From af1568e61e0029c192b6938925d8716b1416d1ba Mon Sep 17 00:00:00 2001
From: yukavio <kavioyu@gmail.com>
Date: Sun, 27 Sep 2020 13:12:45 +0000
Subject: [PATCH] fix generate_proposals and affine grid error info

---
 .../operators/affine_grid_cudnn_op.cu.cc      | 26 +++++++++++++------
 .../detection/generate_proposals_op.cu        |  2 --
 2 files changed, 18 insertions(+), 10 deletions(-)

diff --git a/paddle/fluid/operators/affine_grid_cudnn_op.cu.cc b/paddle/fluid/operators/affine_grid_cudnn_op.cu.cc
index ed71594ba5..a2027e3506 100644
--- a/paddle/fluid/operators/affine_grid_cudnn_op.cu.cc
+++ b/paddle/fluid/operators/affine_grid_cudnn_op.cu.cc
@@ -26,8 +26,11 @@ template <typename T>
 class CUDNNAffineGridOpKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    PADDLE_ENFORCE(platform::is_gpu_place(ctx.GetPlace()),
-                   "It must use CUDAPlace.");
+    PADDLE_ENFORCE_EQ(platform::is_gpu_place(ctx.GetPlace()), true,
+                   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 handle = dev_ctx.cudnn_handle();
     auto* theta = ctx.Input<Tensor>("Theta");
@@ -56,8 +59,10 @@ class CUDNNAffineGridOpKernel : public framework::OpKernel<T> {
     cudnnSpatialTransformerDescriptor_t cudnn_st_desc =
         st_desc.descriptor<T>(4, h_size_data);
 
-    PADDLE_ENFORCE(platform::dynload::cudnnSpatialTfGridGeneratorForward(
-        handle, cudnn_st_desc, theta_data, output_data));
+    PADDLE_ENFORCE_EQ(platform::dynload::cudnnSpatialTfGridGeneratorForward(
+        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>
 class CUDNNAffineGridGradOpKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
-    PADDLE_ENFORCE(platform::is_gpu_place(ctx.GetPlace()),
-                   "It must use CUDAPlace.");
+    PADDLE_ENFORCE_EQ(platform::is_gpu_place(ctx.GetPlace()), true,
+                   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 handle = dev_ctx.cudnn_handle();
     auto output_grad = ctx.Input<Tensor>(framework::GradVarName("Output"));
@@ -95,8 +103,10 @@ class CUDNNAffineGridGradOpKernel : public framework::OpKernel<T> {
     const T* output_grad_data = output_grad->data<T>();
     T* theta_grad_data = theta_grad->mutable_data<T>(ctx.GetPlace());
 
-    PADDLE_ENFORCE(platform::dynload::cudnnSpatialTfGridGeneratorBackward(
-        handle, cudnn_st_desc, output_grad_data, theta_grad_data));
+    PADDLE_ENFORCE_EQ(platform::dynload::cudnnSpatialTfGridGeneratorBackward(
+        handle, cudnn_st_desc, output_grad_data, theta_grad_data),
+        true, "Some errors "
+        "has occurred during forward computation in cudnn;");
   }
 };
 
diff --git a/paddle/fluid/operators/detection/generate_proposals_op.cu b/paddle/fluid/operators/detection/generate_proposals_op.cu
index 485136d8e2..71323ea966 100644
--- a/paddle/fluid/operators/detection/generate_proposals_op.cu
+++ b/paddle/fluid/operators/detection/generate_proposals_op.cu
@@ -247,8 +247,6 @@ static void NMS(const platform::CUDADeviceContext &ctx, const Tensor &proposals,
                 const Tensor &sorted_indices, const T nms_threshold,
                 Tensor *keep_out) {
   int boxes_num = proposals.dims()[0];
-  PADDLE_ENFORCE_EQ(boxes_num, sorted_indices.dims()[0]);
-
   const int col_blocks = DIVUP(boxes_num, kThreadsPerBlock);
   dim3 blocks(DIVUP(boxes_num, kThreadsPerBlock),
               DIVUP(boxes_num, kThreadsPerBlock));
-- 
GitLab