/* 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. */
#include <algorithm>
#include <map>
#include <set>
#include <vector>

#include "paddle/phi/kernels/funcs/blas/blas.h"
#include "paddle/phi/kernels/funcs/selected_rows_functor.h"

#include "paddle/common/ddim.h"
#include "paddle/phi/core/mixed_vector.h"

#ifdef PADDLE_WITH_XPU
#include "paddle/phi/backends/xpu/enforce_xpu.h"
#endif

#ifdef PADDLE_WITH_DNNL
#include "paddle/phi/backends/onednn/axpy_handler.h"
#endif

#include "glog/logging.h"

namespace phi::funcs {
template <typename T>
struct SelectedRowsAdd<CPUContext, T> {
  void operator()(const CPUContext& dev_ctx,
                  const SelectedRows& input1,
                  const SelectedRows& input2,
                  SelectedRows* output) {
    auto in1_height = input1.height();
    PADDLE_ENFORCE_EQ(
        in1_height,
        input2.height(),
        common::errors::InvalidArgument("The two inputs height must be equal."
                                        "But received first input height  = "
                                        "[%d], second input height = [%d]",
                                        in1_height,
                                        input2.height()));
    output->set_height(in1_height);

    auto& in1_rows = input1.rows();
    auto& in2_rows = input2.rows();
    std::vector<int64_t> out_rows;
    out_rows.reserve(in1_rows.size() + in2_rows.size());

    // concat rows
    out_rows.insert(out_rows.end(), in1_rows.begin(), in1_rows.end());
    out_rows.insert(out_rows.end(), in2_rows.begin(), in2_rows.end());
    output->set_rows(out_rows);

    auto* out_value = output->mutable_value();
    auto& in1_value = input1.value();
    auto& in2_value = input2.value();

    auto in1_row_numel = in1_value.numel() / in1_rows.size();
    PADDLE_ENFORCE_EQ(
        in1_row_numel,
        in2_value.numel() / in2_rows.size(),
        common::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But received first input width = [%d], second input width = [%d]",
            in1_row_numel,
            in2_value.numel() / in2_rows.size()));
    PADDLE_ENFORCE_EQ(
        in1_row_numel,
        out_value->numel() / out_rows.size(),
        common::errors::InvalidArgument(
            "The input and output width must be equal."
            "But received input width = [%d], output width = [%d]",
            in1_row_numel,
            out_value->numel() / out_rows.size()));

    auto in1_place = input1.place();
    PADDLE_ENFORCE_EQ(in1_place.GetType() == AllocationType::CPU,
                      true,
                      common::errors::InvalidArgument(
                          "The running environment is not on the CPU place."));
    auto in2_place = input2.place();
    PADDLE_ENFORCE_EQ(in2_place.GetType() == AllocationType::CPU,
                      true,
                      common::errors::InvalidArgument(
                          "The running environment is not on the CPU place."));
    auto out_place = dev_ctx.GetPlace();
    PADDLE_ENFORCE_EQ(out_place.GetType() == AllocationType::CPU,
                      true,
                      common::errors::InvalidArgument(
                          "The running environment is not on the CPU place."));

    auto* out_data = out_value->data<T>();
    auto* in1_data = in1_value.data<T>();
    memory_utils::Copy(out_place,
                       out_data,
                       in1_place,
                       in1_data,
                       in1_value.numel() * sizeof(T));

    auto* in2_data = in2_value.data<T>();
    memory_utils::Copy(out_place,
                       out_data + in1_value.numel(),
                       in2_place,
                       in2_data,
                       in2_value.numel() * sizeof(T));
  }
};

template struct PADDLE_API SelectedRowsAdd<CPUContext, float>;
template struct PADDLE_API SelectedRowsAdd<CPUContext, double>;

template <typename T>
struct SelectedRowsAddTensor<CPUContext, T> {
  void operator()(const CPUContext& dev_ctx,
                  const SelectedRows& input1,
                  const DenseTensor& input2,
                  DenseTensor* output) {
    auto in1_height = input1.height();
    const auto& in2_dims = input2.dims();
    const auto& out_dims = output->dims();
    PADDLE_ENFORCE_EQ(
        in1_height,
        in2_dims[0],
        common::errors::InvalidArgument("The two inputs height must be equal."
                                        "But received first input height = "
                                        "[%d], second input height = [%d]",
                                        in1_height,
                                        in2_dims[0]));
    PADDLE_ENFORCE_EQ(
        in1_height,
        out_dims[0],
        common::errors::InvalidArgument(
            "The input and output height must be equal."
            "But received input height = [%d], output height = [%d]",
            in1_height,
            out_dims[0]));

    auto& in1_value = input1.value();
    auto& in1_rows = input1.rows();

    int64_t in1_row_numel =
        static_cast<int64_t>(in1_value.numel() / in1_rows.size());
    PADDLE_ENFORCE_EQ(
        in1_row_numel,
        input2.numel() / in1_height,
        common::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But received first input width = [%d], second input width = [%d]",
            in1_row_numel,
            input2.numel() / in1_height));
    PADDLE_ENFORCE_EQ(
        in1_row_numel,
        output->numel() / in1_height,
        common::errors::InvalidArgument(
            "The input and output width must be equal."
            "But received input width = [%d], output width = [%d]",
            in1_row_numel,
            output->numel() / in1_height));

    phi::funcs::SetConstant<CPUContext, T> functor;
    functor(dev_ctx, output, static_cast<T>(0.0));

    auto* in1_data = in1_value.data<T>();
    auto* out_data = output->data<T>();

    for (size_t i = 0; i < in1_rows.size(); i++) {
      for (int64_t j = 0; j < in1_row_numel; j++) {
        out_data[in1_rows[i] * in1_row_numel + j] +=
            in1_data[i * in1_row_numel + j];
      }
    }

    auto out_eigen = EigenVector<T>::Flatten(*output);
    auto in2_eigen = EigenVector<T>::Flatten(input2);
    out_eigen.device(*dev_ctx.eigen_device()) = out_eigen + in2_eigen;
  }
};

template struct PADDLE_API SelectedRowsAddTensor<CPUContext, float>;
template struct PADDLE_API SelectedRowsAddTensor<CPUContext, double>;

template <typename T>
struct SelectedRowsAddTo<CPUContext, T> {
  void operator()(const CPUContext& dev_ctx UNUSED,
                  const SelectedRows& input1,
                  const int64_t input2_offset,
                  SelectedRows* input2) {
    auto in1_height = input1.height();
    PADDLE_ENFORCE_EQ(
        in1_height,
        input2->height(),
        common::errors::InvalidArgument("The two inputs height must be equal."
                                        "But received first input height = "
                                        "[%d], second input height = [%d]",
                                        in1_height,
                                        input2->height()));

    auto& in1_rows = input1.rows();
    auto& in2_rows = *(input2->mutable_rows());

    auto& in1_value = input1.value();
    auto* in2_value = input2->mutable_value();

    // concat rows
    phi::MixVector<int64_t> mixv_in2_rows(&in2_rows);
    mixv_in2_rows.Extend(in1_rows.begin(), in1_rows.end());

    auto in1_place = input1.place();
    PADDLE_ENFORCE_EQ(in1_place.GetType() == AllocationType::CPU,
                      true,
                      common::errors::InvalidArgument(
                          "The running environment is not on the CPU place."));
    auto in2_place = input2->place();
    PADDLE_ENFORCE_EQ(in2_place.GetType() == AllocationType::CPU,
                      true,
                      common::errors::InvalidArgument(
                          "The running environment is not on the CPU place."));

    auto* in1_data = in1_value.data<T>();
    auto* in2_data = in2_value->data<T>();
    memory_utils::Copy(in2_place,
                       in2_data + input2_offset,
                       in1_place,
                       in1_data,
                       in1_value.numel() * sizeof(T));
  }
};

template struct PADDLE_API SelectedRowsAddTo<CPUContext, float>;
template struct PADDLE_API SelectedRowsAddTo<CPUContext, double>;
template struct PADDLE_API SelectedRowsAddTo<CPUContext, int>;
template struct PADDLE_API SelectedRowsAddTo<CPUContext, int64_t>;

template <typename T>
struct SelectedRowsSumTo<CPUContext, T> {
  void operator()(const CPUContext& dev_ctx,
                  const std::vector<SelectedRows*>& input1,
                  const std::vector<int64_t>& input2_offsets,
                  SelectedRows* input2) {
    // Ensure all selected rows have the same height
    size_t size = 0u;
    for (auto item : input1) {
      auto& in_rows = item->rows();
      size += in_rows.end() - in_rows.begin();
      auto in1_height = item->height();
      PADDLE_ENFORCE_EQ(in1_height,
                        input2->height(),
                        common::errors::InvalidArgument(
                            "The two inputs height must be equal."
                            "But received first input height = [%d], second "
                            "input height = [%d]",
                            in1_height,
                            input2->height()));
    }
    // concat rows
    std::vector<int64_t> in2_rows;
    in2_rows.reserve(in2_rows.size() + size);
    for (auto item : input1) {
      const Vector<int64_t>& in_rows = item->rows();
      in2_rows.insert(in2_rows.end(), in_rows.begin(), in_rows.end());
    }
    input2->set_rows(in2_rows);

