aarch64: Use memcpy to copy structures in bfloat vst* intrinsics

Use __builtin_memcpy to copy vector structures instead of using a
union - or constructing a new opaque structure one vector at a time -
in each of the vst[234][q] and vst1[q]_x[234] bfloat Neon intrinsics
in arm_neon.h.

Add new code generation tests to verify that superfluous move
instructions are not generated for the vst[234]q or vst1q_x[234]
bfloat intrinsics.

gcc/ChangeLog:

2021-07-30  Jonathan Wright  <jonathan.wright@arm.com>

	* config/aarch64/arm_neon.h (vst1_bf16_x2): Use
	__builtin_memcpy instead of constructing an additional
	__builtin_aarch64_simd_oi one vector at a time.
	(vst1q_bf16_x2): Likewise.
	(vst1_bf16_x3): Use __builtin_memcpy instead of constructing
	an additional __builtin_aarch64_simd_ci one vector at a time.
	(vst1q_bf16_x3): Likewise.
	(vst1_bf16_x4): Use __builtin_memcpy instead of a union.
	(vst1q_bf16_x4): Likewise.
	(vst2_bf16): Use __builtin_memcpy instead of constructing an
	additional __builtin_aarch64_simd_oi one vector at a time.
	(vst2q_bf16): Likewise.
	(vst3_bf16): Use __builtin_memcpy instead of constructing an
	additional __builtin_aarch64_simd_ci mode one vector at a
	time.
	(vst3q_bf16): Likewise.
	(vst4_bf16): Use __builtin_memcpy instead of constructing an
	additional __builtin_aarch64_simd_xi one vector at a time.
	(vst4q_bf16): Likewise.

gcc/testsuite/ChangeLog:

	* gcc.target/aarch64/vector_structure_intrinsics.c: Add new
	tests.

2 files changed, 33 insertions, 38 deletions

diff --git a/gcc/config/aarch64/arm_neon.h b/gcc/config/aarch64/arm_neon.h
index cbae61d..390cf9a 100644
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -33839,8 +33839,7 @@ vst1_bf16_x2 (bfloat16_t * __a, bfloat16x4x2_t __val)
   bfloat16x8x2_t __temp;
   __temp.val[0] = vcombine_bf16 (__val.val[0], vcreate_bf16 (__AARCH64_UINT64_C (0)));
   __temp.val[1] = vcombine_bf16 (__val.val[1], vcreate_bf16 (__AARCH64_UINT64_C (0)));
-  __o = __builtin_aarch64_set_qregoiv8bf (__o, __temp.val[0], 0);
-  __o = __builtin_aarch64_set_qregoiv8bf (__o, __temp.val[1], 1);
+  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
   __builtin_aarch64_st1x2v4bf (__a, __o);
 }
 
@@ -33849,8 +33848,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_bf16_x2 (bfloat16_t * __a, bfloat16x8x2_t __val)
 {
   __builtin_aarch64_simd_oi __o;
-  __o = __builtin_aarch64_set_qregoiv8bf (__o, __val.val[0], 0);
-  __o = __builtin_aarch64_set_qregoiv8bf (__o, __val.val[1], 1);
+  __builtin_memcpy (&__o, &__val, sizeof (__val));
   __builtin_aarch64_st1x2v8bf (__a, __o);
 }
 
@@ -33863,9 +33861,7 @@ vst1_bf16_x3 (bfloat16_t * __a, bfloat16x4x3_t __val)
   __temp.val[0] = vcombine_bf16 (__val.val[0], vcreate_bf16 (__AARCH64_UINT64_C (0)));
   __temp.val[1] = vcombine_bf16 (__val.val[1], vcreate_bf16 (__AARCH64_UINT64_C (0)));
   __temp.val[2] = vcombine_bf16 (__val.val[2], vcreate_bf16 (__AARCH64_UINT64_C (0)));
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __temp.val[0], 0);
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __temp.val[1], 1);
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __temp.val[2], 2);
+  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
   __builtin_aarch64_st1x3v4bf ((__builtin_aarch64_simd_bf *) __a, __o);
 }
 
@@ -33874,26 +33870,31 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst1q_bf16_x3 (bfloat16_t * __a, bfloat16x8x3_t __val)
 {
   __builtin_aarch64_simd_ci __o;
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __val.val[0], 0);
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __val.val[1], 1);
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __val.val[2], 2);
+  __builtin_memcpy (&__o, &__val, sizeof (__val));
   __builtin_aarch64_st1x3v8bf ((__builtin_aarch64_simd_bf *) __a, __o);
 }
 
