got "for_each: failed to synchronize" in every position using "thrust::for_each"

[Hansehun]

Excuse me, I got an exception about "for_each: failed to synchronize" when the program runs to "covariance_regularization" in "vgicp_cuda_->calculate_source_covariances_rbf".I think "thrust::for_each(first, last, svd_reconstruction_kernel(diag_matrix_ptr, covs));" throw this exception. Then I followed the instructions here "https://stackoverflow.com/questions/64441827/cuda-thrustsort-met-memory-problem-when-i-still-have-enough-memory" to expand CUDAHeapSize, but it doesn't seem to work.
My operating environment is Windows11 cuda11.6. I use CMake-GUI to build fast_gicp, then I compile it by Visual Studio. By the way, I also have tried cuda11.7 and cuda11.1, but the same exception still occurs.
Here is the Visual Studio call stack information:

Do you have any idea? Thank you very much.

[koide3]

Can you see if the error occurs when running gicp_align with the example data?

[Hansehun]

Thank you for your reply.I tried gicp_align with the example data.An exception occurs in the same position like this

and the shell shows like this