    auto* in2_value = input2->mutable_value();
    auto* in2_data = in2_value->data<T>();
    auto blas = phi::funcs::GetBlas<CPUContext, T>(dev_ctx);
    size_t offset = 0u;
    for (size_t i = 0u; i != input1.size(); ++i) {
      auto& in_value = input1[i]->value();
      const auto* in_data = in_value.data<T>();
      offset += input2_offsets[i];
      blas.VCOPY(in_value.numel(), in_data, in2_data + offset);
    }
  }
};

template struct PADDLE_API SelectedRowsSumTo<CPUContext, float>;
template struct PADDLE_API SelectedRowsSumTo<CPUContext, double>;

template <typename T>
struct SelectedRowsAddToTensor<CPUContext, T> {
  void operator()(const CPUContext& dev_ctx UNUSED,
                  const SelectedRows& input1,
                  DenseTensor* input2) {
    if (UNLIKELY(input1.rows().empty())) {
      LOG(WARNING) << "input selected rows is empty!";
      return;
    }
    auto in1_height = input1.height();
    const auto& in2_dims = input2->dims();
    PADDLE_ENFORCE_EQ(
        in1_height,
        in2_dims[0],
        common::errors::InvalidArgument("The two inputs height must be equal."
                                        "But received first input height = "
                                        "[%d], second input height = [%d]",
                                        in1_height,
                                        in2_dims[0]));

    auto& in1_value = input1.value();
    auto& in1_rows = input1.rows();

    int64_t in1_row_numel =
        static_cast<int64_t>(in1_value.numel() / in1_rows.size());
    PADDLE_ENFORCE_EQ(
        in1_row_numel,
        input2->numel() / in1_height,
        common::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But received first input width = [%d], second input width = [%d]",
            in1_row_numel,
            input2->numel() / in1_height));

    auto* in1_data = in1_value.data<T>();
    auto* input2_data = input2->data<T>();

    for (size_t i = 0; i < in1_rows.size(); i++) {
      for (int64_t j = 0; j < in1_row_numel; j++) {
        input2_data[in1_rows[i] * in1_row_numel + j] +=
            in1_data[i * in1_row_numel + j];
      }
    }
  }
};

#ifdef PADDLE_WITH_XPU
template <typename T>
struct SelectedRowsAddToTensor<XPUContext, T> {
  void operator()(const XPUContext& dev_ctx,
                  const SelectedRows& input1,
                  DenseTensor* input2) {
    if (UNLIKELY(input1.rows().size() == 0)) {
      LOG(WARNING) << "input selected rows is empty!";
      return;
    }
    using XPUType = typename XPUTypeTrait<T>::Type;
    auto in1_height = input1.height();
    const auto& in2_dims = input2->dims();
    PADDLE_ENFORCE_EQ(
        in1_height,
        in2_dims[0],
        common::errors::InvalidArgument("The two inputs height must be equal."
                                        "But received first input height = "
                                        "[%d], second input height = [%d]",
                                        in1_height,
                                        in2_dims[0]));

    auto& in1_value = input1.value();
    auto& in1_rows = input1.rows();
    int64_t* in1_rows_data = nullptr;
    xpu::VectorParam<int64_t> in1_rows_vec{
        in1_rows.data(), static_cast<int64_t>(in1_rows.size()), in1_rows_data};

    int64_t in1_row_numel = in1_value.numel() / in1_rows.size();
    PADDLE_ENFORCE_EQ(
        in1_row_numel,
        input2->numel() / in1_height,
        common::errors::InvalidArgument(
            "The two inputs width must be equal."
            "But received first input width = [%d], second input width = [%d]",
            in1_row_numel,
            input2->numel() / in1_height));

    auto* in1_data = in1_value.data<T>();
    auto* out_data = input2->data<T>();

    int64_t h = in1_rows.size();
    int64_t w = in1_row_numel;
    const std::vector<int64_t> xshape{h, w};

    int r = xpu::scatter<XPUType, int64_t>(
        dev_ctx.x_context(),
        nullptr,
        reinterpret_cast<const XPUType*>(in1_data),
        reinterpret_cast<XPUType*>(out_data),
        in1_rows_vec,
        xshape,
        0,
        false);
    PADDLE_ENFORCE_XDNN_SUCCESS(r, "scatter");
  }
};

