AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

[HIPIFY][DNN] Revise and sync with CUDNN 10 and hipDNN

Telusuri Sumber 
+ CUDNN_API_supported_by_HIP.md update
+ CUDA_Runtime_API_functions_supported_by_HIP.md update
+ libraryPropertyType_t is moved to Runtime
+ test update
This commit is contained in:
Evgeny Mankov
2019-02-13 19:40:59 +03:00
orang tua 31c4a0dff1
melakukan d26dc3cea7
5 mengubah file dengan 96 tambahan dan 18 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
docs/markdown/CUDA_Runtime_API_functions_supported_by_HIP.md
+5
Melihat File
@@ -1070,3 +1070,8 @@
| typedef |`cudaGraphicsResource_t` | |
| struct |`cudaGraphicsResource` | |
| typedef |`cudaHostFn_t` | |
| enum |***`libraryPropertyType`*** | |
| typedef |***`libraryPropertyType_t`*** | |
| 0 |*`MAJOR_VERSION`* | |
| 1 |*`MINOR_VERSION`* | |
| 2 |*`PATCH_LEVEL`* | |
docs/markdown/CUDNN_API_supported_by_HIP.md
+42 -9
Melihat File
@@ -27,7 +27,6 @@
| 0 |*`CUDNN_ERRQUERY_RAWCODE`* | |
| 1 |*`CUDNN_ERRQUERY_NONBLOCKING`* | |
| 2 |*`CUDNN_ERRQUERY_BLOCKING`* | |
| enum |***`libraryPropertyType_t`*** | |
| struct |`cudnnTensorStruct` | |
| struct* |`cudnnTensorDescriptor_t` |`hipdnnTensorDescriptor_t` |
| struct |`cudnnConvolutionStruct` | |
@@ -55,8 +54,9 @@
| 3 |*`CUDNN_DATA_INT8`* |*`HIPDNN_DATA_INT8`* |
| 4 |*`CUDNN_DATA_INT32`* |*`HIPDNN_DATA_INT32`* |
| 5 |*`CUDNN_DATA_INT8x4`* |*`HIPDNN_DATA_INT8x4`* |
| 6 |*`CUDNN_DATA_UINT8`* |*`HIPDNN_DATA_UINT8`* |
| 7 |*`CUDNN_DATA_UINT8x4`* |*`HIPDNN_DATA_UINT8x4`* |
| 6 |*`CUDNN_DATA_UINT8`* | |
| 7 |*`CUDNN_DATA_UINT8x4`* | |
| 8 |*`CUDNN_DATA_INT8x32`* | |
| enum |***`cudnnMathType_t`*** |***`hipdnnMathType_t`*** |
| 0 |*`CUDNN_DEFAULT_MATH`* |*`HIPDNN_DEFAULT_MATH`* |
| 1 |*`CUDNN_TENSOR_OP_MATH`* |*`HIPDNN_TENSOR_OP_MATH`* |
@@ -77,7 +77,7 @@
| 2 |*`CUDNN_OP_TENSOR_MIN`* |*`HIPDNN_OP_TENSOR_MIN`* |
| 3 |*`CUDNN_OP_TENSOR_MAX`* |*`HIPDNN_OP_TENSOR_MAX`* |
| 4 |*`CUDNN_OP_TENSOR_SQRT`* |*`HIPDNN_OP_TENSOR_SQRT`* |
| 5 |*`CUDNN_OP_TENSOR_NOT`* | |
| 5 |*`CUDNN_OP_TENSOR_NOT`* |*`HIPDNN_OP_TENSOR_NOT`* |
| enum |***`cudnnReduceTensorOp_t`*** |***`hipdnnReduceTensorOp_t`*** |
| 0 |*`CUDNN_REDUCE_TENSOR_ADD`* |*`HIPDNN_REDUCE_TENSOR_ADD`* |
| 1 |*`CUDNN_REDUCE_TENSOR_MUL`* |*`HIPDNN_REDUCE_TENSOR_MUL`* |
@@ -199,6 +199,22 @@
| define |`CUDNN_SEV_INFO_EN` | |
| struct |`cudnnDebug_t` | |
| struct |`cudnnCallback_t` | |
| enum |***`cudnnBatchNormOps_t`*** | |
| 0 |*`CUDNN_BATCHNORM_OPS_BN`* | |
| 1 |*`CUDNN_BATCHNORM_OPS_BN_ACTIVATION`* | |
| 2 |*`CUDNN_BATCHNORM_OPS_BN_ADD_ACTIVATION`* | |
| enum |***`cudnnRNNClipMode_t`*** | |
| 0 |*`CUDNN_RNN_CLIP_NONE`* | |
| 1 |*`CUDNN_RNN_CLIP_MINMAX`* | |
| struct |`cudnnRNNDataStruct` | |
| struct* |`cudnnRNNDataDescriptor_t` | |
| enum |***`cudnnRNNDataLayout_t`*** | |
| 0 |*`CUDNN_RNN_DATA_LAYOUT_SEQ_MAJOR_UNPACKED`* | |
| 1 |*`CUDNN_RNN_DATA_LAYOUT_SEQ_MAJOR_PACKED`* | |
| 2 |*`CUDNN_RNN_DATA_LAYOUT_BATCH_MAJOR_UNPACKED`* | |
| enum |***`cudnnRNNPaddingMode_t`*** | |
| 0 |*`CUDNN_RNN_PADDED_IO_DISABLED`* | |
| 1 |*`CUDNN_RNN_PADDED_IO_ENABLED`* | |
## **2. CUDNN API functions**
@@ -215,7 +231,7 @@
|`cudnnSetStream` |`hipdnnGetStream` |
|`cudnnCreateTensorDescriptor` |`hipdnnCreateTensorDescriptor` |
|`cudnnSetTensor4dDescriptor` |`hipdnnSetTensor4dDescriptor` |
|`cudnnSetTensor4dDescriptorEx` | |
|`cudnnSetTensor4dDescriptorEx` |`hipdnnSetTensor4dDescriptorEx` |
|`cudnnGetTensor4dDescriptor` |`hipdnnGetTensor4dDescriptor` |
|`cudnnSetTensorNdDescriptor` |`hipdnnSetTensorNdDescriptor` |
|`cudnnSetTensorNdDescriptorEx` | |
@@ -239,15 +255,15 @@
|`cudnnSetTensor` |`hipdnnSetTensor` |
|`cudnnScaleTensor` |`hipdnnScaleTensor` |
|`cudnnCreateFilterDescriptor` |`hipdnnCreateFilterDescriptor` |
|`cudnnSetFilter4dDescriptor` |`hipdnnSetFilter4dDescriptor` |
|`cudnnGetFilter4dDescriptor` |`hipdnnGetFilter4dDescriptor` |
|`cudnnSetFilter4dDescriptor` | |
|`cudnnGetFilter4dDescriptor` | |
|`cudnnSetFilterNdDescriptor` |`hipdnnSetFilterNdDescriptor` |
|`cudnnGetFilterNdDescriptor` |`hipdnnGetFilterNdDescriptor` |
|`cudnnDestroyFilterDescriptor` |`hipdnnDestroyFilterDescriptor` |
|`cudnnCreateConvolutionDescriptor` |`hipdnnCreateConvolutionDescriptor` |
|`cudnnSetConvolutionMathType` |`hipdnnSetConvolutionMathType` |
|`cudnnGetConvolutionMathType` | |
|`cudnnSetConvolutionGroupCount` | |
|`cudnnSetConvolutionGroupCount` |`hipdnnSetConvolutionGroupCount` |
|`cudnnGetConvolutionGroupCount` | |
|`cudnnSetConvolution2dDescriptor` |`hipdnnSetConvolution2dDescriptor` |
|`cudnnGetConvolution2dDescriptor` |`hipdnnGetConvolution2dDescriptor` |
@@ -308,8 +324,10 @@
|`cudnnDivisiveNormalizationBackward` | |
|`cudnnDeriveBNTensorDescriptor` |`hipdnnDeriveBNTensorDescriptor` |
|`cudnnBatchNormalizationForwardTraining` |`hipdnnBatchNormalizationForwardTraining` |
|`cudnnBatchNormalizationForwardTrainingEx` | |
|`cudnnBatchNormalizationForwardInference` |`hipdnnBatchNormalizationForwardInference` |
|`cudnnBatchNormalizationBackward` |`hipdnnBatchNormalizationBackward` |
|`cudnnBatchNormalizationBackwardEx` | |
|`cudnnCreateSpatialTransformerDescriptor` | |
|`cudnnSetSpatialTransformerNdDescriptor` | |
|`cudnnDestroySpatialTransformerDescriptor` | |
@@ -340,7 +358,7 @@
|`cudnnSetPersistentRNNPlan` |`hipdnnSetPersistentRNNPlan` |
|`cudnnDestroyPersistentRNNPlan` |`hipdnnDestroyPersistentRNNPlan` |
|`cudnnSetRNNDescriptor` |`hipdnnSetRNNDescriptor` |
|`cudnnGetRNNDescriptor` | |
|`cudnnGetRNNDescriptor` |`hipdnnGetRNNDescriptor` |
|`cudnnSetRNNProjectionLayers` | |
|`cudnnGetRNNProjectionLayers` | |
|`cudnnSetRNNAlgorithmDescriptor` | |
@@ -352,9 +370,19 @@
|`cudnnGetRNNLinLayerMatrixParams` |`hipdnnGetRNNLinLayerMatrixParams` |
|`cudnnGetRNNLinLayerBiasParams` |`hipdnnGetRNNLinLayerBiasParams` |
|`cudnnRNNForwardInference` |`hipdnnRNNForwardInference` |
|`cudnnRNNForwardInferenceEx` | |
|`cudnnRNNForwardTraining` |`hipdnnRNNForwardTraining` |
|`cudnnRNNForwardTrainingEx` | |
|`cudnnRNNBackwardData` |`hipdnnRNNBackwardData` |
|`cudnnRNNBackwardDataEx` | |
|`cudnnRNNBackwardWeights` |`hipdnnRNNBackwardWeights` |
|`cudnnRNNBackwardWeightsEx` | |
|`cudnnSetRNNPaddingMode` | |
|`cudnnGetRNNPaddingMode` | |
|`cudnnCreateRNNDataDescriptor` | |
|`cudnnDestroyRNNDataDescriptor` | |
|`cudnnSetRNNDataDescriptor` | |
|`cudnnGetRNNDataDescriptor` | |
|`cudnnCreateCTCLossDescriptor` | |
|`cudnnSetCTCLossDescriptor` | |
|`cudnnGetCTCLossDescriptor` | |
@@ -377,3 +405,8 @@
|`cudnnSetRNNDescriptor_v6` |`hipdnnSetRNNDescriptor_v6` |
|`cudnnSetCallback` | |
|`cudnnGetCallback` | |
|`cudnnGetBatchNormalizationForwardTrainingExWorkspaceSize` | |
|`cudnnGetBatchNormalizationBackwardExWorkspaceSize` | |
|`cudnnGetBatchNormalizationTrainingExReserveSpaceSize` | |
|`cudnnRNNSetClip` | |
|`cudnnRNNGetClip` | |