| 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_lane_0_f16_tied1: |
| 14 |
** msr fpmr, x0 |
| 15 |
** fdot z0\.h, z4\.b, z5\.b\[0\] |
| 16 |
** ret |
| 17 |
*/ |
| 18 |
TEST_DUAL_Z (dot_lane_0_f16_tied1, svfloat16_t, svmfloat8_t, |
| 19 |
z0 = svdot_lane_f16_mf8_fpm (z0, z4, z5, 0, fpm0), |
| 20 |
z0 = svdot_lane_fpm (z0, z4, z5, 0, fpm0)) |
| 21 |
|
| 22 |
/* |
| 23 |
** dot_lane_0_f16_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\[0\] |
| 28 |
** ret |
| 29 |
*/ |
| 30 |
TEST_DUAL_Z_REV (dot_lane_0_f16_tied2, svfloat16_t, svmfloat8_t, |
| 31 |
z0_res = svdot_lane_f16_mf8_fpm (z4, z0, z1, 0, fpm0), |
| 32 |
z0_res = svdot_lane_fpm (z4, z0, z1, 0, fpm0)) |
| 33 |
|
| 34 |
/* |
| 35 |
** dot_lane_0_f16_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\[0\] |
| 40 |
** ret |
| 41 |
*/ |
| 42 |
TEST_DUAL_Z_REV (dot_lane_0_f16_tied3, svfloat16_t, svmfloat8_t, |
| 43 |
z0_res = svdot_lane_f16_mf8_fpm (z4, z1, z0, 0, fpm0), |
| 44 |
z0_res = svdot_lane_fpm (z4, z1, z0, 0, fpm0)) |
| 45 |
|
| 46 |
/* |
| 47 |
** dot_lane_0_f16_untied: |
| 48 |
** msr fpmr, x0 |
| 49 |
** movprfx z0, z1 |
| 50 |
** fdot z0\.h, z4\.b, z5\.b\[0\] |
| 51 |
** ret |
| 52 |
*/ |
| 53 |
TEST_DUAL_Z (dot_lane_0_f16_untied, svfloat16_t, svmfloat8_t, |
| 54 |
z0 = svdot_lane_f16_mf8_fpm (z1, z4, z5, 0, fpm0), |
| 55 |
z0 = svdot_lane_fpm (z1, z4, z5, 0, fpm0)) |
| 56 |
|
| 57 |
/* |
| 58 |
** dot_lane_1_f16: |
| 59 |
** msr fpmr, x0 |
| 60 |
** fdot z0\.h, z4\.b, z5\.b\[1\] |
| 61 |
** ret |
| 62 |
*/ |
| 63 |
TEST_DUAL_Z (dot_lane_1_f16, svfloat16_t, svmfloat8_t, |
| 64 |
z0 = svdot_lane_f16_mf8_fpm (z0, z4, z5, 1, fpm0), |
| 65 |
z0 = svdot_lane_fpm (z0, z4, z5, 1, fpm0)) |
| 66 |
|
| 67 |
/* |
| 68 |
** dot_lane_z8_f16: |
| 69 |
** ... |
| 70 |
** msr fpmr, x0 |
| 71 |
** mov (z[0-7])\.d, z8\.d |
| 72 |
** fdot z0\.h, z1\.b, \1\.b\[1\] |
| 73 |
** ldr d8, \[sp\], 16 |
| 74 |
** ret |
| 75 |
*/ |
| 76 |
TEST_DUAL_LANE_REG (dot_lane_z8_f16, svfloat16_t, svmfloat8_t, z8, |
| 77 |
z0 = svdot_lane_f16_mf8_fpm (z0, z1, z8, 1, fpm0), |
| 78 |
z0 = svdot_lane_fpm (z0, z1, z8, 1, fpm0)) |
| 79 |
|
| 80 |
/* |
| 81 |
** dot_lane_z16_f16: |
| 82 |
** ... |
| 83 |
** msr fpmr, x0 |
| 84 |
** mov (z[0-7])\.d, z16\.d |
| 85 |
** fdot z0\.h, z1\.b, \1\.b\[7\] |
| 86 |
** ... |
| 87 |
** ret |
| 88 |
*/ |
| 89 |
TEST_DUAL_LANE_REG (dot_lane_z16_f16, svfloat16_t, svmfloat8_t, z16, |
| 90 |
z0 = svdot_lane_f16_mf8_fpm (z0, z1, z16, 7, fpm0), |
| 91 |
z0 = svdot_lane_fpm (z0, z1, z16, 7, fpm0)) |
| 92 |
|
| 93 |
/* |
| 94 |
** dot_lane_0_f32_tied1: |
| 95 |
** msr fpmr, x0 |
| 96 |
** fdot z0\.s, z4\.b, z5\.b\[0\] |
| 97 |
** ret |
| 98 |
*/ |
| 99 |
TEST_DUAL_Z (dot_lane_0_f32_tied1, svfloat32_t, svmfloat8_t, |
| 100 |
z0 = svdot_lane_f32_mf8_fpm (z0, z4, z5, 0, fpm0), |
| 101 |
z0 = svdot_lane_fpm (z0, z4, z5, 0, fpm0)) |
| 102 |
|
| 103 |
/* |
| 104 |
** dot_lane_0_f32_tied2: |
| 105 |
** msr fpmr, x0 |
| 106 |
** mov (z[0-9]+)\.d, z0\.d |
| 107 |
** movprfx z0, z4 |
| 108 |
** fdot z0\.s, \1\.b, z1\.b\[0\] |
| 109 |
** ret |
| 110 |
*/ |
| 111 |
TEST_DUAL_Z_REV (dot_lane_0_f32_tied2, svfloat32_t, svmfloat8_t, |
| 112 |
z0_res = svdot_lane_f32_mf8_fpm (z4, z0, z1, 0, fpm0), |
| 113 |
z0_res = svdot_lane_fpm (z4, z0, z1, 0, fpm0)) |
| 114 |
|
| 115 |
/* |
| 116 |
** dot_lane_0_f32_tied3: |
| 117 |
** msr fpmr, x0 |
| 118 |
** mov (z[0-9]+)\.d, z0\.d |
| 119 |
** movprfx z0, z4 |
| 120 |
** fdot z0\.s, z1\.b, \1\.b\[0\] |
| 121 |
** ret |
| 122 |
*/ |
| 123 |
TEST_DUAL_Z_REV (dot_lane_0_f32_tied3, svfloat32_t, svmfloat8_t, |
| 124 |
z0_res = svdot_lane_f32_mf8_fpm (z4, z1, z0, 0, fpm0), |
| 125 |
z0_res = svdot_lane_fpm (z4, z1, z0, 0, fpm0)) |
| 126 |
|
| 127 |
/* |
| 128 |
** dot_lane_0_f32_untied: |
| 129 |
** msr fpmr, x0 |
| 130 |
** movprfx z0, z1 |
| 131 |
** fdot z0\.s, z4\.b, z5\.b\[0\] |
| 132 |
** ret |
| 133 |
*/ |
| 134 |
TEST_DUAL_Z (dot_lane_0_f32_untied, svfloat32_t, svmfloat8_t, |
| 135 |
z0 = svdot_lane_f32_mf8_fpm (z1, z4, z5, 0, fpm0), |
| 136 |
z0 = svdot_lane_fpm (z1, z4, z5, 0, fpm0)) |
| 137 |
|
| 138 |
/* |
| 139 |
** dot_lane_1_f32: |
| 140 |
** msr fpmr, x0 |
| 141 |
** fdot z0\.s, z4\.b, z5\.b\[1\] |
| 142 |
** ret |
| 143 |
*/ |
| 144 |
TEST_DUAL_Z (dot_lane_1_f32, svfloat32_t, svmfloat8_t, |
| 145 |
z0 = svdot_lane_f32_mf8_fpm (z0, z4, z5, 1, fpm0), |
| 146 |
z0 = svdot_lane_fpm (z0, z4, z5, 1, fpm0)) |
| 147 |
|
| 148 |
/* |
| 149 |
** dot_lane_z8_f32: |
| 150 |
** ... |
| 151 |
** msr fpmr, x0 |
| 152 |
** mov (z[0-7])\.d, z8\.d |
| 153 |
** fdot z0\.s, z1\.b, \1\.b\[1\] |
| 154 |
** ldr d8, \[sp\], 16 |
| 155 |
** ret |
| 156 |
*/ |
| 157 |
TEST_DUAL_LANE_REG (dot_lane_z8_f32, svfloat32_t, svmfloat8_t, z8, |
| 158 |
z0 = svdot_lane_f32_mf8_fpm (z0, z1, z8, 1, fpm0), |
| 159 |
z0 = svdot_lane_fpm (z0, z1, z8, 1, fpm0)) |
| 160 |
|
| 161 |
/* |
| 162 |
** dot_lane_z32_f32: |
| 163 |
** ... |
| 164 |
** msr fpmr, x0 |
| 165 |
** mov (z[0-7])\.d, z16\.d |
| 166 |
** fdot z0\.s, z1\.b, \1\.b\[3\] |
| 167 |
** ... |
| 168 |
** ret |
| 169 |
*/ |
| 170 |
TEST_DUAL_LANE_REG (dot_lane_z32_f32, svfloat32_t, svmfloat8_t, z16, |
| 171 |
z0 = svdot_lane_f32_mf8_fpm (z0, z1, z16, 3, fpm0), |
| 172 |
z0 = svdot_lane_fpm (z0, z1, z16, 3, fpm0)) |