[PATCH] D15059: [IndVars] Have getInsertPointForUses preserve LCSSA

Sanjoy Das via llvm-commits llvm-commits at lists.llvm.org
Fri Dec 4 12:14:29 PST 2015

sanjoy added a comment.

In http://reviews.llvm.org/D15059#302745, @mzolotukhin wrote:

> Hi Sanjoy,
> Thanks for fixing this, and sorry for the delayed response - I was on vacation. The patch looks good to me!
> One question: are you certain that ind-vars was only breaking LCSSA for inner loops, and it's been always preserving LCSSA for other loops?

No, I'm far from certain that IndVars preserves LCSSA for outer loops always.  This is change just fixing the case that we have seen in practice.  My guess / intuition is that we'll eventually have to move to a more principled and more obviously correct approach around preserving LCSSA in IndVars in the future.

> Michael

Comment at: lib/Analysis/LoopInfo.cpp:203
@@ -202,1 +202,3 @@
+bool Loop::isRecursivelyLCSSAForm(DominatorTree &DT) const {
+  if (!isLCSSAForm(DT))
mzolotukhin wrote:
> Could you add a comment for this function please?
There is already a comment on the header.  I think we're moving away from repeating comments in both the .h and .cpp files.


More information about the llvm-commits mailing list