Re: [GHC] #8766: length [Integer] is twice as slow but length [Int] is 10 times faster

10 Mar 2014

      #8766: length [Integer] is twice as slow but length [Int] is 10 times faster
--------------------------------------------+------------------------------
        Reporter:  George                   |            Owner:
            Type:  bug                      |           Status:  new
        Priority:  normal                   |        Milestone:  7.8.1
       Component:  Compiler                 |          Version:  7.8.1-rc1
      Resolution:                           |         Keywords:
Operating System:  Unknown/Multiple         |     Architecture:
 Type of failure:  Runtime performance bug  |  Unknown/Multiple
       Test Case:  T8755                    |       Difficulty:  Unknown
        Blocking:                           |       Blocked By:
                                            |  Related Tickets:
--------------------------------------------+------------------------------
Changes (by nomeata):

 * owner:  nomeata =>
 * status:  closed => new
 * resolution:  fixed =>

Comment:

 Hmm, this is not fully fixed, it seems. Consider

 {{{
 f :: Integer -> Integer
 f n = n
 {-# NOINLINE f #-}

 main :: IO ()
 main = sum (map f [0 .. 10000]) `seq` return ()
 }}}

 Here, the rule
 {{{
 {-# RULES
 "enumDeltaToInteger1"   [0] forall c n x . enumDeltaToIntegerFB c n x 1 =
 up_fb c n x 1
  #-}
 }}}
 does not fire and we end up with
 {{{
 main3 :: Integer -> Integer
 main3 =
   enumDeltaToIntegerFB
     @ (Integer -> Integer) main6 (id @ Integer) main2 main5 main4

 main5 :: Integer
 main5 = __integer 1
 }}}
 and that despite that we have this code in `Phase = 0`:

 {{{
     enumDeltaToIntegerFB
            @ (Integer -> Integer)
            (\ (x :: Integer) (ys [OS=OneShot] :: Integer -> Integer) ->
               let {
                 ds [OS=ProbOneShot] :: Integer
                 ds = f x } in
               \ (ds2 :: Integer) -> ys (plusInteger ds2 ds))
            (id @ Integer)
            (__integer 0)
            (__integer 1)
            (__integer 10000)
            (__integer 0)
 }}}

 The rule engine is still a bit of a mystery to me. Why does the rule not
 match reliably here?

--
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8766#comment:14
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler