#7994: Make foldl into a good consumer -------------------------------------+------------------------------------ Reporter: simonpj | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.3 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Unknown/Multiple Type of failure: None/Unknown | Difficulty: Unknown Test Case: | Blocked By: Blocking: | Related Tickets: -------------------------------------+------------------------------------ Comment (by nomeata): With the patch above, the benefits from foldl-via-buildr and the call arity analysis are {{{ Min -0.2% -74.5% -9.6% -10.6% -50.0% Max +0.9% +1.7% +4.1% +4.1% +6.3% Geometric Mean -0.0% -4.0% -0.6% -0.7% -0.7% }}} with only one increase: `nucleic2`. It comes from a use of `maximum`, and it goes away when that is changed from `NOINLINE [1]` to `INLINE [1]` (GHC was inlining `maximum` while `foldl` was implemented naively, but not any more, so let's give it a push). Then we get the very pleasing result of no single benchmark with allocation increase: {{{ Min -0.1% -74.5% -6.8% -8.3% -50.0% Max +0.2% 0.0% +38.5% +38.5% 0.0% Geometric Mean -0.0% -4.1% +7.7% +7.7% -0.8% }}} This is the current state in the branch `wip/T7994-calledArity` of ghc, and the branch `wip/T7994` of base. The effect on compilation times needs to be measured, and quite possibly improved (e.g. not a full simplifier run after it, but only eta-expansion where possible and then a gentle simplification of the RHS). -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/7994#comment:15 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler