CudaPoolingDeliverKernel.hpp

/*
 * CudaPoolingRecvPost.hpp
 *
 *  Created on: Aug 2, 2016
 *      Author: pschultz
 */

#ifndef CUDAPOOLINGDELIVERKERNEL_HPP_
#define CUDAPOOLINGDELIVERKERNEL_HPP_

#include "arch/cuda/CudaKernel.hpp"
#include "include/PVLayerLoc.h"

namespace PVCuda {

class CudaPoolingDeliverKernel : public CudaKernel {
   struct Params {
      PVLayerLoc const *preLoc;
      PVLayerLoc const *postLoc;
      int diffX;
      int diffY;
      cudnnPoolingMode_t poolingMode;
      float multiplier;
      void *poolingDescriptor;
      void *dataStoreDescriptor;
      float *dataStore;
      void *gSynDescriptor;
      float *gSyn;
   };

  public:
   CudaPoolingDeliverKernel(CudaDevice *inDevice);
   virtual ~CudaPoolingDeliverKernel();
   void setArgs(
         PVLayerLoc const *preLoc,
         PVLayerLoc const *postLoc,
         int nxpPost,
         int nypPost,
         cudnnPoolingMode_t poolingMode,
         int multiplier,
         CudaBuffer *inputBuffer,
         CudaBuffer *outputBuffer,
         int channel);
   static int calcStride(int preRestricted, int postRestricted);

  protected:
   virtual int do_run() override;
   int calcBorderExcess(
         int preRestricted,
         int postRestricted,
         int border,
         int patchSizePostPerspective);
   int calcManyScale(int preRestricted, int postRestricted);

  protected:
   PVLayerLoc const *mPreLoc                    = nullptr;
   PVLayerLoc const *mPostLoc                   = nullptr;
   int mBorderExcessX                           = 0;
   int mBorderExcessY                           = 0;
   cudnnPoolingMode_t mPoolingMode              = CUDNN_POOLING_MAX;
   float mMultiplier                            = 1.0f;
   cudnnPoolingDescriptor_t mPoolingDescriptor  = nullptr;
   cudnnTensorDescriptor_t mDataStoreDescriptor = nullptr;
   float *mDataStore                            = nullptr;
   CudaBuffer *mCudnnDataStore                  = nullptr;

   cudnnTensorDescriptor_t mGSynDescriptor = nullptr;
   float *mGSyn                            = nullptr;
   CudaBuffer *mCudnnGSyn                  = nullptr;
};

} /* namespace PVCuda */

#endif /* CUDAPOOLINGDELIVERKERNEL_HPP_ */