1 /* 2 * Test program for MSA instruction MSUB_Q.W 3 * 4 * Copyright (C) 2019 Wave Computing, Inc. 5 * Copyright (C) 2019 Aleksandar Markovic <amarkovic@wavecomp.com> 6 * Copyright (C) 2019 RT-RK Computer Based Systems LLC 7 * Copyright (C) 2019 Mateja Marjanovic <mateja.marjanovic@rt-rk.com> 8 * 9 * This program is free software: you can redistribute it and/or modify 10 * it under the terms of the GNU General Public License as published by 11 * the Free Software Foundation, either version 2 of the License, or 12 * (at your option) any later version. 13 * 14 * This program is distributed in the hope that it will be useful, 15 * but WITHOUT ANY WARRANTY; without even the implied warranty of 16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 17 * GNU General Public License for more details. 18 * 19 * You should have received a copy of the GNU General Public License 20 * along with this program. If not, see <https://www.gnu.org/licenses/>. 21 * 22 */ 23 24 #include <sys/time.h> 25 #include <stdint.h> 26 27 #include "../../../../include/wrappers_msa.h" 28 #include "../../../../include/test_inputs_128.h" 29 #include "../../../../include/test_utils_128.h" 30 31 #define TEST_COUNT_TOTAL ( \ 32 (PATTERN_INPUTS_SHORT_COUNT) * (PATTERN_INPUTS_SHORT_COUNT) + \ 33 3 * (RANDOM_INPUTS_SHORT_COUNT) * (RANDOM_INPUTS_SHORT_COUNT)) 34 35 36 int32_t main(void) 37 { 38 char *isa_ase_name = "MSA"; 39 char *group_name = "Fixed Multiply"; 40 char *instruction_name = "MSUB_Q.W"; 41 int32_t ret; 42 uint32_t i, j; 43 struct timeval start, end; 44 double elapsed_time; 45 46 uint64_t b128_result[TEST_COUNT_TOTAL][2]; 47 uint64_t b128_expect[TEST_COUNT_TOTAL][2] = { 48 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, /* 0 */ 49 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 50 { 0xfffffffefffffffeULL, 0xfffffffefffffffeULL, }, 51 { 0xfffffffefffffffeULL, 0xfffffffefffffffeULL, }, 52 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 53 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 54 { 0xfffffffcfffffffcULL, 0xfffffffdfffffffcULL, }, 55 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, 56 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, /* 8 */ 57 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, 58 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, 59 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, 60 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, 61 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, 62 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, 63 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, 64 { 0xfffffffbfffffffbULL, 0xfffffffbfffffffbULL, }, /* 16 */ 65 { 0xfffffffbfffffffbULL, 0xfffffffbfffffffbULL, }, 66 { 0xc71c71c1c71c71c1ULL, 0xc71c71c1c71c71c1ULL, }, 67 { 0xfffffffafffffffaULL, 0xfffffffafffffffaULL, }, 68 { 0xddddddd7ddddddd7ULL, 0xddddddd7ddddddd7ULL, }, 69 { 0xfffffff9fffffff9ULL, 0xfffffff9fffffff9ULL, }, 70 { 0xed097b3ab425ed01ULL, 0x25ed0973ed097b3aULL, }, 71 { 0xfffffff7fffffff7ULL, 0xfffffff7fffffff7ULL, }, 72 { 0xfffffff7fffffff7ULL, 0xfffffff7fffffff7ULL, }, /* 24 */ 73 { 0xfffffff7fffffff7ULL, 0xfffffff7fffffff7ULL, }, 74 { 0x38e38e3038e38e30ULL, 0x38e38e3038e38e30ULL, }, 75 { 0xfffffff7fffffff7ULL, 0xfffffff7fffffff7ULL, }, 