#endif

template struct PADDLE_API SelectedRowsAddToTensor<CPUContext, float>;
template struct PADDLE_API SelectedRowsAddToTensor<CPUContext, double>;
template struct PADDLE_API SelectedRowsAddToTensor<CPUContext, int>;
template struct PADDLE_API SelectedRowsAddToTensor<CPUContext, int64_t>;
template struct PADDLE_API SelectedRowsAddToTensor<CPUContext, phi::float16>;
template struct PADDLE_API SelectedRowsAddToTensor<CPUContext, phi::bfloat16>;
template struct PADDLE_API SelectedRowsAddToTensor<CPUContext, phi::complex64>;
template struct PADDLE_API SelectedRowsAddToTensor<CPUContext, phi::complex128>;
#ifdef PADDLE_WITH_XPU
template struct SelectedRowsAddToTensor<XPUContext, float>;
#endif
// This is a separated namespace for manipulate SelectedRows typed
// data. Like merge duplicated rows, adding two SelectedRows etc.
//
// Another group of functors is called "scatter updates", which means
// use SelectedRows to update a dense tensor with different Ops, like
// add or mul.
}  // namespace phi::funcs
namespace phi::funcs::scatter {

template <typename T, typename DeviceContext>
typename std::enable_if<!std::is_integral<T>::value>::type elementwise_add_to(
    phi::funcs::BlasT<DeviceContext, T>* blas,
    size_t data_len,
    const T* in,
    T* out) {
  blas->AXPY(data_len, T(1.f), in, out);
}

template <typename T, typename DeviceContext>
typename std::enable_if<std::is_integral<T>::value>::type elementwise_add_to(
    phi::funcs::BlasT<DeviceContext, T>* blas UNUSED,
    size_t data_len,
    const T* in,
    T* out) {
  for (size_t i = 0; i < data_len; i++) {
    out[i] += in[i];
  }
}

template <typename T, typename DeviceContext>
typename std::enable_if<std::is_same<T, phi::bfloat16>::value>::type
add_sparse_inputs(const std::vector<const SelectedRows*>& inputs,
                  const std::unordered_map<int64_t, size_t>& rows_to_id,
                  int64_t input_width,
                  const DeviceContext& dev_ctx,
                  T* out_data) {
#ifndef PADDLE_WITH_DNNL
  auto blas = phi::funcs::GetBlas<DeviceContext, T>(dev_ctx);
#endif
  for (auto* input : inputs) {
    if (input->rows().empty()) {
      continue;
    }
    auto* input_data = input->value().data<T>();
    auto& input_rows = input->rows();

#ifdef PADDLE_WITH_DNNL
    OneDNNContext onednn_context(dev_ctx.GetPlace());
    funcs::OneDNNAXPYHandler<T> axpy_handler(
        input_width, T(1.f), onednn_context.GetEngine());
    for (size_t i = 0; i < input_rows.size(); i++) {
      size_t out_i = rows_to_id.at(input_rows[i]);
      axpy_handler(&input_data[i * input_width],
                   &out_data[out_i * input_width]);
    }
#else
    for (size_t i = 0; i < input_rows.size(); i++) {
      size_t out_i = rows_to_id.at(input_rows[i]);
      elementwise_add_to<T, DeviceContext>(&blas,
                                           static_cast<size_t>(input_width),
                                           &input_data[i * input_width],
                                           &out_data[out_i * input_width]);
    }
#endif
  }
}

template <typename T, typename DeviceContext>
typename std::enable_if<!std::is_same<T, phi::bfloat16>::value>::type
add_sparse_inputs(const std::vector<const SelectedRows*>& inputs,
                  const std::unordered_map<int64_t, size_t>& rows_to_id,
                  int64_t input_width,
                  const DeviceContext& dev_ctx,
                  T* out_data) {
  VLOG(4) << "[CPU] add_sparse_inputs <" << typeid(T).name();
  auto blas = phi::funcs::GetBlas<DeviceContext, T>(dev_ctx);
  for (auto* input : inputs) {
    if (input->rows().empty()) {
      continue;
    }
    auto* input_data = input->value().data<T>();
    auto& input_rows = input->rows();

    for (size_t i = 0; i < input_rows.size(); i++) {
      size_t out_i = rows_to_id.at(input_rows[i]);
      elementwise_add_to<T, DeviceContext>(&blas,
                                           static_cast<size_t>(input_width),
                                           &input_data[i * input_width],
                                           &out_data[out_i * input_width]);
    }
  }
}

template <typename DeviceContext, typename T>
struct MergeAddImpl {
  SelectedRows operator()(const DeviceContext& dev_ctx,
                          const SelectedRows& input,
                          const bool sorted_result = false) {
    SelectedRows out;
    (*this)(dev_ctx, input, &out, sorted_result);
    return out;
  }

  void operator()(const DeviceContext& dev_ctx,
                  const SelectedRows& input,
                  SelectedRows* output,
                  const bool sorted_result = false) {
    std::vector<const SelectedRows*> inputs;
    inputs.push_back(&input);
    (*this)(dev_ctx, inputs, output, sorted_result);
  }

  void operator()(const DeviceContext& dev_ctx,
                  const std::vector<const SelectedRows*>& inputs,
                  SelectedRows* output,
                  const bool sorted_result = false) {
    if (inputs.empty()) {
      VLOG(3) << "no input! return";
      return;
    }
    const SelectedRows* has_value_input = nullptr;
    for (auto* in : inputs) {
      if (!in->rows().empty()) {
        has_value_input = in;
        break;
      }
    }
    if (has_value_input == nullptr) {
      VLOG(3) << "no input has value! just return" << std::endl;
      return;
    }
    auto input_width = has_value_input->value().dims()[1];
    auto input_height = has_value_input->height();
    SelectedRows& out = *output;
    std::set<int64_t> merged_row_set;
    size_t row_num = 0;
    for (auto* input : inputs) {
      if (input->rows().empty()) {
        continue;
      }
      PADDLE_ENFORCE_EQ(input_width,
                        input->value().dims()[1],
                        common::errors::InvalidArgument(
                            "All inputs should have same "
                            "dimension except for the first one."));
      PADDLE_ENFORCE_EQ(input_height,
                        input->height(),
                        common::errors::InvalidArgument(
                            "All inputs should have same height."));
      row_num += input->rows().size();
      merged_row_set.insert(input->rows().begin(), input->rows().end());
    }

    out.set_height(input_height);
    DenseTensor* out_tensor = out.mutable_value();
    out_tensor->Resize(
        make_ddim({static_cast<int64_t>(merged_row_set.size()), input_width}));
    auto* out_data = dev_ctx.template Alloc<T>(out_tensor);

    if (merged_row_set.size() == row_num && !sorted_result) {
      // no duplicated ids, just concat the result together
      std::vector<int64_t> merge_rows;
      merge_rows.reserve(row_num);
      // concat rows
      for (auto* in : inputs) {
        merge_rows.insert(
            merge_rows.end(), in->rows().begin(), in->rows().end());
      }
      out.set_rows(merge_rows);
      auto in_place = inputs[0]->place();
      auto out_place = out.place();
      int64_t copied_numel = 0;
      for (auto* in : inputs) {
        auto* in_data = in->value().data<T>();
        auto in_numel = in->rows().size() * input_width;
        memory_utils::Copy(out_place,
                           out_data + copied_numel,
                           in_place,
                           in_data,
                           in_numel * sizeof(T));
        copied_numel += static_cast<int64_t>(in_numel);
      }
    } else {
      std::vector<int64_t> merge_rows(merged_row_set.begin(),
                                      merged_row_set.end());

      if (sorted_result) {
        std::sort(merge_rows.begin(), merge_rows.end());
      }

      out.set_rows(merge_rows);

      phi::funcs::SetConstant<DeviceContext, T> constant_functor;
      constant_functor(dev_ctx, out.mutable_value(), static_cast<T>(0.f));

      std::unordered_map<int64_t, size_t> rows_to_id;
      for (size_t i = 0; i < merge_rows.size(); ++i) {
        rows_to_id[merge_rows[i]] = i;
      }

      add_sparse_inputs<T, DeviceContext>(
          inputs, rows_to_id, input_width, dev_ctx, out_data);
    }
  }
};

template <typename T>
struct MergeAdd<CPUContext, T> {
  // unary functor, merge by adding duplicated rows in
  // the input SelectedRows object.
  SelectedRows operator()(const CPUContext& dev_ctx,
                          const SelectedRows& input,
                          const bool sorted_result) {
    return MergeAddImpl<CPUContext, T>()(dev_ctx, input, sorted_result);
  }

  void operator()(const CPUContext& dev_ctx,
                  const SelectedRows& input,
                  SelectedRows* output,
                  const bool sorted_result) {
    MergeAddImpl<CPUContext, T>()(dev_ctx, input, output, sorted_result);
  }

  void operator()(const CPUContext& dev_ctx,
                  const std::vector<const SelectedRows*>& inputs,
                  SelectedRows* output,
                  const bool sorted_result) {
    MergeAddImpl<CPUContext, T>()(dev_ctx, inputs, output, sorted_result);
  }
};

#define TEMPLATE_SPECIALIZED_FOR_MERGEADD_CPU(dtype) \
  template struct MergeAddImpl<CPUContext, dtype>;   \
  template struct PADDLE_API MergeAdd<CPUContext, dtype>;

