Haicheng Wu comments

Results 323 comments of


                                            Haicheng Wu

Fix potential out-of-bounds access in 3.x grouped gemm kernel

We will upstream 3.5.1 first

[BUG] Failing to build on MSVC due to call to _div128

@mhoemmen

Extend profiler search space for SM89 GEMM

After 3.5.1

[FEA] Add cuTensorMapEncodeTiled to CudaHostAdapter

@kerrmudgeon

Add Ampere GEMM example using Cute and CUTLASS 3.x

@thakkarV @ccecka

Fix illegal memory accesses in multistage `Mma's` for `k=0`

@jackkosaian

Fix illegal memory accesses in multistage `Mma's` for `k=0`

What is the name of the kernel used by cublas?

[QST] GEMM Epilogue Fusion: Element-wise Ops and Two-Tensor Element-wise Multiplication

You can write your own via epilogue visitor tree (evt). We have an example of using evt.

[QST] custom kernel integrated in Pytorch

Thanks for your interest. Actually, it is better asking this question in Pytorch github. They integrated many cutlass kernels (https://github.com/pytorch/pytorch/pulls?q=cutlass) already. cc += @jackkosaian

[QST]How to write column major output epilogue in cutlass2.x

cutlass already supports fp8 sparse gemm. as all other sparse gemm, it only has row major output. if you want to implement col major output, first thing is that you...