forked from Theano/libgpuarray
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathgpuarray_blas_opencl_clblas.c
More file actions
421 lines (356 loc) · 12 KB
/
gpuarray_blas_opencl_clblas.c
File metadata and controls
421 lines (356 loc) · 12 KB
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
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
#include "private.h"
#include "private_opencl.h"
#include <clBLAS.h>
#include "gpuarray/buffer_blas.h"
#include "gpuarray/error.h"
static inline clblasOrder convO(cb_order order) {
switch (order) {
case cb_row:
return clblasRowMajor;
case cb_column:
return clblasColumnMajor;
default:
return -1;
}
}
static inline clblasTranspose convT(cb_transpose trans) {
switch (trans) {
case cb_no_trans:
return clblasNoTrans;
case cb_trans:
return clblasTrans;
case cb_conj_trans:
return clblasConjTrans;
default:
return -1;
}
}
static unsigned int refcnt = 0;
static int setup(gpucontext *ctx) {
clblasStatus err;
if (refcnt == 0) {
err = clblasSetup();
if (err != clblasSuccess)
return GA_BLAS_ERROR;
}
if (ctx->blas_handle == NULL)
ctx->blas_handle = &refcnt;
refcnt++;
return GA_NO_ERROR;
}
static void teardown(gpucontext *ctx) {
if (ctx->blas_handle != NULL) {
ctx->blas_handle = NULL;
refcnt--;
}
if (refcnt == 0)
clblasTeardown();
}
static const char *error(gpucontext *ctx) {
return "(clblas) error in blas call, no details for now.";
}
#define ARRAY_INIT(A) \
if (A->ev != NULL) \
evl[num_ev++] = A->ev
#define ARRAY_FINI(A) \
if (A->ev != NULL) \
clReleaseEvent(A->ev); \
A->ev = ev; \
clRetainEvent(A->ev)
static int hgemmBatch(cb_order order, cb_transpose transA, cb_transpose transB,
size_t M, size_t N, size_t K, float alpha,
gpudata **A, size_t *offA, size_t lda,
gpudata **B, size_t *offB, size_t ldb,
float beta, gpudata **C, size_t *offC, size_t ldc,
size_t batchCount) {
return GA_DEVSUP_ERROR;
}
static int sgemmBatch(cb_order order, cb_transpose transA, cb_transpose transB,
size_t M, size_t N, size_t K, float alpha,
gpudata **A, size_t *offA, size_t lda,
gpudata **B, size_t *offB, size_t ldb,
float beta, gpudata **C, size_t *offC, size_t ldc,
size_t batchCount) {
cl_ctx *ctx = A[0]->ctx;
cl_event evl[3];
cl_event ev;
size_t i;
cl_uint num_ev = 0;
clblasStatus err;
for (i = 0; i < batchCount; i++) {
ARRAY_INIT(A[i]);
ARRAY_INIT(B[i]);
ARRAY_INIT(C[i]);
err = clblasSgemm(convO(order), convT(transA), convT(transB), M, N, K,
alpha, A[i]->buf, offA[i], lda, B[i]->buf, offB[i], ldb,
beta, C[i]->buf, offB[i], ldc, 1, &ctx->q,
num_ev, num_ev == 0 ? NULL : evl, &ev);
if (err != clblasSuccess)
return GA_BLAS_ERROR;
ARRAY_FINI(A[i]);
ARRAY_FINI(B[i]);
ARRAY_FINI(C[i]);
clReleaseEvent(ev);
}
return GA_NO_ERROR;
}
static int dgemmBatch(cb_order order, cb_transpose transA, cb_transpose transB,
size_t M, size_t N, size_t K, double alpha,
gpudata **A, size_t *offA, size_t lda,
gpudata **B, size_t *offB, size_t ldb,
double beta, gpudata **C, size_t *offC, size_t ldc,
size_t batchCount) {
cl_ctx *ctx = A[0]->ctx;
cl_event evl[3];
cl_event ev;
size_t i;
cl_uint num_ev = 0;
clblasStatus err;
for (i = 0; i < batchCount; i++) {
ARRAY_INIT(A[i]);
ARRAY_INIT(B[i]);
ARRAY_INIT(C[i]);
err = clblasDgemm(convO(order), convT(transA), convT(transB), M, N, K,
alpha, A[i]->buf, offA[i], lda, B[i]->buf, offB[i], ldb,
beta, C[i]->buf, offB[i], ldc, 1, &ctx->q,
num_ev, num_ev == 0 ? NULL : evl, &ev);
if (err != clblasSuccess)
return GA_BLAS_ERROR;
ARRAY_FINI(A[i]);
ARRAY_FINI(B[i]);
ARRAY_FINI(C[i]);
clReleaseEvent(ev);
}
return GA_NO_ERROR;
}
static int hgemvBatch(cb_order order, cb_transpose transA,
size_t M, size_t N, float alpha,
gpudata **A, size_t *offA, size_t lda,
gpudata **x, size_t *offX, size_t incX,
float beta, gpudata **y, size_t *offY, size_t incY,
size_t batchCount, int flags) {
return GA_DEVSUP_ERROR;
}
static int sgemvBatch(cb_order order, cb_transpose transA,
size_t M, size_t N, float alpha,
gpudata **A, size_t *offA, size_t lda,
gpudata **x, size_t *offX, size_t incX,
float beta, gpudata **y, size_t *offY, size_t incY,
size_t batchCount, int flags) {
return GA_DEVSUP_ERROR;
}
static int dgemvBatch(cb_order order, cb_transpose transA,
size_t M, size_t N, double alpha,
gpudata **A, size_t *offA, size_t lda,
gpudata **x, size_t *offX, size_t incX,
double beta, gpudata **y, size_t *offY, size_t incY,
size_t batchCount, int flags) {
return GA_DEVSUP_ERROR;
}
static int hgerBatch(cb_order order, size_t M, size_t N, float alpha,
gpudata **x, size_t *offX, size_t incX,
gpudata **y, size_t *offY, size_t incY,
gpudata **A, size_t *offA, size_t lda,
size_t batchCount, int flags) {
return GA_DEVSUP_ERROR;
}
static int sgerBatch(cb_order order, size_t M, size_t N, float alpha,
gpudata **x, size_t *offX, size_t incX,
gpudata **y, size_t *offY, size_t incY,
gpudata **A, size_t *offA, size_t lda,
size_t batchCount, int flags) {
return GA_DEVSUP_ERROR;
}
static int dgerBatch(cb_order order, size_t M, size_t N, double alpha,
gpudata **x, size_t *offX, size_t incX,
gpudata **y, size_t *offY, size_t incY,
gpudata **A, size_t *offA, size_t lda,
size_t batchCount, int flags) {
return GA_DEVSUP_ERROR;
}
static int hgemv(cb_order order, cb_transpose transA, size_t M, size_t N,
float alpha, gpudata *A, size_t offA, size_t lda,
gpudata *X, size_t offX, int incX, float beta,
gpudata *Y, size_t offY, int incY) {
return GA_DEVSUP_ERROR;
}
static int sgemv(cb_order order, cb_transpose transA, size_t M, size_t N,
float alpha, gpudata *A, size_t offA, size_t lda,
gpudata *X, size_t offX, int incX, float beta,
gpudata *Y, size_t offY, int incY) {
cl_ctx *ctx = A->ctx;
clblasStatus err;
cl_uint num_ev = 0;
cl_event evl[3];
cl_event ev;
ARRAY_INIT(A);
ARRAY_INIT(X);
ARRAY_INIT(Y);
err = clblasSgemv(convO(order), convT(transA), M, N, alpha,
A->buf, offA, lda, X->buf, offX, incX,
beta, Y->buf, offY, incY, 1, &ctx->q,
num_ev, num_ev == 0 ? NULL : evl, &ev);
if (err != clblasSuccess)
return GA_BLAS_ERROR;
ARRAY_FINI(A);
ARRAY_FINI(X);
ARRAY_FINI(Y);
clReleaseEvent(ev);
return GA_NO_ERROR;
}
static int dgemv(cb_order order, cb_transpose transA, size_t M, size_t N,
double alpha, gpudata *A, size_t offA, size_t lda,
gpudata *X, size_t offX, int incX, double beta,
gpudata *Y, size_t offY, int incY) {
cl_ctx *ctx = A->ctx;
clblasStatus err;
cl_uint num_ev = 0;
cl_event evl[3];
cl_event ev;
ARRAY_INIT(A);
ARRAY_INIT(X);
ARRAY_INIT(Y);
err = clblasDgemv(convO(order), convT(transA), M, N, alpha,
A->buf, offA, lda, X->buf, offX, incX,
beta, Y->buf, offY, incY, 1, &ctx->q,
num_ev, num_ev == 0 ? NULL : evl, &ev);
if (err != clblasSuccess)
return GA_BLAS_ERROR;
ARRAY_FINI(A);
ARRAY_FINI(X);
ARRAY_FINI(Y);
clReleaseEvent(ev);
return GA_NO_ERROR;
}
static int hgemm(cb_order order, cb_transpose transA, cb_transpose transB,
size_t M, size_t N, size_t K, float alpha,
gpudata *A, size_t offA, size_t lda,
gpudata *B, size_t offB, size_t ldb, float beta,
gpudata *C, size_t offC, size_t ldc) {
return GA_DEVSUP_ERROR;
}
static int sgemm(cb_order order, cb_transpose transA, cb_transpose transB,
size_t M, size_t N, size_t K, float alpha,
gpudata *A, size_t offA, size_t lda,
gpudata *B, size_t offB, size_t ldb, float beta,
gpudata *C, size_t offC, size_t ldc) {
cl_ctx *ctx = A->ctx;
clblasStatus err;
cl_uint num_ev = 0;
cl_event evl[3];
cl_event ev;
ARRAY_INIT(A);
ARRAY_INIT(B);
ARRAY_INIT(C);
err = clblasSgemm(convO(order), convT(transA), convT(transB), M, N, K,
alpha, A->buf, offA, lda, B->buf, offB, ldb,
beta, C->buf, offC, ldc, 1, &ctx->q,
num_ev, num_ev == 0 ? NULL : evl, &ev);
if (err != clblasSuccess)
return GA_BLAS_ERROR;
ARRAY_FINI(A);
ARRAY_FINI(B);
ARRAY_FINI(C);
clReleaseEvent(ev);
return GA_NO_ERROR;
}
static int dgemm(cb_order order, cb_transpose transA, cb_transpose transB,
size_t M, size_t N, size_t K, double alpha,
gpudata *A, size_t offA, size_t lda,
gpudata *B, size_t offB, size_t ldb, double beta,
gpudata *C, size_t offC, size_t ldc) {
cl_ctx *ctx = A->ctx;
clblasStatus err;
cl_uint num_ev = 0;
cl_event evl[3];
cl_event ev;
ARRAY_INIT(A);
ARRAY_INIT(B);
ARRAY_INIT(C);
err = clblasDgemm(convO(order), convT(transA), convT(transB), M, N, K,
alpha, A->buf, offA, lda, B->buf, offB, ldb,
beta, C->buf, offC, ldc, 1, &ctx->q,
num_ev, num_ev == 0 ? NULL : evl, &ev);
if (err != clblasSuccess)
return GA_BLAS_ERROR;
ARRAY_FINI(A);
ARRAY_FINI(B);
ARRAY_FINI(C);
clReleaseEvent(ev);
return GA_NO_ERROR;
}
static int hger(cb_order order, size_t M, size_t N, float alpha,
gpudata *X, size_t offX, int incX,
gpudata *Y, size_t offY, int incY,
gpudata *A, size_t offA, size_t lda) {
return GA_DEVSUP_ERROR;
}
static int sger(cb_order order, size_t M, size_t N, float alpha,
gpudata *X, size_t offX, int incX,
gpudata *Y, size_t offY, int incY,
gpudata *A, size_t offA, size_t lda) {
cl_ctx *ctx = X->ctx;
cl_event evl[3];
cl_event ev;
cl_uint num_ev = 0;
clblasStatus err;
ARRAY_INIT(X);
ARRAY_INIT(Y);
ARRAY_INIT(A);
err = clblasSger(convO(order), M, N, alpha, X->buf, offX, incX,
Y->buf, offY, incY, A->buf, offA, lda, 1, &ctx->q,
num_ev, num_ev == 0 ? NULL : evl, &ev);
if (err != clblasSuccess)
return GA_BLAS_ERROR;
ARRAY_FINI(X);
ARRAY_FINI(Y);
ARRAY_FINI(A);
clReleaseEvent(ev);
return GA_NO_ERROR;
}
static int dger(cb_order order, size_t M, size_t N, double alpha,
gpudata *X, size_t offX, int incX,
gpudata *Y, size_t offY, int incY,
gpudata *A, size_t offA, size_t lda) {
cl_ctx *ctx = X->ctx;
cl_event evl[3];
cl_event ev;
cl_uint num_ev = 0;
clblasStatus err;
ARRAY_INIT(X);
ARRAY_INIT(Y);
ARRAY_INIT(A);
err = clblasDger(convO(order), M, N, alpha, X->buf, offX, incX,
Y->buf, offY, incY, A->buf, offA, lda, 1, &ctx->q,
num_ev, num_ev == 0 ? NULL : evl, &ev);
if (err != clblasSuccess)
return GA_BLAS_ERROR;
ARRAY_FINI(X);
ARRAY_FINI(Y);
ARRAY_FINI(A);
clReleaseEvent(ev);
return GA_NO_ERROR;
}
GPUARRAY_LOCAL gpuarray_blas_ops clblas_ops = {
setup,
teardown,
error,
hgemv, /* TODO */
sgemv,
dgemv,
hgemm, /* TODO */
sgemm,
dgemm,
hger, /* TODO */
sger,
dger,
hgemmBatch, /* TODO */
sgemmBatch,
dgemmBatch,
hgemvBatch, /* TODO */
sgemvBatch, /* TODO */
dgemvBatch, /* TODO */
hgerBatch, /* TODO */
sgerBatch, /* TODO */
dgerBatch, /* TODO */
};