1 /*
2  *  Test program for MSA instruction HADD_U.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_U.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         { 0x0001fffe0001fffeULL, 0x0001fffe0001fffeULL, },    /*   0  */
49         { 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL, },
50         { 0x0001aaa90001aaa9ULL, 0x0001aaa90001aaa9ULL, },
51         { 0x0001555400015554ULL, 0x0001555400015554ULL, },
52         { 0x0001cccb0001cccbULL, 0x0001cccb0001cccbULL, },
53         { 0x0001333200013332ULL, 0x0001333200013332ULL, },
54         { 0x000138e20001e38dULL, 0x00018e37000138e2ULL, },
55         { 0x0001c71b00011c70ULL, 0x000171c60001c71bULL, },
56         { 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL, },    /*   8  */
57         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
58         { 0x0000aaaa0000aaaaULL, 0x0000aaaa0000aaaaULL, },
59         { 0x0000555500005555ULL, 0x0000555500005555ULL, },
60         { 0x0000cccc0000ccccULL, 0x0000cccc0000ccccULL, },
61         { 0x0000333300003333ULL, 0x0000333300003333ULL, },
62         { 0x000038e30000e38eULL, 0x00008e38000038e3ULL, },
63         { 0x0000c71c00001c71ULL, 0x000071c70000c71cULL, },
64         { 0x0001aaa90001aaa9ULL, 0x0001aaa90001aaa9ULL, },    /*  16  */
65         { 0x0000aaaa0000aaaaULL, 0x0000aaaa0000aaaaULL, },
66         { 0x0001555400015554ULL, 0x0001555400015554ULL, },
67         { 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL, },
68         { 0x0001777600017776ULL, 0x0001777600017776ULL, },
69         { 0x0000dddd0000ddddULL, 0x0000dddd0000ddddULL, },
70         { 0x0000e38d00018e38ULL, 0x000138e20000e38dULL, },
71         { 0x000171c60000c71bULL, 0x00011c71000171c6ULL, },
72         { 0x0001555400015554ULL, 0x0001555400015554ULL, },    /*  24  */
73         { 0x0000555500005555ULL, 0x0000555500005555ULL, },
74         { 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL, },
75         { 0x0000aaaa0000aaaaULL, 0x0000aaaa0000aaaaULL, },
76         { 0x0001222100012221ULL, 0x0001222100012221ULL, },
77         { 0x0000888800008888ULL, 0x0000888800008888ULL, },
78         { 0x00008e38000138e3ULL, 0x0000e38d00008e38ULL, },
79         { 0x00011c71000071c6ULL, 0x0000c71c00011c71ULL, },
80         { 0x0001cccb0001cccbULL, 0x0001cccb0001cccbULL, },    /*  32  */
81         { 0x0000cccc0000ccccULL, 0x0000cccc0000ccccULL, },
82         { 0x0001777600017776ULL, 0x0001777600017776ULL, },
83         { 0x0001222100012221ULL, 0x0001222100012221ULL, },
84         { 0x0001999800019998ULL, 0x0001999800019998ULL, },
85         { 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL, },
86         { 0x000105af0001b05aULL, 0x00015b04000105afULL, },
87         { 0x000193e80000e93dULL, 0x00013e93000193e8ULL, },
88         { 0x0001333200013332ULL, 0x0001333200013332ULL, },    /*  40  */
89         { 0x0000333300003333ULL, 0x0000333300003333ULL, },
90         { 0x0000dddd0000ddddULL, 0x0000dddd0000ddddULL, },
91         { 0x0000888800008888ULL, 0x0000888800008888ULL, },
92         { 0x0000ffff0000ffffULL, 0x0000ffff0000ffffULL, },
93         { 0x0000666600006666ULL, 0x0000666600006666ULL, },
94         { 0x00006c16000116c1ULL, 0x0000c16b00006c16ULL, },
95         { 0x0000fa4f00004fa4ULL, 0x0000a4fa0000fa4fULL, },
96         { 0x0001e38d00018e37ULL, 0x000138e20001e38dULL, },    /*  48  */
97         { 0x0000e38e00008e38ULL, 0x000038e30000e38eULL, },
98         { 0x00018e38000138e2ULL, 0x0000e38d00018e38ULL, },
99         { 0x000138e30000e38dULL, 0x00008e38000138e3ULL, },
100         { 0x0001b05a00015b04ULL, 0x000105af0001b05aULL, },
101         { 0x000116c10000c16bULL, 0x00006c16000116c1ULL, },
102         { 0x00011c71000171c6ULL, 0x0000c71b00011c71ULL, },
103         { 0x0001aaaa0000aaa9ULL, 0x0000aaaa0001aaaaULL, },
104         { 0x00011c70000171c6ULL, 0x0001c71b00011c70ULL, },    /*  56  */
105         { 0x00001c71000071c7ULL, 0x0000c71c00001c71ULL, },
106         { 0x0000c71b00011c71ULL, 0x000171c60000c71bULL, },
107         { 0x000071c60000c71cULL, 0x00011c71000071c6ULL, },
108         { 0x0000e93d00013e93ULL, 0x000193e80000e93dULL, },
109         { 0x00004fa40000a4faULL, 0x0000fa4f00004fa4ULL, },
110         { 0x0000555400015555ULL, 0x0001555400005554ULL, },
111         { 0x0000e38d00008e38ULL, 0x000138e30000e38dULL, },
112         { 0x00016f3600007da2ULL, 0x000056c50001ae87ULL, },    /*  64  */
113         { 0x000088cd0000ef6aULL, 0x0001068100015177ULL, },
114         { 0x000137140000b3e2ULL, 0x000112660001238fULL, },
115         { 0x00009eb700010ab0ULL, 0x0000d43f0001e11bULL, },
116         { 0x0001e28a0000a2d3ULL, 0x00001e550000c54bULL, },
117         { 0x0000fc210001149bULL, 0x0000ce110000683bULL, },
118         { 0x0001aa680000d913ULL, 0x0000d9f600003a53ULL, },
119         { 0x0001120b00012fe1ULL, 0x00009bcf0000f7dfULL, },
120         { 0x0001932600010f0fULL, 0x0000333600015b37ULL, },    /*  72  */
121         { 0x0000acbd000180d7ULL, 0x0000e2f20000fe27ULL, },
122         { 0x00015b040001454fULL, 0x0000eed70000d03fULL, },
123         { 0x0000c2a700019c1dULL, 0x0000b0b000018dcbULL, },
124         { 0x0001571b0000b371ULL, 0x0000994f0001594eULL, },
125         { 0x000070b200012539ULL, 0x0001490b0000fc3eULL, },
126         { 0x00011ef90000e9b1ULL, 0x000154f00000ce56ULL, },
127         { 0x0000869c0001407fULL, 0x000116c900018be2ULL, },
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_U_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_U_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