| 1 |
/* { dg-additional-options "-march=armv8.2-a+sve+bf16" } */ |
| 2 |
/* { dg-require-effective-target aarch64_asm_bf16_ok } */ |
| 3 |
/* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */ |
| 4 |
|
| 5 |
#include "test_sve_acle.h" |
| 6 |
|
| 7 |
/* |
| 8 |
** bfmlalb_f32_tied1: |
| 9 |
** bfmlalb z0\.s, z4\.h, z5\.h |
| 10 |
** ret |
| 11 |
*/ |
| 12 |
TEST_DUAL_Z (bfmlalb_f32_tied1, svfloat32_t, svbfloat16_t, |
| 13 |
z0 = svbfmlalb_f32 (z0, z4, z5), |
| 14 |
z0 = svbfmlalb (z0, z4, z5)) |
| 15 |
|
| 16 |
/* |
| 17 |
** bfmlalb_f32_tied2: |
| 18 |
** mov (z[0-9]+)\.d, z0\.d |
| 19 |
** movprfx z0, z4 |
| 20 |
** bfmlalb z0\.s, \1\.h, z1\.h |
| 21 |
** ret |
| 22 |
*/ |
| 23 |
TEST_DUAL_Z_REV (bfmlalb_f32_tied2, svfloat32_t, svbfloat16_t, |
| 24 |
z0_res = svbfmlalb_f32 (z4, z0, z1), |
| 25 |
z0_res = svbfmlalb (z4, z0, z1)) |
| 26 |
|
| 27 |
/* |
| 28 |
** bfmlalb_f32_tied3: |
| 29 |
** mov (z[0-9]+)\.d, z0\.d |
| 30 |
** movprfx z0, z4 |
| 31 |
** bfmlalb z0\.s, z1\.h, \1\.h |
| 32 |
** ret |
| 33 |
*/ |
| 34 |
TEST_DUAL_Z_REV (bfmlalb_f32_tied3, svfloat32_t, svbfloat16_t, |
| 35 |
z0_res = svbfmlalb_f32 (z4, z1, z0), |
| 36 |
z0_res = svbfmlalb (z4, z1, z0)) |
| 37 |
|
| 38 |
/* |
| 39 |
** bfmlalb_f32_untied: |
| 40 |
** movprfx z0, z1 |
| 41 |
** bfmlalb z0\.s, z4\.h, z5\.h |
| 42 |
** ret |
| 43 |
*/ |
| 44 |
TEST_DUAL_Z (bfmlalb_f32_untied, svfloat32_t, svbfloat16_t, |
| 45 |
z0 = svbfmlalb_f32 (z1, z4, z5), |
| 46 |
z0 = svbfmlalb (z1, z4, z5)) |
| 47 |
|
| 48 |
/* |
| 49 |
** bfmlalb_h7_f32_tied1: |
| 50 |
** mov (z[0-9]+\.h), h7 |
| 51 |
** bfmlalb z0\.s, z4\.h, \1 |
| 52 |
** ret |
| 53 |
*/ |
| 54 |
TEST_DUAL_ZD (bfmlalb_h7_f32_tied1, svfloat32_t, svbfloat16_t, bfloat16_t, |
| 55 |
z0 = svbfmlalb_n_f32 (z0, z4, d7), |
| 56 |
z0 = svbfmlalb (z0, z4, d7)) |
| 57 |
|
| 58 |
/* |
| 59 |
** bfmlalb_h7_f32_untied: |
| 60 |
** mov (z[0-9]+\.h), h7 |
| 61 |
** movprfx z0, z1 |
| 62 |
** bfmlalb z0\.s, z4\.h, \1 |
| 63 |
** ret |
| 64 |
*/ |
| 65 |
TEST_DUAL_ZD (bfmlalb_h7_f32_untied, svfloat32_t, svbfloat16_t, bfloat16_t, |
| 66 |
z0 = svbfmlalb_n_f32 (z1, z4, d7), |
| 67 |
z0 = svbfmlalb (z1, z4, d7)) |