Repository navigation
Conversation
…e clearing the maps A lane reuses its scratch memory for its next sector. At the first fragment the charge and peak maps were cleared without waiting for the gather of the previous sector, which runs on another stream, and the maps can overlap that sector's clusterByRow: on pbpb30 the peak map of sector 3 ends at row 0 of sector 0. When the gather starts late, row 0 of the previous sector is gathered as zeros: on a 20 kHz Pb-Pb time frame on Metal, the first run in a process found 0.42 % fewer sector tracks. The existing wait on the gather sat on the clusterInRow clear, after the map clears. Move it to the first map clear at fragment 0.
Member
Author
|
@davidrohr on metal this seems to be a genuine issue, with less tracks consistently missing when running on the 20 kHz reference data. Could you have a look if it makes sense? |
Collaborator
|
I think the error is genuine, but I do not like the fix, since it adds synchronization and affects GPU performance. Interesting that I never saw this problem on AMD / NVIDIA. But that is the advantage of testing on different hardware :). |
Member
Author
|
Yours works and it's obviously better. Closing. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
A lane reuses its scratch memory for its next sector. At the first
fragment the charge and peak maps were cleared without waiting for the
gather of the previous sector, which runs on another stream, and the maps
can overlap that sector's clusterByRow: on pbpb30 the peak map of sector 3
ends at row 0 of sector 0. When the gather starts late, row 0 of the
previous sector is gathered as zeros: on a 20 kHz Pb-Pb time frame on
Metal, the first run in a process found 0.42 % fewer sector tracks.
The existing wait on the gather sat on the clusterInRow clear, after the
map clears. Move it to the first map clear at fragment 0.