ngen: workaround for PVC SWSB HW bug #2443
Open
Conversation
This file contains 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
|
make test |
|
make test perf-gpu |
echeresh
approved these changes
Jan 18, 2025
|
make test perf-gpu |
rjoursler
approved these changes
Jan 21, 2025
| // Handle PVC HW bug with WAR dependencies on send instructions. | ||
| PVCWARWA pww; | ||
| if (pvcWARWA && !regions[0].empty() && computeSWSB) | ||
| pww = analyzePVCWARWA(hw, program, bb, phase, consumeOp, pvcWARWADeps); |
There was a problem hiding this comment.
Can we move this if statement into analyzePVCWARWA to prevent invalid use of that function?
There was a problem hiding this comment.
Sure, I can do that.
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Outdated suggestions cannot be applied.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.
Closes MFDNN-12992. Introduces a workaround for a PVC HW bug related to SWSB dependencies on send instruction src operands in the presence of heavy double precision usage. This is not an expected use case for oneDNN, but could theoretically occur if another f64-heavy kernel is running in parallel.
The workaround is carefully written to minimize performance impacts, but if performance impacts are still too high, we can enable the WA on an as-needs basis.