Haicheng Wu

icon company @nvidia

Results 323 comments of Haicheng Wu

[FEA] CVT F32 -> TF32 PTX for sm80

okay, could you please file a PR to us?

[BUG] Logic issue in nondeterministic reduction mode of Stream-K tile scheduler.

@jackkosaian

[BUG] Logic issue in nondeterministic reduction mode of Stream-K tile scheduler.

this should be fixed now.

[QST] How to implement a fused mixed precision matrix multiplication such as w4a4 + w16a16?

Could you please elaborate the input and output of every step? Do you want to fuse two gemms into one kernel similar as what our ex.13 does?

Support N={48, 80, 96, 112, ...} for SM100 EpilogueTileAuto

have you tested them if they are working? we would need some unit tests or adding them to the profiler to verify them.

[QST]Using customized operators in gemm kernel

you could check this https://github.com/hpcgarage/cuASR cc @thakkarV

[QST]Behavior of TMA Store and Wait Mechanism in CUTLASS

@Junkai-Wu

[QST] Why does GenerateSM90_TensorOp_16b_WGMMA_alignx_gemm not generate C.element = DataType.void?

@Junkai-Wu

[QST] Unify The Color for Latex and SVG Print

@ccecka

Can we speed up the cutlass python generator?

we are open to suggestions and prs.