[PATCH] D20776: [CFLAA] Teach cfl-aa to understand heap memory allocation

[George Burgess IV via llvm-commits]

george.burgess.iv added a comment.

Danny - ...Good point. I forgot that we were super conservative when something isn't in the sets.

Jia - Can we please change this to just add an edge from the malloc/calloc result (and free argument) to itself with zero stratified attributes, so that we have the relevant pointer in our sets? :)

We need to do this because, if the result of the malloc is otherwise unused, we'll not see it in the stratified sets. If something's not in the stratified sets, we'll assume it was added after we ran on the function, so we're conservative and always just say MayAlias. If we're not doing the same thing with otherwise-unused allocas, we probably should be.


http://reviews.llvm.org/D20776