TEMPLATE_SPECIALIZED_FOR_MERGEADD_CPU(float)
TEMPLATE_SPECIALIZED_FOR_MERGEADD_CPU(double)
TEMPLATE_SPECIALIZED_FOR_MERGEADD_CPU(int)
TEMPLATE_SPECIALIZED_FOR_MERGEADD_CPU(int64_t)
TEMPLATE_SPECIALIZED_FOR_MERGEADD_CPU(phi::bfloat16)
TEMPLATE_SPECIALIZED_FOR_MERGEADD_CPU(phi::complex64)
TEMPLATE_SPECIALIZED_FOR_MERGEADD_CPU(phi::complex128)

#ifdef PADDLE_WITH_XPU
template <typename T>
struct MergeAdd<XPUContext, T> {
  SelectedRows operator()(const XPUContext& dev_ctx,
                          const SelectedRows& input,
                          const bool sorted_result = false) {
    SelectedRows out;
    (*this)(dev_ctx, input, &out, sorted_result);
    return out;
  }

  void operator()(const XPUContext& dev_ctx,
                  const SelectedRows& input,
                  SelectedRows* output,
                  const bool sorted_result = false) {
    Vector<int64_t> input_rows(input.rows());
    if (input_rows.size() == 0) {
      return;
    }

    SelectedRows& out = *output;
    std::set<int64_t> row_set(input_rows.begin(), input_rows.end());
    std::vector<int64_t> merge_rows(row_set.begin(), row_set.end());
    auto input_width = input.value().dims()[1];

    out.set_rows(merge_rows);
    out.set_height(input.height());
    DenseTensor* out_tensor = out.mutable_value();
    out_tensor->Resize(
        make_ddim({static_cast<int64_t>(merge_rows.size()), input_width}));
    dev_ctx.template Alloc<T>(out_tensor);

    std::unordered_map<int64_t, size_t> rows_to_id;
    for (size_t i = 0; i < merge_rows.size(); ++i) {
      rows_to_id[merge_rows[i]] = i;
    }

    auto* y_data = out.mutable_value()->data<T>();
    auto* x_data = input.value().data<T>();
    int xm = input_rows.size();
    int ym = merge_rows.size();
    int n = input_width;

    xpu::ctx_guard RAII_GUARD(dev_ctx.x_context());
    int64_t* x_rows_data = RAII_GUARD.alloc_l3_or_gm<int64_t>(xm);
    int64_t* y_rows_data = RAII_GUARD.alloc_l3_or_gm<int64_t>(ym);
    memory_utils::Copy(dev_ctx.GetPlace(),
                       y_rows_data,
                       CPUPlace(),
                       merge_rows.data(),
                       ym * sizeof(int64_t));
    memory_utils::Copy(dev_ctx.GetPlace(),
                       x_rows_data,
                       CPUPlace(),
                       input_rows.data(),
                       xm * sizeof(int64_t));
    int r = xpu::merge_dup_rows<T, int64_t>(dev_ctx.x_context(),
                                            x_data,
                                            y_data,
                                            x_rows_data,
                                            y_rows_data,
                                            xm,
                                            n,
                                            ym);
    PADDLE_ENFORCE_XDNN_SUCCESS(r, "merge_dup_rows");
  }

  void operator()(const XPUContext& dev_ctx,
                  const std::vector<const SelectedRows*>& inputs,
                  SelectedRows* output,
                  const bool sorted_result = false) {
    if (inputs.size() == 0) {
      VLOG(3) << "no input! return";
      return;
    }
    const SelectedRows* has_value_input = nullptr;
    for (auto* in : inputs) {
      if (in->rows().size() > 0) {
        has_value_input = in;
        break;
      }
    }
    if (has_value_input == nullptr) {
      VLOG(3) << "no input has value! just return" << std::endl;
      return;
    }
    auto input_width = has_value_input->value().dims()[1];
    auto input_height = has_value_input->height();
    SelectedRows& out = *output;
    std::set<int64_t> merged_row_set;
    size_t row_num = 0;
    for (auto* input : inputs) {
      if (input->rows().size() == 0) {
        continue;
      }
      PADDLE_ENFORCE_EQ(input_width,
                        input->value().dims()[1],
                        common::errors::InvalidArgument(
                            "All inputs should have same "
                            "dimension except for the first one."));
      PADDLE_ENFORCE_EQ(input_height,
                        input->height(),
                        common::errors::InvalidArgument(
                            "All inputs should have same height."));
      row_num += input->rows().size();
      merged_row_set.insert(input->rows().begin(), input->rows().end());
    }

    std::vector<int64_t> merge_rows(merged_row_set.begin(),
                                    merged_row_set.end());

    if (sorted_result) {
      std::sort(merge_rows.begin(), merge_rows.end());
    }

    out.set_rows(merge_rows);
    out.set_height(input_height);

