sequence_conv_op.h

/* 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. */

#pragma once
#include "paddle/framework/eigen.h"
#include "paddle/framework/op_registry.h"
#include "paddle/operators/math/math_function.h"
#include "paddle/operators/math/sequence_project.h"

namespace paddle {
namespace operators {

using Tensor = framework::Tensor;
using LoDTensor = framework::LoDTensor;
// template <typename T, int MajorType = Eigen::RowMajor,
//          typename IndexType = Eigen::DenseIndex>
// using EigenVector = framework::EigenVector<T, MajorType, IndexType>;
template <typename T, int MajorType = Eigen::RowMajor,
          typename IndexType = Eigen::DenseIndex>
using EigenMatrix = framework::EigenMatrix<T, MajorType, IndexType>;

template <typename Place, typename T>
class SequenceConvKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& context) const override {
    auto* in = context.Input<LoDTensor>("X");
    auto* out = context.Output<LoDTensor>("Out");
    auto filter = *context.Input<LoDTensor>("Filter");

    out->mutable_data<T>(context.GetPlace());

    int context_start = context.Attr<int>("context_start");
    int context_length = context.Attr<int>("context_length");
    int context_stride = context.Attr<int>("context_stride");
    bool padding_trainable = context.Attr<bool>("padding_trainable");

    // InferShape by in_lod
    PADDLE_ENFORCE_EQ(in->lod().size(), 1UL,
                      "Only support one level sequence now.");

    const LoDTensor* padding_data = nullptr;
    if (padding_trainable) {
      padding_data = context.Input<LoDTensor>("PaddingData");
    }

    int up_pad = std::max(0, -context_start);
    int down_pad = std::max(0, context_start + context_length - 1);
    int sequence_width;
    sequence_width = static_cast<int>(in->dims()[1]);

    // use col_shape in the im2col calculation
    framework::DDim col_shape = {in->dims()[0],
                                 sequence_width * context_length};
    LoDTensor col;
    col.mutable_data<T>(col_shape, context.GetPlace());
    // Because if padding_trainable is false, padding data should be zeros.
    auto temp = framework::EigenVector<T>::Flatten(col);
    temp.device(context.GetEigenDevice<Place>()) =
        temp.constant(static_cast<T>(0));

    paddle::operators::math::SequenceProjectFunctor<Place, T>
        seq_project_functor;

    seq_project_functor(context.device_context(), in, padding_data, &col,
                        padding_trainable, context_start, context_length,
                        context_stride, up_pad, down_pad);

    filter.Resize(framework::make_ddim({context_length * sequence_width, 1}));
    math::matmul<Place, T>(context.device_context(), col, false, filter, false,
                           T(1.0), out, T(0.0));
  }
};

template <typename Place, typename T>
class SequenceConvGradKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& context) const override {
    auto* out_g = context.Input<LoDTensor>(framework::GradVarName("Out"));
    auto* in_g = context.Output<LoDTensor>(framework::GradVarName("X"));
    auto* filter_g =
        context.Output<LoDTensor>(framework::GradVarName("Filter"));
    auto* padding_data_g =
        context.Output<LoDTensor>(framework::GradVarName("PaddingData"));
    auto* in = context.Input<LoDTensor>("X");
    auto* filter = context.Input<LoDTensor>("Filter");

    auto place = context.GetEigenDevice<Place>();

    int context_start = context.Attr<int>("context_start");
    int context_length = context.Attr<int>("context_length");
    int context_stride = context.Attr<int>("context_stride");
    bool padding_trainable = context.Attr<bool>("padding_trainable");

    // InferShape by in_lod
    PADDLE_ENFORCE_EQ(in->lod().size(), 1UL,
                      "Only support one level sequence now.");
    auto lod_g_level_0 = in->lod()[0];

    int up_pad = std::max(0, -context_start);
    int down_pad = std::max(0, context_start + context_length - 1);
    int sequence_height, sequence_width;
    int input_row_begin, input_row_end;

    sequence_width = static_cast<int>(in->dims()[1]);

    // use col_shape in the im2col calculation
    framework::DDim col_shape = {in->dims()[0],
                                 sequence_width * context_length};
    LoDTensor col;

    if (in_g || filter_g || (padding_trainable && padding_data_g)) {
      col.mutable_data<T>(col_shape, context.GetPlace());
      // Because if padding_trainable is false, padding data should be zeros.
      auto temp = framework::EigenVector<T>::Flatten(col);
      temp.device(context.GetEigenDevice<Place>()) =
          temp.constant(static_cast<T>(0));
      math::matmul<Place, T>(context.device_context(), *out_g, false, *filter,
                             true, T(1.0), &col, T(1.0));
    }

    if (in_g) {
      in_g->mutable_data<T>(context.GetPlace());

      math::SetConstant<Place, T> functor;
      functor(context.device_context(), in_g, 0);

      paddle::operators::math::Col2ImFunctor<
          paddle::operators::math::ColFormat::kOCF, Place, float>
          col2im_ocf;

      for (int i = 0; i < static_cast<int>(lod_g_level_0.size()) - 1; ++i) {
        input_row_begin =
            (context_start > 0)
                ? static_cast<int>(lod_g_level_0[i]) + context_start
                : static_cast<int>(lod_g_level_0[i]);
        input_row_end = static_cast<int>(lod_g_level_0[i + 1]);

