i#6471 sched idle: Add replay fallback and fix #6481
Merged
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
Adds a fallback to return EOF when all outputs are past the last record in replay mode, to handle a schedule where not every input reaches EOF or has regions of interest. Adds a unit test for this case, which hangs without the fix. Fixes a record bug which seems to cause the final segment entry for an input to have a stop instruction ordinal prior to the input's real endpoint. This is what caused hangs in actual usage of PR #6472 and led to adding the fallback just described (which is useful on its own). On selecting a new input which turned out to be at EOF, the original code called close_schedule_segment() on *prev_input* for some reason. That replaces the default -1 sentinel (read to EOF) with the current ordinal. The code then tries again for a new input, but if it ends up finding the prior one it would keep executing that without a new schedule segment entry. If that's the last input on that core, close_schedule_segment() is not called (the default is relied upon). I observed missing sentinels for one input in some runs and this is my theory as to what happened; those cases are gone now with this fix. Unfortunately it is difficult to create a unit test for this, but testing on the original hangs (without the fallback) show it fixed. Issue: #6471
abhinav92003
approved these changes
Nov 29, 2023
…s to test schedule
|
x32 failure is AMD #6417 |
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Outdated suggestions cannot be applied.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.
Adds a fallback to return EOF when all outputs are past the last record in replay mode, to handle a schedule where not every input reaches EOF or has regions of interest. Adds a unit test for this case, which hangs without the fix.
Fixes a record bug which seems to cause the final segment entry for an input to have a stop instruction ordinal prior to the input's real endpoint. This is what caused hangs in actual usage of PR #6472 and led to adding the fallback just described (which is useful on its own). On selecting a new input which turned out to be at EOF, the original code called close_schedule_segment() on prev_input for some reason. That replaces the default -1 sentinel (read to EOF) with the current ordinal. The code then tries again for a new input, but if it ends up finding the prior one it would keep executing that without a new schedule segment entry. If that's the last input on that core, close_schedule_segment() is not called (the default is relied upon). I observed missing sentinels for one input in some runs and this is my theory as to what happened; those cases are gone now with this fix. Unfortunately it is difficult to create a unit test for this, but testing on the original hangs (without the fallback) show it fixed.
Issue: #6471