D39346: [LV] [ScalarEvolution] Fix PR34965 - Cache pointer stride information before LV code gen

Saito, Hideki via llvm-commits llvm-commits at lists.llvm.org
Wed Nov 22 11:46:58 PST 2017


Great. Thanks!

-----Original Message-----
From: Renato Golin via Phabricator [mailto:reviews at reviews.llvm.org] 
Sent: Wednesday, November 22, 2017 10:25 AM
To: Caballero, Diego <diego.caballero at intel.com>; mkuper at google.com; hfinkel at anl.gov; Rapaport, Gil <gil.rapaport at intel.com>; Zaks, Ayal <ayal.zaks at intel.com>; Saito, Hideki <hideki.saito at intel.com>
Cc: siddu.druid at gmail.com; Elovikov, Andrei <andrei.elovikov at intel.com>; roger.ferreribanez at arm.com; llvm-commits at lists.llvm.org
Subject: [PATCH] D39346: [LV] [ScalarEvolution] Fix PR34965 - Cache pointer stride information before LV code gen

rengolin accepted this revision.
rengolin added a comment.
This revision is now accepted and ready to land.

Hi Diego,

Ok, I agree with Hideki that caching is good as well as re-evaluating and making sure that we don't degrade. However, as you both said, detecting degradation will require SCEV to go beyond its current design and new cases in the vectoriser we don't have yet.

From that point of view, I agree with Hideki that this can be a baby-step towards a better model, so the patch would be ok for now.

Can you please add some comments to `collectIsConsecutivePtr` to reflect that? Otherwise, LGTM. Thanks!


https://reviews.llvm.org/D39346





More information about the llvm-commits mailing list