cudnn_softmax.h

// 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 <cudnn.h>

#include <string>
#include <vector>

#include "lite/api/paddle_place.h"
#include "lite/backends/cuda/cuda_utils.h"
#include "lite/core/context.h"
#include "lite/core/target_wrapper.h"
#include "lite/operators/op_params.h"

namespace paddle {
namespace lite {
namespace cuda {
namespace math {

template <typename T, PrecisionType Ptype>
class CudnnSoftmax {
 public:
  CudnnSoftmax()
      : handle_(nullptr),
        bottom_desc_(nullptr),
        top_desc_(nullptr),
        handle_setup_(false) {}
  virtual ~CudnnSoftmax() {
    if (!handle_setup_) return;
    cudnnDestroyTensorDescriptor(bottom_desc_);
    cudnnDestroyTensorDescriptor(top_desc_);
    cudnnDestroy(handle_);
  }
  bool Init(const operators::SoftmaxParam& param, Context<TARGET(kCUDA)>* ctx);

  bool Create(const operators::SoftmaxParam& param,
              Context<TARGET(kCUDA)>* ctx);

  bool Run(const operators::SoftmaxParam& param);

 private:
  cudaStream_t stream_;
  cudnnHandle_t handle_;
  cudnnTensorDescriptor_t bottom_desc_;
  cudnnTensorDescriptor_t top_desc_;
  bool handle_setup_;
};

}  // namespace math
}  // namespace cuda
}  // namespace lite
}  // namespace paddle