cudaUtility.h

/*
 * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 * DEALINGS IN THE SOFTWARE.
 */

#ifndef __CUDA_UTILITY_H_
#define __CUDA_UTILITY_H_


#include <cuda_runtime.h>
#include <cuda.h>
#include <stdio.h>
#include <string.h>
#include <stdint.h>

#include "logging.h"


/**
 * Execute a CUDA call and print out any errors
 * @return the original cudaError_t result
 * @ingroup cudaError
 */
#define CUDA(x)				cudaCheckError((x), #x, __FILE__, __LINE__)

/**
 * Evaluates to true on success
 * @ingroup cudaError
 */
#define CUDA_SUCCESS(x)			(CUDA(x) == cudaSuccess)

/**
 * Evaluates to true on failure
 * @ingroup cudaError
 */
#define CUDA_FAILED(x)			(CUDA(x) != cudaSuccess)

/**
 * Return from the boolean function if CUDA call fails
 * @ingroup cudaError
 */
#define CUDA_VERIFY(x)			if(CUDA_FAILED(x))	return false;

/**
 * Return on CUDA errors, continue on cudaSuccess.
 * @ingroup cudaError
 */
#define CUDA_ASSERT(x)          { const cudaError_t _retval = CUDA(x); if(_retval != cudaSuccess) return _retval; }

/**
 * LOG_CUDA string.
 * @ingroup cudaError
 */
#define LOG_CUDA "[cuda]   "

/*
 * define this if you want all cuda calls to be printed
 * @ingroup cudaError
 */
//#define CUDA_TRACE



/**
 * cudaCheckError
 * @ingroup cudaError
 */
inline cudaError_t cudaCheckError(cudaError_t retval, const char* txt, const char* file, int line )
{
	if( retval == cudaSuccess )
	{
	#if !defined(CUDA_TRACE)
		return cudaSuccess;
	#else
		LogDebug(LOG_CUDA "%s\n", txt);
	#endif
	}
	else
	{
		LogError(LOG_CUDA "%s\n", txt);
		LogError(LOG_CUDA "   %s (error %u) (hex 0x%02X)\n", cudaGetErrorString(retval), retval, retval);
		LogError(LOG_CUDA "   %s:%i\n", file, line);	
	}

	return retval;
}


/**
 * Check for non-NULL pointer before freeing it, and then set the pointer to NULL.
 * @ingroup cudaError
 */
#define CUDA_FREE(x) 		if(x != NULL) { cudaFree(x); x = NULL; }

/**
 * Check for non-NULL pointer before freeing it, and then set the pointer to NULL.
 * @ingroup cudaError
 */
#define CUDA_FREE_HOST(x)	if(x != NULL) { cudaFreeHost(x); x = NULL; }

/**
 * Check for non-NULL pointer before deleting it, and then set the pointer to NULL.
 * @ingroup util
 */
#define SAFE_DELETE(x) 		if(x != NULL) { delete x; x = NULL; }

/**
 * Check for non-NULL pointer before freeing it, and then set the pointer to NULL.
 * @ingroup util
 */
#define SAFE_FREE(x) 		if(x != NULL) { free(x); x = NULL; }


/**
 * If a / b has a remainder, round up.  This function is commonly using when launching 
 * CUDA kernels, to compute a grid size inclusive of the entire dataset if it's dimensions
 * aren't evenly divisible by the block size.
 *
 * For example:
 *
 *    const dim3 blockDim(8,8);
 *    const dim3 gridDim(iDivUp(imgWidth,blockDim.x), iDivUp(imgHeight,blockDim.y));
 *
 * Then inside the CUDA kernel, there is typically a check that thread index is in-bounds.
 *
 * Without the use of iDivUp(), if the data dimensions weren't evenly divisible by the
 * block size, parts of the data wouldn't be covered by the grid and not processed.
 *
 * @ingroup cuda
 */
inline __device__ __host__ int iDivUp( int a, int b )  		{ return (a % b != 0) ? (a / b + 1) : (a / b); }


#endif