1 /* 2 * Test program for MSA instruction MADD_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 = "MADD_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 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, /* 0 */ 49 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, 50 { 0x0000000000000000ULL, 0x0000000000000000ULL, }, 51 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 52 { 0xffffffffffffffffULL, 0xffffffffffffffffULL, }, 53 { 0xfffffffefffffffeULL, 0xfffffffefffffffeULL, }, 54 { 0xfffffffefffffffeULL, 0xfffffffdfffffffeULL, }, 55 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 56 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, /* 8 */ 57 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 58 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 59 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 60 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 61 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 62 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 63 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 64 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, /* 16 */ 65 { 0xfffffffdfffffffdULL, 0xfffffffdfffffffdULL, }, 66 { 0x38e38e3638e38e36ULL, 0x38e38e3638e38e36ULL, }, 67 { 0xfffffffcfffffffcULL, 0xfffffffcfffffffcULL, }, 68 { 0x2222221e2222221eULL, 0x2222221e2222221eULL, }, 69 { 0xfffffffbfffffffbULL, 0xfffffffbfffffffbULL, }, 70 { 0x12f684b94bda12f2ULL, 0xda12f68012f684b9ULL, }, 71 { 0xfffffffbfffffffbULL, 0xfffffffbfffffffbULL, }, 72 { 0xfffffffafffffffaULL, 0xfffffffafffffffaULL, }, /* 24 */ 73 { 0xfffffffafffffffaULL, 0xfffffffafffffffaULL, }, 74 { 0xc71c71c0c71c71c0ULL, 0xc71c71c0c71c71c0ULL, }, 75 { 0xfffffff8fffffff8ULL, 0xfffffff8fffffff8ULL, }, 76 { 0xddddddd5ddddddd5ULL, 0xddddddd5ddddddd5ULL, }, 77 { 0xfffffff6fffffff6ULL, 0xfffffff6fffffff6ULL, }, 78 { 0xed097b38b425ecffULL, 0x25ed0970ed097b38ULL, }, 79 { 0xfffffff5fffffff4ULL, 0xfffffff4fffffff5ULL, }, 80 { 0xfffffff5fffffff4ULL, 0xfffffff4fffffff5ULL, }, /* 32 */ 81 { 0xfffffff5fffffff4ULL, 0xfffffff4fffffff5ULL, }, 82 { 0x2222221722222216ULL, 0x2222221622222217ULL, }, 83 { 0xfffffff4fffffff3ULL, 0xfffffff3fffffff4ULL, }, 84 { 0x147ae13c147ae13bULL, 0x147ae13b147ae13cULL, }, 85 { 0xfffffff4fffffff3ULL, 0xfffffff3fffffff4ULL, }, 86 { 0x0b60b5ff2d82d821ULL, 0xe93e93dc0b60b5ffULL, }, 87 { 0xfffffff3fffffff3ULL, 0xfffffff3fffffff3ULL, }, 88 { 0xfffffff2fffffff2ULL, 0xfffffff2fffffff2ULL, }, /* 40 */ 89 { 0xfffffff2fffffff2ULL, 0xfffffff2fffffff2ULL, }, 90 { 0xddddddcfddddddcfULL, 0xddddddcfddddddcfULL, }, 91 { 0xfffffff0fffffff0ULL, 0xfffffff0fffffff0ULL, }, 92 { 0xeb851ea8eb851ea8ULL, 0xeb851ea8eb851ea8ULL, }, 93 { 0xffffffefffffffefULL, 0xffffffefffffffefULL, }, 94 { 0xf49f49e3d27d27c1ULL, 0x16c16c05f49f49e3ULL, }, 95 { 0xffffffeeffffffeeULL, 0xffffffeeffffffeeULL, }, 96 { 0xffffffeeffffffeeULL, 0xffffffedffffffeeULL, }, /* 48 */ 97 { 0xffffffeeffffffeeULL, 0xffffffedffffffeeULL, }, 98 { 0x12f684ac4bda12e5ULL, 0xda12f67212f684acULL, }, 99 { 0xffffffeeffffffeeULL, 0xffffffecffffffeeULL, }, 100 { 0x0b60b5f92d82d81cULL, 0xe93e93d50b60b5f9ULL, }, 101 { 0xffffffedffffffeeULL, 0xffffffebffffffedULL, }, 102 { 0x06522c2c6522c3e1ULL, 0x1948b0e706522c2cULL, }, 103 { 0xffffffecffffffeeULL, 0xffffffeaffffffecULL, }, 104 { 0xffffffebffffffedULL, 0xffffffeaffffffebULL, }, /* 56 */ 105 { 