From 9b588cfb4288ef41eb59edc6ab8cb83056b3b6a4 Mon Sep 17 00:00:00 2001
From: Kyrylo Tkachov <kyrylo.tkachov@arm.com>
Date: Fri, 29 Jan 2021 13:10:46 +0000
Subject: [PATCH] aarch64: Reimplement vabdl_high* intrinsics using builtins

This patch reimplements the vabdl_high intrinsics using builtins.
It slightly cleans up the RTL pattern (the mode iterators) but nothing
interesting apart from that.

gcc/ChangeLog:

	* config/aarch64/aarch64-simd-builtins.def (sabdl2, uabdl2):
	Define builtins.
	* config/aarch64/aarch64-simd.md (aarch64_<sur>abdl2<mode>_3):
	Rename to...
	(aarch64_<sur>abdl2<mode>): ... This.
	(<sur>sadv16qi): Adjust use of above.
	* config/aarch64/arm_neon.h (vabdl_high_s8): Reimplement using
	builtin.
	(vabdl_high_s16): Likewise.
	(vabdl_high_s32): Likewise.
	(vabdl_high_u8): Likewise.
	(vabdl_high_u16): Likewise.
	(vabdl_high_u32): Likewise.
---
 gcc/config/aarch64/aarch64-simd-builtins.def |  4 ++
 gcc/config/aarch64/aarch64-simd.md           | 10 ++---
 gcc/config/aarch64/arm_neon.h                | 42 +++-----------------
 3 files changed, 15 insertions(+), 41 deletions(-)

diff --git a/gcc/config/aarch64/aarch64-simd-builtins.def b/gcc/config/aarch64/aarch64-simd-builtins.def
index f3aace38577..9db259a2967 100644
--- a/gcc/config/aarch64/aarch64-simd-builtins.def
+++ b/gcc/config/aarch64/aarch64-simd-builtins.def
@@ -172,6 +172,10 @@
   BUILTIN_VQW (TERNOP, sabal2, 0, NONE)
   BUILTIN_VQW (TERNOPU, uabal2, 0, NONE)
 
+  /* Implemented by aarch64_<sur>abdl2<mode>.  */
+  BUILTIN_VQW (BINOP, sabdl2, 0, NONE)
+  BUILTIN_VQW (BINOPU, uabdl2, 0, NONE)
+
   /* Implemented by aarch64_<sur><addsub>hn<mode>.  */
   BUILTIN_VQN (BINOP, addhn, 0, NONE)
   BUILTIN_VQN (BINOP, subhn, 0, NONE)
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index 98d510b6eb6..2f41d7aaa9b 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -818,10 +818,10 @@
   [(set_attr "type" "neon_abd<q>")]
 )
 
-(define_insn "aarch64_<sur>abdl2<mode>_3"
+(define_insn "aarch64_<sur>abdl2<mode>"
   [(set (match_operand:<VDBLW> 0 "register_operand" "=w")
-	(unspec:<VDBLW> [(match_operand:VDQV_S 1 "register_operand" "w")
-			  (match_operand:VDQV_S 2 "register_operand" "w")]
+	(unspec:<VDBLW> [(match_operand:VQW 1 "register_operand" "w")
+			 (match_operand:VQW 2 "register_operand" "w")]
 	ABDL2))]
   "TARGET_SIMD"
   "<sur>abdl2\t%0.<Vwtype>, %1.<Vtype>, %2.<Vtype>"
@@ -897,8 +897,8 @@
 	DONE;
       }
     rtx reduc = gen_reg_rtx (V8HImode);
-    emit_insn (gen_aarch64_<sur>abdl2v16qi_3 (reduc, operands[1],
-					       operands[2]));
+    emit_insn (gen_aarch64_<sur>abdl2v16qi (reduc, operands[1],
+					    operands[2]));
     emit_insn (gen_aarch64_<sur>abalv8qi (reduc, reduc,
 					  gen_lowpart (V8QImode, operands[1]),
 					  gen_lowpart (V8QImode,
diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index 66feec3cce0..2297e5b8d41 100644
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -6831,72 +6831,42 @@ __extension__ extern __inline int16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_s8 (int8x16_t __a, int8x16_t __b)
 {
-  int16x8_t __result;
-  __asm__ ("sabdl2 %0.8h,%1.16b,%2.16b"
-           : "=w"(__result)
-           : "w"(__a), "w"(__b)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_sabdl2v16qi (__a, __b);
 }
 
 __extension__ extern __inline int32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_s16 (int16x8_t __a, int16x8_t __b)
 {
-  int32x4_t __result;
-  __asm__ ("sabdl2 %0.4s,%1.8h,%2.8h"
-           : "=w"(__result)
-           : "w"(__a), "w"(__b)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_sabdl2v8hi (__a, __b);
 }
 
 __extension__ extern __inline int64x2_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_s32 (int32x4_t __a, int32x4_t __b)
 {
-  int64x2_t __result;
-  __asm__ ("sabdl2 %0.2d,%1.4s,%2.4s"
-           : "=w"(__result)
-           : "w"(__a), "w"(__b)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_sabdl2v4si (__a, __b);
 }
 
 __extension__ extern __inline uint16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_u8 (uint8x16_t __a, uint8x16_t __b)
 {
-  uint16x8_t __result;
-  __asm__ ("uabdl2 %0.8h,%1.16b,%2.16b"
-           : "=w"(__result)
-           : "w"(__a), "w"(__b)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_uabdl2v16qi_uuu (__a, __b);
 }
 
 __extension__ extern __inline uint32x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_u16 (uint16x8_t __a, uint16x8_t __b)
 {
-  uint32x4_t __result;
-  __asm__ ("uabdl2 %0.4s,%1.8h,%2.8h"
-           : "=w"(__result)
-           : "w"(__a), "w"(__b)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_uabdl2v8hi_uuu (__a, __b);
 }
 
 __extension__ extern __inline uint64x2_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vabdl_high_u32 (uint32x4_t __a, uint32x4_t __b)
 {
-  uint64x2_t __result;
-  __asm__ ("uabdl2 %0.2d,%1.4s,%2.4s"
-           : "=w"(__result)
-           : "w"(__a), "w"(__b)
-           : /* No clobbers */);
-  return __result;
+  return __builtin_aarch64_uabdl2v4si_uuu (__a, __b);
 }
 
 __extension__ extern __inline int16x8_t
-- 
2.30.2