index_sample_op_npu.cc

/* Copyright (c) 2021 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/operators/index_sample_op.h"

#include "paddle/fluid/operators/npu_op_runner.h"

namespace paddle {
namespace operators {
using Tensor = framework::Tensor;

template <typename T>
class IndexSampleNPUKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
    auto& dev_ctx =
        ctx.template device_context<paddle::platform::NPUDeviceContext>();
    auto* input = ctx.Input<framework::LoDTensor>("X");
    auto* index = ctx.Input<framework::LoDTensor>("Index");
    auto* out = ctx.Output<framework::LoDTensor>("Out");
    out->mutable_data<T>(ctx.GetPlace());

    Tensor transformed_index;
    const auto& index_type = index->type();
    bool index_type_match = index_type == framework::proto::VarType::INT32 ||
                            index_type == framework::proto::VarType::INT64;
    PADDLE_ENFORCE_EQ(index_type_match, true,
                      platform::errors::InvalidArgument(
                          "Input(Index) holds the wrong type, it holds %s, but "
                          "desires to be %s or %s",
                          paddle::framework::DataTypeToString(index_type),
                          paddle::framework::DataTypeToString(
                              framework::proto::VarType::INT32),
                          paddle::framework::DataTypeToString(
                              framework::proto::VarType::INT64)));
    if (index_type == framework::proto::VarType::INT32) {
      transformed_index.mutable_data<int64_t>(index->dims(),
                                              dev_ctx.GetPlace());
      const auto& cast_runner = NpuOpRunner(
          "Cast", {*index}, {transformed_index}, {{"dst_type", ACL_INT64}});
      cast_runner.Run(dev_ctx.stream());
    } else {
      transformed_index.ShareDataWith(*index);
    }

    const auto& runner = NpuOpRunner(
        "GatherElements", {*input, transformed_index}, {*out}, {{"dim", 1}});
    runner.Run(dev_ctx.stream());
  }
};

template <typename IndexT>
void IndexSampleGradScatter(const paddle::platform::NPUDeviceContext& dev_ctx,
                            const Tensor* index, const Tensor* out_grad,
                            Tensor* x_grad) {
  auto index_dims = index->dims();
  auto input_dims = x_grad->dims();
  auto batch_size = input_dims[0];
  auto index_length = index_dims[1];

  std::vector<IndexT> scatter_index_vec;
  std::vector<IndexT> index_vec;
  framework::TensorToVector(*index, dev_ctx, &index_vec);
  for (auto i = 0; i < batch_size; ++i) {
    for (auto j = 0; j < index_length; j++) {
      scatter_index_vec.push_back(i);
      scatter_index_vec.push_back(index_vec[i * index_length + j]);
    }
  }
  Tensor scatter_index;
  framework::TensorFromVector(scatter_index_vec, dev_ctx, &scatter_index);
  scatter_index.Resize({batch_size, index_length, 2});

  NpuOpRunner runner;
  runner.SetType("ScatterNd")
      .AddInput(scatter_index)
      .AddInput(*out_grad)
      .AddInput(framework::vectorize<IndexT>(x_grad->dims()))
      .AddOutput(*x_grad);
  runner.Run(dev_ctx.stream());
}

template <typename T>
class IndexSampleGradNPUKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
    auto& dev_ctx =
        ctx.template device_context<paddle::platform::NPUDeviceContext>();
    auto* index = ctx.Input<framework::LoDTensor>("Index");
    auto* out_grad =
        ctx.Input<framework::LoDTensor>(framework::GradVarName("Out"));
    auto* x_grad =
        ctx.Output<framework::LoDTensor>(framework::GradVarName("X"));
    x_grad->mutable_data<T>(ctx.GetPlace());

    const auto& index_type = index->type();
    if (index_type == framework::proto::VarType::INT32) {
      IndexSampleGradScatter<int32_t>(dev_ctx, index, out_grad, x_grad);
    } else {
      IndexSampleGradScatter<int64_t>(dev_ctx, index, out_grad, x_grad);
    }
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
namespace plat = paddle::platform;

REGISTER_OP_NPU_KERNEL(index_sample, ops::IndexSampleNPUKernel<plat::float16>,
                       ops::IndexSampleNPUKernel<float>,
                       ops::IndexSampleNPUKernel<int32_t>,
                       ops::IndexSampleNPUKernel<int64_t>);
REGISTER_OP_NPU_KERNEL(index_sample_grad,
                       ops::IndexSampleGradNPUKernel<plat::float16>,
                       ops::IndexSampleGradNPUKernel<float>,
                       ops::IndexSampleGradNPUKernel<int32_t>,
                       ops::IndexSampleGradNPUKernel<int64_t>);