The segmented top-k tests use cub::DeviceSegmentedRadixSort only as a verification reference. The compilation of the sort algorithm specializations is contributing a majority of the compilation time. We want to explore compiling the shared reference instantiations once and extern template-ing them in the consuming TUs. We want to explore this applicability beyond just top-k.
The segmented top-k tests use
cub::DeviceSegmentedRadixSortonly as a verification reference. The compilation of the sort algorithm specializations is contributing a majority of the compilation time. We want to explore compiling the shared reference instantiations once and extern template-ing them in the consuming TUs. We want to explore this applicability beyond just top-k.