/* Copyright (c) 2016 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. */

#pragma once
#include <string>
#include <vector>

#include "paddle/fluid/framework/eigen.h"
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/operators/activation_op.h"
#include "paddle/fluid/platform/place.h"
#include "paddle/fluid/platform/transform.h"
#include "paddle/phi/kernels/funcs/blas/blas.h"
#include "paddle/phi/kernels/funcs/detail/activation_functions.h"
#include "paddle/phi/kernels/funcs/lstm_compute.h"
#include "paddle/phi/kernels/funcs/sequence2batch.h"

namespace paddle {
namespace operators {

using platform::Transform;

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

template <typename T>
class _ClipFunctor {
 public:
  explicit _ClipFunctor(const T min, const T max) : min_(min), max_(max) {}
  HOSTDEVICE T operator()(const T& x) const {
    if (x < min_)
      return min_;
    else if (x > max_)
      return max_;
    else
      return x;
  }

 private:
  T min_;
  T max_;
};

template <typename T>
class _ClipGradFunctor {
 public:
  explicit _ClipGradFunctor(const T min, const T max) : min_(min), max_(max) {}
  HOSTDEVICE T operator()(const T& x, const T& y) const {
    return (y > min_ && y < max_) ? x : 0;
  }

 private:
  T min_;
  T max_;
};

template <typename DeviceContext, typename T>
inline void ReorderInitState(const DeviceContext& ctx,
                             const phi::DenseTensor& src,
                             phi::Vector<size_t> index,
                             phi::DenseTensor* dst,
                             bool indexed_src) {
  phi::funcs::CopyMatrixRowsFunctor<DeviceContext, T> row_shuffle;
  dst->mutable_data<T>(src.dims(), ctx.GetPlace());
  row_shuffle(ctx, src, index, dst, indexed_src);
}

template <typename DeviceContext, typename T>
class LSTMPKernel : public framework::OpKernel<T> {
 public:
  template <typename Device, typename X, typename Y>
  void ActCompute(const phi::funcs::detail::ActivationType act_type,
                  const Device& d,
                  X x,
                  Y y,
                  platform::Place place) const {
    if (act_type == phi::funcs::detail::ActivationType::kIdentity) {
      y.device(d) = x;
    } else if (act_type == phi::funcs::detail::ActivationType::kSigmoid) {
      SigmoidFunctor<T>()(d, x, y);
    } else if (act_type == phi::funcs::detail::ActivationType::kTanh) {
      TanhFunctor<T>()(d, x, y);
    } else if (act_type == phi::funcs::detail::ActivationType::kReLU) {
      if (place == platform::CPUPlace())
        ReluCPUFunctor<T>()(d, x, y);
      else
        ReluCUDAFunctor<T>()(d, x, y);
    } else {
      PADDLE_THROW(
          platform::errors::InvalidArgument("unsupported activation type"));
    }
  }

  void Compute(const framework::ExecutionContext& ctx) const override {
    auto* input = ctx.Input<phi::DenseTensor>("Input");
    auto* weight = ctx.Input<phi::DenseTensor>("Weight");
    auto* proj_weight = ctx.Input<phi::DenseTensor>("ProjWeight");
    auto* bias = ctx.Input<phi::DenseTensor>("Bias");

    auto* hidden_t0 = ctx.Input<phi::DenseTensor>("H0");
    auto* cell_t0 = ctx.Input<phi::DenseTensor>("C0");

    auto proj_clip = static_cast<T>(ctx.Attr<float>("proj_clip"));
    auto cell_clip = static_cast<T>(ctx.Attr<float>("cell_clip"));

    auto* batch_gate = ctx.Output<phi::DenseTensor>("BatchGate");
    batch_gate->mutable_data<T>(ctx.GetPlace());
    auto* proj_out = ctx.Output<phi::DenseTensor>("Projection");
    proj_out->mutable_data<T>(ctx.GetPlace());
    auto* cell_out = ctx.Output<phi::DenseTensor>("Cell");
    cell_out->mutable_data<T>(ctx.GetPlace());

    bool is_reverse = ctx.Attr<bool>("is_reverse");
    phi::funcs::LoDTensor2BatchFunctor<DeviceContext, T> to_batch;
    auto& device_ctx = ctx.template device_context<DeviceContext>();
    to_batch(device_ctx, *input, batch_gate, true, is_reverse);