76 { 0x2222221922222219ULL, 0x2222221922222219ULL, }, 77 { 0xfffffff7fffffff7ULL, 0xfffffff7fffffff7ULL, }, 78 { 0x12f684b44bda12edULL, 0xda12f67c12f684b4ULL, }, 79 { 0xfffffff6fffffff7ULL, 0xfffffff7fffffff6ULL, }, 80 { 0xfffffff5fffffff6ULL, 0xfffffff6fffffff5ULL, }, /* 32 */ 81 { 0xfffffff5fffffff6ULL, 0xfffffff6fffffff5ULL, }, 82 { 0xddddddd2ddddddd3ULL, 0xddddddd3ddddddd2ULL, }, 83 { 0xfffffff4fffffff5ULL, 0xfffffff5fffffff4ULL, }, 84 { 0xeb851eabeb851eacULL, 0xeb851eaceb851eabULL, }, 85 { 0xfffffff2fffffff3ULL, 0xfffffff3fffffff2ULL, }, 86 { 0xf49f49e6d27d27c4ULL, 0x16c16c09f49f49e6ULL, }, 87 { 0xfffffff1fffffff1ULL, 0xfffffff1fffffff1ULL, }, 88 { 0xfffffff1fffffff1ULL, 0xfffffff1fffffff1ULL, }, /* 40 */ 89 { 0xfffffff1fffffff1ULL, 0xfffffff1fffffff1ULL, }, 90 { 0x2222221322222213ULL, 0x2222221322222213ULL, }, 91 { 0xfffffff1fffffff1ULL, 0xfffffff1fffffff1ULL, }, 92 { 0x147ae138147ae138ULL, 0x147ae138147ae138ULL, }, 93 { 0xfffffff0fffffff0ULL, 0xfffffff0fffffff0ULL, }, 94 { 0x0b60b5fb2d82d81dULL, 0xe93e93d90b60b5fbULL, }, 95 { 0xffffffefffffffefULL, 0xffffffefffffffefULL, }, 96 { 0xffffffeeffffffeeULL, 0xffffffefffffffeeULL, }, /* 48 */ 97 { 0xffffffeeffffffeeULL, 0xffffffefffffffeeULL, }, 98 { 0xed097b2fb425ecf6ULL, 0x25ed0969ed097b2fULL, }, 99 { 0xffffffecffffffecULL, 0xffffffeeffffffecULL, }, 100 { 0xf49f49e0d27d27bdULL, 0x16c16c04f49f49e0ULL, }, 101 { 0xffffffebffffffeaULL, 0xffffffedffffffebULL, }, 102 { 0xf9add3ab9add3bf6ULL, 0xe6b74ef0f9add3abULL, }, 103 { 0xffffffeaffffffe8ULL, 0xffffffecffffffeaULL, }, 104 { 0xffffffeaffffffe8ULL, 0xffffffebffffffeaULL, }, /* 56 */ 105 { 0xffffffeaffffffe8ULL, 0xffffffebffffffeaULL, }, 106 { 0x12f684a74bda12deULL, 0xda12f66f12f684a7ULL, }, 107 { 0xffffffe9ffffffe8ULL, 0xffffffeaffffffe9ULL, }, 108 { 0x0b60b5f42d82d815ULL, 0xe93e93d20b60b5f4ULL, }, 109 { 0xffffffe8ffffffe7ULL, 0xffffffe8ffffffe8ULL, }, 110 { 0x06522c276522c3d9ULL, 0x1948b0e406522c27ULL, }, 111 { 0xffffffe7ffffffe7ULL, 0xffffffe7ffffffe7ULL, }, 112 { 0x9048175df3423f14ULL, 0xd394eba0fffb65e2ULL, }, /* 64 */ 113 { 0x8c4dc60edac87812ULL, 0xc8687ef6003bdb1bULL, }, 114 { 0x80000000f0ed8852ULL, 0xb0ef6662ff3a80e6ULL, }, 115 { 0xe8ec58e8d33594acULL, 0xf41fb1f8fe335d76ULL, }, 116 { 0xe4f20799babbcdaaULL, 0xe8f3454efe73d2afULL, }, 117 { 0xe4cdc5978bb75798ULL, 0xe623b939faecec20ULL, }, 118 { 0xe2057a0fb6418676ULL, 0xe03c1eae0901cfcdULL, }, 119 { 0xe5c1db3280000000ULL, 0xf122e6111767bfefULL, }, 120 { 0x979cbaab96251040ULL, 0xd9a9cd7d166665baULL, }, /* 72 */ 121 { 0x94d46f23c0af3f1eULL, 0xd3c232f2247b4967ULL, }, 122 { 0x800000009a322d5aULL, 0xc75aaa8dec42881aULL, }, 123 { 0xc96455c6cdd91d8cULL, 0xeadc3c95b2c62f40ULL, }, 124 { 0x3250aeaeb02129e6ULL, 0x2e0c882bb1bf0bd0ULL, }, 125 { 0x360d0fd180000000ULL, 0x3ef34f8ec024fbf2ULL, }, 126 { 0x7f716597b3a6f032ULL, 0x6274e19686a8a318ULL, }, 127 { 0x1ce6cdb280000000ULL, 0xfcd31bb480000000ULL, }, 128 { 0x37e70b49a8625540ULL, 0xfeb1f7e080000000ULL, }, /* 80 */ 129 { 0x39c31699dd7c5546ULL, 0xfee37780953f52fcULL, }, 130 { 0x5f82316fca8f431eULL, 0xff3c0af780000000ULL, }, 131 { 0x0bb5432ff1e2e177ULL, 0xfe8d6e9580000000ULL, }, 132 { 0x16a56af3f656d2b3ULL, 0xff67ba1b80000000ULL, }, 133 { 0x17664384fc31bf42ULL, 0xff7e4aa4953f52fcULL, }, 134 { 