b, const int __c)
 {
   return __builtin_aarch64_ssra_ndi (__a, __b, __c);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsrad_n_u64 (uint64_t __a, uint64_t __b, const int __c)
 {
   return __builtin_aarch64_usra_ndi_uuus (__a, __b, __c);
@@ -22781,109 +25590,127 @@ vsrad_n_u64 (uint64_t __a, uint64_t __b, const int __c)
 
 /* vsri */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsri_n_s8 (int8x8_t __a, int8x8_t __b, const int __c)
 {
   return (int8x8_t) __builtin_aarch64_ssri_nv8qi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsri_n_s16 (int16x4_t __a, int16x4_t __b, const int __c)
 {
   return (int16x4_t) __builtin_aarch64_ssri_nv4hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsri_n_s32 (int32x2_t __a, int32x2_t __b, const int __c)
 {
   return (int32x2_t) __builtin_aarch64_ssri_nv2si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsri_n_s64 (int64x1_t __a, int64x1_t __b, const int __c)
 {
   return (int64x1_t) {__builtin_aarch64_ssri_ndi (__a[0], __b[0], __c)};
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsri_n_u8 (uint8x8_t __a, uint8x8_t __b, const int __c)
 {
   return __builtin_aarch64_usri_nv8qi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsri_n_u16 (uint16x4_t __a, uint16x4_t __b, const int __c)
 {
   return __builtin_aarch64_usri_nv4hi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsri_n_u32 (uint32x2_t __a, uint32x2_t __b, const int __c)
 {
   return __builtin_aarch64_usri_nv2si_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsri_n_u64 (uint64x1_t __a, uint64x1_t __b, const int __c)
 {
   return (uint64x1_t) {__builtin_aarch64_usri_ndi_uuus (__a[0], __b[0], __c)};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsriq_n_s8 (int8x16_t __a, int8x16_t __b, const int __c)
 {
   return (int8x16_t) __builtin_aarch64_ssri_nv16qi (__a, __b, __c);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsriq_n_s16 (int16x8_t __a, int16x8_t __b, const int __c)
 {
   return (int16x8_t) __builtin_aarch64_ssri_nv8hi (__a, __b, __c);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsriq_n_s32 (int32x4_t __a, int32x4_t __b, const int __c)
 {
   return (int32x4_t) __builtin_aarch64_ssri_nv4si (__a, __b, __c);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsriq_n_s64 (int64x2_t __a, int64x2_t __b, const int __c)
 {
   return (int64x2_t) __builtin_aarch64_ssri_nv2di (__a, __b, __c);
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsriq_n_u8 (uint8x16_t __a, uint8x16_t __b, const int __c)
 {
   return __builtin_aarch64_usri_nv16qi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsriq_n_u16 (uint16x8_t __a, uint16x8_t __b, const int __c)
 {
   return __builtin_aarch64_usri_nv8hi_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsriq_n_u32 (uint32x4_t __a, uint32x4_t __b, const int __c)
 {
   return __builtin_aarch64_usri_nv4si_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsriq_n_u64 (uint64x2_t __a, uint64x2_t __b, const int __c)
 {
   return __builtin_aarch64_usri_nv2di_uuus (__a, __b, __c);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsrid_n_s64 (int64_t __a, int64_t __b, const int __c)
 {
   return __builtin_aarch64_ssri_ndi (__a, __b, __c);
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsrid_n_u64 (uint64_t __a, uint64_t __b, const int __c)
 {
   return __builtin_aarch64_usri_ndi_uuus (__a, __b, __c);
@@ -22891,84 +25718,97 @@ vsrid_n_u64 (uint64_t __a, uint64_t __b, const int __c)
 
 /* vst1 */
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_f16 (float16_t *__a, float16x4_t __b)
 {
   __builtin_aarch64_st1v4hf (__a, __b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_f32 (float32_t *a, float32x2_t b)
 {
   __builtin_aarch64_st1v2sf ((__builtin_aarch64_simd_sf *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_f64 (float64_t *a, float64x1_t b)
 {
   *a = b[0];
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_p8 (poly8_t *a, poly8x8_t b)
 {
   __builtin_aarch64_st1v8qi ((__builtin_aarch64_simd_qi *) a,
                             (int8x8_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_p16 (poly16_t *a, poly16x4_t b)
 {
   __builtin_aarch64_st1v4hi ((__builtin_aarch64_simd_hi *) a,
                             (int16x4_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_s8 (int8_t *a, int8x8_t b)
 {
   __builtin_aarch64_st1v8qi ((__builtin_aarch64_simd_qi *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_s16 (int16_t *a, int16x4_t b)
 {
   __builtin_aarch64_st1v4hi ((__builtin_aarch64_simd_hi *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_s32 (int32_t *a, int32x2_t b)
 {
   __builtin_aarch64_st1v2si ((__builtin_aarch64_simd_si *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_s64 (int64_t *a, int64x1_t b)
 {
   *a = b[0];
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_u8 (uint8_t *a, uint8x8_t b)
 {
   __builtin_aarch64_st1v8qi ((__builtin_aarch64_simd_qi *) a,
                             (int8x8_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_u16 (uint16_t *a, uint16x4_t b)
 {
   __builtin_aarch64_st1v4hi ((__builtin_aarch64_simd_hi *) a,
                             (int16x4_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_u32 (uint32_t *a, uint32x2_t b)
 {
   __builtin_aarch64_st1v2si ((__builtin_aarch64_simd_si *) a,
                             (int32x2_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_u64 (uint64_t *a, uint64x1_t b)
 {
   *a = b[0];
@@ -22976,84 +25816,97 @@ vst1_u64 (uint64_t *a, uint64x1_t b)
 
 /* vst1q */
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_f16 (float16_t *__a, float16x8_t __b)
 {
   __builtin_aarch64_st1v8hf (__a, __b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_f32 (float32_t *a, float32x4_t b)
 {
   __builtin_aarch64_st1v4sf ((__builtin_aarch64_simd_sf *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_f64 (float64_t *a, float64x2_t b)
 {
   __builtin_aarch64_st1v2df ((__builtin_aarch64_simd_df *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_p8 (poly8_t *a, poly8x16_t b)
 {
   __builtin_aarch64_st1v16qi ((__builtin_aarch64_simd_qi *) a,
                              (int8x16_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_p16 (poly16_t *a, poly16x8_t b)
 {
   __builtin_aarch64_st1v8hi ((__builtin_aarch64_simd_hi *) a,
                             (int16x8_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_s8 (int8_t *a, int8x16_t b)
 {
   __builtin_aarch64_st1v16qi ((__builtin_aarch64_simd_qi *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_s16 (int16_t *a, int16x8_t b)
 {
   __builtin_aarch64_st1v8hi ((__builtin_aarch64_simd_hi *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_s32 (int32_t *a, int32x4_t b)
 {
   __builtin_aarch64_st1v4si ((__builtin_aarch64_simd_si *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_s64 (int64_t *a, int64x2_t b)
 {
   __builtin_aarch64_st1v2di ((__builtin_aarch64_simd_di *) a, b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_u8 (uint8_t *a, uint8x16_t b)
 {
   __builtin_aarch64_st1v16qi ((__builtin_aarch64_simd_qi *) a,
                              (int8x16_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_u16 (uint16_t *a, uint16x8_t b)
 {
   __builtin_aarch64_st1v8hi ((__builtin_aarch64_simd_hi *) a,
                             (int16x8_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_u32 (uint32_t *a, uint32x4_t b)
 {
   __builtin_aarch64_st1v4si ((__builtin_aarch64_simd_si *) a,
                             (int32x4_t) b);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_u64 (uint64_t *a, uint64x2_t b)
 {
   __builtin_aarch64_st1v2di ((__builtin_aarch64_simd_di *) a,
@@ -23062,79 +25915,92 @@ vst1q_u64 (uint64_t *a, uint64x2_t b)
 
 /* vst1_lane */
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_f16 (float16_t *__a, float16x4_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_f32 (float32_t *__a, float32x2_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_f64 (float64_t *__a, float64x1_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_p8 (poly8_t *__a, poly8x8_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_p16 (poly16_t *__a, poly16x4_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_s8 (int8_t *__a, int8x8_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_s16 (int16_t *__a, int16x4_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_s32 (int32_t *__a, int32x2_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_s64 (int64_t *__a, int64x1_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_u8 (uint8_t *__a, uint8x8_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_u16 (uint16_t *__a, uint16x4_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_u32 (uint32_t *__a, uint32x2_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1_lane_u64 (uint64_t *__a, uint64x1_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
@@ -23142,79 +26008,92 @@ vst1_lane_u64 (uint64_t *__a, uint64x1_t __b, const int __lane)
 
 /* vst1q_lane */
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_f16 (float16_t *__a, float16x8_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_f32 (float32_t *__a, float32x4_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_f64 (float64_t *__a, float64x2_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_p8 (poly8_t *__a, poly8x16_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_p16 (poly16_t *__a, poly16x8_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_s8 (int8_t *__a, int8x16_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_s16 (int16_t *__a, int16x8_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_s32 (int32_t *__a, int32x4_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_s64 (int64_t *__a, int64x2_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_u8 (uint8_t *__a, uint8x16_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_u16 (uint16_t *__a, uint16x8_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_u32 (uint32_t *__a, uint32x4_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_lane_u64 (uint64_t *__a, uint64x2_t __b, const int __lane)
 {
   *__a = __aarch64_vget_lane_any (__b, __lane);
@@ -23222,7 +26101,7 @@ vst1q_lane_u64 (uint64_t *__a, uint64x2_t __b, const int __lane)
 
