[cjbgkagh]

I presumed they’d fold cutlass improvements into cuBASE so cuBASE would be at least as fast. Maybe some compile time optimizations?