cuda stream

[whuzs]

“thrust::for_each(thrust::cuda::par.on(stream), points.begin(), points.end(), accumulate_points_kernel(voxelmap_info_ptr.data(), buckets, num_points, voxel_means, voxel_covs));

thrust::for_each(thrust::counting_iterator(0), thrust::counting_iterator(voxelmap_info.num_voxels), ndt_finalize_voxels_kernel(num_points, voxel_means, voxel_covs));”

Hello, I would like to ask if these two statements are executed simultaneously. They are in different streams, but they should be executed in order logically. Why not put them in the same stream? Thank you.

[koide3]

Ah, it should be a bug. Fortunately, recent versions of thrust perform for_each in a synchronized way (regardless of the use of streams), and this does not affect programs on a recent version of CUDA. Anyways, I'll push a fix soon.

[JACKLiuDay]

Ah, it should be a bug. Fortunately, recent versions of thrust perform for_each in a synchronized way (regardless of the use of streams), and this does not affect programs on a recent version of CUDA. Anyways, I'll push a fix soon.

Hi, thank you for your work about fast gicp, I am trying to use it on my Nvidia jetson NX board. Does this fast gicp project support for Nvidia jetson board?

[whuzs]

Nvidia jetson NX

i run it successfully in orin and Xavier

[JACKLiuDay]

Hi, thank you for your quick reply. You mean you run it successfully on Orin and AGX xavier? That is a good news to me. I just have a AGX xavier. And I will try it on the AGX xavier. Could you tell which jetpack you use? I just tried jetpack5.1 on NX and it did not work.

[whuzs]

Orin(JetPack5.0.2.3,other default) and MIIVII Apex Xavier(default), maybe you cak take a look at #107

[JACKLiuDay]

Orin(JetPack5.0.2.3,other default) and MIIVII Apex Xavier(default), maybe you cak take a look at #107

Hi, thank you for your advice. I just looked at #107, it looks like the fix had been merged to the project master code. But my NX board still can not run gicp_kitty with fastvgivpcuda. It shows identity matrix when run gicp_kitty with fastvgicpcuda.

[JACKLiuDay]

@whuzs @koide3 Hi, I just tried fast_gicp project on AGX Xavier with jtpack5.1.1. I changed the align method with fastvgicpcuda. When I run the fast_gicp with gicp_kitti code. The result shows correct at first. Then it shows nan nan nan all the time. Could you give me some advice. Something wrong caused by arm64?

[PigBroA]

@JACKLiuDay I think you should check out whether there are nan(of inf) value in target or source cloud. you can use pcl::removeNaNFromPointCloud to remove them.