projects / gcc.git / commitdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (parent: a7c8ed0)
[ARM,AArch64] more poly64 intrinsics and tests
authorChristophe Lyon <christophe.lyon@linaro.org>
Fri, 10 Feb 2017 13:18:16 +0000 (13:18 +0000)
committerChristophe Lyon <clyon@gcc.gnu.org>
Fri, 10 Feb 2017 13:18:16 +0000 (14:18 +0100)
2017-02-06  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/
* config/aarch64/arm_neon.h (vtst_p8): Rewrite without asm.
(vtst_p16): Likewise.
(vtstq_p8): Likewise.
(vtstq_p16): Likewise.
(vtst_p64): New.
(vtstq_p64): Likewise.
* config/arm/arm_neon.h (vgetq_lane_p64): New.
(vset_lane_p64): New.
(vsetq_lane_p64): New.

2017-02-06  Christophe Lyon  <christophe.lyon@linaro.org>

gcc/testsuite/
* gcc.target/aarch64/advsimd-intrinsics/p64_p128.c
(vget_lane_expected, vset_lane_expected, vtst_expected_poly64x1):
New.
(vmov_n_expected0, vmov_n_expected1, vmov_n_expected2)
(expected_vld_st2_0, expected_vld_st2_1, expected_vld_st3_0)
(expected_vld_st3_1, expected_vld_st3_2, expected_vld_st4_0)
(expected_vld_st4_1, expected_vld_st4_2, expected_vld_st4_3)
(vtst_expected_poly64x2): Move to aarch64-only section.
(vget_lane_p64, vgetq_lane_p64, vset_lane_p64, vsetq_lane_p64)
(vtst_p64, vtstq_p64): New tests.

From-SVN: r245328

gcc/ChangeLog patch | blob | history
gcc/config/aarch64/arm_neon.h patch | blob | history
gcc/config/arm/arm_neon.h patch | blob | history
gcc/testsuite/ChangeLog patch | blob | history
gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/p64_p128.c patch | blob | history

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 7a06399de9a095d55dd9b9a732e55fcbc219a494..6bdfddf1fb557b4c637834432227f8eeebf8f8b9 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,15 @@
+2017-02-10  Christophe Lyon  <christophe.lyon@linaro.org>
+
+       * config/aarch64/arm_neon.h (vtst_p8): Rewrite without asm.
+       (vtst_p16): Likewise.
+       (vtstq_p8): Likewise.
+       (vtstq_p16): Likewise.
+       (vtst_p64): New.
+       (vtstq_p64): Likewise.
+       * config/arm/arm_neon.h (vgetq_lane_p64): New.
+       (vset_lane_p64): New.
+       (vsetq_lane_p64): New.
+
 2017-02-10  Jakub Jelinek  <jakub@redhat.com>
 
        PR tree-optimization/79411
diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index a54c0be04d398f7cc5402e086298b572837588c6..0753da32f59f1c56580f5d4b04a8d8275cf3c0d3 100644 (file)
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -10862,48 +10862,47 @@ __extension__ extern __inline uint8x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_p8 (poly8x8_t a, poly8x8_t b)
 {
-  uint8x8_t result;
-  __asm__ ("cmtst %0.8b, %1.8b, %2.8b"
-           : "=w"(result)
-           : "w"(a), "w"(b)
-           : /* No clobbers */);
-  return result;
+  return (uint8x8_t) ((((uint8x8_t) a) & ((uint8x8_t) b))
+                      != 0);
 }
 
 __extension__ extern __inline uint16x4_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtst_p16 (poly16x4_t a, poly16x4_t b)
 {
-  uint16x4_t result;
-  __asm__ ("cmtst %0.4h, %1.4h, %2.4h"
-           : "=w"(result)
-           : "w"(a), "w"(b)
-           : /* No clobbers */);
-  return result;
+  return (uint16x4_t) ((((uint16x4_t) a) & ((uint16x4_t) b))
+                      != 0);
+}
+
+__extension__ extern __inline uint64x1_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vtst_p64 (poly64x1_t a, poly64x1_t b)
+{
+  return (uint64x1_t) ((a & b) != __AARCH64_INT64_C (0));
 }
 
 __extension__ extern __inline uint8x16_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_p8 (poly8x16_t a, poly8x16_t b)
 {
-  uint8x16_t result;
-  __asm__ ("cmtst %0.16b, %1.16b, %2.16b"
-           : "=w"(result)
-           : "w"(a), "w"(b)
-           : /* No clobbers */);
-  return result;
+  return (uint8x16_t) ((((uint8x16_t) a) & ((uint8x16_t) b))
+                      != 0);
 }
 
 __extension__ extern __inline uint16x8_t
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vtstq_p16 (poly16x8_t a, poly16x8_t b)
 {
-  uint16x8_t result;
-  __asm__ ("cmtst %0.8h, %1.8h, %2.8h"
-           : "=w"(result)
-           : "w"(a), "w"(b)
-           : /* No clobbers */);
-  return result;
+  return (uint16x8_t) ((((uint16x8_t) a) & ((uint16x8_t) b))
+                      != 0);
+}
+
+__extension__ extern __inline uint64x2_t
+__attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
+vtstq_p64 (poly64x2_t a, poly64x2_t b)
+{
+  return (uint64x2_t) ((((uint64x2_t) a) & ((uint64x2_t) b))
+                      != __AARCH64_INT64_C (0));
 }
 
 /* End of temporary inline asm implementations.  */
diff --git a/gcc/config/arm/arm_neon.h b/gcc/config/arm/arm_neon.h
index 16bf8c5006740e1eeec2862642194ca26f990325..f81d77ef741f829c19395e82cc0df11652b302fe 100644 (file)
--- a/gcc/config/arm/arm_neon.h
+++ b/gcc/config/arm/arm_neon.h
@@ -6309,6 +6309,16 @@ vgetq_lane_s64 (int64x2_t __a, const int __b)
   return (int64_t)__builtin_neon_vget_lanev2di (__a, __b);
 }
 
+#pragma GCC push_options
+#pragma GCC target ("fpu=crypto-neon-fp-armv8")
+__extension__ extern __inline poly64_t
+__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
+vgetq_lane_p64 (poly64x2_t __a, const int __b)
+{
+  return (poly64_t)__builtin_neon_vget_lanev2di ((int64x2_t) __a, __b);
+}
+
+#pragma GCC pop_options
 __extension__ extern __inline uint64_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vgetq_lane_u64 (uint64x2_t __a, const int __b)
@@ -6405,6 +6415,16 @@ vset_lane_u64 (uint64_t __a, uint64x1_t __b, const int __c)
   return (uint64x1_t)__builtin_neon_vset_lanedi ((__builtin_neon_di) __a, (int64x1_t) __b, __c);
 }
 
+#pragma GCC push_options
+#pragma GCC target ("fpu=crypto-neon-fp-armv8")
+__extension__ extern __inline poly64x1_t
+__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
+vset_lane_p64 (poly64_t __a, poly64x1_t __b, const int __c)
+{
+  return (poly64x1_t)__builtin_neon_vset_lanedi ((__builtin_neon_di) __a, (int64x1_t) __b, __c);
+}
+
+#pragma GCC pop_options
 __extension__ extern __inline int8x16_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vsetq_lane_s8 (int8_t __a, int8x16_t __b, const int __c)
@@ -6496,6 +6516,13 @@ vsetq_lane_u64 (uint64_t __a, uint64x2_t __b, const int __c)
 
 #pragma GCC push_options
 #pragma GCC target ("fpu=crypto-neon-fp-armv8")
+__extension__ extern __inline poly64x2_t
+__attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
+vsetq_lane_p64 (poly64_t __a, poly64x2_t __b, const int __c)
+{
+  return (poly64x2_t)__builtin_neon_vset_lanev2di ((__builtin_neon_di) __a, (int64x2_t) __b, __c);
+}
+
 __extension__ extern __inline poly64x1_t
 __attribute__  ((__always_inline__, __gnu_inline__, __artificial__))
 vcreate_p64 (uint64_t __a)
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 050bd52b15b597333eff5724d63438905a250c66..a1cee1b28126a360fb24e2bf9bf6a2e07088e084 100644 (file)
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,16 @@
+2017-02-10  Christophe Lyon  <christophe.lyon@linaro.org>
+
+       * gcc.target/aarch64/advsimd-intrinsics/p64_p128.c
+       (vget_lane_expected, vset_lane_expected, vtst_expected_poly64x1):
+       New.
+       (vmov_n_expected0, vmov_n_expected1, vmov_n_expected2)
+       (expected_vld_st2_0, expected_vld_st2_1, expected_vld_st3_0)
+       (expected_vld_st3_1, expected_vld_st3_2, expected_vld_st4_0)
+       (expected_vld_st4_1, expected_vld_st4_2, expected_vld_st4_3)
+       (vtst_expected_poly64x2): Move to aarch64-only section.
+       (vget_lane_p64, vgetq_lane_p64, vset_lane_p64, vsetq_lane_p64)
+       (vtst_p64, vtstq_p64): New tests.
+
 2017-02-10  Paolo Carlini  <paolo.carlini@oracle.com>
 
        PR c++/71737
diff --git a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/p64_p128.c b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/p64_p128.c
index 7c5bca2ef70cf3dfce03956ba6591d89ab8341bb..a3210a94b1dcd00ea8b9f687a7805386ef5c0b08 100644 (file)
--- a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/p64_p128.c
+++ b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/p64_p128.c
@@ -39,17 +39,6 @@ VECT_VAR_DECL(vdup_n_expected2,poly,64,1) [] = { 0xfffffffffffffff2 };
 VECT_VAR_DECL(vdup_n_expected2,poly,64,2) [] = { 0xfffffffffffffff2,
                                                 0xfffffffffffffff2 };
 
-/* Expected results: vmov_n.  */
-VECT_VAR_DECL(vmov_n_expected0,poly,64,1) [] = { 0xfffffffffffffff0 };
-VECT_VAR_DECL(vmov_n_expected0,poly,64,2) [] = { 0xfffffffffffffff0,
-                                                0xfffffffffffffff0 };
-VECT_VAR_DECL(vmov_n_expected1,poly,64,1) [] = { 0xfffffffffffffff1 };
-VECT_VAR_DECL(vmov_n_expected1,poly,64,2) [] = { 0xfffffffffffffff1,
-                                                0xfffffffffffffff1 };
-VECT_VAR_DECL(vmov_n_expected2,poly,64,1) [] = { 0xfffffffffffffff2 };
-VECT_VAR_DECL(vmov_n_expected2,poly,64,2) [] = { 0xfffffffffffffff2,
-                                                0xfffffffffffffff2 };
-
 /* Expected results: vext.  */
 VECT_VAR_DECL(vext_expected,poly,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(vext_expected,poly,64,2) [] = { 0xfffffffffffffff1, 0x88 };
@@ -124,6 +113,29 @@ VECT_VAR_DECL(vst1_lane_expected,poly,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(vst1_lane_expected,poly,64,2) [] = { 0xfffffffffffffff0,
                                                   0x3333333333333333 };
 
+/* Expected results: vget_lane.  */
+VECT_VAR_DECL(vget_lane_expected,poly,64,1) = 0xfffffffffffffff0;
+VECT_VAR_DECL(vget_lane_expected,poly,64,2) = 0xfffffffffffffff0;
+
+/* Expected results: vset_lane.  */
+VECT_VAR_DECL(vset_lane_expected,poly,64,1) [] = { 0x88 };
+VECT_VAR_DECL(vset_lane_expected,poly,64,2) [] = { 0xfffffffffffffff0, 0x11 };
+
+/* Expected results: vtst.  */
+VECT_VAR_DECL(vtst_expected,uint,64,1) [] = { 0x0 };
+
+#ifdef __aarch64__
+/* Expected results: vmov_n.  */
+VECT_VAR_DECL(vmov_n_expected0,poly,64,1) [] = { 0xfffffffffffffff0 };
+VECT_VAR_DECL(vmov_n_expected0,poly,64,2) [] = { 0xfffffffffffffff0,
+                                                0xfffffffffffffff0 };
+VECT_VAR_DECL(vmov_n_expected1,poly,64,1) [] = { 0xfffffffffffffff1 };
+VECT_VAR_DECL(vmov_n_expected1,poly,64,2) [] = { 0xfffffffffffffff1,
+                                                0xfffffffffffffff1 };
+VECT_VAR_DECL(vmov_n_expected2,poly,64,1) [] = { 0xfffffffffffffff2 };
+VECT_VAR_DECL(vmov_n_expected2,poly,64,2) [] = { 0xfffffffffffffff2,
+                                                0xfffffffffffffff2 };
+
 /* Expected results: vldX_lane.  */
 VECT_VAR_DECL(expected_vld_st2_0,poly,64,1) [] = { 0xfffffffffffffff0 };
 VECT_VAR_DECL(expected_vld_st2_0,poly,64,2) [] = { 0xfffffffffffffff0,
@@ -153,9 +165,9 @@ VECT_VAR_DECL(expected_vld_st4_3,poly,64,1) [] = { 0xfffffffffffffff3 };
 VECT_VAR_DECL(expected_vld_st4_3,poly,64,2) [] = { 0xaaaaaaaaaaaaaaaa,
                                                   0xaaaaaaaaaaaaaaaa };
 
