[GHC] #14068: Loopification using join points

[GHC]

#14068: Loopification using join points
-------------------------------------+-------------------------------------
        Reporter:  nomeata           |                Owner:  nomeata
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.0.1
      Resolution:                    |             Keywords:  JoinPoints
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #13966 #14067     |  Differential Rev(s):  Phab:D3811
  #14827                             |
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by nomeata):

 > > Without loopification we get a specialization for drop, with
 loopification we do not.
 >
 > I wonder why. Are you happy to pursue, or would you like help?

 It seems rather clear to me *why*: We get this code for `drop`, which does
 not satisfy the conditions for !SpecConstr; in particular, `drop` is not
 recursive (!SpecConstr only creates specializations for recursive
 functions):

 {{{
 drop n xs =
   case xs of
     []     -> []
     (y:ys) -> case n of
        I# n# -> case n# of
          0# -> []
          _ ->
            joinrec j xs n# = case xs of
               []     -> []
               (y:ys) -> case n# of
                  0# -> []
                  _ -> call j ys (n# -# 1#)
            in j ys (n# -# 1#)
 }}}

 But I think I need your guidance as to what we should do about it!

 Should we simply enable !SpecConstr even for all non-recursive functions?
 (My gut feeling is that this will blow up code size and compilation times
 too much.)

 Maybe we need to float the `joinrec … in …` out out of `drop`, so that the
 remaining code in  `drop` (which does the case analysis on `n`) is small
 enough to be always inlined?

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14068#comment:33>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler