-
Notifications
You must be signed in to change notification settings - Fork 156
/
ffi.lua
1630 lines (1402 loc) · 95.2 KB
/
ffi.lua
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
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
require 'cutorch'
local ffi = require 'ffi'
ffi.cdef[[
typedef enum {
CUDNN_MAJOR = 5,
CUDNN_MINOR = 0,
CUDNN_PATCHLEVEL = 4,
CUDNN_VERSION = (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL)
} cudnnVerFakeEnum;
struct cudnnContext;
typedef struct cudnnContext *cudnnHandle_t;
size_t cudnnGetVersion(void);
/*
* CUDNN return codes
*/
typedef enum
{
CUDNN_STATUS_SUCCESS = 0,
CUDNN_STATUS_NOT_INITIALIZED = 1,
CUDNN_STATUS_ALLOC_FAILED = 2,
CUDNN_STATUS_BAD_PARAM = 3,
CUDNN_STATUS_INTERNAL_ERROR = 4,
CUDNN_STATUS_INVALID_VALUE = 5,
CUDNN_STATUS_ARCH_MISMATCH = 6,
CUDNN_STATUS_MAPPING_ERROR = 7,
CUDNN_STATUS_EXECUTION_FAILED = 8,
CUDNN_STATUS_NOT_SUPPORTED = 9,
CUDNN_STATUS_LICENSE_ERROR = 10
} cudnnStatus_t;
/* human-readable error messages*/
const char * cudnnGetErrorString(cudnnStatus_t status);
cudnnStatus_t cudnnCreate (cudnnHandle_t *handle);
cudnnStatus_t cudnnDestroy (cudnnHandle_t handle);
cudnnStatus_t cudnnSetStream (cudnnHandle_t handle, cudaStream_t streamId);
cudnnStatus_t cudnnGetStream (cudnnHandle_t handle, cudaStream_t *streamId);
/* Data structures to represent Image/Filter and the Neural Network Layer */
typedef struct cudnnTensorStruct* cudnnTensorDescriptor_t;
typedef struct cudnnConvolutionStruct* cudnnConvolutionDescriptor_t;
typedef struct cudnnPoolingStruct* cudnnPoolingDescriptor_t;
typedef struct cudnnFilterStruct* cudnnFilterDescriptor_t;
typedef struct cudnnLRNStruct* cudnnLRNDescriptor_t;
typedef struct cudnnActivationStruct* cudnnActivationDescriptor_t;
typedef struct cudnnSpatialTransformerStruct* cudnnSpatialTransformerDescriptor_t;
typedef struct cudnnOpTensorStruct* cudnnOpTensorDescriptor_t;
/*
* CUDNN data type
*/
typedef enum
{
CUDNN_DATA_FLOAT = 0,
CUDNN_DATA_DOUBLE = 1,
CUDNN_DATA_HALF = 2,
} cudnnDataType_t;
/*
* CUDNN propagate Nan
*/
typedef enum{
CUDNN_NOT_PROPAGATE_NAN = 0,
CUDNN_PROPAGATE_NAN = 1,
} cudnnNanPropagation_t;
/* Maximum supported number of tensor dimensions */
typedef enum { CUDNN_DIM_MAX = 8 } cudnnDimMaxFakeEnum;
/* Create an instance of a generic Tensor descriptor */
cudnnStatus_t cudnnCreateTensorDescriptor(
cudnnTensorDescriptor_t *tensorDesc );
typedef enum
{
CUDNN_TENSOR_NCHW = 0, /* row major (wStride = 1, hStride = w) */
CUDNN_TENSOR_NHWC = 1 /* feature maps interleaved ( cStride = 1 )*/
} cudnnTensorFormat_t;
cudnnStatus_t cudnnSetTensor4dDescriptor(
cudnnTensorDescriptor_t tensorDesc,
cudnnTensorFormat_t format,
cudnnDataType_t dataType, /* image data type*/
int n, /* number of inputs (batch size)*/
int c, /* number of input feature maps*/
int h, /* height of input section*/
int w ); /* width of input section*/
cudnnStatus_t cudnnSetTensor4dDescriptorEx(
cudnnTensorDescriptor_t tensorDesc,
cudnnDataType_t dataType, /* image data type*/
int n, /* number of inputs (batch size)*/
int c, /* number of input feature maps*/
int h, /* height of input section*/
int w, /* width of input section*/
int nStride,
int cStride,
int hStride,
int wStride );
cudnnStatus_t cudnnGetTensor4dDescriptor(
const cudnnTensorDescriptor_t tensorDesc,
cudnnDataType_t *dataType, /* image data type*/
int *n, /* number of inputs (batch size)*/
int *c, /* number of input feature maps*/
int *h, /* height of input section*/
int *w, /* width of input section*/
int *nStride,
int *cStride,
int *hStride,
int *wStride );
cudnnStatus_t cudnnSetTensorNdDescriptor(
cudnnTensorDescriptor_t tensorDesc,
cudnnDataType_t dataType,
int nbDims,
const int dimA[],
const int strideA[] );
cudnnStatus_t cudnnGetTensorNdDescriptor(
const cudnnTensorDescriptor_t tensorDesc,
int nbDimsRequested,
cudnnDataType_t *dataType,
int *nbDims,
int dimA[],
int strideA[] );
/* PixelOffset( n, c, h, w ) = n *input_stride + c * feature_stride + h * h_stride + w * w_stride
1)Example of all images in row major order one batch of features after the other (with an optional padding on row)
input_stride : c x h x h_stride
feature_stride : h x h_stride
h_stride : >= w ( h_stride = w if no padding)
w_stride : 1
2)Example of all images in row major with features maps interleaved
input_stride : c x h x h_stride
feature_stride : 1
h_stride : w x c
w_stride : c
3)Example of all images in column major order one batch of features after the other (with optional padding on column)
input_stride : c x w x w_stride
feature_stride : w x w_stride
h_stride : 1
w_stride : >= h
*/
/* Destroy an instance of Tensor4d descriptor */
cudnnStatus_t cudnnDestroyTensorDescriptor(
cudnnTensorDescriptor_t tensorDesc );
/* Tensor layout conversion helper (y = alpha * x + beta * y) */
cudnnStatus_t cudnnTransformTensor(
cudnnHandle_t handle,
const void *alpha,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *beta,
const cudnnTensorDescriptor_t yDesc,
void *y );
/* Tensor Bias addition : C = alpha * A + beta * C */
cudnnStatus_t cudnnAddTensor(
cudnnHandle_t handle,
const void *alpha,
const cudnnTensorDescriptor_t aDesc,
const void *A,
const void *beta,
const cudnnTensorDescriptor_t cDesc,
void *C );
/*
* CUDNN OpTensor op type
*/
typedef enum
{
CUDNN_OP_TENSOR_ADD = 0,
CUDNN_OP_TENSOR_MUL = 1,
CUDNN_OP_TENSOR_MIN = 2,
CUDNN_OP_TENSOR_MAX = 3,
} cudnnOpTensorOp_t;
cudnnStatus_t cudnnCreateOpTensorDescriptor(
cudnnOpTensorDescriptor_t *opTensorDesc );
cudnnStatus_t cudnnSetOpTensorDescriptor(
cudnnOpTensorDescriptor_t opTensorDesc,
cudnnOpTensorOp_t opTensorOp,
cudnnDataType_t opTensorCompType,
cudnnNanPropagation_t opTensorNanOpt );
cudnnStatus_t cudnnGetOpTensorDescriptor(
const cudnnOpTensorDescriptor_t opTensorDesc,
cudnnOpTensorOp_t *opTensorOp,
cudnnDataType_t *opTensorCompType,
cudnnNanPropagation_t *opTensorNanOpt );
cudnnStatus_t cudnnDestroyOpTensorDescriptor(
cudnnOpTensorDescriptor_t opTensorDesc );
/* Tensor Bias operation : C = op( alpha1 * A, alpha2 * B ) + beta * C */
cudnnStatus_t cudnnOpTensor(
cudnnHandle_t handle,
const cudnnOpTensorDescriptor_t opTensorDesc,
const void *alpha1,
const cudnnTensorDescriptor_t aDesc,
const void *A,
const void *alpha2,
const cudnnTensorDescriptor_t bDesc,
const void *B,
const void *beta,
const cudnnTensorDescriptor_t cDesc,
void *C );
/* Set all values of a tensor to a given value : y[i] = value[0] */
cudnnStatus_t cudnnSetTensor(
cudnnHandle_t handle,
const cudnnTensorDescriptor_t yDesc,
void *y,
const void *valuePtr );
/* Scale all values of a tensor by a given factor : y[i] = alpha * y[i] */
cudnnStatus_t cudnnScaleTensor(
cudnnHandle_t handle,
const cudnnTensorDescriptor_t yDesc,
void *y,
const void *alpha );
/*
* convolution mode
*/
typedef enum
{
CUDNN_CONVOLUTION = 0,
CUDNN_CROSS_CORRELATION = 1
} cudnnConvolutionMode_t;
/* Create an instance of FilterStruct */
cudnnStatus_t cudnnCreateFilterDescriptor(
cudnnFilterDescriptor_t *filterDesc );
cudnnStatus_t cudnnSetFilter4dDescriptor(
cudnnFilterDescriptor_t filterDesc,
cudnnDataType_t dataType, /* image data type*/
cudnnTensorFormat_t format,
int k, /* number of output feature maps*/
int c, /* number of input feature maps*/
int h, /* height of each input filter*/
int w ); /* width of each input filter*/
cudnnStatus_t cudnnGetFilter4dDescriptor(
const cudnnFilterDescriptor_t filterDesc,
cudnnDataType_t *dataType, /* image data type*/
cudnnTensorFormat_t *format,
int *k, /* number of output feature maps*/
int *c, /* number of input feature maps*/
int *h, /* height of each input filter*/
int *w ); /* width of each input filter*/
cudnnStatus_t cudnnSetFilterNdDescriptor(
cudnnFilterDescriptor_t filterDesc,
cudnnDataType_t dataType, /* image data type*/
cudnnTensorFormat_t format,
int nbDims,
const int filterDimA[] );
cudnnStatus_t cudnnGetFilterNdDescriptor(
const cudnnFilterDescriptor_t filterDesc,
int nbDimsRequested,
cudnnDataType_t *dataType, /* image data type*/
cudnnTensorFormat_t *format,
int *nbDims,
int filterDimA[] );
cudnnStatus_t cudnnDestroyFilterDescriptor(
cudnnFilterDescriptor_t filterDesc );
/* Create an instance of convolution descriptor */
cudnnStatus_t cudnnCreateConvolutionDescriptor(
cudnnConvolutionDescriptor_t *convDesc );
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 );
cudnnStatus_t cudnnSetConvolution2dDescriptor_v5( 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 dataType
);
cudnnStatus_t cudnnGetConvolution2dDescriptor(
const 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 );
cudnnStatus_t cudnnGetConvolution2dDescriptor_v5( const 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 *dataType
);
/* Helper function to return the dimensions of the output tensor given a convolution descriptor */
cudnnStatus_t cudnnGetConvolution2dForwardOutputDim(
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t inputTensorDesc,
const cudnnFilterDescriptor_t filterDesc,
int *n,
int *c,
int *h,
int *w );
cudnnStatus_t cudnnSetConvolutionNdDescriptor(
cudnnConvolutionDescriptor_t convDesc,
int arrayLength, /* nbDims-2 size */
const int padA[],
const int filterStrideA[],
const int upscaleA[],
cudnnConvolutionMode_t mode,
cudnnDataType_t dataType ); /* convolution data type*/
cudnnStatus_t cudnnGetConvolutionNdDescriptor(
const cudnnConvolutionDescriptor_t convDesc,
int arrayLengthRequested,
int *arrayLength,
int padA[],
int strideA[],
int upscaleA[],
cudnnConvolutionMode_t *mode,
cudnnDataType_t *dataType ); /* convolution data type*/
/* Helper function to return the dimensions of the output tensor given a convolution descriptor */
