softmax_op.h

/* 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 <vector>
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/operators/math/softmax.h"
#include "paddle/fluid/operators/transpose_op.h"

namespace paddle {
namespace operators {

using Tensor = framework::Tensor;

template <typename DeviceContext, typename T>
static inline void TransposeAxisToEnd(const Tensor& x, const Tensor& out,
                                      Tensor* x_trans, Tensor* out_trans,
                                      const int axis, std::vector<int> perm,
                                      const framework::ExecutionContext& ctx) {
  auto dim_x = x.dims();
  int rank = dim_x.size();

  if (axis == -1 || axis == rank - 1) {
    *x_trans = x;
    *out_trans = out;
    return;
  }

  auto& dev_ctx = ctx.template device_context<DeviceContext>();
  std::vector<int> shape;
  for (int i = 0; i < rank - 1; i++) {
    if (i == axis) {
      perm.push_back(rank - 1);
      shape.push_back(dim_x[rank - 1]);
    } else {
      perm.push_back(i);
      shape.push_back(dim_x[i]);
    }
  }
  perm.push_back(axis);
  shape.push_back(dim_x[axis]);

  x_trans->mutable_data<T>(framework::make_ddim(shape), ctx.GetPlace());
  out_trans->mutable_data<T>(framework::make_ddim(shape), ctx.GetPlace());
  TransCompute<DeviceContext, T>(rank, dev_ctx, x, x_trans, perm);
  TransCompute<DeviceContext, T>(rank, dev_ctx, out, out_trans, perm);
}

template <typename DeviceContext, typename T>
class SoftmaxKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& context) const override {
    auto* X = context.Input<Tensor>("X");
    auto* Out = context.Output<Tensor>("Out");
    const int axis = context.Attr<int>("axis");

    // allocate memory on device.
    Out->mutable_data<T>(context.GetPlace());

    Tensor X_trans, Out_trans;
    std::vector<int> perm;
    TransposeAxisToEnd<DeviceContext, T>(*X, *Out, &X_trans, &Out_trans, axis,
                                         perm, context);

    int rank = X->dims().size();
    Tensor X_2d = framework::ReshapeToMatrix(X_trans, rank - 1);
    Tensor Out_2d = framework::ReshapeToMatrix(Out_trans, rank - 1);

#ifdef PADDLE_ON_INFERENCE
    math::SoftmaxFunctor<DeviceContext, T, true>()(
        context.template device_context<DeviceContext>(), &X_2d, &Out_2d);
#else
    math::SoftmaxFunctor<DeviceContext, T, false>()(
        context.template device_context<DeviceContext>(), &X_2d, &Out_2d);
#endif

    if (axis != -1 && axis != rank - 1) {
      auto& dev_ctx = context.template device_context<DeviceContext>();
      TransCompute<DeviceContext, T>(rank, dev_ctx, Out_trans, Out, perm);
    }
  }
};

template <typename DeviceContext, typename T>
class SoftmaxGradKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& context) const override {
    auto* Out = context.Input<Tensor>("Out");
    auto* dOut = context.Input<Tensor>(framework::GradVarName("Out"));
    auto* dX = context.Output<Tensor>(framework::GradVarName("X"));

    // allocate memory on device.
    dX->mutable_data<T>(context.GetPlace());

    int rank = Out->dims().size();
    Tensor Out_2d = framework::ReshapeToMatrix(*Out, rank - 1);
    Tensor dOut_2d = framework::ReshapeToMatrix(*dOut, rank - 1);
    Tensor dX_2d = framework::ReshapeToMatrix(*dX, rank - 1);

    math::SoftmaxGradFunctor<DeviceContext, T>()(
        context.template device_context<DeviceContext>(), &Out_2d, &dOut_2d,
        &dX_2d);
  }
};

}  // namespace operators
}  // namespace paddle