1 /*
2  *  Test program for MSA instruction DPADD_S.D
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             3 * (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 = "Int Dot Product";
38     char *instruction_name =  "DPADD_S.D";
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          { 0x0000000000000002ULL, 0x0000000000000002ULL, },    /*   0  */
47         { 0x0000000000000002ULL, 0x0000000000000002ULL, },
48         { 0x00000000aaaaaaaeULL, 0x00000000aaaaaaaeULL, },
49         { 0x0000000000000004ULL, 0x0000000000000004ULL, },
50         { 0x000000006666666cULL, 0x000000006666666cULL, },
51         { 0x0000000000000006ULL, 0x0000000000000006ULL, },
52         { 0x000000008e38e395ULL, 0xffffffffe38e38ebULL, },
53         { 0x0000000000000008ULL, 0x0000000000000008ULL, },
54         { 0x0000000000000008ULL, 0x0000000000000008ULL, },    /*   8  */
55         { 0x0000000000000008ULL, 0x0000000000000008ULL, },
56         { 0x0000000000000008ULL, 0x0000000000000008ULL, },
57         { 0x0000000000000008ULL, 0x0000000000000008ULL, },
58         { 0x0000000000000008ULL, 0x0000000000000008ULL, },
59         { 0x0000000000000008ULL, 0x0000000000000008ULL, },
60         { 0x0000000000000008ULL, 0x0000000000000008ULL, },
61         { 0x0000000000000008ULL, 0x0000000000000008ULL, },
62         { 0x00000000aaaaaab4ULL, 0x00000000aaaaaab4ULL, },    /*  16  */
63         { 0x00000000aaaaaab4ULL, 0x00000000aaaaaab4ULL, },
64         { 0x38e38e3a71c71c7cULL, 0x38e38e3a71c71c7cULL, },
65         { 0x0000000155555560ULL, 0x0000000155555560ULL, },
66         { 0x2222222444444450ULL, 0x2222222444444450ULL, },
67         { 0x000000020000000cULL, 0x000000020000000cULL, },
68         { 0x2f684bdcb425ed16ULL, 0xf684bda397b425faULL, },
69         { 0x00000002aaaaaab8ULL, 0x00000002aaaaaab8ULL, },
70         { 0x000000020000000eULL, 0x000000020000000eULL, },    /*  24  */
71         { 0x000000020000000eULL, 0x000000020000000eULL, },
72         { 0xc71c71c8e38e38f2ULL, 0xc71c71c8e38e38f2ULL, },
73         { 0x0000000155555564ULL, 0x0000000155555564ULL, },
74         { 0xdddddddeccccccdcULL, 0xdddddddeccccccdcULL, },
75         { 0x00000000aaaaaabaULL, 0x00000000aaaaaabaULL, },
76         { 0xd097b42684bda13fULL, 0x097b425ef684bdb1ULL, },
77         { 0x0000000000000010ULL, 0x0000000000000010ULL, },
78         { 0x0000000066666678ULL, 0x0000000066666678ULL, },    /*  32  */
79         { 0x0000000066666678ULL, 0x0000000066666678ULL, },
80         { 0x2222222355555568ULL, 0x2222222355555568ULL, },
81         { 0x00000000cccccce0ULL, 0x00000000cccccce0ULL, },
82         { 0x147ae1491eb85200ULL, 0x147ae1491eb85200ULL, },
83         { 0x0000000133333348ULL, 0x0000000133333348ULL, },
84         { 0x1c71c71e3e93e954ULL, 0xfa4fa4fbb60b60ccULL, },
85         { 0x00000001999999b0ULL, 0x00000001999999b0ULL, },
86         { 0x000000013333334aULL, 0x000000013333334aULL, },    /*  40  */
87         { 0x000000013333334aULL, 0x000000013333334aULL, },
88         { 0xdddddddeeeeeef06ULL, 0xdddddddeeeeeef06ULL, },
89         { 0x00000000cccccce4ULL, 0x00000000cccccce4ULL, },
90         { 0xeb851eb8e147ae2cULL, 0xeb851eb8e147ae2cULL, },
91         { 0x000000006666667eULL, 0x000000006666667eULL, },
92         { 0xe38e38e3e93e9401ULL, 0x05b05b05c71c71dfULL, },
93         { 0x0000000000000018ULL, 0x0000000000000018ULL, },
