1 /*
2  *  Test program for MSA instruction MAX_A.B
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 *isa_ase_name = "MSA";
39     char *group_name = "Int Max Min";
40     char *instruction_name =  "MAX_A.B";
41     int32_t ret;
42     uint32_t i, j;
43     struct timeval start, end;
44     double elapsed_time;
45 
46     uint64_t b128_result[TEST_COUNT_TOTAL][2];
47     uint64_t b128_expect[TEST_COUNT_TOTAL][2] = {
48         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*   0  */
49         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
50         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
51         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
52         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
53         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
54         { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, },
55         { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, },
56         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*   8  */
57         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
58         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
59         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
60         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
61         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
62         { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, },
63         { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, },
64         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },    /*  16  */
65         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
66         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
67         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
68         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
69         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
70         { 0xaa8eaaaa8eaaaa8eULL, 0xaaaa8eaaaa8eaaaaULL, },
71         { 0xaa71aaaa71aaaa71ULL, 0xaaaa71aaaa71aaaaULL, },
72         { 0x5555555555555555ULL, 0x5555555555555555ULL, },    /*  24  */
73         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
74         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
75         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
76         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
77         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
78         { 0x558e55558e55558eULL, 0x55558e55558e5555ULL, },
79         { 0x5571555571555571ULL, 0x5555715555715555ULL, },
80         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },    /*  32  */
81         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
82         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
83         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
84         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
85         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
86         { 0xcc8e38cc8e38cc8eULL, 0x38cc8e38cc8e38ccULL, },
87         { 0xcc71c7cc71c7cc71ULL, 0xc7cc71c7cc71c7ccULL, },
88         { 0x3333333333333333ULL, 0x3333333333333333ULL, },    /*  40  */
89         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
90         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
91         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
92         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
93         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
94         { 0x338e38338e38338eULL, 0x38338e38338e3833ULL, },
95         { 0x3371c73371c73371ULL, 0xc73371c73371c733ULL, },
96         { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, },    /*  48  */
97         { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, },
98         { 0xaa8eaaaa8eaaaa8eULL, 0xaaaa8eaaaa8eaaaaULL, },
99         { 0x558e55558e55558eULL, 0x55558e55558e5555ULL, },
100         { 0xcc8e38cc8e38cc8eULL, 0x38cc8e38cc8e38ccULL, },
101         { 0x338e38338e38338eULL, 0x38338e38338e3833ULL, },
102         { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, },
103         { 0xe38ec7e38ec7e38eULL, 0xc7e38ec7e38ec7e3ULL, },
104         { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, },    /*  56  */
105         { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, },
106         { 0xaa71aaaa71aaaa71ULL, 0xaaaa71aaaa71aaaaULL, },
107         { 0x5571555571555571ULL, 0x5555715555715555ULL, },
108         { 0xcc71c7cc71c7cc71ULL, 0xc7cc71c7cc71c7ccULL, },
109         { 0x3371c73371c73371ULL, 0xc73371c73371c733ULL, },
110         { 0xe38ec7e38ec7e38eULL, 0xc7e38ec7e38ec7e3ULL, },
111         { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, },
112         { 0x886ae6cc28625540ULL, 0x4b670b5efe7bb00cULL, },    /*  64  */
113         { 0x886ae6634d935540ULL, 0x4b67bb5e157b520cULL, },
114         { 0x886aaeaab9628b80ULL, 0x4b67c65eab7bb014ULL, },
115         { 0x886ae64d5e62554eULL, 0x8d67885ea97bb0a0ULL, },
116         { 0x886ae6634d935540ULL, 0x4b67bb5e157b520cULL, },
117         { 0xfbbe00634d93c708ULL, 0x12f7bb1a153f52fcULL, },
118         { 0xac5aae634d938b80ULL, 0x27d8bb1aab3f5214ULL, },
119         { 0x704f16635e93c74eULL, 0x8df188d8a94252a0ULL, },
120         { 0x886aaeaab9628b80ULL, 0x4b67c65eab7bb014ULL, },    /*  72  */
121         { 0xac5aae634d938b80ULL, 0x27d8bb1aab3f5214ULL, },
122         { 0xac5aaeaab9cf8b80ULL, 0x27d8c6ffab2b2514ULL, },
123         { 0x705aaeaa5e318b80ULL, 0x8dd888d8a94225a0ULL, },
124         { 0x886ae64d5e62554eULL, 0x8d67885ea97bb0a0ULL, },
125         { 0x704f16635e93c74eULL, 0x8df188d8a94252a0ULL, },
126         { 0x705aaeaa5ecf8b80ULL, 0x8dd888d8a94225a0ULL, },
127         { 0x704f164d5e31e24eULL, 0x8df188d8a942e2a0ULL, },
128     };
129 
130     reset_msa_registers();
131 
132     gettimeofday(&start, NULL);
133 
134     for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) {
135         for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) {
136             do_msa_MAX_A_B(b128_pattern[i], b128_pattern[j],
137                            b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]);
138         }
139     }
140 
141     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
142         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
143             do_msa_MAX_A_B(b128_random[i], b128_random[j],
144                            b128_result[((PATTERN_INPUTS_SHORT_COUNT) *
145                                         (PATTERN_INPUTS_SHORT_COUNT)) +
146                                        RANDOM_INPUTS_SHORT_COUNT * i + j]);
147         }
148     }
149 
150     gettimeofday(&end, NULL);
151 
152     elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0;
153     elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0;
154 
155     ret = check_results_128(isa_ase_name, group_name, instruction_name,
156                             TEST_COUNT_TOTAL, elapsed_time,
157                             &b128_result[0][0], &b128_expect[0][0]);
158 
159     return ret;
160 }
161