#1643: Unnecessary dictionaries
-------------------------+--------------------------------------------------
Reporter: guest | Owner:
Type: bug | Status: new
Priority: normal | Milestone:
Component: Compiler | Version: 6.7
Severity: normal | Resolution:
Keywords: | Difficulty: Unknown
Os: MacOS X | Testcase:
Architecture: x86 |
-------------------------+--------------------------------------------------
Comment (by [EMAIL PROTECTED]):
If you look at the Core, you can see that the definition of the $dMonad
inside qsort looks like this:
{{{
$dMonad_s5Hx :: forall s_a5pv.
GHC.Base.Monad (CTBase.E' CTBase.RValue (GHC.ST.ST
s_a5pv))
[]
$dMonad_s5Hx =
\ (@ s_a5pv) ->
case CT.$wic
@ (GHC.ST.ST s_a5pv)
@ (GHC.STRef.STRef s_a5pv)
@ GHC.Base.Int
(MonadRef.$f1 @ s_a5pv)
of ww_s3pb [Dead Nothing] { (# ww1_s3pd, ww2_s3pe, ww3_s3pf, ww4_s3pg
#) ->
GHC.Base.:DMonad
@ (CTBase.E' CTBase.RValue (GHC.ST.ST s_a5pv))
ww1_s3pd
ww2_s3pe
ww3_s3pf
ww4_s3pg
}
}}}
Because the rhs of {{{$dMonad_s5Hx}}} isn't cheap, it doesn't get inlined
inside a lambda (e.g., in the body of {{{qsort}}}. If the call to
{{{CT.$wic}}} inside {{{$dMonad_s5Hx}}} were inlined, then
{{{$dMonad_s5Hx}}} would be inlined as well. But by default, GHC thinks
{{{CT.$wic}}} is too big to inline. If you pass in the flag
{{{-funfolding-use-threshold36}}} to increase the threshold for the
allowable size of an unfolding, then you get the results you want, but I
haven't measured whether that actually increases performance here.
-Tim
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/1643>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
