pystencils.codegen.config.GpuOptions.use_cub_reductions

GpuOptions.use_cub_reductions: BasicOption[bool] = None

Specifies if CUB is used as a reduction back-end (CUDA/HIP). Its back-end platforms depend on the block size used, i.e. default_block_size, requiring it to match the launch configuration’s block size.

If this option is not set, warp reductions or pure atomics are used.