        Tensor col_t = col.Slice(static_cast<int>(lod_g_level_0[i]),
                                 static_cast<int>(lod_g_level_0[i + 1]));

        sequence_height = static_cast<int>(col_t.dims()[0]);

        if (input_row_begin < input_row_end) {
          Tensor in_t = in_g->Slice(input_row_begin, input_row_end);

          std::vector<int64_t> output_shape(
              {sequence_height, 1, 1, context_length,
               sequence_width});  // output_height, output_width,
                                  // input_channels, filter_height, filter_width
          col_t.Resize(framework::make_ddim(output_shape));

          std::vector<int64_t> input_shape(
              {1, input_row_end - input_row_begin,
               sequence_width});  // input_channels, input_height, input_width
          in_t.Resize(framework::make_ddim(input_shape));

          col2im_ocf(context.device_context(), in_t, col_t,
                     /*stride_height*/ context_stride, /*stride_width*/ 0,
                     up_pad, down_pad);
        }
        col_t.Resize(framework::make_ddim(
            {sequence_height, context_length * sequence_width}));
      }
    }

    if (padding_trainable && padding_data_g) {
      padding_data_g->mutable_data<T>(context.GetPlace());

      math::SetConstant<Place, T> functor;
      functor(context.device_context(), padding_data_g, 0);

      for (int i = 0; i < static_cast<int>(lod_g_level_0.size()) - 1; ++i) {
        input_row_begin =
            (context_start > 0)
                ? static_cast<int>(lod_g_level_0[i]) + context_start
                : static_cast<int>(lod_g_level_0[i]);
        input_row_end = static_cast<int>(lod_g_level_0[i + 1]);

        Tensor col_t = col.Slice(static_cast<int>(lod_g_level_0[i]),
                                 static_cast<int>(lod_g_level_0[i + 1]));

        sequence_height = static_cast<int>(col_t.dims()[0]);

        col_t.Resize(framework::make_ddim(
            {sequence_height * context_length, sequence_width}));

        if (up_pad > 0) {  // add up pad
          int padding_rows = std::min(
              up_pad,
              static_cast<int>(lod_g_level_0[i + 1] - lod_g_level_0[i]));

          for (int k = 0; k < padding_rows; ++k) {
            int padding_size =
                k + context_length < up_pad ? context_length : up_pad - k;
            Tensor out_t_sub = col_t.Slice(k * context_length,
                                           k * context_length + padding_size);
            Tensor w_sub = padding_data_g->Slice(k, k + padding_size);
            // in this block, using EigenVector<T>::Flatten is ok too.
            auto out_t_sub_e = EigenMatrix<T>::From(out_t_sub);
            auto w_sub_e = EigenMatrix<T>::From(w_sub);
            w_sub_e.device(place) = w_sub_e + out_t_sub_e;
          }
        }
        if (down_pad > 0) {  // add down pad
          int down_pad_begin_row =
              std::max(0,
                       (sequence_height - context_start - context_length) + 1) +
              1;
          int padding_begin = std::max(0, context_start - sequence_height);
          int padding_size =
              sequence_height - context_start >= context_length
                  ? 1
                  : context_length - (sequence_height - context_start);
          if (context_start >= sequence_height) padding_size = context_length;
          int padding_idx = padding_begin;
          for (int t = 0; t + down_pad_begin_row <= sequence_height;
               ++t, ++padding_size) {
            if (context_start >= sequence_height) padding_size = context_length;
            if (padding_size > context_length) {
              padding_size = context_length;
              padding_idx++;
            }
            if (padding_begin > 0 || sequence_height == context_start)
              padding_idx = padding_begin + t;
            Tensor out_t_sub = col_t.Slice(
                (down_pad_begin_row + t) * context_length - padding_size,
                (down_pad_begin_row + t) * context_length);
            Tensor w_sub = padding_data_g->Slice(
                up_pad + padding_idx, up_pad + padding_idx + padding_size);
            auto out_t_sub_e = EigenMatrix<T>::From(out_t_sub);
            auto w_sub_e = EigenMatrix<T>::From(w_sub);
            w_sub_e.device(place) = w_sub_e + out_t_sub_e;
          }
        }
        col_t.Resize(framework::make_ddim(
            {sequence_height, context_length * sequence_width}));
      }
    }

    if (filter_g) {
      filter_g->mutable_data<T>(context.GetPlace());

      math::SetConstant<Place, T> functor;
      functor(context.device_context(), filter_g, 0);

      Tensor filter_grad_ = *filter_g;
      Tensor out_grad_ = *out_g;

      const LoDTensor* padding_data = nullptr;
      if (padding_trainable) {
        padding_data = context.Input<LoDTensor>("PaddingData");
      }

      sequence_width = static_cast<int>(in->dims()[1]);

      paddle::operators::math::SequenceProjectFunctor<Place, T>
          seq_project_functor;

      seq_project_functor(context.device_context(), in, padding_data, &col,
                          padding_trainable, context_start, context_length,
                          context_stride, up_pad, down_pad);

      filter_grad_.Resize(
          framework::make_ddim({context_length * sequence_width, 1}));

      math::matmul<Place, T>(context.device_context(), col, true, out_grad_,
                             false, T(1.0), &filter_grad_, T(1.0));
    }
  }
};

}  // namespace operators
}  // namespace paddle