Merge CK validation to release/2.6 #2016
Open
+65
−79
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…eate link target. Enable USE_CK_FLASH_ATTENTION based on USE_FLASH_ATTENTION option.
… using lintrunner.
akashveramd
requested review from
pragupta,
pruthvistony,
jithunnair-amd and
alugorey
pruthvistony
force-pushed
the
release/2.6
branch
from
b47638e to
20ad852
Compare
|
Jenkins build for 6943678cdc3d79c37906161e88df783828f8d403 commit finished as FAILURE Detected error during Pytorch building: |
|
Jenkins build for 6943678cdc3d79c37906161e88df783828f8d403 commit finished as FAILURE Detected error during Pytorch building: |
xinyazhang
reviewed
Apr 3, 2025
| include(${CMAKE_SOURCE_DIR}/cmake/public/LoadHIP.cmake) | ||
|
|
||
| # full path for CK library on compute-artifactory.amd.com | ||
| set(url "https://compute-artifactory.amd.com/artifactory/rocm-generic-local") |
There was a problem hiding this comment.
Consider reuse the release page of https://github.com/ROCm/composable_kernel
(There is no need to create new releases, just re-use any existing release to store the assets)
There was a problem hiding this comment.
Hmm, I don't think https://github.com/ROCm/composable_kernel is an option, since we don't own that repo and I don't think it'd be appropriate to store these artifacts on their repo, given that they are generated using scripts that are not even in their repo.
Alternately, we could instead create releases on https://github.com/ROCm/CK_kernels, since we own that repo.
@pruthvistony, can we file an OSRB request for CK_kernels repo (if we want to use the same solution for upstream)?
jithunnair-amd
requested changes
Apr 3, 2025
CMakeLists.txt
Outdated
| @@ -888,6 +895,13 @@ if(USE_ROCM) | |||
| endif() | |||
| endif() | |||
|
|
|||
| # CK shared lib linkage | |||
| if(USE_ROCM) | |||
| if(UNIX AND (USE_CK_FLASH_ATTENTION)) | |||
There was a problem hiding this comment.
Do we need the UNIX part since USE_ROCM is defined as cmake_dependent_option(USE_ROCM "Use ROCm" ON "LINUX" OFF)?
Also, please correct indentation
There was a problem hiding this comment.
aotriton.cmake is getting linked similarly using condition if(UNIX AND (USE_FLASH_ATTENTION...
I borrowed the logic from there and linked in a similar way.
Will correct the indentation and the spacing though.
| include(${CMAKE_SOURCE_DIR}/cmake/public/LoadHIP.cmake) | ||
|
|
||
| # full path for CK library on compute-artifactory.amd.com | ||
| set(url "https://compute-artifactory.amd.com/artifactory/rocm-generic-local") |
There was a problem hiding this comment.
@pruthvistony @jeffdaily Should this be a concern about putting artifactory links in a public repo?
… and code cleanup.
|
Jenkins build for 507e0aa5b054253f13bc8cdfe6826baea425f464 commit finished as FAILURE Detected error during Pytorch building: |
|
Jenkins build for 507e0aa5b054253f13bc8cdfe6826baea425f464 commit finished as FAILURE |
|
Jenkins build for 507e0aa5b054253f13bc8cdfe6826baea425f464 commit finished as FAILURE |
|
Jenkins build for 507e0aa5b054253f13bc8cdfe6826baea425f464 commit finished as FAILURE |
|
Jenkins build for 507e0aa5b054253f13bc8cdfe6826baea425f464 commit finished as FAILURE |
|
Jenkins build for 507e0aa5b054253f13bc8cdfe6826baea425f464 commit finished as FAILURE |
|
Jenkins build for 507e0aa5b054253f13bc8cdfe6826baea425f464 commit is in progress |
|
Jenkins build for 507e0aa5b054253f13bc8cdfe6826baea425f464 commit is in progress |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Refiled version of #2007
These are the changes made for this PR-
Removed generating and building CK kernels when building pytorch.
Download CK library from compute-artifactory and link with pytorch.
Enabled USE_CK_FLASH_ATTENTION based on USE_FLASH_ATTENTION option.
Added USE_CK_FLASH_ATTENTION as a cmake variable
Fixed lint/NIT errors using lintrunner.
Steps for validating CK library-
Note: I used MI-200, hence, used gfx90a.
As a proof for validation the sdpa.py should successfully run to completion.