[PATCH] D43457: [Evaluator] Improve evaluation of load/store

[Teresa Johnson via Phabricator via llvm-commits]

tejohnson added a comment.

Sorry for the delay. I don't feel super confident reviewing these change. I looked back at who touched the ComputeLoadResult code last (besides pcc who simply extracted it out of GlobalOpt into Utils), and it was Chris Lattner in 2005! So I am not completely sure who to add to as a reviewer at this point.
Poking around LLVM though,  I see that there is some very similar handling in llvm::ConstantFoldLoadFromConstPtr (ConstantFolding.cpp). In particular, the handling of P as a GlobalVariable or GEP appears identical to what is already in Evaluator::ComputeLoadResult. It already has handling for bitcasts, although that handling appears quite a bit more complex that what you added to ComputeLoadResult. I'm wondering whether ComputeLoadResult can be refactored to just call ConstantFoldLoadFromConstPtr, or whether there is a reason it needs to remain distinct (and why the bitcast handling added here is simpler).


https://reviews.llvm.org/D43457