 /* vstn */
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst2_s64 (int64_t * __a, int64x1x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23234,7 +26113,7 @@ vst2_s64 (int64_t * __a, int64x1x2_t val)
   __builtin_aarch64_st2di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst2_u64 (uint64_t * __a, uint64x1x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23246,7 +26125,7 @@ vst2_u64 (uint64_t * __a, uint64x1x2_t val)
   __builtin_aarch64_st2di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst2_f64 (float64_t * __a, float64x1x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23258,7 +26137,7 @@ vst2_f64 (float64_t * __a, float64x1x2_t val)
   __builtin_aarch64_st2df ((__builtin_aarch64_simd_df *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst2_s8 (int8_t * __a, int8x8x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23270,7 +26149,8 @@ vst2_s8 (int8_t * __a, int8x8x2_t val)
   __builtin_aarch64_st2v8qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2_p8 (poly8_t * __a, poly8x8x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23282,7 +26162,8 @@ vst2_p8 (poly8_t * __a, poly8x8x2_t val)
   __builtin_aarch64_st2v8qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2_s16 (int16_t * __a, int16x4x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23294,7 +26175,8 @@ vst2_s16 (int16_t * __a, int16x4x2_t val)
   __builtin_aarch64_st2v4hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2_p16 (poly16_t * __a, poly16x4x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23306,7 +26188,8 @@ vst2_p16 (poly16_t * __a, poly16x4x2_t val)
   __builtin_aarch64_st2v4hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2_s32 (int32_t * __a, int32x2x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23318,7 +26201,8 @@ vst2_s32 (int32_t * __a, int32x2x2_t val)
   __builtin_aarch64_st2v2si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2_u8 (uint8_t * __a, uint8x8x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23330,7 +26214,8 @@ vst2_u8 (uint8_t * __a, uint8x8x2_t val)
   __builtin_aarch64_st2v8qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2_u16 (uint16_t * __a, uint16x4x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23342,7 +26227,8 @@ vst2_u16 (uint16_t * __a, uint16x4x2_t val)
   __builtin_aarch64_st2v4hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2_u32 (uint32_t * __a, uint32x2x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23354,7 +26240,8 @@ vst2_u32 (uint32_t * __a, uint32x2x2_t val)
   __builtin_aarch64_st2v2si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2_f16 (float16_t * __a, float16x4x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23366,7 +26253,8 @@ vst2_f16 (float16_t * __a, float16x4x2_t val)
   __builtin_aarch64_st2v4hf (__a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2_f32 (float32_t * __a, float32x2x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23378,7 +26266,8 @@ vst2_f32 (float32_t * __a, float32x2x2_t val)
   __builtin_aarch64_st2v2sf ((__builtin_aarch64_simd_sf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_s8 (int8_t * __a, int8x16x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23387,7 +26276,8 @@ vst2q_s8 (int8_t * __a, int8x16x2_t val)
   __builtin_aarch64_st2v16qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_p8 (poly8_t * __a, poly8x16x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23396,7 +26286,8 @@ vst2q_p8 (poly8_t * __a, poly8x16x2_t val)
   __builtin_aarch64_st2v16qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_s16 (int16_t * __a, int16x8x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23405,7 +26296,8 @@ vst2q_s16 (int16_t * __a, int16x8x2_t val)
   __builtin_aarch64_st2v8hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_p16 (poly16_t * __a, poly16x8x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23414,7 +26306,8 @@ vst2q_p16 (poly16_t * __a, poly16x8x2_t val)
   __builtin_aarch64_st2v8hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_s32 (int32_t * __a, int32x4x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23423,7 +26316,8 @@ vst2q_s32 (int32_t * __a, int32x4x2_t val)
   __builtin_aarch64_st2v4si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_s64 (int64_t * __a, int64x2x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23432,7 +26326,8 @@ vst2q_s64 (int64_t * __a, int64x2x2_t val)
   __builtin_aarch64_st2v2di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_u8 (uint8_t * __a, uint8x16x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23441,7 +26336,8 @@ vst2q_u8 (uint8_t * __a, uint8x16x2_t val)
   __builtin_aarch64_st2v16qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_u16 (uint16_t * __a, uint16x8x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23450,7 +26346,8 @@ vst2q_u16 (uint16_t * __a, uint16x8x2_t val)
   __builtin_aarch64_st2v8hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_u32 (uint32_t * __a, uint32x4x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23459,7 +26356,8 @@ vst2q_u32 (uint32_t * __a, uint32x4x2_t val)
   __builtin_aarch64_st2v4si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_u64 (uint64_t * __a, uint64x2x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23468,7 +26366,8 @@ vst2q_u64 (uint64_t * __a, uint64x2x2_t val)
   __builtin_aarch64_st2v2di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_f16 (float16_t * __a, float16x8x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23477,7 +26376,8 @@ vst2q_f16 (float16_t * __a, float16x8x2_t val)
   __builtin_aarch64_st2v8hf (__a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_f32 (float32_t * __a, float32x4x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23486,7 +26386,8 @@ vst2q_f32 (float32_t * __a, float32x4x2_t val)
   __builtin_aarch64_st2v4sf ((__builtin_aarch64_simd_sf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_f64 (float64_t * __a, float64x2x2_t val)
 {
   __builtin_aarch64_simd_oi __o;
@@ -23495,7 +26396,7 @@ vst2q_f64 (float64_t * __a, float64x2x2_t val)
   __builtin_aarch64_st2v2df ((__builtin_aarch64_simd_df *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst3_s64 (int64_t * __a, int64x1x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23509,7 +26410,7 @@ vst3_s64 (int64_t * __a, int64x1x3_t val)
   __builtin_aarch64_st3di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst3_u64 (uint64_t * __a, uint64x1x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23523,7 +26424,7 @@ vst3_u64 (uint64_t * __a, uint64x1x3_t val)
   __builtin_aarch64_st3di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst3_f64 (float64_t * __a, float64x1x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23537,7 +26438,7 @@ vst3_f64 (float64_t * __a, float64x1x3_t val)
   __builtin_aarch64_st3df ((__builtin_aarch64_simd_df *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst3_s8 (int8_t * __a, int8x8x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23551,7 +26452,8 @@ vst3_s8 (int8_t * __a, int8x8x3_t val)
   __builtin_aarch64_st3v8qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3_p8 (poly8_t * __a, poly8x8x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23565,7 +26467,8 @@ vst3_p8 (poly8_t * __a, poly8x8x3_t val)
   __builtin_aarch64_st3v8qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3_s16 (int16_t * __a, int16x4x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23579,7 +26482,8 @@ vst3_s16 (int16_t * __a, int16x4x3_t val)
   __builtin_aarch64_st3v4hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3_p16 (poly16_t * __a, poly16x4x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23593,7 +26497,8 @@ vst3_p16 (poly16_t * __a, poly16x4x3_t val)
   __builtin_aarch64_st3v4hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3_s32 (int32_t * __a, int32x2x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23607,7 +26512,8 @@ vst3_s32 (int32_t * __a, int32x2x3_t val)
   __builtin_aarch64_st3v2si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3_u8 (uint8_t * __a, uint8x8x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23621,7 +26527,8 @@ vst3_u8 (uint8_t * __a, uint8x8x3_t val)
   __builtin_aarch64_st3v8qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3_u16 (uint16_t * __a, uint16x4x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23635,7 +26542,8 @@ vst3_u16 (uint16_t * __a, uint16x4x3_t val)
   __builtin_aarch64_st3v4hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3_u32 (uint32_t * __a, uint32x2x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23649,7 +26557,8 @@ vst3_u32 (uint32_t * __a, uint32x2x3_t val)
   __builtin_aarch64_st3v2si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3_f16 (float16_t * __a, float16x4x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23663,7 +26572,8 @@ vst3_f16 (float16_t * __a, float16x4x3_t val)
   __builtin_aarch64_st3v4hf ((__builtin_aarch64_simd_hf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3_f32 (float32_t * __a, float32x2x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23677,7 +26587,8 @@ vst3_f32 (float32_t * __a, float32x2x3_t val)
   __builtin_aarch64_st3v2sf ((__builtin_aarch64_simd_sf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_s8 (int8_t * __a, int8x16x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23687,7 +26598,8 @@ vst3q_s8 (int8_t * __a, int8x16x3_t val)
   __builtin_aarch64_st3v16qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_p8 (poly8_t * __a, poly8x16x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23697,7 +26609,8 @@ vst3q_p8 (poly8_t * __a, poly8x16x3_t val)
   __builtin_aarch64_st3v16qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_s16 (int16_t * __a, int16x8x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23707,7 +26620,8 @@ vst3q_s16 (int16_t * __a, int16x8x3_t val)
   __builtin_aarch64_st3v8hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_p16 (poly16_t * __a, poly16x8x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23717,7 +26631,8 @@ vst3q_p16 (poly16_t * __a, poly16x8x3_t val)
   __builtin_aarch64_st3v8hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_s32 (int32_t * __a, int32x4x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23727,7 +26642,8 @@ vst3q_s32 (int32_t * __a, int32x4x3_t val)
   __builtin_aarch64_st3v4si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_s64 (int64_t * __a, int64x2x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23737,7 +26653,8 @@ vst3q_s64 (int64_t * __a, int64x2x3_t val)
   __builtin_aarch64_st3v2di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_u8 (uint8_t * __a, uint8x16x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23747,7 +26664,8 @@ vst3q_u8 (uint8_t * __a, uint8x16x3_t val)
   __builtin_aarch64_st3v16qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_u16 (uint16_t * __a, uint16x8x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23757,7 +26675,8 @@ vst3q_u16 (uint16_t * __a, uint16x8x3_t val)
   __builtin_aarch64_st3v8hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_u32 (uint32_t * __a, uint32x4x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23767,7 +26686,8 @@ vst3q_u32 (uint32_t * __a, uint32x4x3_t val)
   __builtin_aarch64_st3v4si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_u64 (uint64_t * __a, uint64x2x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23777,7 +26697,8 @@ vst3q_u64 (uint64_t * __a, uint64x2x3_t val)
   __builtin_aarch64_st3v2di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_f16 (float16_t * __a, float16x8x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23787,7 +26708,8 @@ vst3q_f16 (float16_t * __a, float16x8x3_t val)
   __builtin_aarch64_st3v8hf ((__builtin_aarch64_simd_hf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_f32 (float32_t * __a, float32x4x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23797,7 +26719,8 @@ vst3q_f32 (float32_t * __a, float32x4x3_t val)
   __builtin_aarch64_st3v4sf ((__builtin_aarch64_simd_sf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_f64 (float64_t * __a, float64x2x3_t val)
 {
   __builtin_aarch64_simd_ci __o;
@@ -23807,7 +26730,7 @@ vst3q_f64 (float64_t * __a, float64x2x3_t val)
   __builtin_aarch64_st3v2df ((__builtin_aarch64_simd_df *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst4_s64 (int64_t * __a, int64x1x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23823,7 +26746,7 @@ vst4_s64 (int64_t * __a, int64x1x4_t val)
   __builtin_aarch64_st4di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst4_u64 (uint64_t * __a, uint64x1x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23839,7 +26762,7 @@ vst4_u64 (uint64_t * __a, uint64x1x4_t val)
   __builtin_aarch64_st4di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst4_f64 (float64_t * __a, float64x1x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23855,7 +26778,7 @@ vst4_f64 (float64_t * __a, float64x1x4_t val)
   __builtin_aarch64_st4df ((__builtin_aarch64_simd_df *) __a, __o);
 }
 
