clip_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/clip_op.h"
#include "paddle/fluid/platform/device/npu/npu_op_runner.h"

namespace paddle {
namespace operators {

using Tensor = framework::Tensor;

template <typename DeviceContext, typename T>
class ClipNPUKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
    auto* x = ctx.Input<Tensor>("X");
    auto* out = ctx.Output<Tensor>("Out");
    out->mutable_data<T>(ctx.GetPlace());

    auto min_tensor = ctx.HasInput("Min") ? ctx.Input<Tensor>("Min") : nullptr;
    auto max_tensor = ctx.HasInput("Max") ? ctx.Input<Tensor>("Max") : nullptr;

    Tensor min_tensor_temp(x->type());
    Tensor max_tensor_temp(x->type());
    if (min_tensor == nullptr) {
      auto min_value = static_cast<T>(ctx.Attr<float>("min"));
      min_tensor_temp.mutable_data<T>({1}, ctx.GetPlace());
      FillNpuTensorWithConstant<T>(&min_tensor_temp, min_value);
      min_tensor = &min_tensor_temp;
    }

    if (max_tensor == nullptr) {
      auto max_value = static_cast<T>(ctx.Attr<float>("max"));
      max_tensor_temp.mutable_data<T>({1}, ctx.GetPlace());
      FillNpuTensorWithConstant<T>(&max_tensor_temp, max_value);
      max_tensor = &max_tensor_temp;
    }

    auto stream =
        ctx.template device_context<paddle::platform::NPUDeviceContext>()
            .stream();
    const auto& runner =
        NpuOpRunner("ClipByValue", {*x, *min_tensor, *max_tensor}, {*out}, {});
    runner.Run(stream);
  }
};

template <typename DeviceContext, typename T>
class ClipGradNPUKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
    auto* x = ctx.Input<Tensor>("X");
    auto* dout = ctx.Input<Tensor>(framework::GradVarName("Out"));
    auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
    dx->mutable_data<T>(ctx.GetPlace());

    auto* min_tensor = ctx.HasInput("Min") ? ctx.Input<Tensor>("Min") : nullptr;
    auto* max_tensor = ctx.HasInput("Max") ? ctx.Input<Tensor>("Max") : nullptr;

    auto min_val = ctx.Attr<float>("min");
    if (min_tensor) {
      Tensor min_data;
      framework::TensorCopy(
          *min_tensor, platform::CPUPlace(),
          ctx.template device_context<platform::DeviceContext>(), &min_data);
      ctx.template device_context<paddle::platform::NPUDeviceContext>().Wait();
      min_val = static_cast<float>(min_data.data<T>()[0]);
    }

    auto max_val = ctx.Attr<float>("max");
    if (max_tensor) {
      Tensor max_data;
      framework::TensorCopy(
          *max_tensor, platform::CPUPlace(),
          ctx.template device_context<platform::DeviceContext>(), &max_data);
      ctx.template device_context<paddle::platform::NPUDeviceContext>().Wait();
      max_val = static_cast<float>(max_data.data<T>()[0]);
    }

    auto stream =
        ctx.template device_context<paddle::platform::NPUDeviceContext>()
            .stream();
    const auto& runner =
        NpuOpRunner("HardtanhGrad", {*x, *dout}, {*dx},
                    {{"min_val", min_val}, {"max_val", max_val}});
    runner.Run(stream);
  }
};

}  // namespace operators
}  // namespace paddle

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

REGISTER_OP_NPU_KERNEL(
    clip, ops::ClipNPUKernel<plat::NPUDeviceContext, float>,
    ops::ClipNPUKernel<plat::NPUDeviceContext, plat::float16>);

REGISTER_OP_NPU_KERNEL(
    clip_grad, ops::ClipGradNPUKernel<plat::NPUDeviceContext, float>,
    ops::ClipGradNPUKernel<plat::NPUDeviceContext, plat::float16>);