1*a8f91dd9SMateja Marjanovic /*
2*a8f91dd9SMateja Marjanovic  *  Test program for MSA instruction DOTP_S.W
3*a8f91dd9SMateja Marjanovic  *
4*a8f91dd9SMateja Marjanovic  *  Copyright (C) 2018  Wave Computing, Inc.
5*a8f91dd9SMateja Marjanovic  *  Copyright (C) 2018  Mateja Marjanovic <mateja.marjanovic@rt-rk.com>
6*a8f91dd9SMateja Marjanovic  *
7*a8f91dd9SMateja Marjanovic  *  This program is free software: you can redistribute it and/or modify
8*a8f91dd9SMateja Marjanovic  *  it under the terms of the GNU General Public License as published by
9*a8f91dd9SMateja Marjanovic  *  the Free Software Foundation, either version 2 of the License, or
10*a8f91dd9SMateja Marjanovic  *  (at your option) any later version.
11*a8f91dd9SMateja Marjanovic  *
12*a8f91dd9SMateja Marjanovic  *  This program is distributed in the hope that it will be useful,
13*a8f91dd9SMateja Marjanovic  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
14*a8f91dd9SMateja Marjanovic  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15*a8f91dd9SMateja Marjanovic  *  GNU General Public License for more details.
16*a8f91dd9SMateja Marjanovic  *
17*a8f91dd9SMateja Marjanovic  *  You should have received a copy of the GNU General Public License
18*a8f91dd9SMateja Marjanovic  *  along with this program.  If not, see <https://www.gnu.org/licenses/>.
19*a8f91dd9SMateja Marjanovic  *
20*a8f91dd9SMateja Marjanovic  */
21*a8f91dd9SMateja Marjanovic 
22*a8f91dd9SMateja Marjanovic #include <sys/time.h>
23*a8f91dd9SMateja Marjanovic #include <stdint.h>
24*a8f91dd9SMateja Marjanovic 
25*a8f91dd9SMateja Marjanovic #include "../../../../include/wrappers_msa.h"
26*a8f91dd9SMateja Marjanovic #include "../../../../include/test_inputs.h"
27*a8f91dd9SMateja Marjanovic #include "../../../../include/test_utils.h"
28*a8f91dd9SMateja Marjanovic 
29*a8f91dd9SMateja Marjanovic #define TEST_COUNT_TOTAL (                                                \
30*a8f91dd9SMateja Marjanovic             (PATTERN_INPUTS_SHORT_COUNT) * (PATTERN_INPUTS_SHORT_COUNT) + \
31*a8f91dd9SMateja Marjanovic             (RANDOM_INPUTS_SHORT_COUNT) * (RANDOM_INPUTS_SHORT_COUNT))
32*a8f91dd9SMateja Marjanovic 
33*a8f91dd9SMateja Marjanovic 
34*a8f91dd9SMateja Marjanovic int32_t main(void)
35*a8f91dd9SMateja Marjanovic {
36*a8f91dd9SMateja Marjanovic     char *instruction_name = "DOTP_S.W";
37*a8f91dd9SMateja Marjanovic     int32_t ret;
38*a8f91dd9SMateja Marjanovic     uint32_t i, j;
39*a8f91dd9SMateja Marjanovic     struct timeval start, end;
40*a8f91dd9SMateja Marjanovic     double elapsed_time;
41*a8f91dd9SMateja Marjanovic 
42*a8f91dd9SMateja Marjanovic     uint64_t b128_result[TEST_COUNT_TOTAL][2];
43*a8f91dd9SMateja Marjanovic     uint64_t b128_expect[TEST_COUNT_TOTAL][2] = {
44*a8f91dd9SMateja Marjanovic         { 0x0000000200000002ULL, 0x0000000200000002ULL, },    /*   0  */
45*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
46*a8f91dd9SMateja Marjanovic         { 0x0000aaac0000aaacULL, 0x0000aaac0000aaacULL, },
47*a8f91dd9SMateja Marjanovic         { 0xffff5556ffff5556ULL, 0xffff5556ffff5556ULL, },
48*a8f91dd9SMateja Marjanovic         { 0x0000666800006668ULL, 0x0000666800006668ULL, },
49*a8f91dd9SMateja Marjanovic         { 0xffff999affff999aULL, 0xffff999affff999aULL, },
50*a8f91dd9SMateja Marjanovic         { 0xffffe38f00008e3aULL, 0x000038e5ffffe38fULL, },
51*a8f91dd9SMateja Marjanovic         { 0x00001c73ffff71c8ULL, 0xffffc71d00001c73ULL, },
52*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },    /*   8  */
53*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
54*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
55*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
56*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
57*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
58*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
59*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
60*a8f91dd9SMateja Marjanovic         { 0x0000aaac0000aaacULL, 0x0000aaac0000aaacULL, },    /*  16  */
61*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
62*a8f91dd9SMateja Marjanovic         { 0x38e471c838e471c8ULL, 0x38e471c838e471c8ULL, },
63*a8f91dd9SMateja Marjanovic         { 0xc71c38e4c71c38e4ULL, 0xc71c38e4c71c38e4ULL, },
64*a8f91dd9SMateja Marjanovic         { 0x2222eef02222eef0ULL, 0x2222eef02222eef0ULL, },
65*a8f91dd9SMateja Marjanovic         { 0xddddbbbcddddbbbcULL, 0xddddbbbcddddbbbcULL, },
66*a8f91dd9SMateja Marjanovic         { 0xf684ed0a2f69097cULL, 0x12f725eef684ed0aULL, },
67*a8f91dd9SMateja Marjanovic         { 0x097bbda2d097a130ULL, 0xed0984be097bbda2ULL, },
68*a8f91dd9SMateja Marjanovic         { 0xffff5556ffff5556ULL, 0xffff5556ffff5556ULL, },    /*  24  */
69*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
70*a8f91dd9SMateja Marjanovic         { 0xc71c38e4c71c38e4ULL, 0xc71c38e4c71c38e4ULL, },
71*a8f91dd9SMateja Marjanovic         { 0x38e31c7238e31c72ULL, 0x38e31c7238e31c72ULL, },
72*a8f91dd9SMateja Marjanovic         { 0xdddd7778dddd7778ULL, 0xdddd7778dddd7778ULL, },
73*a8f91dd9SMateja Marjanovic         { 0x2221ddde2221dddeULL, 0x2221ddde2221dddeULL, },
74*a8f91dd9SMateja Marjanovic         { 0x097af685d09784beULL, 0xed0912f7097af685ULL, },
75*a8f91dd9SMateja Marjanovic         { 0xf6845ed12f67d098ULL, 0x12f6425ff6845ed1ULL, },
76*a8f91dd9SMateja Marjanovic         { 0x0000666800006668ULL, 0x0000666800006668ULL, },    /*  32  */
77*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
78*a8f91dd9SMateja Marjanovic         { 0x2222eef02222eef0ULL, 0x2222eef02222eef0ULL, },
79*a8f91dd9SMateja Marjanovic         { 0xdddd7778dddd7778ULL, 0xdddd7778dddd7778ULL, },
80*a8f91dd9SMateja Marjanovic         { 0x147b8520147b8520ULL, 0x147b8520147b8520ULL, },
81*a8f91dd9SMateja Marjanovic         { 0xeb84e148eb84e148ULL, 0xeb84e148eb84e148ULL, },
82*a8f91dd9SMateja Marjanovic         { 0xfa4fb60c1c7271c8ULL, 0x0b612d84fa4fb60cULL, },
83*a8f91dd9SMateja Marjanovic         { 0x05b0b05ce38df4a0ULL, 0xf49f38e405b0b05cULL, },
84*a8f91dd9SMateja Marjanovic         { 0xffff999affff999aULL, 0xffff999affff999aULL, },    /*  40  */
85*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
86*a8f91dd9SMateja Marjanovic         { 0xddddbbbcddddbbbcULL, 0xddddbbbcddddbbbcULL, },
87*a8f91dd9SMateja Marjanovic         { 0x2221ddde2221dddeULL, 0x2221ddde2221dddeULL, },
88*a8f91dd9SMateja Marjanovic         { 0xeb84e148eb84e148ULL, 0xeb84e148eb84e148ULL, },
89*a8f91dd9SMateja Marjanovic         { 0x147ab852147ab852ULL, 0x147ab852147ab852ULL, },
90*a8f91dd9SMateja Marjanovic         { 0x05b02d83e38e1c72ULL, 0xf49f0b6105b02d83ULL, },
91*a8f91dd9SMateja Marjanovic         { 0xfa4f6c171c717d28ULL, 0x0b608e39fa4f6c17ULL, },
92*a8f91dd9SMateja Marjanovic         { 0xffffe38f00008e3aULL, 0x000038e5ffffe38fULL, },    /*  48  */
93*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
94*a8f91dd9SMateja Marjanovic         { 0xf684ed0a2f69097cULL, 0x12f725eef684ed0aULL, },
95*a8f91dd9SMateja Marjanovic         { 0x097af685d09784beULL, 0xed0912f7097af685ULL, },
96*a8f91dd9SMateja Marjanovic         { 0xfa4fb60c1c7271c8ULL, 0x0b612d84fa4fb60cULL, },
97*a8f91dd9SMateja Marjanovic         { 0x05b02d83e38e1c72ULL, 0xf49f0b6105b02d83ULL, },
98*a8f91dd9SMateja Marjanovic         { 0x0fcd3c0d35bb4f04ULL, 0x3f3645890fcd3c0dULL, },
99*a8f91dd9SMateja Marjanovic         { 0xf032a782ca453f36ULL, 0xc0c9f35cf032a782ULL, },
100*a8f91dd9SMateja Marjanovic         { 0x00001c73ffff71c8ULL, 0xffffc71d00001c73ULL, },    /*  56  */
101*a8f91dd9SMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
102*a8f91dd9SMateja Marjanovic         { 0x097bbda2d097a130ULL, 0xed0984be097bbda2ULL, },
103*a8f91dd9SMateja Marjanovic         { 0xf6845ed12f67d098ULL, 0x12f6425ff6845ed1ULL, },
104*a8f91dd9SMateja Marjanovic         { 0x05b0b05ce38df4a0ULL, 0xf49f38e405b0b05cULL, },
105*a8f91dd9SMateja Marjanovic         { 0xfa4f6c171c717d28ULL, 0x0b608e39fa4f6c17ULL, },
106*a8f91dd9SMateja Marjanovic         { 0xf032a782ca453f36ULL, 0xc0c9f35cf032a782ULL, },
107*a8f91dd9SMateja Marjanovic         { 0x0fcd74f135ba3292ULL, 0x3f35d3c10fcd74f1ULL, },
108*a8f91dd9SMateja Marjanovic         { 0x3a57fe7422c25584ULL, 0x16b6b9f518facfa9ULL, },    /*  64  */
109*a8f91dd9SMateja Marjanovic         { 0x01f36d90f9441446ULL, 0x0286cfede5f4db15ULL, },
110*a8f91dd9SMateja Marjanovic         { 0x2f1518bcce21d93eULL, 0x0934568af4ec6499ULL, },
111*a8f91dd9SMateja Marjanovic         { 0xc9576c1204f83042ULL, 0xd91d3e4709b06e36ULL, },
112*a8f91dd9SMateja Marjanovic         { 0x01f36d90f9441446ULL, 0x0286cfede5f4db15ULL, },
113*a8f91dd9SMateja Marjanovic         { 0x0012474d242f32a9ULL, 0x13f2a8f51ca9cd91ULL, },
114*a8f91dd9SMateja Marjanovic         { 0x0144b48a04a7d0ddULL, 0x124b1c4e04fa8e45ULL, },
115*a8f91dd9SMateja Marjanovic         { 0xfe2a6f6923268793ULL, 0x179e9377ef4766beULL, },
116*a8f91dd9SMateja Marjanovic         { 0x2f1518bcce21d93eULL, 0x0934568af4ec6499ULL, },    /*  72  */
117*a8f91dd9SMateja Marjanovic         { 0x0144b48a04a7d0ddULL, 0x124b1c4e04fa8e45ULL, },
118*a8f91dd9SMateja Marjanovic         { 0x352c988848431561ULL, 0x12e4f841217b42c9ULL, },
119*a8f91dd9SMateja Marjanovic         { 0xd437b4e8f3b0139fULL, 0x08c7d980187d5896ULL, },
120*a8f91dd9SMateja Marjanovic         { 0xc9576c1204f83042ULL, 0xd91d3e4709b06e36ULL, },
121*a8f91dd9SMateja Marjanovic         { 0xfe2a6f6923268793ULL, 0x179e9377ef4766beULL, },
122*a8f91dd9SMateja Marjanovic };
123*a8f91dd9SMateja Marjanovic 
124*a8f91dd9SMateja Marjanovic     gettimeofday(&start, NULL);
125*a8f91dd9SMateja Marjanovic 
126*a8f91dd9SMateja Marjanovic     for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) {
127*a8f91dd9SMateja Marjanovic         for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) {
128*a8f91dd9SMateja Marjanovic             do_msa_DOTP_S_W(b128_pattern[i], b128_pattern[j],
129*a8f91dd9SMateja Marjanovic                            b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]);
130*a8f91dd9SMateja Marjanovic         }
131*a8f91dd9SMateja Marjanovic     }
132*a8f91dd9SMateja Marjanovic 
133*a8f91dd9SMateja Marjanovic     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
134*a8f91dd9SMateja Marjanovic         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
135*a8f91dd9SMateja Marjanovic             do_msa_DOTP_S_W(b128_random[i], b128_random[j],
136*a8f91dd9SMateja Marjanovic                            b128_result[((PATTERN_INPUTS_SHORT_COUNT) *
137*a8f91dd9SMateja Marjanovic                                         (PATTERN_INPUTS_SHORT_COUNT)) +
138*a8f91dd9SMateja Marjanovic                                        RANDOM_INPUTS_SHORT_COUNT * i + j]);
139*a8f91dd9SMateja Marjanovic         }
140*a8f91dd9SMateja Marjanovic     }
141*a8f91dd9SMateja Marjanovic 
142*a8f91dd9SMateja Marjanovic     gettimeofday(&end, NULL);
143*a8f91dd9SMateja Marjanovic 
144*a8f91dd9SMateja Marjanovic     elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0;
145*a8f91dd9SMateja Marjanovic     elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0;
146*a8f91dd9SMateja Marjanovic 
147*a8f91dd9SMateja Marjanovic     ret = check_results(instruction_name, TEST_COUNT_TOTAL, elapsed_time,
148*a8f91dd9SMateja Marjanovic                         &b128_result[0][0], &b128_expect[0][0]);
149*a8f91dd9SMateja Marjanovic 
150*a8f91dd9SMateja Marjanovic     return ret;
151*a8f91dd9SMateja Marjanovic }
152