Hacker News Reader

LLVM recently introduced a SelectOptimize pass, which is used for select -> branch conversion in the presence of PGO information. This should replace the current non-PGO-aware X86 cmov conversion pass.

The result of this should be that without PGO cmov is preferred, which is the right default (incorrectly picking a cmov is a bit worse than picking a branch, incorrectly picking a branch is catastrophically worse than picking a cmov).

View on HN

Is it the right default? If the vast majority of branches are predictable, then without PGO a branch would be the right choice, no?

If they're gonna make this behavior the default, I presume they have lots of benchmarks showing it's generally the correct choice..

The vast majority of branches don't fit in the branch predictor, so it is taking no more than a naive guess about them. If you tell LLVM that the branch is likely or unlikely, they shouldn't use a CMOV, and should instead lay out the code so that the naive branch predictor is correct most of the time. Otherwise, the CMOV actually is likely to be beneficial.

Reply on news.ycombinator.com