[PATCH] D22634: [CFLAA] Add support for field offset in CFLAnders::FunctionInfo

[Jia Chen via llvm-commits]

grievejia added inline comments.

================
Comment at: lib/Analysis/CFLAndersAliasAnalysis.cpp:527
@@ +526,3 @@
+        if (LHSEnd > RHSStart && LHSStart < RHSEnd)
+          return (LHSStart == 0) ? MayAlias : PartialAlias;
+      }
----------------
george.burgess.iv wrote:
> `PartialAlias` doesn't seem quite right here. `PartialAlias` is essentially `MustAlias` (at some offset, as you've noted), and, unless I'm missing something obvious, I think we'd need to do many more checks to prove `MustAlias`.
> 
> (If I *am* wrong, s/`MayAlias`/`MustAlias`/, please)
My original thought was that PartialAlias = MayAlias at nonzero offset.

If PartialAlias = MustAlias at nonzero offset, then we probably don't have anything smarter than MayAlias, since the analysis only handles may info. 


https://reviews.llvm.org/D22634