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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
|
require 'cutorch'
local ffi = require 'ffi'
ffi.cdef[[
typedef enum
{
MAJOR_VERSION,
MINOR_VERSION,
PATCH_LEVEL
} libraryPropertyType;
typedef enum {
CUDNN_MAJOR = 7,
CUDNN_MINOR = 0,
CUDNN_PATCHLEVEL = 15,
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,
CUDNN_STATUS_RUNTIME_PREREQUISITE_MISSING = 11
} cudnnStatus_t;
/* human-readable error messages*/
const char * cudnnGetErrorString(cudnnStatus_t status);
cudnnStatus_t cudnnGetProperty(libraryPropertyType type, int *value);
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;
typedef struct cudnnReduceTensorStruct* cudnnReduceTensorDescriptor_t;
typedef struct cudnnCTCLossStruct* cudnnCTCLossDescriptor_t;
/*
* CUDNN data type
*/
typedef enum
{
CUDNN_DATA_FLOAT = 0,
CUDNN_DATA_DOUBLE = 1,
CUDNN_DATA_HALF = 2,
CUDNN_DATA_INT8 = 3,
CUDNN_DATA_INT32 = 4,
CUDNN_DATA_INT8x4 = 5
} cudnnDataType_t;
/*
* CUDNN math type
*/
typedef enum {
CUDNN_DEFAULT_MATH = 0,
CUDNN_TENSOR_OP_MATH = 1,
} cudnnMathType_t;
/*
* CUDNN propagate Nan
*/
typedef enum{
CUDNN_NOT_PROPAGATE_NAN = 0,
CUDNN_PROPAGATE_NAN = 1,
} cudnnNanPropagation_t;
/*
* CUDNN Determinism
*/
typedef enum {
CUDNN_NON_DETERMINISTIC = 0,
CUDNN_DETERMINISTIC = 1,
} cudnnDeterminism_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 )*/
CUDNN_TENSOR_NCHW_VECT_C = 2 /* each image point is vector of element of C : the length of the vector is carried by the data type*/
} 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 cudnnSetTensorNdDescriptorEx(
cudnnTensorDescriptor_t tensorDesc,
cudnnTensorFormat_t format,
cudnnDataType_t dataType,
int nbDims,
const int dimA[] );
cudnnStatus_t cudnnGetTensorNdDescriptor(
const cudnnTensorDescriptor_t tensorDesc,
int nbDimsRequested,
cudnnDataType_t *dataType,
int *nbDims,
int dimA[],
int strideA[] );
cudnnStatus_t cudnnGetTensorSizeInBytes(
const cudnnTensorDescriptor_t tensorDesc,
size_t *size);
/* 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,
CUDNN_OP_TENSOR_SQRT = 4,
CUDNN_OP_TENSOR_NOT = 5,
} 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 operation : C = op( alpha1 * A, alpha2 * B ) + beta * C */
/* B tensor is ignored for CUDNN_OP_TENSOR_SQRT, CUDNN_OP_TENSOR_NOT. */
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 );
/*
* CUDNN ReduceTensor op type
*/
typedef enum
{
CUDNN_REDUCE_TENSOR_ADD = 0,
CUDNN_REDUCE_TENSOR_MUL = 1,
CUDNN_REDUCE_TENSOR_MIN = 2,
CUDNN_REDUCE_TENSOR_MAX = 3,
CUDNN_REDUCE_TENSOR_AMAX = 4,
CUDNN_REDUCE_TENSOR_AVG = 5,
CUDNN_REDUCE_TENSOR_NORM1 = 6,
CUDNN_REDUCE_TENSOR_NORM2 = 7,
CUDNN_REDUCE_TENSOR_MUL_NO_ZEROS = 8,
} cudnnReduceTensorOp_t;
/*
* CUDNN ReduceTensor indices type
*/
typedef enum
{
CUDNN_REDUCE_TENSOR_NO_INDICES = 0,
CUDNN_REDUCE_TENSOR_FLATTENED_INDICES = 1,
} cudnnReduceTensorIndices_t;
/*
* CUDNN tensor indices type size (all unsigned)
* Currently not supported, default is 32 bit unsigned.
*/
typedef enum
{
CUDNN_32BIT_INDICES = 0,
CUDNN_64BIT_INDICES = 1,
CUDNN_16BIT_INDICES = 2,
CUDNN_8BIT_INDICES = 3,
} cudnnIndicesType_t;
cudnnStatus_t cudnnCreateReduceTensorDescriptor(
cudnnReduceTensorDescriptor_t *reduceTensorDesc );
cudnnStatus_t cudnnSetReduceTensorDescriptor(
cudnnReduceTensorDescriptor_t reduceTensorDesc,
cudnnReduceTensorOp_t reduceTensorOp,
cudnnDataType_t reduceTensorCompType,
cudnnNanPropagation_t reduceTensorNanOpt,
cudnnReduceTensorIndices_t reduceTensorIndices,
cudnnIndicesType_t reduceTensorIndicesType );
cudnnStatus_t cudnnGetReduceTensorDescriptor(
const cudnnReduceTensorDescriptor_t reduceTensorDesc,
cudnnReduceTensorOp_t *reduceTensorOp,
cudnnDataType_t *reduceTensorCompType,
cudnnNanPropagation_t *reduceTensorNanOpt,
cudnnReduceTensorIndices_t *reduceTensorIndices,
cudnnIndicesType_t *reduceTensorIndicesType );
cudnnStatus_t cudnnDestroyReduceTensorDescriptor(
cudnnReduceTensorDescriptor_t reduceTensorDesc );
/* Helper function to return the minimum size of the index space to be passed to the reduction given the input and output tensors */
cudnnStatus_t cudnnGetReductionIndicesSize(
cudnnHandle_t handle,
const cudnnReduceTensorDescriptor_t reduceTensorDesc,
const cudnnTensorDescriptor_t aDesc,
const cudnnTensorDescriptor_t cDesc,
size_t *sizeInBytes );
/* Helper function to return the minimum size of the workspace to be passed to the reduction given the input and output tensors */
cudnnStatus_t cudnnGetReductionWorkspaceSize(
cudnnHandle_t handle,
const cudnnReduceTensorDescriptor_t reduceTensorDesc,
const cudnnTensorDescriptor_t aDesc,
const cudnnTensorDescriptor_t cDesc,
size_t *sizeInBytes );
/* Tensor operation : C = reduce op( alpha * A ) + beta * C */
/* The NaN propagation enum applies to only the min and max reduce ops; the other reduce ops propagate NaN as usual. */
/* The indices space is ignored for reduce ops other than min or max. */
cudnnStatus_t cudnnReduceTensor(
cudnnHandle_t handle,
const cudnnReduceTensorDescriptor_t reduceTensorDesc,
void *indices,
size_t indicesSizeInBytes,
void *workspace,
size_t workspaceSizeInBytes,
const void *alpha,
const cudnnTensorDescriptor_t aDesc,
const void *A,
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 cudnnSetConvolutionMathType( cudnnConvolutionDescriptor_t convDesc,
cudnnMathType_t mathType );
cudnnStatus_t cudnnGetConvolutionMathType( cudnnConvolutionDescriptor_t convDesc,
cudnnMathType_t *mathType );
cudnnStatus_t cudnnSetConvolutionGroupCount( cudnnConvolutionDescriptor_t convDesc,
int groupCount );
cudnnStatus_t cudnnGetConvolutionGroupCount( cudnnConvolutionDescriptor_t convDesc,
int *groupCount );
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 dilation_h, /* filter dilation in the vertical dimension */
int dilation_w, /* filter dilation in the horizontal dimension */
cudnnConvolutionMode_t mode,
cudnnDataType_t computeType
);
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* dilation_h, // filter dilation in the vertical dimension
int* dilation_w, // filter dilation in the horizontal dimension
cudnnConvolutionMode_t* mode,
cudnnDataType_t *computeType
);
/* 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 dilationA[],
cudnnConvolutionMode_t mode,
cudnnDataType_t computeType ); // convolution data type
cudnnStatus_t cudnnGetConvolutionNdDescriptor(
const cudnnConvolutionDescriptor_t convDesc,
int arrayLengthRequested,
int *arrayLength,
int padA[],
int strideA[],
int dilationA[],
cudnnConvolutionMode_t *mode,
cudnnDataType_t *computeType ); // 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,
CUDNN_CONVOLUTION_FWD_ALGO_COUNT = 8,
} cudnnConvolutionFwdAlgo_t;
typedef struct {
cudnnConvolutionFwdAlgo_t algo;
cudnnStatus_t status;
float time;
size_t memory;
cudnnDeterminism_t determinism;
cudnnMathType_t mathType;
int reserved[3];
} cudnnConvolutionFwdAlgoPerf_t;
cudnnStatus_t cudnnGetConvolutionForwardAlgorithmMaxCount( cudnnHandle_t handle,
int *count);
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 );