-__extension__ static __inline void
+__extension__ extern __inline void
 vst4_s8 (int8_t * __a, int8x8x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23871,7 +26794,8 @@ vst4_s8 (int8_t * __a, int8x8x4_t val)
   __builtin_aarch64_st4v8qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4_p8 (poly8_t * __a, poly8x8x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23887,7 +26811,8 @@ vst4_p8 (poly8_t * __a, poly8x8x4_t val)
   __builtin_aarch64_st4v8qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4_s16 (int16_t * __a, int16x4x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23903,7 +26828,8 @@ vst4_s16 (int16_t * __a, int16x4x4_t val)
   __builtin_aarch64_st4v4hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4_p16 (poly16_t * __a, poly16x4x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23919,7 +26845,8 @@ vst4_p16 (poly16_t * __a, poly16x4x4_t val)
   __builtin_aarch64_st4v4hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4_s32 (int32_t * __a, int32x2x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23935,7 +26862,8 @@ vst4_s32 (int32_t * __a, int32x2x4_t val)
   __builtin_aarch64_st4v2si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4_u8 (uint8_t * __a, uint8x8x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23951,7 +26879,8 @@ vst4_u8 (uint8_t * __a, uint8x8x4_t val)
   __builtin_aarch64_st4v8qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4_u16 (uint16_t * __a, uint16x4x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23967,7 +26896,8 @@ vst4_u16 (uint16_t * __a, uint16x4x4_t val)
   __builtin_aarch64_st4v4hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4_u32 (uint32_t * __a, uint32x2x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23983,7 +26913,8 @@ vst4_u32 (uint32_t * __a, uint32x2x4_t val)
   __builtin_aarch64_st4v2si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4_f16 (float16_t * __a, float16x4x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -23999,7 +26930,8 @@ vst4_f16 (float16_t * __a, float16x4x4_t val)
   __builtin_aarch64_st4v4hf ((__builtin_aarch64_simd_hf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4_f32 (float32_t * __a, float32x2x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24015,7 +26947,8 @@ vst4_f32 (float32_t * __a, float32x2x4_t val)
   __builtin_aarch64_st4v2sf ((__builtin_aarch64_simd_sf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_s8 (int8_t * __a, int8x16x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24026,7 +26959,8 @@ vst4q_s8 (int8_t * __a, int8x16x4_t val)
   __builtin_aarch64_st4v16qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_p8 (poly8_t * __a, poly8x16x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24037,7 +26971,8 @@ vst4q_p8 (poly8_t * __a, poly8x16x4_t val)
   __builtin_aarch64_st4v16qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_s16 (int16_t * __a, int16x8x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24048,7 +26983,8 @@ vst4q_s16 (int16_t * __a, int16x8x4_t val)
   __builtin_aarch64_st4v8hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_p16 (poly16_t * __a, poly16x8x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24059,7 +26995,8 @@ vst4q_p16 (poly16_t * __a, poly16x8x4_t val)
   __builtin_aarch64_st4v8hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_s32 (int32_t * __a, int32x4x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24070,7 +27007,8 @@ vst4q_s32 (int32_t * __a, int32x4x4_t val)
   __builtin_aarch64_st4v4si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_s64 (int64_t * __a, int64x2x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24081,7 +27019,8 @@ vst4q_s64 (int64_t * __a, int64x2x4_t val)
   __builtin_aarch64_st4v2di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_u8 (uint8_t * __a, uint8x16x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24092,7 +27031,8 @@ vst4q_u8 (uint8_t * __a, uint8x16x4_t val)
   __builtin_aarch64_st4v16qi ((__builtin_aarch64_simd_qi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_u16 (uint16_t * __a, uint16x8x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24103,7 +27043,8 @@ vst4q_u16 (uint16_t * __a, uint16x8x4_t val)
   __builtin_aarch64_st4v8hi ((__builtin_aarch64_simd_hi *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_u32 (uint32_t * __a, uint32x4x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24114,7 +27055,8 @@ vst4q_u32 (uint32_t * __a, uint32x4x4_t val)
   __builtin_aarch64_st4v4si ((__builtin_aarch64_simd_si *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_u64 (uint64_t * __a, uint64x2x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24125,7 +27067,8 @@ vst4q_u64 (uint64_t * __a, uint64x2x4_t val)
   __builtin_aarch64_st4v2di ((__builtin_aarch64_simd_di *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_f16 (float16_t * __a, float16x8x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24136,7 +27079,8 @@ vst4q_f16 (float16_t * __a, float16x8x4_t val)
   __builtin_aarch64_st4v8hf ((__builtin_aarch64_simd_hf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_f32 (float32_t * __a, float32x4x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24147,7 +27091,8 @@ vst4q_f32 (float32_t * __a, float32x4x4_t val)
   __builtin_aarch64_st4v4sf ((__builtin_aarch64_simd_sf *) __a, __o);
 }
 
-__extension__ static __inline void __attribute__ ((__always_inline__))
+__extension__ extern __inline void
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_f64 (float64_t * __a, float64x2x4_t val)
 {
   __builtin_aarch64_simd_xi __o;
@@ -24160,13 +27105,15 @@ vst4q_f64 (float64_t * __a, float64x2x4_t val)
 
 /* vsub */
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubd_s64 (int64_t __a, int64_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubd_u64 (uint64_t __a, uint64_t __b)
 {
   return __a - __b;
@@ -24174,7 +27121,8 @@ vsubd_u64 (uint64_t __a, uint64_t __b)
 
 /* vtbx1  */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx1_s8 (int8x8_t __r, int8x8_t __tab, int8x8_t __idx)
 {
   uint8x8_t __mask = vclt_u8 (vreinterpret_u8_s8 (__idx),
@@ -24184,7 +27132,8 @@ vtbx1_s8 (int8x8_t __r, int8x8_t __tab, int8x8_t __idx)
   return vbsl_s8 (__mask, __tbl, __r);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx1_u8 (uint8x8_t __r, uint8x8_t __tab, uint8x8_t __idx)
 {
   uint8x8_t __mask = vclt_u8 (__idx, vmov_n_u8 (8));
@@ -24193,7 +27142,8 @@ vtbx1_u8 (uint8x8_t __r, uint8x8_t __tab, uint8x8_t __idx)
   return vbsl_u8 (__mask, __tbl, __r);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx1_p8 (poly8x8_t __r, poly8x8_t __tab, uint8x8_t __idx)
 {
   uint8x8_t __mask = vclt_u8 (__idx, vmov_n_u8 (8));
@@ -24204,7 +27154,8 @@ vtbx1_p8 (poly8x8_t __r, poly8x8_t __tab, uint8x8_t __idx)
 
 /* vtbx3  */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx3_s8 (int8x8_t __r, int8x8x3_t __tab, int8x8_t __idx)
 {
   uint8x8_t __mask = vclt_u8 (vreinterpret_u8_s8 (__idx),
@@ -24214,7 +27165,8 @@ vtbx3_s8 (int8x8_t __r, int8x8x3_t __tab, int8x8_t __idx)
   return vbsl_s8 (__mask, __tbl, __r);
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx3_u8 (uint8x8_t __r, uint8x8x3_t __tab, uint8x8_t __idx)
 {
   uint8x8_t __mask = vclt_u8 (__idx, vmov_n_u8 (24));
@@ -24223,7 +27175,8 @@ vtbx3_u8 (uint8x8_t __r, uint8x8x3_t __tab, uint8x8_t __idx)
   return vbsl_u8 (__mask, __tbl, __r);
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx3_p8 (poly8x8_t __r, poly8x8x3_t __tab, uint8x8_t __idx)
 {
   uint8x8_t __mask = vclt_u8 (__idx, vmov_n_u8 (24));
@@ -24234,7 +27187,8 @@ vtbx3_p8 (poly8x8_t __r, poly8x8x3_t __tab, uint8x8_t __idx)
 
