1 /*
2  *  Test program for MSA instruction ILVR.H
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.H";
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         { 0xffff0000ffff0000ULL, 0xffff0000ffff0000ULL, },
48         { 0xffffaaaaffffaaaaULL, 0xffffaaaaffffaaaaULL, },
49         { 0xffff5555ffff5555ULL, 0xffff5555ffff5555ULL, },
50         { 0xffffccccffffccccULL, 0xffffccccffffccccULL, },
51         { 0xffff3333ffff3333ULL, 0xffff3333ffff3333ULL, },
52         { 0xffff8e38ffffe38eULL, 0xffffe38effff38e3ULL, },
53         { 0xffff71c7ffff1c71ULL, 0xffff1c71ffffc71cULL, },
54         { 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL, },    /*   8  */
55         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
56         { 0x0000aaaa0000aaaaULL, 0x0000aaaa0000aaaaULL, },
57         { 0x0000555500005555ULL, 0x0000555500005555ULL, },
58         { 0x0000cccc0000ccccULL, 0x0000cccc0000ccccULL, },
59         { 0x0000333300003333ULL, 0x0000333300003333ULL, },
60         { 0x00008e380000e38eULL, 0x0000e38e000038e3ULL, },
61         { 0x000071c700001c71ULL, 0x00001c710000c71cULL, },
62         { 0xaaaaffffaaaaffffULL, 0xaaaaffffaaaaffffULL, },    /*  16  */
63         { 0xaaaa0000aaaa0000ULL, 0xaaaa0000aaaa0000ULL, },
64         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
65         { 0xaaaa5555aaaa5555ULL, 0xaaaa5555aaaa5555ULL, },
66         { 0xaaaaccccaaaaccccULL, 0xaaaaccccaaaaccccULL, },
67         { 0xaaaa3333aaaa3333ULL, 0xaaaa3333aaaa3333ULL, },
68         { 0xaaaa8e38aaaae38eULL, 0xaaaae38eaaaa38e3ULL, },
69         { 0xaaaa71c7aaaa1c71ULL, 0xaaaa1c71aaaac71cULL, },
70         { 0x5555ffff5555ffffULL, 0x5555ffff5555ffffULL, },    /*  24  */
71         { 0x5555000055550000ULL, 0x5555000055550000ULL, },
72         { 0x5555aaaa5555aaaaULL, 0x5555aaaa5555aaaaULL, },
73         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
74         { 0x5555cccc5555ccccULL, 0x5555cccc5555ccccULL, },
75         { 0x5555333355553333ULL, 0x5555333355553333ULL, },
76         { 0x55558e385555e38eULL, 0x5555e38e555538e3ULL, },
77         { 0x555571c755551c71ULL, 0x55551c715555c71cULL, },
78         { 0xccccffffccccffffULL, 0xccccffffccccffffULL, },    /*  32  */
79         { 0xcccc0000cccc0000ULL, 0xcccc0000cccc0000ULL, },
80         { 0xccccaaaaccccaaaaULL, 0xccccaaaaccccaaaaULL, },
81         { 0xcccc5555cccc5555ULL, 0xcccc5555cccc5555ULL, },
82         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
83         { 0xcccc3333cccc3333ULL, 0xcccc3333cccc3333ULL, },
84         { 0xcccc8e38cccce38eULL, 0xcccce38ecccc38e3ULL, },
85         { 0xcccc71c7cccc1c71ULL, 0xcccc1c71ccccc71cULL, },
86         { 0x3333ffff3333ffffULL, 0x3333ffff3333ffffULL, },    /*  40  */
87         { 0x3333000033330000ULL, 0x3333000033330000ULL, },
88         { 0x3333aaaa3333aaaaULL, 0x3333aaaa3333aaaaULL, },
89         { 0x3333555533335555ULL, 0x3333555533335555ULL, },
90         { 0x3333cccc3333ccccULL, 0x3333cccc3333ccccULL, },
91         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
92         { 0x33338e383333e38eULL, 0x3333e38e333338e3ULL, },
93         { 0x333371c733331c71ULL, 0x33331c713333c71cULL, },
94         { 0x8e38ffffe38effffULL, 0xe38effff38e3ffffULL, },    /*  48  */
95         { 0x8e380000e38e0000ULL, 0xe38e000038e30000ULL, },
96         { 0x8e38aaaae38eaaaaULL, 0xe38eaaaa38e3aaaaULL, },
97         { 0x8e385555e38e5555ULL, 0xe38e555538e35555ULL, },
98         { 0x8e38cccce38eccccULL, 0xe38ecccc38e3ccccULL, },
99         { 0x8e383333e38e3333ULL, 0xe38e333338e33333ULL, },
100         { 0x8e388e38e38ee38eULL, 0xe38ee38e38e338e3ULL, },
101         { 0x8e3871c7e38e1c71ULL, 0xe38e1c7138e3c71cULL, },
102         { 0x71c7ffff1c71ffffULL, 0x1c71ffffc71cffffULL, },    /*  56  */
103         { 0x71c700001c710000ULL, 0x1c710000c71c0000ULL, },
104         { 0x71c7aaaa1c71aaaaULL, 0x1c71aaaac71caaaaULL, },
105         { 0x71c755551c715555ULL, 0x1c715555c71c5555ULL, },
106         { 0x71c7cccc1c71ccccULL, 0x1c71ccccc71cccccULL, },
107         { 0x71c733331c713333ULL, 0x1c713333c71c3333ULL, },
108         { 0x71c78e381c71e38eULL, 0x1c71e38ec71c38e3ULL, },
109         { 0x71c771c71c711c71ULL, 0x1c711c71c71cc71cULL, },
110         { 0x2862286255405540ULL, 0x886a886ae6cce6ccULL, },    /*  64  */
111         { 0x28624d935540c708ULL, 0x886afbbee6cc0063ULL, },
112         { 0x2862b9cf55408b80ULL, 0x886aac5ae6ccaeaaULL, },
113         { 0x28625e315540e24eULL, 0x886a704fe6cc164dULL, },
114         { 0x4d932862c7085540ULL, 0xfbbe886a0063e6ccULL, },
115         { 0x4d934d93c708c708ULL, 0xfbbefbbe00630063ULL, },
116         { 0x4d93b9cfc7088b80ULL, 0xfbbeac5a0063aeaaULL, },
117         { 0x4d935e31c708e24eULL, 0xfbbe704f0063164dULL, },
118         { 0xb9cf28628b805540ULL, 0xac5a886aaeaae6ccULL, },    /*  72  */
119         { 0xb9cf4d938b80c708ULL, 0xac5afbbeaeaa0063ULL, },
120         { 0xb9cfb9cf8b808b80ULL, 0xac5aac5aaeaaaeaaULL, },
121         { 0xb9cf5e318b80e24eULL, 0xac5a704faeaa164dULL, },
122         { 0x5e312862e24e5540ULL, 0x704f886a164de6ccULL, },
123         { 0x5e314d93e24ec708ULL, 0x704ffbbe164d0063ULL, },
124         { 0x5e31b9cfe24e8b80ULL, 0x704fac5a164daeaaULL, },
125         { 0x5e315e31e24ee24eULL, 0x704f704f164d164dULL, },
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_H(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_H(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