    DenseTensor* out_tensor = out.mutable_value();
    out_tensor->Resize(
        make_ddim({static_cast<int64_t>(merged_row_set.size()), input_width}));
    dev_ctx.template Alloc<T>(out_tensor);

    float* y_data = reinterpret_cast<float*>(out_tensor->data<T>());

    std::unordered_map<int64_t, size_t> rows_to_id;
    for (size_t i = 0; i < merge_rows.size(); ++i) {
      rows_to_id[merge_rows[i]] = i;
    }

    for (auto* input : inputs) {
      if (input->rows().size() == 0) {
        continue;
      }
      auto& input_rows = input->rows();

      auto* x_data = input->value().data<T>();
      int xm = input_rows.size();
      int ym = merge_rows.size();
      int n = input_width;

      xpu::ctx_guard RAII_GUARD(dev_ctx.x_context());
      int64_t* x_rows_data = RAII_GUARD.alloc_l3_or_gm<int64_t>(xm);
      int64_t* y_rows_data = RAII_GUARD.alloc_l3_or_gm<int64_t>(ym);
      memory_utils::Copy(dev_ctx.GetPlace(),
                         y_rows_data,
                         CPUPlace(),
                         merge_rows.data(),
                         ym * sizeof(int64_t));
      memory_utils::Copy(dev_ctx.GetPlace(),
                         x_rows_data,
                         CPUPlace(),
                         input_rows.data(),
                         xm * sizeof(int64_t));
      int r = xpu::merge_dup_rows<T, int64_t>(dev_ctx.x_context(),
                                              x_data,
                                              y_data,
                                              x_rows_data,
                                              y_rows_data,
                                              xm,
                                              n,
                                              ym);
      PADDLE_ENFORCE_XDNN_SUCCESS(r, "merge_dup_rows");
    }
  }
};

#endif
template <typename T>
struct MergeAverage<CPUContext, T> {
  SelectedRows operator()(const CPUContext& dev_ctx,
                          const SelectedRows& input) {
    SelectedRows out;
    (*this)(dev_ctx, input, &out);
    return out;
  }

  void operator()(const CPUContext& dev_ctx,
                  const SelectedRows& input,
                  SelectedRows* output) {
    std::vector<const SelectedRows*> inputs;
    inputs.push_back(&input);
    (*this)(dev_ctx, inputs, output);
  }

  void operator()(const CPUContext& dev_ctx,
                  const std::vector<const SelectedRows*>& inputs,
                  SelectedRows* output) {
    if (inputs.empty()) {
      VLOG(3) << "no input! return";
      return;
    }
    const SelectedRows* has_value_input = nullptr;
    for (auto* in : inputs) {
      if (!in->rows().empty()) {
        has_value_input = in;
        break;
      }
    }
    if (has_value_input == nullptr) {
      VLOG(3) << "no input has value! just return" << std::endl;
      return;
    }
    auto input_width = has_value_input->value().dims()[1];
    auto input_height = has_value_input->height();
    SelectedRows& out = *output;
    std::set<int64_t> merged_row_set;
    for (auto* input : inputs) {
      if (input->rows().empty()) {
        continue;
      }
      PADDLE_ENFORCE_EQ(input_width,
                        input->value().dims()[1],
                        common::errors::InvalidArgument(
                            "All inputs should have same "
                            "dimension except for the first one."));
      PADDLE_ENFORCE_EQ(input_height,
                        input->height(),
                        common::errors::InvalidArgument(
                            "All input should have same height."));
      merged_row_set.insert(input->rows().begin(), input->rows().end());
    }

    out.set_height(input_height);

