1 /*
2  *  Test program for MSA instruction MSUB_Q.H
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             3 * (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 = "Fixed Multiply";
40     char *instruction_name =  "MSUB_Q.H";
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         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },    /*   0  */
49         { 0xffffffffffffffffULL, 0xffffffffffffffffULL, },
50         { 0xfffefffefffefffeULL, 0xfffefffefffefffeULL, },
51         { 0xfffefffefffefffeULL, 0xfffefffefffefffeULL, },
52         { 0xfffdfffdfffdfffdULL, 0xfffdfffdfffdfffdULL, },
53         { 0xfffdfffdfffdfffdULL, 0xfffdfffdfffdfffdULL, },
54         { 0xfffcfffdfffcfffcULL, 0xfffdfffcfffcfffdULL, },
55         { 0xfffcfffcfffcfffcULL, 0xfffcfffcfffcfffcULL, },
56         { 0xfffcfffcfffcfffcULL, 0xfffcfffcfffcfffcULL, },    /*   8  */
57         { 0xfffcfffcfffcfffcULL, 0xfffcfffcfffcfffcULL, },
58         { 0xfffcfffcfffcfffcULL, 0xfffcfffcfffcfffcULL, },
59         { 0xfffcfffcfffcfffcULL, 0xfffcfffcfffcfffcULL, },
60         { 0xfffcfffcfffcfffcULL, 0xfffcfffcfffcfffcULL, },
61         { 0xfffcfffcfffcfffcULL, 0xfffcfffcfffcfffcULL, },
62         { 0xfffcfffcfffcfffcULL, 0xfffcfffcfffcfffcULL, },
63         { 0xfffcfffcfffcfffcULL, 0xfffcfffcfffcfffcULL, },
64         { 0xfffbfffbfffbfffbULL, 0xfffbfffbfffbfffbULL, },    /*  16  */
65         { 0xfffbfffbfffbfffbULL, 0xfffbfffbfffbfffbULL, },
66         { 0xc716c716c716c716ULL, 0xc716c716c716c716ULL, },
67         { 0xfff9fff9fff9fff9ULL, 0xfff9fff9fff9fff9ULL, },
68         { 0xddd6ddd6ddd6ddd6ULL, 0xddd6ddd6ddd6ddd6ULL, },
69         { 0xfff8fff8fff8fff8ULL, 0xfff8fff8fff8fff8ULL, },
70         { 0xed0125e4b41ced01ULL, 0x25e4b41ced0125e4ULL, },
71         { 0xfff7fff6fff6fff7ULL, 0xfff6fff6fff7fff6ULL, },
72         { 0xfff7fff6fff6fff7ULL, 0xfff6fff6fff7fff6ULL, },    /*  24  */
73         { 0xfff7fff6fff6fff7ULL, 0xfff6fff6fff7fff6ULL, },
74         { 0x38da38d938d938daULL, 0x38d938d938da38d9ULL, },
75         { 0xfff6fff5fff5fff6ULL, 0xfff5fff5fff6fff5ULL, },
76         { 0x2218221722172218ULL, 0x2217221722182217ULL, },
77         { 0xfff6fff5fff5fff6ULL, 0xfff5fff5fff6fff5ULL, },
78         { 0x12ecda084bcf12ecULL, 0xda084bcf12ecda08ULL, },
79         { 0xfff6fff5fff5fff6ULL, 0xfff5fff5fff6fff5ULL, },
80         { 0xfff5fff4fff4fff5ULL, 0xfff4fff4fff5fff4ULL, },    /*  32  */
81         { 0xfff5fff4fff4fff5ULL, 0xfff4fff4fff5fff4ULL, },
82         { 0xddd2ddd1ddd1ddd2ULL, 0xddd1ddd1ddd2ddd1ULL, },
83         { 0xfff4fff3fff3fff4ULL, 0xfff3fff3fff4fff3ULL, },
84         { 0xeb78eb77eb77eb78ULL, 0xeb77eb77eb78eb77ULL, },
85         { 0xfff3fff2fff2fff3ULL, 0xfff2fff2fff3fff2ULL, },
86         { 0xf49216b3d26ef492ULL, 0x16b3d26ef49216b3ULL, },
87         { 0xfff2fff1fff1fff2ULL, 0xfff1fff1fff2fff1ULL, },
88         { 0xfff2fff1fff1fff2ULL, 0xfff1fff1fff2fff1ULL, },    /*  40  */
89         { 0xfff2fff1fff1fff2ULL, 0xfff1fff1fff2fff1ULL, },
90         { 0x2214221322132214ULL, 0x2213221322142213ULL, },
91         { 0xfff2fff1fff1fff2ULL, 0xfff1fff1fff2fff1ULL, },
92         { 0x146d146c146c146dULL, 0x146c146c146d146cULL, },
93         { 0xfff2fff1fff1fff2ULL, 0xfff1fff1fff2fff1ULL, },
94         { 0x0b52e92f2d740b52ULL, 0xe92f2d740b52e92fULL, },
95         { 0xfff1fff0fff1fff1ULL, 0xfff0fff1fff1fff0ULL, },
