[PATCH] D38952: [ARM] Allow unrolling on multi-block loops.

[Sam Parker via Phabricator via llvm-commits]

samparker added a comment.

We don't run LNT on these cores, but I have numbers from running some industry standard benchmarks, but unfortunately I cannot share exact details. I've observed that half of the benchmarks gain a speedup and there's only one regression. The speedups are between 1 and 20%, with half of those being 5% and over.


https://reviews.llvm.org/D38952