1 /*
2  *  Test program for MSA instruction DOTP_S.H
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 Dot Product";
40     char *instruction_name =  "DOTP_S.H";
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         { 0x0002000200020002ULL, 0x0002000200020002ULL, },    /*   0  */
49         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
50         { 0x00ac00ac00ac00acULL, 0x00ac00ac00ac00acULL, },
51         { 0xff56ff56ff56ff56ULL, 0xff56ff56ff56ff56ULL, },
52         { 0x0068006800680068ULL, 0x0068006800680068ULL, },
53         { 0xff9aff9aff9aff9aULL, 0xff9aff9aff9aff9aULL, },
54         { 0x008fffe5003a008fULL, 0xffe5003a008fffe5ULL, },
55         { 0xff73001dffc8ff73ULL, 0x001dffc8ff73001dULL, },
56         { 0x0000000000000000ULL, 0x0000000000000000ULL, },    /*   8  */
57         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
58         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
59         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
60         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
61         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
62         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
63         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
64         { 0x00ac00ac00ac00acULL, 0x00ac00ac00ac00acULL, },    /*  16  */
65         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
66         { 0x39c839c839c839c8ULL, 0x39c839c839c839c8ULL, },
67         { 0xc6e4c6e4c6e4c6e4ULL, 0xc6e4c6e4c6e4c6e4ULL, },
68         { 0x22f022f022f022f0ULL, 0x22f022f022f022f0ULL, },
69         { 0xddbcddbcddbcddbcULL, 0xddbcddbcddbcddbcULL, },
70         { 0x300af6ee137c300aULL, 0xf6ee137c300af6eeULL, },
71         { 0xd0a209beed30d0a2ULL, 0x09beed30d0a209beULL, },
72         { 0xff56ff56ff56ff56ULL, 0xff56ff56ff56ff56ULL, },    /*  24  */
73         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
74         { 0xc6e4c6e4c6e4c6e4ULL, 0xc6e4c6e4c6e4c6e4ULL, },
75         { 0x3872387238723872ULL, 0x3872387238723872ULL, },
76         { 0xdd78dd78dd78dd78ULL, 0xdd78dd78dd78dd78ULL, },
77         { 0x21de21de21de21deULL, 0x21de21de21de21deULL, },
78         { 0xd08508f7ecbed085ULL, 0x08f7ecbed08508f7ULL, },
79         { 0x2ed1f65f12982ed1ULL, 0xf65f12982ed1f65fULL, },
80         { 0x0068006800680068ULL, 0x0068006800680068ULL, },    /*  32  */
81         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
82         { 0x22f022f022f022f0ULL, 0x22f022f022f022f0ULL, },
83         { 0xdd78dd78dd78dd78ULL, 0xdd78dd78dd78dd78ULL, },
84         { 0x1520152015201520ULL, 0x1520152015201520ULL, },
85         { 0xeb48eb48eb48eb48ULL, 0xeb48eb48eb48eb48ULL, },
86         { 0x1d0cfa840bc81d0cULL, 0xfa840bc81d0cfa84ULL, },
87         { 0xe35c05e4f4a0e35cULL, 0x05e4f4a0e35c05e4ULL, },
88         { 0xff9aff9aff9aff9aULL, 0xff9aff9aff9aff9aULL, },    /*  40  */
89         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
90         { 0xddbcddbcddbcddbcULL, 0xddbcddbcddbcddbcULL, },
91         { 0x21de21de21de21deULL, 0x21de21de21de21deULL, },
92         { 0xeb48eb48eb48eb48ULL, 0xeb48eb48eb48eb48ULL, },
93         { 0x1452145214521452ULL, 0x1452145214521452ULL, },
94         { 0xe3830561f472e383ULL, 0x0561f472e3830561ULL, },
95         { 0x1c17fa390b281c17ULL, 0xfa390b281c17fa39ULL, },
96         { 0x008fffe5003a008fULL, 0xffe5003a008fffe5ULL, },    /*  48  */
97         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
98         { 0x300af6ee137c300aULL, 0xf6ee137c300af6eeULL, },
99         { 0xd08508f7ecbed085ULL, 0x08f7ecbed08508f7ULL, },
100         { 0x1d0cfa840bc81d0cULL, 0xfa840bc81d0cfa84ULL, },
101         { 0xe3830561f472e383ULL, 0x0561f472e3830561ULL, },
102         { 0x360d0f893f04360dULL, 0x0f893f04360d0f89ULL, },
103         { 0xca82f05cc136ca82ULL, 0xf05cc136ca82f05cULL, },
104         { 0xff73001dffc8ff73ULL, 0x001dffc8ff73001dULL, },    /*  56  */
105         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
106         { 0xd0a209beed30d0a2ULL, 0x09beed30d0a209beULL, },
107         { 0x2ed1f65f12982ed1ULL, 0xf65f12982ed1f65fULL, },
108         { 0xe35c05e4f4a0e35cULL, 0x05e4f4a0e35c05e4ULL, },
109         { 0x1c17fa390b281c17ULL, 0xfa390b281c17fa39ULL, },
110         { 0xca82f05cc136ca82ULL, 0xf05cc136ca82f05cULL, },
111         { 0x34f10fc13e9234f1ULL, 0x0fc13e9234f10fc1ULL, },
112         { 0x64240d342bc42c39ULL, 0x3f6a22fd3b1d1990ULL, },    /*  64  */
113         { 0xe704ebe4e24eef13ULL, 0x01a706951e1be630ULL, },
114         { 0x4ca419cce226b927ULL, 0xfb55fd241553f560ULL, },
115         { 0xec36ee202172098aULL, 0xd846ec28206404e0ULL, },
116         { 0xe704ebe4e24eef13ULL, 0x01a706951e1be630ULL, },
117         { 0x111d264945920cf1ULL, 0x0195153d113a1a54ULL, },
118         { 0xea70debeff82160dULL, 0x04260f88039c0b8aULL, },
119         { 0xe9721dc70769091eULL, 0xf8711c48091bf7e4ULL, },
120         { 0x4ca419cce226b927ULL, 0xfb55fd241553f560ULL, },    /*  72  */
121         { 0xea70debeff82160dULL, 0x04260f88039c0b8aULL, },
122         { 0x3b3437281d127579ULL, 0x0c310d25237206e9ULL, },
123         { 0xf706df16dc8de6b6ULL, 0xf0d31b5827f9f42aULL, },
124         { 0xec36ee202172098aULL, 0xd846ec28206404e0ULL, },
125         { 0xe9721dc70769091eULL, 0xf8711c48091bf7e4ULL, },
126         { 0xf706df16dc8de6b6ULL, 0xf0d31b5827f9f42aULL, },
127         { 0x4961190d2be51b48ULL, 0x348a3e802e952784ULL, },
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_DOTP_S_H(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_DOTP_S_H(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