 /* vtbx4  */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx4_s8 (int8x8_t __r, int8x8x4_t __tab, int8x8_t __idx)
 {
   int8x8_t result;
@@ -24250,7 +27204,8 @@ vtbx4_s8 (int8x8_t __r, int8x8x4_t __tab, int8x8_t __idx)
   return result;
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx4_u8 (uint8x8_t __r, uint8x8x4_t __tab, uint8x8_t __idx)
 {
   uint8x8_t result;
@@ -24267,7 +27222,8 @@ vtbx4_u8 (uint8x8_t __r, uint8x8x4_t __tab, uint8x8_t __idx)
   return result;
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtbx4_p8 (poly8x8_t __r, poly8x8x4_t __tab, uint8x8_t __idx)
 {
   poly8x8_t result;
@@ -24286,7 +27242,8 @@ vtbx4_p8 (poly8x8_t __r, poly8x8x4_t __tab, uint8x8_t __idx)
 
 /* vtrn */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_f16 (float16x4_t __a, float16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24296,7 +27253,8 @@ vtrn1_f16 (float16x4_t __a, float16x4_t __b)
 #endif
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_f32 (float32x2_t __a, float32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24306,7 +27264,8 @@ vtrn1_f32 (float32x2_t __a, float32x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_p8 (poly8x8_t __a, poly8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24316,7 +27275,8 @@ vtrn1_p8 (poly8x8_t __a, poly8x8_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_p16 (poly16x4_t __a, poly16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24326,7 +27286,8 @@ vtrn1_p16 (poly16x4_t __a, poly16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_s8 (int8x8_t __a, int8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24336,7 +27297,8 @@ vtrn1_s8 (int8x8_t __a, int8x8_t __b)
 #endif
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_s16 (int16x4_t __a, int16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24346,7 +27308,8 @@ vtrn1_s16 (int16x4_t __a, int16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_s32 (int32x2_t __a, int32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24356,7 +27319,8 @@ vtrn1_s32 (int32x2_t __a, int32x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_u8 (uint8x8_t __a, uint8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24366,7 +27330,8 @@ vtrn1_u8 (uint8x8_t __a, uint8x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_u16 (uint16x4_t __a, uint16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24376,7 +27341,8 @@ vtrn1_u16 (uint16x4_t __a, uint16x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1_u32 (uint32x2_t __a, uint32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24386,7 +27352,8 @@ vtrn1_u32 (uint32x2_t __a, uint32x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_f16 (float16x8_t __a, float16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24396,7 +27363,8 @@ vtrn1q_f16 (float16x8_t __a, float16x8_t __b)
 #endif
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_f32 (float32x4_t __a, float32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24406,7 +27374,8 @@ vtrn1q_f32 (float32x4_t __a, float32x4_t __b)
 #endif
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_f64 (float64x2_t __a, float64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24416,7 +27385,8 @@ vtrn1q_f64 (float64x2_t __a, float64x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_p8 (poly8x16_t __a, poly8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24428,7 +27398,8 @@ vtrn1q_p8 (poly8x16_t __a, poly8x16_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_p16 (poly16x8_t __a, poly16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24438,7 +27409,8 @@ vtrn1q_p16 (poly16x8_t __a, poly16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_s8 (int8x16_t __a, int8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24450,7 +27422,8 @@ vtrn1q_s8 (int8x16_t __a, int8x16_t __b)
 #endif
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_s16 (int16x8_t __a, int16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24460,7 +27433,8 @@ vtrn1q_s16 (int16x8_t __a, int16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_s32 (int32x4_t __a, int32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24470,7 +27444,8 @@ vtrn1q_s32 (int32x4_t __a, int32x4_t __b)
 #endif
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_s64 (int64x2_t __a, int64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24480,7 +27455,8 @@ vtrn1q_s64 (int64x2_t __a, int64x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_u8 (uint8x16_t __a, uint8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24492,7 +27468,8 @@ vtrn1q_u8 (uint8x16_t __a, uint8x16_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_u16 (uint16x8_t __a, uint16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24502,7 +27479,8 @@ vtrn1q_u16 (uint16x8_t __a, uint16x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_u32 (uint32x4_t __a, uint32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24512,7 +27490,8 @@ vtrn1q_u32 (uint32x4_t __a, uint32x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn1q_u64 (uint64x2_t __a, uint64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24522,7 +27501,8 @@ vtrn1q_u64 (uint64x2_t __a, uint64x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_f16 (float16x4_t __a, float16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24532,7 +27512,8 @@ vtrn2_f16 (float16x4_t __a, float16x4_t __b)
 #endif
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_f32 (float32x2_t __a, float32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24542,7 +27523,8 @@ vtrn2_f32 (float32x2_t __a, float32x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_p8 (poly8x8_t __a, poly8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24552,7 +27534,8 @@ vtrn2_p8 (poly8x8_t __a, poly8x8_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_p16 (poly16x4_t __a, poly16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24562,7 +27545,8 @@ vtrn2_p16 (poly16x4_t __a, poly16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_s8 (int8x8_t __a, int8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24572,7 +27556,8 @@ vtrn2_s8 (int8x8_t __a, int8x8_t __b)
 #endif
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_s16 (int16x4_t __a, int16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24582,7 +27567,8 @@ vtrn2_s16 (int16x4_t __a, int16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_s32 (int32x2_t __a, int32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24592,7 +27578,8 @@ vtrn2_s32 (int32x2_t __a, int32x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_u8 (uint8x8_t __a, uint8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24602,7 +27589,8 @@ vtrn2_u8 (uint8x8_t __a, uint8x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_u16 (uint16x4_t __a, uint16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24612,7 +27600,8 @@ vtrn2_u16 (uint16x4_t __a, uint16x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2_u32 (uint32x2_t __a, uint32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24622,7 +27611,8 @@ vtrn2_u32 (uint32x2_t __a, uint32x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_f16 (float16x8_t __a, float16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24632,7 +27622,8 @@ vtrn2q_f16 (float16x8_t __a, float16x8_t __b)
 #endif
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_f32 (float32x4_t __a, float32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24642,7 +27633,8 @@ vtrn2q_f32 (float32x4_t __a, float32x4_t __b)
 #endif
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_f64 (float64x2_t __a, float64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24652,7 +27644,8 @@ vtrn2q_f64 (float64x2_t __a, float64x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_p8 (poly8x16_t __a, poly8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24664,7 +27657,8 @@ vtrn2q_p8 (poly8x16_t __a, poly8x16_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_p16 (poly16x8_t __a, poly16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24674,7 +27668,8 @@ vtrn2q_p16 (poly16x8_t __a, poly16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_s8 (int8x16_t __a, int8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24686,7 +27681,8 @@ vtrn2q_s8 (int8x16_t __a, int8x16_t __b)
 #endif
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_s16 (int16x8_t __a, int16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24696,7 +27692,8 @@ vtrn2q_s16 (int16x8_t __a, int16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_s32 (int32x4_t __a, int32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24706,7 +27703,8 @@ vtrn2q_s32 (int32x4_t __a, int32x4_t __b)
 #endif
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_s64 (int64x2_t __a, int64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24716,7 +27714,8 @@ vtrn2q_s64 (int64x2_t __a, int64x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_u8 (uint8x16_t __a, uint8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24728,7 +27727,8 @@ vtrn2q_u8 (uint8x16_t __a, uint8x16_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_u16 (uint16x8_t __a, uint16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24738,7 +27738,8 @@ vtrn2q_u16 (uint16x8_t __a, uint16x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_u32 (uint32x4_t __a, uint32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24748,7 +27749,8 @@ vtrn2q_u32 (uint32x4_t __a, uint32x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn2q_u64 (uint64x2_t __a, uint64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -24758,121 +27760,141 @@ vtrn2q_u64 (uint64x2_t __a, uint64x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_f16 (float16x4_t __a, float16x4_t __b)
 {
   return (float16x4x2_t) {vtrn1_f16 (__a, __b), vtrn2_f16 (__a, __b)};
 }
 
-__extension__ static __inline float32x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_f32 (float32x2_t a, float32x2_t b)
 {
   return (float32x2x2_t) {vtrn1_f32 (a, b), vtrn2_f32 (a, b)};
 }
 
-__extension__ static __inline poly8x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_p8 (poly8x8_t a, poly8x8_t b)
 {
   return (poly8x8x2_t) {vtrn1_p8 (a, b), vtrn2_p8 (a, b)};
 }
 
-__extension__ static __inline poly16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_p16 (poly16x4_t a, poly16x4_t b)
 {
   return (poly16x4x2_t) {vtrn1_p16 (a, b), vtrn2_p16 (a, b)};
 }
 
-__extension__ static __inline int8x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_s8 (int8x8_t a, int8x8_t b)
 {
   return (int8x8x2_t) {vtrn1_s8 (a, b), vtrn2_s8 (a, b)};
 }
 
-__extension__ static __inline int16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_s16 (int16x4_t a, int16x4_t b)
 {
   return (int16x4x2_t) {vtrn1_s16 (a, b), vtrn2_s16 (a, b)};
 }
 
-__extension__ static __inline int32x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_s32 (int32x2_t a, int32x2_t b)
 {
   return (int32x2x2_t) {vtrn1_s32 (a, b), vtrn2_s32 (a, b)};
 }
 
-__extension__ static __inline uint8x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_u8 (uint8x8_t a, uint8x8_t b)
 {
   return (uint8x8x2_t) {vtrn1_u8 (a, b), vtrn2_u8 (a, b)};
 }
 
-__extension__ static __inline uint16x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_u16 (uint16x4_t a, uint16x4_t b)
 {
   return (uint16x4x2_t) {vtrn1_u16 (a, b), vtrn2_u16 (a, b)};
 }
 
