[GHC] #13422: INLINE CONLIKE sometimes fails to inline

[GHC]

#13422: INLINE CONLIKE sometimes fails to inline
-------------------------------------+-------------------------------------
        Reporter:  dfeuer            |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.4.1
       Component:  Compiler          |              Version:  8.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Runtime           |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #7206             |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by simonpj):

 Here's the reason.  We get
 {{{
 let xs = case n of I# n1 -> cheapBuild blah
 in ...(foldr k z xs)...(foldr k2 z2 xs)...
 }}}
 So the `cheapBuild` is hidden behind that `case n` and the
 `foldr/cheapBuild` rule does not fire.

 The problem comes from
 {{{
 instance Enum Int where
     ...
     enumFromTo (I# x) (I# y) = eftInt x y
 }}}
 So `[1..n]` desugars into `(case n of I# n' -> eftInt 1# n')`.  This
 doesn't hurt normal foldr/build because if we see
 {{{
    foldr k z (case n of I# n' -> build blah)
 }}}
 we know that `foldr` is strict and so float the case outwards.  This
 doesn't happen with the `cheapBuild` stuff since the producer and consumer
 are further apart.

 But the solution is, I think, easy: move the evaluation of n into
 `eftInt`.  So we have
 {{{
 instance Enum Int where
     ...
     enumFromTo x y = eftInt x y
 }}}
 Now `eftInt` takes boxed `Ints` but it can evaluate them just fine.

 I've pushed a patch to `wip/cheap-build`, and it works just fine.

 '''However''':

 * It needs a serious note (steal the text above)

 * Other uses of `cheapBuild` need similar treatment.

 So the commit needs work.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13422#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler