[PATCH] D27734: [LoopUnroll] Enable PGO-based loop peeling by default

[David Li via Phabricator via llvm-commits]

davidxl added a comment.

Regarding #1, this should not be a concern for PGO. PGO is a mode (same as O3 and above) that aims to maximize program speed at cost of possible size and compile time increase.  besides, for large apps, the size increase should be minimal. bzip2 is an outlier as it is tiny.  For reference, GCC turns on loop peeing with PGO as well.

Regarding Adam's question: I have not yet seen users using Os + PGO combination. If this really is a concern, Os should also be checked (under PGO) to avoid size growth.


https://reviews.llvm.org/D27734