esa-tu-darmstadt / daphne-benchmark Goto Github PK

There is an optimization that could be implemented for OpenCl/points2image leading to reduced data transfer. However this would add atomics support as a requirement. I am not sure whether these changes should be integrated into the existing OpenCL portion, applied to a folder copy or only be available in a separate branch.

We are missing initialization of OCL_objs here. Platform and device queries are left out as well as command queue creation. This is probably related to the declaration:

that has no non-extern counterpart. As a result I get an undefined reference to `OCL_objs' in the compilation process.

For the minimal dataset used in points2image,

The sizes for the first two cases from the function parsePointCloud() are
1 55631 128
32684 1093664768 32

For the second one, the data size is too big and causes bad allocation. Did you have the memory allocation issue ?

Thanks

void  parsePointCloud(std::ifstream& input_file, PointCloud2* pointcloud2) {
  try {
    input_file.read((char*)&(pointcloud2->height), sizeof(int32_t));
    input_file.read((char*)&(pointcloud2->width), sizeof(int32_t));
    const size_t vol = pointcloud2->height * pointcloud2->width * pointcloud2->point_step;
    printf("%d %d %d\n", pointcloud2->height , pointcloud2->width , pointcloud2->point_step);
    input_file.read((char*)&(pointcloud2->point_step), sizeof(uint32_t));
    pointcloud2->data = new float[vol];
    input_file.read((char*)pointcloud2->data, vol);
  } catch (std::ifstream::failure) {
    throw std::ios_base::failure("Error reading the next point cloud.");
  }
}