pub fn tc_route(
m: u32,
n: u32,
k: u32,
enabled: bool,
have8: bool,
have16: bool,
) -> Option<TcKind>Expand description
2026-09-25: The routing decision, without I/O. None keeps the caller’s CUDA-core kernel.
The kernel walks K in 128-element blocks, so a K that is not a multiple of 128 declines.
Any N routes: in a partial last tile the kernel loads zeros for columns past N and never
stores them.
tc8 serves m <= 8; tc16 serves m <= 16, including m <= 8 when tc8 is missing.