 __extension__ extern __inline void
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-vst1_bf16_x4 (bfloat16_t * __a, bfloat16x4x4_t val)
+vst1_bf16_x4 (bfloat16_t * __a, bfloat16x4x4_t __val)
 {
-  union { bfloat16x4x4_t __i; __builtin_aarch64_simd_xi __o; } __u = { val };
-  __builtin_aarch64_st1x4v4bf ((__builtin_aarch64_simd_bf *) __a, __u.__o);
+  __builtin_aarch64_simd_xi __o;
+  bfloat16x8x4_t __temp;
+  __temp.val[0] = vcombine_bf16 (__val.val[0], vcreate_bf16 (__AARCH64_UINT64_C (0)));
+  __temp.val[1] = vcombine_bf16 (__val.val[1], vcreate_bf16 (__AARCH64_UINT64_C (0)));
+  __temp.val[2] = vcombine_bf16 (__val.val[2], vcreate_bf16 (__AARCH64_UINT64_C (0)));
+  __temp.val[3] = vcombine_bf16 (__val.val[3], vcreate_bf16 (__AARCH64_UINT64_C (0)));
+  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
+  __builtin_aarch64_st1x4v4bf ((__builtin_aarch64_simd_bf *) __a, __o);
 }
 
 __extension__ extern __inline void
 __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
-vst1q_bf16_x4 (bfloat16_t * __a, bfloat16x8x4_t val)
+vst1q_bf16_x4 (bfloat16_t * __a, bfloat16x8x4_t __val)
 {
-  union { bfloat16x8x4_t __i; __builtin_aarch64_simd_xi __o; } __u = { val };
-  __builtin_aarch64_st1x4v8bf ((__builtin_aarch64_simd_bf *) __a, __u.__o);
+  __builtin_aarch64_simd_xi __o;
+  __builtin_memcpy (&__o, &__val, sizeof (__val));
+  __builtin_aarch64_st1x4v8bf ((__builtin_aarch64_simd_bf *) __a, __o);
 }
 
 __extension__ extern __inline void
@@ -33925,8 +33926,7 @@ vst2_bf16 (bfloat16_t * __a, bfloat16x4x2_t __val)
   bfloat16x8x2_t __temp;
   __temp.val[0] = vcombine_bf16 (__val.val[0], vcreate_bf16 (__AARCH64_UINT64_C (0)));
   __temp.val[1] = vcombine_bf16 (__val.val[1], vcreate_bf16 (__AARCH64_UINT64_C (0)));
-  __o = __builtin_aarch64_set_qregoiv8bf (__o, __temp.val[0], 0);
-  __o = __builtin_aarch64_set_qregoiv8bf (__o, __temp.val[1], 1);
+  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
   __builtin_aarch64_st2v4bf (__a, __o);
 }
 
@@ -33935,8 +33935,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst2q_bf16 (bfloat16_t * __a, bfloat16x8x2_t __val)
 {
   __builtin_aarch64_simd_oi __o;
-  __o = __builtin_aarch64_set_qregoiv8bf (__o, __val.val[0], 0);
-  __o = __builtin_aarch64_set_qregoiv8bf (__o, __val.val[1], 1);
+  __builtin_memcpy (&__o, &__val, sizeof (__val));
   __builtin_aarch64_st2v8bf (__a, __o);
 }
 
@@ -33949,9 +33948,7 @@ vst3_bf16 (bfloat16_t * __a, bfloat16x4x3_t __val)
   __temp.val[0] = vcombine_bf16 (__val.val[0], vcreate_bf16 (__AARCH64_UINT64_C (0)));
   __temp.val[1] = vcombine_bf16 (__val.val[1], vcreate_bf16 (__AARCH64_UINT64_C (0)));
   __temp.val[2] = vcombine_bf16 (__val.val[2], vcreate_bf16 (__AARCH64_UINT64_C (0)));
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __temp.val[0], 0);
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __temp.val[1], 1);
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __temp.val[2], 2);
+  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
   __builtin_aarch64_st3v4bf ((__builtin_aarch64_simd_bf *) __a, __o);
 }
 
