Optimizing CUDA parallel reduction.

src/core/cuda/tnlCudaReductionBuffer.h

+/***************************************************************************
+                          tnlCudaReductionBuffer.h  -  description
+                             -------------------
+    begin                : June 17, 2015
+    copyright            : (C) 2015 by Tomas Oberhuber
+    email                : tomas.oberhuber@fjfi.cvut.cz
+ ***************************************************************************/
+
+/***************************************************************************
+ *                                                                         *
+ *   This program is free software; you can redistribute it and/or modify  *
+ *   it under the terms of the GNU General Public License as published by  *
+ *   the Free Software Foundation; either version 2 of the License, or     *
+ *   (at your option) any later version.                                   *
+ *                                                                         *
+ ***************************************************************************/
+
+#ifndef TNLCUDAREDUCTIONBUFFER_H
+#define	TNLCUDAREDUCTIONBUFFER_H
+
+#include <core/tnlCuda.h>
+
+class tnlCudaReductionBuffer
+{
+   public:
+      inline tnlCudaReductionBuffer( size_t size = 0 ): data( 0 ), size( 0 )
+      {
+         if( size != 0 ) setSize( size );
+      }
+      
+      inline bool setSize( size_t size )
+      {
+#ifdef HAVE_CUDA         
+         if( size > this->size )
+         {
+            this->size = size;
+            if( cudaMalloc( ( void** ) &this->data, size ) != cudaSuccess )
+            {
+               cerr << "I am not able to allocate reduction buffer on the GPU." << endl;
+               this->data = 0;
+            }
+            return checkCudaDevice;
+         }
+         else
+            return true;
+#else
+         return false;
+#endif         
+      }
+      
+      template< typename Type >
+      Type* getData() { return ( Type* ) this->data; }
+      
+      inline ~tnlCudaReductionBuffer()
+      {
+#ifdef HAVE_CUDA         
+         if( data ) cudaFree( data );
+#endif         
+      }
+      
+   protected:
+      
+      void* data;
+      
+      size_t size;
+};
+
+
+#endif	/* TNLCUDAREDUCTIONBUFFER_H */
+