[SWDEV-539215] - Autotune support for persistent reduction and no_x_dim removal #2417
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
|
Jenkins build for 34ad38c7edb1f03037c4c124daa0f5ee3c8b4ddf commit finished as FAILURE |
|
Jenkins build for 21a8eded2cbea28dce93e962ae0ed253432b3e22 commit finished as FAILURE |
|
UTs seem good |
|
@jataylo can you please cherry-pick this into release/2.8 as well? This is not cherry-picking cleanly. |
|
! cherry-pick --onto release/2.8 |
|
Created branch autogenerated/release/2.8_cherry-pick_pr-2417 and #2454. It contains a merge conflict. Please resolve it |
jataylo
added a commit
to jataylo/pytorch
that referenced
this pull request
Aug 11, 2025
…im removal (ROCm#2417) We noticed persistent reduction kernels can be extremely poor performing https://ontrack-internal.amd.com/browse/SWDEV-539215 The root cause is that in certain size restrictions and kernels "no_x_dim" mode is enabled, which embeds static XBLOCK=1 into the kernel. This means tuning is not optimal. Removing this mode and enabling autotune we achieve 2x performance proving that new heuristics must be made. We will bring this into 2.7 for perf uplift, discussion is undergoing with upstream on removing no_x_dim, if there is no perf regression they are in agreement. Draft PR shows no perf loss on ROCm for any inductor benchmark pytorch#159048 Removing tests because no longer relevant. (cherry picked from commit 6c845c6)
jataylo
added a commit
that referenced
this pull request
Aug 11, 2025
…ersistent reduction and no_x_dim removal (#2454) Cherry-pick of #2417 Need to resolve conflicts --------- Co-authored-by: Jack Taylor <[email protected]>
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.
We noticed persistent reduction kernels can be extremely poor performing https://ontrack-internal.amd.com/browse/SWDEV-539215
The root cause is that in certain size restrictions and kernels "no_x_dim" mode is enabled, which embeds static XBLOCK=1 into the kernel. This means tuning is not optimal. Removing this mode and enabling autotune we achieve 2x performance proving that new heuristics must be made.
We will bring this into 2.7 for perf uplift, discussion is undergoing with upstream on removing no_x_dim, if there is no perf regression they are in agreement. Draft PR shows no perf loss on ROCm for any inductor benchmark pytorch#159048
Removing tests because no longer relevant.
Cherry-picked to release/2.8 branch via #2454