[PATCH] D15828: [PGO] Passmanagerbuilder change that enable IR level PGO instrumentation

[David Li via llvm-commits]

davidxl added inline comments.

================
Comment at: lib/Transforms/IPO/PassManagerBuilder.cpp:281
@@ -239,1 +280,3 @@
 
+  addPGOInstrPasses(MPM);
+
----------------
chandlerc wrote:
> davidxl wrote:
> > joker.eph wrote:
> > > There will be an inliner SCC pass manager created in the PGO InstrPasses. It is not clear to me why you don't integrate this with the regular inliner a few line below.
> > This is the pre-inline pass that needs to happen before instrumentation. See the original RFC: http://lists.llvm.org/pipermail/llvm-dev/2015-August/089425.html for details.
> Yes, and there was not consensus in the RFC for the pre-inlining. Instead, the approach suggested was to work on independent components which seem non-controversial and have a separate focused discussion about pre-inlining and other interactions.
> 
> I think there are still a lot of unanswered questions around the pre-inlining phase (based on my reading of the RFC thread you pointed at). If there is some other thread that I've not seen, feel free to direct there...
The original RFC discussion is here: https://groups.google.com/forum/#!topic/llvm-dev/ncONLhVJ3L0




http://reviews.llvm.org/D15828