-/* Expected results: vget_lane.  */
-VECT_VAR_DECL(vget_lane_expected,poly,64,1) = 0xfffffffffffffff0;
-VECT_VAR_DECL(vget_lane_expected,poly,64,2) = 0xfffffffffffffff0;
+/* Expected results: vtst.  */
+VECT_VAR_DECL(vtst_expected,uint,64,2) [] = { 0x0, 0xffffffffffffffff };
+#endif
 
 int main (void)
 {
@@ -727,7 +739,105 @@ int main (void)
   CHECK_POLY(TEST_MSG, poly, 64, 1, PRIx64, vst1_lane_expected, "");
   CHECK_POLY(TEST_MSG, poly, 64, 2, PRIx64, vst1_lane_expected, "");
 
+  /* vget_lane_p64 tests.  */
+#undef TEST_MSG
+#define TEST_MSG "VGET_LANE/VGETQ_LANE"
+
+#define TEST_VGET_LANE(Q, T1, T2, W, N, L)                                \
+  VECT_VAR(vget_lane_vector, T1, W, N) = vget##Q##_lane_##T2##W(VECT_VAR(vget_lane_vector1, T1, W, N), L); \
+  if (VECT_VAR(vget_lane_vector, T1, W, N) != VECT_VAR(vget_lane_expected, T1, W, N)) {                \
+    fprintf(stderr,                                                       \
+           "ERROR in %s (%s line %d in result '%s') at type %s "          \
+           "got 0x%" PRIx##W " != 0x%" PRIx##W "\n",                      \
+           TEST_MSG, __FILE__, __LINE__,                                  \
+           STR(VECT_VAR(vget_lane_expected, T1, W, N)),                   \
+           STR(VECT_NAME(T1, W, N)),                                      \
+           VECT_VAR(vget_lane_vector, T1, W, N),                          \
+           VECT_VAR(vget_lane_expected, T1, W, N));                       \
+    abort ();                                                             \
+  }
+
+  /* Initialize input values.  */
+  DECL_VARIABLE(vget_lane_vector1, poly, 64, 1);
+  DECL_VARIABLE(vget_lane_vector1, poly, 64, 2);
+
+  VLOAD(vget_lane_vector1, buffer,  , poly, p, 64, 1);
+  VLOAD(vget_lane_vector1, buffer, q, poly, p, 64, 2);
+
+  VECT_VAR_DECL(vget_lane_vector, poly, 64, 1);
+  VECT_VAR_DECL(vget_lane_vector, poly, 64, 2);
+
+  TEST_VGET_LANE( , poly, p, 64, 1, 0);
+  TEST_VGET_LANE(q, poly, p, 64, 2, 0);
+
+
+  /* vset_lane_p64 tests.  */
+#undef TEST_MSG
+#define TEST_MSG "VSET_LANE/VSETQ_LANE"
+
+#define TEST_VSET_LANE(Q, T1, T2, W, N, V, L)                          \
+  VECT_VAR(vset_lane_vector, T1, W, N) =                                               \
+    vset##Q##_lane_##T2##W(V,                                          \
+                          VECT_VAR(vset_lane_vector, T1, W, N),                        \
+                          L);                                          \
+  vst1##Q##_##T2##W(VECT_VAR(result, T1, W, N), VECT_VAR(vset_lane_vector, T1, W, N))
+
+  /* Initialize input values.  */
+  DECL_VARIABLE(vset_lane_vector, poly, 64, 1);
+  DECL_VARIABLE(vset_lane_vector, poly, 64, 2);
+
+  CLEAN(result, uint, 64, 1);
+  CLEAN(result, uint, 64, 2);
+
+  VLOAD(vset_lane_vector, buffer, , poly, p, 64, 1);
+  VLOAD(vset_lane_vector, buffer, q, poly, p, 64, 2);
+
+  /* Choose value and lane arbitrarily.  */
+  TEST_VSET_LANE(, poly, p, 64, 1, 0x88, 0);
+  TEST_VSET_LANE(q, poly, p, 64, 2, 0x11, 1);
+
+  CHECK(TEST_MSG, poly, 64, 1, PRIx64, vset_lane_expected, "");
+  CHECK(TEST_MSG, poly, 64, 2, PRIx64, vset_lane_expected, "");
+
+
+  /* vtst_p64 tests.  */
+#undef TEST_MSG
+#define TEST_MSG "VTST"
+  
+#define TEST_VTST1(INSN, Q, T1, T2, W, N)                      \
+  VECT_VAR(vtst_vector_res, uint, W, N) =                      \
+    INSN##Q##_##T2##W(VECT_VAR(vtst_vector, T1, W, N),         \
+                     VECT_VAR(vtst_vector2, T1, W, N));        \
+    vst1##Q##_u##W(VECT_VAR(result, uint, W, N),               \
+                  VECT_VAR(vtst_vector_res, uint, W, N))
+
+#define TEST_VTST(INSN, Q, T1, T2, W, N)       \
+  TEST_VTST1(INSN, Q, T1, T2, W, N)            \
+
+  /* Initialize input values.  */
+  DECL_VARIABLE(vtst_vector, poly, 64, 1);
+  DECL_VARIABLE(vtst_vector2, poly, 64, 1);
+  DECL_VARIABLE(vtst_vector_res, uint, 64, 1);
+
+  CLEAN(result, uint, 64, 1);
+
+  VLOAD(vtst_vector, buffer,  , poly, p, 64, 1);
+  VDUP(vtst_vector2, , poly, p, 64, 1, 5);
+
+  TEST_VTST(vtst, , poly, p, 64, 1);
+
+  CHECK(TEST_MSG, uint, 64, 1, PRIx64, vtst_expected, "");
+
+  /* vtstq_p64 is supported by aarch64 only.  */
 #ifdef __aarch64__
+  DECL_VARIABLE(vtst_vector, poly, 64, 2);
+  DECL_VARIABLE(vtst_vector2, poly, 64, 2);
+  DECL_VARIABLE(vtst_vector_res, uint, 64, 2);
+  CLEAN(result, uint, 64, 2);
+  VLOAD(vtst_vector, buffer, q, poly, p, 64, 2);
+  VDUP(vtst_vector2, q, poly, p, 64, 2, 5);
+  TEST_VTST(vtst, q, poly, p, 64, 2);
+  CHECK(TEST_MSG, uint, 64, 2, PRIx64, vtst_expected, "");
 
