At present the micromagnetic part only supports CUDA 10.2.
The reason is that cusparseScsrmv is deprecated in CUDA 11.0. The solution is to change to cusparseSpMV but this requires modifying MagTenseCudaBlas.cu extensively. The sparse matrices must be recoded to comply with the new format. An example can be seen in https://docs.nvidia.com/cuda/cusparse/index.html#cusparse-generic-example-spmv .
It would be useful to implement the change so that the dBuffer that must be allocated for the computation is only allocated once, when the sparse matrices are also constructed.
At present the micromagnetic part only supports CUDA 10.2.
The reason is that cusparseScsrmv is deprecated in CUDA 11.0. The solution is to change to cusparseSpMV but this requires modifying MagTenseCudaBlas.cu extensively. The sparse matrices must be recoded to comply with the new format. An example can be seen in https://docs.nvidia.com/cuda/cusparse/index.html#cusparse-generic-example-spmv .
It would be useful to implement the change so that the dBuffer that must be allocated for the computation is only allocated once, when the sparse matrices are also constructed.