[PATCH] D17337: __builtin_object_size problem with pointer as argument

[Ahmed Bougacha via llvm-commits]

ab added a subscriber: ab.
ab added a comment.

While I don't think the builtin is intended to be used this way (that is, this behavior is intentional), the only argument I have against this patch is that it's basically wasting time on "dead" code.

That said, why not try to evaluate the objectsize call when we're about to remove it?
In the testcase, I think that we erase the objectsize call (and the alloca) before evaluating it only because InstCombine examines the alloca first.


Repository:
  rL LLVM

http://reviews.llvm.org/D17337