1 /*
2  *  Test program for MSA instruction HADD_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 Add";
40     char *instruction_name =  "HADD_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         { 0xfffffffefffffffeULL, 0xfffffffefffffffeULL, },    /*   0  */
49         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
50         { 0xffffaaa9ffffaaa9ULL, 0xffffaaa9ffffaaa9ULL, },
51         { 0x0000555400005554ULL, 0x0000555400005554ULL, },
52         { 0xffffcccbffffcccbULL, 0xffffcccbffffcccbULL, },
53         { 0x0000333200003332ULL, 0x0000333200003332ULL, },
54         { 0x000038e2ffffe38dULL, 0xffff8e37000038e2ULL, },
55         { 0xffffc71b00001c70ULL, 0x000071c6ffffc71bULL, },
56         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*   8  */
57         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
58         { 0xffffaaaaffffaaaaULL, 0xffffaaaaffffaaaaULL, },
59         { 0x0000555500005555ULL, 0x0000555500005555ULL, },
60         { 0xffffccccffffccccULL, 0xffffccccffffccccULL, },
61         { 0x0000333300003333ULL, 0x0000333300003333ULL, },
62         { 0x000038e3ffffe38eULL, 0xffff8e38000038e3ULL, },
63         { 0xffffc71c00001c71ULL, 0x000071c7ffffc71cULL, },
64         { 0xffffaaa9ffffaaa9ULL, 0xffffaaa9ffffaaa9ULL, },    /*  16  */
65         { 0xffffaaaaffffaaaaULL, 0xffffaaaaffffaaaaULL, },
66         { 0xffff5554ffff5554ULL, 0xffff5554ffff5554ULL, },
67         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
68         { 0xffff7776ffff7776ULL, 0xffff7776ffff7776ULL, },
69         { 0xffffddddffffddddULL, 0xffffddddffffddddULL, },
70         { 0xffffe38dffff8e38ULL, 0xffff38e2ffffe38dULL, },
71         { 0xffff71c6ffffc71bULL, 0x00001c71ffff71c6ULL, },
72         { 0x0000555400005554ULL, 0x0000555400005554ULL, },    /*  24  */
73         { 0x0000555500005555ULL, 0x0000555500005555ULL, },
74         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
75         { 0x0000aaaa0000aaaaULL, 0x0000aaaa0000aaaaULL, },
76         { 0x0000222100002221ULL, 0x0000222100002221ULL, },
77         { 0x0000888800008888ULL, 0x0000888800008888ULL, },
78         { 0x00008e38000038e3ULL, 0xffffe38d00008e38ULL, },
79         { 0x00001c71000071c6ULL, 0x0000c71c00001c71ULL, },
80         { 0xffffcccbffffcccbULL, 0xffffcccbffffcccbULL, },    /*  32  */
81         { 0xffffccccffffccccULL, 0xffffccccffffccccULL, },
82         { 0xffff7776ffff7776ULL, 0xffff7776ffff7776ULL, },
83         { 0x0000222100002221ULL, 0x0000222100002221ULL, },
84         { 0xffff9998ffff9998ULL, 0xffff9998ffff9998ULL, },
85         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
86         { 0x000005afffffb05aULL, 0xffff5b04000005afULL, },
87         { 0xffff93e8ffffe93dULL, 0x00003e93ffff93e8ULL, },
88         { 0x0000333200003332ULL, 0x0000333200003332ULL, },    /*  40  */
89         { 0x0000333300003333ULL, 0x0000333300003333ULL, },
90         { 0xffffddddffffddddULL, 0xffffddddffffddddULL, },
91         { 0x0000888800008888ULL, 0x0000888800008888ULL, },
92         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
93         { 0x0000666600006666ULL, 0x0000666600006666ULL, },
94         { 0x00006c16000016c1ULL, 0xffffc16b00006c16ULL, },
95         { 0xfffffa4f00004fa4ULL, 0x0000a4fafffffa4fULL, },
96         { 0xffffe38dffff8e37ULL, 0x000038e2ffffe38dULL, },    /*  48  */
97         { 0xffffe38effff8e38ULL, 0x000038e3ffffe38eULL, },
98         { 0xffff8e38ffff38e2ULL, 0xffffe38dffff8e38ULL, },
99         { 0x000038e3ffffe38dULL, 0x00008e38000038e3ULL, },
100         { 0xffffb05affff5b04ULL, 0x000005afffffb05aULL, },
101         { 0x000016c1ffffc16bULL, 0x00006c16000016c1ULL, },
102         { 0x00001c71ffff71c6ULL, 0xffffc71b00001c71ULL, },
103         { 0xffffaaaaffffaaa9ULL, 0x0000aaaaffffaaaaULL, },
104         { 0x00001c70000071c6ULL, 0xffffc71b00001c70ULL, },    /*  56  */
105         { 0x00001c71000071c7ULL, 0xffffc71c00001c71ULL, },
106         { 0xffffc71b00001c71ULL, 0xffff71c6ffffc71bULL, },
107         { 0x000071c60000c71cULL, 0x00001c71000071c6ULL, },
108         { 0xffffe93d00003e93ULL, 0xffff93e8ffffe93dULL, },
109         { 0x00004fa40000a4faULL, 0xfffffa4f00004fa4ULL, },
110         { 0x0000555400005555ULL, 0xffff555400005554ULL, },
111         { 0xffffe38d00008e38ULL, 0x000038e3ffffe38dULL, },
112         { 0xffff6f3600007da2ULL, 0x000056c5ffffae87ULL, },    /*  64  */
113         { 0xffff88cdffffef6aULL, 0x0000068100005177ULL, },
114         { 0xffff3714ffffb3e2ULL, 0x000012660000238fULL, },
115         { 0xffff9eb700000ab0ULL, 0xffffd43fffffe11bULL, },
116         { 0xffffe28a0000a2d3ULL, 0x00001e55ffffc54bULL, },
117         { 0xfffffc210000149bULL, 0xffffce110000683bULL, },
118         { 0xffffaa68ffffd913ULL, 0xffffd9f600003a53ULL, },
119         { 0x0000120b00002fe1ULL, 0xffff9bcffffff7dfULL, },
120         { 0xffff932600000f0fULL, 0x00003336ffff5b37ULL, },    /*  72  */
121         { 0xffffacbdffff80d7ULL, 0xffffe2f2fffffe27ULL, },
122         { 0xffff5b04ffff454fULL, 0xffffeed7ffffd03fULL, },
123         { 0xffffc2a7ffff9c1dULL, 0xffffb0b0ffff8dcbULL, },
124         { 0x0000571b0000b371ULL, 0xffff994fffff594eULL, },
125         { 0x000070b200002539ULL, 0xffff490bfffffc3eULL, },
126         { 0x00001ef9ffffe9b1ULL, 0xffff54f0ffffce56ULL, },
127         { 0x0000869c0000407fULL, 0xffff16c9ffff8be2ULL, },
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_HADD_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_HADD_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