@@ -33960,9 +33957,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst3q_bf16 (bfloat16_t * __a, bfloat16x8x3_t __val)
 {
   __builtin_aarch64_simd_ci __o;
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __val.val[0], 0);
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __val.val[1], 1);
-  __o = __builtin_aarch64_set_qregciv8bf (__o, (bfloat16x8_t) __val.val[2], 2);
+  __builtin_memcpy (&__o, &__val, sizeof (__val));
   __builtin_aarch64_st3v8bf ((__builtin_aarch64_simd_bf *) __a, __o);
 }
 
@@ -33976,10 +33971,7 @@ vst4_bf16 (bfloat16_t * __a, bfloat16x4x4_t __val)
   __temp.val[1] = vcombine_bf16 (__val.val[1], vcreate_bf16 (__AARCH64_UINT64_C (0)));
   __temp.val[2] = vcombine_bf16 (__val.val[2], vcreate_bf16 (__AARCH64_UINT64_C (0)));
   __temp.val[3] = vcombine_bf16 (__val.val[3], vcreate_bf16 (__AARCH64_UINT64_C (0)));
-  __o = __builtin_aarch64_set_qregxiv8bf (__o, (bfloat16x8_t) __temp.val[0], 0);
-  __o = __builtin_aarch64_set_qregxiv8bf (__o, (bfloat16x8_t) __temp.val[1], 1);
-  __o = __builtin_aarch64_set_qregxiv8bf (__o, (bfloat16x8_t) __temp.val[2], 2);
-  __o = __builtin_aarch64_set_qregxiv8bf (__o, (bfloat16x8_t) __temp.val[3], 3);
+  __builtin_memcpy (&__o, &__temp, sizeof (__temp));
   __builtin_aarch64_st4v4bf ((__builtin_aarch64_simd_bf *) __a, __o);
 }
 
@@ -33988,10 +33980,7 @@ __attribute__ ((__always_inline__, __gnu_inline__, __artificial__))
 vst4q_bf16 (bfloat16_t * __a, bfloat16x8x4_t __val)
 {
   __builtin_aarch64_simd_xi __o;
-  __o = __builtin_aarch64_set_qregxiv8bf (__o, (bfloat16x8_t) __val.val[0], 0);
-  __o = __builtin_aarch64_set_qregxiv8bf (__o, (bfloat16x8_t) __val.val[1], 1);
-  __o = __builtin_aarch64_set_qregxiv8bf (__o, (bfloat16x8_t) __val.val[2], 2);
-  __o = __builtin_aarch64_set_qregxiv8bf (__o, (bfloat16x8_t) __val.val[3], 3);
+  __builtin_memcpy (&__o, &__val, sizeof (__val));
   __builtin_aarch64_st4v8bf ((__builtin_aarch64_simd_bf *) __a, __o);
 }
 
