This is an archive of the discontinued LLVM Phabricator instance.

Change default implementation of doMemOpsHaveSameBasePtr
AbandonedPublic

Authored by rampitec on Sep 12 2017, 12:09 PM.

Details

Summary

This changes default implementation of the doMemOpsHaveSameBasePtr
target callback used for load/store clustering as requested in the
D37698.

The implementation is switched to the old one for the AArch64
target as it seems to expect same base registers with ordered
offsets for further combining purposes.

The change is based on the D37698.

Diff Detail

Event Timeline

rampitec created this revision.Sep 12 2017, 12:09 PM
hfinkel edited edge metadata.Sep 13 2017, 2:08 AM

Please see my comments in D37698. If the primary purpose of this on AArch64 is to enable the AArch64LoadStoreOptimizer, I'm not sure that this makes sense.

Please see my comments in D37698. If the primary purpose of this on AArch64 is to enable the AArch64LoadStoreOptimizer, I'm not sure that this makes sense.

Right. After I did it and run testing I have realized only to targets in the tree use these mutations, and AArch64 needs exactly current implementation. So I would keep it AMD specific untill there are other users.

rampitec abandoned this revision.Sep 13 2017, 11:50 AM