1 /*
2  *  Test program for MSA instruction PCNT.W
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 (PATTERN_INPUTS_COUNT + RANDOM_INPUTS_COUNT)
30 
31 
32 int32_t main(void)
33 {
34     char *isa_ase_name = "MSA";
35     char *group_name = "Bit Count";
36     char *instruction_name =  "PCNT.W";
37     int32_t ret;
38     uint32_t i;
39     struct timeval start, end;
40     double elapsed_time;
41 
42     uint64_t b128_result[TEST_COUNT_TOTAL][2];
43     uint64_t b128_expect[TEST_COUNT_TOTAL][2] = {
44         { 0x0000002000000020ULL, 0x0000002000000020ULL, },    /*   0  */
45         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
46         { 0x0000001000000010ULL, 0x0000001000000010ULL, },
47         { 0x0000001000000010ULL, 0x0000001000000010ULL, },
48         { 0x0000001000000010ULL, 0x0000001000000010ULL, },
49         { 0x0000001000000010ULL, 0x0000001000000010ULL, },
50         { 0x0000001100000010ULL, 0x0000000f00000011ULL, },
51         { 0x0000000f00000010ULL, 0x000000110000000fULL, },
52         { 0x0000001000000010ULL, 0x0000001000000010ULL, },    /*   8  */
53         { 0x0000001000000010ULL, 0x0000001000000010ULL, },
54         { 0x0000001100000011ULL, 0x000000100000000fULL, },
55         { 0x0000000f0000000fULL, 0x0000001000000011ULL, },
56         { 0x0000001200000010ULL, 0x0000000e00000012ULL, },
57         { 0x0000000e00000010ULL, 0x000000120000000eULL, },
58         { 0x0000001200000011ULL, 0x0000000e00000010ULL, },
59         { 0x0000000e0000000fULL, 0x0000001200000010ULL, },
60         { 0x0000001000000010ULL, 0x0000001000000010ULL, },    /*  16  */
61         { 0x0000001000000010ULL, 0x0000001000000010ULL, },
62         { 0x0000001200000012ULL, 0x0000000f0000000eULL, },
63         { 0x0000000e0000000eULL, 0x0000001100000012ULL, },
64         { 0x000000140000000eULL, 0x0000001000000012ULL, },
65         { 0x0000000c00000012ULL, 0x000000100000000eULL, },
66         { 0x000000150000000cULL, 0x000000130000000eULL, },
67         { 0x0000000b00000014ULL, 0x0000000d00000012ULL, },
68         { 0x0000001400000010ULL, 0x0000000c00000014ULL, },    /*  24  */
69         { 0x0000000c00000010ULL, 0x000000140000000cULL, },
70         { 0x0000001300000013ULL, 0x0000000e0000000dULL, },
71         { 0x0000000d0000000dULL, 0x0000001200000013ULL, },
72         { 0x0000001200000012ULL, 0x0000001200000010ULL, },
73         { 0x0000000e0000000eULL, 0x0000000e00000010ULL, },
74         { 0x0000001100000011ULL, 0x0000001100000011ULL, },
75         { 0x0000000f0000000fULL, 0x0000000f0000000fULL, },
76         { 0x0000001000000010ULL, 0x0000001000000010ULL, },    /*  32  */
77         { 0x0000001000000010ULL, 0x0000001000000010ULL, },
78         { 0x0000001100000011ULL, 0x0000001100000011ULL, },
79         { 0x0000000f0000000fULL, 0x0000000f0000000fULL, },
80         { 0x0000001200000012ULL, 0x0000001200000012ULL, },
81         { 0x0000000e0000000eULL, 0x0000000e0000000eULL, },
82         { 0x0000001300000013ULL, 0x000000130000000eULL, },
83         { 0x0000000d0000000dULL, 0x0000000d00000012ULL, },
84         { 0x0000001400000014ULL, 0x000000100000000cULL, },    /*  40  */
85         { 0x0000000c0000000cULL, 0x0000001000000014ULL, },
86         { 0x0000001500000015ULL, 0x0000000c0000000bULL, },
