#6082: Program compiled with 7.4.1 runs many times slower than compiled with
7.2.2
--------------------------------------+-------------------------------------
Reporter: gchrupala | Owner: simonpj
Type: bug | Status: new
Priority: high | Milestone: 7.4.3
Component: libraries (other) | Version: 7.4.1
Resolution: | Keywords:
Os: Unknown/Multiple | Architecture: Unknown/Multiple
Failure: Runtime performance bug | Difficulty: Unknown
Testcase: | Blockedby:
Blocking: | Related:
--------------------------------------+-------------------------------------
Comment(by milan):
I am not a specialist, but considering the code
{{{
module Data.Array.MArray
{-# INLINE unsafeFreeze #-}
unsafeFreeze :: (Ix i, MArray a e m, IArray b e) => a i e -> m (b i e)
unsafeFreeze = Data.Array.Base.unsafeFreeze
module Data.Array.Unsafe
(exports Data.Array.Base.unsafeFreeze)
module Data.Array.Base
{-# INLINE unsafeFreeze #-}
unsafeFreeze :: (Ix i, MArray a e m, IArray b e) => a i e -> m (b i e)
unsafeFreeze = freeze
{-# RULES
"unsafeFreeze/STArray" unsafeFreeze = ArrST.unsafeFreezeSTArray
"unsafeFreeze/STUArray" unsafeFreeze = unsafeFreezeSTUArray
#-}
}}}
then when calling Data.Array.Base.unsafeFreeze, RULES have chance to fire
before the INLINing happen. But when calling
Data.Array.MArray.unsafeFreeze, RULES do not fire (as they fire only on
Data.Array.Base.unsafeFreeze) and then '''both''' INLINing happen at once,
replacing the Data.Array.MArray.unsafeFreeze by Data.Array.Base.freeze.
When the Data.Array.Base module postpones the INLINing of unsafeFreeze to
phase 1, as in
{{{
module Data.Array.Base
{-# INLINE[1] unsafeFreeze #-}
unsafeFreeze :: (Ix i, MArray a e m, IArray b e) => a i e -> m (b i e)
unsafeFreeze = freeze
}}}
everything works as expected.
I am not a RULES expert, so I do not know, what the right behaviour is. In
a situation like this:
{{{
{-# INLINE f #-}
f = g
{-# INLINE g #-}
g = h
{-# RULES "g/gspec" g = gspec #-}
}}}
is f expected to be rewrittenn to h or gspec? In our array situation, GHC
follows the f -> g -> h path.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/6082#comment:10>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