94         { 0x000000008e38e3a7ULL, 0xffffffffe38e38fdULL, },    /*  48  */
95         { 0x000000008e38e3a7ULL, 0xffffffffe38e38fdULL, },
96         { 0x2f684bdb425ed0b1ULL, 0xf684bda17b425eebULL, },
97         { 0x000000011c71c736ULL, 0xffffffffc71c71e2ULL, },
98         { 0x1c71c71e27d27d42ULL, 0xfa4fa4fa49f49f66ULL, },
99         { 0x00000001aaaaaac5ULL, 0xffffffffaaaaaac7ULL, },
100         { 0x35ba781b4587e6d2ULL, 0x0fcd6e9d6b74f050ULL, },
101         { 0x0000000238e38e54ULL, 0xffffffff8e38e3acULL, },
102         { 0x00000001aaaaaac7ULL, 0xffffffffaaaaaac9ULL, },    /*  56  */
103         { 0x00000001aaaaaac7ULL, 0xffffffffaaaaaac9ULL, },
104         { 0xd097b427a12f6869ULL, 0x097b425ebda12f87ULL, },
105         { 0x000000011c71c73aULL, 0xffffffffc71c71e6ULL, },
106         { 0xe38e38e477777796ULL, 0x05b05b05aaaaaacaULL, },
107         { 0x000000008e38e3adULL, 0xffffffffe38e3903ULL, },
108         { 0xca4587e781948b2fULL, 0xf032916206522c5fULL, },
109         { 0x0000000000000020ULL, 0x0000000000000020ULL, },
110         { 0x3e3ad4ae1266c2b0ULL, 0x1637d725aebdb734ULL, },    /*  64  */
111         { 0x4c74e0d60a3d6d94ULL, 0x1badd2dd9f4dac90ULL, },
112         { 0x6874e8f94205b90cULL, 0x27eb0c41af2c3022ULL, },
113         { 0x42dab657e16f25e8ULL, 0x06d6782e137656f2ULL, },
114         { 0x5114c27fd945d0ccULL, 0x0c4c73e604064c4eULL, },
115         { 0x68a91e898c276755ULL, 0x0f77ad378bdfb302ULL, },
116         { 0x54c82cde41d1cf13ULL, 0x0b6108a5f38e1598ULL, },
117         { 0x6f755d3eddd1234aULL, 0xfbbaace2f5421908ULL, },
118         { 0x8b75656215996ec2ULL, 0x07f7e64705209c9aULL, },    /*  72  */
119         { 0x779473b6cb43d680ULL, 0x03e141b56cceff30ULL, },
120         { 0xa6279a1866fb9f64ULL, 0x2631668db9e53ac1ULL, },
121         { 0x67a1f71bd99e4586ULL, 0x312ec9f6206e6e69ULL, },
122         { 0x4207c47a7907b262ULL, 0x101a35e284b89539ULL, },
123         { 0x5cb4f4db15070699ULL, 0x0073da1f866c98a9ULL, },
124         { 0x1e2f51de87a9acbbULL, 0x0b713d87ecf5cc51ULL, },
125         { 0x721d49ba5f0acfa8ULL, 0x5ba5bbe9afeae691ULL, },
126         { 0x4bcd68690d995de0ULL, 0x771da6b4b6c967ebULL, },    /*  80  */
127         { 0x4ea9a2cfbb5acd7bULL, 0x79dd6a73439e6387ULL, },
128         { 0x47c800b999dd2371ULL, 0x766d25914ef7a7a0ULL, },
129         { 0x41b0fa10eb77cf84ULL, 0x26e85189458965f8ULL, },
130         { 0x1fc448ce062c2944ULL, 0x31f490a9422a80e6ULL, },
131         { 0x211bdfadfd79770eULL, 0x3b25f4cac5763378ULL, },
132         { 0x16fbb87edd87b6f0ULL, 0x57c0b65fabdda20eULL, },
133         { 0x14621091eac4a5f6ULL, 0x4d29a25d32fa9ef6ULL, },
134         { 0x07832ded1c464b02ULL, 0x6396905709e3cfa4ULL, },    /*  88  */
135         { 0x0ff4a84eab8df3b9ULL, 0x6bc9a7d8c6adf2eaULL, },
136         { 0x21e53326bfbd0b05ULL, 0x8f8f3b9c679dff5aULL, },
137         { 0x191ed6a24e1576f9ULL, 0x9e8c2e402760373aULL, },
138         { 0x19b438400fc27751ULL, 0x819c4bbfd3ee6972ULL, },
139         { 0x1e0d5dc1094ae999ULL, 0x7496a289f5eff010ULL, },
140         { 0x11af620b7bc03943ULL, 0x8a11f229836addc7ULL, },
141         { 0x46fa45d0e84440fcULL, 0xe8d2c0211fb042bfULL, },
142         { 0x22142516b5a8adbcULL, 0xe1cf1923e186aad1ULL, },    /*  96  */
143         { 0x066ebbbb4ff6da44ULL, 0xd918d7e6a7e61877ULL, },
144         { 0x100acc9d22839a48ULL, 0xce291932929e367fULL, },
145         { 0x0dfe419d62a62f64ULL, 0xc020fe45a8cf7acfULL, },
146         { 0x2ba79b6ffbf3c63bULL, 0xb428f52c49fce695ULL, },
147         { 0x29b3b85200bdf100ULL, 0xb4ae7ea2f52aa5b9ULL, },
148         { 0x293bb84d6360c0b6ULL, 0xae33b26e4c493c49ULL, },
149         { 0x46a99fdf54f4862dULL, 0xae790dc5055f6f51ULL, },
150         { 0x18480e0fd728c7c3ULL, 0xa000ad7b15f8ebe0ULL, },    /* 104  */
151         { 0x1b8b97aa205e1239ULL, 0x89c78b8909c4a8e5ULL, },
152         { 0x09abb26b05ef649dULL, 0x74242fa1bd49e740ULL, },
153         { 0x04e233bc861d272bULL, 0x9c5343ab30f62f9fULL, },
154         { 0xda2da0d0884dc3d1ULL, 0xb824f201640b4147ULL, },
155         { 0x9d8b22ee1b9a2e0fULL, 0xb642ddf1edb0747fULL, },
156         { 0x7c81956533686a37ULL, 0xdd5181781dc3ad37ULL, },
157         { 0xc60b1905717ff25aULL, 0xe2af726e71ad7ad7ULL, },
158     };
159 
160     reset_msa_registers();
161 
162     gettimeofday(&start, NULL);
163 
164     for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) {
165         for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) {
166             do_msa_DPADD_S_D(b128_pattern[i], b128_pattern[j],
167                              b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]);
168         }
169     }
170 
171     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
172         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
173             do_msa_DPADD_S_D(b128_random[i], b128_random[j],
174                              b128_result[((PATTERN_INPUTS_SHORT_COUNT) *
175                                           (PATTERN_INPUTS_SHORT_COUNT)) +
176                                          RANDOM_INPUTS_SHORT_COUNT * i + j]);
177         }
178     }
179 
180     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
181         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
182             do_msa_DPADD_S_D__DDT(b128_random[i], b128_random[j],
183                                   b128_result[
184                                       ((PATTERN_INPUTS_SHORT_COUNT) *
185                                        (PATTERN_INPUTS_SHORT_COUNT)) +
186                                       ((RANDOM_INPUTS_SHORT_COUNT) *
187                                        (RANDOM_INPUTS_SHORT_COUNT)) +
188                                       RANDOM_INPUTS_SHORT_COUNT * i + j]);
189         }
190     }
191 
192     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
193         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
194             do_msa_DPADD_S_D__DSD(b128_random[i], b128_random[j],
195                                   b128_result[
196                                       ((PATTERN_INPUTS_SHORT_COUNT) *
197                                        (PATTERN_INPUTS_SHORT_COUNT)) +
198                                       (2 * (RANDOM_INPUTS_SHORT_COUNT) *
199                                        (RANDOM_INPUTS_SHORT_COUNT)) +
200                                       RANDOM_INPUTS_SHORT_COUNT * i + j]);
201         }
202     }
203 
204     gettimeofday(&end, NULL);
205 
206     elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0;
207     elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0;
208 
209     ret = check_results_128(isa_ase_name, group_name, instruction_name,
210                             TEST_COUNT_TOTAL, elapsed_time,
211                             &b128_result[0][0], &b128_expect[0][0]);
212 
213     return ret;
214 }
215