-
Notifications
You must be signed in to change notification settings - Fork 0
/
cublas_benchmark.cu
254 lines (235 loc) · 8.52 KB
/
cublas_benchmark.cu
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
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
/*
* Copyright (C) 2012 NISHIMURA Ryohei
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include <cstdio>
#include <cstdlib>
#include <cublas_v2.h>
#define CUDA_SAFE_CALL(E) do { \
cudaError_t e = (E); \
if (e != cudaSuccess) { \
printf("line %d: CUDA error: %s\n", __LINE__, cudaGetErrorString(e)); \
exit(-2); \
} \
} while (false)
#define CUBLAS_SAFE_CALL(S) do { \
cublasStatus_t s = (S); \
if (s != CUBLAS_STATUS_SUCCESS) { \
printf("line %d: CUBLAS error: %d\n", __LINE__, s); \
exit(-3); \
} \
} while (false)
#define CHECKZERO(X) do { \
if ((X) <= 0) { \
printf("error: " #X " = %d <= 0!\n", (X)); \
exit(-1); \
} \
} while(false)
#define CHECKMINMAX(X, Y) do { \
if ((X) > (Y)) { \
printf("error: " #X " = %d > %d = " #Y "!\n", (X), (Y)); \
exit(-1); \
} \
} while(false)
template<typename T> void
gemm(cublasHandle_t handle, int m, int n, int k,
const T *a, int lda, const T *b, int ldb,
T *c, int ldc);
template<> void
gemm<float>(cublasHandle_t handle, int m, int n, int k,
const float *a, int lda,
const float *b, int ldb,
float *c, int ldc)
{
float alpha = 1.0f;
float beta = 1.0f;
CUBLAS_SAFE_CALL(cublasSgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N,
m, n, k,
&alpha, a, lda, b, ldb, &beta, c, ldc));
}
template<> void
gemm<double>(cublasHandle_t handle, int m, int n, int k,
const double *a, int lda,
const double *b, int ldb,
double *c, int ldc)
{
double alpha = 1.0;
double beta = 1.0;
CUBLAS_SAFE_CALL(cublasDgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N,
m, n, k,
&alpha, a, lda, b, ldb, &beta, c, ldc));
}
template<> void
gemm<cuComplex>(cublasHandle_t handle, int m, int n, int k,
const cuComplex *a, int lda,
const cuComplex *b, int ldb,
cuComplex *c, int ldc)
{
cuComplex alpha = make_cuComplex(1.0f, 0.0f);
cuComplex beta = make_cuComplex(1.0f, 0.0f);
CUBLAS_SAFE_CALL(cublasCgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N,
m, n, k,
&alpha, a, lda, b, ldb, &beta, c, ldc));
}
template<> void
gemm<cuDoubleComplex>(cublasHandle_t handle, int m, int n, int k,
const cuDoubleComplex *a, int lda,
const cuDoubleComplex *b, int ldb,
cuDoubleComplex *c, int ldc)
{
cuDoubleComplex alpha = make_cuDoubleComplex(1.0, 0.0);
cuDoubleComplex beta = make_cuDoubleComplex(1.0, 0.0);
CUBLAS_SAFE_CALL(cublasZgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N,
m, n, k,
&alpha, a, lda, b, ldb, &beta, c, ldc));
}
template<typename T> double
calc_gflops(int m, int n, int k, float ms);
template<> double
calc_gflops<float>(int m, int n, int k, float ms)
{
return 2.0 * m * n * k / ms * 1.0e-6;
}
template<> double
calc_gflops<double>(int m, int n, int k, float ms)
{
return 2.0 * m * n * k / ms * 1.0e-6;
}
template<> double
calc_gflops<cuComplex>(int m, int n, int k, float ms)
{
return 8.0 * m * n * k / ms * 1.0e-6;
}
template<> double
calc_gflops<cuDoubleComplex>(int m, int n, int k, float ms)
{
return 8.0 * m * n * k / ms * 1.0e-6;
}
template<typename T> void
benchmark(int m, int n, int k,
int ldamin, int ldamax, int ldastep,
int ldbmin, int ldbmax, int ldbstep,
int ldcmin, int ldcmax, int ldcstep)
{
cublasHandle_t handle;
CUBLAS_SAFE_CALL(cublasCreate(&handle));
T *a, *b, *c;
CUDA_SAFE_CALL(cudaMalloc(reinterpret_cast<void **>(&a),
k * ldamax * sizeof(T)));
CUDA_SAFE_CALL(cudaMemset(a, 0, k * ldamax * sizeof(T)));
CUDA_SAFE_CALL(cudaMalloc(reinterpret_cast<void **>(&b),
n * ldbmax * sizeof(T)));
CUDA_SAFE_CALL(cudaMemset(b, 0, n * ldbmax * sizeof(T)));
CUDA_SAFE_CALL(cudaMalloc(reinterpret_cast<void **>(&c),
n * ldcmax * sizeof(T)));
CUDA_SAFE_CALL(cudaMemset(c, 0, n * ldcmax * sizeof(T)));
cudaStream_t stream;
CUDA_SAFE_CALL(cudaStreamCreate(&stream));
CUBLAS_SAFE_CALL(cublasSetStream(handle, stream));
cudaEvent_t start, end;
CUDA_SAFE_CALL(cudaEventCreate(&start));
CUDA_SAFE_CALL(cudaEventCreate(&end));
printf("M: %d\tN: %d\tK: %d\n", m, n, k);
printf("LDA\tLDB\tLDC\tSecond\tGFLOPS\n");
for (int ldc = ldcmin; ldc <= ldcmax; ldc += ldcstep) {
for (int ldb = ldbmin; ldb <= ldbmax; ldb += ldbstep) {
for (int lda = ldamin; lda <= ldamax; lda += ldastep) {
CUDA_SAFE_CALL(cudaEventRecord(start, stream));
gemm(handle, m, n, k, a, lda, b, ldb, c, ldc);
CUDA_SAFE_CALL(cudaEventRecord(end, stream));
CUDA_SAFE_CALL(cudaEventSynchronize(end));
float ms;
CUDA_SAFE_CALL(cudaEventElapsedTime(&ms, start, end));
double gflops = calc_gflops<T>(m, n, k, ms);
printf("%d\t%d\t%d\t%.6f\t%.6f\n", lda, ldb, ldc, ms, gflops);
}
}
}
CUDA_SAFE_CALL(cudaEventDestroy(end));
CUDA_SAFE_CALL(cudaEventDestroy(start));
CUDA_SAFE_CALL(cudaStreamDestroy(stream));
CUBLAS_SAFE_CALL(cublasDestroy(handle));
}
int
main(int argc, char **argv)
{
if (argc < 14) {
printf("usage: %s typename([sdcz]) m n k ldamin ldamax ldastep ldbmin ldbmax ldbstep ldcmin ldcmax ldcstep\n", argv[0]);
return 0;
}
ptrdiff_t iarg = 0;
char *type = argv[++iarg];
int m = atoi(argv[++iarg]);
int n = atoi(argv[++iarg]);
int k = atoi(argv[++iarg]);
int ldamin = atoi(argv[++iarg]);
int ldamax = atoi(argv[++iarg]);
int ldastep = atoi(argv[++iarg]);
int ldbmin = atoi(argv[++iarg]);
int ldbmax = atoi(argv[++iarg]);
int ldbstep = atoi(argv[++iarg]);
int ldcmin = atoi(argv[++iarg]);
int ldcmax = atoi(argv[++iarg]);
int ldcstep = atoi(argv[++iarg]);
CHECKZERO(m);
CHECKZERO(n);
CHECKZERO(k);
CHECKZERO(ldamin);
CHECKMINMAX(m, ldamin);
CHECKMINMAX(ldamin, ldamax);
CHECKZERO(ldastep);
CHECKZERO(ldbmin);
CHECKMINMAX(k, ldbmin);
CHECKMINMAX(ldbmin, ldbmax);
CHECKZERO(ldbstep);
CHECKZERO(ldcmin);
CHECKMINMAX(m, ldcmin);
CHECKMINMAX(ldcmin, ldcmax);
CHECKZERO(ldcstep);
switch (type[0]) {
case 'S':
case 's':
benchmark<float>(m, n, k,
ldamin, ldamax, ldastep,
ldbmin, ldbmax, ldbstep,
ldcmin, ldcmax, ldcstep);
break;
case 'D':
case 'd':
benchmark<double>(m, n, k,
ldamin, ldamax, ldastep,
ldbmin, ldbmax, ldbstep,
ldcmin, ldcmax, ldcstep);
break;
case 'C':
case 'c':
benchmark<cuComplex>(m, n, k,
ldamin, ldamax, ldastep,
ldbmin, ldbmax, ldbstep,
ldcmin, ldcmax, ldcstep);
break;
case 'Z':
case 'z':
benchmark<cuDoubleComplex>(m, n, k,
ldamin, ldamax, ldastep,
ldbmin, ldbmax, ldbstep,
ldcmin, ldcmax, ldcstep);
break;
default:
printf("invaild typename: %s\n", type);
return -1;
}
return 0;
}