Now that numba 0.57 has been released with support for CUDA atomic compare and swap functionality (numba/numba#8790), we need to use this to give us faster CUDA reductions such as max_n, building on top of #1196. It will use a check on the version of numba available to choose whether to use the new or old mutex functionality.
Now that numba 0.57 has been released with support for CUDA atomic compare and swap functionality (numba/numba#8790), we need to use this to give us faster CUDA reductions such as
max_n, building on top of #1196. It will use a check on the version of numba available to choose whether to use the new or old mutex functionality.