conv2d_fusion.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/anakin/convert/conv2d_fusion.h"
#include <algorithm>
#include <memory>
#include <vector>
#include "paddle/fluid/inference/anakin/convert/helper.h"

using anakin::PTuple;

namespace paddle {
namespace inference {
namespace anakin {

template <typename TargetT, ::anakin::Precision PrecisionT>
void Conv2dFusionOpConverter<TargetT, PrecisionT>::operator()(
    const framework::proto::OpDesc &op, const framework::BlockDesc &block_desc,
    const framework::Scope &scope, bool test_mode) {
  framework::OpDesc op_desc(op, nullptr);
  PADDLE_ENFORCE_EQ(op_desc.Input("Input").size(), 1UL);
  PADDLE_ENFORCE_EQ(op_desc.Input("Filter").size(), 1UL);
  PADDLE_ENFORCE_EQ(op_desc.Input("Bias").size(), 1UL);
  PADDLE_ENFORCE_EQ(op_desc.Output("Output").size(), 1UL);

  auto input_name = op_desc.Input("Input").front();
  auto output_name = op_desc.Output("Output").front();
  auto op_name = op_desc.Type() + ":" + op_desc.Output("Output").front();
  this->engine_->AddOp(op_name, "Convolution", {input_name}, {output_name});

  auto *filter_v = scope.FindVar(op_desc.Input("Filter").front());
  PADDLE_ENFORCE_NOT_NULL(filter_v);

  auto weight_tensor = tensor_from_var(*filter_v, platform::CPUPlace());
  auto weight_shape = framework::vectorize2int(weight_tensor->dims());

  auto *b_v = scope.FindVar(op_desc.Input("Bias").front());
  PADDLE_ENFORCE_NOT_NULL(b_v);

  PADDLE_ENFORCE_EQ(weight_tensor->dims().size(), 4UL);
  const int filter_h = weight_tensor->dims()[2];
  const int filter_w = weight_tensor->dims()[3];
  auto filter_num = weight_tensor->dims()[0];
  this->engine_->template AddOpAttr<int>(op_name, "filter_num", filter_num);
  this->engine_->template AddOpAttr<PTuple<int>>(op_name, "kernel_size",
                                                 {filter_h, filter_w});
  auto strides = boost::get<std::vector<int>>(op_desc.GetAttr("strides"));
  this->engine_->template AddOpAttr<PTuple<int>>(op_name, "strides", strides);
  auto paddings = boost::get<std::vector<int>>(op_desc.GetAttr("paddings"));
  this->engine_->template AddOpAttr<PTuple<int>>(op_name, "padding", paddings);
  auto dilations = boost::get<std::vector<int>>(op_desc.GetAttr("dilations"));
  this->engine_->template AddOpAttr<PTuple<int>>(op_name, "dilation_rate",
                                                 dilations);
  const int groups = boost::get<int>(op_desc.GetAttr("groups"));
  this->engine_->AddOpAttr(op_name, "group", groups);
  this->engine_->AddOpAttr(op_name, "axis", 1);
  this->engine_->AddOpAttr(op_name, "bias_term", true);

  ::anakin::saber::Shape anakin_shape(weight_shape);
  bool enable_int8 = boost::get<bool>(op_desc.HasAttr("enable_int8"));
  if (enable_int8) {
    const float int8_range = 127.;
    float in_scale = boost::get<float>(op_desc.GetAttr("input_scale"));
    float weight_scale = boost::get<float>(op_desc.GetAttr("weight_scale"));
    PBlock<TargetT> *weight1 =
        new PBlock<TargetT>(anakin_shape, ::anakin::AK_INT8);
    this->engine_->RegistBlock(weight1);
    float *weight_data = weight_tensor->data<float>();
    std::vector<char> weight_int8;
    int weight_num = weight_tensor->numel();
    for (int i = 0; i < weight_tensor->numel(); i++) {
      bool is_valid_int8 =
          ((weight_data[i] >= -128) && (weight_data[i] <= 127));
      PADDLE_ENFORCE(is_valid_int8,
                     "We are in anakin subgraph int8 mode, the weight of conv "
                     "should be in range [-128, 127]");
      weight_int8.push_back(static_cast<char>(weight_data[i]));
    }
    memcpy(static_cast<void *>(weight1->h_tensor().mutable_data()),
           static_cast<void *>(weight_int8.data()), sizeof(char) * weight_num);
    weight1->d_tensor().set_shape(anakin_shape);
    weight1->d_tensor().copy_from(weight1->h_tensor());
    this->engine_->AddOpAttr(op_name, "weight_1", *weight1);
    this->engine_->Graph()->SetOpPrec(op_name, ::anakin::AK_INT8);
    this->engine_->Graph()->SetWeightsScale(op_name,
                                            {weight_scale / int8_range}, false);
    this->engine_->AddTensorScale(input_name, in_scale / int8_range);
  } else {
    auto weight_tensor = tensor_from_var(*filter_v, platform::CPUPlace());
    auto weight_shape = framework::vectorize2int(weight_tensor->dims());
    auto *weight1 = pblock_from_tensor<TargetT, PrecisionT>(
        *weight_tensor, weight_shape, this->engine_);
    this->engine_->AddOpAttr(op_name, "weight_1", *weight1);
    auto weight2 = pblock_from_var<TargetT, PrecisionT>(*b_v, this->engine_);
    this->engine_->AddOpAttr(op_name, "weight_2", *weight2);
  }
}

}  // namespace anakin
}  // namespace inference
}  // namespace paddle

REGISTER_ANAKIN_OP_CONVERTER(conv2d_fusion, Conv2dFusionOpConverter);