| 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 |
** bfmlalt_lane_0_f32_tied1: |
| 9 |
** bfmlalt z0\.s, z4\.h, z5\.h\[0\] |
| 10 |
** ret |
| 11 |
*/ |
| 12 |
TEST_DUAL_Z (bfmlalt_lane_0_f32_tied1, svfloat32_t, svbfloat16_t, |
| 13 |
z0 = svbfmlalt_lane_f32 (z0, z4, z5, 0), |
| 14 |
z0 = svbfmlalt_lane (z0, z4, z5, 0)) |
| 15 |
|
| 16 |
/* |
| 17 |
** bfmlalt_lane_0_f32_tied2: |
| 18 |
** mov (z[0-9]+)\.d, z0\.d |
| 19 |
** movprfx z0, z4 |
| 20 |
** bfmlalt z0\.s, \1\.h, z1\.h\[0\] |
| 21 |
** ret |
| 22 |
*/ |
| 23 |
TEST_DUAL_Z_REV (bfmlalt_lane_0_f32_tied2, svfloat32_t, svbfloat16_t, |
| 24 |
z0_res = svbfmlalt_lane_f32 (z4, z0, z1, 0), |
| 25 |
z0_res = svbfmlalt_lane (z4, z0, z1, 0)) |
| 26 |
|
| 27 |
/* |
| 28 |
** bfmlalt_lane_0_f32_tied3: |
| 29 |
** mov (z[0-9]+)\.d, z0\.d |
| 30 |
** movprfx z0, z4 |
| 31 |
** bfmlalt z0\.s, z1\.h, \1\.h\[0\] |
| 32 |
** ret |
| 33 |
*/ |
| 34 |
TEST_DUAL_Z_REV (bfmlalt_lane_0_f32_tied3, svfloat32_t, svbfloat16_t, |
| 35 |
z0_res = svbfmlalt_lane_f32 (z4, z1, z0, 0), |
| 36 |
z0_res = svbfmlalt_lane (z4, z1, z0, 0)) |
| 37 |
|
| 38 |
/* |
| 39 |
** bfmlalt_lane_0_f32_untied: |
| 40 |
** movprfx z0, z1 |
| 41 |
** bfmlalt z0\.s, z4\.h, z5\.h\[0\] |
| 42 |
** ret |
| 43 |
*/ |
| 44 |
TEST_DUAL_Z (bfmlalt_lane_0_f32_untied, svfloat32_t, svbfloat16_t, |
| 45 |
z0 = svbfmlalt_lane_f32 (z1, z4, z5, 0), |
| 46 |
z0 = svbfmlalt_lane (z1, z4, z5, 0)) |
| 47 |
|
| 48 |
/* |
| 49 |
** bfmlalt_lane_1_f32: |
| 50 |
** bfmlalt z0\.s, z4\.h, z5\.h\[1\] |
| 51 |
** ret |
| 52 |
*/ |
| 53 |
TEST_DUAL_Z (bfmlalt_lane_1_f32, svfloat32_t, svbfloat16_t, |
| 54 |
z0 = svbfmlalt_lane_f32 (z0, z4, z5, 1), |
| 55 |
z0 = svbfmlalt_lane (z0, z4, z5, 1)) |
| 56 |
|
| 57 |
/* |
| 58 |
** bfmlalt_lane_7_f32: |
| 59 |
** bfmlalt z0\.s, z4\.h, z5\.h\[7\] |
| 60 |
** ret |
| 61 |
*/ |
| 62 |
TEST_DUAL_Z (bfmlalt_lane_7_f32, svfloat32_t, svbfloat16_t, |
| 63 |
z0 = svbfmlalt_lane_f32 (z0, z4, z5, 7), |
| 64 |
z0 = svbfmlalt_lane (z0, z4, z5, 7)) |
| 65 |
|
| 66 |
/* |
| 67 |
** bfmlalt_lane_z8_f32: |
| 68 |
** str d8, \[sp, -16\]! |
| 69 |
** mov (z[0-7])\.d, z8\.d |
| 70 |
** bfmlalt z0\.s, z1\.h, \1\.h\[1\] |
| 71 |
** ldr d8, \[sp\], 16 |
| 72 |
** ret |
| 73 |
*/ |
| 74 |
TEST_DUAL_LANE_REG (bfmlalt_lane_z8_f32, svfloat32_t, svbfloat16_t, z8, |
| 75 |
z0 = svbfmlalt_lane_f32 (z0, z1, z8, 1), |
| 76 |
z0 = svbfmlalt_lane (z0, z1, z8, 1)) |
| 77 |
|
| 78 |
/* |
| 79 |
** bfmlalt_lane_z16_f32: |
| 80 |
** mov (z[0-7])\.d, z16\.d |
| 81 |
** bfmlalt z0\.s, z1\.h, \1\.h\[1\] |
| 82 |
** ret |
| 83 |
*/ |
| 84 |
TEST_DUAL_LANE_REG (bfmlalt_lane_z16_f32, svfloat32_t, svbfloat16_t, z16, |
| 85 |
z0 = svbfmlalt_lane_f32 (z0, z1, z16, 1), |
| 86 |
z0 = svbfmlalt_lane (z0, z1, z16, 1)) |