1 /*
2  *  Test program for MSA instruction ILVR.W
3  *
4  *  Copyright (C) 2019  Wave Computing, Inc.
5  *  Copyright (C) 2019  Aleksandar Markovic <amarkovic@wavecomp.com>
6  *
7  *  This program is free software: you can redistribute it and/or modify
8  *  it under the terms of the GNU General Public License as published by
9  *  the Free Software Foundation, either version 2 of the License, or
10  *  (at your option) any later version.
11  *
12  *  This program is distributed in the hope that it will be useful,
13  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
14  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  *  GNU General Public License for more details.
16  *
17  *  You should have received a copy of the GNU General Public License
18  *  along with this program.  If not, see <https://www.gnu.org/licenses/>.
19  *
20  */
21 
22 #include <sys/time.h>
23 #include <stdint.h>
24 
25 #include "../../../../include/wrappers_msa.h"
26 #include "../../../../include/test_inputs_128.h"
27 #include "../../../../include/test_utils_128.h"
28 
29 #define TEST_COUNT_TOTAL (                                                \
30             (PATTERN_INPUTS_SHORT_COUNT) * (PATTERN_INPUTS_SHORT_COUNT) + \
31             (RANDOM_INPUTS_SHORT_COUNT) * (RANDOM_INPUTS_SHORT_COUNT))
32 
33 
34 int32_t main(void)
35 {
36     char *isa_ase_name = "MSA";
37     char *group_name = "Interleave";
38     char *instruction_name =  "ILVR.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         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*   0  */
47         { 0xffffffff00000000ULL, 0xffffffff00000000ULL, },
48         { 0xffffffffaaaaaaaaULL, 0xffffffffaaaaaaaaULL, },
49         { 0xffffffff55555555ULL, 0xffffffff55555555ULL, },
50         { 0xffffffffccccccccULL, 0xffffffffccccccccULL, },
51         { 0xffffffff33333333ULL, 0xffffffff33333333ULL, },
52         { 0xffffffff8e38e38eULL, 0xffffffffe38e38e3ULL, },
53         { 0xffffffff71c71c71ULL, 0xffffffff1c71c71cULL, },
54         { 0x00000000ffffffffULL, 0x00000000ffffffffULL, },    /*   8  */
55         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
56         { 0x00000000aaaaaaaaULL, 0x00000000aaaaaaaaULL, },
57         { 0x0000000055555555ULL, 0x0000000055555555ULL, },
58         { 0x00000000ccccccccULL, 0x00000000ccccccccULL, },
59         { 0x0000000033333333ULL, 0x0000000033333333ULL, },
60         { 0x000000008e38e38eULL, 0x00000000e38e38e3ULL, },
61         { 0x0000000071c71c71ULL, 0x000000001c71c71cULL, },
62         { 0xaaaaaaaaffffffffULL, 0xaaaaaaaaffffffffULL, },    /*  16  */
63         { 0xaaaaaaaa00000000ULL, 0xaaaaaaaa00000000ULL, },
64         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
65         { 0xaaaaaaaa55555555ULL, 0xaaaaaaaa55555555ULL, },
66         { 0xaaaaaaaaccccccccULL, 0xaaaaaaaaccccccccULL, },
67         { 0xaaaaaaaa33333333ULL, 0xaaaaaaaa33333333ULL, },
68         { 0xaaaaaaaa8e38e38eULL, 0xaaaaaaaae38e38e3ULL, },
69         { 0xaaaaaaaa71c71c71ULL, 0xaaaaaaaa1c71c71cULL, },
70         { 0x55555555ffffffffULL, 0x55555555ffffffffULL, },    /*  24  */
71         { 0x5555555500000000ULL, 0x5555555500000000ULL, },
72         { 0x55555555aaaaaaaaULL, 0x55555555aaaaaaaaULL, },
73         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
74         { 0x55555555ccccccccULL, 0x55555555ccccccccULL, },
75         { 0x5555555533333333ULL, 0x5555555533333333ULL, },
76         { 0x555555558e38e38eULL, 0x55555555e38e38e3ULL, },
