Matthew Nicely comments

Results 113 comments of


                                            Matthew Nicely

[QST] How to use slicedK in GEMM?

@Arsmart123 were you able to resolve your issues?

[BUG] Fused GEMM example gives wrong result with some shapes

@jwang323 @danthe3rd are we able to close this issue?

add Conv singlestage

@leiwen83 can you provide performance results to justify this PR?

[BUG] batch GEMM execution via cutlass_profiler gives weird outputs

@TUMSchieben were you able to work past your issues?

[QST] how to undersrtand a cutlass kernel name?

Most of the information you're looking for is in the `functionality.md`. https://github.com/NVIDIA/cutlass/blob/master/media/docs/functionality.md

[QST] How slice K reduce the value?

@Arsmart123, I think you would benefit greatly from some of the CUTLASS sessions that have been presented at GTC. [CUTLASS: Software Primitives for Dense Linear Algebra at All Levels and...

[QST] set different stages has different accuracy

@yuxgis > I just did gemm(MxK KxN) TF32 relative error compare with FP64. when K increase, relative errors are decrease. It is unreasonable. Are you sure you're not run 3xTF32?...

[FEA] FP8 GEMM implementation

@jianyuh is your interest in dense, sparse, or both?

[DOC] Could you write document for how to use cutlass on windows?

Hi @Arsmart123, have you tried using CUTLASS through WSL2?

[DOC] Where does cutlass’ detailed GEMM kernel?

@Arsmart123, that blog is pretty outdated. To get started please look at our [examples](https://github.com/NVIDIA/cutlass/tree/master/examples), specifically [_basic_gemm_](https://github.com/NVIDIA/cutlass/blob/master/examples/00_basic_gemm/basic_gemm.cu)