/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. */ #include "paddle/operators/conv_op.h" namespace paddle { namespace operators { void ConvOp::InferShape(framework::InferShapeContext* ctx) const { PADDLE_ENFORCE(ctx->HasInput("Input"), "Input(Input) of ConvOp should not be null."); PADDLE_ENFORCE(ctx->HasInput("Filter"), "Input(Filter) of ConvOp should not be null."); PADDLE_ENFORCE(ctx->HasOutput("Output"), "Output(Output) of ConvOp should not be null."); auto in_dims = ctx->GetInputDim("Input"); auto filter_dims = ctx->GetInputDim("Filter"); std::vector strides = ctx->Attrs().Get>("strides"); std::vector paddings = ctx->Attrs().Get>("paddings"); int groups = ctx->Attrs().Get("groups"); std::vector dilations = ctx->Attrs().Get>("dilations"); PADDLE_ENFORCE(in_dims.size() == 4 || in_dims.size() == 5, "Conv intput should be 4-D or 5-D tensor."); PADDLE_ENFORCE_EQ( in_dims.size(), filter_dims.size(), "Conv input dimension and filter dimension should be the same."); PADDLE_ENFORCE( in_dims.size() - strides.size() == 2U, "Conv input dimension and strides dimension should be consistent."); PADDLE_ENFORCE_EQ( paddings.size(), strides.size(), "Conv paddings dimension and Conv strides dimension should be the same."); PADDLE_ENFORCE_EQ(in_dims[1], filter_dims[1] * groups, "The number of input channels should be equal to filter " "channels * groups."); PADDLE_ENFORCE_EQ( filter_dims[0] % groups, 0, "The number of output channels should be divided by groups."); std::vector output_shape({in_dims[0], filter_dims[0]}); for (size_t i = 0; i < strides.size(); ++i) { PADDLE_ENFORCE(in_dims[i + 2] + 2 * paddings[i] - (dilations[i] * (filter_dims[i + 2] - 1) + 1) > 0, "Due to the settings of paddings, filter_dims and " "dilations, the output size is less than 0, please check " "again."); output_shape.push_back(OutputSize(in_dims[i + 2], filter_dims[i + 2], dilations[i], paddings[i], strides[i])); } ctx->SetOutputDim("Output", framework::make_ddim(output_shape)); ctx->ShareLoD("Input", "Output"); } framework::OpKernelType ConvOp::GetExpectedKernelType( const framework::ExecutionContext& ctx) const { bool use_cudnn = ctx.Attr("use_cudnn"); use_cudnn &= platform::is_gpu_place(ctx.GetPlace()); framework::LibraryType library_; if (use_cudnn) { library_ = framework::LibraryType::kCUDNN; } else { library_ = framework::LibraryType::kPlain; } std::string data_format = ctx.Attr("data_format"); framework::DataLayout layout_ = framework::StringToDataLayout(data_format); return framework::OpKernelType( framework::ToDataType(ctx.Input("Input")->type()), ctx.GetPlace(), layout_, library_); } Conv2DOpMaker::Conv2DOpMaker(OpProto* proto, OpAttrChecker* op_checker) : OpProtoAndCheckerMaker(proto, op_checker) { AddInput( "Input", "(Tensor) The input tensor of convolution operator. " "The format of input tensor is NCHW, where N is batch size, C is the " "number of channels, H is the height of the feature, " "and W is the width of the feature."); AddInput("Filter", "(Tensor) The filter tensor of convolution operator. " "The format of the filter tensor is MCHW, where M is the number of " "output image channels, C is the number of input image channels, " "H is the height of the filter, and W is the width of the filter. " "If the groups attribute is greater than 1, C equals the number of " "input image channels divided by the groups."); AddOutput("Output", "(Tensor) The output tensor of convolution operator. " "The format of output tensor is also NCHW."); AddAttr>("strides", "(vector default:{1, 1}), the " "strides(h_stride, w_stride) of " "convolution operator.") .SetDefault({1, 1}); AddAttr>("paddings", "(vector default:{0, 0}), the " "paddings(h_pad, w_pad) of " "convolution operator.") .SetDefault({0, 0}); AddAttr( "groups", "(int default:1), the groups number of the convolution operator. " "According to grouped convolution in Alex Krizhevsky's Deep CNN paper: " "when group=2, the first half of the filters is only connected to the " "first half of the input channels, while the second half of the filters " "is only connected to the second half of the input channels.") .SetDefault(1); AddAttr>("dilations", "(vector default:{1, 1}), the " "dilations(h_dilation, w_dilation) of " "convolution operator.") .SetDefault({1, 1}); AddAttr( "use_cudnn", "(bool, default false) Only used in cudnn kernel, need install cudnn") .SetDefault(false); AddAttr( "data_format", "(string, default NCHW) Only used in " "An optional string from: \"NHWC\", \"NCHW\". " "Defaults to \"NHWC\". Specify the data format of the output data, " "the input will be transformed automatically. ") .SetDefault("AnyLayout"); // TODO(dzhwinter): need to registered layout transform function AddAttr("workspace_size_MB", "Only used in cudnn kernel. Need set use_cudnn to true." "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 hardware. This size should be chosen carefully.") .SetDefault(4096); AddComment(R"DOC( Convolution Operator. The convolution operation calculates the output based on the input, filter and strides, paddings, dilations, groups parameters. The size of each dimension of the parameters is checked in the infer-shape. Input(Input) and Output(Output) are in NCHW format. Where N is batch size, C is the number of channels, H is the height of the feature, and W is the width of the feature. Filters(Input) is MCHW format. Where M is the number of output image channels, C is the number of input image channels, H is the height of the filter, and W is the width of the filter. Parameters(strides, paddings, dilations) are two elements. These two elements represent height and width, respectively. The input(X) size and output(Out) size may be different. Example: Input: Input shape: $(N, C_{in}, H_{in}, W_{in})$ Filter shape: $(C_{out}, C_{in}, H_f, W_f)$ Output: Output shape: $(N, C_{out}, H_{out}, W_{out})$ Where $$ H_{out}= \frac{(H_{in} + 2 * paddings[0] - (dilations[0] * (H_f - 1) + 1))}{strides[0]}+ 1 \\ W_{out}= \frac{(W_{in} + 2 * paddings[1] - (dilations[1] * (W_f - 1) + 1))}{strides[1]}+ 1 $$ )DOC"); } Conv3DOpMaker::Conv3DOpMaker(OpProto* proto, OpAttrChecker* op_checker) : OpProtoAndCheckerMaker(proto, op_checker) { AddInput( "Input", "(Tensor) The input tensor of convolution operator. " "The format of input tensor is NCDHW. Where N is batch size, C is the " "number of channels, D is the depth of the feature, H is the height of " "the feature, " "and W is the width of the feature."); AddInput("Filter", "(Tensor) The filter tensor of convolution operator. " "The format of the filter tensor is MCDHW, where M is the number of " "output image channels, C is the number of input image channels, " "D is the depth of the filter, H is the height of the filter, and W " "is the width of the filter." "If the groups attribute is greater than 1, C equals the number of " "input image channels divided by the groups."); AddOutput("Output", "(Tensor) The output tensor of convolution operator." "The format of output tensor is also NCDHW."); AddAttr>("strides", "(vector, default:{1, 1, 1}), the " "strides(d_stride, h_stride, w_stride) of " "convolution operator.") .SetDefault({1, 1, 1}); AddAttr>("paddings", "(vector, default:{0, 0, 0}), the " "paddings(d_pad, h_pad, w_pad) of convolution " "operator.") .SetDefault({0, 0, 0}); AddAttr( "groups", "(int default:1), the groups number of the convolution operator. " "According to grouped convolution in Alex Krizhevsky's Deep CNN paper: " "when group=2, the first half of the filters is only connected to the " "first half of the input channels, while the second half of the