#14672: Make likelyhood of branches/conditions available throughout the compiler. -------------------------------------+------------------------------------- Reporter: AndreasK | Owner: (none) Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.2 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D4316 Wiki Page: | Phab:D4324 Phab:D4327 -------------------------------------+------------------------------------- Comment (by svenpanne): Replying to [comment:10 AndreasK]:
I was talking about annotations primarily because they seem like the low hanging fruit.
[...] Once we have a way to pass/use the information throughout the
I assumed that, and it's definitely the right way to start. And annotations can sometimes be really useful, e.g. GCC has ''_builtin_expect'' for a reason. :-) In general, it's just a bit hard to find the places for such annotations: You have to profile your code, identify the hot spots, and figure out if a branch hint helps. PGO automates that process. passes adding PGO is the next logical step and should be possible without touching anything past core. PGO is not about branch hints alone, it is e.g. tremendously helpful to have an estimate how often a looping construct actually loops, know the usual target of an indirect branch etc. (Well, in a sense these are all some kind of branch hint, but in a much broader sense.) So it might be a good idea to plan for something more general than a relative branch frequency. But I guess even threading that simple information through the compiler and use it wisely is not a simple task, so let's not overengineer in the beginning... -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14672#comment:11 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler