p04896132
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

!806 GPU update kernel bn 

Merge pull request !806 from VectorSL/gpu-fix-bn
This commit is contained in:
mindspore-ci-bot 2020-04-28 21:24:14 +08:00 committed by Gitee
parent e655e0e04c 25af911ed9
commit d5d3bf7565
2 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
mindspore/ccsrc/kernel/gpu/nn/fused_batch_norm_gpu_kernel.h
View File

@ -82,6 +82,7 @@ class FusedBatchNormGpuKernel : public GpuKernel {
}
bool Init(const CNodePtr &kernel_node) override {
InitResource();
cudnn_data_type_ = kCudnnDtypeMap[TypeIdLabel(AnfAlgo::GetInputDeviceDataType(kernel_node, 0))];
size_t input_num = AnfAlgo::GetInputTensorNum(kernel_node);
if (input_num != 5) {
MS_LOG(EXCEPTION) << "input tensor size is " << input_num << ", FusedBatchNormGpuKernel should be 5";
@ -112,11 +113,11 @@ class FusedBatchNormGpuKernel : public GpuKernel {
}
CHECK_CUDNN_RET_WITH_EXCEPT(
cudnnSetTensor4dDescriptor(x_desc_, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, batch_, channel_, height_, width_),
cudnnSetTensor4dDescriptor(x_desc_, CUDNN_TENSOR_NCHW, cudnn_data_type_, batch_, channel_, height_, width_),
"Set x desc failed");
CHECK_CUDNN_RET_WITH_EXCEPT(
cudnnSetTensor4dDescriptor(y_desc_, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, batch_, channel_, height_, width_),
cudnnSetTensor4dDescriptor(y_desc_, CUDNN_TENSOR_NCHW, cudnn_data_type_, batch_, channel_, height_, width_),
"Set y desc failed");
CHECK_CUDNN_RET_WITH_EXCEPT(

2
mindspore/ccsrc/kernel/gpu/nn/fused_batchnorm_grad_gpu_kernel.h
View File

@ -110,7 +110,7 @@ class FusedBatchNormGradGpuKernel : public GpuKernel {
cudnnSetTensor4dDescriptor(dx_desc_, CUDNN_TENSOR_NCHW, cudnn_data_type_, batch_, channel_, height_, width_),
"Set dx desc failed");
CHECK_CUDNN_RET_WITH_EXCEPT(
cudnnSetTensor4dDescriptor(scale_bias_desc_, CUDNN_TENSOR_NCHW, cudnn_data_type_, 1, channel_, 1, 1),
cudnnSetTensor4dDescriptor(scale_bias_desc_, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, 1, channel_, 1, 1),
"Set para desc failed");
InitSizeLists();