cudnnStatus_t cudnnGetConvolutionForwardAlgorithm_v7(
cudnnHandle_t handle,
const cudnnTensorDescriptor_t srcDesc,
const cudnnFilterDescriptor_t filterDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t destDesc,
const int requestedAlgoCount,
int *returnedAlgoCount,
cudnnConvolutionFwdAlgoPerf_t *perfResults);
/*
* 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 );
/* Fused conv/bias/activation operation : y = Act( alpha1 * conv(x) + alpha2 * z + bias ) */
cudnnStatus_t cudnnConvolutionBiasActivationForward(
cudnnHandle_t handle,
const void *alpha1,
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 *alpha2,
const cudnnTensorDescriptor_t zDesc,
const void *z,
const cudnnTensorDescriptor_t biasDesc,
const void *bias,
const cudnnActivationDescriptor_t activationDesc,
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 */
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_WINOGRAD = 4, /* not implemented */
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_WINOGRAD_NONFUSED = 5,
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_FFT_TILING = 6,
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_COUNT = 7
} cudnnConvolutionBwdFilterAlgo_t;
typedef struct {
cudnnConvolutionBwdFilterAlgo_t algo;
cudnnStatus_t status;
float time;
size_t memory;
cudnnDeterminism_t determinism;
cudnnMathType_t mathType;
int reserved[3];
} cudnnConvolutionBwdFilterAlgoPerf_t;
cudnnStatus_t cudnnGetConvolutionBackwardFilterAlgorithmMaxCount( cudnnHandle_t handle,
int *count);
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 );
cudnnStatus_t cudnnGetConvolutionBackwardFilterAlgorithm_v7(
cudnnHandle_t handle,
const cudnnTensorDescriptor_t srcDesc,
const cudnnTensorDescriptor_t diffDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnFilterDescriptor_t gradDesc,
const int requestedAlgoCount,
int *returnedAlgoCount,
cudnnConvolutionBwdFilterAlgoPerf_t *perfResults);
/*
* 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,
CUDNN_CONVOLUTION_BWD_DATA_ALGO_COUNT = 6
} cudnnConvolutionBwdDataAlgo_t;
typedef struct {
cudnnConvolutionBwdDataAlgo_t algo;
cudnnStatus_t status;
float time;
size_t memory;
cudnnDeterminism_t determinism;
cudnnMathType_t mathType;
int reserved[3];
} cudnnConvolutionBwdDataAlgoPerf_t;
cudnnStatus_t cudnnGetConvolutionBackwardDataAlgorithmMaxCount( cudnnHandle_t handle,
int *count);
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 );
cudnnStatus_t cudnnGetConvolutionBackwardDataAlgorithm_v7(
cudnnHandle_t handle,
const cudnnFilterDescriptor_t filterDesc,
const cudnnTensorDescriptor_t diffDesc,
const cudnnConvolutionDescriptor_t convDesc,
const cudnnTensorDescriptor_t gradDesc,
const int requestedAlgoCount,
int *returnedAlgoCount,
cudnnConvolutionBwdDataAlgoPerf_t *perfResults);
/* 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,
CUDNN_POOLING_MAX_DETERMINISTIC = 3
} 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,
CUDNN_ACTIVATION_ELU = 4
} 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 coef ); /* ceiling for clipped RELU, alpha for ELU */
cudnnStatus_t cudnnGetActivationDescriptor(
const cudnnActivationDescriptor_t activationDesc,
cudnnActivationMode_t *mode,
cudnnNanPropagation_t *reluNanOpt,
double* coef ); /* ceiling for clipped RELU, alpha for ELU */
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) */
CUDNN_BATCHNORM_SPATIAL = 1,
/*
* bnScale, bnBias tensor dims are 1xCx1x1 (one value per C-dim normalized over Nx1xHxW subtensors).
* May be faster than CUDNN_BATCHNORM_SPATIAL but imposes some limits on the range of values
*/
CUDNN_BATCHNORM_SPATIAL_PERSISTENT = 2,
} cudnnBatchNormMode_t;
/* static const float CUDNN_BN_MIN_EPSILON = 1e-5; */ /* Minimum epsilon allowed to be used in the Batch Normalization formula*/
/*
* Derives a tensor descriptor from layer data descriptor for BatchNormalization
* scale, invVariance, bnBias, bnScale tensors. Use this tensor desc for
* bnScaleBiasMeanVarDesc and bnScaleBiasDiffDesc in Batch Normalization forward and backward functions.
*/
cudnnStatus_t cudnnDeriveBNTensorDescriptor(
cudnnTensorDescriptor_t derivedBnDesc,
const cudnnTensorDescriptor_t xDesc,
cudnnBatchNormMode_t mode );
/* Computes y = BN(x). Also accumulates moving averages of mean and inverse variances */
cudnnStatus_t cudnnBatchNormalizationForwardTraining(
cudnnHandle_t handle,
cudnnBatchNormMode_t mode,
const void *alpha, /* alpha[0] = result blend factor*/
const void *beta, /* beta[0] = dest layer blend factor*/
const cudnnTensorDescriptor_t xDesc,
const void *x, /* NxCxHxW*/
const cudnnTensorDescriptor_t yDesc,
void *y, /* NxCxHxW*/
/* Shared desc for the next 6 tensors in the argument list.
Data type to be set as follows:
type = (typeOf(x) == double) ? double : float
Dimensions for this descriptor depend on normalization mode
- Spatial Normalization : tensors are expected to have dims 1xCx1x1
(normalization is performed across NxHxW)
- Per-Activation Normalization : tensors are expected to have dims of 1xCxHxW
(normalization is performed across N) */
const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc,
/* 'Gamma' and 'Beta' respectively in Ioffe and Szegedy's paper's notation*/
const void *bnScale,
const void *bnBias,
/* MUST use factor=1 in the very first call of a complete training cycle.
Use a factor=1/(1+n) at N-th call to the function to get
Cumulative Moving Average (CMA) behavior
CMA[n] = (x[1]+...+x[n])/n
Since CMA[n+1] = (n*CMA[n]+x[n+1])/(n+1) =
((n+1)*CMA[n]-CMA[n])/(n+1) + x[n+1]/(n+1) =
CMA[n]*(1-1/(n+1)) + x[n+1]*1/(n+1) */
double exponentialAverageFactor,
/* Used in Training phase only.
runningMean = newMean*factor + runningMean*(1-factor) */
void *resultRunningMean,
/* Output in training mode, input in inference. Is the moving average
of variance[x] (factor is applied in the same way as for runningMean) */
void *resultRunningVariance,
/* Has to be >= CUDNN_BN_MIN_EPSILON. Should be the same in forward and backward functions. */
double epsilon,
/* Optionally save intermediate results from the forward pass here
- can be reused to speed up backward pass. NULL if unused */
void *resultSaveMean,
void *resultSaveInvVariance );
/*
* Performs Batch Normalization during Inference:
* y[i] = bnScale[k]*(x[i]-estimatedMean[k])/sqrt(epsilon+estimatedVariance[k]) + bnBias[k]
* with bnScale, bnBias, runningMean, runningInvVariance tensors indexed
* according to spatial or per-activation mode. Refer to cudnnBatchNormalizationForwardTraining
* above for notes on function arguments.
*/
cudnnStatus_t cudnnBatchNormalizationForwardInference(
cudnnHandle_t handle,
cudnnBatchNormMode_t mode,
const void *alpha, /* alpha[0] = result blend factor*/
const void *beta, /* beta[0] = dest layer blend factor*/
const cudnnTensorDescriptor_t xDesc,
const void *x, /* NxCxHxW*/
const cudnnTensorDescriptor_t yDesc,
void *y, /* NxCxHxW*/
const cudnnTensorDescriptor_t bnScaleBiasMeanVarDesc,
const void *bnScale,
const void *bnBias,
const void *estimatedMean,
const void *estimatedVariance,
double epsilon );
/* Performs backward pass of Batch Normalization layer. Returns x gradient,
* bnScale gradient and bnBias gradient */
cudnnStatus_t cudnnBatchNormalizationBackward(
cudnnHandle_t handle,
cudnnBatchNormMode_t mode,
const void *alphaDataDiff,
const void *betaDataDiff,
const void *alphaParamDiff,
const void *betaParamDiff,
const cudnnTensorDescriptor_t xDesc, /* same desc for x, dx, dy*/
const void *x,
const cudnnTensorDescriptor_t dyDesc,
const void *dy,
const cudnnTensorDescriptor_t dxDesc,
void *dx,
/* Shared tensor desc for the 4 tensors below */
const cudnnTensorDescriptor_t dBnScaleBiasDesc,
const void *bnScale, /* bnBias doesn't affect backpropagation*/
/* scale and bias diff are not backpropagated below this layer */
void *dBnScaleResult,
void *dBnBiasResult,
/* Same epsilon as forward pass */
double epsilon,
/* Optionally cached intermediate results from
forward pass */
const void *savedMean,
const void *savedInvVariance );
/* APIs for spatial transformer network*/
typedef enum {
CUDNN_SAMPLER_BILINEAR=0,
} cudnnSamplerType_t;
cudnnStatus_t cudnnCreateSpatialTransformerDescriptor(
cudnnSpatialTransformerDescriptor_t *stDesc);
cudnnStatus_t cudnnSetSpatialTransformerNdDescriptor(
cudnnSpatialTransformerDescriptor_t stDesc,
cudnnSamplerType_t samplerType,
cudnnDataType_t dataType,
const int nbDims,
const int dimA[]);
cudnnStatus_t cudnnDestroySpatialTransformerDescriptor(
cudnnSpatialTransformerDescriptor_t stDesc);
cudnnStatus_t cudnnSpatialTfGridGeneratorForward(
cudnnHandle_t handle,
const cudnnSpatialTransformerDescriptor_t stDesc,
const void *theta,
void *grid);
cudnnStatus_t cudnnSpatialTfGridGeneratorBackward(
cudnnHandle_t handle,
const cudnnSpatialTransformerDescriptor_t stDesc,
const void *dgrid,
void *dtheta);
cudnnStatus_t cudnnSpatialTfSamplerForward(
cudnnHandle_t handle,
cudnnSpatialTransformerDescriptor_t stDesc,
const void *alpha,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *grid,
const void *beta,
cudnnTensorDescriptor_t yDesc,
void *y);
cudnnStatus_t cudnnSpatialTfSamplerBackward(
cudnnHandle_t handle,
cudnnSpatialTransformerDescriptor_t stDesc,
const void *alpha,
const cudnnTensorDescriptor_t xDesc,
const void *x,
const void *beta,
const cudnnTensorDescriptor_t dxDesc,
void *dx,
const void *alphaDgrid,
const cudnnTensorDescriptor_t dyDesc,
const void *dy,
const void *grid,
const void *betaDgrid,
void *dgrid);
typedef struct cudnnDropoutStruct * cudnnDropoutDescriptor_t;
cudnnStatus_t cudnnCreateDropoutDescriptor(cudnnDropoutDescriptor_t * dropoutDesc);
cudnnStatus_t cudnnDestroyDropoutDescriptor(cudnnDropoutDescriptor_t dropoutDesc);
/*helper function to determine size of the states to be passed to cudnnSetDropoutDescriptor */
cudnnStatus_t cudnnDropoutGetStatesSize(cudnnHandle_t handle, size_t * sizeInBytes);
/*helper function to determine size of the reserve space to be passed to dropout forward/backward calls */
cudnnStatus_t cudnnDropoutGetReserveSpaceSize(cudnnTensorDescriptor_t xdesc, size_t * sizeInBytes);
cudnnStatus_t cudnnSetDropoutDescriptor(cudnnDropoutDescriptor_t dropoutDesc,
cudnnHandle_t handle,
float dropout,
void * states,
size_t stateSizeInBytes,
unsigned long long seed);
// Restores the dropout descriptor to a previously saved-off state
cudnnStatus_t cudnnRestoreDropoutDescriptor(cudnnDropoutDescriptor_t dropoutDesc,
cudnnHandle_t handle,
float dropout,
void * states,
size_t stateSizeInBytes,
unsigned long long seed);
cudnnStatus_t cudnnGetDropoutDescriptor(cudnnDropoutDescriptor_t dropoutDesc,
cudnnHandle_t handle,
float * dropout,
void ** states,
unsigned long long * seed);
cudnnStatus_t cudnnDropoutForward(cudnnHandle_t handle,
const cudnnDropoutDescriptor_t dropoutDesc,
const cudnnTensorDescriptor_t xdesc,
const void * x,
const cudnnTensorDescriptor_t ydesc,
void * y,
void * reserveSpace,
size_t reserveSpaceSizeInBytes);
cudnnStatus_t cudnnDropoutBackward(cudnnHandle_t handle,
const cudnnDropoutDescriptor_t dropoutDesc,
const cudnnTensorDescriptor_t dydesc,
const void * dy,
const cudnnTensorDescriptor_t dxdesc,
void * dx,
void * reserveSpace,
size_t reserveSpaceSizeInBytes);
/* RNN API */
typedef enum
{
CUDNN_RNN_RELU = 0, /* Stock RNN with ReLu activation*/
CUDNN_RNN_TANH = 1, /* Stock RNN with tanh activation*/
CUDNN_LSTM = 2, /* LSTM with no peephole connections*/
CUDNN_GRU = 3 /* Using h' = tanh(r * Uh(t-1) + Wx) and h = (1 - z) * h' + z * h(t-1);*/
} cudnnRNNMode_t;
typedef enum
{
CUDNN_UNIDIRECTIONAL = 0,
CUDNN_BIDIRECTIONAL = 1 /* Using output concatination at each step. Do we also want to support output sum?*/
} cudnnDirectionMode_t;
typedef enum
{
CUDNN_LINEAR_INPUT = 0,
CUDNN_SKIP_INPUT = 1
} cudnnRNNInputMode_t;
typedef enum
{
CUDNN_RNN_ALGO_STANDARD = 0,
CUDNN_RNN_ALGO_PERSIST_STATIC = 1,
CUDNN_RNN_ALGO_PERSIST_DYNAMIC = 2
} cudnnRNNAlgo_t;
struct cudnnRNNStruct;
typedef struct cudnnRNNStruct* cudnnRNNDescriptor_t;
cudnnStatus_t cudnnCreateRNNDescriptor(cudnnRNNDescriptor_t * rnnDesc);
cudnnStatus_t cudnnDestroyRNNDescriptor(cudnnRNNDescriptor_t rnnDesc);
struct cudnnPersistentRNNPlan;
typedef struct cudnnPersistentRNNPlan *cudnnPersistentRNNPlan_t;
/* Expensive. Creates the plan for the specific settings. */
cudnnStatus_t cudnnCreatePersistentRNNPlan(cudnnRNNDescriptor_t rnnDesc,
const int minibatch,
const cudnnDataType_t dataType,
cudnnPersistentRNNPlan_t * plan);
/* Attaches the plan to the descriptor. */
cudnnStatus_t cudnnSetPersistentRNNPlan(cudnnRNNDescriptor_t rnnDesc,
cudnnPersistentRNNPlan_t plan);
cudnnStatus_t cudnnDestroyPersistentRNNPlan(cudnnPersistentRNNPlan_t plan);
cudnnStatus_t cudnnSetRNNDescriptor(cudnnHandle_t handle,
cudnnRNNDescriptor_t rnnDesc,
const int hiddenSize,
const int numLayers,
cudnnDropoutDescriptor_t dropoutDesc, /* Between layers, not between recurrent steps. */
cudnnRNNInputMode_t inputMode,
cudnnDirectionMode_t direction,
cudnnRNNMode_t mode,
cudnnRNNAlgo_t algo,
cudnnDataType_t dataType);
cudnnStatus_t cudnnGetRNNDescriptor(cudnnHandle_t cudnnHandle,
cudnnRNNDescriptor_t rnnDesc,
int * hiddenSize,
int * numLayers,
cudnnDropoutDescriptor_t * dropoutDesc,
cudnnRNNInputMode_t * inputMode,
cudnnDirectionMode_t * direction,
cudnnRNNMode_t * mode,
cudnnRNNAlgo_t * algo,
cudnnDataType_t * dataType);
cudnnStatus_t cudnnSetRNNMatrixMathType (cudnnRNNDescriptor_t desc, cudnnMathType_t math);
/* dataType in the RNN descriptor is used to determine math precision */
/* dataType in weight descriptors and input descriptors is used to describe storage */
cudnnStatus_t cudnnGetRNNWorkspaceSize( cudnnHandle_t handle,
const cudnnRNNDescriptor_t rnnDesc,
const int seqLength,
const cudnnTensorDescriptor_t *xDesc,
size_t *sizeInBytes);
cudnnStatus_t cudnnGetRNNTrainingReserveSize( cudnnHandle_t handle,
const cudnnRNNDescriptor_t rnnDesc,
const int seqLength,
const cudnnTensorDescriptor_t *xDesc,
size_t *sizeInBytes);
cudnnStatus_t cudnnGetRNNParamsSize( cudnnHandle_t handle,
const cudnnRNNDescriptor_t rnnDesc,
const cudnnTensorDescriptor_t xDesc,
size_t *sizeInBytes,
cudnnDataType_t dataType);
cudnnStatus_t cudnnGetRNNLinLayerMatrixParams( cudnnHandle_t handle,
const cudnnRNNDescriptor_t rnnDesc,
const int layer,
const cudnnTensorDescriptor_t xDesc,
const cudnnFilterDescriptor_t wDesc,
const void * w,
const int linLayerID,
cudnnFilterDescriptor_t linLayerMatDesc,
void ** linLayerMat);
cudnnStatus_t cudnnGetRNNLinLayerBiasParams( cudnnHandle_t handle,
const cudnnRNNDescriptor_t rnnDesc,
const int layer,
const cudnnTensorDescriptor_t xDesc,
const cudnnFilterDescriptor_t wDesc,
const void * w,
const int linLayerID,
cudnnFilterDescriptor_t linLayerBiasDesc,
void ** linLayerBias);
cudnnStatus_t cudnnRNNForwardInference( cudnnHandle_t handle,
const cudnnRNNDescriptor_t rnnDesc,
const int seqLength,
const cudnnTensorDescriptor_t * xDesc,
const void * x,
const cudnnTensorDescriptor_t hxDesc,
const void * hx,
const cudnnTensorDescriptor_t cxDesc,
const void * cx,
const cudnnFilterDescriptor_t wDesc,
const void * w,
const cudnnTensorDescriptor_t *yDesc,
void * y,
const cudnnTensorDescriptor_t hyDesc,
void * hy,
const cudnnTensorDescriptor_t cyDesc,
void * cy,
void * workspace,
size_t workSpaceSizeInBytes);
cudnnStatus_t cudnnRNNForwardTraining( cudnnHandle_t handle,
const cudnnRNNDescriptor_t rnnDesc,
const int seqLength,
const cudnnTensorDescriptor_t *xDesc,
const void * x,
const cudnnTensorDescriptor_t hxDesc,
const void * hx,
const cudnnTensorDescriptor_t cxDesc,
const void * cx,
const cudnnFilterDescriptor_t wDesc,
const void * w,
const cudnnTensorDescriptor_t *yDesc,
void * y,
const cudnnTensorDescriptor_t hyDesc,
void * hy,
const cudnnTensorDescriptor_t cyDesc,
void * cy,
void * workspace,
size_t workSpaceSizeInBytes,
void * reserveSpace,
size_t reserveSpaceSizeInBytes);
cudnnStatus_t cudnnRNNBackwardData( cudnnHandle_t handle,
const cudnnRNNDescriptor_t rnnDesc,
const int seqLength,
const cudnnTensorDescriptor_t * yDesc,
const void * y,
const cudnnTensorDescriptor_t * dyDesc,
const void * dy,
const cudnnTensorDescriptor_t dhyDesc,
const void * dhy,
const cudnnTensorDescriptor_t dcyDesc,
const void * dcy,
const cudnnFilterDescriptor_t wDesc,
const void * w,
const cudnnTensorDescriptor_t hxDesc,
const void * hx,
const cudnnTensorDescriptor_t cxDesc,
const void * cx,
const cudnnTensorDescriptor_t * dxDesc,
void * dx,
const cudnnTensorDescriptor_t dhxDesc,
void * dhx,
const cudnnTensorDescriptor_t dcxDesc,
void * dcx,
void * workspace,
size_t workSpaceSizeInBytes,
void * reserveSpace,
size_t reserveSpaceSizeInBytes );
cudnnStatus_t cudnnRNNBackwardWeights( cudnnHandle_t handle,
const cudnnRNNDescriptor_t rnnDesc,
const int seqLength,
const cudnnTensorDescriptor_t * xDesc,
const void * x,
const cudnnTensorDescriptor_t hxDesc,
const void * hx,
const cudnnTensorDescriptor_t * yDesc,
const void * y,
const void * workspace,
size_t workSpaceSizeInBytes,
const cudnnFilterDescriptor_t dwDesc,
void * dw,
const void * reserveSpace,
size_t reserveSpaceSizeInBytes );
typedef enum
{
CUDNN_CTC_LOSS_ALGO_DETERMINISTIC = 0,
CUDNN_CTC_LOSS_ALGO_NON_DETERMINISTIC = 1
}cudnnCTCLossAlgo_t;
/*
* Create an instance of a CTC (Connectionist Temporal Classification) loss descriptor
*/
cudnnStatus_t cudnnCreateCTCLossDescriptor( cudnnCTCLossDescriptor_t* ctcLossDesc );
cudnnStatus_t cudnnSetCTCLossDescriptor(
cudnnCTCLossDescriptor_t ctcLossDesc,
cudnnDataType_t compType );
cudnnStatus_t cudnnGetCTCLossDescriptor(
cudnnCTCLossDescriptor_t ctcLossDesc,
cudnnDataType_t* compType );
cudnnStatus_t cudnnDestroyCTCLossDescriptor( cudnnCTCLossDescriptor_t ctcLossDesc );
/* return the ctc costs and gradients, given the probabilities and labels */
cudnnStatus_t cudnnCTCLoss( cudnnHandle_t handle,
const cudnnTensorDescriptor_t probsDesc, /* Tensor descriptor for probabilities, the dimensions are T,N,A (T is the timing steps, N is the mini batch size, A is the alphabet size) */
const void * probs, /* probabilities after softmax, in GPU memory */
const int * labels, /* labels, in CPU memory */
const int * labelLengths, /* the length of each label, in CPU memory */
const int * inputLengths, /* the lengths of timing steps in each batch, in CPU memory */
void * costs, /* the returned costs of CTC, in GPU memory */
const cudnnTensorDescriptor_t gradientsDesc, /* Tensor descriptor for gradients, the dimensions are T,N,A */
const void * gradients, /* the returned CTC gradients, in GPU memory, to compute costs only, set it to NULL */
cudnnCTCLossAlgo_t algo, /* algorithm selected, supported now 0 and 1 */
cudnnCTCLossDescriptor_t ctcLossDesc,
void * workspace, /* pointer to the workspace, in GPU memory */
size_t workSpaceSizeInBytes); /* the workspace size needed */
/* return the workspace size needed for ctc */
cudnnStatus_t cudnnGetCTCLossWorkspaceSize(
cudnnHandle_t handle,
const cudnnTensorDescriptor_t probsDesc, /* Tensor descriptor for probabilities, the dimensions are T,N,A (T is the timing steps, N is the mini batch size, A is the alphabet size) */
const cudnnTensorDescriptor_t gradientsDesc, /* Tensor descriptor for gradients, the dimensions are T,N,A. To compute costs only, set it to NULL */
const int * labels, /* labels, in CPU memory */
const int * labelLengths, /* the length of each label, in CPU memory */
const int * inputLengths, /* the lengths of timing steps in each batch, in CPU memory */
cudnnCTCLossAlgo_t algo, /* algorithm selected, supported now 0 and 1 */
cudnnCTCLossDescriptor_t ctcLossDesc,
size_t *sizeInBytes ); /* pointer to the returned workspace size */
/* DEPRECATED routines to be removed next release :
User should use the non-suffixed version (which has the API and functionality of _v6 version)
Routines with _v5 suffix has the functionality of the non-suffixed routines in the CUDNN V6
*/
cudnnStatus_t cudnnSetRNNDescriptor_v6(cudnnHandle_t handle,
cudnnRNNDescriptor_t rnnDesc,
const int hiddenSize,
const int numLayers,
cudnnDropoutDescriptor_t dropoutDesc, /* Between layers, not between recurrent steps. */
cudnnRNNInputMode_t inputMode,
cudnnDirectionMode_t direction,
cudnnRNNMode_t mode,
cudnnRNNAlgo_t algo,
cudnnDataType_t dataType);
cudnnStatus_t cudnnSetRNNDescriptor_v5(cudnnRNNDescriptor_t rnnDesc,
int hiddenSize,
int numLayers,
cudnnDropoutDescriptor_t dropoutDesc, /* Between layers, not between recurrent steps. */
cudnnRNNInputMode_t inputMode,
cudnnDirectionMode_t direction,
cudnnRNNMode_t mode,
cudnnDataType_t dataType);
cudnnStatus_t cudnnGetConvolution2dDescriptor_v4(
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 *dilation_h, // filter dilation in the vertical dimension
int *dilation_w, // filter dilation in the horizontal dimension
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* dilation_h, // filter dilation in the vertical dimension
int* dilation_w, // filter dilation in the horizontal dimension
cudnnConvolutionMode_t* mode,
cudnnDataType_t *computeType
);
]]
local CUDNN_PATH = os.getenv('CUDNN_PATH')
if CUDNN_PATH then
io.stderr:write('Found Environment variable CUDNN_PATH = ' .. CUDNN_PATH)
cudnn.C = ffi.load(CUDNN_PATH)
else
local libnames = {'libcudnn.so.7', 'libcudnn.7.dylib', 'cudnn64_6.dll'}
local ok = false
for i=1,#libnames do
ok = pcall(function () cudnn.C = ffi.load(libnames[i]) end)
if ok then break; end
end
if not ok then
error([['libcudnn (R7\) not found in library path.
Please install CuDNN from https://developer.nvidia.com/cuDNN
Then make sure files named as libcudnn.so.7 or libcudnn.7.dylib are placed in
your library load path (for example /usr/local/lib , or manually add a path to LD_LIBRARY_PATH)
Alternatively, set the path to libcudnn.so.7 or libcudnn.7.dylib
to the environment variable CUDNN_PATH and rerun torch.
For example: export CUDNN_PATH = "/usr/local/cuda/lib64/libcudnn.so.7"
]])
end
end
-- check cuDNN version
cudnn.version = tonumber(cudnn.C.cudnnGetVersion())
if cudnn.version < 7000 then
error('These bindings are for version 7000 or above, '
.. 'while the loaded CuDNN is version: ' .. cudnn.version
.. ' \nAre you using an older or newer version of CuDNN?')
end
-- check GPU driver version
local props = cutorch.getDeviceProperties(cutorch.getDevice())
if cutorch.driverVersion and -- for backward compatiblity
not(cutorch.driverVersion >= 7050 -- desktop GPUs
or (props.major == 5 and props.minor == 3 and cutorch.driverVersion >= 7000) ) -- Tegra X1
then
error('Insufficient GPU driver version.')
end
|