-__extension__ static __inline uint32x2x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrn_u32 (uint32x2_t a, uint32x2_t b)
 {
   return (uint32x2x2_t) {vtrn1_u32 (a, b), vtrn2_u32 (a, b)};
 }
 
-__extension__ static __inline float16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return (float16x8x2_t) {vtrn1q_f16 (__a, __b), vtrn2q_f16 (__a, __b)};
 }
 
-__extension__ static __inline float32x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_f32 (float32x4_t a, float32x4_t b)
 {
   return (float32x4x2_t) {vtrn1q_f32 (a, b), vtrn2q_f32 (a, b)};
 }
 
-__extension__ static __inline poly8x16x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_p8 (poly8x16_t a, poly8x16_t b)
 {
   return (poly8x16x2_t) {vtrn1q_p8 (a, b), vtrn2q_p8 (a, b)};
 }
 
-__extension__ static __inline poly16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_p16 (poly16x8_t a, poly16x8_t b)
 {
   return (poly16x8x2_t) {vtrn1q_p16 (a, b), vtrn2q_p16 (a, b)};
 }
 
-__extension__ static __inline int8x16x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_s8 (int8x16_t a, int8x16_t b)
 {
   return (int8x16x2_t) {vtrn1q_s8 (a, b), vtrn2q_s8 (a, b)};
 }
 
-__extension__ static __inline int16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_s16 (int16x8_t a, int16x8_t b)
 {
   return (int16x8x2_t) {vtrn1q_s16 (a, b), vtrn2q_s16 (a, b)};
 }
 
-__extension__ static __inline int32x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_s32 (int32x4_t a, int32x4_t b)
 {
   return (int32x4x2_t) {vtrn1q_s32 (a, b), vtrn2q_s32 (a, b)};
 }
 
-__extension__ static __inline uint8x16x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_u8 (uint8x16_t a, uint8x16_t b)
 {
   return (uint8x16x2_t) {vtrn1q_u8 (a, b), vtrn2q_u8 (a, b)};
 }
 
-__extension__ static __inline uint16x8x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_u16 (uint16x8_t a, uint16x8_t b)
 {
   return (uint16x8x2_t) {vtrn1q_u16 (a, b), vtrn2q_u16 (a, b)};
 }
 
-__extension__ static __inline uint32x4x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtrnq_u32 (uint32x4_t a, uint32x4_t b)
 {
   return (uint32x4x2_t) {vtrn1q_u32 (a, b), vtrn2q_u32 (a, b)};
@@ -24880,109 +27902,127 @@ vtrnq_u32 (uint32x4_t a, uint32x4_t b)
 
 /* vtst */
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_s8 (int8x8_t __a, int8x8_t __b)
 {
   return (uint8x8_t) ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_s16 (int16x4_t __a, int16x4_t __b)
 {
   return (uint16x4_t) ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_s32 (int32x2_t __a, int32x2_t __b)
 {
   return (uint32x2_t) ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_s64 (int64x1_t __a, int64x1_t __b)
 {
   return (uint64x1_t) ((__a & __b) != __AARCH64_INT64_C (0));
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_u8 (uint8x8_t __a, uint8x8_t __b)
 {
   return ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_u16 (uint16x4_t __a, uint16x4_t __b)
 {
   return ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_u32 (uint32x2_t __a, uint32x2_t __b)
 {
   return ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_u64 (uint64x1_t __a, uint64x1_t __b)
 {
   return ((__a & __b) != __AARCH64_UINT64_C (0));
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_s8 (int8x16_t __a, int8x16_t __b)
 {
   return (uint8x16_t) ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_s16 (int16x8_t __a, int16x8_t __b)
 {
   return (uint16x8_t) ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_s32 (int32x4_t __a, int32x4_t __b)
 {
   return (uint32x4_t) ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_s64 (int64x2_t __a, int64x2_t __b)
 {
   return (uint64x2_t) ((__a & __b) != __AARCH64_INT64_C (0));
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_u8 (uint8x16_t __a, uint8x16_t __b)
 {
   return ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_u16 (uint16x8_t __a, uint16x8_t __b)
 {
   return ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_u32 (uint32x4_t __a, uint32x4_t __b)
 {
   return ((__a & __b) != 0);
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_u64 (uint64x2_t __a, uint64x2_t __b)
 {
   return ((__a & __b) != __AARCH64_UINT64_C (0));
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstd_s64 (int64_t __a, int64_t __b)
 {
   return (__a & __b) ? -1ll : 0ll;
 }
 
-__extension__ static __inline uint64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstd_u64 (uint64_t __a, uint64_t __b)
 {
   return (__a & __b) ? -1ll : 0ll;
@@ -24990,81 +28030,93 @@ vtstd_u64 (uint64_t __a, uint64_t __b)
 
 /* vuqadd */
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqadd_s8 (int8x8_t __a, uint8x8_t __b)
 {
   return __builtin_aarch64_suqaddv8qi_ssu (__a,  __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqadd_s16 (int16x4_t __a, uint16x4_t __b)
 {
   return __builtin_aarch64_suqaddv4hi_ssu (__a,  __b);
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqadd_s32 (int32x2_t __a, uint32x2_t __b)
 {
   return __builtin_aarch64_suqaddv2si_ssu (__a,  __b);
 }
 
-__extension__ static __inline int64x1_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqadd_s64 (int64x1_t __a, uint64x1_t __b)
 {
   return (int64x1_t) {__builtin_aarch64_suqadddi_ssu (__a[0], __b[0])};
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqaddq_s8 (int8x16_t __a, uint8x16_t __b)
 {
   return __builtin_aarch64_suqaddv16qi_ssu (__a,  __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqaddq_s16 (int16x8_t __a, uint16x8_t __b)
 {
   return __builtin_aarch64_suqaddv8hi_ssu (__a,  __b);
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqaddq_s32 (int32x4_t __a, uint32x4_t __b)
 {
   return __builtin_aarch64_suqaddv4si_ssu (__a,  __b);
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqaddq_s64 (int64x2_t __a, uint64x2_t __b)
 {
   return __builtin_aarch64_suqaddv2di_ssu (__a,  __b);
 }
 
-__extension__ static __inline int8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqaddb_s8 (int8_t __a, uint8_t __b)
 {
   return __builtin_aarch64_suqaddqi_ssu (__a,  __b);
 }
 
-__extension__ static __inline int16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqaddh_s16 (int16_t __a, uint16_t __b)
 {
   return __builtin_aarch64_suqaddhi_ssu (__a,  __b);
 }
 
-__extension__ static __inline int32_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqadds_s32 (int32_t __a, uint32_t __b)
 {
   return __builtin_aarch64_suqaddsi_ssu (__a,  __b);
 }
 
-__extension__ static __inline int64_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuqaddd_s64 (int64_t __a, uint64_t __b)
 {
   return __builtin_aarch64_suqadddi_ssu (__a,  __b);
 }
 
 #define __DEFINTERLEAVE(op, rettype, intype, funcsuffix, Q)            \
-  __extension__ static __inline rettype                                        \
-  __attribute__ ((__always_inline__))                                  \
+  __extension__ extern __inline rettype                                        \
+  __attribute__ ((__always_inline__, __gnu_inline__, __artificial__)) \
   v ## op ## Q ## _ ## funcsuffix (intype a, intype b)                 \
   {                                                                    \
     return (rettype) {v ## op ## 1 ## Q ## _ ## funcsuffix (a, b),     \
@@ -25095,7 +28147,8 @@ vuqaddd_s64 (int64_t __a, uint64_t __b)
 
