[llvm-dev] PBI: ORE in Greedy RA

[Serguei Katkov via llvm-dev]

Hi All,

Please be informed that I plan to extend the statistic reported by Greedy RA to ORE.
The following changes are proposed:

  1.  Compute stats only if ORE allows extra analysis to save compile time
  2.  Add function level stats (to existed loop level)
  3.  Separate zero cost folded reloads from folded reloads (zero cost reloads are those which used in var area of patch points where stack usage is usually zero cost)
  4.  Add stats for number of copy virtual after RA.
  5.  For each stats, report also cost of locations where corresponding instruction are used basing on branch frequency information.
  6.  Some clean-up possibly - used Analysis remark instead of Missed and reduce the size of YAML output be removing extra reported args.

I'm starting with https://reviews.llvm.org/D100010. If you are using these remarks or have custom YAML parser and the changes might affect you, feel free to add yourself as a reviewer of D1000010 and I will add you as a reviewer to all further patches related to these changes.

Thank you,
Serguei.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20210407/362320ce/attachment.html>