1 /*
2  *  Test program for MSA instruction DOTP_S.W
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.W";
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         { 0x0000000200000002ULL, 0x0000000200000002ULL, },    /*   0  */
49         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
50         { 0x0000aaac0000aaacULL, 0x0000aaac0000aaacULL, },
51         { 0xffff5556ffff5556ULL, 0xffff5556ffff5556ULL, },
52         { 0x0000666800006668ULL, 0x0000666800006668ULL, },
53         { 0xffff999affff999aULL, 0xffff999affff999aULL, },
54         { 0xffffe38f00008e3aULL, 0x000038e5ffffe38fULL, },
55         { 0x00001c73ffff71c8ULL, 0xffffc71d00001c73ULL, },
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         { 0x0000aaac0000aaacULL, 0x0000aaac0000aaacULL, },    /*  16  */
65         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
66         { 0x38e471c838e471c8ULL, 0x38e471c838e471c8ULL, },
67         { 0xc71c38e4c71c38e4ULL, 0xc71c38e4c71c38e4ULL, },
68         { 0x2222eef02222eef0ULL, 0x2222eef02222eef0ULL, },
69         { 0xddddbbbcddddbbbcULL, 0xddddbbbcddddbbbcULL, },
70         { 0xf684ed0a2f69097cULL, 0x12f725eef684ed0aULL, },
71         { 0x097bbda2d097a130ULL, 0xed0984be097bbda2ULL, },
72         { 0xffff5556ffff5556ULL, 0xffff5556ffff5556ULL, },    /*  24  */
73         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
74         { 0xc71c38e4c71c38e4ULL, 0xc71c38e4c71c38e4ULL, },
75         { 0x38e31c7238e31c72ULL, 0x38e31c7238e31c72ULL, },
76         { 0xdddd7778dddd7778ULL, 0xdddd7778dddd7778ULL, },
77         { 0x2221ddde2221dddeULL, 0x2221ddde2221dddeULL, },
78         { 0x097af685d09784beULL, 0xed0912f7097af685ULL, },
79         { 0xf6845ed12f67d098ULL, 0x12f6425ff6845ed1ULL, },
80         { 0x0000666800006668ULL, 0x0000666800006668ULL, },    /*  32  */
81         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
82         { 0x2222eef02222eef0ULL, 0x2222eef02222eef0ULL, },
83         { 0xdddd7778dddd7778ULL, 0xdddd7778dddd7778ULL, },
84         { 0x147b8520147b8520ULL, 0x147b8520147b8520ULL, },
85         { 0xeb84e148eb84e148ULL, 0xeb84e148eb84e148ULL, },
86         { 0xfa4fb60c1c7271c8ULL, 0x0b612d84fa4fb60cULL, },
87         { 0x05b0b05ce38df4a0ULL, 0xf49f38e405b0b05cULL, },
88         { 0xffff999affff999aULL, 0xffff999affff999aULL, },    /*  40  */
89         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
90         { 0xddddbbbcddddbbbcULL, 0xddddbbbcddddbbbcULL, },
91         { 0x2221ddde2221dddeULL, 0x2221ddde2221dddeULL, },
92         { 0xeb84e148eb84e148ULL, 0xeb84e148eb84e148ULL, },
93         { 0x147ab852147ab852ULL, 0x147ab852147ab852ULL, },
94         { 0x05b02d83e38e1c72ULL, 0xf49f0b6105b02d83ULL, },
95         { 0xfa4f6c171c717d28ULL, 0x0b608e39fa4f6c17ULL, },
96         { 0xffffe38f00008e3aULL, 0x000038e5ffffe38fULL, },    /*  48  */
97         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
98         { 0xf684ed0a2f69097cULL, 0x12f725eef684ed0aULL, },
99         { 0x097af685d09784beULL, 0xed0912f7097af685ULL, },
100         { 0xfa4fb60c1c7271c8ULL, 0x0b612d84fa4fb60cULL, },
101         { 0x05b02d83e38e1c72ULL, 0xf49f0b6105b02d83ULL, },
102         { 0x0fcd3c0d35bb4f04ULL, 0x3f3645890fcd3c0dULL, },
103         { 0xf032a782ca453f36ULL, 0xc0c9f35cf032a782ULL, },
104         { 0x00001c73ffff71c8ULL, 0xffffc71d00001c73ULL, },    /*  56  */
105         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
106         { 0x097bbda2d097a130ULL, 0xed0984be097bbda2ULL, },
107         { 0xf6845ed12f67d098ULL, 0x12f6425ff6845ed1ULL, },
108         { 0x05b0b05ce38df4a0ULL, 0xf49f38e405b0b05cULL, },
109         { 0xfa4f6c171c717d28ULL, 0x0b608e39fa4f6c17ULL, },
110         { 0xf032a782ca453f36ULL, 0xc0c9f35cf032a782ULL, },
111         { 0x0fcd74f135ba3292ULL, 0x3f35d3c10fcd74f1ULL, },
112         { 0x3a57fe7422c25584ULL, 0x16b6b9f518facfa9ULL, },    /*  64  */
113         { 0x01f36d90f9441446ULL, 0x0286cfede5f4db15ULL, },
114         { 0x2f1518bcce21d93eULL, 0x0934568af4ec6499ULL, },
115         { 0xc9576c1204f83042ULL, 0xd91d3e4709b06e36ULL, },
116         { 0x01f36d90f9441446ULL, 0x0286cfede5f4db15ULL, },
117         { 0x0012474d242f32a9ULL, 0x13f2a8f51ca9cd91ULL, },
118         { 0x0144b48a04a7d0ddULL, 0x124b1c4e04fa8e45ULL, },
119         { 0xfe2a6f6923268793ULL, 0x179e9377ef4766beULL, },
120         { 0x2f1518bcce21d93eULL, 0x0934568af4ec6499ULL, },    /*  72  */
121         { 0x0144b48a04a7d0ddULL, 0x124b1c4e04fa8e45ULL, },
122         { 0x352c988848431561ULL, 0x12e4f841217b42c9ULL, },
123         { 0xd437b4e8f3b0139fULL, 0x08c7d980187d5896ULL, },
124         { 0xc9576c1204f83042ULL, 0xd91d3e4709b06e36ULL, },
125         { 0xfe2a6f6923268793ULL, 0x179e9377ef4766beULL, },
126         { 0xd437b4e8f3b0139fULL, 0x08c7d980187d5896ULL, },
127         { 0x33368b8a2619d525ULL, 0x6a47932120c31904ULL, },
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_W(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_W(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