    auto in_dims = input->dims();
    int frame_size = static_cast<int>(in_dims[1] / 4);
    framework::DDim dims({in_dims[0], frame_size});
    framework::DDim proj_dims({in_dims[0], proj_weight->dims()[1]});

    if (bias) {
      phi::DenseTensor b = *bias;
      b.Resize({bias->numel(), 1});
      phi::DenseTensor gate_bias = b.Slice(0, 4 * frame_size);
      phi::funcs::RowwiseAdd<DeviceContext, T> add_bias;
      add_bias(device_ctx, *batch_gate, gate_bias, batch_gate);
    }

    phi::funcs::LstmMetaValue<T> lstmp_value;
    if (bias && ctx.Attr<bool>("use_peepholes")) {
      T* bias_data = const_cast<T*>(bias->data<T>());
      // the code style in LstmpMetaValue will be updated later.

      lstmp_value.check_ig = bias_data + 4 * frame_size;
      lstmp_value.check_fg = lstmp_value.check_ig + frame_size;
      lstmp_value.check_og = lstmp_value.check_fg + frame_size;
    } else {
      lstmp_value.check_ig = nullptr;
      lstmp_value.check_fg = nullptr;
      lstmp_value.check_og = nullptr;
    }
    lstmp_value.prev_state_value = nullptr;
    phi::DenseTensor ordered_c0;
    phi::DenseTensor ordered_h0;

    phi::Vector<size_t> order(batch_gate->lod()[2]);

    if (cell_t0) {
      // Since the batch computing for LSTMP reorders the input sequence
      // according to their length. The initialized cell state also needs
      // to reorder.
      ReorderInitState<DeviceContext, T>(
          device_ctx, *cell_t0, order, &ordered_c0, true);
      lstmp_value.prev_state_value = ordered_c0.data<T>();
    }

    // Use the local variable as here.
    phi::DenseTensor batch_proj, batch_cell;
    auto* batch_cell_pre_act = ctx.Output<phi::DenseTensor>("BatchCellPreAct");
    batch_cell_pre_act->mutable_data<T>(dims, ctx.GetPlace());
    auto* batch_hidden = ctx.Output<phi::DenseTensor>("BatchHidden");
    batch_hidden->mutable_data<T>(dims, ctx.GetPlace());    // T x D
    batch_proj.mutable_data<T>(proj_dims, ctx.GetPlace());  // T x P
    batch_cell.mutable_data<T>(dims, ctx.GetPlace());       // T x D

    auto batch_starts = batch_gate->lod()[0];
    size_t num_batch = batch_starts.size() - 1;
    auto gate_act = phi::funcs::detail::GetActivationType(
        ctx.Attr<std::string>("gate_activation"));
    auto cell_act = phi::funcs::detail::GetActivationType(
        ctx.Attr<std::string>("cell_activation"));
    auto cand_act = phi::funcs::detail::GetActivationType(
        ctx.Attr<std::string>("candidate_activation"));
    auto proj_act = phi::funcs::detail::GetActivationType(
        ctx.Attr<std::string>("proj_activation"));
    auto& place = *ctx.template device_context<DeviceContext>().eigen_device();
    auto blas = phi::funcs::GetBlas<DeviceContext, T>(device_ctx);
    for (size_t n = 0; n < num_batch; n++) {
      int bstart = static_cast<int>(batch_starts[n]);
      int bend = static_cast<int>(batch_starts[n + 1]);

      phi::DenseTensor gate_t = batch_gate->Slice(bstart, bend);
      phi::DenseTensor hidden_t = batch_hidden->Slice(bstart, bend);
      phi::DenseTensor proj_t = batch_proj.Slice(bstart, bend);
      phi::DenseTensor cell_t = batch_cell.Slice(bstart, bend);
      phi::DenseTensor cell_pre_act_t = batch_cell_pre_act->Slice(bstart, bend);

      int cur_batch_size = bend - bstart;

      if (n > 0) {
        int pre_h_start = static_cast<int>(batch_starts[n - 1]);
        int pre_h_end = pre_h_start + cur_batch_size;
        auto pre_proj_t = batch_proj.Slice(pre_h_start, pre_h_end);
        blas.MatMul(pre_proj_t,
                    false,
                    *weight,
                    false,
                    static_cast<T>(1.0),
                    &gate_t,
                    static_cast<T>(1.0));
      } else if (hidden_t0) {
        // If n == 0 and there is no initialized hidden state, that is to say
        // the H0 is zeros, the calculation W_h * H0 will be skiped.
        // If n == 0 and there is initialized hidden state, calculate W_h * H0.

