1a8f91dd9SMateja Marjanovic /*
2a8f91dd9SMateja Marjanovic * Test program for MSA instruction DOTP_U.D
3a8f91dd9SMateja Marjanovic *
4750541c4SAleksandar Markovic * Copyright (C) 2019 Wave Computing, Inc.
5750541c4SAleksandar Markovic * Copyright (C) 2019 Aleksandar Markovic <amarkovic@wavecomp.com>
6750541c4SAleksandar Markovic * Copyright (C) 2019 RT-RK Computer Based Systems LLC
7750541c4SAleksandar Markovic * Copyright (C) 2019 Mateja Marjanovic <mateja.marjanovic@rt-rk.com>
8a8f91dd9SMateja Marjanovic *
9a8f91dd9SMateja Marjanovic * This program is free software: you can redistribute it and/or modify
10a8f91dd9SMateja Marjanovic * it under the terms of the GNU General Public License as published by
11a8f91dd9SMateja Marjanovic * the Free Software Foundation, either version 2 of the License, or
12a8f91dd9SMateja Marjanovic * (at your option) any later version.
13a8f91dd9SMateja Marjanovic *
14a8f91dd9SMateja Marjanovic * This program is distributed in the hope that it will be useful,
15a8f91dd9SMateja Marjanovic * but WITHOUT ANY WARRANTY; without even the implied warranty of
16a8f91dd9SMateja Marjanovic * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17a8f91dd9SMateja Marjanovic * GNU General Public License for more details.
18a8f91dd9SMateja Marjanovic *
19a8f91dd9SMateja Marjanovic * You should have received a copy of the GNU General Public License
20a8f91dd9SMateja Marjanovic * along with this program. If not, see <https://www.gnu.org/licenses/>.
21a8f91dd9SMateja Marjanovic *
22a8f91dd9SMateja Marjanovic */
23a8f91dd9SMateja Marjanovic
24a8f91dd9SMateja Marjanovic #include <sys/time.h>
25a8f91dd9SMateja Marjanovic #include <stdint.h>
26a8f91dd9SMateja Marjanovic
27a8f91dd9SMateja Marjanovic #include "../../../../include/wrappers_msa.h"
28750541c4SAleksandar Markovic #include "../../../../include/test_inputs_128.h"
29750541c4SAleksandar Markovic #include "../../../../include/test_utils_128.h"
30a8f91dd9SMateja Marjanovic
31a8f91dd9SMateja Marjanovic #define TEST_COUNT_TOTAL ( \
32a8f91dd9SMateja Marjanovic (PATTERN_INPUTS_SHORT_COUNT) * (PATTERN_INPUTS_SHORT_COUNT) + \
33a8f91dd9SMateja Marjanovic (RANDOM_INPUTS_SHORT_COUNT) * (RANDOM_INPUTS_SHORT_COUNT))
34a8f91dd9SMateja Marjanovic
35a8f91dd9SMateja Marjanovic
main(void)36a8f91dd9SMateja Marjanovic int32_t main(void)
37a8f91dd9SMateja Marjanovic {
38*8e6fe6b8SAleksandar Markovic char *isa_ase_name = "MSA";
39*8e6fe6b8SAleksandar Markovic char *group_name = "Int Dot Product";
40a8f91dd9SMateja Marjanovic char *instruction_name = "DOTP_U.D";
41a8f91dd9SMateja Marjanovic int32_t ret;
42a8f91dd9SMateja Marjanovic uint32_t i, j;
43a8f91dd9SMateja Marjanovic struct timeval start, end;
44a8f91dd9SMateja Marjanovic double elapsed_time;
45a8f91dd9SMateja Marjanovic
46a8f91dd9SMateja Marjanovic uint64_t b128_result[TEST_COUNT_TOTAL][2];
47a8f91dd9SMateja Marjanovic uint64_t b128_expect[TEST_COUNT_TOTAL][2] = {
48a8f91dd9SMateja Marjanovic { 0xfffffffc00000002ULL, 0xfffffffc00000002ULL, }, /* 0 */
49a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
50a8f91dd9SMateja Marjanovic { 0x55555552aaaaaaacULL, 0x55555552aaaaaaacULL, },
51a8f91dd9SMateja Marjanovic { 0xaaaaaaa955555556ULL, 0xaaaaaaa955555556ULL, },
52a8f91dd9SMateja Marjanovic { 0x9999999666666668ULL, 0x9999999666666668ULL, },
53a8f91dd9SMateja Marjanovic { 0x666666659999999aULL, 0x666666659999999aULL, },
54a8f91dd9SMateja Marjanovic { 0x71c71c6f8e38e38fULL, 0x1c71c719e38e38e5ULL, },
55a8f91dd9SMateja Marjanovic { 0x8e38e38c71c71c73ULL, 0xe38e38e21c71c71dULL, },
56a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, }, /* 8 */
57a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
58a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
59a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
60a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
61a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
62a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
63a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
64a8f91dd9SMateja Marjanovic { 0x55555552aaaaaaacULL, 0x55555552aaaaaaacULL, }, /* 16 */
65a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
66a8f91dd9SMateja Marjanovic { 0xe38e38e1c71c71c8ULL, 0xe38e38e1c71c71c8ULL, },
67a8f91dd9SMateja Marjanovic { 0x71c71c70e38e38e4ULL, 0x71c71c70e38e38e4ULL, },
68a8f91dd9SMateja Marjanovic { 0x1111110eeeeeeef0ULL, 0x1111110eeeeeeef0ULL, },
69a8f91dd9SMateja Marjanovic { 0x44444443bbbbbbbcULL, 0x44444443bbbbbbbcULL, },
70a8f91dd9SMateja Marjanovic { 0xf684bd9fb425ed0aULL, 0xbda12f6697b425eeULL, },
71a8f91dd9SMateja Marjanovic { 0x5ed097b2f684bda2ULL, 0x97b425ec12f684beULL, },
72a8f91dd9SMateja Marjanovic { 0xaaaaaaa955555556ULL, 0xaaaaaaa955555556ULL, }, /* 24 */
73a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
74a8f91dd9SMateja Marjanovic { 0x71c71c70e38e38e4ULL, 0x71c71c70e38e38e4ULL, },
75a8f91dd9SMateja Marjanovic { 0x38e38e3871c71c72ULL, 0x38e38e3871c71c72ULL, },
76a8f91dd9SMateja Marjanovic { 0x8888888777777778ULL, 0x8888888777777778ULL, },
77a8f91dd9SMateja Marjanovic { 0x22222221dddddddeULL, 0x22222221dddddddeULL, },
78a8f91dd9SMateja Marjanovic { 0x7b425ecfda12f685ULL, 0x5ed097b34bda12f7ULL, },
79a8f91dd9SMateja Marjanovic { 0x2f684bd97b425ed1ULL, 0x4bda12f6097b425fULL, },
80a8f91dd9SMateja Marjanovic { 0x9999999666666668ULL, 0x9999999666666668ULL, }, /* 32 */
81a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
82a8f91dd9SMateja Marjanovic { 0x1111110eeeeeeef0ULL, 0x1111110eeeeeeef0ULL, },
83a8f91dd9SMateja Marjanovic { 0x8888888777777778ULL, 0x8888888777777778ULL, },
84a8f91dd9SMateja Marjanovic { 0x47ae147851eb8520ULL, 0x47ae147851eb8520ULL, },
85a8f91dd9SMateja Marjanovic { 0x51eb851e147ae148ULL, 0x51eb851e147ae148ULL, },
86a8f91dd9SMateja Marjanovic { 0x27d27d260b60b60cULL, 0xe38e38e182d82d84ULL, },
87a8f91dd9SMateja Marjanovic { 0x71c71c705b05b05cULL, 0xb60b60b4e38e38e4ULL, },
88a8f91dd9SMateja Marjanovic { 0x666666659999999aULL, 0x666666659999999aULL, }, /* 40 */
89a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
90a8f91dd9SMateja Marjanovic { 0x44444443bbbbbbbcULL, 0x44444443bbbbbbbcULL, },
91a8f91dd9SMateja Marjanovic { 0x22222221dddddddeULL, 0x22222221dddddddeULL, },
92a8f91dd9SMateja Marjanovic { 0x51eb851e147ae148ULL, 0x51eb851e147ae148ULL, },
93a8f91dd9SMateja Marjanovic { 0x147ae147851eb852ULL, 0x147ae147851eb852ULL, },
94a8f91dd9SMateja Marjanovic { 0x49f49f4982d82d83ULL, 0x38e38e3860b60b61ULL, },
95a8f91dd9SMateja Marjanovic { 0x1c71c71c16c16c17ULL, 0x2d82d82d38e38e39ULL, },
96a8f91dd9SMateja Marjanovic { 0x71c71c6f8e38e38fULL, 0x1c71c719e38e38e5ULL, }, /* 48 */
97a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
98a8f91dd9SMateja Marjanovic { 0xf684bd9fb425ed0aULL, 0xbda12f6697b425eeULL, },
99a8f91dd9SMateja Marjanovic { 0x7b425ecfda12f685ULL, 0x5ed097b34bda12f7ULL, },
100a8f91dd9SMateja Marjanovic { 0x27d27d260b60b60cULL, 0xe38e38e182d82d84ULL, },
101a8f91dd9SMateja Marjanovic { 0x49f49f4982d82d83ULL, 0x38e38e3860b60b61ULL, },
102a8f91dd9SMateja Marjanovic { 0x1948b0fb9add3c0dULL, 0xd6e9e063c0ca4589ULL, },
103a8f91dd9SMateja Marjanovic { 0x587e6b73f35ba782ULL, 0x4587e6b622c3f35cULL, },
104a8f91dd9SMateja Marjanovic { 0x8e38e38c71c71c73ULL, 0xe38e38e21c71c71dULL, }, /* 56 */
105a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
106a8f91dd9SMateja Marjanovic { 0x5ed097b2f684bda2ULL, 0x97b425ec12f684beULL, },
107a8f91dd9SMateja Marjanovic { 0x2f684bd97b425ed1ULL, 0x4bda12f6097b425fULL, },
108a8f91dd9SMateja Marjanovic { 0x71c71c705b05b05cULL, 0xb60b60b4e38e38e4ULL, },
109a8f91dd9SMateja Marjanovic { 0x1c71c71c16c16c17ULL, 0x2d82d82d38e38e39ULL, },
110a8f91dd9SMateja Marjanovic { 0x587e6b73f35ba782ULL, 0x4587e6b622c3f35cULL, },
111a8f91dd9SMateja Marjanovic { 0x35ba78187e6b74f1ULL, 0x9e06522bf9add3c1ULL, },
112a8f91dd9SMateja Marjanovic { 0x4f10a2461266c290ULL, 0x132f373daebdb714ULL, }, /* 64 */
113a8f91dd9SMateja Marjanovic { 0x9262f356f7d6aae4ULL, 0x1ab54eb3f08ff55cULL, },
114a8f91dd9SMateja Marjanovic { 0x7927f2d937c84b78ULL, 0xb5e40e840fde8392ULL, },
115a8f91dd9SMateja Marjanovic { 0x4ab4e3ab9f696cdcULL, 0xd21109f6644a26d0ULL, },
116a8f91dd9SMateja Marjanovic { 0x9262f356f7d6aae4ULL, 0x1ab54eb3f08ff55cULL, },
117a8f91dd9SMateja Marjanovic { 0x0f105ccfb2e19689ULL, 0x032b395187d966b4ULL, },
118a8f91dd9SMateja Marjanovic { 0xe1cb8469b5aa67beULL, 0x1128ae6a67ae6296ULL, },
119a8f91dd9SMateja Marjanovic { 0x8afc46ad9bff5437ULL, 0x1890b25301b40370ULL, },
120a8f91dd9SMateja Marjanovic { 0x7927f2d937c84b78ULL, 0xb5e40e840fde8392ULL, }, /* 72 */
121a8f91dd9SMateja Marjanovic { 0xe1cb8469b5aa67beULL, 0x1128ae6a67ae6296ULL, },
122a8f91dd9SMateja Marjanovic { 0xfae79ab59bb7c8e4ULL, 0x78a66f004d163b91ULL, },
123a8f91dd9SMateja Marjanovic { 0x8ffb559e72a2a622ULL, 0x8744321b668933a8ULL, },
124a8f91dd9SMateja Marjanovic { 0x4ab4e3ab9f696cdcULL, 0xd21109f6644a26d0ULL, },
125a8f91dd9SMateja Marjanovic { 0x8afc46ad9bff5437ULL, 0x1890b25301b40370ULL, },
126750541c4SAleksandar Markovic { 0x8ffb559e72a2a622ULL, 0x8744321b668933a8ULL, },
127750541c4SAleksandar Markovic { 0x53edf7dbd76122edULL, 0xbe9d5551c2f51a40ULL, },
128a8f91dd9SMateja Marjanovic };
129a8f91dd9SMateja Marjanovic
130ca693d1cSAleksandar Markovic reset_msa_registers();
131ca693d1cSAleksandar Markovic
132a8f91dd9SMateja Marjanovic gettimeofday(&start, NULL);
133a8f91dd9SMateja Marjanovic
134a8f91dd9SMateja Marjanovic for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) {
135a8f91dd9SMateja Marjanovic for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) {
136a8f91dd9SMateja Marjanovic do_msa_DOTP_U_D(b128_pattern[i], b128_pattern[j],
137a8f91dd9SMateja Marjanovic b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]);
138a8f91dd9SMateja Marjanovic }
139a8f91dd9SMateja Marjanovic }
140a8f91dd9SMateja Marjanovic
141a8f91dd9SMateja Marjanovic for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
142a8f91dd9SMateja Marjanovic for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
143a8f91dd9SMateja Marjanovic do_msa_DOTP_U_D(b128_random[i], b128_random[j],
144a8f91dd9SMateja Marjanovic b128_result[((PATTERN_INPUTS_SHORT_COUNT) *
145a8f91dd9SMateja Marjanovic (PATTERN_INPUTS_SHORT_COUNT)) +
146a8f91dd9SMateja Marjanovic RANDOM_INPUTS_SHORT_COUNT * i + j]);
147a8f91dd9SMateja Marjanovic }
148a8f91dd9SMateja Marjanovic }
149a8f91dd9SMateja Marjanovic
150a8f91dd9SMateja Marjanovic gettimeofday(&end, NULL);
151a8f91dd9SMateja Marjanovic
152a8f91dd9SMateja Marjanovic elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0;
153a8f91dd9SMateja Marjanovic elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0;
154a8f91dd9SMateja Marjanovic
155*8e6fe6b8SAleksandar Markovic ret = check_results_128(isa_ase_name, group_name, instruction_name,
156*8e6fe6b8SAleksandar Markovic TEST_COUNT_TOTAL, elapsed_time,
157a8f91dd9SMateja Marjanovic &b128_result[0][0], &b128_expect[0][0]);
158a8f91dd9SMateja Marjanovic
159a8f91dd9SMateja Marjanovic return ret;
160a8f91dd9SMateja Marjanovic }
161