 /* vuzp */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_f16 (float16x4_t __a, float16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25105,7 +28158,8 @@ vuzp1_f16 (float16x4_t __a, float16x4_t __b)
 #endif
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_f32 (float32x2_t __a, float32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25115,7 +28169,8 @@ vuzp1_f32 (float32x2_t __a, float32x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_p8 (poly8x8_t __a, poly8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25125,7 +28180,8 @@ vuzp1_p8 (poly8x8_t __a, poly8x8_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_p16 (poly16x4_t __a, poly16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25135,7 +28191,8 @@ vuzp1_p16 (poly16x4_t __a, poly16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_s8 (int8x8_t __a, int8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25145,7 +28202,8 @@ vuzp1_s8 (int8x8_t __a, int8x8_t __b)
 #endif
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_s16 (int16x4_t __a, int16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25155,7 +28213,8 @@ vuzp1_s16 (int16x4_t __a, int16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_s32 (int32x2_t __a, int32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25165,7 +28224,8 @@ vuzp1_s32 (int32x2_t __a, int32x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_u8 (uint8x8_t __a, uint8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25175,7 +28235,8 @@ vuzp1_u8 (uint8x8_t __a, uint8x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_u16 (uint16x4_t __a, uint16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25185,7 +28246,8 @@ vuzp1_u16 (uint16x4_t __a, uint16x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1_u32 (uint32x2_t __a, uint32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25195,7 +28257,8 @@ vuzp1_u32 (uint32x2_t __a, uint32x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_f16 (float16x8_t __a, float16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25205,7 +28268,8 @@ vuzp1q_f16 (float16x8_t __a, float16x8_t __b)
 #endif
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_f32 (float32x4_t __a, float32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25215,7 +28279,8 @@ vuzp1q_f32 (float32x4_t __a, float32x4_t __b)
 #endif
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_f64 (float64x2_t __a, float64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25225,7 +28290,8 @@ vuzp1q_f64 (float64x2_t __a, float64x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_p8 (poly8x16_t __a, poly8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25237,7 +28303,8 @@ vuzp1q_p8 (poly8x16_t __a, poly8x16_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_p16 (poly16x8_t __a, poly16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25247,7 +28314,8 @@ vuzp1q_p16 (poly16x8_t __a, poly16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_s8 (int8x16_t __a, int8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25259,7 +28327,8 @@ vuzp1q_s8 (int8x16_t __a, int8x16_t __b)
 #endif
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_s16 (int16x8_t __a, int16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25269,7 +28338,8 @@ vuzp1q_s16 (int16x8_t __a, int16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_s32 (int32x4_t __a, int32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25279,7 +28349,8 @@ vuzp1q_s32 (int32x4_t __a, int32x4_t __b)
 #endif
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_s64 (int64x2_t __a, int64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25289,7 +28360,8 @@ vuzp1q_s64 (int64x2_t __a, int64x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_u8 (uint8x16_t __a, uint8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25301,7 +28373,8 @@ vuzp1q_u8 (uint8x16_t __a, uint8x16_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_u16 (uint16x8_t __a, uint16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25311,7 +28384,8 @@ vuzp1q_u16 (uint16x8_t __a, uint16x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_u32 (uint32x4_t __a, uint32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25321,7 +28395,8 @@ vuzp1q_u32 (uint32x4_t __a, uint32x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp1q_u64 (uint64x2_t __a, uint64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25331,7 +28406,8 @@ vuzp1q_u64 (uint64x2_t __a, uint64x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_f16 (float16x4_t __a, float16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25341,7 +28417,8 @@ vuzp2_f16 (float16x4_t __a, float16x4_t __b)
 #endif
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_f32 (float32x2_t __a, float32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25351,7 +28428,8 @@ vuzp2_f32 (float32x2_t __a, float32x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_p8 (poly8x8_t __a, poly8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25361,7 +28439,8 @@ vuzp2_p8 (poly8x8_t __a, poly8x8_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_p16 (poly16x4_t __a, poly16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25371,7 +28450,8 @@ vuzp2_p16 (poly16x4_t __a, poly16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_s8 (int8x8_t __a, int8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25381,7 +28461,8 @@ vuzp2_s8 (int8x8_t __a, int8x8_t __b)
 #endif
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_s16 (int16x4_t __a, int16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25391,7 +28472,8 @@ vuzp2_s16 (int16x4_t __a, int16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_s32 (int32x2_t __a, int32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25401,7 +28483,8 @@ vuzp2_s32 (int32x2_t __a, int32x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_u8 (uint8x8_t __a, uint8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25411,7 +28494,8 @@ vuzp2_u8 (uint8x8_t __a, uint8x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_u16 (uint16x4_t __a, uint16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25421,7 +28505,8 @@ vuzp2_u16 (uint16x4_t __a, uint16x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2_u32 (uint32x2_t __a, uint32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25431,7 +28516,8 @@ vuzp2_u32 (uint32x2_t __a, uint32x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_f16 (float16x8_t __a, float16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25441,7 +28527,8 @@ vuzp2q_f16 (float16x8_t __a, float16x8_t __b)
 #endif
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_f32 (float32x4_t __a, float32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25451,7 +28538,8 @@ vuzp2q_f32 (float32x4_t __a, float32x4_t __b)
 #endif
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_f64 (float64x2_t __a, float64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25461,7 +28549,8 @@ vuzp2q_f64 (float64x2_t __a, float64x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_p8 (poly8x16_t __a, poly8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25473,7 +28562,8 @@ vuzp2q_p8 (poly8x16_t __a, poly8x16_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_p16 (poly16x8_t __a, poly16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25483,7 +28573,8 @@ vuzp2q_p16 (poly16x8_t __a, poly16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_s8 (int8x16_t __a, int8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25495,7 +28586,8 @@ vuzp2q_s8 (int8x16_t __a, int8x16_t __b)
 #endif
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_s16 (int16x8_t __a, int16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25505,7 +28597,8 @@ vuzp2q_s16 (int16x8_t __a, int16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_s32 (int32x4_t __a, int32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25515,7 +28608,8 @@ vuzp2q_s32 (int32x4_t __a, int32x4_t __b)
 #endif
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_s64 (int64x2_t __a, int64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25525,7 +28619,8 @@ vuzp2q_s64 (int64x2_t __a, int64x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_u8 (uint8x16_t __a, uint8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25537,7 +28632,8 @@ vuzp2q_u8 (uint8x16_t __a, uint8x16_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_u16 (uint16x8_t __a, uint16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25547,7 +28643,8 @@ vuzp2q_u16 (uint16x8_t __a, uint16x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_u32 (uint32x4_t __a, uint32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25557,7 +28654,8 @@ vuzp2q_u32 (uint32x4_t __a, uint32x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vuzp2q_u64 (uint64x2_t __a, uint64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25571,7 +28669,8 @@ __INTERLEAVE_LIST (uzp)
 
 /* vzip */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_f16 (float16x4_t __a, float16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25581,7 +28680,8 @@ vzip1_f16 (float16x4_t __a, float16x4_t __b)
 #endif
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_f32 (float32x2_t __a, float32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25591,7 +28691,8 @@ vzip1_f32 (float32x2_t __a, float32x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_p8 (poly8x8_t __a, poly8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25601,7 +28702,8 @@ vzip1_p8 (poly8x8_t __a, poly8x8_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_p16 (poly16x4_t __a, poly16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25611,7 +28713,8 @@ vzip1_p16 (poly16x4_t __a, poly16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_s8 (int8x8_t __a, int8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25621,7 +28724,8 @@ vzip1_s8 (int8x8_t __a, int8x8_t __b)
 #endif
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_s16 (int16x4_t __a, int16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25631,7 +28735,8 @@ vzip1_s16 (int16x4_t __a, int16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_s32 (int32x2_t __a, int32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25641,7 +28746,8 @@ vzip1_s32 (int32x2_t __a, int32x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_u8 (uint8x8_t __a, uint8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25651,7 +28757,8 @@ vzip1_u8 (uint8x8_t __a, uint8x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_u16 (uint16x4_t __a, uint16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25661,7 +28768,8 @@ vzip1_u16 (uint16x4_t __a, uint16x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1_u32 (uint32x2_t __a, uint32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25671,7 +28779,8 @@ vzip1_u32 (uint32x2_t __a, uint32x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_f16 (float16x8_t __a, float16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25683,7 +28792,8 @@ vzip1q_f16 (float16x8_t __a, float16x8_t __b)
 #endif
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_f32 (float32x4_t __a, float32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25693,7 +28803,8 @@ vzip1q_f32 (float32x4_t __a, float32x4_t __b)
 #endif
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_f64 (float64x2_t __a, float64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25703,7 +28814,8 @@ vzip1q_f64 (float64x2_t __a, float64x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_p8 (poly8x16_t __a, poly8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25715,7 +28827,8 @@ vzip1q_p8 (poly8x16_t __a, poly8x16_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_p16 (poly16x8_t __a, poly16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25726,7 +28839,8 @@ vzip1q_p16 (poly16x8_t __a, poly16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_s8 (int8x16_t __a, int8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25738,7 +28852,8 @@ vzip1q_s8 (int8x16_t __a, int8x16_t __b)
 #endif
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_s16 (int16x8_t __a, int16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25749,7 +28864,8 @@ vzip1q_s16 (int16x8_t __a, int16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_s32 (int32x4_t __a, int32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25759,7 +28875,8 @@ vzip1q_s32 (int32x4_t __a, int32x4_t __b)
 #endif
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_s64 (int64x2_t __a, int64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25769,7 +28886,8 @@ vzip1q_s64 (int64x2_t __a, int64x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_u8 (uint8x16_t __a, uint8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25781,7 +28899,8 @@ vzip1q_u8 (uint8x16_t __a, uint8x16_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_u16 (uint16x8_t __a, uint16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25792,7 +28911,8 @@ vzip1q_u16 (uint16x8_t __a, uint16x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_u32 (uint32x4_t __a, uint32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25802,7 +28922,8 @@ vzip1q_u32 (uint32x4_t __a, uint32x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip1q_u64 (uint64x2_t __a, uint64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25812,7 +28933,8 @@ vzip1q_u64 (uint64x2_t __a, uint64x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_f16 (float16x4_t __a, float16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25822,7 +28944,8 @@ vzip2_f16 (float16x4_t __a, float16x4_t __b)
 #endif
 }
 
-__extension__ static __inline float32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_f32 (float32x2_t __a, float32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25832,7 +28955,8 @@ vzip2_f32 (float32x2_t __a, float32x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_p8 (poly8x8_t __a, poly8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25842,7 +28966,8 @@ vzip2_p8 (poly8x8_t __a, poly8x8_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_p16 (poly16x4_t __a, poly16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25852,7 +28977,8 @@ vzip2_p16 (poly16x4_t __a, poly16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_s8 (int8x8_t __a, int8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25862,7 +28988,8 @@ vzip2_s8 (int8x8_t __a, int8x8_t __b)
 #endif
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_s16 (int16x4_t __a, int16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25872,7 +28999,8 @@ vzip2_s16 (int16x4_t __a, int16x4_t __b)
 #endif
 }
 
-__extension__ static __inline int32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_s32 (int32x2_t __a, int32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25882,7 +29010,8 @@ vzip2_s32 (int32x2_t __a, int32x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_u8 (uint8x8_t __a, uint8x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25892,7 +29021,8 @@ vzip2_u8 (uint8x8_t __a, uint8x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_u16 (uint16x4_t __a, uint16x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25902,7 +29032,8 @@ vzip2_u16 (uint16x4_t __a, uint16x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2_u32 (uint32x2_t __a, uint32x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25912,7 +29043,8 @@ vzip2_u32 (uint32x2_t __a, uint32x2_t __b)
 #endif
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_f16 (float16x8_t __a, float16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25924,7 +29056,8 @@ vzip2q_f16 (float16x8_t __a, float16x8_t __b)
 #endif
 }
 
