| 1 |
/* { dg-do assemble { target aarch64_asm_fp8dot2_ok } } */ |
| 2 |
/* { dg-do compile { target { ! aarch64_asm_fp8dot2_ok } } } */ |
| 3 |
/* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */ |
| 4 |
|
| 5 |
#include "test_sve_acle.h" |
| 6 |
|
| 7 |
#pragma GCC target "+fp8dot2+fp8dot4" |
| 8 |
#ifdef STREAMING_COMPATIBLE |
| 9 |
#pragma GCC target "+ssve-fp8dot2+ssve-fp8dot4" |
| 10 |
#endif |
| 11 |
|
| 12 |
/* |
| 13 |
** dot_f16_mf8_tied1: |
| 14 |
** msr fpmr, x0 |
| 15 |
** fdot z0\.h, z4\.b, z5\.b |
| 16 |
** ret |
| 17 |
*/ |
| 18 |
TEST_DUAL_Z (dot_f16_mf8_tied1, svfloat16_t, svmfloat8_t, |
| 19 |
z0 = svdot_f16_mf8_fpm (z0, z4, z5, fpm0), |
| 20 |
z0 = svdot_fpm (z0, z4, z5, fpm0)) |
| 21 |
|
| 22 |
/* |
| 23 |
** dot_f16_mf8_tied2: |
| 24 |
** msr fpmr, x0 |
| 25 |
** mov (z[0-9]+)\.d, z0\.d |
| 26 |
** movprfx z0, z4 |
| 27 |
** fdot z0\.h, \1\.b, z1\.b |
| 28 |
** ret |
| 29 |
*/ |
| 30 |
TEST_DUAL_Z_REV (dot_f16_mf8_tied2, svfloat16_t, svmfloat8_t, |
| 31 |
z0_res = svdot_f16_mf8_fpm (z4, z0, z1, fpm0), |
| 32 |
z0_res = svdot_fpm (z4, z0, z1, fpm0)) |
| 33 |
|
| 34 |
/* |
| 35 |
** dot_f16_mf8_tied3: |
| 36 |
** msr fpmr, x0 |
| 37 |
** mov (z[0-9]+)\.d, z0\.d |
| 38 |
** movprfx z0, z4 |
| 39 |
** fdot z0\.h, z1\.b, \1\.b |
| 40 |
** ret |
| 41 |
*/ |
| 42 |
TEST_DUAL_Z_REV (dot_f16_mf8_tied3, svfloat16_t, svmfloat8_t, |
| 43 |
z0_res = svdot_f16_mf8_fpm (z4, z1, z0, fpm0), |
| 44 |
z0_res = svdot_fpm (z4, z1, z0, fpm0)) |
| 45 |
|
| 46 |
/* |
| 47 |
** dot_f16_mf8_untied: |
| 48 |
** msr fpmr, x0 |
| 49 |
** movprfx z0, z1 |
| 50 |
** fdot z0\.h, z4\.b, z5\.b |
| 51 |
** ret |
| 52 |
*/ |
| 53 |
TEST_DUAL_Z (dot_f16_mf8_untied, svfloat16_t, svmfloat8_t, |
| 54 |
z0 = svdot_f16_mf8_fpm (z1, z4, z5, fpm0), |
| 55 |
z0 = svdot_fpm (z1, z4, z5, fpm0)) |
| 56 |
|
| 57 |
/* |
| 58 |
** dot_f32_mf8_tied1: |
| 59 |
** msr fpmr, x0 |
| 60 |
** fdot z0\.s, z4\.b, z5\.b |
| 61 |
** ret |
| 62 |
*/ |
| 63 |
TEST_DUAL_Z (dot_f32_mf8_tied1, svfloat32_t, svmfloat8_t, |
| 64 |
z0 = svdot_f32_mf8_fpm (z0, z4, z5, fpm0), |
| 65 |
z0 = svdot_fpm (z0, z4, z5, fpm0)) |
| 66 |
|
| 67 |
/* |
| 68 |
** dot_f32_mf8_tied2: |
| 69 |
** msr fpmr, x0 |
| 70 |
** mov (z[0-9]+)\.d, z0\.d |
| 71 |
** movprfx z0, z4 |
| 72 |
** fdot z0\.s, \1\.b, z1\.b |
| 73 |
** ret |
| 74 |
*/ |
| 75 |
TEST_DUAL_Z_REV (dot_f32_mf8_tied2, svfloat32_t, svmfloat8_t, |
| 76 |
z0_res = svdot_f32_mf8_fpm (z4, z0, z1, fpm0), |
| 77 |
z0_res = svdot_fpm (z4, z0, z1, fpm0)) |
| 78 |
|
| 79 |
/* |
| 80 |
** dot_f32_mf8_tied3: |
| 81 |
** msr fpmr, x0 |
| 82 |
** mov (z[0-9]+)\.d, z0\.d |
| 83 |
** movprfx z0, z4 |
| 84 |
** fdot z0\.s, z1\.b, \1\.b |
| 85 |
** ret |
| 86 |
*/ |
| 87 |
TEST_DUAL_Z_REV (dot_f32_mf8_tied3, svfloat32_t, svmfloat8_t, |
| 88 |
z0_res = svdot_f32_mf8_fpm (z4, z1, z0, fpm0), |
| 89 |
z0_res = svdot_fpm (z4, z1, z0, fpm0)) |
| 90 |
|
| 91 |
/* |
| 92 |
** dot_f32_mf8_untied: |
| 93 |
** msr fpmr, x0 |
| 94 |
** movprfx z0, z1 |
| 95 |
** fdot z0\.s, z4\.b, z5\.b |
| 96 |
** ret |
| 97 |
*/ |
| 98 |
TEST_DUAL_Z (dot_f32_mf8_untied, svfloat32_t, svmfloat8_t, |
| 99 |
z0 = svdot_f32_mf8_fpm (z1, z4, z5, fpm0), |
| 100 |
z0 = svdot_fpm (z1, z4, z5, fpm0)) |
| 101 |
|