0x26b0cbfdfa1b830bULL, 0xffa6ab9180000000ULL, }, 135 { 0x04be31a4fe719ab1ULL, 0xff57124580000000ULL, }, 136 { 0x092c8a1ffeef4c68ULL, 0xffba958e80000000ULL, }, /* 88 */ 137 { 0x097aa960ff949347ULL, 0xffc4dede953f52fcULL, }, 138 { 0x0fac7158ff59ab27ULL, 0xffd7471a80000000ULL, }, 139 { 0x01ebdf01ffd41248ULL, 0xffb2fdd380000000ULL, }, 140 { 0x03b76546ffe1ee50ULL, 0xffe05b1780000000ULL, }, 141 { 0x03d70afdfff427aaULL, 0xffe50b86953f52fcULL, }, 142 { 0x065971c1ffeda8dfULL, 0xffed6fa980000000ULL, }, 143 { 0x00c741e8fffb2801ULL, 0xffdce50280000000ULL, }, 144 { 0x01816947fffcaf39ULL, 0xfff1931580000000ULL, }, /* 96 */ 145 { 0x02e97a17fffdbb03ULL, 0xfffa128380000000ULL, }, 146 { 0x05a1edf3fffe7250ULL, 0xfffd906f80000000ULL, }, 147 { 0x0ae508c5fffeefc8ULL, 0xfffeffc380000000ULL, }, 148 { 0x0b41cf1bffff94c3ULL, 0xffff25bb953f52fcULL, }, 149 { 0x0ba1ab79ffffd5c1ULL, 0xffff4613a6f7bf69ULL, }, 150 { 0x0c04b828ffffef5bULL, 0xffff61a0b5bf25caULL, }, 151 { 0x0c6b104efffff971ULL, 0xffff7918c21285a5ULL, }, 152 { 0x148886c7fffff5d8ULL, 0xffffa3179907b21bULL, }, /* 104 */ 153 { 0x21f39335fffff046ULL, 0xffffc00380000000ULL, }, 154 { 0x38235e38ffffe7a6ULL, 0xffffd3ee80000000ULL, }, 155 { 0x5cd2ce93ffffda4bULL, 0xffffe1a680000000ULL, }, 156 { 0x0b60ff8afffff60aULL, 0xffffc69a80000000ULL, }, 157 { 0x01651818fffffd5eULL, 0xffff937480000000ULL, }, 158 { 0x002bc65fffffff4dULL, 0xffff32bb80000000ULL, }, 159 { 0x00055dbfffffffd0ULL, 0xfffe7bd280000000ULL, }, 160 }; 161 162 reset_msa_registers(); 163 164 gettimeofday(&start, NULL); 165 166 for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) { 167 for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) { 168 do_msa_MSUB_Q_W(b128_pattern[i], b128_pattern[j], 169 b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]); 170 } 171 } 172 173 for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) { 174 for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) { 175 do_msa_MSUB_Q_W(b128_random[i], b128_random[j], 176 b128_result[((PATTERN_INPUTS_SHORT_COUNT) * 177 (PATTERN_INPUTS_SHORT_COUNT)) + 178 RANDOM_INPUTS_SHORT_COUNT * i + j]); 179 } 180 } 181 182 for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) { 183 for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) { 184 do_msa_MSUB_Q_W__DDT(b128_random[i], b128_random[j], 185 b128_result[ 186 ((PATTERN_INPUTS_SHORT_COUNT) * 187 (PATTERN_INPUTS_SHORT_COUNT)) + 188 ((RANDOM_INPUTS_SHORT_COUNT) * 189 (RANDOM_INPUTS_SHORT_COUNT)) + 190 RANDOM_INPUTS_SHORT_COUNT * i + j]); 191 } 192 } 193 194 for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) { 195 for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) { 196 do_msa_MSUB_Q_W__DSD(b128_random[i], b128_random[j], 197 b128_result[ 198 ((PATTERN_INPUTS_SHORT_COUNT) * 199 (PATTERN_INPUTS_SHORT_COUNT)) + 200 (2 * (RANDOM_INPUTS_SHORT_COUNT) * 201 (RANDOM_INPUTS_SHORT_COUNT)) + 202 RANDOM_INPUTS_SHORT_COUNT * i + j]); 203 } 204 } 205 206 gettimeofday(&end, NULL); 207 208 elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0; 209 elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0; 210 211 ret = check_results_128(isa_ase_name, group_name, instruction_name, 212 TEST_COUNT_TOTAL, elapsed_time, 213 &b128_result[0][0], &b128_expect[0][0]); 214 215 return ret; 216 } 217