1 /*
2  *  Test program for MSA instruction HADD_S.D
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 = "HADD_S.D";
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         { 0xfffffffffffffffeULL, 0xfffffffffffffffeULL, },    /*   0  */
47         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
48         { 0xffffffffaaaaaaa9ULL, 0xffffffffaaaaaaa9ULL, },
49         { 0x0000000055555554ULL, 0x0000000055555554ULL, },
50         { 0xffffffffcccccccbULL, 0xffffffffcccccccbULL, },
51         { 0x0000000033333332ULL, 0x0000000033333332ULL, },
52         { 0xffffffff8e38e38dULL, 0xffffffffe38e38e2ULL, },
53         { 0x0000000071c71c70ULL, 0x000000001c71c71bULL, },
54         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*   8  */
55         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
56         { 0xffffffffaaaaaaaaULL, 0xffffffffaaaaaaaaULL, },
57         { 0x0000000055555555ULL, 0x0000000055555555ULL, },
58         { 0xffffffffccccccccULL, 0xffffffffccccccccULL, },
59         { 0x0000000033333333ULL, 0x0000000033333333ULL, },
60         { 0xffffffff8e38e38eULL, 0xffffffffe38e38e3ULL, },
61         { 0x0000000071c71c71ULL, 0x000000001c71c71cULL, },
62         { 0xffffffffaaaaaaa9ULL, 0xffffffffaaaaaaa9ULL, },    /*  16  */
63         { 0xffffffffaaaaaaaaULL, 0xffffffffaaaaaaaaULL, },
64         { 0xffffffff55555554ULL, 0xffffffff55555554ULL, },
65         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
66         { 0xffffffff77777776ULL, 0xffffffff77777776ULL, },
67         { 0xffffffffddddddddULL, 0xffffffffddddddddULL, },
68         { 0xffffffff38e38e38ULL, 0xffffffff8e38e38dULL, },
69         { 0x000000001c71c71bULL, 0xffffffffc71c71c6ULL, },
70         { 0x0000000055555554ULL, 0x0000000055555554ULL, },    /*  24  */
71         { 0x0000000055555555ULL, 0x0000000055555555ULL, },
72         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
73         { 0x00000000aaaaaaaaULL, 0x00000000aaaaaaaaULL, },
74         { 0x0000000022222221ULL, 0x0000000022222221ULL, },
75         { 0x0000000088888888ULL, 0x0000000088888888ULL, },
76         { 0xffffffffe38e38e3ULL, 0x0000000038e38e38ULL, },
77         { 0x00000000c71c71c6ULL, 0x0000000071c71c71ULL, },
78         { 0xffffffffcccccccbULL, 0xffffffffcccccccbULL, },    /*  32  */
79         { 0xffffffffccccccccULL, 0xffffffffccccccccULL, },
80         { 0xffffffff77777776ULL, 0xffffffff77777776ULL, },
81         { 0x0000000022222221ULL, 0x0000000022222221ULL, },
82         { 0xffffffff99999998ULL, 0xffffffff99999998ULL, },
83         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
84         { 0xffffffff5b05b05aULL, 0xffffffffb05b05afULL, },
85         { 0x000000003e93e93dULL, 0xffffffffe93e93e8ULL, },
86         { 0x0000000033333332ULL, 0x0000000033333332ULL, },    /*  40  */
87         { 0x0000000033333333ULL, 0x0000000033333333ULL, },
88         { 0xffffffffddddddddULL, 0xffffffffddddddddULL, },
89         { 0x0000000088888888ULL, 0x0000000088888888ULL, },
90         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
91         { 0x0000000066666666ULL, 0x0000000066666666ULL, },
92         { 0xffffffffc16c16c1ULL, 0x0000000016c16c16ULL, },
93         { 0x00000000a4fa4fa4ULL, 0x000000004fa4fa4fULL, },
94         { 0xffffffffe38e38e2ULL, 0x0000000038e38e37ULL, },    /*  48  */
95         { 0xffffffffe38e38e3ULL, 0x0000000038e38e38ULL, },
96         { 0xffffffff8e38e38dULL, 0xffffffffe38e38e2ULL, },
97         { 0x0000000038e38e38ULL, 0x000000008e38e38dULL, },
98         { 0xffffffffb05b05afULL, 0x0000000005b05b04ULL, },
99         { 0x0000000016c16c16ULL, 0x000000006c16c16bULL, },
100         { 0xffffffff71c71c71ULL, 0x000000001c71c71bULL, },
101         { 0x0000000055555554ULL, 0x0000000055555554ULL, },
102         { 0x000000001c71c71bULL, 0xffffffffc71c71c6ULL, },    /*  56  */
103         { 0x000000001c71c71cULL, 0xffffffffc71c71c7ULL, },
104         { 0xffffffffc71c71c6ULL, 0xffffffff71c71c71ULL, },
105         { 0x0000000071c71c71ULL, 0x000000001c71c71cULL, },
106         { 0xffffffffe93e93e8ULL, 0xffffffff93e93e93ULL, },
107         { 0x000000004fa4fa4fULL, 0xfffffffffa4fa4faULL, },
108         { 0xffffffffaaaaaaaaULL, 0xffffffffaaaaaaaaULL, },
109         { 0x000000008e38e38dULL, 0xffffffffe38e38e3ULL, },
110         { 0xffffffffb0cd3c0cULL, 0x0000000049e2bb6aULL, },    /*  64  */
111         { 0xffffffffd5feadd4ULL, 0x0000000060a65e5aULL, },
112         { 0xffffffff423a724cULL, 0xfffffffff6923072ULL, },
113         { 0xffffffffe69cc91aULL, 0xfffffffff4a9edfeULL, },
114         { 0x00000000242055a3ULL, 0x0000000011736b26ULL, },
115         { 0x000000004951c76bULL, 0x0000000028370e16ULL, },
116         { 0xffffffffb58d8be3ULL, 0xffffffffbe22e02eULL, },
117         { 0x0000000059efe2b1ULL, 0xffffffffbc3a9dbaULL, },
118         { 0xffffffffd4bd03eaULL, 0x000000002654770bULL, },    /*  72  */
119         { 0xfffffffff9ee75b2ULL, 0x000000003d1819fbULL, },
120         { 0xffffffff662a3a2aULL, 0xffffffffd303ec13ULL, },
121         { 0x000000000a8c90f8ULL, 0xffffffffd11ba99fULL, },
122         { 0x0000000098b16b8dULL, 0xffffffff8c6d38e4ULL, },
123         { 0x00000000bde2dd55ULL, 0xffffffffa330dbd4ULL, },
124         { 0x000000002a1ea1cdULL, 0xffffffff391cadecULL, },
125         { 0x00000000ce80f89bULL, 0xffffffff37346b78ULL, },
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_HADD_S_D(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_HADD_S_D(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