1 /*
2  *  Test program for MSA instruction DOTP_U.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             (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 = "Int Dot Product";
40     char *instruction_name =  "DOTP_U.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         { 0xfffc0002fffc0002ULL, 0xfffc0002fffc0002ULL, },    /*   0  */
49         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
50         { 0x5552aaac5552aaacULL, 0x5552aaac5552aaacULL, },
51         { 0xaaa95556aaa95556ULL, 0xaaa95556aaa95556ULL, },
52         { 0x9996666899966668ULL, 0x9996666899966668ULL, },
53         { 0x6665999a6665999aULL, 0x6665999a6665999aULL, },
54         { 0x1c6fe38f71c48e3aULL, 0xc71a38e51c6fe38fULL, },
55         { 0xe38c1c738e3771c8ULL, 0x38e1c71de38c1c73ULL, },
56         { 0x0000000000000000ULL, 0x0000000000000000ULL, },    /*   8  */
57         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
58         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
59         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
60         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
61         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
62         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
63         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
64         { 0x5552aaac5552aaacULL, 0x5552aaac5552aaacULL, },    /*  16  */
65         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
66         { 0xe38c71c8e38c71c8ULL, 0xe38c71c8e38c71c8ULL, },
67         { 0x71c638e471c638e4ULL, 0x71c638e471c638e4ULL, },
68         { 0x110eeef0110eeef0ULL, 0x110eeef0110eeef0ULL, },
69         { 0x4443bbbc4443bbbcULL, 0x4443bbbc4443bbbcULL, },
70         { 0xbd9fed0af683097cULL, 0x84bc25eebd9fed0aULL, },
71         { 0x97b2bda25ecfa130ULL, 0xd09684be97b2bda2ULL, },
72         { 0xaaa95556aaa95556ULL, 0xaaa95556aaa95556ULL, },    /*  24  */
73         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
74         { 0x71c638e471c638e4ULL, 0x71c638e471c638e4ULL, },
75         { 0x38e31c7238e31c72ULL, 0x38e31c7238e31c72ULL, },
76         { 0x8887777888877778ULL, 0x8887777888877778ULL, },
77         { 0x2221ddde2221dddeULL, 0x2221ddde2221dddeULL, },
78         { 0x5ecff6857b4184beULL, 0x425e12f75ecff685ULL, },
79         { 0x4bd95ed12f67d098ULL, 0x684b425f4bd95ed1ULL, },
80         { 0x9996666899966668ULL, 0x9996666899966668ULL, },    /*  32  */
81         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
82         { 0x110eeef0110eeef0ULL, 0x110eeef0110eeef0ULL, },
83         { 0x8887777888877778ULL, 0x8887777888877778ULL, },
84         { 0x47ab852047ab8520ULL, 0x47ab852047ab8520ULL, },
85         { 0x51eae14851eae148ULL, 0x51eae14851eae148ULL, },
86         { 0xe38cb60c27d071c8ULL, 0x9f482d84e38cb60cULL, },
87         { 0xb609b05c71c5f4a0ULL, 0xfa4e38e4b609b05cULL, },
88         { 0x6665999a6665999aULL, 0x6665999a6665999aULL, },    /*  40  */
89         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
90         { 0x4443bbbc4443bbbcULL, 0x4443bbbc4443bbbcULL, },
91         { 0x2221ddde2221dddeULL, 0x2221ddde2221dddeULL, },
92         { 0x51eae14851eae148ULL, 0x51eae14851eae148ULL, },
93         { 0x147ab852147ab852ULL, 0x147ab852147ab852ULL, },
94         { 0x38e32d8349f41c72ULL, 0x27d20b6138e32d83ULL, },
95         { 0x2d826c171c717d28ULL, 0x3e938e392d826c17ULL, },
