[PATCH] D105996: [AArch64] Enable Upper bound unrolling universally

[Dave Green via Phabricator via llvm-commits]

dmgreen added a comment.

Do you have number for other benchmarks too? I believe that whether the option is beneficial - although it can depend on the sophistication of the cpu's branch predictor - is more dependent on the code that is being compiled. Some code is better suited for this type of unrolling, in cases because it helps further simplification or it just happens not to run into the drawbacks of the extra branches. It appears that SPEC is a win - I'm wondering if the llvm-test-suite would be the same, or geekbench as Florian mentioned.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105996/new/

https://reviews.llvm.org/D105996