1a8f91dd9SMateja Marjanovic /*
2a8f91dd9SMateja Marjanovic * Test program for MSA instruction DOTP_U.W
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.W";
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 { 0xfffc0002fffc0002ULL, 0xfffc0002fffc0002ULL, }, /* 0 */
49a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
50a8f91dd9SMateja Marjanovic { 0x5552aaac5552aaacULL, 0x5552aaac5552aaacULL, },
51a8f91dd9SMateja Marjanovic { 0xaaa95556aaa95556ULL, 0xaaa95556aaa95556ULL, },
52a8f91dd9SMateja Marjanovic { 0x9996666899966668ULL, 0x9996666899966668ULL, },
53a8f91dd9SMateja Marjanovic { 0x6665999a6665999aULL, 0x6665999a6665999aULL, },
54a8f91dd9SMateja Marjanovic { 0x1c6fe38f71c48e3aULL, 0xc71a38e51c6fe38fULL, },
55a8f91dd9SMateja Marjanovic { 0xe38c1c738e3771c8ULL, 0x38e1c71de38c1c73ULL, },
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 { 0x5552aaac5552aaacULL, 0x5552aaac5552aaacULL, }, /* 16 */
65a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
66a8f91dd9SMateja Marjanovic { 0xe38c71c8e38c71c8ULL, 0xe38c71c8e38c71c8ULL, },
67a8f91dd9SMateja Marjanovic { 0x71c638e471c638e4ULL, 0x71c638e471c638e4ULL, },
68a8f91dd9SMateja Marjanovic { 0x110eeef0110eeef0ULL, 0x110eeef0110eeef0ULL, },
69a8f91dd9SMateja Marjanovic { 0x4443bbbc4443bbbcULL, 0x4443bbbc4443bbbcULL, },
70a8f91dd9SMateja Marjanovic { 0xbd9fed0af683097cULL, 0x84bc25eebd9fed0aULL, },
71a8f91dd9SMateja Marjanovic { 0x97b2bda25ecfa130ULL, 0xd09684be97b2bda2ULL, },
72a8f91dd9SMateja Marjanovic { 0xaaa95556aaa95556ULL, 0xaaa95556aaa95556ULL, }, /* 24 */
73a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
74a8f91dd9SMateja Marjanovic { 0x71c638e471c638e4ULL, 0x71c638e471c638e4ULL, },
75a8f91dd9SMateja Marjanovic { 0x38e31c7238e31c72ULL, 0x38e31c7238e31c72ULL, },
76a8f91dd9SMateja Marjanovic { 0x8887777888877778ULL, 0x8887777888877778ULL, },
77a8f91dd9SMateja Marjanovic { 0x2221ddde2221dddeULL, 0x2221ddde2221dddeULL, },
78a8f91dd9SMateja Marjanovic { 0x5ecff6857b4184beULL, 0x425e12f75ecff685ULL, },
79a8f91dd9SMateja Marjanovic { 0x4bd95ed12f67d098ULL, 0x684b425f4bd95ed1ULL, },
80a8f91dd9SMateja Marjanovic { 0x9996666899966668ULL, 0x9996666899966668ULL, }, /* 32 */
81a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
82a8f91dd9SMateja Marjanovic { 0x110eeef0110eeef0ULL, 0x110eeef0110eeef0ULL, },
83a8f91dd9SMateja Marjanovic { 0x8887777888877778ULL, 0x8887777888877778ULL, },
84a8f91dd9SMateja Marjanovic { 0x47ab852047ab8520ULL, 0x47ab852047ab8520ULL, },
85a8f91dd9SMateja Marjanovic { 0x51eae14851eae148ULL, 0x51eae14851eae148ULL, },
86a8f91dd9SMateja Marjanovic { 0xe38cb60c27d071c8ULL, 0x9f482d84e38cb60cULL, },
87a8f91dd9SMateja Marjanovic { 0xb609b05c71c5f4a0ULL, 0xfa4e38e4b609b05cULL, },
88a8f91dd9SMateja Marjanovic { 0x6665999a6665999aULL, 0x6665999a6665999aULL, }, /* 40 */
89a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
90a8f91dd9SMateja Marjanovic { 0x4443bbbc4443bbbcULL, 0x4443bbbc4443bbbcULL, },
91a8f91dd9SMateja Marjanovic { 0x2221ddde2221dddeULL, 0x2221ddde2221dddeULL, },
