| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-require-effective-target arm_v8_2a_fp16_neon_ok } */ |
| 3 |
/* { dg-options "-O2 -ffast-math" } */ |
| 4 |
/* { dg-add-options arm_v8_2a_fp16_neon } */ |
| 5 |
|
| 6 |
/* Test compiler use of FP16 FMA/FMS instructions with -ffast-math. */ |
| 7 |
|
| 8 |
#include <arm_neon.h> |
| 9 |
|
| 10 |
float16x4_t |
| 11 |
test_vfma_1 (float16x4_t a, float16x4_t b, float16x4_t c) |
| 12 |
{ |
| 13 |
return vadd_f16 (vmul_f16 (a, b), c); |
| 14 |
} |
| 15 |
|
| 16 |
float16x4_t |
| 17 |
test_vfma_2 (float16x4_t a, float16x4_t b, float16x4_t c) |
| 18 |
{ |
| 19 |
return vsub_f16 (vmul_f16 (a, b), vneg_f16 (c)); |
| 20 |
} |
| 21 |
|
| 22 |
float16x4_t |
| 23 |
test_vfma_3 (float16x4_t a, float16x4_t b, float16x4_t c) |
| 24 |
{ |
| 25 |
return vsub_f16 (vmul_f16 (vneg_f16 (a), vneg_f16 (b)), vneg_f16 (c)); |
| 26 |
} |
| 27 |
|
| 28 |
float16x4_t |
| 29 |
test_vfma_4 (float16x4_t a, float16x4_t b, float16x4_t c) |
| 30 |
{ |
| 31 |
return vsub_f16 (vmul_f16 (a, b), vneg_f16 (c)); |
| 32 |
} |
| 33 |
/* { dg-final { scan-assembler-times {vfma\.f16\td[0-9]+, d[0-9]+, d[0-9]+} 4 } } */ |
| 34 |
|
| 35 |
float16x8_t |
| 36 |
test_vfmaq_1 (float16x8_t a, float16x8_t b, float16x8_t c) |
| 37 |
{ |
| 38 |
return vaddq_f16 (vmulq_f16 (a, b), c); |
| 39 |
} |
| 40 |
|
| 41 |
float16x8_t |
| 42 |
test_vfmaq_2 (float16x8_t a, float16x8_t b, float16x8_t c) |
| 43 |
{ |
| 44 |
return vsubq_f16 (vmulq_f16 (a, b), vnegq_f16 (c)); |
| 45 |
} |
| 46 |
|
| 47 |
float16x8_t |
| 48 |
test_vfmaq_3 (float16x8_t a, float16x8_t b, float16x8_t c) |
| 49 |
{ |
| 50 |
return vsubq_f16 (vmulq_f16 (vnegq_f16 (a), vnegq_f16 (b)), vnegq_f16 (c)); |
| 51 |
} |
| 52 |
|
| 53 |
float16x8_t |
| 54 |
test_vfmaq_4 (float16x8_t a, float16x8_t b, float16x8_t c) |
| 55 |
{ |
| 56 |
return vsubq_f16 (vmulq_f16 (a, b), vnegq_f16 (c)); |
| 57 |
} |
| 58 |
/* { dg-final { scan-assembler-times {vfma\.f16\tq[0-9]+, q[0-9]+, q[0-9]+} 4 } } */ |
| 59 |
|
| 60 |
float16x4_t |
| 61 |
test_vfms_1 (float16x4_t a, float16x4_t b, float16x4_t c) |
| 62 |
{ |
| 63 |
return vsub_f16 (c, vmul_f16 (a, b)); |
| 64 |
} |
| 65 |
|
| 66 |
float16x4_t |
| 67 |
test_vfms_2 (float16x4_t a, float16x4_t b, float16x4_t c) |
| 68 |
{ |
| 69 |
return vsub_f16 (a, vmul_f16 (b, c)); |
| 70 |
} |
| 71 |
|
| 72 |
float16x4_t |
| 73 |
test_vfms_3 (float16x4_t a, float16x4_t b, float16x4_t c) |
| 74 |
{ |
| 75 |
return vadd_f16 (vmul_f16 (vneg_f16 (a), b), c); |
| 76 |
} |
| 77 |
|
| 78 |
float16x4_t |
| 79 |
test_vfms_4 (float16x4_t a, float16x4_t b, float16x4_t c) |
| 80 |
{ |
| 81 |
return vadd_f16 (vmul_f16 (a, vneg_f16 (b)), c); |
| 82 |
} |
| 83 |
/* { dg-final { scan-assembler-times {vfms\.f16\td[0-9]+, d[0-9]+, d[0-9]+} 4 } } */ |
| 84 |
|
| 85 |
float16x8_t |
| 86 |
test_vfmsq_1 (float16x8_t a, float16x8_t b, float16x8_t c) |
| 87 |
{ |
| 88 |
return vsubq_f16 (c, vmulq_f16 (a, b)); |
| 89 |
} |
| 90 |
|
| 91 |
float16x8_t |
| 92 |
test_vfmsq_2 (float16x8_t a, float16x8_t b, float16x8_t c) |
| 93 |
{ |
| 94 |
return vsubq_f16 (a, vmulq_f16 (b, c)); |
| 95 |
} |
| 96 |
|
| 97 |
float16x8_t |
| 98 |
test_vfmsq_3 (float16x8_t a, float16x8_t b, float16x8_t c) |
| 99 |
{ |
| 100 |
return vaddq_f16 (vmulq_f16 (vnegq_f16 (a), b), c); |
| 101 |
} |
| 102 |
|
| 103 |
float16x8_t |
| 104 |
test_vfmsq_4 (float16x8_t a, float16x8_t b, float16x8_t c) |
| 105 |
{ |
| 106 |
return vaddq_f16 (vmulq_f16 (a, vnegq_f16 (b)), c); |
| 107 |
} |
| 108 |
/* { dg-final { scan-assembler-times {vfms\.f16\tq[0-9]+, q[0-9]+, q[0-9]+} 4 } } */ |