Skip to content

[PHI] Fix conv_transpose for cudnn big tensor #74096

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jul 18, 2025
+12 −6
Merged

[PHI] Fix conv_transpose for cudnn big tensor #74096

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 6 additions & 3 deletions paddle/phi/kernels/gpudnn/conv_transpose_grad_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -179,6 +179,12 @@ void ConvTransposeGradRawGPUDNNKernel(const Context& dev_ctx,
out_vec = common::vectorize<int>(transformed_dout.dims());

// ------------------- cudnn descriptors ---------------------
#ifndef PADDLE_WITH_HIP
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(transformed_dout);
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(filter);
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(x_transpose);
#endif

GPUDNNDataLayout layout;

if (strides.size() == 2U) {
Expand Down Expand Up @@ -223,9 +229,6 @@ void ConvTransposeGradRawGPUDNNKernel(const Context& dev_ctx,
SearchResult<miopenConvFwdAlgorithm_t> fwd_result;
SearchResult<miopenConvBwdWeightsAlgorithm_t> filter_result;
#else
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(transformed_dout);
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(filter);
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(x_transpose);
SearchResult<cudnnConvolutionFwdAlgo_t> fwd_result;
SearchResult<cudnnConvolutionBwdFilterAlgo_t> filter_result;
#endif
Expand Down
9 changes: 6 additions & 3 deletions paddle/phi/kernels/gpudnn/conv_transpose_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -179,6 +179,12 @@ void ConvTransposeRawGPUDNNKernel(const Context& dev_ctx,
}
T* transformed_out_data = transformed_out.data<T>();

#ifndef PADDLE_WITH_HIP
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(transformed_x);
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(filter);
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(transformed_out);
#endif

GPUDNNDataLayout layout;

int iwo_groups = groups;
Expand Down Expand Up @@ -235,9 +241,6 @@ void ConvTransposeRawGPUDNNKernel(const Context& dev_ctx,
bwd_result.algo =
search::Find<T>(args, false, deterministic, workspace_size, dev_ctx);
#else
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(transformed_x);
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(filter);
CUDNN_ENFORCE_TENSOR_SIZE_SUPPORTED(transformed_out);
SearchResult<cudnnConvolutionBwdDataAlgo_t> bwd_result;
using search = SearchAlgorithm<ConvKind::kBackwardData>;
bwd_result = search::Find<T>(dev_ctx, args, false, deterministic, false);
Expand Down