clEnqueueNDRangeKernel上的OpenCl分段错误

时间:2016-02-16 10:05:38

标签: opencl convolution

我一直在使用Eclipse上的OpenCL开发Convolution。它在enqueueNDRangeKernel之后给出了分段错误。 这是我的主机代码: -

我使用OpenCV获取输入图像,然后: -

const int width = image.size().width;
const int height = image.size().height;
std::cout<<"width: \t"<<width<<"\t height: "<<height<<std::endl;
std::size_t in_imagesize = (width*height)*sizeof(float);

std::vector<float> ptr(width*height,0);

const float filter[3] = {1,2,3};
float filter_size = 3*sizeof(float);
const int FilterRadius = 1;

cv::Mat result_image = cv::Mat(cvSize(width,height), CV_32FC1);
std::size_t out_imagesize = sizeof(float)*(width*height);
std::vector<float> read_buffer(width*height,0);

然后是上下文,命令队列,内核程序,然后: -

cl::Buffer input_dev, filter_kernel, output_dev;

input_dev = cl::Buffer(ctx,CL_MEM_READ_ONLY|CL_MEM_USE_HOST_PT R,in_imagesize,image.data,&err);
if(error!= CL_SUCCESS){
std::cout<<"Input Buffer Failed "<<std::endl;
}

output_dev =cl::Buffer(ctx,CL_MEM_READ_WRITE,out_imagesize,NU LL,&err);
if(error!= CL_SUCCESS){
std::cout<<"Output Buffer Failed "<<std::endl;
}

filter_kernel = cl::Buffer(ctx,CL_MEM_READ_ONLY,filter_size,NULL,& err);
if(error!= CL_SUCCESS){
std::cout<<"Output Buffer Failed "<<std::endl;


std::cout<<"filter_kernel write buffer "<<std::endl;
queue.enqueueWriteBuffer(filter_kernel,CL_TRUE,0,3 *sizeof(float),filter,NULL,NULL);


// Create Kernel

std::cout<<"Now try create kernel objects .."<<std::endl;

cl::Kernel kernel(prg,"ConvH_naive",&err);
if(error!= CL_SUCCESS)
{
std::cout<<"create Kernel_naive failed \n"<<std::endl;
}

然后是内核参数,之后: -

cl::NDRange globalsize(width,height);
cl::NDRange localsize(1,1);
cl::NDRange offset(0,0);


std::cout<<"Enqueuing the Kernel"<<std::endl;
if(queue.enqueueNDRangeKernel(kernel,offset,global size,localsize,NULL,NULL)!=CL_SUCCESS)
{
std::cout<<"Failed enqueuing the Kernel"<<std::endl;
}
queue.finish();

此Readbuffer和imshow之后。但是代码在此语句之后停止,给出了分段错误。

任何人都可以帮忙吗?是否有问题是内核代码?我也加上吗?

1 个答案:

答案 0 :(得分:0)

  1. (1,1)的本地大小通常是一个非常糟糕的选择
  2. 你在哪个平台上运行?什么设备(例如CPU,GPU)?
  3. 可能是你是segfaulting,因为你没有处理边界条件和访问缓冲区越界。