-__extension__ static __inline float32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_f32 (float32x4_t __a, float32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25934,7 +29067,8 @@ vzip2q_f32 (float32x4_t __a, float32x4_t __b)
 #endif
 }
 
-__extension__ static __inline float64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_f64 (float64x2_t __a, float64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25944,7 +29078,8 @@ vzip2q_f64 (float64x2_t __a, float64x2_t __b)
 #endif
 }
 
-__extension__ static __inline poly8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_p8 (poly8x16_t __a, poly8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25956,7 +29091,8 @@ vzip2q_p8 (poly8x16_t __a, poly8x16_t __b)
 #endif
 }
 
-__extension__ static __inline poly16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline poly16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_p16 (poly16x8_t __a, poly16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25967,7 +29103,8 @@ vzip2q_p16 (poly16x8_t __a, poly16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_s8 (int8x16_t __a, int8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25979,7 +29116,8 @@ vzip2q_s8 (int8x16_t __a, int8x16_t __b)
 #endif
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_s16 (int16x8_t __a, int16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -25990,7 +29128,8 @@ vzip2q_s16 (int16x8_t __a, int16x8_t __b)
 #endif
 }
 
-__extension__ static __inline int32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_s32 (int32x4_t __a, int32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -26000,7 +29139,8 @@ vzip2q_s32 (int32x4_t __a, int32x4_t __b)
 #endif
 }
 
-__extension__ static __inline int64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_s64 (int64x2_t __a, int64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -26010,7 +29150,8 @@ vzip2q_s64 (int64x2_t __a, int64x2_t __b)
 #endif
 }
 
-__extension__ static __inline uint8x16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint8x16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_u8 (uint8x16_t __a, uint8x16_t __b)
 {
 #ifdef __AARCH64EB__
@@ -26022,7 +29163,8 @@ vzip2q_u8 (uint8x16_t __a, uint8x16_t __b)
 #endif
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_u16 (uint16x8_t __a, uint16x8_t __b)
 {
 #ifdef __AARCH64EB__
@@ -26033,7 +29175,8 @@ vzip2q_u16 (uint16x8_t __a, uint16x8_t __b)
 #endif
 }
 
-__extension__ static __inline uint32x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint32x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_u32 (uint32x4_t __a, uint32x4_t __b)
 {
 #ifdef __AARCH64EB__
@@ -26043,7 +29186,8 @@ vzip2q_u32 (uint32x4_t __a, uint32x4_t __b)
 #endif
 }
 
-__extension__ static __inline uint64x2_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vzip2q_u64 (uint64x2_t __a, uint64x2_t __b)
 {
 #ifdef __AARCH64EB__
@@ -26071,349 +29215,407 @@ __INTERLEAVE_LIST (zip)
 
 /* ARMv8.2-A FP16 one operand vector intrinsics.  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabs_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_absv4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabsq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_absv8hf (__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqz_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_cmeqv4hf_uss (__a, vdup_n_f16 (0.0f));
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqzq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_cmeqv8hf_uss (__a, vdupq_n_f16 (0.0f));
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgez_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_cmgev4hf_uss (__a, vdup_n_f16 (0.0f));
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgezq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_cmgev8hf_uss (__a, vdupq_n_f16 (0.0f));
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtz_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_cmgtv4hf_uss (__a, vdup_n_f16 (0.0f));
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtzq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_cmgtv8hf_uss (__a, vdupq_n_f16 (0.0f));
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclez_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_cmlev4hf_uss (__a, vdup_n_f16 (0.0f));
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclezq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_cmlev8hf_uss (__a, vdupq_n_f16 (0.0f));
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltz_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_cmltv4hf_uss (__a, vdup_n_f16 (0.0f));
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltzq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_cmltv8hf_uss (__a, vdupq_n_f16 (0.0f));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f16_s16 (int16x4_t __a)
 {
   return __builtin_aarch64_floatv4hiv4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_f16_s16 (int16x8_t __a)
 {
   return __builtin_aarch64_floatv8hiv8hf (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_f16_u16 (uint16x4_t __a)
 {
   return __builtin_aarch64_floatunsv4hiv4hf ((int16x4_t) __a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_f16_u16 (uint16x8_t __a)
 {
   return __builtin_aarch64_floatunsv8hiv8hf ((int16x8_t) __a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_s16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lbtruncv4hfv4hi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_s16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lbtruncv8hfv8hi (__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_u16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lbtruncuv4hfv4hi_us (__a);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_u16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lbtruncuv8hfv8hi_us (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvta_s16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lroundv4hfv4hi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtaq_s16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lroundv8hfv8hi (__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvta_u16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lrounduv4hfv4hi_us (__a);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtaq_u16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lrounduv8hfv8hi_us (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtm_s16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lfloorv4hfv4hi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtmq_s16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lfloorv8hfv8hi (__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtm_u16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lflooruv4hfv4hi_us (__a);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtmq_u16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lflooruv8hfv8hi_us (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtn_s16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lfrintnv4hfv4hi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtnq_s16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lfrintnv8hfv8hi (__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtn_u16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lfrintnuv4hfv4hi_us (__a);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtnq_u16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lfrintnuv8hfv8hi_us (__a);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtp_s16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lceilv4hfv4hi (__a);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtpq_s16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lceilv8hfv8hi (__a);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtp_u16_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_lceiluv4hfv4hi_us (__a);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtpq_u16_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_lceiluv8hfv8hi_us (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vneg_f16 (float16x4_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vnegq_f16 (float16x8_t __a)
 {
   return -__a;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpe_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_frecpev4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpeq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_frecpev8hf (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrnd_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_btruncv4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_btruncv8hf (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrnda_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_roundv4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndaq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_roundv8hf (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndi_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_nearbyintv4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndiq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_nearbyintv8hf (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndm_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_floorv4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndmq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_floorv8hf (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndn_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_frintnv4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndnq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_frintnv8hf (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndp_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_ceilv4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndpq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_ceilv8hf (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndx_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_rintv4hf (__a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrndxq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_rintv8hf (__a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrte_f16 (float16x4_t a)
 {
   return __builtin_aarch64_rsqrtev4hf (a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrteq_f16 (float16x8_t a)
 {
   return __builtin_aarch64_rsqrtev8hf (a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqrt_f16 (float16x4_t a)
 {
   return __builtin_aarch64_sqrtv4hf (a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsqrtq_f16 (float16x8_t a)
 {
   return __builtin_aarch64_sqrtv8hf (a);
@@ -26421,361 +29623,421 @@ vsqrtq_f16 (float16x8_t a)
 
 /* ARMv8.2-A FP16 two operands vector intrinsics.  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vadd_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vaddq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __a + __b;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabd_f16 (float16x4_t a, float16x4_t b)
 {
   return __builtin_aarch64_fabdv4hf (a, b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdq_f16 (float16x8_t a, float16x8_t b)
 {
   return __builtin_aarch64_fabdv8hf (a, b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcage_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_facgev4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcageq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_facgev8hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcagt_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_facgtv4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcagtq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_facgtv8hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcale_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_faclev4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcaleq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_faclev8hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcalt_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_facltv4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcaltq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_facltv8hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceq_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_cmeqv4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vceqq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_cmeqv8hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcge_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_cmgev4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgeq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_cmgev8hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgt_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_cmgtv4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcgtq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_cmgtv8hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcle_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_cmlev4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcleq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_cmlev8hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vclt_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_cmltv4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcltq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_cmltv8hf_uss (__a, __b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_f16_s16 (int16x4_t __a, const int __b)
 {
   return __builtin_aarch64_scvtfv4hi (__a, __b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_f16_s16 (int16x8_t __a, const int __b)
 {
   return __builtin_aarch64_scvtfv8hi (__a, __b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_f16_u16 (uint16x4_t __a, const int __b)
 {
   return __builtin_aarch64_ucvtfv4hi_sus (__a, __b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_f16_u16 (uint16x8_t __a, const int __b)
 {
   return __builtin_aarch64_ucvtfv8hi_sus (__a, __b);
 }
 
-__extension__ static __inline int16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_s16_f16 (float16x4_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzsv4hf (__a, __b);
 }
 
-__extension__ static __inline int16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline int16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_s16_f16 (float16x8_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzsv8hf (__a, __b);
 }
 
-__extension__ static __inline uint16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvt_n_u16_f16 (float16x4_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzuv4hf_uss (__a, __b);
 }
 
