| 1 |
/* { dg-do assemble { target { aarch64*-*-* } } } */ |
| 2 |
/* { dg-skip-if "no optimizations" { *-*-* } { "-O0" } { "" } } */ |
| 3 |
/* { dg-require-effective-target arm_v8_2a_bf16_neon_ok } */ |
| 4 |
/* { dg-add-options arm_v8_2a_bf16_neon } */ |
| 5 |
/* { dg-additional-options "-save-temps" } */ |
| 6 |
|
| 7 |
#include <arm_neon.h> |
| 8 |
|
| 9 |
float32x2_t test_vcreate (float32x2_t r, uint64_t a, uint64_t b) |
| 10 |
{ |
| 11 |
bfloat16x4_t _a = vcreate_bf16(a); |
| 12 |
bfloat16x4_t _b = vcreate_bf16(b); |
| 13 |
|
| 14 |
return vbfdot_f32 (r, _a, _b); |
| 15 |
} |
| 16 |
/* { dg-final { scan-assembler {bfdot\tv[0-9]+.2s, v[0-9]+.4h, v[0-9]+.4h} } } */ |
| 17 |
|
| 18 |
bfloat16x4_t test_vset_lane_bf16 (bfloat16_t a, bfloat16x4_t b) |
| 19 |
{ |
| 20 |
return vset_lane_bf16 (a, b, 3); |
| 21 |
} |
| 22 |
|
| 23 |
bfloat16x8_t test_vsetq_lane_bf16 (bfloat16_t a, bfloat16x8_t b) |
| 24 |
{ |
| 25 |
return vsetq_lane_bf16 (a, b, 7); |
| 26 |
} |
| 27 |
/* { dg-final { scan-assembler-times "ins\\t" 2 } } */ |
| 28 |
|
| 29 |
bfloat16x4_t vdup_test (bfloat16_t a) |
| 30 |
{ |
| 31 |
return vdup_n_bf16 (a); |
| 32 |
} |
| 33 |
/* { dg-final { scan-assembler "dup\\tv\[0-9\]+\.4h, v\[0-9\]+.h\\\[0\\\]" } } */ |
| 34 |
|
| 35 |
bfloat16x8_t vdupq_test (bfloat16_t a) |
| 36 |
{ |
| 37 |
return vdupq_n_bf16 (a); |
| 38 |
} |
| 39 |
|
| 40 |
bfloat16x8_t test_vdupq_lane_bf16 (bfloat16x4_t a) |
| 41 |
{ |
| 42 |
return vdupq_lane_bf16 (a, 1); |
| 43 |
} |
| 44 |
/* { dg-final { scan-assembler-times "dup\\tv\[0-9\]+\.8h, v\[0-9\]+.h\\\[0\\\]" 2 } } */ |
| 45 |
|
| 46 |
bfloat16_t test_vget_lane_bf16 (bfloat16x4_t a) |
| 47 |
{ |
| 48 |
return vget_lane_bf16 (a, 1); |
| 49 |
} |
| 50 |
/* { dg-final { scan-assembler-times "dup\\th\[0-9\]+, v\[0-9\]+\.h\\\[1\\\]" 2 } } */ |
| 51 |
|
| 52 |
bfloat16x4_t test_vdup_lane_bf16 (bfloat16x4_t a) |
| 53 |
{ |
| 54 |
return vdup_lane_bf16 (a, 1); |
| 55 |
} |
| 56 |
/* { dg-final { scan-assembler "dup\\tv\[0-9\]+\.4h, v\[0-9\]+\.h\\\[1\\\]" } } */ |
| 57 |
|
| 58 |
bfloat16x4_t test_vdup_laneq_bf16 (bfloat16x8_t a) |
| 59 |
{ |
| 60 |
return vdup_laneq_bf16 (a, 7); |
| 61 |
} |
| 62 |
/* { dg-final { scan-assembler "dup\\tv\[0-9\]+\.8h, v\[0-9\]+\.h\\\[7\\\]" } } */ |
| 63 |
|
| 64 |
bfloat16x8_t test_vdupq_laneq_bf16 (bfloat16x8_t a) |
| 65 |
{ |
| 66 |
return vdupq_laneq_bf16 (a, 5); |
| 67 |
} |
| 68 |
/* { dg-final { scan-assembler "dup\\tv\[0-9\]+\.8h, v\[0-9\]+\.h\\\[5\\\]" } } */ |
| 69 |
|
| 70 |
bfloat16_t test_vduph_lane_bf16 (bfloat16x4_t a) |
| 71 |
{ |
| 72 |
return vduph_lane_bf16 (a, 3); |
| 73 |
} |
| 74 |
/* { dg-final { scan-assembler "dup\\th\[0-9\]+, v\[0-9\]+\.h\\\[3\\\]" } } */ |
| 75 |
|
| 76 |
bfloat16_t test_vgetq_lane_bf16 (bfloat16x8_t a) |
| 77 |
{ |
| 78 |
return vgetq_lane_bf16 (a, 7); |
| 79 |
} |
| 80 |
|
| 81 |
bfloat16_t test_vduph_laneq_bf16 (bfloat16x8_t a) |
| 82 |
{ |
| 83 |
return vduph_laneq_bf16 (a, 7); |
| 84 |
} |
| 85 |
/* { dg-final { scan-assembler-times "dup\\th\[0-9\]+, v\[0-9\]+\.h\\\[7\\\]" 2 } } */ |