1 /*
2  *  Test program for MSA instruction VSHF.B
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 
main(void)34 int32_t main(void)
35 {
36     char *isa_ase_name = "MSA";
37     char *group_name = "Pack";
38     char *instruction_name =  "VSHF.B";
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         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*   0  */
47         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
48         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
49         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
50         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
51         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
52         { 0x8e8e8e8e8e8e8e8eULL, 0x8e8e8e8e8e8e8e8eULL, },
53         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
54         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*   8  */
55         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
56         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
57         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
58         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
59         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
60         { 0x8e8e8e8e8e8e8e8eULL, 0x8e8e8e8e8e8e8e8eULL, },
61         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
62         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*  16  */
63         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
64         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
65         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
66         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
67         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
68         { 0x8e8e8e8e8e8e8e8eULL, 0x8e8e8e8e8e8e8e8eULL, },
69         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
70         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*  24  */
71         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
72         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
73         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
74         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
75         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
76         { 0x8e8e8e8e8e8e8e8eULL, 0x8e8e8e8e8e8e8e8eULL, },
77         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
78         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*  32  */
79         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
80         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
81         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
82         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
83         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
84         { 0x8e8e8e8e8e8e8e8eULL, 0x8e8e8e8e8e8e8e8eULL, },
85         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
86         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*  40  */
87         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
88         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
89         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
90         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
91         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
92         { 0x8e8e8e8e8e8e8e8eULL, 0x8e8e8e8e8e8e8e8eULL, },
93         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
94         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*  48  */
95         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
96         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
97         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
98         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
99         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
100         { 0x8e8e8e8e8e8e8e8eULL, 0x8e8e8e8e8e8e8e8eULL, },
101         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
102         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*  56  */
103         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
104         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
105         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
106         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
107         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
108         { 0x8e8e8e8e8e8e8e8eULL, 0x8e8e8e8e8e8e8e8eULL, },
109         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
110         { 0x4040404040404040ULL, 0x4040404040404040ULL, },    /*  64  */
111         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
112         { 0x8080808080808080ULL, 0x8080808080808080ULL, },
113         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
114         { 0x4040404040404040ULL, 0x4040404040404040ULL, },
115         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
116         { 0x8080808080808080ULL, 0x8080808080808080ULL, },
117         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
118         { 0x4040404040404040ULL, 0x4040404040404040ULL, },    /*  72  */
119         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
120         { 0x8080808080808080ULL, 0x8080808080808080ULL, },
121         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
122         { 0x4040404040404040ULL, 0x4040404040404040ULL, },
123         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
124         { 0x8080808080808080ULL, 0x8080808080808080ULL, },
125         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
126         { 0x4040404040404040ULL, 0x4040404040404040ULL, },    /*  80  */
127         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
128         { 0x8080808080808080ULL, 0x8080808080808080ULL, },
129         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
130         { 0x4040404040404040ULL, 0x4040404040404040ULL, },
131         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
132         { 0x8080808080808080ULL, 0x8080808080808080ULL, },
133         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
134         { 0x4040404040404040ULL, 0x4040404040404040ULL, },    /*  88  */
135         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
136         { 0x8080808080808080ULL, 0x8080808080808080ULL, },
137         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
138         { 0x4040404040404040ULL, 0x4040404040404040ULL, },
139         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
140         { 0x8080808080808080ULL, 0x8080808080808080ULL, },
141         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
142         { 0x0000000000000000ULL, 0x0000000000000000ULL, },    /*  96  */
143         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
144         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
145         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
146         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
147         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
148         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
149         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
150         { 0x0000000000000000ULL, 0x0000000000000000ULL, },    /* 104  */
151         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
152         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
153         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
154         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
155         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
156         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
157         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
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_VSHF_B(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_VSHF_B(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_VSHF_B__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_VSHF_B__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