This is an archive of the discontinued LLVM Phabricator instance.

[AArch64] Allocate the modified and used regs once per function.
ClosedPublic

Authored by mcrosier on Feb 1 2016, 2:36 PM.

Details

Summary

Rather than once per load/store in the function. NFC, but hopefully this has a non-trivial improvement on compile-time.

Chad

Diff Detail

Event Timeline

mcrosier updated this revision to Diff 46580.Feb 1 2016, 2:36 PM
mcrosier retitled this revision from to [AArch64] Allocate the modified and used regs once per function..
mcrosier updated this object.
mcrosier added subscribers: llvm-commits, junbuml, mssimpso and 2 others.

Looks good to me.

mgrang added a subscriber: mgrang.Feb 1 2016, 3:38 PM
mgrang added inline comments.
lib/Target/AArch64/AArch64LoadStoreOptimizer.cpp
1020

Can you please update these comments to reflect that the bitfield trackers are being reset here?

mcrosier accepted this revision.Feb 2 2016, 7:07 AM
mcrosier added a reviewer: mcrosier.

Looks good to me.

This revision is now accepted and ready to land.Feb 2 2016, 7:07 AM

Committed r259510.

lib/Target/AArch64/AArch64LoadStoreOptimizer.cpp
1020

Sure, Mandeep. I think the call to reset makes it pretty clear we're resetting the vector at this point, so I didn't add a comment here. However, I did add an additional comment where the vectors are being resized.

mcrosier closed this revision.Feb 2 2016, 7:09 AM