    DenseTensor* out_tensor = out.mutable_value();
    out_tensor->Resize(
        make_ddim({static_cast<int64_t>(merged_row_set.size()), input_width}));
    auto* out_data = dev_ctx.template Alloc<T>(out_tensor);

    std::vector<int64_t> merge_rows(merged_row_set.begin(),
                                    merged_row_set.end());
    std::sort(merge_rows.begin(), merge_rows.end());

    out.set_rows(merge_rows);

    phi::funcs::SetConstant<CPUContext, T> constant_functor;
    constant_functor(dev_ctx, out.mutable_value(), static_cast<T>(0.0));

    std::unordered_map<int64_t, size_t> rows_to_id;
    for (size_t i = 0; i < merge_rows.size(); ++i) {
      rows_to_id[merge_rows[i]] = i;
    }

    auto blas = phi::funcs::GetBlas<CPUContext, T>(dev_ctx);
    for (auto* input : inputs) {
      if (input->rows().empty()) {
        continue;
      }
      auto* input_data = input->value().data<T>();
      auto& input_rows = input->rows();

      for (size_t i = 0; i < input_rows.size(); i++) {
        size_t out_i = rows_to_id[input_rows[i]];
        elementwise_add_to<T>(&blas,
                              static_cast<size_t>(input_width),
                              &input_data[i * input_width],
                              &out_data[out_i * input_width]);
      }
    }
    size_t input_width_cast = static_cast<size_t>(input_width);
    T count = static_cast<T>(inputs.size());
    for (size_t i = 0; i < merge_rows.size(); i++) {
      for (size_t j = 0; j < input_width_cast; j++) {
        out_data[i * input_width + j] = out_data[i * input_width + j] / count;
      }
    }
  }
};

#ifdef PADDLE_WITH_XPU
template struct MergeAdd<XPUContext, float>;
#endif

template struct PADDLE_API MergeAverage<CPUContext, int>;
template struct PADDLE_API MergeAverage<CPUContext, int64_t>;
template struct PADDLE_API MergeAverage<CPUContext, float>;
template struct PADDLE_API MergeAverage<CPUContext, double>;

template <typename T>
struct UpdateToTensor<CPUContext, T> {
  void operator()(const CPUContext& dev_ctx,
                  const ScatterOps& op,
                  const SelectedRows& input1,
                  DenseTensor* input2) {
    auto in1_height = input1.height();
    const auto& in2_dims = input2->dims();
    PADDLE_ENFORCE_EQ(
        in1_height,
        in2_dims[0],
        common::errors::InvalidArgument("The two inputs height must be equal."
                                        "But received first input height = "
                                        "[%d], second input height = [%d]",
                                        in1_height,
                                        in2_dims[0]));

    auto& in1_value = input1.value();
    auto& in1_rows = input1.rows();

    int64_t in1_row_numel =
        static_cast<int64_t>(in1_value.numel() / in1_rows.size());
    PADDLE_ENFORCE_EQ(in1_row_numel,
                      input2->numel() / in1_height,
                      common::errors::InvalidArgument(
                          "The two inputs width must be equal."
                          "But received first input width = [%d], "
                          "second input width = [%d]",
                          in1_row_numel,
                          input2->numel() / in1_height));

    auto* in1_data = in1_value.data<T>();
    auto* input2_data = input2->data<T>();

    // FIXME(typhoonzero): use macro fix the below messy code.
    switch (op) {
      case ScatterOps::ASSIGN:
        INLINE_FOR2(in1_rows.size(), in1_row_numel)
        input2_data[in1_rows[i] * in1_row_numel + j] =
            in1_data[i * in1_row_numel + j];
        break;
      case ScatterOps::ADD:
        INLINE_FOR2(in1_rows.size(), in1_row_numel)
        input2_data[in1_rows[i] * in1_row_numel + j] +=
            in1_data[i * in1_row_numel + j];
        break;
      case ScatterOps::SUB:
        INLINE_FOR2(in1_rows.size(), in1_row_numel)
        input2_data[in1_rows[i] * in1_row_numel + j] -=
            in1_data[i * in1_row_numel + j];
        break;
      case ScatterOps::SUBBY:
        INLINE_FOR2(in1_rows.size(), in1_row_numel)
        input2_data[in1_rows[i] * in1_row_numel + j] =
            in1_data[i * in1_row_numel + j] -
            input2_data[in1_rows[i] * in1_row_numel + j];
        break;
      case ScatterOps::MUL:
        INLINE_FOR2(in1_rows.size(), in1_row_numel)
        input2_data[in1_rows[i] * in1_row_numel + j] *=
            in1_data[i * in1_row_numel + j];
        break;
      case ScatterOps::DIV:
        INLINE_FOR2(in1_rows.size(), in1_row_numel)
        input2_data[in1_rows[i] * in1_row_numel + j] /=
            in1_data[i * in1_row_numel + j];
        break;
      case ScatterOps::DIVBY:
        INLINE_FOR2(in1_rows.size(), in1_row_numel)
        input2_data[in1_rows[i] * in1_row_numel + j] =
            in1_data[i * in1_row_numel + j] /
            input2_data[in1_rows[i] * in1_row_numel + j];
        break;
    }
  }
};

}  // namespace phi::funcs::scatter