Here's a mystery.
* I took `t-10-getline.hs` and `t-10-getine-appl.hs` from `examples.zip` just before comment:44
* I compiled both with `-O -dshow-passes -ddump-simpl`.
* Both produced ''identical'' Core. In the early stages `getline-appl` is a little bigger, but they become identical after the first run of the simplifier following specialise/float-out. (I.e. early)
I am at a loss for how to reconcile these results with the dramatically worse compile times for `getline` compared with `getline-appl` reported in
#7258: Compiling DynFlags is jolly slow -------------------------------------+------------------------------------- Reporter: simonpj | Owner: simonpj Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.1 Resolution: | Keywords: deriving-perf Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple performance bug | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by tdammers): Replying to [comment:49 simonpj]: the graph plot in comment:43.
Tobias, any ideas?
Hmm, maybe `-O` vs. `-O2`? I'll run a few trials myself. Other thought, though probably not really an explanation: the t-10-XXXX examples are maybe too small to produce useful metrics, the performance differences only start to get dramatic past 100 fields or so. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/7258#comment:51 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler