-
Notifications
You must be signed in to change notification settings - Fork 3
/
tests-512.c
236 lines (208 loc) · 8.19 KB
/
tests-512.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
#include "assert.h"
#include "stdatomic.h"
#include "stdint.h"
#include "stdio.h"
#include "sys/sysinfo.h"
#include "immintrin.h"
#include "cblas.h"
#include "SIMDxorshift/include/simdxorshift128plus.h"
static uint64_t CPUS;
static _Atomic(uint64_t)* ERRORS;
static avx512_xorshift128plus_key_t rng;
#define FOR_EACH_BLOCK_UP \
_Pragma("omp parallel for schedule(static)") \
for(ssize_t i = 0; i < CPUS; i++) \
for(ssize_t j = 0; j < (size / CPUS); j += 64)
#define FOR_EACH_BLOCK_DOWN \
_Pragma("omp parallel for schedule(static)") \
for(ssize_t i = CPUS - 1; i >= 0; i--) \
for(ssize_t j = (size / CPUS) - 64; j >= 0; j -= 64)
#define BLOCK_IDX (j + i * (size / CPUS))
void avx512_tests_init(size_t cpus, _Atomic(uint64_t)* errors) {
CPUS = cpus;
ERRORS = errors;
unsigned long long r1 = 0, r2 = 0;
while(r1 == 0 && r2 == 0) {
assert(_rdrand64_step(&r1));
assert(_rdrand64_step(&r2));
}
avx512_xorshift128plus_init(r1, r2, &rng);
}
static inline void get(const char* const restrict mem, const size_t idx, const __m512i expected) {
const __m512i actual = _mm512_load_si512(&mem[idx]);
_mm_lfence();
const __mmask64 result = _mm512_cmp_epu8_mask(expected, actual, _MM_CMPINT_NE);
if(__builtin_expect(result != 0, 0)) {
const uint64_t error_total = _mm_popcnt_u64(result);
fprintf(stderr, "%ld errors detected at offset 0x%016lx [error mask: 0x%016llx]\n", error_total, idx, result);
atomic_fetch_add(ERRORS, error_total);
}
}
static inline void get_all_up(const char* const restrict mem, const size_t size, const __m512i expected) {
FOR_EACH_BLOCK_UP {
get(mem, BLOCK_IDX, expected);
}
}
static inline void get_all_down(const char* const restrict mem, const size_t size, const __m512i expected) {
FOR_EACH_BLOCK_DOWN {
get(mem, BLOCK_IDX, expected);
}
}
static inline void set(char* const restrict mem, const size_t idx, const __m512i val) {
_mm512_stream_si512((__m512i*)&mem[idx], val);
}
static inline void set_all_up(char* const restrict mem, const size_t size, const __m512i val) {
FOR_EACH_BLOCK_UP {
set(mem, BLOCK_IDX, val);
}
}
static inline void set_all_down(char* const restrict mem, const size_t size, const __m512i val) {
FOR_EACH_BLOCK_DOWN {
set(mem, BLOCK_IDX, val);
}
}
void avx512_basic_tests(void* const restrict mem, const size_t size) {
const uint8_t patterns[] = { 0x00, 0xFF, 0x0F, 0xF0, 0x55, 0xAA, };
for(size_t i = 0; i < sizeof(patterns) / sizeof(uint8_t); i++) {
const __m512i pattern = _mm512_set1_epi8(patterns[i]);
set_all_up(mem, size, pattern);
get_all_up(mem, size, pattern);
set_all_down(mem, size, pattern);
get_all_down(mem, size, pattern);
}
}
void avx512_march(void* const restrict mem, const size_t size) {
for(size_t _ = 0; _ < 2; _++) {
const __m512i ones = _mm512_set1_epi8(0xFF);
const __m512i zeroes = _mm512_set1_epi8(0x00);
FOR_EACH_BLOCK_DOWN {
set(mem, j + i * (size / CPUS), zeroes);
}
FOR_EACH_BLOCK_UP {
get(mem, BLOCK_IDX, zeroes);
set(mem, BLOCK_IDX, ones);
get(mem, BLOCK_IDX, ones);
set(mem, BLOCK_IDX, zeroes);
get(mem, BLOCK_IDX, zeroes);
set(mem, BLOCK_IDX, ones);
}
FOR_EACH_BLOCK_UP {
get(mem, BLOCK_IDX, ones);
set(mem, BLOCK_IDX, zeroes);
set(mem, BLOCK_IDX, ones);
}
FOR_EACH_BLOCK_DOWN {
get(mem, BLOCK_IDX, ones);
set(mem, BLOCK_IDX, zeroes);
set(mem, BLOCK_IDX, ones);
set(mem, BLOCK_IDX, zeroes);
}
FOR_EACH_BLOCK_DOWN {
get(mem, BLOCK_IDX, zeroes);
set(mem, BLOCK_IDX, ones);
set(mem, BLOCK_IDX, zeroes);
}
}
}
void avx512_random_inversions(void* const restrict mem, const size_t size) {
for(size_t i = 0; i < 16; i++) {
const __m512i pattern = avx512_xorshift128plus(&rng);
set_all_up(mem, size, pattern);
get_all_up(mem, size, pattern);
const __m512i not_pattern = _mm512_xor_epi64(pattern, _mm512_set1_epi8(0xFF));
set_all_up(mem, size, not_pattern);
get_all_up(mem, size, not_pattern);
}
}
static void moving_inversions_template(void* const restrict mem, const size_t size, const size_t iters, __m512i (*shift)(__m512i, unsigned int), const __m512i initial) {
for(size_t i = 0; i < iters; i++) {
const __m512i pattern = shift(initial, i);
set_all_up(mem, size, pattern);
get_all_up(mem, size, pattern);
const __m512i not_pattern = _mm512_xor_epi64(pattern, _mm512_set1_epi8(0xFF));
set_all_up(mem, size, not_pattern);
get_all_up(mem, size, not_pattern);
}
}
void avx512_moving_inversions_left_64(void* const restrict mem, const size_t size) {
return moving_inversions_template(mem, size, 64, _mm512_slli_epi64, _mm512_set1_epi64(0x0000000000000001));
}
void avx512_moving_inversions_right_32(void* const restrict mem, const size_t size) {
return moving_inversions_template(mem, size, 32, _mm512_srli_epi64, _mm512_set1_epi32(0x80000000));
}
void avx512_moving_inversions_left_16(void* const restrict mem, const size_t size) {
return moving_inversions_template(mem, size, 16, _mm512_slli_epi64, _mm512_set1_epi16(0x0001));
}
void avx512_moving_inversions_right_8(void* const restrict mem, const size_t size) {
return moving_inversions_template(mem, size, 8, _mm512_srli_epi64, _mm512_set1_epi8(0x80));
}
void avx512_moving_inversions_left_4(void* const restrict mem, const size_t size) {
return moving_inversions_template(mem, size, 4, _mm512_slli_epi64, _mm512_set1_epi8(0x11));
}
void avx512_moving_saturations_right_16(void* const restrict mem, const size_t size) {
for(size_t i = 0; i < 16; i++) {
const __m512i pattern = _mm512_srli_epi16(_mm512_set1_epi16(0x8000), i);
set_all_up(mem, size, pattern);
get_all_up(mem, size, pattern);
const __m512i zeroes = _mm512_set1_epi8(0x00);
set_all_up(mem, size, zeroes);
get_all_up(mem, size, zeroes);
set_all_up(mem, size, pattern);
get_all_up(mem, size, pattern);
const __m512i ones = _mm512_set1_epi8(0xFF);
set_all_up(mem, size, ones);
get_all_up(mem, size, ones);
}
}
void avx512_moving_saturations_left_8(void* const restrict mem, const size_t size) {
for(size_t i = 0; i < 8; i++) {
const __m512i pattern = _mm512_srli_epi16(_mm512_set1_epi16(0x01), i);
set_all_up(mem, size, pattern);
get_all_up(mem, size, pattern);
const __m512i zeroes = _mm512_set1_epi8(0x00);
set_all_up(mem, size, zeroes);
get_all_up(mem, size, zeroes);
set_all_up(mem, size, pattern);
get_all_up(mem, size, pattern);
const __m512i ones = _mm512_set1_epi8(0xFF);
set_all_up(mem, size, ones);
get_all_up(mem, size, ones);
}
}
void avx512_addressing(void* const restrict mem, const size_t size) {
for(size_t _ = 0; _ < 16; _++) {
__m512i increasing = _mm512_set_epi64(56, 48, 40, 32, 24, 16, 8, 0);
FOR_EACH_BLOCK_UP {
set(mem, BLOCK_IDX, _mm512_add_epi64(_mm512_set1_epi64(BLOCK_IDX), increasing));
}
FOR_EACH_BLOCK_UP {
get(mem, BLOCK_IDX, _mm512_add_epi64(_mm512_set1_epi64(BLOCK_IDX), increasing));
}
FOR_EACH_BLOCK_DOWN {
set(mem, BLOCK_IDX, _mm512_add_epi64(_mm512_set1_epi64(BLOCK_IDX), increasing));
}
FOR_EACH_BLOCK_DOWN {
get(mem, BLOCK_IDX, _mm512_add_epi64(_mm512_set1_epi64(BLOCK_IDX), increasing));
}
}
}
void avx512_sgemm(char* const restrict mem, const size_t size) {
const __m512 zeroes = _mm512_set1_ps(0.0f);
set_all_down(mem, size, (__m512i) zeroes);
for(ssize_t _ = 0; _ < 32; _++) {
_Pragma("omp parallel for schedule(static)")
for(ssize_t i = 0; i < CPUS; i++) {
for(ssize_t j = 64 * 64 * 4 * 2; j < (size / CPUS); j += 64 * 64 * 4) {
float* const a = (float*) &mem[BLOCK_IDX - 64 * 64 * 4 * 2];
float* const b = (float*) &mem[BLOCK_IDX - 64 * 64 * 4 * 1];
float* const c = (float*) &mem[BLOCK_IDX - 64 * 64 * 4 * 0];
cblas_sgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, 64, 64, 64, 1.0, a, 64, b, 64, 0.0, c, 64);
for(ssize_t k = 0; k < 64 * 64 * 4; k += 64) {
_mm_clflushopt(&mem[BLOCK_IDX + k]);
}
_mm_sfence();
}
}
}
get_all_up(mem, size, (__m512i) zeroes);
}