-__extension__ static __inline uint16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline uint16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vcvtq_n_u16_f16 (float16x8_t __a, const int __b)
 {
   return __builtin_aarch64_fcvtzuv8hf_uss (__a, __b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdiv_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __a / __b;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vdivq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __a / __b;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmax_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_smax_nanv4hf (__a, __b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_smax_nanv8hf (__a, __b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnm_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_fmaxv4hf (__a, __b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnmq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_fmaxv8hf (__a, __b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmin_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_smin_nanv4hf (__a, __b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_smin_nanv8hf (__a, __b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnm_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_fminv4hf (__a, __b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnmq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_fminv8hf (__a, __b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __a * __b;
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_fmulxv4hf (__a, __b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_fmulxv8hf (__a, __b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpadd_f16 (float16x4_t a, float16x4_t b)
 {
   return __builtin_aarch64_faddpv4hf (a, b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpaddq_f16 (float16x8_t a, float16x8_t b)
 {
   return __builtin_aarch64_faddpv8hf (a, b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmax_f16 (float16x4_t a, float16x4_t b)
 {
   return __builtin_aarch64_smax_nanpv4hf (a, b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxq_f16 (float16x8_t a, float16x8_t b)
 {
   return __builtin_aarch64_smax_nanpv8hf (a, b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxnm_f16 (float16x4_t a, float16x4_t b)
 {
   return __builtin_aarch64_smaxpv4hf (a, b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmaxnmq_f16 (float16x8_t a, float16x8_t b)
 {
   return __builtin_aarch64_smaxpv8hf (a, b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpmin_f16 (float16x4_t a, float16x4_t b)
 {
   return __builtin_aarch64_smin_nanpv4hf (a, b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminq_f16 (float16x8_t a, float16x8_t b)
 {
   return __builtin_aarch64_smin_nanpv8hf (a, b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminnm_f16 (float16x4_t a, float16x4_t b)
 {
   return __builtin_aarch64_sminpv4hf (a, b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vpminnmq_f16 (float16x8_t a, float16x8_t b)
 {
   return __builtin_aarch64_sminpv8hf (a, b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecps_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __builtin_aarch64_frecpsv4hf (__a, __b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrecpsq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __builtin_aarch64_frecpsv8hf (__a, __b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrts_f16 (float16x4_t a, float16x4_t b)
 {
   return __builtin_aarch64_rsqrtsv4hf (a, b);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vrsqrtsq_f16 (float16x8_t a, float16x8_t b)
 {
   return __builtin_aarch64_rsqrtsv8hf (a, b);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsub_f16 (float16x4_t __a, float16x4_t __b)
 {
   return __a - __b;
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vsubq_f16 (float16x8_t __a, float16x8_t __b)
 {
   return __a - __b;
@@ -26783,25 +30045,29 @@ vsubq_f16 (float16x8_t __a, float16x8_t __b)
 
 /* ARMv8.2-A FP16 three operands vector intrinsics.  */
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_f16 (float16x4_t __a, float16x4_t __b, float16x4_t __c)
 {
   return __builtin_aarch64_fmav4hf (__b, __c, __a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_f16 (float16x8_t __a, float16x8_t __b, float16x8_t __c)
 {
   return __builtin_aarch64_fmav8hf (__b, __c, __a);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_f16 (float16x4_t __a, float16x4_t __b, float16x4_t __c)
 {
   return __builtin_aarch64_fnmav4hf (__b, __c, __a);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_f16 (float16x8_t __a, float16x8_t __b, float16x8_t __c)
 {
   return __builtin_aarch64_fnmav8hf (__b, __c, __a);
@@ -26809,205 +30075,237 @@ vfmsq_f16 (float16x8_t __a, float16x8_t __b, float16x8_t __c)
 
 /* ARMv8.2-A FP16 lane vector intrinsics.  */
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmah_lane_f16 (float16_t __a, float16_t __b,
                float16x4_t __c, const int __lane)
 {
   return vfmah_f16 (__a, __b, __aarch64_vget_lane_any (__c, __lane));
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmah_laneq_f16 (float16_t __a, float16_t __b,
                 float16x8_t __c, const int __lane)
 {
   return vfmah_f16 (__a, __b, __aarch64_vget_lane_any (__c, __lane));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_lane_f16 (float16x4_t __a, float16x4_t __b,
               float16x4_t __c, const int __lane)
 {
   return vfma_f16 (__a, __b, __aarch64_vdup_lane_f16 (__c, __lane));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_lane_f16 (float16x8_t __a, float16x8_t __b,
                float16x4_t __c, const int __lane)
 {
   return vfmaq_f16 (__a, __b, __aarch64_vdupq_lane_f16 (__c, __lane));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_laneq_f16 (float16x4_t __a, float16x4_t __b,
                float16x8_t __c, const int __lane)
 {
   return vfma_f16 (__a, __b, __aarch64_vdup_laneq_f16 (__c, __lane));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_laneq_f16 (float16x8_t __a, float16x8_t __b,
                 float16x8_t __c, const int __lane)
 {
   return vfmaq_f16 (__a, __b, __aarch64_vdupq_laneq_f16 (__c, __lane));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfma_n_f16 (float16x4_t __a, float16x4_t __b, float16_t __c)
 {
   return vfma_f16 (__a, __b, vdup_n_f16 (__c));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmaq_n_f16 (float16x8_t __a, float16x8_t __b, float16_t __c)
 {
   return vfmaq_f16 (__a, __b, vdupq_n_f16 (__c));
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsh_lane_f16 (float16_t __a, float16_t __b,
                float16x4_t __c, const int __lane)
 {
   return vfmsh_f16 (__a, __b, __aarch64_vget_lane_any (__c, __lane));
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsh_laneq_f16 (float16_t __a, float16_t __b,
                 float16x8_t __c, const int __lane)
 {
   return vfmsh_f16 (__a, __b, __aarch64_vget_lane_any (__c, __lane));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_lane_f16 (float16x4_t __a, float16x4_t __b,
               float16x4_t __c, const int __lane)
 {
   return vfms_f16 (__a, __b, __aarch64_vdup_lane_f16 (__c, __lane));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_lane_f16 (float16x8_t __a, float16x8_t __b,
                float16x4_t __c, const int __lane)
 {
   return vfmsq_f16 (__a, __b, __aarch64_vdupq_lane_f16 (__c, __lane));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_laneq_f16 (float16x4_t __a, float16x4_t __b,
                float16x8_t __c, const int __lane)
 {
   return vfms_f16 (__a, __b, __aarch64_vdup_laneq_f16 (__c, __lane));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_laneq_f16 (float16x8_t __a, float16x8_t __b,
                 float16x8_t __c, const int __lane)
 {
   return vfmsq_f16 (__a, __b, __aarch64_vdupq_laneq_f16 (__c, __lane));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfms_n_f16 (float16x4_t __a, float16x4_t __b, float16_t __c)
 {
   return vfms_f16 (__a, __b, vdup_n_f16 (__c));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vfmsq_n_f16 (float16x8_t __a, float16x8_t __b, float16_t __c)
 {
   return vfmsq_f16 (__a, __b, vdupq_n_f16 (__c));
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulh_lane_f16 (float16_t __a, float16x4_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_lane_f16 (float16x4_t __a, float16x4_t __b, const int __lane)
 {
   return vmul_f16 (__a, vdup_n_f16 (__aarch64_vget_lane_any (__b, __lane)));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_lane_f16 (float16x8_t __a, float16x4_t __b, const int __lane)
 {
   return vmulq_f16 (__a, vdupq_n_f16 (__aarch64_vget_lane_any (__b, __lane)));
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulh_laneq_f16 (float16_t __a, float16x8_t __b, const int __lane)
 {
   return __a * __aarch64_vget_lane_any (__b, __lane);
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_laneq_f16 (float16x4_t __a, float16x8_t __b, const int __lane)
 {
   return vmul_f16 (__a, vdup_n_f16 (__aarch64_vget_lane_any (__b, __lane)));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_laneq_f16 (float16x8_t __a, float16x8_t __b, const int __lane)
 {
   return vmulq_f16 (__a, vdupq_n_f16 (__aarch64_vget_lane_any (__b, __lane)));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmul_n_f16 (float16x4_t __a, float16_t __b)
 {
   return vmul_lane_f16 (__a, vdup_n_f16 (__b), 0);
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulq_n_f16 (float16x8_t __a, float16_t __b)
 {
   return vmulq_laneq_f16 (__a, vdupq_n_f16 (__b), 0);
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxh_lane_f16 (float16_t __a, float16x4_t __b, const int __lane)
 {
   return vmulxh_f16 (__a, __aarch64_vget_lane_any (__b, __lane));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_lane_f16 (float16x4_t __a, float16x4_t __b, const int __lane)
 {
   return vmulx_f16 (__a, __aarch64_vdup_lane_f16 (__b, __lane));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_lane_f16 (float16x8_t __a, float16x4_t __b, const int __lane)
 {
   return vmulxq_f16 (__a, __aarch64_vdupq_lane_f16 (__b, __lane));
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxh_laneq_f16 (float16_t __a, float16x8_t __b, const int __lane)
 {
   return vmulxh_f16 (__a, __aarch64_vget_lane_any (__b, __lane));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_laneq_f16 (float16x4_t __a, float16x8_t __b, const int __lane)
 {
   return vmulx_f16 (__a, __aarch64_vdup_laneq_f16 (__b, __lane));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_laneq_f16 (float16x8_t __a, float16x8_t __b, const int __lane)
 {
   return vmulxq_f16 (__a, __aarch64_vdupq_laneq_f16 (__b, __lane));
 }
 
-__extension__ static __inline float16x4_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x4_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulx_n_f16 (float16x4_t __a, float16_t __b)
 {
   return vmulx_f16 (__a, vdup_n_f16 (__b));
 }
 
-__extension__ static __inline float16x8_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16x8_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmulxq_n_f16 (float16x8_t __a, float16_t __b)
 {
   return vmulxq_f16 (__a, vdupq_n_f16 (__b));
@@ -27015,49 +30313,57 @@ vmulxq_n_f16 (float16x8_t __a, float16_t __b)
 
 /* ARMv8.2-A FP16 reduction vector intrinsics.  */
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxv_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_reduc_smax_nan_scal_v4hf (__a);
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxvq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_reduc_smax_nan_scal_v8hf (__a);
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminv_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_reduc_smin_nan_scal_v4hf (__a);
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminvq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_reduc_smin_nan_scal_v8hf (__a);
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnmv_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v4hf (__a);
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vmaxnmvq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_reduc_smax_scal_v8hf (__a);
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnmv_f16 (float16x4_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v4hf (__a);
 }
 
-__extension__ static __inline float16_t __attribute__ ((__always_inline__))
+__extension__ extern __inline float16_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vminnmvq_f16 (float16x8_t __a)
 {
   return __builtin_aarch64_reduc_smin_scal_v8hf (__a);