1 /*
2  *  Test program for MSA instruction ILVOD.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 =  "ILVOD.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         { 0xffffe38effff8e38ULL, 0xffff38e3ffffe38eULL, },
53         { 0xffff1c71ffff71c7ULL, 0xffffc71cffff1c71ULL, },
54         { 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL, },    /*   8  */
55         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
56         { 0x0000aaaa0000aaaaULL, 0x0000aaaa0000aaaaULL, },
57         { 0x0000555500005555ULL, 0x0000555500005555ULL, },
58         { 0x0000cccc0000ccccULL, 0x0000cccc0000ccccULL, },
59         { 0x0000333300003333ULL, 0x0000333300003333ULL, },
60         { 0x0000e38e00008e38ULL, 0x000038e30000e38eULL, },
61         { 0x00001c71000071c7ULL, 0x0000c71c00001c71ULL, },
62         { 0xaaaaffffaaaaffffULL, 0xaaaaffffaaaaffffULL, },    /*  16  */
63         { 0xaaaa0000aaaa0000ULL, 0xaaaa0000aaaa0000ULL, },
64         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
65         { 0xaaaa5555aaaa5555ULL, 0xaaaa5555aaaa5555ULL, },
66         { 0xaaaaccccaaaaccccULL, 0xaaaaccccaaaaccccULL, },
67         { 0xaaaa3333aaaa3333ULL, 0xaaaa3333aaaa3333ULL, },
68         { 0xaaaae38eaaaa8e38ULL, 0xaaaa38e3aaaae38eULL, },
69         { 0xaaaa1c71aaaa71c7ULL, 0xaaaac71caaaa1c71ULL, },
70         { 0x5555ffff5555ffffULL, 0x5555ffff5555ffffULL, },    /*  24  */
71         { 0x5555000055550000ULL, 0x5555000055550000ULL, },
72         { 0x5555aaaa5555aaaaULL, 0x5555aaaa5555aaaaULL, },
73         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
74         { 0x5555cccc5555ccccULL, 0x5555cccc5555ccccULL, },
75         { 0x5555333355553333ULL, 0x5555333355553333ULL, },
76         { 0x5555e38e55558e38ULL, 0x555538e35555e38eULL, },
77         { 0x55551c71555571c7ULL, 0x5555c71c55551c71ULL, },
78         { 0xccccffffccccffffULL, 0xccccffffccccffffULL, },    /*  32  */
79         { 0xcccc0000cccc0000ULL, 0xcccc0000cccc0000ULL, },
80         { 0xccccaaaaccccaaaaULL, 0xccccaaaaccccaaaaULL, },
81         { 0xcccc5555cccc5555ULL, 0xcccc5555cccc5555ULL, },
82         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
83         { 0xcccc3333cccc3333ULL, 0xcccc3333cccc3333ULL, },
84         { 0xcccce38ecccc8e38ULL, 0xcccc38e3cccce38eULL, },
85         { 0xcccc1c71cccc71c7ULL, 0xccccc71ccccc1c71ULL, },
86         { 0x3333ffff3333ffffULL, 0x3333ffff3333ffffULL, },    /*  40  */
87         { 0x3333000033330000ULL, 0x3333000033330000ULL, },
88         { 0x3333aaaa3333aaaaULL, 0x3333aaaa3333aaaaULL, },
89         { 0x3333555533335555ULL, 0x3333555533335555ULL, },
90         { 0x3333cccc3333ccccULL, 0x3333cccc3333ccccULL, },
91         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
92         { 0x3333e38e33338e38ULL, 0x333338e33333e38eULL, },
93         { 0x33331c71333371c7ULL, 0x3333c71c33331c71ULL, },
94         { 0xe38effff8e38ffffULL, 0x38e3ffffe38effffULL, },    /*  48  */
95         { 0xe38e00008e380000ULL, 0x38e30000e38e0000ULL, },
96         { 0xe38eaaaa8e38aaaaULL, 0x38e3aaaae38eaaaaULL, },
97         { 0xe38e55558e385555ULL, 0x38e35555e38e5555ULL, },
98         { 0xe38ecccc8e38ccccULL, 0x38e3cccce38eccccULL, },
99         { 0xe38e33338e383333ULL, 0x38e33333e38e3333ULL, },
100         { 0xe38ee38e8e388e38ULL, 0x38e338e3e38ee38eULL, },
101         { 0xe38e1c718e3871c7ULL, 0x38e3c71ce38e1c71ULL, },
102         { 0x1c71ffff71c7ffffULL, 0xc71cffff1c71ffffULL, },    /*  56  */
103         { 0x1c71000071c70000ULL, 0xc71c00001c710000ULL, },
104         { 0x1c71aaaa71c7aaaaULL, 0xc71caaaa1c71aaaaULL, },
105         { 0x1c71555571c75555ULL, 0xc71c55551c715555ULL, },
106         { 0x1c71cccc71c7ccccULL, 0xc71ccccc1c71ccccULL, },
107         { 0x1c71333371c73333ULL, 0xc71c33331c713333ULL, },
108         { 0x1c71e38e71c78e38ULL, 0xc71c38e31c71e38eULL, },
109         { 0x1c711c7171c771c7ULL, 0xc71cc71c1c711c71ULL, },
110         { 0x886a886a28622862ULL, 0x4b674b67fe7bfe7bULL, },    /*  64  */
111         { 0x886afbbe28624d93ULL, 0x4b6712f7fe7b153fULL, },
112         { 0x886aac5a2862b9cfULL, 0x4b6727d8fe7bab2bULL, },
113         { 0x886a704f28625e31ULL, 0x4b678df1fe7ba942ULL, },
114         { 0xfbbe886a4d932862ULL, 0x12f74b67153ffe7bULL, },
115         { 0xfbbefbbe4d934d93ULL, 0x12f712f7153f153fULL, },
116         { 0xfbbeac5a4d93b9cfULL, 0x12f727d8153fab2bULL, },
117         { 0xfbbe704f4d935e31ULL, 0x12f78df1153fa942ULL, },
118         { 0xac5a886ab9cf2862ULL, 0x27d84b67ab2bfe7bULL, },    /*  72  */
119         { 0xac5afbbeb9cf4d93ULL, 0x27d812f7ab2b153fULL, },
120         { 0xac5aac5ab9cfb9cfULL, 0x27d827d8ab2bab2bULL, },
121         { 0xac5a704fb9cf5e31ULL, 0x27d88df1ab2ba942ULL, },
122         { 0x704f886a5e312862ULL, 0x8df14b67a942fe7bULL, },
123         { 0x704ffbbe5e314d93ULL, 0x8df112f7a942153fULL, },
124         { 0x704fac5a5e31b9cfULL, 0x8df127d8a942ab2bULL, },
125         { 0x704f704f5e315e31ULL, 0x8df18df1a942a942ULL, },
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_ILVOD_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_ILVOD_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