-
Notifications
You must be signed in to change notification settings - Fork 548
Expand file tree
/
Copy pathcudnn.cpp
More file actions
307 lines (279 loc) · 12.2 KB
/
cudnn.cpp
File metadata and controls
307 lines (279 loc) · 12.2 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
/*******************************************************
* Copyright (c) 2018, ArrayFire
* All rights reserved.
*
* This file is distributed under 3-clause BSD license.
* The complete license agreement can be obtained at:
* http://arrayfire.com/licenses/BSD-3-Clause
********************************************************/
#include <cudnn.hpp>
#include <err_cuda.hpp>
using af::dim4;
namespace arrayfire {
namespace cuda {
const char *errorString(cudnnStatus_t err) {
switch (err) {
case CUDNN_STATUS_SUCCESS: return "CUDNN_STATUS_SUCCESS";
case CUDNN_STATUS_NOT_INITIALIZED:
return "CUDNN_STATUS_NOT_INITIALIZED";
case CUDNN_STATUS_ALLOC_FAILED: return "CUDNN_STATUS_ALLOC_FAILED";
case CUDNN_STATUS_BAD_PARAM: return "CUDNN_STATUS_BAD_PARAM";
case CUDNN_STATUS_INTERNAL_ERROR: return "CUDNN_STATUS_INTERNAL_ERROR";
case CUDNN_STATUS_INVALID_VALUE: return "CUDNN_STATUS_INVALID_VALUE";
case CUDNN_STATUS_ARCH_MISMATCH: return "CUDNN_STATUS_ARCH_MISMATCH";
case CUDNN_STATUS_MAPPING_ERROR: return "CUDNN_STATUS_MAPPING_ERROR";
case CUDNN_STATUS_EXECUTION_FAILED:
return "CUDNN_STATUS_EXECUTION_FAILED";
case CUDNN_STATUS_NOT_SUPPORTED: return "CUDNN_STATUS_NOT_SUPPORTED";
case CUDNN_STATUS_LICENSE_ERROR: return "CUDNN_STATUS_LICENSE_ERROR";
#if CUDNN_VERSION >= 6000
case CUDNN_STATUS_RUNTIME_PREREQUISITE_MISSING:
return "CUDNN_STATUS_RUNTIME_PREREQUISITE_MISSING";
#if CUDNN_VERSION >= 7000
case CUDNN_STATUS_RUNTIME_IN_PROGRESS:
return "CUDNN_STATUS_RUNTIME_IN_PROGRESS";
case CUDNN_STATUS_RUNTIME_FP_OVERFLOW:
return "CUDNN_STATUS_RUNTIME_FP_OVERFLOW";
#if CUDNN_VERSION >= 8000
case CUDNN_STATUS_VERSION_MISMATCH:
return "CUDNN_STATUS_VERSION_MISMATCH";
#endif
#endif
#endif
default: return "UNKNOWN";
}
}
template<>
cudnnDataType_t getCudnnDataType<float>() {
return CUDNN_DATA_FLOAT;
}
template<>
cudnnDataType_t getCudnnDataType<double>() {
return CUDNN_DATA_DOUBLE;
}
#if CUDNN_VERSION >= 6000
template<>
cudnnDataType_t getCudnnDataType<int>() {
return CUDNN_DATA_INT32;
}
#if CUDNN_VERSION >= 7100
/// TODONT COMMIT
template<>
cudnnDataType_t getCudnnDataType<signed char>() {
return CUDNN_DATA_INT8;
}
template<>
cudnnDataType_t getCudnnDataType<unsigned char>() {
return CUDNN_DATA_UINT8;
}
#endif
#endif
template<>
cudnnDataType_t getCudnnDataType<common::half>() {
return CUDNN_DATA_HALF;
}
void cudnnSet(cudnnTensorDescriptor_t desc, cudnnDataType_t cudnn_dtype,
dim4 dims) {
CUDNN_CHECK(cuda::cudnnSetTensor4dDescriptor(desc, CUDNN_TENSOR_NCHW,
cudnn_dtype, dims[3], dims[2],
dims[1], dims[0]));
}
void cudnnSet(cudnnFilterDescriptor_t desc, cudnnDataType_t cudnn_dtype,
dim4 dims) {
CUDNN_CHECK(cuda::cudnnSetFilter4dDescriptor(desc, cudnn_dtype,
CUDNN_TENSOR_NCHW, dims[3],
dims[2], dims[1], dims[0]));
}
cudnnStatus_t cudnnSetConvolution2dDescriptor(
cudnnConvolutionDescriptor_t convDesc,
int pad_h, // zero-padding height
int pad_w, // zero-padding width
int u, // vertical filter stride
int v, // horizontal filter stride
int upscalex, // upscale the input in x-direction
int upscaley, // upscale the input in y-direction
cudnnConvolutionMode_t mode, cudnnDataType_t computeType) {
return
#if CUDNN_VERSION >= 6000
getCudnnPlugin().cudnnSetConvolution2dDescriptor(
convDesc, pad_h, pad_w, u, v, upscalex, upscaley, mode,
computeType);
#elif CUDNN_VERSION >= 4000
getCudnnPlugin().cudnnSetConvolution2dDescriptor(
convDesc, pad_h, pad_w, u, v, upscalex, upscaley, mode);
#else
static_assert(1 != 1, "cuDNN version not supported");
#endif
}
cudnnStatus_t cudnnSetFilter4dDescriptor(cudnnFilterDescriptor_t filterDesc,
cudnnDataType_t dataType,
cudnnTensorFormat_t format, int k,
int c, int h, int w) {
#if CUDNN_VERSION >= 6000
int version = getCudnnPlugin().cudnnGetVersion();
if (version >= 6000) {
return getCudnnPlugin().cudnnSetFilter4dDescriptor(filterDesc, dataType,
format, k, c, h, w);
} else if (version == 4000) {
return getCudnnPlugin().cudnnSetFilter4dDescriptor_v4(
filterDesc, dataType, format, k, c, h, w);
}
CUDA_NOT_SUPPORTED(
"cudnnSetFilter4dDescriptor not supported for the current version of "
"cuDNN");
#elif CUDNN_VERSION == 4000
return getCudnnPlugin().cudnnSetFilter4dDescriptor_v4(filterDesc, dataType,
format, k, c, h, w);
#else
static_assert(1 != 1, "cuDNN version not supported");
#endif
}
cudnnStatus_t cudnnSetTensor4dDescriptor(cudnnTensorDescriptor_t tensorDesc,
cudnnTensorFormat_t format,
cudnnDataType_t dataType, int n, int c,
int h, int w) {
return getCudnnPlugin().cudnnSetTensor4dDescriptor(tensorDesc, format,
dataType, n, c, h, w);
}
cudnnStatus_t cudnnGetConvolutionBackwardDataWorkspaceSize(
cudnnHandle_t handle, const cudnnFilterDescriptor_t wDesc,
const cudnnTensorDescriptor_t dyDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t dxDesc, cudnnConvolutionBwdDataAlgo_t algo,
size_t *sizeInBytes) {
return getCudnnPlugin().cudnnGetConvolutionBackwardDataWorkspaceSize(
handle, wDesc, dyDesc, convDesc, dxDesc, algo, sizeInBytes);
}
cudnnStatus_t cudnnConvolutionBackwardData(
cudnnHandle_t handle, const void *alpha,
const cudnnFilterDescriptor_t wDesc, const void *w,
const cudnnTensorDescriptor_t dyDesc, const void *dy,
const cudnnConvolutionDescriptor_t convDesc,
cudnnConvolutionBwdDataAlgo_t algo, void *workSpace,
size_t workSpaceSizeInBytes, const void *beta,
const cudnnTensorDescriptor_t dxDesc, void *dx) {
return getCudnnPlugin().cudnnConvolutionBackwardData(
handle, alpha, wDesc, w, dyDesc, dy, convDesc, algo, workSpace,
workSpaceSizeInBytes, beta, dxDesc, dx);
}
cudnnStatus_t cudnnGetConvolutionNdForwardOutputDim(
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t inputTensorDesc,
const cudnnFilterDescriptor_t filterDesc, int nbDims,
int tensorOuputDimA[]) {
return getCudnnPlugin().cudnnGetConvolutionNdForwardOutputDim(
convDesc, inputTensorDesc, filterDesc, nbDims, tensorOuputDimA);
}
cudnnStatus_t cudnnGetConvolutionForwardAlgorithmMaxCount(cudnnHandle_t handle,
int *count) {
return getCudnnPlugin().cudnnGetConvolutionForwardAlgorithmMaxCount(handle,
count);
}
cudnnStatus_t cudnnGetConvolutionBackwardFilterAlgorithmMaxCount(
cudnnHandle_t handle, int *count) {
return getCudnnPlugin().cudnnGetConvolutionBackwardFilterAlgorithmMaxCount(
handle, count);
}
cudnnStatus_t cudnnGetConvolutionForwardWorkspaceSize(
cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
const cudnnFilterDescriptor_t wDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t yDesc, cudnnConvolutionFwdAlgo_t algo,
size_t *sizeInBytes) {
return getCudnnPlugin().cudnnGetConvolutionForwardWorkspaceSize(
handle, xDesc, wDesc, convDesc, yDesc, algo, sizeInBytes);
}
cudnnStatus_t cudnnGetConvolutionBackwardFilterWorkspaceSize(
cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
const cudnnTensorDescriptor_t dyDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnFilterDescriptor_t gradDesc,
cudnnConvolutionBwdFilterAlgo_t algo, size_t *sizeInBytes) {
return getCudnnPlugin().cudnnGetConvolutionBackwardFilterWorkspaceSize(
handle, xDesc, dyDesc, convDesc, gradDesc, algo, sizeInBytes);
}
cudnnStatus_t cudnnFindConvolutionForwardAlgorithm(
cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
const cudnnFilterDescriptor_t wDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t yDesc, const int requestedAlgoCount,
int *returnedAlgoCount, cudnnConvolutionFwdAlgoPerf_t *perfResults) {
return getCudnnPlugin().cudnnFindConvolutionForwardAlgorithm(
handle, xDesc, wDesc, convDesc, yDesc, requestedAlgoCount,
returnedAlgoCount, perfResults);
}
cudnnStatus_t cudnnFindConvolutionBackwardFilterAlgorithm(
cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
const cudnnTensorDescriptor_t dyDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnFilterDescriptor_t dwDesc, const int requestedAlgoCount,
int *returnedAlgoCount, cudnnConvolutionBwdFilterAlgoPerf_t *perfResults) {
return getCudnnPlugin().cudnnFindConvolutionBackwardFilterAlgorithm(
handle, xDesc, dyDesc, convDesc, dwDesc, requestedAlgoCount,
returnedAlgoCount, perfResults);
}
cudnnStatus_t cudnnGetConvolutionForwardAlgorithm(
cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
const cudnnFilterDescriptor_t wDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t yDesc,
cudnnConvolutionFwdPreference_t preference, size_t memoryLimitInBytes,
cudnnConvolutionFwdAlgo_t *algo) {
auto version = getCudnnPlugin().getVersion();
if (version.major() < 8) {
return getCudnnPlugin().cudnnGetConvolutionForwardAlgorithm(
handle, xDesc, wDesc, convDesc, yDesc, preference,
memoryLimitInBytes, algo);
} else {
AF_ERROR(
"cudnnGetConvolutionForwardAlgorithm has been removed since cuDNN "
"8",
AF_ERR_NOT_SUPPORTED);
return CUDNN_STATUS_SUCCESS;
}
}
cudnnStatus_t cudnnGetConvolutionBackwardFilterAlgorithm(
cudnnHandle_t handle, const cudnnTensorDescriptor_t xDesc,
const cudnnTensorDescriptor_t dyDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnFilterDescriptor_t dwDesc,
cudnnConvolutionBwdFilterPreference_t preference, size_t memoryLimitInBytes,
cudnnConvolutionBwdFilterAlgo_t *algo) {
auto version = getCudnnPlugin().getVersion();
if (version.major() < 8) {
return getCudnnPlugin().cudnnGetConvolutionBackwardFilterAlgorithm(
handle, xDesc, dyDesc, convDesc, dwDesc, preference,
memoryLimitInBytes, algo);
} else {
AF_ERROR(
"cudnnGetConvolutionBackwardFilterAlgorithm has been removed since "
"cuDNN 8",
AF_ERR_NOT_SUPPORTED);
return CUDNN_STATUS_SUCCESS;
}
}
cudnnStatus_t cudnnConvolutionForward(
cudnnHandle_t handle, const void *alpha,
const cudnnTensorDescriptor_t xDesc, const void *x,
const cudnnFilterDescriptor_t wDesc, const void *w,
const cudnnConvolutionDescriptor_t convDesc, cudnnConvolutionFwdAlgo_t algo,
void *workSpace, size_t workSpaceSizeInBytes, const void *beta,
const cudnnTensorDescriptor_t yDesc, void *y) {
return getCudnnPlugin().cudnnConvolutionForward(
handle, alpha, xDesc, x, wDesc, w, convDesc, algo, workSpace,
workSpaceSizeInBytes, beta, yDesc, y);
}
cudnnStatus_t cudnnConvolutionBackwardFilter(
cudnnHandle_t handle, const void *alpha,
const cudnnTensorDescriptor_t xDesc, const void *x,
const cudnnTensorDescriptor_t dyDesc, const void *dy,
const cudnnConvolutionDescriptor_t convDesc,
cudnnConvolutionBwdFilterAlgo_t algo, void *workSpace,
size_t workSpaceSizeInBytes, const void *beta,
const cudnnFilterDescriptor_t dwDesc, void *dw) {
return getCudnnPlugin().cudnnConvolutionBackwardFilter(
handle, alpha, xDesc, x, dyDesc, dy, convDesc, algo, workSpace,
workSpaceSizeInBytes, beta, dwDesc, dw);
}
} // namespace cuda
} // namespace arrayfire