#9342: Branchless arithmetic operations -------------------------------------+------------------------------------- Reporter: hvr | Owner: Type: feature request | Status: new Priority: normal | Milestone: 8.0.1 Component: Compiler | Version: 7.8.3 (CodeGen) | Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): -------------------------------------+------------------------------------- Comment (by svenpanne): I'd like to point out my previous email on a similar topic: https://mail.haskell.org/pipermail/ghc-devs/2015-April/008858.html In a nutshell: It's far from clear that code with branches is "sub- optimal", toy benchmarks are totally meaningless (they don't stress e.g. the very limited number of shifting units in CPUs), and benchmarks on a single (micro-)architecture are even less meaningful (what's good for one HW makes things worse on another HW). Applying rules of thumb from previous decades doesn't really work anymore today, this has already been learned the hard way in several other compilers/JITs, I just wanted to warn about this again... -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/9342#comment:6 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler