CUDA程序错误检测

所有CUDA的API函数都有一个类型为cudaError_t的返回值，代表了一种错误信息；只有返回cudaSuccess时，才是成功调用。

• cudaGetLastError()用来检测核函数的执行是否出错
• cudaGetErrorString()输出错误信息

#include <stdio.h>
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include<math.h>
#include <malloc.h> 
#include <opencv2/opencv.hpp>
#include <stdlib.h>



#define BLOCK_SIZE 1


//图像卷积 GPU
__global__ void sobel_gpu(unsigned char* in, unsigned char* out, const int Height, const int Width)
{
	int x = blockDim.x * blockIdx.x + threadIdx.x;
	int y = blockDim.y + blockIdx.y + threadIdx.y;
	int index = y * Width + x;

	int Gx = 0;
	int Gy = 0;

	unsigned char x0, x1, x2, x3, x4, x5, x6, x7, x8;

	if (x>0 && x<(Width-1) && y>0 && y<(Height-1))
	{
		x0 = in[(y - 1)*Width + (x - 1)];
		x1 = in[(y - 1)*Width + (x)];
		x2 = in[(y - 1)*Width + (x + 1)];
		x3 = in[(y)*Width + (x - 1)];

		x5 = in[(y)*Width + (x + 1)];
		x6 = in[(y + 1)*Width + (x - 1)];
		x7 = in[(y + 1)*Width + (x)];
		x8 = in[(y + 1)*Width + (x + 1)];

		Gx = (x0 + 2 * x3 + x6) - (x2 + 2 * x5 + x8);
		Gy = (x0 + 2 * x1 + x2) - (x6 + 2 * x7 + x8);

		out[index] = (abs(Gx) + abs(Gy)) / 2;
	}
}



int main()
{
	cv::Mat src;
	src = cv::imread("complete004.jpg");

	cv::Mat grayImg,gaussImg;
	cv::cvtColor(src, grayImg, cv::COLOR_BGR2GRAY);
	cv::GaussianBlur(grayImg, gaussImg, cv::Size(3,3), 0, 0, cv::BORDER_DEFAULT);

	int height = src.rows;
	int width = src.cols;
	//输出图像
	cv::Mat dst_gpu(height, width, CV_8UC1, cv::Scalar(0));
	//GPU存储空间
	int memsize = height * width * sizeof(unsigned char);
	//输入 输出
	unsigned char* in_gpu;
	unsigned char* out_gpu;

	cudaMalloc((void**)&in_gpu, memsize);
	cudaMalloc((void**)&out_gpu, memsize);
	cudaError_t error_code;

	dim3 threadsPreBlock(BLOCK_SIZE, BLOCK_SIZE);
	dim3 blocksPreGrid((width + threadsPreBlock.x - 1)/threadsPreBlock.x, (height + threadsPreBlock.y - 1)/threadsPreBlock.y);
	
	cudaMemcpy(in_gpu, gaussImg.data, memsize, cudaMemcpyHostToDevice);

	sobel_gpu <<<blocksPreGrid, threadsPreBlock>>> (in_gpu, out_gpu, height, width);
	error_code = cudaGetLastError();
	
	printf("Error: %s\n", cudaGetErrorString(error_code));
	printf("FILE: %s\n", __FILE__);
	printf("LINE: %d\n", __LINE__);
	printf("Error code: %d\n", error_code);

	cudaMemcpy(dst_gpu.data, out_gpu, memsize, cudaMemcpyDeviceToHost);

	cv::imwrite("dst_gpu_save.png", dst_gpu);

	//cv::namedWindow("src", cv::WINDOW_NORMAL);
	cv::imshow("src", src);
	cv::imshow("dst_gpu", dst_gpu);
	cv::waitKey();

	cudaFree(in_gpu);
	cudaFree(out_gpu);

	return 0;
}

樊哲勇大牛的检测CUDA运行时错误的宏函数：

#pragma once
#include<stdio.h>

#define CHECK(call)																\
do                                                                              \
{																				\
	const cudaError_t error_code = call;										\
	if (error_code != cudaSuccess)												\
	{																			\
		printf("CUDA Error:\n");												\
		printf("	File:	%s\n", __FILE__);									\
		printf("	Line:	%d\n",__LINE__);									\
		printf("	Error code: %d\n",error_code);								\
		printf("	Error text: %s\n", cudaGetErrorString(error_code));			\
		exit(1);																\
	}																			\
} while (0)																		

采用检测CUDA运行时错误的宏函数：

#include <stdio.h>
#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include<math.h>
#include <malloc.h> 
#include <opencv2/opencv.hpp>
#include <stdlib.h>

#include "error.cuh"

#define BLOCK_SIZE 1


//图像卷积 GPU
__global__ void sobel_gpu(unsigned char* in, unsigned char* out, const int Height, const int Width)
{
	int x = blockDim.x * blockIdx.x + threadIdx.x;
	int y = blockDim.y + blockIdx.y + threadIdx.y;
	int index = y * Width + x;

	int Gx = 0;
	int Gy = 0;

	unsigned char x0, x1, x2, x3, x4, x5, x6, x7, x8;

	if (x>0 && x<(Width-1) && y>0 && y<(Height-1))
	{
		x0 = in[(y - 1)*Width + (x - 1)];
		x1 = in[(y - 1)*Width + (x)];
		x2 = in[(y - 1)*Width + (x + 1)];
		x3 = in[(y)*Width + (x - 1)];

		x5 = in[(y)*Width + (x + 1)];
		x6 = in[(y + 1)*Width + (x - 1)];
		x7 = in[(y + 1)*Width + (x)];
		x8 = in[(y + 1)*Width + (x + 1)];

		Gx = (x0 + 2 * x3 + x6) - (x2 + 2 * x5 + x8);
		Gy = (x0 + 2 * x1 + x2) - (x6 + 2 * x7 + x8);

		out[index] = (abs(Gx) + abs(Gy)) / 2;
	}
}



int main()
{
	cv::Mat src;
	src = cv::imread("complete004.jpg");

	cv::Mat grayImg,gaussImg;
	cv::cvtColor(src, grayImg, cv::COLOR_BGR2GRAY);
	cv::GaussianBlur(grayImg, gaussImg, cv::Size(3,3), 0, 0, cv::BORDER_DEFAULT);

	int height = src.rows;
	int width = src.cols;
	//输出图像
	cv::Mat dst_gpu(height, width, CV_8UC1, cv::Scalar(0));
	//GPU存储空间
	int memsize = height * width * sizeof(unsigned char);
	//输入 输出
	unsigned char* in_gpu;
	unsigned char* out_gpu;

	cudaMalloc((void**)&in_gpu, memsize);
	cudaMalloc((void**)&out_gpu, memsize);

	dim3 threadsPreBlock(BLOCK_SIZE, BLOCK_SIZE);
	dim3 blocksPreGrid((width + threadsPreBlock.x - 1)/threadsPreBlock.x, (height + threadsPreBlock.y - 1)/threadsPreBlock.y);
	
	cudaMemcpy(in_gpu, gaussImg.data, memsize, cudaMemcpyHostToDevice);

	sobel_gpu <<<blocksPreGrid, threadsPreBlock>>> (in_gpu, out_gpu, height, width);

	CHECK(cudaMemcpy(dst_gpu.data, out_gpu, memsize*10, cudaMemcpyDeviceToHost));//增大size值  引起报错

	cv::imwrite("dst_gpu_save.png", dst_gpu);

	//cv::namedWindow("src", cv::WINDOW_NORMAL);
	cv::imshow("src", src);
	cv::imshow("dst_gpu", dst_gpu);
	cv::waitKey();

	cudaFree(in_gpu);
	cudaFree(out_gpu);

	return 0;
}