        // Since the batch computing for LSTMP reorders the input sequence
        // according to their length. The initialized hidden state also needs
        // to reorder.
        ReorderInitState<DeviceContext, T>(
            device_ctx, *hidden_t0, order, &ordered_h0, true);
        blas.MatMul(ordered_h0,
                    false,
                    *weight,
                    false,
                    static_cast<T>(1.0),
                    &gate_t,
                    static_cast<T>(1.0));
      }

      lstmp_value.gate_value = gate_t.data<T>();
      lstmp_value.output_value = hidden_t.data<T>();
      lstmp_value.state_value = cell_t.data<T>();
      lstmp_value.state_active_value = cell_pre_act_t.data<T>();
      phi::funcs::LstmUnitFunctor<DeviceContext, T>::compute(device_ctx,
                                                             lstmp_value,
                                                             frame_size,
                                                             cur_batch_size,
                                                             cell_clip,
                                                             gate_act,
                                                             cell_act,
                                                             cand_act);
      lstmp_value.prev_state_value = lstmp_value.state_value;
      blas.MatMul(hidden_t,
                  false,
                  *proj_weight,
                  false,
                  static_cast<T>(1.0),
                  &proj_t,
                  static_cast<T>(0.0));
      if (proj_act != phi::funcs::detail::ActivationType::kIdentity) {
        auto proj_t_dev = EigenMatrix<T>::From(proj_t);
        ActCompute(cell_act, place, proj_t_dev, proj_t_dev, ctx.GetPlace());
      }
      if (proj_clip && proj_clip > 0.0) {
        T* x_data = proj_t.data<T>();
        int64_t numel = proj_t.numel();
        Transform<DeviceContext> trans;
        trans(ctx.template device_context<DeviceContext>(),
              x_data,
              x_data + numel,
              x_data,
              _ClipFunctor<T>(-1.0 * proj_clip, proj_clip));
      }
    }

    phi::funcs::Batch2LoDTensorFunctor<DeviceContext, T> to_seq;
    batch_proj.set_lod(batch_gate->lod());
    // restore the output hidden in phi::DenseTensor from the batch hidden
    to_seq(device_ctx, batch_proj, proj_out);

    batch_cell.set_lod(batch_gate->lod());
    // restore the output cell state in phi::DenseTensor from the batch cell
    to_seq(device_ctx, batch_cell, cell_out);
  }
};

template <typename DeviceContext, typename T>
class LSTMPGradKernel : public framework::OpKernel<T> {
 public:
  template <typename Device, typename X, typename Y, typename DX, typename DY>
  void ActGradCompute(const phi::funcs::detail::ActivationType act_type,
                      const Device& d,
                      X x,
                      Y y,
                      DX dx,
                      DY dy) const {
    // x is dummy and won't be used even in Relu(use y instead)
    if (act_type == phi::funcs::detail::ActivationType::kIdentity)
      dx.device(d) = dy;
    else if (act_type == phi::funcs::detail::ActivationType::kSigmoid)
      SigmoidGradFunctor<T>()(d, x, y, dy, dx);
    else if (act_type == phi::funcs::detail::ActivationType::kTanh)
      TanhGradFunctor<T>()(d, x, y, dy, dx);
    else if (act_type == phi::funcs::detail::ActivationType::kReLU)
      ReluGradFunctor<T>()(d, x, y, dy, dx);
    else
      PADDLE_THROW(
          platform::errors::InvalidArgument("unsupported activation type"));
  }

  void Compute(const framework::ExecutionContext& ctx) const override {
    auto* weight = ctx.Input<phi::DenseTensor>("Weight");
    auto* proj_weight = ctx.Input<phi::DenseTensor>("ProjWeight");
    auto* bias = ctx.Input<phi::DenseTensor>("Bias");

    auto* proj_out = ctx.Input<phi::DenseTensor>("Projection");
    auto* cell_out = ctx.Input<phi::DenseTensor>("Cell");

    auto proj_clip = static_cast<T>(ctx.Attr<float>("proj_clip"));
    auto cell_clip = static_cast<T>(ctx.Attr<float>("cell_clip"));

    auto* batch_gate = ctx.Input<phi::DenseTensor>("BatchGate");
    auto* batch_cell_pre_act = ctx.Input<phi::DenseTensor>("BatchCellPreAct");
    auto* batch_hidden = ctx.Input<phi::DenseTensor>("BatchHidden");

    auto* projection_g =
        ctx.Input<phi::DenseTensor>(framework::GradVarName("Projection"));

    auto* in_g = ctx.Output<phi::DenseTensor>(framework::GradVarName("Input"));
    auto* weight_g =
        ctx.Output<phi::DenseTensor>(framework::GradVarName("Weight"));
    auto* proj_weight_g =
        ctx.Output<phi::DenseTensor>(framework::GradVarName("ProjWeight"));
    auto* bias_g = ctx.Output<phi::DenseTensor>(framework::GradVarName("Bias"));

    auto* h0 = ctx.Input<phi::DenseTensor>("H0");
    auto* c0 = ctx.Input<phi::DenseTensor>("C0");

    auto* h0_g = ctx.Output<phi::DenseTensor>(framework::GradVarName("H0"));
    auto* c0_g = ctx.Output<phi::DenseTensor>(framework::GradVarName("C0"));

    auto& device_ctx = ctx.template device_context<DeviceContext>();
    phi::funcs::SetConstant<DeviceContext, T> zero;
    if (weight_g) {
      weight_g->mutable_data<T>(ctx.GetPlace());
      zero(device_ctx, weight_g, static_cast<T>(0.0));
    }
    if (proj_weight_g) {
      proj_weight_g->mutable_data<T>(ctx.GetPlace());
      zero(device_ctx, proj_weight_g, static_cast<T>(0.0));
    }

    // ordered_h0/c0 is the reordered hidden/cell initialization.
    // ordered_h0_g/c0_g is the reordered gradient of hidden/cell
    // initialization.
    phi::DenseTensor ordered_h0, ordered_c0, ordered_h0_g, ordered_c0_g;

    phi::Vector<size_t> order(batch_gate->lod()[2]);

    if (c0) {
      ReorderInitState<DeviceContext, T>(
          device_ctx, *c0, order, &ordered_c0, true);
    }
    if (c0 && c0_g) {
      ordered_c0_g.mutable_data<T>(c0_g->dims(), ctx.GetPlace());
    }

    // batch_gate dims equal to input dims
    auto in_dims = batch_gate->dims();
    auto out_dims = cell_out->dims();
    framework::DDim proj_dims({in_dims[0], proj_weight->dims()[1]});
    int frame_size = static_cast<int>(in_dims[1] / 4);
    PADDLE_ENFORCE_EQ(frame_size,
                      out_dims[1],
                      platform::errors::InvalidArgument(
                          "The second dimension of Input(Cell) should be %d, "
                          "but received %d in LSTMP@Grad operator.",
                          frame_size,
                          out_dims[1]));

    phi::funcs::LstmMetaValue<T> lstmp_value;
    if (bias && ctx.Attr<bool>("use_peepholes")) {
      T* bias_data = const_cast<T*>(bias->data<T>());
      lstmp_value.check_ig = bias_data + 4 * frame_size;
      lstmp_value.check_fg = lstmp_value.check_ig + frame_size;
      lstmp_value.check_og = lstmp_value.check_fg + frame_size;
    } else {
      lstmp_value.check_ig = nullptr;
      lstmp_value.check_fg = nullptr;
      lstmp_value.check_og = nullptr;
    }

    phi::funcs::LstmMetaGrad<T> lstmp_grad;

    if (bias && bias_g) {
      bias_g->mutable_data<T>(ctx.GetPlace());
      zero(device_ctx, bias_g, static_cast<T>(0.0));
    }
    if (bias && bias_g && ctx.Attr<bool>("use_peepholes")) {
      T* bias_g_data = bias_g->data<T>();
      lstmp_grad.check_ig_grad = bias_g_data + 4 * frame_size;
      lstmp_grad.check_fg_grad = lstmp_grad.check_ig_grad + frame_size;
      lstmp_grad.check_og_grad = lstmp_grad.check_fg_grad + frame_size;
    } else {
      lstmp_grad.check_ig_grad = nullptr;
      lstmp_grad.check_fg_grad = nullptr;
      lstmp_grad.check_og_grad = nullptr;
    }

    phi::funcs::LoDTensor2BatchFunctor<DeviceContext, T> to_batch;

