Ilegal memory access in code (Jetson AGX Xavier)

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Xavier
1

Hi,

For some reason when I run the attached code in my jetson AGX xavier, I get the following error:
GPUassert: an illegal memory access was encountered /home/folder/test.cu 55

Can someone please help me figure out what is wrong with my memory access pattern? The size of my pointcloud is 14316 and it is initialized and filled before I enter this function.

Here is the code:

#include <iostream>
#include <math.h>

#define PCL_NO_PRECOMPILE
#include <pcl/point_cloud.h>
#include <pcl/point_types.h>


#define gpuErrchk(ans) { gpuAssert((ans), __FILE__, __LINE__); }
inline void gpuAssert(cudaError_t code, const char *file, int line, bool abort=true)
{
   if (code != cudaSuccess) 
   {
      fprintf(stderr,"GPUassert: %s %s %d\n", cudaGetErrorString(code), file, line);
      if (abort) exit(code);
   }
}


__global__ 
void pointCloud2Float3(const int nfloat3, const float* point, float3* point3)
{
    int idx = blockIdx.x * blockDim.x + threadIdx.x;
    if(idx < nfloat3)
    {
        point3[idx] = make_float3(point[4*idx], point[4*idx+1], point[4*idx+2]); 
    }
    __syncthreads();

}

void test(pcl::PointCloud<pcl::PointXYZ>::Ptr target_cloud)
{
    int max_iterations_ = 100;
    int nr_samples_ = 4;

    // Allocate Unified Memory – accessible from CPU or GPU
    float* points = NULL;
    gpuErrchk( cudaMallocManaged(&points, target_cloud->points.size()*sizeof(float)* 4) );
    points = (float *)target_cloud->points.data();

    float3* points3 = NULL;
    gpuErrchk( cudaMallocManaged(&points3, target_cloud->points.size()*sizeof(float3)) );


    // Define kernel parameters
    const int gridSize = 1024;
    const int blocks = ceil(static_cast<float>(target_cloud->points.size())/static_cast<float>(gridSize));


    const int nfloat3 = static_cast<int>(target_cloud->points.size());
    pointCloud2Float3<<<gridSize, blocks>>>(nfloat3, points, points3);
    gpuErrchk( cudaPeekAtLastError() );
    // Wait for GPU to finish before accessing on host
    gpuErrchk( cudaDeviceSynchronize() );

    cudaFree(points);
    cudaFree(points3);

}
5

Hi,

points = (float *)target_cloud->points.data();

You will need to do memcpy or use the buffer to load the PCL point cloud.

The above assignment will overwrite the unified memory pointer into a pre-allocated PCL buffer.
And it seems that the target_cloud->points.data is a CPU buffer so causes the illegal access error.

Thanks.

1 Like