1*53e116feSMateja Marjanovic /*
2*53e116feSMateja Marjanovic  *  Test program for MSA instruction SUBSUS_U.B
3*53e116feSMateja Marjanovic  *
4*53e116feSMateja Marjanovic  *  Copyright (C) 2018  Wave Computing, Inc.
5*53e116feSMateja Marjanovic  *  Copyright (C) 2018  Mateja Marjanovic <mateja.marjanovic@rt-rk.com>
6*53e116feSMateja Marjanovic  *
7*53e116feSMateja Marjanovic  *  This program is free software: you can redistribute it and/or modify
8*53e116feSMateja Marjanovic  *  it under the terms of the GNU General Public License as published by
9*53e116feSMateja Marjanovic  *  the Free Software Foundation, either version 2 of the License, or
10*53e116feSMateja Marjanovic  *  (at your option) any later version.
11*53e116feSMateja Marjanovic  *
12*53e116feSMateja Marjanovic  *  This program is distributed in the hope that it will be useful,
13*53e116feSMateja Marjanovic  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
14*53e116feSMateja Marjanovic  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15*53e116feSMateja Marjanovic  *  GNU General Public License for more details.
16*53e116feSMateja Marjanovic  *
17*53e116feSMateja Marjanovic  *  You should have received a copy of the GNU General Public License
18*53e116feSMateja Marjanovic  *  along with this program.  If not, see <https://www.gnu.org/licenses/>.
19*53e116feSMateja Marjanovic  *
20*53e116feSMateja Marjanovic  */
21*53e116feSMateja Marjanovic 
22*53e116feSMateja Marjanovic #include <sys/time.h>
23*53e116feSMateja Marjanovic #include <stdint.h>
24*53e116feSMateja Marjanovic 
25*53e116feSMateja Marjanovic #include "../../../../include/wrappers_msa.h"
26*53e116feSMateja Marjanovic #include "../../../../include/test_inputs.h"
27*53e116feSMateja Marjanovic #include "../../../../include/test_utils.h"
28*53e116feSMateja Marjanovic 
29*53e116feSMateja Marjanovic #define TEST_COUNT_TOTAL (                                                \
30*53e116feSMateja Marjanovic             (PATTERN_INPUTS_SHORT_COUNT) * (PATTERN_INPUTS_SHORT_COUNT) + \
31*53e116feSMateja Marjanovic             (RANDOM_INPUTS_SHORT_COUNT) * (RANDOM_INPUTS_SHORT_COUNT))
32*53e116feSMateja Marjanovic 
33*53e116feSMateja Marjanovic 
34*53e116feSMateja Marjanovic int32_t main(void)
35*53e116feSMateja Marjanovic {
36*53e116feSMateja Marjanovic     char *instruction_name = "SUBSUS_U.B";
37*53e116feSMateja Marjanovic     int32_t ret;
38*53e116feSMateja Marjanovic     uint32_t i, j;
39*53e116feSMateja Marjanovic     struct timeval start, end;
40*53e116feSMateja Marjanovic     double elapsed_time;
41*53e116feSMateja Marjanovic 
42*53e116feSMateja Marjanovic     uint64_t b128_result[TEST_COUNT_TOTAL][2];
43*53e116feSMateja Marjanovic     uint64_t b128_expect[TEST_COUNT_TOTAL][2] = {
44*53e116feSMateja Marjanovic         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*   0  */
45*53e116feSMateja Marjanovic         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
46*53e116feSMateja Marjanovic         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
47*53e116feSMateja Marjanovic         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
48*53e116feSMateja Marjanovic         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
49*53e116feSMateja Marjanovic         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
50*53e116feSMateja Marjanovic         { 0xffffc7ffffc7ffffULL, 0xc7ffffc7ffffc7ffULL, },
51*53e116feSMateja Marjanovic         { 0xe38effe38effe38eULL, 0xffe38effe38effe3ULL, },
52*53e116feSMateja Marjanovic         { 0x0101010101010101ULL, 0x0101010101010101ULL, },    /*   8  */
53*53e116feSMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
54*53e116feSMateja Marjanovic         { 0x5656565656565656ULL, 0x5656565656565656ULL, },
55*53e116feSMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
56*53e116feSMateja Marjanovic         { 0x3434343434343434ULL, 0x3434343434343434ULL, },
57*53e116feSMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
58*53e116feSMateja Marjanovic         { 0x1d72001d72001d72ULL, 0x001d72001d72001dULL, },
59*53e116feSMateja Marjanovic         { 0x0000390000390000ULL, 0x3900003900003900ULL, },
60*53e116feSMateja Marjanovic         { 0xababababababababULL, 0xababababababababULL, },    /*  16  */
61*53e116feSMateja Marjanovic         { 0xaaaaaaaaaaaaaaaaULL, 0xaaaaaaaaaaaaaaaaULL, },
62*53e116feSMateja Marjanovic         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
63*53e116feSMateja Marjanovic         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
64*53e116feSMateja Marjanovic         { 0xdedededededededeULL, 0xdedededededededeULL, },
65*53e116feSMateja Marjanovic         { 0x7777777777777777ULL, 0x7777777777777777ULL, },
66*53e116feSMateja Marjanovic         { 0xc7ff72c7ff72c7ffULL, 0x72c7ff72c7ff72c7ULL, },
67*53e116feSMateja Marjanovic         { 0x8e39e38e39e38e39ULL, 0xe38e39e38e39e38eULL, },
68*53e116feSMateja Marjanovic         { 0x5656565656565656ULL, 0x5656565656565656ULL, },    /*  24  */
69*53e116feSMateja Marjanovic         { 0x5555555555555555ULL, 0x5555555555555555ULL, },
70*53e116feSMateja Marjanovic         { 0xababababababababULL, 0xababababababababULL, },
71*53e116feSMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
72*53e116feSMateja Marjanovic         { 0x8989898989898989ULL, 0x8989898989898989ULL, },
73*53e116feSMateja Marjanovic         { 0x2222222222222222ULL, 0x2222222222222222ULL, },
74*53e116feSMateja Marjanovic         { 0x72c71d72c71d72c7ULL, 0x1d72c71d72c71d72ULL, },
75*53e116feSMateja Marjanovic         { 0x39008e39008e3900ULL, 0x8e39008e39008e39ULL, },
76*53e116feSMateja Marjanovic         { 0xcdcdcdcdcdcdcdcdULL, 0xcdcdcdcdcdcdcdcdULL, },    /*  32  */
77*53e116feSMateja Marjanovic         { 0xccccccccccccccccULL, 0xccccccccccccccccULL, },
78*53e116feSMateja Marjanovic         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
79*53e116feSMateja Marjanovic         { 0x7777777777777777ULL, 0x7777777777777777ULL, },
80*53e116feSMateja Marjanovic         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
81*53e116feSMateja Marjanovic         { 0x9999999999999999ULL, 0x9999999999999999ULL, },
82*53e116feSMateja Marjanovic         { 0xe9ff94e9ff94e9ffULL, 0x94e9ff94e9ff94e9ULL, },
83*53e116feSMateja Marjanovic         { 0xb05bffb05bffb05bULL, 0xffb05bffb05bffb0ULL, },
84*53e116feSMateja Marjanovic         { 0x3434343434343434ULL, 0x3434343434343434ULL, },    /*  40  */
85*53e116feSMateja Marjanovic         { 0x3333333333333333ULL, 0x3333333333333333ULL, },
86*53e116feSMateja Marjanovic         { 0x8989898989898989ULL, 0x8989898989898989ULL, },
87*53e116feSMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
88*53e116feSMateja Marjanovic         { 0x6767676767676767ULL, 0x6767676767676767ULL, },
89*53e116feSMateja Marjanovic         { 0x0000000000000000ULL, 0x0000000000000000ULL, },
90*53e116feSMateja Marjanovic         { 0x50a50050a50050a5ULL, 0x0050a50050a50050ULL, },
91*53e116feSMateja Marjanovic         { 0x17006c17006c1700ULL, 0x6c17006c17006c17ULL, },
92*53e116feSMateja Marjanovic         { 0xe48f39e48f39e48fULL, 0x39e48f39e48f39e4ULL, },    /*  48  */
93*53e116feSMateja Marjanovic         { 0xe38e38e38e38e38eULL, 0x38e38e38e38e38e3ULL, },
94*53e116feSMateja Marjanovic         { 0xffe48effe48effe4ULL, 0x8effe48effe48effULL, },
95*53e116feSMateja Marjanovic         { 0x8e39008e39008e39ULL, 0x008e39008e39008eULL, },
96*53e116feSMateja Marjanovic         { 0xffc26cffc26cffc2ULL, 0x6cffc26cffc26cffULL, },
97*53e116feSMateja Marjanovic         { 0xb05b05b05b05b05bULL, 0x05b05b05b05b05b0ULL, },
98*53e116feSMateja Marjanovic         { 0xffff00ffff00ffffULL, 0x00ffff00ffff00ffULL, },
99*53e116feSMateja Marjanovic         { 0xc71d71c71d71c71dULL, 0x71c71d71c71d71c7ULL, },
100*53e116feSMateja Marjanovic         { 0x1d72c81d72c81d72ULL, 0xc81d72c81d72c81dULL, },    /*  56  */
101*53e116feSMateja Marjanovic         { 0x1c71c71c71c71c71ULL, 0xc71c71c71c71c71cULL, },
102*53e116feSMateja Marjanovic         { 0x72c7ff72c7ff72c7ULL, 0xff72c7ff72c7ff72ULL, },
103*53e116feSMateja Marjanovic         { 0x001c72001c72001cULL, 0x72001c72001c7200ULL, },
104*53e116feSMateja Marjanovic         { 0x50a5fb50a5fb50a5ULL, 0xfb50a5fb50a5fb50ULL, },
105*53e116feSMateja Marjanovic         { 0x003e94003e94003eULL, 0x94003e94003e9400ULL, },
106*53e116feSMateja Marjanovic         { 0x39e38f39e38f39e3ULL, 0x8f39e38f39e38f39ULL, },
107*53e116feSMateja Marjanovic         { 0x0000ff0000ff0000ULL, 0xff0000ff0000ff00ULL, },
108*53e116feSMateja Marjanovic         { 0xff00ffff00000000ULL, 0x00000000ff00ff00ULL, },    /*  64  */
109*53e116feSMateja Marjanovic         { 0x8dace66900cf8e38ULL, 0x39705044e93c5e10ULL, },
110*53e116feSMateja Marjanovic         { 0xdc10ffff6f93cac0ULL, 0x248f455fff508b00ULL, },
111*53e116feSMateja Marjanovic         { 0x181bd07f00317300ULL, 0xbe768386ff39ce6cULL, },
112*53e116feSMateja Marjanovic         { 0xff541a9725317200ULL, 0x0090b0001700a2f0ULL, },
113*53e116feSMateja Marjanovic         { 0xffff000000ffff00ULL, 0x00ffff00000000ffULL, },
114*53e116feSMateja Marjanovic         { 0xff6452b994c4ff88ULL, 0x00fff51b6a142de8ULL, },
115*53e116feSMateja Marjanovic         { 0x8b6f00160062e500ULL, 0x85ffff426c0070ffULL, },
116*53e116feSMateja Marjanovic         { 0xff00c8de916d3640ULL, 0x0071bba1ad007508ULL, },    /*  72  */
117*53e116feSMateja Marjanovic         { 0xb19cae476cffc478ULL, 0x15e1ffe596000018ULL, },
118*53e116feSMateja Marjanovic         { 0xff00ffffffffffffULL, 0x00ffffffff000000ULL, },
119*53e116feSMateja Marjanovic         { 0x3c0b985d5b9ea932ULL, 0x9ae7ffffff004374ULL, },
120*53e116feSMateja Marjanovic         { 0xe800308136008d0eULL, 0x428a7d7aab00ff94ULL, },
121*53e116feSMateja Marjanovic         { 0x75911600119eff46ULL, 0x7bfacdbe940390a4ULL, },
122*53e116feSMateja Marjanovic };
123*53e116feSMateja Marjanovic 
124*53e116feSMateja Marjanovic     gettimeofday(&start, NULL);
125*53e116feSMateja Marjanovic 
126*53e116feSMateja Marjanovic     for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) {
127*53e116feSMateja Marjanovic         for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) {
128*53e116feSMateja Marjanovic             do_msa_SUBSUS_U_B(b128_pattern[i], b128_pattern[j],
129*53e116feSMateja Marjanovic                            b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]);
130*53e116feSMateja Marjanovic         }
131*53e116feSMateja Marjanovic     }
132*53e116feSMateja Marjanovic 
133*53e116feSMateja Marjanovic     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
134*53e116feSMateja Marjanovic         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
135*53e116feSMateja Marjanovic             do_msa_SUBSUS_U_B(b128_random[i], b128_random[j],
136*53e116feSMateja Marjanovic                            b128_result[((PATTERN_INPUTS_SHORT_COUNT) *
137*53e116feSMateja Marjanovic                                         (PATTERN_INPUTS_SHORT_COUNT)) +
138*53e116feSMateja Marjanovic                                        RANDOM_INPUTS_SHORT_COUNT * i + j]);
139*53e116feSMateja Marjanovic         }
140*53e116feSMateja Marjanovic     }
141*53e116feSMateja Marjanovic 
142*53e116feSMateja Marjanovic     gettimeofday(&end, NULL);
143*53e116feSMateja Marjanovic 
144*53e116feSMateja Marjanovic     elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0;
145*53e116feSMateja Marjanovic     elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0;
146*53e116feSMateja Marjanovic 
147*53e116feSMateja Marjanovic     ret = check_results(instruction_name, TEST_COUNT_TOTAL, elapsed_time,
148*53e116feSMateja Marjanovic                         &b128_result[0][0], &b128_expect[0][0]);
149*53e116feSMateja Marjanovic 
150*53e116feSMateja Marjanovic     return ret;
151*53e116feSMateja Marjanovic }
152