96         { 0xfff0fff0fff0fff0ULL, 0xfff0fff0fff0fff0ULL, },    /*  48  */
97         { 0xfff0fff0fff0fff0ULL, 0xfff0fff0fff0fff0ULL, },
98         { 0xecf925dcb414ecf9ULL, 0x25dcb414ecf925dcULL, },
99         { 0xffefffefffeeffefULL, 0xffefffeeffefffefULL, },
100         { 0xf48e16b0d26af48eULL, 0x16b0d26af48e16b0ULL, },
101         { 0xffeeffeeffedffeeULL, 0xffeeffedffeeffeeULL, },
102         { 0xf99be6a59ac8f99bULL, 0xe6a59ac8f99be6a5ULL, },
103         { 0xffedffedffebffedULL, 0xffedffebffedffedULL, },
104         { 0xffedffecffebffedULL, 0xffecffebffedffecULL, },    /*  56  */
105         { 0xffedffecffebffedULL, 0xffecffebffedffecULL, },
106         { 0x12e3d9fe4bc512e3ULL, 0xd9fe4bc512e3d9feULL, },
107         { 0xffedffebffebffedULL, 0xffebffebffedffebULL, },
108         { 0x0b4de9292d6e0b4dULL, 0xe9292d6e0b4de929ULL, },
109         { 0xffecffeaffebffecULL, 0xffeaffebffecffeaULL, },
110         { 0x063e1932650e063eULL, 0x1932650e063e1932ULL, },
111         { 0xffecffe8ffebffecULL, 0xffe8ffebffecffe8ULL, },
112         { 0x9032faf1f32dc724ULL, 0xd37cfee8ffe7cdf6ULL, },    /*  64  */
113         { 0x8c37fb04dab3ed15ULL, 0xc8500506002701cbULL, },
114         { 0x8000eb00f0d83aacULL, 0xb0d70a15ff2518f4ULL, },
115         { 0xe8edef64d3204e73ULL, 0xf40714a9fe1d069aULL, },
116         { 0xe4f2ef77baa67464ULL, 0xe8db1ac7fe5d3a6fULL, },
117         { 0xe4cdef768ba25b09ULL, 0xe60bf5b1fad604a2ULL, },
118         { 0xe204efb4b62c272fULL, 0xe023d70208eaec98ULL, },
119         { 0xe5c0efa2800019f7ULL, 0xf10996de174fffa3ULL, },
120         { 0x9799df9e9625678eULL, 0xd9909bed164d16ccULL, },    /*  72  */
121         { 0x94d0dfdcc0af33b4ULL, 0xd3a880002461fec2ULL, },
122         { 0x8000ac2c9a31c9abULL, 0xc7408000ec28f404ULL, },
123         { 0xc964ba57cdd7aea3ULL, 0xeac18000b2aafc86ULL, },
124         { 0x3251bebbb01fc26aULL, 0x2df18a94b1a2ea2cULL, },
125         { 0x360dbea98000b532ULL, 0x3ed78000c007fd37ULL, },
126         { 0x7f71ccd4b3a69a2aULL, 0x62588000868905b9ULL, },
127         { 0x1ce6c8f180009346ULL, 0xfcb580008000fefbULL, },
128         { 0x37e5be19a862dbafULL, 0xfea58b5e8000fe57ULL, },    /*  80  */
129         { 0x39c0be4bdd7bcb85ULL, 0xfed88000953fff6aULL, },
130         { 0x5f7d948aca8d9bc1ULL, 0xff3480008000ff95ULL, },
131         { 0x0bb4a742f1e1847fULL, 0xfe7e80008000ff7cULL, },
132         { 0x16a395c8f655d6c0ULL, 0xff618b5e8000ff29ULL, },
133         { 0x1763961afc30c464ULL, 0xff788000953fffb4ULL, },
134         { 0x26ab8000fa188e23ULL, 0xffa280008000ffcaULL, },
135         { 0x04bd964dfe708000ULL, 0xff4e80008000ffbdULL, },
136         { 0x092a817dfeeed540ULL, 0xffb68b5e8000ff93ULL, },    /*  88  */
137         { 0x097881deff94c239ULL, 0xffc08000953fffd9ULL, },
138         { 0x0fa88000ff5889feULL, 0xffd380008000ffe4ULL, },
139         { 0x01eb964dffd38000ULL, 0xffaa80008000ffddULL, },
140         { 0x03b5817dffe1d540ULL, 0xffdc8b5e8000ffc7ULL, },
141         { 0x03d481defff3c239ULL, 0xffe18000953fffebULL, },
142         { 0x06548000ffeb89feULL, 0xffea80008000fff1ULL, },
143         { 0x00c6964dfffa8000ULL, 0xffd680008000ffedULL, },
144         { 0x017e817dfffbd540ULL, 0xffee8b5e8000ffe1ULL, },    /*  96  */
145         { 0x02e28000fffcf1b8ULL, 0xfff895b98000ffcdULL, },
146         { 0x05938000fffdfb3aULL, 0xfffc9f298000ffadULL, },
147         { 0x0ac88000fffdfe67ULL, 0xfffea7c28000ff79ULL, },
148         { 0x0b238063fffefdb0ULL, 0xfffe8000953fffd0ULL, },
149         { 0x0b8180c5fffffca8ULL, 0xfffe8000a6f7ffefULL, },
150         { 0x0be28127fffffb2bULL, 0xfffe8000b5befffaULL, },
151         { 0x0c478189fffff904ULL, 0xfffe8000c211fffdULL, },
152         { 0x144c8000fffef2a8ULL, 0xfffe80009905fffdULL, },    /* 104  */
153         { 0x218f8000fffce682ULL, 0xfffe80008000fffdULL, },
154         { 0x377d8000fff9cf4eULL, 0xfffe80008000fffdULL, },
155         { 0x5bc08000fff5a2fbULL, 0xfffe80008000fffdULL, },
156         { 0x0b3f964dfffd8d66ULL, 0xfffc80008000fffcULL, },
157         { 0x0160a8b7ffff8000ULL, 0xfff880008000fffbULL, },
158         { 0x002bb7ecffff8000ULL, 0xfff080008000fff9ULL, },
159         { 0x0005c47affff8000ULL, 0xffe180008000fff7ULL, },
160 };
161 
162     reset_msa_registers();
163 
164     gettimeofday(&start, NULL);
165 
166     for (i = 0; i < PATTERN_INPUTS_SHORT_COUNT; i++) {
167         for (j = 0; j < PATTERN_INPUTS_SHORT_COUNT; j++) {
168             do_msa_MSUB_Q_H(b128_pattern[i], b128_pattern[j],
169                             b128_result[PATTERN_INPUTS_SHORT_COUNT * i + j]);
170         }
171     }
172 
173     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
174         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
175             do_msa_MSUB_Q_H(b128_random[i], b128_random[j],
176                             b128_result[((PATTERN_INPUTS_SHORT_COUNT) *
177                                          (PATTERN_INPUTS_SHORT_COUNT)) +
178                                         RANDOM_INPUTS_SHORT_COUNT * i + j]);
179         }
180     }
181 
182     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
183         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
184             do_msa_MSUB_Q_H__DDT(b128_random[i], b128_random[j],
185                                  b128_result[
186                                      ((PATTERN_INPUTS_SHORT_COUNT) *
187                                       (PATTERN_INPUTS_SHORT_COUNT)) +
188                                      ((RANDOM_INPUTS_SHORT_COUNT) *
189                                       (RANDOM_INPUTS_SHORT_COUNT)) +
190                                      RANDOM_INPUTS_SHORT_COUNT * i + j]);
191         }
192     }
193 
194     for (i = 0; i < RANDOM_INPUTS_SHORT_COUNT; i++) {
195         for (j = 0; j < RANDOM_INPUTS_SHORT_COUNT; j++) {
196             do_msa_MSUB_Q_H__DSD(b128_random[i], b128_random[j],
197                                  b128_result[
198                                      ((PATTERN_INPUTS_SHORT_COUNT) *
199                                       (PATTERN_INPUTS_SHORT_COUNT)) +
200                                      (2 * (RANDOM_INPUTS_SHORT_COUNT) *
201                                       (RANDOM_INPUTS_SHORT_COUNT)) +
202                                      RANDOM_INPUTS_SHORT_COUNT * i + j]);
203         }
204     }
205 
206     gettimeofday(&end, NULL);
207 
208     elapsed_time = (end.tv_sec - start.tv_sec) * 1000.0;
209     elapsed_time += (end.tv_usec - start.tv_usec) / 1000.0;
210 
211     ret = check_results_128(isa_ase_name, group_name, instruction_name,
212                             TEST_COUNT_TOTAL, elapsed_time,
213                             &b128_result[0][0], &b128_expect[0][0]);
214 
215     return ret;
216 }
217