1 /* 2 * Test program for MSA instruction DPSUB_U.W 3 * 4 * Copyright (C) 2019 Wave Computing, Inc. 5 * Copyright (C) 2019 Aleksandar Markovic <amarkovic@wavecomp.com> 6 * 7 * This program is free software: you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License as published by 9 * the Free Software Foundation, either version 2 of the License, or 10 * (at your option) any later version. 11 *` 12 * This program is distributed in the hope that it will be useful, 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 15 * GNU General Public License for more details. 16 * 17 * You should have received a copy of the GNU General Public License 18 * along with this program. If not, see <https://www.gnu.org/licenses/>. 19 * 20 */ 21 22 #include <sys/time.h> 23 #include <stdint.h> 24 25 #include "../../../../include/wrappers_msa.h" 26 #include "../../../../include/test_inputs_128.h" 27 #include "../../../../include/test_utils_128.h" 28 29 #define TEST_COUNT_TOTAL ( \ 30 (PATTERN_INPUTS_SHORT_COUNT) * (PATTERN_INPUTS_SHORT_COUNT) + \ 31 3 * (RANDOM_INPUTS_SHORT_COUNT) * (RANDOM_INPUTS_SHORT_COUNT)) 32 33 34 int32_t main(void) 35 { 36 char *isa_ase_name = "MSA"; 37 char *group_name = "Int Dot Product"; 38 char *instruction_name = "DPSUB_U.W"; 39 int32_t ret; 40 uint32_t i, j; 41 struct timeval start, end; 42 double elapsed_time; 43 44 uint64_t b128_result[TEST_COUNT_TOTAL][2]; 45 uint64_t b128_expect[TEST_COUNT_TOTAL][2] = { 46 { 0x0003fffe0003fffeULL, 0x0003fffe0003fffeULL, }, /* 0 */ 47 { 0x0003fffe0003fffeULL, 0x0003fffe0003fffeULL, }, 48 { 0xaab15552aab15552ULL, 0xaab15552aab15552ULL, }, 49 { 0x0007fffc0007fffcULL, 0x0007fffc0007fffcULL, }, 50 { 0x6671999466719994ULL, 0x6671999466719994ULL, }, 51 { 0x000bfffa000bfffaULL, 0x000bfffa000bfffaULL, }, 52 { 0xe39c1c6b8e4771c0ULL, 0x38f1c715e39c1c6bULL, }, 53 { 0x000ffff8000ffff8ULL, 0x000ffff8000ffff8ULL, }, 54 { 0x000ffff8000ffff8ULL, 0x000ffff8000ffff8ULL, }, /* 8 */ 55 { 0x000ffff8000ffff8ULL, 0x000ffff8000ffff8ULL, }, 56 { 0x000ffff8000ffff8ULL, 0x000ffff8000ffff8ULL, }, 57 { 0x000ffff8000ffff8ULL, 0x000ffff8000ffff8ULL, }, 58 { 0x000ffff8000ffff8ULL, 0x000ffff8000ffff8ULL, }, 59 { 0x000ffff8000ffff8ULL, 0x000ffff8000ffff8ULL, }, 60 { 0x000ffff8000ffff8ULL, 0x000ffff8000ffff8ULL, }, 61 { 0x000ffff8000ffff8ULL, 0x000ffff8000ffff8ULL, }, 62 { 0xaabd554caabd554cULL, 0xaabd554caabd554cULL, }, /* 16 */ 63 { 0xaabd554caabd554cULL, 0xaabd554caabd554cULL, }, 64 { 0xc730e384c730e384ULL, 0xc730e384c730e384ULL, }, 65 { 0x556aaaa0556aaaa0ULL, 0x556aaaa0556aaaa0ULL, }, 66 { 0x445bbbb0445bbbb0ULL, 0x445bbbb0445bbbb0ULL, }, 67 { 0x0017fff40017fff4ULL, 0x0017fff40017fff4ULL, }, 68 { 0x427812ea0994f678ULL, 0x7b5bda06427812eaULL, }, 69 { 0xaac55548aac55548ULL, 0xaac55548aac55548ULL, }, 70 { 0x001bfff2001bfff2ULL, 0x001bfff2001bfff2ULL, }, /* 24 */ 71 { 0x001bfff2001bfff2ULL, 0x001bfff2001bfff2ULL, }, 72 { 0x8e55c70e8e55c70eULL, 0x8e55c70e8e55c70eULL, }, 73 { 0x5572aa9c5572aa9cULL, 0x5572aa9c5572aa9cULL, }, 74 { 0xcceb3324cceb3324ULL, 0xcceb3324cceb3324ULL, }, 75 { 0xaac95546aac95546ULL, 0xaac95546aac95546ULL, }, 76 { 0x4bf95ec12f87d088ULL, 0x686b424f4bf95ec1ULL, }, 77 { 0x001ffff0001ffff0ULL, 0x001ffff0001ffff0ULL, }, 78 { 0x6689998866899988ULL, 0x6689998866899988ULL, }, /* 32 */ 79 { 0x6689998866899988ULL, 0x6689998866899988ULL, }, 80 { 0x557aaa98557aaa98ULL, 0x557aaa98557aaa98ULL, }, 81 { 0xccf33320ccf33320ULL, 0xccf33320ccf33320ULL, }, 82 { 0x8547ae008547ae00ULL, 0x8547ae008547ae00ULL, }, 83 { 0x335cccb8335cccb8ULL, 0x335cccb8335cccb8ULL, }, 84 { 0x4fd016ac0b8c5af0ULL, 0x94149f344fd016acULL, }, 85 { 0x99c6665099c66650ULL, 0x99c6665099c66650ULL, }, 86 { 0x3360ccb63360ccb6ULL, 0x3360ccb63360ccb6ULL, }, /* 40 */ 87 { 0x3360ccb63360ccb6ULL, 0x3360ccb63360ccb6ULL, }, 88 { 0xef1d10faef1d10faULL, 0xef1d10faef1d10faULL, }, 89 { 0xccfb331cccfb331cULL, 0xccfb331cccfb331cULL, }, 90 { 0x7b1051d47b1051d4ULL, 0x7b1051d47b1051d4ULL, }, 91 { 0x6695998266959982ULL, 0x6695998266959982ULL, }, 92 { 0x2db26bff1ca17d10ULL, 0x3ec38e212db26bffULL, }, 93 { 0x002fffe8002fffe8ULL, 0x002fffe8002fffe8ULL, }, 94 { 0xe3c01c598e6b71aeULL, 0x3915c703e3c01c59ULL, }, /* 48 */ 95 { 0xe3c01c598e6b71aeULL, 0x3915c703e3c01c59ULL, }, 96 { 0x26202f4f97e86832ULL, 0xb459a11526202f4fULL, }, 97 { 0xc75038ca1ca6e374ULL, 0x71fb8e1ec75038caULL, }, 98 { 0xe3c382bef4d671acULL, 0xd2b3609ae3c382beULL, }, 99 { 0xaae0553baae2553aULL, 0xaae15539aae0553bULL, }, 100 { 0xd3f7192e919b0636ULL, 0x4f3b0fb0d3f7192eULL, }, 101 { 0x8e7071ac391dc700ULL, 0xe3c71c548e7071acULL, }, 102 { 0xaae45539aae65538ULL, 0xaae55537aae45539ULL, }, /* 56 */ 103 { 0xaae45539aae65538ULL, 0xaae55537aae45539ULL, }, 104 { 0x133197974c16b408ULL, 0xda4ed07913319797ULL, }, 105 { 0xc75838c61caee370ULL, 0x72038e1ac75838c6ULL, }, 106 { 0x114e886aaae8eed0ULL, 0x77b55536114e886aULL, }, 107 { 0xe3cc1c538e7771a8ULL, 0x3921c6fde3cc1c53ULL, }, 108 { 0x9e4574d135fa3272ULL, 0xcdadd3a19e4574d1ULL, }, 109 { 0x003fffe0003fffe0ULL, 0x003fffe0003fffe0ULL, }, 110 { 0xe77c016cdd7daa5cULL, 0xe98945eb8a373037ULL, }, /* 64 */ 111 { 0x60fd93dc8ef99616ULL, 0xdba475fe3c075522ULL, }, 112 { 0x67ae7b204335bcd8ULL, 0xc7121f747860f089ULL, }, 113 { 0x17bb0f0ee8fd8c96ULL, 0x972fe12d34478253ULL, }, 114 { 0x913ca17e9a797850ULL, 0x894b1140e617a73eULL, }, 115 { 0x99ae5a31e83a45a7ULL, 0xff24684bc96dd9adULL, }, 116 { 0xefeea5a7437774caULL, 0x6ac04bfdaf344b68ULL, }, 117 { 0x8175363e76faed37ULL, 0xfc38b88657b1e4aaULL, }, 118 { 0x88261d822b3713f9ULL, 0xe7a661fc940b8011ULL, }, /* 72 */ 119 { 0xde6668f88674431cULL, 0x534245ae79d1f1ccULL, }, 120 { 0xf331d070b3932dbbULL, 0xb25f4d6d0200af03ULL, }, 121 { 0x985e1b88f3e41a1cULL, 0x31e873ed7002566dULL, }, 122 { 0x486aaf7699abe9daULL, 0x020635a62be8e837ULL, }, 123 { 0xd9f1400dcd2f6247ULL, 0x937ea22fd4668179ULL, }, 124 { 0x7f1d8b250d804ea8ULL, 0x1307c8af426828e3ULL, }, 125 { 0x4be6ff9b22ca7983ULL, 0x7b2e358e09e10fdfULL, }, 126 { 0x3d0470dbf4d6b86fULL, 0x548567e8f5250450ULL, }, /* 80 */ 127 { 0x00d897321b41b715ULL, 0x02517c05df66c875ULL, }, 128 { 0x991ec80ea3b5c306ULL, 0xa18dc9b22cff8e2fULL, }, 129 { 0x44850796bb133f8dULL, 0xdc2a4cc591614211ULL, }, 130 { 0x192b30fc8866f607ULL, 0x97e8c289d36e61aaULL, }, 131 { 0x0058689e9fcad43dULL, 0xfe7a0cc7a239bc40ULL, }, 132 { 0xb8bc4cc2b8296867ULL, 0xccf01b9e1a7e74adULL, }, 133 { 0x61014864181c5d2cULL, 0x4c8bc05ea1b0cc11ULL, }, 134 { 0xec0d0e4af547db74ULL, 0x2d758eed74a13bb5ULL, }, /* 88 */ 135 { 0x03e797060056a10fULL, 0xc1a1d5f8579892eaULL, }, 136 { 0x9a3ca5d4a8548905ULL, 0xfd2bfd1807c0081aULL, }, 137 { 0x4820b48cf1454f6bULL, 0xe982ac5dfb74445aULL, }, 138 { 0x7eec2fbcb0c3c941ULL, 0x9d1459e9d27d4766ULL, }, 139 { 0x020a22e0debbd140ULL, 0x4fbb0ef3a9e0453bULL, }, 140 { 0xe8df4a9ccb0c350bULL, 0x37b3761e2e442cffULL, }, 141 { 0x7c3604df51731065ULL, 0xd9add64be7d81e17ULL, }, 142 { 0x35a1aacf3f24481fULL, 0x900caa26ecaf303bULL, }, /* 96 */ 143 { 0x7f0fd7311d2a2997ULL, 0x5e11155ee03d0362ULL, }, 144 { 0x7959c1ef0ab6e6c3ULL, 0x41695f03ff01377bULL, }, 145 { 0x89d8f6a1bc2ded57ULL, 0x29ed46aadb5c8a3cULL, }, 146 { 0x01ec800ecaa24ac8ULL, 0xf32ccdbb9c58b788ULL, }, 147 { 0xffd7297c53176782ULL, 0x4acc984953e0cc00ULL, }, 148 { 0x04316ff6e9707c3dULL, 0xd5f54b0b0ac9f7e0ULL, }, 149 { 0xffe6fc76421c7405ULL, 0x8f42f98ab98b12e9ULL, }, 150 { 0xa75ea33ed2e809e1ULL, 0xb6fdbf643abee85cULL, }, /* 104 */ 151 { 0xc75019063471bcc9ULL, 0x05bcd250f1d0ad42ULL, }, 152 { 0x300d94eaa78224eaULL, 0x615cfa00370a0c2aULL, }, 153 { 0xaa1a04f419d03dccULL, 0x8fe0ca60107a1a34ULL, }, 154 { 0x5f0bb18ad9b000d4ULL, 0xd3ed3780ee630840ULL, }, 155 { 0x25e24aa388dc4d8cULL, 0x40c1586349788fbaULL, }, 156 { 0x0ec344de11f41ac8ULL, 0xed9aea2a99a95e8aULL, }, 157 { 0x02499bebf3ac5a24ULL, 0xecb186c0e06045b8ULL, }, 158 }; 159 160 reset_msa_registers(); 161 162 gettimeofday(&start, NULL); 163 164 for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) { 165 for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) { 166 do_msa_DPSUB_U_W(b128_pattern[i], b128_pattern[j], 167 b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]); 168 } 169 } 170 171 for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) { 172 for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) { 173 do_msa_DPSUB_U_W(b128_random[i], b128_random[j], 174 b128_result[((PATTERN_INPUTS_SHORT_COUNT) * 175 (PATTERN_INPUTS_SHORT_COUNT)) + 176 RANDOM_INPUTS_SHORT_COUNT * i + j]); 177 } 178 } 179 180 for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) { 181 for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) { 182 do_msa_DPSUB_U_W__DDT(b128_random[i], b128_random[j], 183 b128_result[ 184 ((PATTERN_INPUTS_SHORT_COUNT) * 185 (PATTERN_INPUTS_SHORT_COUNT)) + 186 ((RANDOM_INPUTS_SHORT_COUNT) * 187 (RANDOM_INPUTS_SHORT_COUNT)) + 188 RANDOM_INPUTS_SHORT_COUNT * i + j]); 189 } 190 } 191 192 for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) { 193 for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) { 194 do_msa_DPSUB_U_W__DSD(b128_random[i], b128_random[j], 195 b128_result[ 196 ((PATTERN_INPUTS_SHORT_COUNT) * 197 (PATTERN_INPUTS_SHORT_COUNT)) + 198 (2 * (RANDOM_INPUTS_SHORT_COUNT) * 199 (RANDOM_INPUTS_SHORT_COUNT)) + 200 RANDOM_INPUTS_SHORT_COUNT * i + j]); 201 } 202 } 203 204 gettimeofday(&end, NULL); 205 206 elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0; 207 elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0; 208 209 ret = check_results_128(isa_ase_name, group_name, instruction_name, 210 TEST_COUNT_TOTAL, elapsed_time, 211 &b128_result[0][0], &b128_expect[0][0]); 212 213 return ret; 214 } 215