Skip to content

Make gshareN_ahead slightly more accurate.#27

Merged
alindsay-ampere merged 1 commit into
AmpereComputing:mainfrom
PierreMichaud-Inria:main
Mar 23, 2026
Merged

Make gshareN_ahead slightly more accurate.#27
alindsay-ampere merged 1 commit into
AmpereComputing:mainfrom
PierreMichaud-Inria:main

Conversation

@PierreMichaud-Inria

Copy link
Copy Markdown
Contributor

The optimal global history length is a tradeoff between footprint (that is, PHT aliasing) and branch correlations. Global history footprint is a function of the global history length in branches, not in blocks. So we shift the global history by the number of branches in the block. NB: the same optimization could be applied to other gshare predictors.

The optimal global history length is a tradeoff between footprint (that is, PHT aliasing)
and branch correlations.  Global history footprint is a function of the global history length
in branches, not in blocks. So we shift the global history by the number of branches in the block.
NB: the same optimization could be applied to other gshare predictors.
@alindsay-ampere alindsay-ampere merged commit 2b6ac38 into AmpereComputing:main Mar 23, 2026
30 checks passed
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.

2 participants