0xffffffebffffffedULL, 0xffffffeaffffffebULL, }, 106 { 0xed097b2db425ecf6ULL, 0x25ed0965ed097b2dULL, }, 107 { 0xffffffeaffffffebULL, 0xffffffe9ffffffeaULL, }, 108 { 0xf49f49ded27d27bdULL, 0x16c16c00f49f49deULL, }, 109 { 0xffffffe9ffffffeaULL, 0xffffffe9ffffffe9ULL, }, 110 { 0xf9add3a99add3bf7ULL, 0xe6b74eecf9add3a9ULL, }, 111 { 0xffffffe8ffffffe8ULL, 0xffffffe8ffffffe8ULL, }, 112 { 0x6fb7e8710cbdc0baULL, 0x2c6b142e000499ecULL, }, /* 64 */ 113 { 0x73b239bf253787bbULL, 0x379780d7ffc424b2ULL, }, 114 { 0x7fffffff0f12777aULL, 0x4f10996a00c57ee6ULL, }, 115 { 0x1713a7162cca6b1fULL, 0x0be04dd301cca255ULL, }, 116 { 0x1b0df86445443220ULL, 0x170cba7c018c2d1bULL, }, 117 { 0x1b323a657448a831ULL, 0x19dc4690051313a9ULL, }, 118 { 0x1dfa85ec49be7952ULL, 0x1fc3e11af6fe2ffbULL, }, 119 { 0x1a3e24c87fffffffULL, 0x0edd19b6e8983fd8ULL, }, 120 { 0x6863454e69daefbeULL, 0x26563249e9999a0cULL, }, /* 72 */ 121 { 0x6b2b90d53f50c0dfULL, 0x2c3dccd3db84b65eULL, }, 122 { 0x7fffffff65cdd2a2ULL, 0x38a5553713bd77aaULL, }, 123 { 0x369baa383226e26fULL, 0x1523c32e4d39d083ULL, }, 124 { 0xcdaf514f4fded614ULL, 0xd1f377974e40f3f2ULL, }, 125 { 0xc9f2f02b7fffffffULL, 0xc10cb0333fdb03cfULL, }, 126 { 0x808e9a644c590fccULL, 0x9d8b1e2a79575ca8ULL, }, 127 { 0xe31932487fffffffULL, 0x032ce40b7fffffffULL, }, 128 { 0xfe196fe57fffffffULL, 0x050bc0117e7bb00bULL, }, /* 80 */ 129 { 0xfe299f467fffffffULL, 0x05cb2b207fffffffULL, }, 130 { 0xff5d018239cf8b7fULL, 0x0798e21e2b2b2513ULL, }, 131 { 0xfecdfe1e645a7d99ULL, 0x00d3dcf00dea608dULL, }, 132 { 0xffebe0507fffffffULL, 0x0150aaf30dc02967ULL, }, 133 { 0xffec8bad7fffffffULL, 0x01828e9310087db0ULL, }, 134 { 0xfff9423b39cf8b7fULL, 0x01fae4ad056841b8ULL, }, 135 { 0xfff35804645a7d99ULL, 0x003737ba01be3861ULL, }, 136 { 0xffff2aee7fffffffULL, 0x0057bed401b8eeafULL, }, /* 88 */ 137 { 0xffff32047fffffffULL, 0x0064bf7c02021ffbULL, }, 138 { 0xffffb89f39cf8b7fULL, 0x00841c7300ad6409ULL, }, 139 { 0xffff79fe645a7d99ULL, 0x000e642e0037e4a5ULL, }, 140 { 0xfffff72f7fffffffULL, 0x0016de7600373b15ULL, }, 141 { 0xfffff77a7fffffffULL, 0x001a420100406619ULL, }, 142 { 0xfffffd0b39cf8b7fULL, 0x00226e950015b801ULL, }, 143 { 0xfffffa72645a7d99ULL, 0x0003c03400070049ULL, }, 144 { 0xffffffa27fffffffULL, 0x0005f5d70006eb0bULL, }, /* 96 */ 145 { 0xfffffff97fffffffULL, 0x000978af0006d60eULL, }, 146 { 0xffffffff7fffffffULL, 0x000f0d050006c150ULL, }, 147 { 0xffffffff7fffffffULL, 0x0017eac30006acd1ULL, }, 148 { 0xffffffff7fffffffULL, 0x001b76100007c878ULL, }, 149 { 0xffffffff7fffffffULL, 0x001f87d000091335ULL, }, 150 { 0xffffffff7fffffffULL, 0x002433ef000a94d9ULL, }, 151 { 0xffffffff7fffffffULL, 0x0029914d000c5680ULL, }, 152 { 0xffffffff39cf8b7fULL, 0x003681f800042937ULL, }, /* 104 */ 153 { 0xffffffff1a1c28c3ULL, 0x004779e10001673fULL, }, 154 { 0xffffffff0bcae025ULL, 0x005dba1000007928ULL, }, 155 { 0xffffffff055376c1ULL, 0x007ae77c000028dcULL, }, 156 { 0xfffffffe093ed554ULL, 0x000d636d00000d2bULL, }, 157 { 0xfffffffc100c9463ULL, 0x0001755c0000043eULL, }, 158 { 0xfffffff81bdc128cULL, 0x000028ab0000015eULL, }, 159 { 0xfffffff0305c8babULL, 0x0000046e00000070ULL, }, 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_MADD_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_MADD_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_MADD_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_MADD_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