cudnnStatus_t cudnnGetConvolutionNdForwardOutputDim(
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t inputTensorDesc,
const cudnnFilterDescriptor_t filterDesc,
int nbDims,
int tensorOuputDimA[] );
/* Destroy an instance of convolution descriptor */
cudnnStatus_t cudnnDestroyConvolutionDescriptor(
cudnnConvolutionDescriptor_t convDesc );
/* helper function to provide the convolution algo that fit best the requirement */
typedef enum
{
CUDNN_CONVOLUTION_FWD_NO_WORKSPACE = 0,
CUDNN_CONVOLUTION_FWD_PREFER_FASTEST = 1,
CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT = 2,
} cudnnConvolutionFwdPreference_t;
typedef enum
{
CUDNN_CONVOLUTION_FWD_ALGO_IMPLICIT_GEMM = 0,
CUDNN_CONVOLUTION_FWD_ALGO_IMPLICIT_PRECOMP_GEMM = 1,
CUDNN_CONVOLUTION_FWD_ALGO_GEMM = 2,
CUDNN_CONVOLUTION_FWD_ALGO_DIRECT = 3,
CUDNN_CONVOLUTION_FWD_ALGO_FFT = 4,
CUDNN_CONVOLUTION_FWD_ALGO_FFT_TILING = 5,
CUDNN_CONVOLUTION_FWD_ALGO_WINOGRAD = 6,
CUDNN_CONVOLUTION_FWD_ALGO_WINOGRAD_NONFUSED = 7
} cudnnConvolutionFwdAlgo_t;
typedef struct {
cudnnConvolutionFwdAlgo_t algo;
cudnnStatus_t status;
float time;
size_t memory;
} cudnnConvolutionFwdAlgoPerf_t;
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 );
cudnnStatus_t cudnnFindConvolutionForwardAlgorithmEx(
cudnnHandle_t handle,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const cudnnFilterDescriptor_t wDesc,
const void *w,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t yDesc,
void *y,
const int requestedAlgoCount,
int *returnedAlgoCount,
cudnnConvolutionFwdAlgoPerf_t *perfResults,
void *workSpace,
size_t workSpaceSizeInBytes );
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 );
/*
* convolution algorithm (which requires potentially some workspace)
*/
/* Helper function to return the minimum size of the workspace to be passed to the convolution given an algo*/
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 );
/* Convolution functions: All of the form "output = alpha * Op(inputs) + beta * output" */
/* Function to perform the forward pass for batch convolution */
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 );
/* Function to compute the bias gradient for batch convolution */
cudnnStatus_t cudnnConvolutionBackwardBias(
cudnnHandle_t handle,
const void *alpha,
const cudnnTensorDescriptor_t dyDesc,
const void *dy,
const void *beta,
const cudnnTensorDescriptor_t dbDesc,
void *db );
/* helper function to provide the convolution algo that fit best the requirement */
typedef enum
{
CUDNN_CONVOLUTION_BWD_FILTER_NO_WORKSPACE = 0,
CUDNN_CONVOLUTION_BWD_FILTER_PREFER_FASTEST = 1,
CUDNN_CONVOLUTION_BWD_FILTER_SPECIFY_WORKSPACE_LIMIT = 2,
} cudnnConvolutionBwdFilterPreference_t;
typedef enum
{
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_0 = 0, /* non-deterministic*/
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_1 = 1,
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_FFT = 2,
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_3 = 3, /* non-deterministic, algo0 with workspace*/
/* CUDNN_CONVOLUTION_BWD_FILTER_ALGO_WINOGRAD = 4, not implemented */
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_WINOGRAD_NONFUSED = 5
} cudnnConvolutionBwdFilterAlgo_t;
typedef struct {
cudnnConvolutionBwdFilterAlgo_t algo;
cudnnStatus_t status;
float time;
size_t memory;
} cudnnConvolutionBwdFilterAlgoPerf_t;
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 );
cudnnStatus_t cudnnFindConvolutionBackwardFilterAlgorithmEx(
cudnnHandle_t handle,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const cudnnTensorDescriptor_t dyDesc,
const void *y,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnFilterDescriptor_t dwDesc,
void *dw,
const int requestedAlgoCount,
int *returnedAlgoCount,
cudnnConvolutionBwdFilterAlgoPerf_t *perfResults,
void *workSpace,
size_t workSpaceSizeInBytes );
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 );
/*
* convolution algorithm (which requires potentially some workspace)
*/
/* Helper function to return the minimum size of the workspace to be passed to the convolution given an algo*/
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 );
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 );
/*********************************************************/
/* helper function to provide the convolution algo that fit best the requirement */
typedef enum
{
CUDNN_CONVOLUTION_BWD_DATA_NO_WORKSPACE = 0,
CUDNN_CONVOLUTION_BWD_DATA_PREFER_FASTEST = 1,
CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT = 2,
} cudnnConvolutionBwdDataPreference_t;
typedef enum
{
CUDNN_CONVOLUTION_BWD_DATA_ALGO_0 = 0, /* non-deterministic*/
CUDNN_CONVOLUTION_BWD_DATA_ALGO_1 = 1,
CUDNN_CONVOLUTION_BWD_DATA_ALGO_FFT = 2,
CUDNN_CONVOLUTION_BWD_DATA_ALGO_FFT_TILING = 3,
CUDNN_CONVOLUTION_BWD_DATA_ALGO_WINOGRAD = 4,
CUDNN_CONVOLUTION_BWD_DATA_ALGO_WINOGRAD_NONFUSED = 5
} cudnnConvolutionBwdDataAlgo_t;
typedef struct {
cudnnConvolutionBwdDataAlgo_t algo;
cudnnStatus_t status;
float time;
size_t memory;
} cudnnConvolutionBwdDataAlgoPerf_t;
cudnnStatus_t cudnnFindConvolutionBackwardDataAlgorithm(
cudnnHandle_t handle,
const cudnnFilterDescriptor_t wDesc,
const cudnnTensorDescriptor_t dyDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t dxDesc,
const int requestedAlgoCount,
int *returnedAlgoCount,
cudnnConvolutionBwdDataAlgoPerf_t *perfResults );
cudnnStatus_t cudnnFindConvolutionBackwardDataAlgorithmEx(
cudnnHandle_t handle,
const cudnnFilterDescriptor_t wDesc,
const void *w,
const cudnnTensorDescriptor_t dyDesc,
const void *dy,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t dxDesc,
void *dx,
const int requestedAlgoCount,
int *returnedAlgoCount,
cudnnConvolutionBwdDataAlgoPerf_t *perfResults,
void *workSpace,
size_t workSpaceSizeInBytes );
cudnnStatus_t cudnnGetConvolutionBackwardDataAlgorithm(
cudnnHandle_t handle,
const cudnnFilterDescriptor_t wDesc,
const cudnnTensorDescriptor_t dyDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t dxDesc,
cudnnConvolutionBwdDataPreference_t preference,
size_t memoryLimitInBytes,
cudnnConvolutionBwdDataAlgo_t *algo );
/* Helper function to return the minimum size of the workspace to be passed to the convolution given an algo*/
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 );
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 );
cudnnStatus_t cudnnIm2Col(
cudnnHandle_t handle,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const cudnnFilterDescriptor_t wDesc,
const cudnnConvolutionDescriptor_t convDesc,
void *colBuffer );
/*
* softmax algorithm
*/
typedef enum
{
CUDNN_SOFTMAX_FAST = 0, /* straightforward implementation */
CUDNN_SOFTMAX_ACCURATE = 1, /* subtract max from every point to avoid overflow */
CUDNN_SOFTMAX_LOG = 2
} cudnnSoftmaxAlgorithm_t;
typedef enum
{
CUDNN_SOFTMAX_MODE_INSTANCE = 0, /* compute the softmax over all C, H, W for each N */
CUDNN_SOFTMAX_MODE_CHANNEL = 1 /* compute the softmax over all C for each H, W, N */
} cudnnSoftmaxMode_t;
/* Softmax functions: All of the form "output = alpha * Op(inputs) + beta * output" */
/* Function to perform forward softmax */
cudnnStatus_t cudnnSoftmaxForward(
cudnnHandle_t handle,
cudnnSoftmaxAlgorithm_t algo,
cudnnSoftmaxMode_t mode,
const void *alpha,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *beta,
const cudnnTensorDescriptor_t yDesc,
void *y );
/* Function to perform backward softmax */
cudnnStatus_t cudnnSoftmaxBackward(
cudnnHandle_t handle,
cudnnSoftmaxAlgorithm_t algo,
cudnnSoftmaxMode_t mode,
const void *alpha,
const cudnnTensorDescriptor_t yDesc,
const void *y,
const cudnnTensorDescriptor_t dyDesc,
const void *dy,
const void *beta,
const cudnnTensorDescriptor_t dxDesc,
void *dx );
/*
* pooling mode
*/
typedef enum
{
CUDNN_POOLING_MAX = 0,
CUDNN_POOLING_AVERAGE_COUNT_INCLUDE_PADDING = 1, /* count for average includes padded values*/
CUDNN_POOLING_AVERAGE_COUNT_EXCLUDE_PADDING = 2, /* count for average does not include padded values*/
CUDNN_POOLING_AVERAGE = CUDNN_POOLING_AVERAGE_COUNT_INCLUDE_PADDING // for backward compatibility
} cudnnPoolingMode_t;
/* Create an instance of pooling descriptor */
cudnnStatus_t cudnnCreatePoolingDescriptor(
cudnnPoolingDescriptor_t *poolingDesc );
cudnnStatus_t cudnnSetPooling2dDescriptor(
cudnnPoolingDescriptor_t poolingDesc,
cudnnPoolingMode_t mode,
cudnnNanPropagation_t maxpoolingNanOpt,
int windowHeight,
int windowWidth,
int verticalPadding,
int horizontalPadding,
int verticalStride,
int horizontalStride );
cudnnStatus_t cudnnGetPooling2dDescriptor(
const cudnnPoolingDescriptor_t poolingDesc,
cudnnPoolingMode_t *mode,
cudnnNanPropagation_t *maxpoolingNanOpt,
int *windowHeight,
int *windowWidth,
int *verticalPadding,
int *horizontalPadding,
int *verticalStride,
int *horizontalStride );
cudnnStatus_t cudnnSetPoolingNdDescriptor(
cudnnPoolingDescriptor_t poolingDesc,
const cudnnPoolingMode_t mode,
const cudnnNanPropagation_t maxpoolingNanOpt,
int nbDims,
const int windowDimA[],
const int paddingA[],
const int strideA[] );
cudnnStatus_t cudnnGetPoolingNdDescriptor(
const cudnnPoolingDescriptor_t poolingDesc,
int nbDimsRequested,
cudnnPoolingMode_t *mode,
cudnnNanPropagation_t *maxpoolingNanOpt,
int *nbDims,
int windowDimA[],
int paddingA[],
int strideA[] );
cudnnStatus_t cudnnGetPoolingNdForwardOutputDim(
const cudnnPoolingDescriptor_t poolingDesc,
const cudnnTensorDescriptor_t inputTensorDesc,
int nbDims,
int outputTensorDimA[] );
cudnnStatus_t cudnnGetPooling2dForwardOutputDim(
const cudnnPoolingDescriptor_t poolingDesc,
const cudnnTensorDescriptor_t inputTensorDesc,