77         { 0x5555555571c71c71ULL, 0x555555551c71c71cULL, },
78         { 0xccccccccffffffffULL, 0xccccccccffffffffULL, },    /*  32  */
79         { 0xcccccccc00000000ULL, 0xcccccccc00000000ULL, },
80         { 0xccccccccaaaaaaaaULL, 0xccccccccaaaaaaaaULL, },
81         { 0xcccccccc55555555ULL, 0xcccccccc55555555ULL, },
82         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
83         { 0xcccccccc33333333ULL, 0xcccccccc33333333ULL, },
84         { 0xcccccccc8e38e38eULL, 0xcccccccce38e38e3ULL, },
85         { 0xcccccccc71c71c71ULL, 0xcccccccc1c71c71cULL, },
86         { 0x33333333ffffffffULL, 0x33333333ffffffffULL, },    /*  40  */
87         { 0x3333333300000000ULL, 0x3333333300000000ULL, },
88         { 0x33333333aaaaaaaaULL, 0x33333333aaaaaaaaULL, },
89         { 0x3333333355555555ULL, 0x3333333355555555ULL, },
90         { 0x33333333ccccccccULL, 0x33333333ccccccccULL, },
91         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
92         { 0x333333338e38e38eULL, 0x33333333e38e38e3ULL, },
93         { 0x3333333371c71c71ULL, 0x333333331c71c71cULL, },
94         { 0x8e38e38effffffffULL, 0xe38e38e3ffffffffULL, },    /*  48  */
95         { 0x8e38e38e00000000ULL, 0xe38e38e300000000ULL, },
96         { 0x8e38e38eaaaaaaaaULL, 0xe38e38e3aaaaaaaaULL, },
97         { 0x8e38e38e55555555ULL, 0xe38e38e355555555ULL, },
98         { 0x8e38e38eccccccccULL, 0xe38e38e3ccccccccULL, },
99         { 0x8e38e38e33333333ULL, 0xe38e38e333333333ULL, },
100         { 0x8e38e38e8e38e38eULL, 0xe38e38e3e38e38e3ULL, },
101         { 0x8e38e38e71c71c71ULL, 0xe38e38e31c71c71cULL, },
102         { 0x71c71c71ffffffffULL, 0x1c71c71cffffffffULL, },    /*  56  */
103         { 0x71c71c7100000000ULL, 0x1c71c71c00000000ULL, },
104         { 0x71c71c71aaaaaaaaULL, 0x1c71c71caaaaaaaaULL, },
105         { 0x71c71c7155555555ULL, 0x1c71c71c55555555ULL, },
106         { 0x71c71c71ccccccccULL, 0x1c71c71cccccccccULL, },
107         { 0x71c71c7133333333ULL, 0x1c71c71c33333333ULL, },
108         { 0x71c71c718e38e38eULL, 0x1c71c71ce38e38e3ULL, },
109         { 0x71c71c7171c71c71ULL, 0x1c71c71c1c71c71cULL, },
110         { 0x2862554028625540ULL, 0x886ae6cc886ae6ccULL, },    /*  64  */
111         { 0x286255404d93c708ULL, 0x886ae6ccfbbe0063ULL, },
112         { 0x28625540b9cf8b80ULL, 0x886ae6ccac5aaeaaULL, },
113         { 0x286255405e31e24eULL, 0x886ae6cc704f164dULL, },
114         { 0x4d93c70828625540ULL, 0xfbbe0063886ae6ccULL, },
115         { 0x4d93c7084d93c708ULL, 0xfbbe0063fbbe0063ULL, },
116         { 0x4d93c708b9cf8b80ULL, 0xfbbe0063ac5aaeaaULL, },
117         { 0x4d93c7085e31e24eULL, 0xfbbe0063704f164dULL, },
118         { 0xb9cf8b8028625540ULL, 0xac5aaeaa886ae6ccULL, },    /*  72  */
119         { 0xb9cf8b804d93c708ULL, 0xac5aaeaafbbe0063ULL, },
120         { 0xb9cf8b80b9cf8b80ULL, 0xac5aaeaaac5aaeaaULL, },
121         { 0xb9cf8b805e31e24eULL, 0xac5aaeaa704f164dULL, },
122         { 0x5e31e24e28625540ULL, 0x704f164d886ae6ccULL, },
123         { 0x5e31e24e4d93c708ULL, 0x704f164dfbbe0063ULL, },
124         { 0x5e31e24eb9cf8b80ULL, 0x704f164dac5aaeaaULL, },
125         { 0x5e31e24e5e31e24eULL, 0x704f164d704f164dULL, },
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_ILVR_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_ILVR_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_128(isa_ase_name, group_name, instruction_name,
154                             TEST_COUNT_TOTAL, elapsed_time,
155                             &b128_result[0][0], &b128_expect[0][0]);
156 
157     return ret;
158 }
159