[PATCH] D44884: Improving 'addArgumentAttrsFromCallsites'

[Buddhika Chamith via Phabricator via llvm-commits]

chamibuddhika created this revision.
chamibuddhika added reviewers: nlopes, philames, echristo.
Herald added subscribers: llvm-commits, mgorny, jfb, mehdi_amini.

This patch aims to fix a TODO inside addArgumentAttrsFromCallsites method in function attribute inference code. This is more like a RFC patch where I need some feedback if the direction taken is valid. Idea was to introduce a PostDominatorTree argument to the method. It could be used to check if the callsite dominates the entry block and if so it is safe to transfer its callee attributes to the callers arguments.

This patch currently does *not* pass regression tests. I suspect this is due to an issue with how PostDominatorTree pass is introduced in the PassManager. I am still not completely familiar with how PassManager works.


Repository:
  rL LLVM

https://reviews.llvm.org/D44884

Files:
  include/llvm/Transforms/IPO/FunctionAttrs.h
  lib/Transforms/CMakeLists.txt
  lib/Transforms/IPO/FunctionAttrs.cpp
  lib/Transforms/IPO/PassManagerBuilder.cpp
  lib/Transforms/IPO/WholeProgramDevirt.cpp

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D44884.139751.patch
Type: text/x-patch
Size: 33846 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180325/c1a2bcca/attachment.bin>