    auto ToBatch = [&batch_gate, &to_batch](const DeviceContext& ctx,
                                            const phi::DenseTensor& src,
                                            const framework::DDim& dims,
                                            phi::DenseTensor& dst) {
      dst.mutable_data<T>(dims, ctx.GetPlace());
      dst.set_lod(batch_gate->lod());
      to_batch(ctx, src, &dst, false);
    };

    phi::DenseTensor batch_hidden_g, batch_proj, batch_proj_g, batch_cell;
    batch_hidden_g.mutable_data<T>(out_dims, ctx.GetPlace());
    ToBatch(device_ctx, *proj_out, proj_dims, batch_proj);        // T x P
    ToBatch(device_ctx, *projection_g, proj_dims, batch_proj_g);  // T x P
    ToBatch(device_ctx, *cell_out, out_dims, batch_cell);         // T x D

    phi::DenseTensor batch_cell_g, batch_gate_g;
    batch_cell_g.mutable_data<T>(out_dims, ctx.GetPlace());
    // TODO(qingqing) support the case output cell has gradient.
    // to_batch(device_ctx, *cell_g, batch_cell_g, false);
    zero(device_ctx, &batch_cell_g, static_cast<T>(0.0));
    batch_gate_g.mutable_data<T>(batch_gate->dims(), ctx.GetPlace());
    batch_gate_g.set_lod(batch_gate->lod());

    auto gate_act = phi::funcs::detail::GetActivationType(
        ctx.Attr<std::string>("gate_activation"));
    auto cell_act = phi::funcs::detail::GetActivationType(
        ctx.Attr<std::string>("cell_activation"));
    auto cand_act = phi::funcs::detail::GetActivationType(
        ctx.Attr<std::string>("candidate_activation"));
    auto proj_act = phi::funcs::detail::GetActivationType(
        ctx.Attr<std::string>("proj_activation"));
    auto& place = *ctx.template device_context<DeviceContext>().eigen_device();

    auto batch_starts = batch_gate->lod()[0];
    size_t num_batch = batch_starts.size() - 1;
    auto blas = phi::funcs::GetBlas<DeviceContext, T>(device_ctx);
    for (int n = static_cast<int>(num_batch) - 1; n >= 0; n--) {
      int bstart = static_cast<int>(batch_starts[n]);
      int bend = static_cast<int>(batch_starts[n + 1]);

      phi::DenseTensor cur_proj = batch_proj.Slice(bstart, bend);
      phi::DenseTensor proj_g = batch_proj_g.Slice(bstart, bend);

      if (proj_clip && proj_clip > 0.0) {
        T* dx_data = proj_g.data<T>();
        T* x_data = cur_proj.data<T>();
        int64_t numel = proj_g.numel();
        Transform<DeviceContext> trans;
        trans(ctx.template device_context<DeviceContext>(),
              dx_data,
              dx_data + numel,
              x_data,
              dx_data,
              _ClipGradFunctor<T>(-1.0 * proj_clip, proj_clip));
      }

      if (proj_act != phi::funcs::detail::ActivationType::kIdentity) {
        auto cur_proj_dev = EigenMatrix<T>::From(cur_proj);
        auto proj_g_dev = EigenMatrix<T>::From(proj_g);
        ActGradCompute(cell_act,
                       place,
                       cur_proj_dev,
                       cur_proj_dev,
                       proj_g_dev,
                       proj_g_dev);
      }
      /* hidden state backwarad */
      phi::DenseTensor out_g = batch_hidden_g.Slice(bstart, bend);
      blas.MatMul(proj_g,
                  false,
                  *proj_weight,
                  true,
                  static_cast<T>(1.0),
                  &out_g,
                  static_cast<T>(0.0));
      /* projection weight backward*/
      if (proj_weight_g) {
        phi::DenseTensor hidden_t = batch_hidden->Slice(bstart, bend);
        blas.MatMul(hidden_t,
                    true,
                    proj_g,
                    false,
                    static_cast<T>(1.0),
                    proj_weight_g,
                    static_cast<T>(1.0));
      }

      phi::DenseTensor gate = batch_gate->Slice(bstart, bend);
      phi::DenseTensor cell = batch_cell.Slice(bstart, bend);
      phi::DenseTensor cell_pre_act = batch_cell_pre_act->Slice(bstart, bend);
      lstmp_value.gate_value = gate.data<T>();
      lstmp_value.state_value = cell.data<T>();
      lstmp_value.state_active_value = cell_pre_act.data<T>();

