composable_kernel

composable_kernel copied to clipboard

qianfengz

zjing14

Added client example for bwd qloop v1, v2, light v1 and light v2. Now we can do profiling for flash attention backward qloop.

guangzlu

[CMake][Doc] Why CK would require cmake 3.27.5at least 3.25

1

- [ ] The justification needs to be made and tracked here - [ ] There should be documentation task to update the readme, and installation guide. Reason: non-default and...

junliume

turneram

Tags for ROCm releases

8

Hi! I'm the maintainer of ROCm community packages for [Arch Linux](https://github.com/rocm-arch/rocm-arch). Would it be possible to add tags / releases that match ROCm and MIOpen releases? This would greatly simplify...

tpkessler

cloudhan

Fix GetWorkspaceSize() in some device operator

1

Just found that 1. `/data/composable_kernel/include/ck/tensor_operation/gpu/device/device_cgemm.hpp` defined `GetWorkspaceSize` 2. `/data/composable_kernel/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp` implemented `GetWorkspaceSize` The correct name should be `GetWorkSpaceSize`

qianfengz

In CK for (conv + bias + activation) we currently have activation as RELU. It would be great to have other activations. We can start with the ones that take...

bghimireamd

fsx950223