[PATCH] D13042: [SCEV] Try to prove predicates by splitting them.

[Sanjoy Das via llvm-commits]

sanjoy added inline comments.

================
Comment at: lib/Analysis/ScalarEvolution.cpp:6981
@@ +6980,3 @@
+  if (Pred == ICmpInst::ICMP_ULT && isKnownNonNegative(RHS) &&
+      isKnownPredicate(CmpInst::ICMP_SGE, LHS,
+                       getConstant(LHS->getType(), 0)) &&
----------------
hfinkel wrote:
> Why do you use isKnownNonNegative(RHS) but not isKnownNonNegative(LHS)?
> 
`isKnownNonNegative` is less powerful than `isKnownPredicate`.  `isKnownPredicate` does a control dependence analysis while `isKnownNonNegative` just looks at `getSignedRange` which does not take most control dependence into account (it does take //some// control dependence into account indirectly, by leveraging the no-wrap flags on SCEV instructions).

The cases we're looking to optimize are of the form `%iv ult %length`.  Here `%length` is a load instruction with `!range` metadata that lets us directly prove that `%length` is positive without control dependence, and `%iv` is an induction variable which takes some control dependence analysis to prove as being always positive.

I cannot try this out before tomorrow, but I think using `isKnownPredicate` to prove `RHS` is positive as well should work (except we'll be burning more compile time).  Another possibility is to keep the code as is, and add what I said here as a comment.


http://reviews.llvm.org/D13042