96         { 0x1c6fe38f71c48e3aULL, 0xc71a38e51c6fe38fULL, },    /*  48  */
97         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
98         { 0xbd9fed0af683097cULL, 0x84bc25eebd9fed0aULL, },
99         { 0x5ecff6857b4184beULL, 0x425e12f75ecff685ULL, },
100         { 0xe38cb60c27d071c8ULL, 0x9f482d84e38cb60cULL, },
101         { 0x38e32d8349f41c72ULL, 0x27d20b6138e32d83ULL, },
102         { 0xd6e93c0d19474f04ULL, 0x5ba64589d6e93c0dULL, },
103         { 0x4586a782587d3f36ULL, 0x6b73f35c4586a782ULL, },
104         { 0xe38c1c738e3771c8ULL, 0x38e1c71de38c1c73ULL, },    /*  56  */
105         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
106         { 0x97b2bda25ecfa130ULL, 0xd09684be97b2bda2ULL, },
107         { 0x4bd95ed12f67d098ULL, 0x684b425f4bd95ed1ULL, },
108         { 0xb609b05c71c5f4a0ULL, 0xfa4e38e4b609b05cULL, },
109         { 0x2d826c171c717d28ULL, 0x3e938e392d826c17ULL, },
110         { 0x4586a782587d3f36ULL, 0x6b73f35c4586a782ULL, },
111         { 0x9e0574f135ba3292ULL, 0xcd6dd3c19e0574f1ULL, },
112         { 0x18c3fe7422c25584ULL, 0x16b6b9f57608cfa9ULL, },    /*  64  */
113         { 0x867e6d904e841446ULL, 0x0de4cfed4e2fdb15ULL, },
114         { 0xf94f18bc4bc3d93eULL, 0x1492568ac3a66499ULL, },
115         { 0x4ff36c125a383042ULL, 0x2fe23e4744196e36ULL, },
116         { 0x867e6d904e841446ULL, 0x0de4cfed4e2fdb15ULL, },
117         { 0xf78e474db23f32a9ULL, 0x8a26a8f51ca9cd91ULL, },
118         { 0xa9bfb48aa4c2d0ddULL, 0x94641c4e1a398e45ULL, },
119         { 0x6e796f69cc7c8793ULL, 0x6e879377578266beULL, },
120         { 0xf94f18bc4bc3d93eULL, 0x1492568ac3a66499ULL, },    /*  72  */
121         { 0xa9bfb48aa4c2d0ddULL, 0x94641c4e1a398e45ULL, },
122         { 0xeb349888d2e11561ULL, 0xa0e2f84177d142c9ULL, },
123         { 0x5ad3b4e8bfaf139fULL, 0x8076d98091fe5896ULL, },
124         { 0x4ff36c125a383042ULL, 0x2fe23e4744196e36ULL, },
125         { 0x6e796f69cc7c8793ULL, 0x6e879377578266beULL, },
126         { 0x5ad3b4e8bfaf139fULL, 0x8076d98091fe5896ULL, },
127         { 0x33368b8aeab5d525ULL, 0x97d9932138871904ULL, },
128 };
129 
130     reset_msa_registers();
131 
132     gettimeofday(&start, NULL);
133 
134     for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) {
135         for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) {
136             do_msa_DOTP_U_W(b128_pattern[i], b128_pattern[j],
137                            b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]);
138         }
139     }
140 
141     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
142         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
143             do_msa_DOTP_U_W(b128_random[i], b128_random[j],
144                            b128_result[((PATTERN_INPUTS_SHORT_COUNT) *
145                                         (PATTERN_INPUTS_SHORT_COUNT)) +
146                                        RANDOM_INPUTS_SHORT_COUNT * i + j]);
147         }
148     }
149 
150     gettimeofday(&end, NULL);
151 
152     elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0;
153     elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0;
154 
155     ret = check_results_128(isa_ase_name, group_name, instruction_name,
156                             TEST_COUNT_TOTAL, elapsed_time,
157                             &b128_result[0][0], &b128_expect[0][0]);
158 
159     return ret;
160 }
161