NMF_GPU_MAX_RANK.Rd
The sweep kernels tier their workgroup width by rank and stop at 128, where the width is already down to one SIMD group on Apple Silicon.
NMF_GPU_MAX_RANK
