1 /*
2  *  Test program for MSA instruction SUBS_S.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 *instruction_name = "SUBS_S.W";
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         { 0x0000000000000000ULL, 0x0000000000000000ULL, },    /*   0  */
47         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
48         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
49         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
50         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
51         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
52         { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, },
53         { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, },
54         { 0x0000000100000001ULL, 0x0000000100000001ULL, },    /*   8  */
55         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
56         { 0x5555555655555556ULL, 0x5555555655555556ULL, },
57         { 0xaaaaaaabaaaaaaabULL, 0xaaaaaaabaaaaaaabULL, },
58         { 0x3333333433333334ULL, 0x3333333433333334ULL, },
59         { 0xcccccccdcccccccdULL, 0xcccccccdcccccccdULL, },
60         { 0x1c71c71d71c71c72ULL, 0xc71c71c81c71c71dULL, },
61         { 0xe38e38e48e38e38fULL, 0x38e38e39e38e38e4ULL, },
62         { 0xaaaaaaabaaaaaaabULL, 0xaaaaaaabaaaaaaabULL, },    /*  16  */
63         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
64         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
65         { 0x8000000080000000ULL, 0x8000000080000000ULL, },
66         { 0xdddddddedddddddeULL, 0xdddddddedddddddeULL, },
67         { 0x8000000080000000ULL, 0x8000000080000000ULL, },
68         { 0xc71c71c71c71c71cULL, 0x80000000c71c71c7ULL, },
69         { 0x8e38e38e80000000ULL, 0xe38e38e38e38e38eULL, },
70         { 0x5555555655555556ULL, 0x5555555655555556ULL, },    /*  24  */
71         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
72         { 0x7fffffff7fffffffULL, 0x7fffffff7fffffffULL, },
73         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
74         { 0x7fffffff7fffffffULL, 0x7fffffff7fffffffULL, },
75         { 0x2222222222222222ULL, 0x2222222222222222ULL, },
76         { 0x71c71c727fffffffULL, 0x1c71c71d71c71c72ULL, },
77         { 0x38e38e39e38e38e4ULL, 0x7fffffff38e38e39ULL, },
78         { 0xcccccccdcccccccdULL, 0xcccccccdcccccccdULL, },    /*  32  */
79         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
80         { 0x2222222222222222ULL, 0x2222222222222222ULL, },
81         { 0x8000000080000000ULL, 0x8000000080000000ULL, },
82         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
83         { 0x9999999999999999ULL, 0x9999999999999999ULL, },
84         { 0xe93e93e93e93e93eULL, 0x93e93e94e93e93e9ULL, },
85         { 0xb05b05b080000000ULL, 0x05b05b05b05b05b0ULL, },
86         { 0x3333333433333334ULL, 0x3333333433333334ULL, },    /*  40  */
87         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
88         { 0x7fffffff7fffffffULL, 0x7fffffff7fffffffULL, },
89         { 0xdddddddedddddddeULL, 0xdddddddedddddddeULL, },
90         { 0x6666666766666667ULL, 0x6666666766666667ULL, },
91         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
92         { 0x4fa4fa507fffffffULL, 0xfa4fa4fb4fa4fa50ULL, },
93         { 0x16c16c17c16c16c2ULL, 0x6c16c16c16c16c17ULL, },
94         { 0xe38e38e48e38e38fULL, 0x38e38e39e38e38e4ULL, },    /*  48  */
95         { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, },
96         { 0x38e38e39e38e38e4ULL, 0x7fffffff38e38e39ULL, },
97         { 0x8e38e38e80000000ULL, 0xe38e38e38e38e38eULL, },
98         { 0x16c16c17c16c16c2ULL, 0x6c16c16c16c16c17ULL, },
99         { 0xb05b05b080000000ULL, 0x05b05b05b05b05b0ULL, },
100         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
101         { 0xc71c71c780000000ULL, 0x71c71c71c71c71c7ULL, },
102         { 0x1c71c71d71c71c72ULL, 0xc71c71c81c71c71dULL, },    /*  56  */
103         { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, },
104         { 0x71c71c727fffffffULL, 0x1c71c71d71c71c72ULL, },
105         { 0xc71c71c71c71c71cULL, 0x80000000c71c71c7ULL, },
106         { 0x4fa4fa507fffffffULL, 0xfa4fa4fb4fa4fa50ULL, },
107         { 0xe93e93e93e93e93eULL, 0x93e93e94e93e93e9ULL, },
108         { 0x38e38e397fffffffULL, 0x8e38e38f38e38e39ULL, },
109         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
110         { 0x0000000000000000ULL, 0x0000000000000000ULL, },    /*  64  */
111         { 0x8cace669dace8e38ULL, 0x386f5044e93c5d10ULL, },
112         { 0xdc1038226e92c9c0ULL, 0x238e445f53508af8ULL, },
113         { 0x80000000ca3072f2ULL, 0x7fffffff5538cd6cULL, },
114         { 0x73531997253171c8ULL, 0xc790afbc16c3a2f0ULL, },
115         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
116         { 0x4f6351b97fffffffULL, 0xeb1ef41b6a142de8ULL, },
117         { 0x8b6eea16ef61e4baULL, 0x7fffffff6bfc705cULL, },
118         { 0x23efc7de916d3640ULL, 0xdc71bba1acaf7508ULL, },    /*  72  */
119         { 0xb09cae4780000000ULL, 0x14e10be595ebd218ULL, },
120         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
121         { 0x8000000080000000ULL, 0x7fffffff01e84274ULL, },
122         { 0x7fffffff35cf8d0eULL, 0x80000000aac73294ULL, },
123         { 0x749115ea109e1b46ULL, 0x8000000094038fa4ULL, },
124         { 0x7fffffff7fffffffULL, 0x80000000fe17bd8cULL, },
125         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
126 };
127 
128     reset_msa_registers();
129 
130     gettimeofday(&start, NULL);
131 
132     for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) {
133         for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) {
134             do_msa_SUBS_S_W(b128_pattern[i], b128_pattern[j],
135                            b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]);
136         }
137     }
138 
139     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
140         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
141             do_msa_SUBS_S_W(b128_random[i], b128_random[j],
142                            b128_result[((PATTERN_INPUTS_SHORT_COUNT) *
143                                         (PATTERN_INPUTS_SHORT_COUNT)) +
144                                        RANDOM_INPUTS_SHORT_COUNT * i + j]);
145         }
146     }
147 
148     gettimeofday(&end, NULL);
149 
150     elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0;
151     elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0;
152 
153     ret = check_results(instruction_name, TEST_COUNT_TOTAL, elapsed_time,
154                         &b128_result[0][0], &b128_expect[0][0]);
155 
156     return ret;
157 }
158