[ BACKEND ] Enable `tl.dot` with TF32 precision on tiles with N=8 and K=8#10234

Merged

ThomasRaoux merged 1 commit into

triton-lang:mainfrom

mwichro:wgmma.m64n8k8

May 13, 2026

Commits on May 7, 2026

Enable `tl.dot` with TF32 precision on tiles with N=8 and K=8 (e.g. `wgmma.mma_async.sync.aligned.m64n8k8.f32.tf32.tf32`) via the standard `tt.dot` → `AccelerateMatmul` path on sm90+.
mwichro
committed