test_msa_bmnz_v.c (10635B)
1 /* 2 * Test program for MSA instruction BMNZ.V 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 = "Bit Move"; 38 char *instruction_name = "BMNZ.V"; 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 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, /* 0 */ 47 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 48 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 49 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 50 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 51 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 52 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 53 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 54 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, /* 8 */ 55 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, 56 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, 57 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, 58 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, 59 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, 60 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, 61 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, 62 { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, }, /* 16 */ 63 { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, }, 64 { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, }, 65 { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, }, 66 { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, }, 67 { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, }, 68 { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, }, 69 { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, }, 70 { 0x5555555555555555ULL, 0x5555555555555555ULL, }, /* 24 */ 71 { 0x5555555555555555ULL, 0x5555555555555555ULL, }, 72 { 0x5555555555555555ULL, 0x5555555555555555ULL, }, 73 { 0x5555555555555555ULL, 0x5555555555555555ULL, }, 74 { 0x5555555555555555ULL, 0x5555555555555555ULL, }, 75 { 0x5555555555555555ULL, 0x5555555555555555ULL, }, 76 { 0x5555555555555555ULL, 0x5555555555555555ULL, }, 77 { 0x5555555555555555ULL, 0x5555555555555555ULL, }, 78 { 0xccccccccccccccccULL, 0xccccccccccccccccULL, }, /* 32 */ 79 { 0xccccccccccccccccULL, 0xccccccccccccccccULL, }, 80 { 0xccccccccccccccccULL, 0xccccccccccccccccULL, }, 81 { 0xccccccccccccccccULL, 0xccccccccccccccccULL, }, 82 { 0xccccccccccccccccULL, 0xccccccccccccccccULL, }, 83 { 0xccccccccccccccccULL, 0xccccccccccccccccULL, }, 84 { 0xccccccccccccccccULL, 0xccccccccccccccccULL, }, 85 { 0xccccccccccccccccULL, 0xccccccccccccccccULL, }, 86 { 0x3333333333333333ULL, 0x3333333333333333ULL, }, /* 40 */ 87 { 0x3333333333333333ULL, 0x3333333333333333ULL, }, 88 { 0x3333333333333333ULL, 0x3333333333333333ULL, }, 89 { 0x3333333333333333ULL, 0x3333333333333333ULL, }, 90 { 0x3333333333333333ULL, 0x3333333333333333ULL, }, 91 { 0x3333333333333333ULL, 0x3333333333333333ULL, }, 92 { 0x3333333333333333ULL, 0x3333333333333333ULL, }, 93 { 0x3333333333333333ULL, 0x3333333333333333ULL, }, 94 { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, }, /* 48 */ 95 { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, }, 96 { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, }, 97 { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, }, 98 { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, }, 99 { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, }, 100 { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, }, 101 { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, }, 102 { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, }, /* 56 */ 103 { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, }, 104 { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, }, 105 { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, }, 106 { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, }, 107 { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, }, 108 { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, }, 109 { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, }, 110 { 0x9c7be7dc79e75d71ULL, 0xcf7f7bdffe7bf71cULL, }, /* 64 */ 111 { 0x8c6be7dc38665d71ULL, 0xcf6f4bdffe7bb50cULL, }, 112 { 0x886be7dc28625571ULL, 0xcb670b5efe7bb00cULL, }, 113 { 0x886ae7dc28625571ULL, 0x4b670b5efe7bb00cULL, }, 114 { 0x882a015008024531ULL, 0x02670b1a143b100cULL, }, 115 { 0xfbbe01734d93c739ULL, 0x12f7bb1a153f52fcULL, }, 116 { 0xfbbe01734d93c739ULL, 0x12f7bb1a153f52fcULL, }, 117 { 0xfbbe01734d93c739ULL, 0x12f7bb1a153f52fcULL, }, 118 { 0xfbdea7bb6dd38339ULL, 0x13d0b25eab2f62f4ULL, }, /* 72 */ 119 { 0xa85aa7ba29c38331ULL, 0x03d0825eab2b2014ULL, }, 120 { 0xac5aafbab9cf8bb1ULL, 0x27d8c6ffab2b2514ULL, }, 121 { 0xac5aafbab9cf8bb1ULL, 0x27d8c6ffab2b2514ULL, }, 122 { 0x245a0f7e99adcaf1ULL, 0x2df9ccf9a942a510ULL, }, 123 { 0x744e0f5ddc3dcaf9ULL, 0x2df9ccf9a942e7a0ULL, }, 124 { 0x704e075d5c31c279ULL, 0x0df188d8a942e2a0ULL, }, 125 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 126 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, /* 80 */ 127 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 128 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 129 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 130 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 131 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 132 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 133 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 134 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, /* 88 */ 135 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 136 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 137 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 138 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 139 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 140 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 141 { 0x704f175d5e31e27fULL, 0x8df188d8a942e2a0ULL, }, 142 { 0x004a064c08204040ULL, 0x09610858a842a000ULL, }, /* 96 */ 143 { 0x004a064c08204040ULL, 0x09610858a842a000ULL, }, 144 { 0x004a064c08204040ULL, 0x09610858a842a000ULL, }, 145 { 0x004a064c08204040ULL, 0x09610858a842a000ULL, }, 146 { 0x000a004008004000ULL, 0x0061081800020000ULL, }, 147 { 0x000a004008004000ULL, 0x0061081800020000ULL, }, 148 { 0x000a004008004000ULL, 0x0061081800020000ULL, }, 149 { 0x000a004008004000ULL, 0x0061081800020000ULL, }, 150 { 0x000a000008000000ULL, 0x0040001800020000ULL, }, /* 104 */ 151 { 0x000a000008000000ULL, 0x0040001800020000ULL, }, 152 { 0x000a000008000000ULL, 0x0040001800020000ULL, }, 153 { 0x000a000008000000ULL, 0x0040001800020000ULL, }, 154 { 0x000a000008000000ULL, 0x0040001800020000ULL, }, 155 { 0x000a000008000000ULL, 0x0040001800020000ULL, }, 156 { 0x000a000008000000ULL, 0x0040001800020000ULL, }, 157 { 0x000a000008000000ULL, 0x0040001800020000ULL, }, 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_BMNZ_V(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_BMNZ_V(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_BMNZ_V__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_BMNZ_V__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 }