      phi::DenseTensor gate_g = batch_gate_g.Slice(bstart, bend);
      phi::DenseTensor cell_g = batch_cell_g.Slice(bstart, bend);
      lstmp_grad.state_grad = cell_g.data<T>();
      lstmp_grad.gate_grad = gate_g.data<T>();
      lstmp_grad.output_grad = out_g.data<T>();

      if (n > 0) {
        int bstart_pre = static_cast<int>(batch_starts[n - 1]);
        phi::DenseTensor cell_pre = batch_cell.Slice(bstart_pre, bstart);
        phi::DenseTensor cell_pre_g = batch_cell_g.Slice(bstart_pre, bstart);
        lstmp_value.prev_state_value = cell_pre.data<T>();
        lstmp_grad.prev_state_grad = cell_pre_g.data<T>();
      } else {
        lstmp_value.prev_state_value = c0 ? ordered_c0.data<T>() : nullptr;
        lstmp_grad.prev_state_grad = c0_g ? ordered_c0_g.data<T>() : nullptr;
      }

      int cur_batch_size = bend - bstart;
      // lstmp_value.output_value not used in bp, set to null
      // lstmp_grad.state_active_grad not used in bp, set to null
      lstmp_value.output_value = nullptr;
      lstmp_grad.state_active_grad = nullptr;

      phi::funcs::LstmUnitGradFunctor<DeviceContext, T>::compute(device_ctx,
                                                                 lstmp_value,
                                                                 lstmp_grad,
                                                                 frame_size,
                                                                 cur_batch_size,
                                                                 cell_clip,
                                                                 gate_act,
                                                                 cell_act,
                                                                 cand_act);

      if (n > 0) {
        int pre_h_start = static_cast<int>(batch_starts[n - 1]);
        int pre_h_end = pre_h_start + cur_batch_size;
        auto pre_proj_g = batch_proj_g.Slice(pre_h_start, pre_h_end);
        blas.MatMul(gate_g,
                    false,
                    *weight,
                    true,
                    static_cast<T>(1.0),
                    &pre_proj_g,
                    static_cast<T>(1.0));
        if (weight_g) {
          /* weight backward*/
          auto pre_proj = batch_proj.Slice(pre_h_start, pre_h_end);
          blas.MatMul(pre_proj,
                      true,
                      gate_g,
                      false,
                      static_cast<T>(1.0),
                      weight_g,
                      static_cast<T>(1.0));
        }
      } else {
        if (h0 && weight_g) {
          ReorderInitState<DeviceContext, T>(
              device_ctx, *h0, order, &ordered_h0, true);
          if (weight_g) {
            blas.MatMul(ordered_h0,
                        true,
                        gate_g,
                        false,
                        static_cast<T>(1.0),
                        weight_g,
                        static_cast<T>(1.0));
          }
        }
        if (h0 && (h0_g || proj_weight_g)) {
          ordered_h0_g.mutable_data<T>(h0_g->dims(), ctx.GetPlace());
          blas.MatMul(gate_g,
                      false,
                      *weight,
                      true,
                      static_cast<T>(1.0),
                      &ordered_h0_g,
                      static_cast<T>(0.0));
        }
      }
    }

    phi::funcs::Batch2LoDTensorFunctor<DeviceContext, T> to_seq;
    if (in_g) {
      /* backward data */
      in_g->mutable_data<T>(ctx.GetPlace());
      to_seq(device_ctx, batch_gate_g, in_g);
    }
    if (bias && bias_g) {
      /* backward bias */
      phi::DenseTensor b_g = *bias_g;
      b_g.Resize({bias_g->numel(), 1});
      phi::DenseTensor gate_bias_g = b_g.Slice(0, 4 * frame_size);
      phi::funcs::ColwiseSum<DeviceContext, T> col_sum;
      col_sum(device_ctx, batch_gate_g, &gate_bias_g);
    }

    if (h0 && h0_g) {
      ReorderInitState<DeviceContext, T>(
          device_ctx, ordered_h0_g, order, h0_g, false);
    }
    if (c0 && c0_g) {
      ReorderInitState<DeviceContext, T>(
          device_ctx, ordered_c0_g, order, c0_g, false);
    }
  }
};

}  // namespace operators
}  // namespace paddle