92a8f91dd9SMateja Marjanovic { 0x51eae14851eae148ULL, 0x51eae14851eae148ULL, },
93a8f91dd9SMateja Marjanovic { 0x147ab852147ab852ULL, 0x147ab852147ab852ULL, },
94a8f91dd9SMateja Marjanovic { 0x38e32d8349f41c72ULL, 0x27d20b6138e32d83ULL, },
95a8f91dd9SMateja Marjanovic { 0x2d826c171c717d28ULL, 0x3e938e392d826c17ULL, },
96a8f91dd9SMateja Marjanovic { 0x1c6fe38f71c48e3aULL, 0xc71a38e51c6fe38fULL, }, /* 48 */
97a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
98a8f91dd9SMateja Marjanovic { 0xbd9fed0af683097cULL, 0x84bc25eebd9fed0aULL, },
99a8f91dd9SMateja Marjanovic { 0x5ecff6857b4184beULL, 0x425e12f75ecff685ULL, },
100a8f91dd9SMateja Marjanovic { 0xe38cb60c27d071c8ULL, 0x9f482d84e38cb60cULL, },
101a8f91dd9SMateja Marjanovic { 0x38e32d8349f41c72ULL, 0x27d20b6138e32d83ULL, },
102a8f91dd9SMateja Marjanovic { 0xd6e93c0d19474f04ULL, 0x5ba64589d6e93c0dULL, },
103a8f91dd9SMateja Marjanovic { 0x4586a782587d3f36ULL, 0x6b73f35c4586a782ULL, },
104a8f91dd9SMateja Marjanovic { 0xe38c1c738e3771c8ULL, 0x38e1c71de38c1c73ULL, }, /* 56 */
105a8f91dd9SMateja Marjanovic { 0x0000000000000000ULL, 0x0000000000000000ULL, },
106a8f91dd9SMateja Marjanovic { 0x97b2bda25ecfa130ULL, 0xd09684be97b2bda2ULL, },
107a8f91dd9SMateja Marjanovic { 0x4bd95ed12f67d098ULL, 0x684b425f4bd95ed1ULL, },
108a8f91dd9SMateja Marjanovic { 0xb609b05c71c5f4a0ULL, 0xfa4e38e4b609b05cULL, },
109a8f91dd9SMateja Marjanovic { 0x2d826c171c717d28ULL, 0x3e938e392d826c17ULL, },
110a8f91dd9SMateja Marjanovic { 0x4586a782587d3f36ULL, 0x6b73f35c4586a782ULL, },
111a8f91dd9SMateja Marjanovic { 0x9e0574f135ba3292ULL, 0xcd6dd3c19e0574f1ULL, },
112a8f91dd9SMateja Marjanovic { 0x18c3fe7422c25584ULL, 0x16b6b9f57608cfa9ULL, }, /* 64 */
113a8f91dd9SMateja Marjanovic { 0x867e6d904e841446ULL, 0x0de4cfed4e2fdb15ULL, },
114a8f91dd9SMateja Marjanovic { 0xf94f18bc4bc3d93eULL, 0x1492568ac3a66499ULL, },
115a8f91dd9SMateja Marjanovic { 0x4ff36c125a383042ULL, 0x2fe23e4744196e36ULL, },
116a8f91dd9SMateja Marjanovic { 0x867e6d904e841446ULL, 0x0de4cfed4e2fdb15ULL, },
117a8f91dd9SMateja Marjanovic { 0xf78e474db23f32a9ULL, 0x8a26a8f51ca9cd91ULL, },
118a8f91dd9SMateja Marjanovic { 0xa9bfb48aa4c2d0ddULL, 0x94641c4e1a398e45ULL, },
119a8f91dd9SMateja Marjanovic { 0x6e796f69cc7c8793ULL, 0x6e879377578266beULL, },
120a8f91dd9SMateja Marjanovic { 0xf94f18bc4bc3d93eULL, 0x1492568ac3a66499ULL, }, /* 72 */
121a8f91dd9SMateja Marjanovic { 0xa9bfb48aa4c2d0ddULL, 0x94641c4e1a398e45ULL, },
122a8f91dd9SMateja Marjanovic { 0xeb349888d2e11561ULL, 0xa0e2f84177d142c9ULL, },
123a8f91dd9SMateja Marjanovic { 0x5ad3b4e8bfaf139fULL, 0x8076d98091fe5896ULL, },
124a8f91dd9SMateja Marjanovic { 0x4ff36c125a383042ULL, 0x2fe23e4744196e36ULL, },
125a8f91dd9SMateja Marjanovic { 0x6e796f69cc7c8793ULL, 0x6e879377578266beULL, },
126750541c4SAleksandar Markovic { 0x5ad3b4e8bfaf139fULL, 0x8076d98091fe5896ULL, },
127750541c4SAleksandar Markovic { 0x33368b8aeab5d525ULL, 0x97d9932138871904ULL, },
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_W(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_W(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