diff --git a/gcc/testsuite/gcc.target/aarch64/vector_structure_intrinsics.c b/gcc/testsuite/gcc.target/aarch64/vector_structure_intrinsics.c
index e491d46..3e7e572 100644
--- a/gcc/testsuite/gcc.target/aarch64/vector_structure_intrinsics.c
+++ b/gcc/testsuite/gcc.target/aarch64/vector_structure_intrinsics.c
@@ -95,6 +95,7 @@ TEST_STX (vst4q, int16x8x4_t, int16_t*, s16);
 TEST_STX (vst4q, uint16x8x4_t, uint16_t*, u16);
 TEST_STX (vst4q, poly16x8x4_t, poly16_t*, p16);
 TEST_STX (vst4q, float16x8x4_t, float16_t*, f16);
+TEST_STX (vst4q, bfloat16x8x4_t, bfloat16_t*, bf16);
 TEST_STX (vst4q, int32x4x4_t, int32_t*, s32);
 TEST_STX (vst4q, uint32x4x4_t, uint32_t*, u32);
 TEST_STX (vst4q, float32x4x4_t, float32_t*, f32);
@@ -110,6 +111,7 @@ TEST_STX (vst2q, int16x8x2_t, int16_t*, s16);
 TEST_STX (vst2q, uint16x8x2_t, uint16_t*, u16);
 TEST_STX (vst2q, poly16x8x2_t, poly16_t*, p16);
 TEST_STX (vst2q, float16x8x2_t, float16_t*, f16);
+TEST_STX (vst2q, bfloat16x8x2_t, bfloat16_t*, bf16);
 TEST_STX (vst2q, int32x4x2_t, int32_t*, s32);
 TEST_STX (vst2q, uint32x4x2_t, uint32_t*, u32);
 TEST_STX (vst2q, float32x4x2_t, float32_t*, f32);
@@ -131,6 +133,7 @@ TEST_ST3 (vst3q, int16x8x3_t, int16_t*, s16);
 TEST_ST3 (vst3q, uint16x8x3_t, uint16_t*, u16);
 TEST_ST3 (vst3q, poly16x8x3_t, poly16_t*, p16);
 TEST_ST3 (vst3q, float16x8x3_t, float16_t*, f16);
+TEST_ST3 (vst3q, bfloat16x8x3_t, bfloat16_t*, bf16);
 TEST_ST3 (vst3q, int32x4x3_t, int32_t*, s32);
 TEST_ST3 (vst3q, uint32x4x3_t, uint32_t*, u32);
 TEST_ST3 (vst3q, float32x4x3_t, float32_t*, f32);
@@ -212,6 +215,7 @@ TEST_ST1xN (vst1q, int16x8x4_t, int16_t*, s16, x4);
 TEST_ST1xN (vst1q, uint16x8x4_t, uint16_t*, u16, x4);
 TEST_ST1xN (vst1q, poly16x8x4_t, poly16_t*, p16, x4);
 TEST_ST1xN (vst1q, float16x8x4_t, float16_t*, f16, x4);
+TEST_ST1xN (vst1q, bfloat16x8x4_t, bfloat16_t*, bf16, x4);
 TEST_ST1xN (vst1q, int32x4x4_t, int32_t*, s32, x4);
 TEST_ST1xN (vst1q, uint32x4x4_t, uint32_t*, u32, x4);
 TEST_ST1xN (vst1q, float32x4x4_t, float32_t*, f32, x4);
@@ -227,6 +231,7 @@ TEST_ST1xN (vst1q, int16x8x2_t, int16_t*, s16, x2);
 TEST_ST1xN (vst1q, uint16x8x2_t, uint16_t*, u16, x2);
 TEST_ST1xN (vst1q, poly16x8x2_t, poly16_t*, p16, x2);
 TEST_ST1xN (vst1q, float16x8x2_t, float16_t*, f16, x2);
+TEST_ST1xN (vst1q, bfloat16x8x2_t, bfloat16_t*, bf16, x2);
 TEST_ST1xN (vst1q, int32x4x2_t, int32_t*, s32, x2);
 TEST_ST1xN (vst1q, uint32x4x2_t, uint32_t*, u32, x2);
 TEST_ST1xN (vst1q, float32x4x2_t, float32_t*, f32, x2);
@@ -249,6 +254,7 @@ TEST_ST1x3 (vst1q, int16x8x3_t, int16_t*, s16, x3);
 TEST_ST1x3 (vst1q, uint16x8x3_t, uint16_t*, u16, x3);
 TEST_ST1x3 (vst1q, poly16x8x3_t, poly16_t*, p16, x3);
 TEST_ST1x3 (vst1q, float16x8x3_t, float16_t*, f16, x3);
+TEST_ST1x3 (vst1q, bfloat16x8x3_t, bfloat16_t*, bf16, x3);
 TEST_ST1x3 (vst1q, int32x4x3_t, int32_t*, s32, x3);
 TEST_ST1x3 (vst1q, uint32x4x3_t, uint32_t*, u32, x3);
 TEST_ST1x3 (vst1q, float32x4x3_t, float32_t*, f32, x3);
@@ -261,7 +267,7 @@ TEST_ST1x3 (vst1q, float64x2x3_t, float64_t*, f64, x3);
 
 /* { dg-final { scan-assembler-times "tbl\\t" 18} }  */
 /* { dg-final { scan-assembler-times "tbx\\t" 18} }  */
-/* { dg-final { scan-assembler-times "st4\\t" 29} }  */
-/* { dg-final { scan-assembler-times "st3\\t" 29} }  */
-/* { dg-final { scan-assembler-times "st2\\t" 29} }  */
-/* { dg-final { scan-assembler-times "st1\\t" 42} }  */
+/* { dg-final { scan-assembler-times "st4\\t" 30} }  */
+/* { dg-final { scan-assembler-times "st3\\t" 30} }  */
+/* { dg-final { scan-assembler-times "st2\\t" 30} }  */
+/* { dg-final { scan-assembler-times "st1\\t" 45} }  */