Re: [GHC] #683: RULES for recursive functions don't work properly

#683: RULES for recursive functions don't work properly
-----------------------+----------------------------------------------------
  Reporter:  simonpj   |          Owner:  simonpj
      Type:  bug       |         Status:  new    
  Priority:  normal    |      Milestone:  6.6    
 Component:  Compiler  |        Version:  6.4.1  
  Severity:  normal    |     Resolution:         
  Keywords:            |             Os:  Unknown
Difficulty:  Unknown   |   Architecture:  Unknown
-----------------------+----------------------------------------------------
Old description:

> someone mentioned to me that this expression:
>
>   mapM_ action [n..m]
>
> isn't being optimised properly, so I thought I'd look into it.  Sure
> enough, we don't get down to a simple loop like we should, although the
> F/B transformation is happening (see ~simonmar/scratch/mapm.hs).  The
> problem is that GHC.Enum.eftIntFB isn't being inlined, it's defined like
> this:
>
> {-# INLINE [0] eftIntFB #-}
> eftIntFB :: (Int -> r -> r) -> r -> Int# -> Int# -> r
> eftIntFB c n x y | x ># y    = n
>                  | otherwise = go x
>                  where
>                    go x = I# x `c` if x ==# y then n else go (x +# 1#)
>
> but, strangely, the inlining doesn't appear in the interface:
>
> eftIntFB :: (GHC.Base.Int -> r -> r) -> r -> GHC.Prim.Int# ->
> GHC.Prim.Int# -> r
>   {- Arity: 4 HasNoCafRefs Strictness: LLLL -}
>
> This is becuase the RULE make eftIntFB look recursive, so its inlining
> isnt' exposed.  Bad, bad.

New description:

 someone mentioned to me that this expression:

   mapM_ action [n..m]

 isn't being optimised properly, so I thought I'd look into it.  Sure
 enough, we don't get down to a simple loop like we should, although the
 F/B transformation is happening (see ~simonmar/scratch/mapm.hs).  The
 problem is that GHC.Enum.eftIntFB isn't being inlined, it's defined like
 this:
 {{{
 {-# INLINE [0] eftIntFB #-}
 eftIntFB :: (Int -> r -> r) -> r -> Int# -> Int# -> r
 eftIntFB c n x y | x ># y    = n
                  | otherwise = go x
                  where
                    go x = I# x `c` if x ==# y then n else go (x +# 1#)
 }}}
 but, strangely, the inlining doesn't appear in the interface:
 {{{
 eftIntFB :: (GHC.Base.Int -> r -> r) -> r -> GHC.Prim.Int# ->
 GHC.Prim.Int# -> r
   {- Arity: 4 HasNoCafRefs Strictness: LLLL -}
 }}}
 This is becuase the RULE make eftIntFB look recursive, so its inlining
 isnt' exposed.  Bad, bad.

Comment (by simonpj):

 Fix formmatting only

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/683>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler_______________________________________________
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@xxxxxxxxxxx
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs