提交 af1568e6 编写于 作者: Y yukavio

fix generate_proposals and affine grid error info

上级 6b727e08
...@@ -26,8 +26,11 @@ template <typename T> ...@@ -26,8 +26,11 @@ template <typename T>
class CUDNNAffineGridOpKernel : public framework::OpKernel<T> { class CUDNNAffineGridOpKernel : public framework::OpKernel<T> {
public: public:
void Compute(const framework::ExecutionContext& ctx) const override { void Compute(const framework::ExecutionContext& ctx) const override {
PADDLE_ENFORCE(platform::is_gpu_place(ctx.GetPlace()), PADDLE_ENFORCE_EQ(platform::is_gpu_place(ctx.GetPlace()), true,
"It must use CUDAPlace."); platform::errors::InvalidArgument("Only "
"support for CUDAPlace.Please switch "
"your context from CPUPlace to "
"CUDAPlace or update your cudnn.");
auto& dev_ctx = ctx.template device_context<platform::CUDADeviceContext>(); auto& dev_ctx = ctx.template device_context<platform::CUDADeviceContext>();
auto handle = dev_ctx.cudnn_handle(); auto handle = dev_ctx.cudnn_handle();
auto* theta = ctx.Input<Tensor>("Theta"); auto* theta = ctx.Input<Tensor>("Theta");
...@@ -56,8 +59,10 @@ class CUDNNAffineGridOpKernel : public framework::OpKernel<T> { ...@@ -56,8 +59,10 @@ class CUDNNAffineGridOpKernel : public framework::OpKernel<T> {
cudnnSpatialTransformerDescriptor_t cudnn_st_desc = cudnnSpatialTransformerDescriptor_t cudnn_st_desc =
st_desc.descriptor<T>(4, h_size_data); st_desc.descriptor<T>(4, h_size_data);
PADDLE_ENFORCE(platform::dynload::cudnnSpatialTfGridGeneratorForward( PADDLE_ENFORCE_EQ(platform::dynload::cudnnSpatialTfGridGeneratorForward(
handle, cudnn_st_desc, theta_data, output_data)); handle, cudnn_st_desc, theta_data, output_data),
true, platform::errors::Fatal("Some errors has occurred "
"during forward computation in cudnn."));
} }
}; };
...@@ -65,8 +70,11 @@ template <typename T> ...@@ -65,8 +70,11 @@ template <typename T>
class CUDNNAffineGridGradOpKernel : public framework::OpKernel<T> { class CUDNNAffineGridGradOpKernel : public framework::OpKernel<T> {
public: public:
void Compute(const framework::ExecutionContext& ctx) const override { void Compute(const framework::ExecutionContext& ctx) const override {
PADDLE_ENFORCE(platform::is_gpu_place(ctx.GetPlace()), PADDLE_ENFORCE_EQ(platform::is_gpu_place(ctx.GetPlace()), true,
"It must use CUDAPlace."); platform::errors::InvalidArgument("Only "
"support for CUDAPlace. Please switch "
"your context from CPUPlace to "
"CUDAPlace or update your cudnn.");
auto& dev_ctx = ctx.template device_context<platform::CUDADeviceContext>(); auto& dev_ctx = ctx.template device_context<platform::CUDADeviceContext>();
auto handle = dev_ctx.cudnn_handle(); auto handle = dev_ctx.cudnn_handle();
auto output_grad = ctx.Input<Tensor>(framework::GradVarName("Output")); auto output_grad = ctx.Input<Tensor>(framework::GradVarName("Output"));
...@@ -95,8 +103,10 @@ class CUDNNAffineGridGradOpKernel : public framework::OpKernel<T> { ...@@ -95,8 +103,10 @@ class CUDNNAffineGridGradOpKernel : public framework::OpKernel<T> {
const T* output_grad_data = output_grad->data<T>(); const T* output_grad_data = output_grad->data<T>();
T* theta_grad_data = theta_grad->mutable_data<T>(ctx.GetPlace()); T* theta_grad_data = theta_grad->mutable_data<T>(ctx.GetPlace());
PADDLE_ENFORCE(platform::dynload::cudnnSpatialTfGridGeneratorBackward( PADDLE_ENFORCE_EQ(platform::dynload::cudnnSpatialTfGridGeneratorBackward(
handle, cudnn_st_desc, output_grad_data, theta_grad_data)); handle, cudnn_st_desc, output_grad_data, theta_grad_data),
true, "Some errors "
"has occurred during forward computation in cudnn;");
} }
}; };
......
...@@ -247,8 +247,6 @@ static void NMS(const platform::CUDADeviceContext &ctx, const Tensor &proposals, ...@@ -247,8 +247,6 @@ static void NMS(const platform::CUDADeviceContext &ctx, const Tensor &proposals,
const Tensor &sorted_indices, const T nms_threshold, const Tensor &sorted_indices, const T nms_threshold,
Tensor *keep_out) { Tensor *keep_out) {
int boxes_num = proposals.dims()[0]; int boxes_num = proposals.dims()[0];
PADDLE_ENFORCE_EQ(boxes_num, sorted_indices.dims()[0]);
const int col_blocks = DIVUP(boxes_num, kThreadsPerBlock); const int col_blocks = DIVUP(boxes_num, kThreadsPerBlock);
dim3 blocks(DIVUP(boxes_num, kThreadsPerBlock), dim3 blocks(DIVUP(boxes_num, kThreadsPerBlock),
DIVUP(boxes_num, kThreadsPerBlock)); DIVUP(boxes_num, kThreadsPerBlock));
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册