int *n,
int *c,
int *h,
int *w );
/* Destroy an instance of pooling descriptor */
cudnnStatus_t cudnnDestroyPoolingDescriptor(
cudnnPoolingDescriptor_t poolingDesc );
/* Pooling functions: All of the form "output = alpha * Op(inputs) + beta * output" */
/* Function to perform forward pooling */
cudnnStatus_t cudnnPoolingForward(
cudnnHandle_t handle,
const cudnnPoolingDescriptor_t poolingDesc,
const void *alpha,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *beta,
const cudnnTensorDescriptor_t yDesc,
void *y );
/* Function to perform backward pooling */
cudnnStatus_t cudnnPoolingBackward(
cudnnHandle_t handle,
const cudnnPoolingDescriptor_t poolingDesc,
const void *alpha,
const cudnnTensorDescriptor_t yDesc,
const void *y,
const cudnnTensorDescriptor_t dyDesc,
const void *dy,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *beta,
const cudnnTensorDescriptor_t dxDesc,
void *dx );
/*
* activation mode
*/
typedef enum
{
CUDNN_ACTIVATION_SIGMOID = 0,
CUDNN_ACTIVATION_RELU = 1,
CUDNN_ACTIVATION_TANH = 2,
CUDNN_ACTIVATION_CLIPPED_RELU = 3
} cudnnActivationMode_t;
/* Activation functions: All of the form "output = alpha * Op(inputs) + beta * output" */
cudnnStatus_t cudnnCreateActivationDescriptor(
cudnnActivationDescriptor_t *activationDesc);
cudnnStatus_t cudnnSetActivationDescriptor(
cudnnActivationDescriptor_t activationDesc,
cudnnActivationMode_t mode,
cudnnNanPropagation_t reluNanOpt,
double reluCeiling );
cudnnStatus_t cudnnGetActivationDescriptor(
const cudnnActivationDescriptor_t activationDesc,
cudnnActivationMode_t *mode,
cudnnNanPropagation_t *reluNanOpt,
double* reluCeiling );
cudnnStatus_t cudnnDestroyActivationDescriptor(
cudnnActivationDescriptor_t activationDesc);
/* Function to perform forward activation */
cudnnStatus_t cudnnActivationForward(
cudnnHandle_t handle,
cudnnActivationDescriptor_t activationDesc,
const void *alpha,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *beta,
const cudnnTensorDescriptor_t yDesc,
void *y );
/* Function to perform backward activation */
cudnnStatus_t cudnnActivationBackward(
cudnnHandle_t handle,
cudnnActivationDescriptor_t activationDesc,
const void *alpha,
const cudnnTensorDescriptor_t yDesc,
const void *y,
const cudnnTensorDescriptor_t dyDesc,
const void *dy,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *beta,
const cudnnTensorDescriptor_t dxDesc,
void *dx );
/*
* Create an instance of LRN (Local Response Normalization) descriptor
* Uses lrnN=5, lrnAlpha=1e-4, lrnBeta=0.75, lrnK=2.0 as defaults from Krizhevsky'12 ImageNet paper
*/
cudnnStatus_t cudnnCreateLRNDescriptor(
cudnnLRNDescriptor_t *normDesc );
typedef enum { CUDNN_LRN_MIN_N = 1, /* minimum allowed lrnN */
CUDNN_LRN_MAX_N = 16 } /* maximum allowed lrnN */
LRN_MinMaxFakeEnum;
/* static const float CUDNN_LRN_MIN_K = 1e-5; */ /* minimum allowed lrnK*/
/* static const float CUDNN_LRN_MIN_BETA = 0.01; */ /* minimum allowed lrnBeta*/
/* LRN layer mode */
typedef enum
{
CUDNN_LRN_CROSS_CHANNEL_DIM1 = 0,/* Normalize across tensor's dimA[1] dimension*/
} cudnnLRNMode_t;
/*
* Uses a window [center-lookBehind, center+lookAhead], where
* lookBehind = floor( (lrnN-1)/2 ), lookAhead = lrnN-lookBehind-1.
* Values of double parameters cast to tensor data type.
*/
cudnnStatus_t cudnnSetLRNDescriptor(
cudnnLRNDescriptor_t normDesc,
unsigned lrnN,
double lrnAlpha,
double lrnBeta,
double lrnK );
/*
* Retrieve the settings currently stored in an LRN layer descriptor
* Any of the provided pointers can be NULL (no corresponding value will be returned)
*/
cudnnStatus_t cudnnGetLRNDescriptor(
cudnnLRNDescriptor_t normDesc,
unsigned* lrnN,
double* lrnAlpha,
double* lrnBeta,
double* lrnK );
/* Destroy an instance of LRN descriptor */
cudnnStatus_t cudnnDestroyLRNDescriptor( cudnnLRNDescriptor_t lrnDesc );
/* LRN functions: output = alpha * normalize(x) + beta * old_y */
/* LRN cross-channel forward computation. Double parameters cast to tensor data type */
cudnnStatus_t cudnnLRNCrossChannelForward(
cudnnHandle_t handle,
cudnnLRNDescriptor_t normDesc,
cudnnLRNMode_t lrnMode,
const void* alpha,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *beta,
const cudnnTensorDescriptor_t yDesc,
void *y );
/* LRN cross-channel backward computation. Double parameters cast to tensor data type */
cudnnStatus_t cudnnLRNCrossChannelBackward(
cudnnHandle_t handle,
cudnnLRNDescriptor_t normDesc,
cudnnLRNMode_t lrnMode,
const void* alpha,
const cudnnTensorDescriptor_t yDesc,
const void *y,
const cudnnTensorDescriptor_t dyDesc,
const void *dy,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *beta,
const cudnnTensorDescriptor_t dxDesc,
void *dx);
typedef enum
{
CUDNN_DIVNORM_PRECOMPUTED_MEANS = 0,
} cudnnDivNormMode_t;
/* LCN/divisive normalization functions: y = alpha * normalize(x) + beta * y */
cudnnStatus_t cudnnDivisiveNormalizationForward(
cudnnHandle_t handle,
cudnnLRNDescriptor_t normDesc,
cudnnDivNormMode_t mode,
const void *alpha,
const cudnnTensorDescriptor_t xDesc, /* same desc for means, temp, temp2*/
const void *x,
const void *means, /* if NULL, means are assumed to be zero*/
void *temp,
void *temp2,
const void *beta,
const cudnnTensorDescriptor_t yDesc,
void *y );
cudnnStatus_t cudnnDivisiveNormalizationBackward(
cudnnHandle_t handle,
cudnnLRNDescriptor_t normDesc,
cudnnDivNormMode_t mode,
const void *alpha,
const cudnnTensorDescriptor_t xDesc, /* same desc for x, means, dy, temp, temp2*/
const void *x,
const void *means, /* if NULL, means are assumed to be zero*/
const void *dy,
void *temp,
void *temp2,
const void *beta,
const cudnnTensorDescriptor_t dXdMeansDesc, /* same desc for dx, dMeans*/
void *dx, /* output x differential*/
void *dMeans ); /* output means differential, can be NULL*/
typedef enum
{
/* bnScale, bnBias tensor dims are 1xCxHxWx.. (one value per CHW...-slice, normalized over N slice)*/
CUDNN_BATCHNORM_PER_ACTIVATION = 0,
/*bnScale, bnBias tensor dims are 1xCx1x1 (one value per C-dim normalized over Nx1xHxW subtensors)*/