// Copyright (c) 2020 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 <unordered_map>
#include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
#include "paddle/fluid/operators/elementwise/elementwise_op_function.h"

#include "paddle/fluid/framework/data_layout_transform.h"
#include "paddle/fluid/platform/mkldnn_reuse.h"

namespace paddle {
namespace operators {

using framework::DataLayout;
using framework::Tensor;
using mkldnn::memory;
using mkldnn::primitive;
using mkldnn::stream;

template <typename T, dnnl::algorithm BINARY_OP>
class EltwiseMKLDNNKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
    const auto& dev_ctx =
        ctx.template device_context<paddle::platform::MKLDNNDeviceContext>();
    const auto& mkldnn_engine = dev_ctx.GetEngine();

    const auto* x = ctx.Input<Tensor>("X");
    const auto* y = ctx.Input<Tensor>("Y");
    auto* z = ctx.Output<Tensor>("Out");

    float scale_x = ctx.Attr<float>("Scale_x");
    float scale_y = ctx.Attr<float>("Scale_y");
    float scale_o = ctx.Attr<float>("Scale_out");
    int axis = ctx.Attr<int>("axis");

    bool is_inplaced = x->IsSharedBufferWith(*z);

    std::string key = is_inplaced
                          ? platform::CreateKey(dev_ctx, ctx.OutputName("Out"),
                                                x->format(), y->format())
                          : ctx.OutputName("Out");

    platform::BinaryMKLDNNHandler<T> handler(
        BINARY_OP, axis, dev_ctx, mkldnn_engine, ctx.GetPlace(), x, y, z,
        scale_x, scale_y, scale_o, key);

    const auto src_x_memory = handler.AcquireSrcMemory(x);
    const auto src_y_memory = handler.AcquireSecondSrcMemory(y);

    // For Inplace src and and dst are the same memory object
    const auto dst_memory =
        is_inplaced ? src_x_memory : handler.AcquireDstMemory(z);

    const auto binary_prim = handler.AcquireForwardPrimitive();

    auto& astream = platform::MKLDNNDeviceContext::tls().get_stream();

    const std::unordered_map<int, dnnl::memory> args = {
        {DNNL_ARG_SRC_0, *src_x_memory},
        {DNNL_ARG_SRC_1, *src_y_memory},
        {DNNL_ARG_DST, *dst_memory}};

    binary_prim->execute(astream, args);
    astream.wait();

    z->set_layout(DataLayout::kMKLDNN);
    z->set_format(platform::GetMKLDNNFormat(*dst_memory));
  }
};
}  // namespace operators
}  // namespace paddle