[PATCH] D28380: [SelectionDAG] Handle inverted conditions when splitting into multiple branches.

[Geoff Berry via Phabricator via llvm-commits]

gberry added a comment.

I did some performance testing on aarch64 (kryo) and there was a slight improvement overall, though almost all of the differences look to be due to indirect I$ effects.
For x86, the static instruction count improves by .02% over all benchmarks, with no significant static instruction count regressions.

My primary motivation for getting this patch in is to address PR31382 in order to get my change (https://reviews.llvm.org/D27722) to move GVNHoist later in the optimization pipeline re-enabled


https://reviews.llvm.org/D28380