filters " "is only connected to the second half of the input channels.") .SetDefault(1); AddAttr>("dilations", "(vector default:{1, 1, 1}), the " "dilations(d_dilation, h_dilation, w_dilation) of " "convolution operator.") .SetDefault({1, 1, 1}); AddAttr( "use_cudnn", "(bool, default false) Only used in cudnn kernel, need install cudnn") .SetDefault(false); AddAttr( "data_format", "(string, default NCHW) Only used in " "An optional string from: \"NHWC\", \"NCHW\". " "Defaults to \"NHWC\". Specify the data format of the output data, " "the input will be transformed automatically. ") .SetDefault("AnyLayout"); // TODO(dzhwinter): need to registered layout transform function AddAttr("workspace_size_MB", "Only used in cudnn kernel. 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 hardware. This size should be chosen carefully.") .SetDefault(4096); AddComment(R"DOC( Convolution3D Operator. The convolution operation calculates the output based on the input, filter and strides, paddings, dilations, groups parameters. The size of each dimension of the parameters is checked in the infer-shape. Input(Input) and output(Output) are in NCDHW format, where N is batch size, C is the number of channels,D is the depth of the feature, H is the height of the feature, and W is the width of the feature. Filters(Input) is MCDHW format, where M is the number of output image channels, C is the number of input image channels, D is the depth of the filter, H is the height of the filter, and W is the width of the filter. Parameters(strides, paddings, dilations) are three elements. These three elements represent depth, height and width, respectively. The input(X) size and output(Out) size may be different. Example: Input: Input shape: $(N, C_{in}, D_{in}, H_{in}, W_{in})$ Filter shape: $(C_{out}, C_{in}, D_f, H_f, W_f)$ Output: Output shape: $(N, C_{out}, D_{out}, H_{out}, W_{out})$ Where $$ D_{out}= \frac{(D_{in} + 2 * paddings[0] - (dilations[0] * (D_f - 1) + 1))}{ strides[0]}+ 1 \\ H_{out}= \frac{(H_{in} + 2 * paddings[1] - (dilations[1] * (H_f - 1) + 1))}{ strides[1]}+ 1 \\ W_{out}= \frac{(W_{in} + 2 * paddings[2] - (dilations[2] * (W_f - 1) + 1))}{ strides[2]}+ 1 $$ )DOC"); } void ConvOpGrad::InferShape(framework::InferShapeContext* ctx) const { auto in_dims = ctx->GetInputDim("Input"); auto filter_dims = ctx->GetInputDim("Filter"); if (ctx->HasOutput(framework::GradVarName("Input"))) { ctx->SetOutputDim(framework::GradVarName("Input"), in_dims); } if (ctx->HasOutput(framework::GradVarName("Filter"))) { ctx->SetOutputDim(framework::GradVarName("Filter"), filter_dims); } } framework::OpKernelType ConvOpGrad::GetExpectedKernelType( const framework::ExecutionContext& ctx) const { bool use_cudnn = ctx.Attr("use_cudnn"); if (paddle::platform::is_cpu_place(ctx.GetPlace())) { use_cudnn = false; } framework::LibraryType library_; if (use_cudnn) { library_ = framework::LibraryType::kCUDNN; } else { library_ = framework::LibraryType::kPlain; } std::string data_format = ctx.Attr("data_format"); framework::DataLayout layout_ = framework::StringToDataLayout(data_format); return framework::OpKernelType( framework::ToDataType(ctx.Input("Input")->type()), ctx.GetPlace(), layout_, library_); } } // namespace operators } // namespace paddle namespace ops = paddle::operators; REGISTER_OP(conv2d, ops::ConvOp, ops::Conv2DOpMaker, conv2d_grad, ops::ConvOpGrad); REGISTER_OP(conv3d, ops::ConvOp, ops::Conv3DOpMaker, conv3d_grad, ops::ConvOpGrad); REGISTER_OP_CPU_KERNEL( conv2d, ops::GemmConvKernel, ops::GemmConvKernel); REGISTER_OP_CPU_KERNEL( conv2d_grad, ops::GemmConvGradKernel, ops::GemmConvGradKernel); REGISTER_OP_CPU_KERNEL( conv3d, ops::GemmConvKernel, ops::GemmConvKernel); REGISTER_OP_CPU_KERNEL( conv3d_grad, ops::GemmConvGradKernel, ops::GemmConvGradKernel);