add double register op_data_type of pad2d and fix compile error, test…

…=develop (PaddlePaddle#22075)
seiriosPlus · Jan 8, 2020 · 64a4044 · 64a4044
1 parent 7ba7acd
commit 64a4044
Show file tree

Hide file tree

Showing 2 changed files with 18 additions and 10 deletions.
diff --git a/paddle/fluid/operators/pad2d_op.cc b/paddle/fluid/operators/pad2d_op.cc
@@ -661,5 +661,8 @@ REGISTER_OPERATOR(pad2d, ops::Pad2dOp, ops::Pad2dOpMaker,
                   ops::Pad2dOpGradMaker<paddle::imperative::OpBase>);
 REGISTER_OPERATOR(pad2d_grad, ops::Pad2dOpGrad,
                   ops::Pad2dOpGradNoNeedBufferVarsInference);
-REGISTER_OP_CPU_KERNEL(pad2d, ops::Pad2dCPUKernel<float>);
-REGISTER_OP_CPU_KERNEL(pad2d_grad, ops::Pad2dGradCPUKernel<float>);
+REGISTER_OP_CPU_KERNEL(pad2d, ops::Pad2dCPUKernel<float>,
+                       ops::Pad2dCPUKernel<double>, ops::Pad2dCPUKernel<int>,
+                       ops::Pad2dCPUKernel<int64_t>);
+REGISTER_OP_CPU_KERNEL(pad2d_grad, ops::Pad2dGradCPUKernel<float>,
+                       ops::Pad2dGradCPUKernel<double>);
diff --git a/paddle/fluid/operators/pad2d_op.cu b/paddle/fluid/operators/pad2d_op.cu
@@ -215,8 +215,9 @@ __global__ void Pad2DGradReflectNCHW(const int out_size, T* d_in_data,
     in_w = max(in_w, -in_w);
     in_h = min(in_h, 2 * in_height - in_h - 2);
     in_w = min(in_w, 2 * in_width - in_w - 2);
-    atomicAdd(&d_in_data[(nc * in_height + in_h) * in_width + in_w],
-              d_out_data[out_index]);
+    platform::CudaAtomicAdd(
+        &d_in_data[(nc * in_height + in_h) * in_width + in_w],
+        d_out_data[out_index]);
   }
 }
 
@@ -240,7 +241,7 @@ __global__ void Pad2DGradReflectNHWC(const int out_size, T* d_in_data,
     in_w = max(in_w, -in_w);
     in_h = min(in_h, in_height * 2 - in_h - 2);
     in_w = min(in_w, in_width * 2 - in_w - 2);
-    atomicAdd(
+    platform::CudaAtomicAdd(
         &d_in_data[((n * in_height + in_h) * in_width + in_w) * channels + c],
         d_out_data[out_index]);
   }
@@ -260,8 +261,9 @@ __global__ void Pad2DGradEdgeNCHW(const int out_size, T* d_in_data,
     nc /= out_height;
     const int in_h = min(in_height - 1, max(out_h - pad_top, 0));
     const int in_w = min(in_width - 1, max(out_w - pad_left, 0));
-    atomicAdd(&d_in_data[(nc * in_height + in_h) * in_width + in_w],
-              d_out_data[out_index]);
+    platform::CudaAtomicAdd(
+        &d_in_data[(nc * in_height + in_h) * in_width + in_w],
+        d_out_data[out_index]);
   }
 }
 
@@ -281,7 +283,7 @@ __global__ void Pad2DGradEdgeNHWC(const int out_size, T* d_in_data,
     n /= out_height;
     const int in_h = min(in_height - 1, max(out_h - pad_top, 0));
     const int in_w = min(in_width - 1, max(out_w - pad_left, 0));
-    atomicAdd(
+    platform::CudaAtomicAdd(
         &d_in_data[((n * in_height + in_h) * in_width + in_w) * channels + c],
         d_out_data[out_index]);
   }
@@ -459,5 +461,8 @@ class Pad2dGradCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP_CUDA_KERNEL(pad2d, ops::Pad2dCUDAKernel<float>);
-REGISTER_OP_CUDA_KERNEL(pad2d_grad, ops::Pad2dGradCUDAKernel<float>);
+REGISTER_OP_CUDA_KERNEL(pad2d, ops::Pad2dCUDAKernel<float>,
+                        ops::Pad2dCUDAKernel<double>, ops::Pad2dCUDAKernel<int>,
+                        ops::Pad2dCUDAKernel<int64_t>);
+REGISTER_OP_CUDA_KERNEL(pad2d_grad, ops::Pad2dGradCUDAKernel<float>,
+                        ops::Pad2dGradCUDAKernel<double>);