87         { 0x0000000b0000000bULL, 0x0000001400000015ULL, },
88         { 0x0000001600000014ULL, 0x0000000a0000000eULL, },
89         { 0x0000000a0000000cULL, 0x0000001600000012ULL, },
90         { 0x0000001700000012ULL, 0x0000000900000013ULL, },
91         { 0x000000090000000eULL, 0x000000170000000dULL, },
92         { 0x0000001800000010ULL, 0x0000000800000018ULL, },    /*  48  */
93         { 0x0000000800000010ULL, 0x0000001800000008ULL, },
94         { 0x000000190000000eULL, 0x0000000b00000019ULL, },
95         { 0x0000000700000012ULL, 0x0000001500000007ULL, },
96         { 0x0000001a0000000cULL, 0x0000000e00000018ULL, },
97         { 0x0000000600000014ULL, 0x0000001200000008ULL, },
98         { 0x0000001b0000000aULL, 0x0000001100000014ULL, },
99         { 0x0000000500000016ULL, 0x0000000f0000000cULL, },
100         { 0x0000001c00000008ULL, 0x0000001400000010ULL, },    /*  56  */
101         { 0x0000000400000018ULL, 0x0000000c00000010ULL, },
102         { 0x0000001d00000006ULL, 0x000000170000000cULL, },
103         { 0x000000030000001aULL, 0x0000000900000014ULL, },
104         { 0x0000001e00000004ULL, 0x0000001a00000008ULL, },
105         { 0x000000020000001cULL, 0x0000000600000018ULL, },
106         { 0x0000001f00000002ULL, 0x0000001d00000004ULL, },
107         { 0x000000010000001eULL, 0x000000030000001cULL, },
108         { 0x0000000f0000000aULL, 0x0000001100000012ULL, },    /*  64  */
109         { 0x000000110000000eULL, 0x0000001200000012ULL, },
110         { 0x0000001100000010ULL, 0x000000140000000eULL, },
111         { 0x0000000f00000010ULL, 0x0000000f0000000cULL, },
112         { 0x0000001200000011ULL, 0x0000000f00000013ULL, },
113         { 0x0000001000000014ULL, 0x0000000b00000011ULL, },
114         { 0x0000000e00000010ULL, 0x0000000900000013ULL, },
115         { 0x0000001200000010ULL, 0x000000140000000eULL, },
116         { 0x0000000c00000011ULL, 0x000000130000000fULL, },    /*  72  */
117         { 0x0000001100000014ULL, 0x0000000e00000010ULL, },
118         { 0x0000001300000011ULL, 0x0000000e0000000fULL, },
119         { 0x000000110000000eULL, 0x0000001100000013ULL, },
120         { 0x0000000f0000000eULL, 0x0000000e00000010ULL, },
121         { 0x000000110000000cULL, 0x0000001100000010ULL, },
122         { 0x0000000b00000013ULL, 0x000000140000000eULL, },
123         { 0x0000001000000014ULL, 0x0000001200000010ULL, },
124     };
125 
126     reset_msa_registers();
127 
128     gettimeofday(&start, NULL);
129 
130     for (i = 0; i < TEST_COUNT_TOTAL; i++) {
131         if (i < PATTERN_INPUTS_COUNT) {
132             do_msa_PCNT_W(b128_pattern[i], b128_result[i]);
133         } else {
134             do_msa_PCNT_W(b128_random[i - PATTERN_INPUTS_COUNT],
135                           b128_result[i]);
136         }
137     }
138 
139     gettimeofday(&end, NULL);
140 
141     elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0;
142     elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0;
143 
144     ret = check_results_128(isa_ase_name, group_name, instruction_name,
145                             TEST_COUNT_TOTAL, elapsed_time,
146                             &b128_result[0][0], &b128_expect[0][0]);
147 
148     return ret;
149 }
150