Open SeyedHamidreza opened 1 month ago
Hi @SeyedHamidreza
The pretrained model is provided directly inside the GitHub repo here.
Thanks for sharing the CUDA fix! It's really helpful.
This project was created back in 2019, and there have been quite a few changes to the Torch and CUDA interface since then. It would be amazing if you could share your fix with the rest of the community by submitting a pull request. This would help ensure compatibility for everyone.
Thanks again for your contribution!
sure, I will do it as soon as I find time. hopefully at some point next week.
Hi, Thank you for sharing the code.
For the newer version of CUDA, we need to replace THCudaCheck with AT_CUDA_CHECK
//THCudaCheck(cudaGetLastError()); AT_CUDA_CHECK(cudaGetLastError());
and we need to comment //THArgCheck(at::cuda::getApplyGrid(totalElements, grid, curDevice), 1, "Too many elements to calculate");
in sampling_kernel.cu and ball_query_kernel.cu we need to do the following
//#include <THC/THC.h>
include <ATen/cuda/CUDAContext.h>
include <ATen/cuda/CUDAEvent.h>
for more info, please check https://github.com/pytorch/pytorch/issues/72807
best regards, Hamidreza