1 /*
2  *  Test program for MSA instruction DPSUB_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 =  "DPSUB_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         { 0xfffffffffffffffeULL, 0xfffffffffffffffeULL, },    /*   0  */
47         { 0xfffffffffffffffeULL, 0xfffffffffffffffeULL, },
48         { 0xffffffff55555552ULL, 0xffffffff55555552ULL, },
49         { 0xfffffffffffffffcULL, 0xfffffffffffffffcULL, },
50         { 0xffffffff99999994ULL, 0xffffffff99999994ULL, },
51         { 0xfffffffffffffffaULL, 0xfffffffffffffffaULL, },
52         { 0xffffffff71c71c6bULL, 0x000000001c71c715ULL, },
53         { 0xfffffffffffffff8ULL, 0xfffffffffffffff8ULL, },
54         { 0xfffffffffffffff8ULL, 0xfffffffffffffff8ULL, },    /*   8  */
55         { 0xfffffffffffffff8ULL, 0xfffffffffffffff8ULL, },
56         { 0xfffffffffffffff8ULL, 0xfffffffffffffff8ULL, },
57         { 0xfffffffffffffff8ULL, 0xfffffffffffffff8ULL, },
58         { 0xfffffffffffffff8ULL, 0xfffffffffffffff8ULL, },
59         { 0xfffffffffffffff8ULL, 0xfffffffffffffff8ULL, },
60         { 0xfffffffffffffff8ULL, 0xfffffffffffffff8ULL, },
61         { 0xfffffffffffffff8ULL, 0xfffffffffffffff8ULL, },
62         { 0xffffffff5555554cULL, 0xffffffff5555554cULL, },    /*  16  */
63         { 0xffffffff5555554cULL, 0xffffffff5555554cULL, },
64         { 0xc71c71c58e38e384ULL, 0xc71c71c58e38e384ULL, },
65         { 0xfffffffeaaaaaaa0ULL, 0xfffffffeaaaaaaa0ULL, },
66         { 0xdddddddbbbbbbbb0ULL, 0xdddddddbbbbbbbb0ULL, },
67         { 0xfffffffdfffffff4ULL, 0xfffffffdfffffff4ULL, },
68         { 0xd097b4234bda12eaULL, 0x097b425c684bda06ULL, },
69         { 0xfffffffd55555548ULL, 0xfffffffd55555548ULL, },
70         { 0xfffffffdfffffff2ULL, 0xfffffffdfffffff2ULL, },    /*  24  */
71         { 0xfffffffdfffffff2ULL, 0xfffffffdfffffff2ULL, },
72         { 0x38e38e371c71c70eULL, 0x38e38e371c71c70eULL, },
73         { 0xfffffffeaaaaaa9cULL, 0xfffffffeaaaaaa9cULL, },
74         { 0x2222222133333324ULL, 0x2222222133333324ULL, },
75         { 0xffffffff55555546ULL, 0xffffffff55555546ULL, },
76         { 0x2f684bd97b425ec1ULL, 0xf684bda1097b424fULL, },
77         { 0xfffffffffffffff0ULL, 0xfffffffffffffff0ULL, },
78         { 0xffffffff99999988ULL, 0xffffffff99999988ULL, },    /*  32  */
79         { 0xffffffff99999988ULL, 0xffffffff99999988ULL, },
80         { 0xdddddddcaaaaaa98ULL, 0xdddddddcaaaaaa98ULL, },
81         { 0xffffffff33333320ULL, 0xffffffff33333320ULL, },
82         { 0xeb851eb6e147ae00ULL, 0xeb851eb6e147ae00ULL, },
83         { 0xfffffffeccccccb8ULL, 0xfffffffeccccccb8ULL, },
84         { 0xe38e38e1c16c16acULL, 0x05b05b0449f49f34ULL, },
85         { 0xfffffffe66666650ULL, 0xfffffffe66666650ULL, },
86         { 0xfffffffeccccccb6ULL, 0xfffffffeccccccb6ULL, },    /*  40  */
87         { 0xfffffffeccccccb6ULL, 0xfffffffeccccccb6ULL, },
88         { 0x22222221111110faULL, 0x22222221111110faULL, },
89         { 0xffffffff3333331cULL, 0xffffffff3333331cULL, },
90         { 0x147ae1471eb851d4ULL, 0x147ae1471eb851d4ULL, },
91         { 0xffffffff99999982ULL, 0xffffffff99999982ULL, },
92         { 0x1c71c71c16c16bffULL, 0xfa4fa4fa38e38e21ULL, },
93         { 0xffffffffffffffe8ULL, 0xffffffffffffffe8ULL, },
94         { 0xffffffff71c71c59ULL, 0x000000001c71c703ULL, },    /*  48  */
95         { 0xffffffff71c71c59ULL, 0x000000001c71c703ULL, },
96         { 0xd097b424bda12f4fULL, 0x097b425e84bda115ULL, },
97         { 0xfffffffee38e38caULL, 0x0000000038e38e1eULL, },
98         { 0xe38e38e1d82d82beULL, 0x05b05b05b60b609aULL, },
99         { 0xfffffffe5555553bULL, 0x0000000055555539ULL, },
100         { 0xca4587e4ba78192eULL, 0xf0329162948b0fb0ULL, },
101         { 0xfffffffdc71c71acULL, 0x0000000071c71c54ULL, },
102         { 0xfffffffe55555539ULL, 0x0000000055555537ULL, },    /*  56  */
