paddlepaddle--paddle/paddle/fluid/inference/tensorrt/convert/conv2d_op.cc

/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.

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/fluid/inference/tensorrt/convert/op_converter.h"
#include "paddle/fluid/inference/tensorrt/engine.h"
#include "paddle/phi/common/data_type.h"

namespace paddle::framework {
class Scope;

}  // namespace paddle::framework
namespace paddle::framework::proto {
class OpDesc;
}  // namespace paddle::framework::proto

namespace paddle::inference::tensorrt {

template <typename RegisterFunc, typename SetDilationFunc>
void ConvertConv2d(TensorRTEngine* engine,
                   const framework::proto::OpDesc& op,
                   const framework::Scope& scope,
                   bool test_mode,
                   RegisterFunc fadd_layer,
                   SetDilationFunc fset_dilation,
                   const std::string& name) {
  VLOG(3) << "convert a " << name << " op to tensorrt layer without bias";

  framework::OpDesc op_desc(op, nullptr);

  auto* X = engine->GetITensor(op_desc.Input("Input").front());
  std::string filter_var_name = op_desc.Input("Filter").front();
  auto* Y_v = scope.FindVar(filter_var_name);
  phi::DenseTensor* Y_t = nullptr;
  nvinfer1::ITensor* filter = nullptr;
  int n_output;
  int n_input;
  int filter_h;
  int filter_w;
  if (Y_v) {
    Y_t = Y_v->GetMutable<phi::DenseTensor>();
    PADDLE_ENFORCE_EQ(
        Y_t->dims().size(),
        4UL,
        common::errors::InvalidArgument(
            "The conv2d filter's dims size should be 4, but got %d",
            Y_t->dims().size()));
    n_output = Y_t->dims()[0];
    n_input = Y_t->dims()[1];
    filter_h = Y_t->dims()[2];
    filter_w = Y_t->dims()[3];
  } else {
    filter = engine->GetITensor(op_desc.Input("Filter").front());
    PADDLE_ENFORCE_EQ(
        filter->getDimensions().nbDims,
        4UL,
        common::errors::InvalidArgument(
            "The conv2d filter's dims size should be 4, but got %d",
            filter->getDimensions().nbDims));
    n_output = filter->getDimensions().d[0];
    n_input = filter->getDimensions().d[1];
    filter_h = filter->getDimensions().d[2];
    filter_w = filter->getDimensions().d[3];
  }

  bool enable_int8 = op_desc.HasAttr("enable_int8");

  if (enable_int8) {
    float in_scale = PADDLE_GET_CONST(float, op_desc.GetAttr("Input_scale"));
    engine->SetTensorDynamicRange(X, in_scale);
  }
  const int groups = PADDLE_GET_CONST(int, op_desc.GetAttr("groups"));
  const std::vector<int> dilations =
      PADDLE_GET_CONST(std::vector<int>, op_desc.GetAttr("dilations"));
  const std::vector<int> strides =
      PADDLE_GET_CONST(std::vector<int>, op_desc.GetAttr("strides"));
  std::vector<int> paddings =
      PADDLE_GET_CONST(std::vector<int>, op_desc.GetAttr("paddings"));
  // for conv2d_transpose
  std::vector<int> output_padding;
  if (op_desc.HasAttr("output_padding")) {
    output_padding =
        PADDLE_GET_CONST(std::vector<int>, op_desc.GetAttr("output_padding"));
  }
  std::string padding_algorithm = "EXPLICIT";
  if (op_desc.HasAttr("padding_algorithm"))
    padding_algorithm =
        PADDLE_GET_CONST(std::string, op_desc.GetAttr("padding_algorithm"));
  if (padding_algorithm == "VALID") {
    for (size_t i = 0; i < paddings.size(); i++) {
      paddings[i] = 0;
    }
  }

  nvinfer1::DimsHW nv_ksize(filter_h, filter_w);
  nvinfer1::DimsHW nv_dilations(dilations[0], dilations[1]);
  nvinfer1::DimsHW nv_strides(strides[0], strides[1]);
  nvinfer1::DimsHW nv_pre_paddings;
  nvinfer1::DimsHW nv_post_paddings;
  if (paddings.size() == 2) {
    nv_pre_paddings.d[0] = paddings[0];
    nv_pre_paddings.d[1] = paddings[1];
    nv_post_paddings.d[0] = paddings[0];
    nv_post_paddings.d[1] = paddings[1];
  } else {
    nv_pre_paddings.d[0] = paddings[0];
    nv_pre_paddings.d[1] = paddings[2];
    nv_post_paddings.d[0] = paddings[1];
    nv_post_paddings.d[1] = paddings[3];
  }
  TensorRTEngine::Weight weight(nvinfer1::DataType::kFLOAT, nullptr, 0);
  if (Y_v) {
    weight = engine->GetTrtWeight(op_desc.Input("Filter").front(), *Y_t);
  }
  TensorRTEngine::Weight bias;
  bias.SetDataType(weight.get().type);
  bias.SetCount(0);
  bias.SetValues(nullptr);
  if (op_desc.Type() == "fused_conv2d_add_act") {
    auto* bias_tensor = scope.GetVar(op_desc.Input("Bias").front());
    auto* bias_tensor_data = bias_tensor->GetMutable<phi::DenseTensor>();
    bias =
        engine->GetTrtWeight(op_desc.Input("Bias").front(), *bias_tensor_data);
  }

  // In conv2d_transpose and depthwise_conv2d_transpose,
  // output channels = filter_dims[1] * groups
  auto* layer = (op_desc.Type() == "conv2d_transpose" ||
                 op_desc.Type() == "depthwise_conv2d_transpose")
                    ? fadd_layer(const_cast<nvinfer1::ITensor*>(X),
                                 n_input * groups,
                                 nv_ksize,
                                 weight,
                                 bias)
                    : fadd_layer(const_cast<nvinfer1::ITensor*>(X),
                                 n_output,
                                 nv_ksize,
                                 weight,
                                 bias);

  PADDLE_ENFORCE_NOT_NULL(
      layer,
      common::errors::Fatal("TensorRT create conv2d/conv2d_transpose"
                            " layer failed."));
  layer->setStrideNd(nv_strides);

  layer->setPrePadding(nv_pre_paddings);

  if (!Y_v) layer->setInput(1, *filter);

  if (!output_padding.empty()) {
    nv_post_paddings.d[0] -= output_padding[0];
    nv_post_paddings.d[1] -= output_padding[1];
  }
  if (nv_post_paddings.d[0] < 0 || nv_post_paddings.d[1] < 0) {
    PADDLE_THROW(common::errors::Fatal(
        "The value in conv2d_transpose's PostPadding should be >= 0."));
  }
  layer->setPostPadding(nv_post_paddings);

  layer->setNbGroups(groups);
  if (padding_algorithm == "SAME") {
    layer->setPaddingMode(nvinfer1::PaddingMode::kSAME_UPPER);
    nv_dilations.d[0] = 1;
    nv_dilations.d[1] = 1;
  }
  // set dilations
  fset_dilation(layer, nv_dilations);

  auto output_name = op_desc.Output("Output").front();
  layer->setName((name + " (Output: " + output_name + ")").c_str());
  layer->getOutput(0)->setName(output_name.c_str());
  engine->SetITensor(output_name, layer->getOutput(0));

  if (test_mode) {
    engine->DeclareOutput(output_name);
  }
}

class Conv2dOpConverter : public OpConverter {
 public:
  void operator()(const framework::proto::OpDesc& op,
                  const framework::Scope& scope,
                  bool test_mode) override {
    framework::OpDesc op_desc(op, nullptr);
    auto output_name = op_desc.Output("Output").front();
    ConvertConv2d(
        engine_,
        op,
        scope,
        test_mode,
        [&](nvinfer1::ITensor* inputs,
            int n_output, /* Conv output maps */
            nvinfer1::DimsHW& ksize,
            TensorRTEngine::Weight& weight,
            TensorRTEngine::Weight& bias) -> nvinfer1::IConvolutionLayer* {
          auto* layer = TRT_ENGINE_ADD_LAYER(engine_,
                                             ConvolutionNd,
                                             *inputs,
                                             n_output,
                                             ksize,
                                             weight.get(),
                                             bias.get());
          SupportFP32MixPrecision(output_name, op_desc.Type(), layer);
          return layer;
        },
        [](nvinfer1::IConvolutionLayer* layer, nvinfer1::DimsHW& dilations) {
          layer->setDilationNd(dilations);
        },
        "conv2d");
  }
};

class Deconv2dOpConverter : public OpConverter {
 public:
  void operator()(const framework::proto::OpDesc& op,
                  const framework::Scope& scope,
                  bool test_mode) override {
    framework::OpDesc op_desc(op, nullptr);
    auto output_name = op_desc.Output("Output").front();
    ConvertConv2d(
        engine_,
        op,
        scope,
        test_mode,
        [&](nvinfer1::ITensor* inputs,
            int n_output, /* Deconv input maps */
            nvinfer1::DimsHW& ksize,
            TensorRTEngine::Weight& weight,
            TensorRTEngine::Weight& bias) -> nvinfer1::IDeconvolutionLayer* {
          auto* layer = TRT_ENGINE_ADD_LAYER(engine_,
                                             DeconvolutionNd,
                                             *inputs,
                                             n_output,
                                             ksize,
                                             weight.get(),
                                             bias.get());
          SupportFP32MixPrecision(output_name, op_desc.Type(), layer);
          return layer;
        },
        [](nvinfer1::IDeconvolutionLayer* layer, nvinfer1::DimsHW& dilations) {
        },
        "conv2d_transpose");
  }
};

}  // namespace paddle::inference::tensorrt

REGISTER_TRT_OP_CONVERTER(conv2d, Conv2dOpConverter);
REGISTER_TRT_OP_CONVERTER(fused_conv2d_add_act, Conv2dOpConverter);
REGISTER_TRT_OP_CONVERTER(conv2d_transpose, Deconv2dOpConverter);