add conv3d_trans_cudnn_op

paddle/operators/CMakeLists.txt

@@ -61,16 +61,30 @@ function(op_library TARGET)
         set(pybind_flag 1)
     endif()
 
+    if ("${TARGET}" STREQUAL "compare_op")
+        set(pybind_flag 1)
+        file(APPEND ${pybind_file} "USE_OP(less_than);\nUSE_OP(equal);\n")
+    endif()
+
+    # conv_op contains several operators
+    if ("${TARGET}" STREQUAL "conv_op")
+        set(pybind_flag 1)
+        # It's enough to just adding one operator to pybind
+        file(APPEND ${pybind_file} "USE_OP(conv2d);\n")
+    endif()
+
     # pool_op contains several operators
     if ("${TARGET}" STREQUAL "pool_op")
         set(pybind_flag 1)
         # It's enough to just adding one operator to pybind
         file(APPEND ${pybind_file} "USE_OP(pool2d);\n")
     endif()
 
-    if ("${TARGET}" STREQUAL "compare_op")
+    # pool_cudnn_op contains several operators
+    if ("${TARGET}" STREQUAL "pool_cudnn_op")
         set(pybind_flag 1)
-        file(APPEND ${pybind_file} "USE_OP(less_than);\nUSE_OP(equal);\n")
+        # It's enough to just adding one operator to pybind
+        file(APPEND ${pybind_file} "USE_OP(pool2d_cudnn);\n")
     endif()
 
     # pool_with_index_op contains several operators
@@ -80,25 +94,18 @@ function(op_library TARGET)
         file(APPEND ${pybind_file} "USE_OP(max_pool2d_with_index);\n")
     endif()
 
-    # conv_op contains several operators
-    if ("${TARGET}" STREQUAL "conv_op")
-        set(pybind_flag 1)
-        # It's enough to just adding one operator to pybind
-        file(APPEND ${pybind_file} "USE_OP(conv2d);\n")
-    endif()
-
     # conv_transpose_op contains several operators
     if ("${TARGET}" STREQUAL "conv_transpose_op")
         set(pybind_flag 1)
         # It's enough to just adding one operator to pybind
         file(APPEND ${pybind_file} "USE_OP(conv2d_transpose);\n")
     endif()
-    
-    # pool_cudnn_op contains several operators
-    if ("${TARGET}" STREQUAL "pool_cudnn_op")
+
+    # conv_transpose_cudnn_op contains two operators
+    if ("${TARGET}" STREQUAL "conv_transpose_cudnn_op")
         set(pybind_flag 1)
         # It's enough to just adding one operator to pybind
-        file(APPEND ${pybind_file} "USE_OP(pool2d_cudnn);\n")
+        file(APPEND ${pybind_file} "USE_OP(conv2d_transpose_cudnn);\n")
     endif()
 
     # save_restore_op contains several operators

paddle/operators/conv_cudnn_op.cu.cc

@@ -226,9 +226,8 @@ class CudnnConvGradOpKernel : public framework::OpKernel<T> {
     T alpha = 1.0f, beta = 0.0f;
     if (input_grad) {
       T* input_grad_data = input_grad->mutable_data<T>(ctx.GetPlace());
-      auto t = framework::EigenVector<T>::Flatten(*input_grad);
-      t.device(ctx.GetEigenDevice<platform::GPUPlace>()) =
-          t.constant(static_cast<T>(0));
+      // Because beta is zero, it is unnecessary to reset input_grad.
+
       for (int i = 0; i < groups; i++) {
         PADDLE_ENFORCE(platform::dynload::cudnnConvolutionBackwardData(
             handle, &alpha, cudnn_filter_desc,
@@ -241,9 +240,8 @@ class CudnnConvGradOpKernel : public framework::OpKernel<T> {
     // ------------------- cudnn conv backward filter ---------------------
     if (filter_grad) {
       T* filter_grad_data = filter_grad->mutable_data<T>(ctx.GetPlace());
-      auto t = framework::EigenVector<T>::Flatten(*filter_grad);
-      t.device(ctx.GetEigenDevice<platform::GPUPlace>()) =
-          t.constant(static_cast<T>(0));
+      // Because beta is zero, it is unnecessary to reset filter_grad.
+
       for (int i = 0; i < groups; i++) {
         PADDLE_ENFORCE(platform::dynload::cudnnConvolutionBackwardFilter(
             handle, &alpha, cudnn_input_desc, input_data + i * group_offset_in,

paddle/operators/conv2d_transpose_cudnn_op.cc → paddle/operators/conv_transpose_cudnn_op.cc

@@ -23,7 +23,24 @@ class CudnnConv2DTransposeOpMaker : public Conv2DTransposeOpMaker {
                               framework::OpAttrChecker* op_checker)
       : Conv2DTransposeOpMaker(proto, op_checker) {
     AddAttr<std::vector<int>>("dilations", "dilations of convolution operator.")
-        .SetDefault(std::vector<int>{1, 1});
+        .SetDefault({1, 1});
+    AddAttr<int>("workspace_size_MB",
+                 "workspace size for cudnn, in MB, "
+                 "workspace is a section of GPU memory which will be "
+                 "allocated/freed each time the operator runs, larger "
+                 "workspace size can increase performance but also requires "
+                 "better hardward. This size should be carefully setted.")
+        .SetDefault(4096);
+  }
+};
+
+class CudnnConv3DTransposeOpMaker : public Conv3DTransposeOpMaker {
+ public:
+  CudnnConv3DTransposeOpMaker(framework::OpProto* proto,
+                              framework::OpAttrChecker* op_checker)
+      : Conv3DTransposeOpMaker(proto, op_checker) {
+    AddAttr<std::vector<int>>("dilations", "dilations of convolution operator.")
+        .SetDefault({1, 1, 1});
     AddAttr<int>("workspace_size_MB",
                  "workspace size for cudnn, in MB, "
                  "workspace is a section of GPU memory which will be "
@@ -48,3 +65,14 @@ REGISTER_OP_CPU_KERNEL(
 REGISTER_OP_CPU_KERNEL(
     conv2d_transpose_cudnn_grad,
     ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>);
+
+REGISTER_OP(conv3d_transpose_cudnn, ops::ConvTransposeOp,
+            ops::CudnnConv3DTransposeOpMaker, conv3d_transpose_cudnn_grad,
+            ops::ConvTransposeOpGrad);
+
+REGISTER_OP_CPU_KERNEL(
+    conv3d_transpose_cudnn,
+    ops::GemmConvTransposeKernel<paddle::platform::CPUPlace, float>);
+REGISTER_OP_CPU_KERNEL(
+    conv3d_transpose_cudnn_grad,
+    ops::GemmConvTransposeGradKernel<paddle::platform::CPUPlace, float>);

paddle/operators/conv2d_transpose_cudnn_op.cu.cc → paddle/operators/conv_transpose_cudnn_op.cu.cc

@@ -54,15 +54,21 @@ class CudnnConvTransposeOpKernel : public framework::OpKernel<T> {
     ScopedTensorDescriptor output_desc;
     ScopedFilterDescriptor filter_desc;
     ScopedConvolutionDescriptor conv_desc;
-    DataLayout layout = DataLayout::kNCHW;
+    DataLayout layout;
+
+    if (strides.size() == 2U) {
+      layout = DataLayout::kNCHW;
+    } else {
+      layout = DataLayout::kNCDHW;
+    }
 
-    // N, M, H, W
+    // (N, M, H, W) or (N, M, D, H, W)
     cudnnTensorDescriptor_t cudnn_input_desc = input_desc.descriptor<T>(
         layout, framework::vectorize2int(input->dims()));
-    // N, C, O_h, O_w
+    // (N, C, O_h, O_w) or (N, C, O_d, O_h, O_w)
     cudnnTensorDescriptor_t cudnn_output_desc = output_desc.descriptor<T>(
         layout, framework::vectorize2int(output->dims()));
-    // M, C, K_h, K_w
+    // (M, C, K_h, K_w) or (M, C, K_d, K_h, K_w)
     cudnnFilterDescriptor_t cudnn_filter_desc = filter_desc.descriptor<T>(
         layout, framework::vectorize2int(filter->dims()));
     cudnnConvolutionDescriptor_t cudnn_conv_desc =
@@ -136,13 +142,13 @@ class CudnnConvTransposeGradOpKernel : public framework::OpKernel<T> {
     ScopedConvolutionDescriptor conv_desc;
     DataLayout layout = DataLayout::kNCHW;
 
-    // Input: (N, M, H, W)
+    // Input: (N, M, H, W) or (N, M, D, H, W)
     cudnnTensorDescriptor_t cudnn_input_desc = input_desc.descriptor<T>(
         layout, framework::vectorize2int(input->dims()));
-    // Output: (N, C, O_H, O_W)
+    // Output: (N, C, O_h, O_w) or (N, C, O_d, O_h, O_w)
     cudnnTensorDescriptor_t cudnn_output_desc = output_desc.descriptor<T>(
         layout, framework::vectorize2int(output_grad->dims()));
-    // Filter (M, C, K_H, K_W)
+    // Filter (M, C, K_h, K_w) or (M, C, K_d K_h, K_w)
     cudnnFilterDescriptor_t cudnn_filter_desc = filter_desc.descriptor<T>(
         layout, framework::vectorize2int(filter->dims()));
 
@@ -200,8 +206,7 @@ class CudnnConvTransposeGradOpKernel : public framework::OpKernel<T> {
     T alpha = 1.0f, beta = 0.0f;
     if (input_grad) {
       T* input_grad_data = input_grad->mutable_data<T>(ctx.GetPlace());
-      math::set_constant(ctx.device_context(), input_grad, 0);
-
+      // Because beta is zero, it is unnecessary to reset input_grad.
       PADDLE_ENFORCE(platform::dynload::cudnnConvolutionForward(
           handle, &alpha, cudnn_output_desc, output_grad_data,
           cudnn_filter_desc, filter_data, cudnn_conv_desc, data_algo,
@@ -212,8 +217,7 @@ class CudnnConvTransposeGradOpKernel : public framework::OpKernel<T> {
     // ------------------- cudnn conv backward filter ---------------------
     if (filter_grad) {
       T* filter_grad_data = filter_grad->mutable_data<T>(ctx.GetPlace());
-      math::set_constant(ctx.device_context(), filter_grad, 0);
-
+      // Because beta is zero, it is unnecessary to reset filter_grad.
       // Gradient with respect to the filter
       PADDLE_ENFORCE(platform::dynload::cudnnConvolutionBackwardFilter(
           handle, &alpha, cudnn_output_desc, output_grad_data, cudnn_input_desc,
@@ -234,3 +238,8 @@ REGISTER_OP_GPU_KERNEL(conv2d_transpose_cudnn,
                        ops::CudnnConvTransposeOpKernel<float>);
 REGISTER_OP_GPU_KERNEL(conv2d_transpose_cudnn_grad,
                        ops::CudnnConvTransposeGradOpKernel<float>);
+
+REGISTER_OP_GPU_KERNEL(conv3d_transpose_cudnn,
+                       ops::CudnnConvTransposeOpKernel<float>);
+REGISTER_OP_GPU_KERNEL(conv3d_transpose_cudnn_grad,
+                       ops::CudnnConvTransposeGradOpKernel<float>);

paddle/operators/pool_cudnn_op.cu.cc

@@ -135,8 +135,7 @@ class PoolCudnnGradOpKernel : public framework::OpKernel<T> {
 
     if (input_grad) {
       T *input_grad_data = input_grad->mutable_data<T>(ctx.GetPlace());
-      math::SetConstant<paddle::platform::GPUPlace, T> set_zero;
-      set_zero(ctx.device_context(), input_grad, static_cast<T>(0));
+      // Because beta is zero, it is unnecessary to reset input_grad.
 
       PADDLE_ENFORCE(platform::dynload::cudnnPoolingBackward(
           handle, cudnn_pool_desc, &alpha, cudnn_output_desc, output_data,

paddle/platform/cudnn_helper.h

@@ -63,9 +63,10 @@ inline const char* cudnnGetErrorString(cudnnStatus_t status) {
     }                                                             \
   } while (false)
 
-enum class DataLayout {
+enum class DataLayout {  // Not use
   kNHWC,
   kNCHW,
+  kNCDHW,
   kNCHW_VECT_C,
 };
 
@@ -107,12 +108,15 @@ class CudnnDataType<double> {
   }
 };
 
-inline cudnnTensorFormat_t GetCudnnTensorFormat(const DataLayout& order) {
+inline cudnnTensorFormat_t GetCudnnTensorFormat(
+    const DataLayout& order) {  // Not use
   switch (order) {
     case DataLayout::kNHWC:
       return CUDNN_TENSOR_NHWC;
     case DataLayout::kNCHW:
       return CUDNN_TENSOR_NCHW;
+    case DataLayout::kNCDHW:
+      return CUDNN_TENSOR_NCHW;  // TODO(chengduoZH) : add CUDNN_TENSOR_NCDHW
     default:
       PADDLE_THROW("Unknown cudnn equivalent for order");
   }
@@ -139,7 +143,7 @@ class ScopedTensorDescriptor {
       strides[i] = dims[i + 1] * strides[i + 1];
     }
     // Update tensor descriptor dims setting if groups > 1
-    // FIXME(typhoonzero): Assume using NCHW order
+    // FIXME(typhoonzero): Assume using NCHW or NCDHW order
     std::vector<int> dims_with_group(dims.begin(), dims.end());  // copy
     if (groups > 1) {
       dims_with_group[1] = dims_with_group[1] / groups;
@@ -176,9 +180,10 @@ class ScopedFilterDescriptor {
                                             const cudnnDataType_t type,
                                             const std::vector<int>& kernel,
                                             const int groups = 1) {
-    // filter layout: MCHW, where M is the number of
+    // filter layout: MCHW(MCDHW), where M is the number of
     // output image channels, C is the number of input image channels,
-    // H and W is height and width of filter.
+    // D is the depth of the filter, H is the height of the filter, and W is the
+    // width of the filter.
     std::vector<int> kernel_with_group(kernel.begin(), kernel.end());
     if (groups > 1) {
       // M /= groups

python/paddle/v2/fluid/tests/test_conv3d_transpose_op.py

@@ -108,5 +108,11 @@ def init_test_case(self):
         self.filter_size = [f_c, 6, 3, 3, 3]
 
 
+# ------------ test_cudnn ------------
+class TestCudnn(TestConv3dTransposeOp):
+    def init_op_type(self):
+        self.op_type = "conv3d_transpose_cudnn"
+
+
 if __name__ == '__main__':
     unittest.main()