103         { 0xfffffffe55555539ULL, 0x0000000055555537ULL, },
104         { 0x2f684bd85ed09797ULL, 0xf684bda1425ed079ULL, },
105         { 0xfffffffee38e38c6ULL, 0x0000000038e38e1aULL, },
106         { 0x1c71c71b8888886aULL, 0xfa4fa4fa55555536ULL, },
107         { 0xffffffff71c71c53ULL, 0x000000001c71c6fdULL, },
108         { 0x35ba78187e6b74d1ULL, 0x0fcd6e9df9add3a1ULL, },
109         { 0xffffffffffffffe0ULL, 0xffffffffffffffe0ULL, },
110         { 0xc1c52b51ed993d50ULL, 0xe9c828da514248ccULL, },    /*  64  */
111         { 0xb38b1f29f5c2926cULL, 0xe4522d2260b25370ULL, },
112         { 0x978b1706bdfa46f4ULL, 0xd814f3be50d3cfdeULL, },
113         { 0xbd2549a81e90da18ULL, 0xf92987d1ec89a90eULL, },
114         { 0xaeeb3d8026ba2f34ULL, 0xf3b38c19fbf9b3b2ULL, },
115         { 0x9756e17673d898abULL, 0xf08852c874204cfeULL, },
116         { 0xab37d321be2e30edULL, 0xf49ef75a0c71ea68ULL, },
117         { 0x908aa2c1222edcb6ULL, 0x0445531d0abde6f8ULL, },
118         { 0x748a9a9dea66913eULL, 0xf80819b8fadf6366ULL, },    /*  72  */
119         { 0x886b8c4934bc2980ULL, 0xfc1ebe4a933100d0ULL, },
120         { 0x59d865e79904609cULL, 0xd9ce9972461ac53fULL, },
121         { 0x985e08e42661ba7aULL, 0xced13609df919197ULL, },
122         { 0xbdf83b8586f84d9eULL, 0xefe5ca1d7b476ac7ULL, },
123         { 0xa34b0b24eaf8f967ULL, 0xff8c25e079936757ULL, },
124         { 0xe1d0ae2178565345ULL, 0xf48ec278130a33afULL, },
125         { 0x8de2b645a0f53058ULL, 0xa45a44165015196fULL, },
126         { 0x6792d4f3d7eea55cULL, 0xbfd22ee1a25aa627ULL, },    /*  80  */
127         { 0x75702d5b9af89c83ULL, 0xcc593d1da09f7be9ULL, },
128         { 0x801c3e1c97724195ULL, 0xb4c868d4067dd2d2ULL, },
129         { 0xdeafd0d6f0bea5c3ULL, 0x957877eb733b98b2ULL, },
130         { 0xd1883629f50ec77bULL, 0xb587d85cf1ffef10ULL, },
131         { 0xd4133b37d7cbfcc8ULL, 0xbc35d373b6f24df8ULL, },
132         { 0xbab344ed957a4c42ULL, 0xae8dcb499ce6cd0bULL, },
133         { 0x004c193eb947b2ddULL, 0x68b0a9907b71a293ULL, },
134         { 0x0b979b74995fc935ULL, 0x4a9602f12aa080cfULL, },    /*  88  */
135         { 0x2ae2653846d12eb1ULL, 0x4185939a2d850f91ULL, },
136         { 0x4c5017cc0eed7401ULL, 0x466840b4575dc0d7ULL, },
137         { 0x255760c7e1e38957ULL, 0x8360b1037a4f3497ULL, },
138         { 0x3b88c1c3a41f6803ULL, 0xa8cf0d07b592cd69ULL, },
139         { 0x585dd51272f3e482ULL, 0xb5723c3756218857ULL, },
140         { 0x94c1c43b5f5b538eULL, 0xdd9794c5786cc9c2ULL, },
141         { 0xa0b80278cc3c6a8bULL, 0xf710a53506ea3e4aULL, },
142         { 0x7c607ecd0201d92bULL, 0xf9bcdab0e105825cULL, },    /*  96  */
143         { 0xb628bad7d2470e0fULL, 0xfb660e974362496cULL, },
144         { 0x9ae11df599c281fbULL, 0xfd2738784b8dbfeaULL, },
145         { 0x7bc5bf3b5e23aeffULL, 0xfe707ab5676dfce2ULL, },
146         { 0x614dabb2dc4e0a36ULL, 0xf5f8795b76d8fd08ULL, },
147         { 0x6dbd1a209fc658b0ULL, 0xecd982bc128c8ceaULL, },
148         { 0x8cb93c5d61b1a8d0ULL, 0xecbaa1839f7e477aULL, },
149         { 0x6d33947e52d25a59ULL, 0xf62aab8428f0bf14ULL, },
150         { 0xa7970469e4259b2dULL, 0x0543881aad9efd08ULL, },    /* 104  */
151         { 0x8310e5e55f8149f3ULL, 0xe925758a04d06282ULL, },
152         { 0x746e208dd13c0f61ULL, 0xee4c7bccbccd15e4ULL, },
153         { 0x8da69743b598403fULL, 0xdac93db8514253e0ULL, },
154         { 0xdb31a0aea0a5cde6ULL, 0xe5bd105b853454a0ULL, },
155         { 0x0e6cfc3a89e7bd7cULL, 0xb06ea3bad3a90bd8ULL, },
156         { 0x338cc47438edb042ULL, 0x7df572596f6dffe8ULL, },
157         { 0x07fce3091840a942ULL, 0xdbd5224936527bd0ULL, },
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_DPSUB_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_DPSUB_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_DPSUB_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_DPSUB_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