ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-02 13:17:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update example/ck_tile/42_unified_attention/README.md

Browse Source 
Co-authored-by: spolifroni-amd <Sandra.Polifroni@amd.com>
This commit is contained in:
Tianxing Wu
2026-01-02 14:23:15 +02:00
committed by GitHub
parent 6a62216c24
commit 788890ac83
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/ck_tile/42_unified_attention/README.md
View File

@@ -17,7 +17,7 @@ The kernel template is `unified_attention.hpp`, this is the grid-wise op in old
There are 2 template parameters for this kernel template.
* `FmhaPipeline` is one of the block_tile_pipeline(under `include/ck_tile/tile_program/block_tile_pipeline`) which is a performance critical component. Indeed, we did a lot of optimization and trials to optimize the pipeline and may still workout more performance pipeline and update into that folder. People only need to replace this pipeline type and would be able to enjoy the benefit of different performant implementations (stay tuned for updated pipeline(s)).
* `EpiloguePipeline` will modify and store out the result in the last phase. People usually will do lot of post-fusion at this stage, so we also abstract this concept. Currently we didn't do much thing at the epilogue stage but leave the room for future possible support.
* `EpiloguePipeline` is the last stage of the pipeline. It modifies and stores the result. Post-fusion can be done at this stage though the example only returns the result.
## codegen
To speed up compile time, we instantiate the kernels into separate file. In this way we can benefit from parallel building from CMake/Make system. This is achieved by `generate.py` script. Besides, you can look into this script to learn how to instantiate a kernel instance step by step, which is described in `FMHA_FWD_KERNEL_BODY` variable.