   /* vmov_n_p64 tests.  */
 #undef TEST_MSG
@@ -767,37 +877,6 @@ int main (void)
     }
   }
 
-  /* vget_lane_p64 tests.  */
-#undef TEST_MSG
-#define TEST_MSG "VGET_LANE/VGETQ_LANE"
-
-#define TEST_VGET_LANE(Q, T1, T2, W, N, L)                                \
-  VECT_VAR(vget_lane_vector, T1, W, N) = vget##Q##_lane_##T2##W(VECT_VAR(vector, T1, W, N), L); \
-  if (VECT_VAR(vget_lane_vector, T1, W, N) != VECT_VAR(vget_lane_expected, T1, W, N)) {                \
-    fprintf(stderr,                                                       \
-           "ERROR in %s (%s line %d in result '%s') at type %s "          \
-           "got 0x%" PRIx##W " != 0x%" PRIx##W "\n",                      \
-           TEST_MSG, __FILE__, __LINE__,                                  \
-           STR(VECT_VAR(vget_lane_expected, T1, W, N)),                   \
-           STR(VECT_NAME(T1, W, N)),                                      \
-           (uint##W##_t)VECT_VAR(vget_lane_vector, T1, W, N),             \
-           (uint##W##_t)VECT_VAR(vget_lane_expected, T1, W, N));          \
-    abort ();                                                             \
-  }
-
-  /* Initialize input values.  */
-  DECL_VARIABLE(vector, poly, 64, 1);
-  DECL_VARIABLE(vector, poly, 64, 2);
-
-  VLOAD(vector, buffer,  , poly, p, 64, 1);
-  VLOAD(vector, buffer, q, poly, p, 64, 2);
-
-  VECT_VAR_DECL(vget_lane_vector, poly, 64, 1);
-  VECT_VAR_DECL(vget_lane_vector, poly, 64, 2);
-
-  TEST_VGET_LANE( , poly, p, 64, 1, 0);
-  TEST_VGET_LANE(q, poly, p, 64, 2, 0);
-
   /* vldx_lane_p64 tests.  */
 #undef TEST_MSG
 #define TEST_MSG "VLDX_LANE/VLDXQ_LANE"