Function emit_wmma_tiled_bench 

pub fn emit_wmma_tiled_bench(
    kind: MmaKind,
    tile: &WmmaTile,
    m: usize,
    k: usize,
    n: usize,
) -> String

Timing bench for the PIPELINED (shared-memory, warp-tiled) WMMA kernel at kind + tile — prints GOPS.