Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Translation fix #297

Merged
merged 4 commits into from
Jan 21, 2025
Merged

Translation fix #297

merged 4 commits into from
Jan 21, 2025

Conversation

angshuman-parashar
Copy link
Collaborator

Major fix for detecting stride changes in classic NestAnalysis code. Check commit log for 002f916 for full details.

… the

classic NestAnalysis code. Its job is to determine how the translation
vector (or stride vector) between dataspace point-sets changes between
consecutive loop iterations. Note that if the translation vector is the
"first-order derivative" of point sets, then this computation is effectively
the second-order derivative.

If this translation vector _changes_, then this code avoids computing the
delta between consecutive point-sets and instead discards the old point set
and installs the new one. This is done to emulate the behavior of simplistic
hardware pattern generators that are unable to describe hierarchical patterns.

However, this code needs a number of special cases, particularly stationarity
detection (translation == 0). This is because we do want to allow patterns
such as translation=<N,0,0,0,N,0,0,0,...>. However, this specific case
was being detected as a translation vector change, resulting in the old
point-set being thrown away and increasing the number of accesses to the next
storage level.

The entire logic has been re-written to carefully handle all cases.
OperationSpace::SaveAndSubtractIfSameStride().
@angshuman-parashar angshuman-parashar merged commit a12fff5 into master Jan 21, 2025
@angshuman-parashar angshuman-parashar deleted the translation_fix branch January 21, 2025 20:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Input Access Count Mismatch in Timeloop ISPASS 2020 Tutorial
1 participant