| 1 |
/* Test the vmulxq_f32 AArch64 SIMD intrinsic. */ |
| 2 |
|
| 3 |
/* { dg-do run } */ |
| 4 |
/* { dg-options "-save-temps -O3" } */ |
| 5 |
|
| 6 |
#include "arm_neon.h" |
| 7 |
|
| 8 |
extern void abort (void); |
| 9 |
|
| 10 |
void __attribute__ ((noinline)) |
| 11 |
test_case (float32_t v1[4], float32_t v2[4], float32_t e[4]) |
| 12 |
{ |
| 13 |
int i; |
| 14 |
float32x4_t vec1_1 = vld1q_f32 (v1); |
| 15 |
float32x4_t vec1_2 = vld1q_f32 (v2); |
| 16 |
|
| 17 |
float32x4_t actual1 = vmulxq_f32 (vec1_1, vec1_2); |
| 18 |
float32_t actual[4]; |
| 19 |
vst1q_f32 (actual, actual1); |
| 20 |
|
| 21 |
for (i = 0; i < 4; ++i) |
| 22 |
if (actual[i] != e[i]) |
| 23 |
abort (); |
| 24 |
} |
| 25 |
|
| 26 |
int |
| 27 |
main (void) |
| 28 |
{ |
| 29 |
float32_t v1 = 3.14159265359; |
| 30 |
float32_t v2 = 1.383894; |
| 31 |
float32_t v3 = -2.71828; |
| 32 |
float32_t v4 = -3.4891931; |
| 33 |
|
| 34 |
float32_t v1_1[] = {v1, v2, v3, v4}; |
| 35 |
float32_t v1_2[] = {v3, v4, v1, v2}; |
| 36 |
float32_t e1[] = {v1 * v3, v2 * v4, v3 * v1, v4 * v2}; |
| 37 |
test_case (v1_1, v1_2, e1); |
| 38 |
|
| 39 |
float32_t v2_1[] = {0, -0.0, 0, -0.0}; |
| 40 |
float32_t v2_2[] = {-__builtin_huge_valf (), -__builtin_huge_valf (), |
| 41 |
__builtin_huge_valf (), __builtin_huge_valf () }; |
| 42 |
float32_t e2[] = {-2.0, 2.0, 2.0, -2.0}; |
| 43 |
test_case (v2_1, v2_2, e2); |
| 44 |
|
| 45 |
return 0; |
| 46 |
} |
| 47 |
|
| 48 |
/* { dg-final { scan-assembler-times "fmulx\[ \t\]+\[vV\]\[0-9\]+\.4s, ?\[vV\]\[0-9\]+\.4s, ?\[vV\]\[0-9\]+\.4s\n" 1 } } */ |