Another odd missed SpecConstr opportunity

Compiling the attached code with HEAD (and without case liberation) has this in the output of SpecConstr:

Rec {
$wa_s1G0 [Occ=LoopBreaker]
  :: forall s_aIU.
     Data.Array.Base.STUArray s_aIU GHC.Types.Int GHC.Types.Int
     -> GHC.Types.Int
     -> GHC.Prim.Int#
     -> GHC.Prim.Int#
     -> GHC.Prim.State# s_aIU
     -> (# GHC.Prim.State# s_aIU, () #)
[LclId, Arity=5, Str=DmdType LLLLL]
$wa_s1G0 =
  \ (@ s_aIU)
    (w_s1FA
       :: Data.Array.Base.STUArray s_aIU GHC.Types.Int GHC.Types.Int)
    (w_s1FB :: GHC.Types.Int)
    (ww_s1FE :: GHC.Prim.Int#)
    (ww_s1FI :: GHC.Prim.Int#)
    (w_s1FK :: GHC.Prim.State# s_aIU) ->
    case GHC.Prim.># ww_s1FE ww_s1FI of wild_Xk [Dmd=Just A] {
      GHC.Types.False ->
        case w_s1FA
        of wild_aRL [Dmd=Just L]
        { Data.Array.Base.STUArray ds2_aRN [Dmd=Just U]
                                   ds3_aRO [Dmd=Just U]
                                   ds4_aRP [Dmd=Just U]
                                   marr#_aRQ [Dmd=Just L] ->
        case GHC.Prim.readIntArray# @ s_aIU marr#_aRQ ww_s1FE w_s1FK
        of wild2_aRX [Dmd=Just A]
        { (# s2#_aRZ [Dmd=Just L], e#_aS0 [Dmd=Just L] #) ->
        case w_s1FB
        of wild1_aSy [Dmd=Just L] { GHC.Types.I# y_aSA [Dmd=Just L] ->
        case GHC.Prim.writeIntArray#
               @ s_aIU marr#_aRQ ww_s1FE (GHC.Prim.+# e#_aS0 y_aSA) s2#_aRZ
        of s2#_aSp [Dmd=Just L] { __DEFAULT ->
        $wa_s1G0
          @ s_aIU w_s1FA w_s1FB (GHC.Prim.+# ww_s1FE 1) ww_s1FI s2#_aSp
        }
        }
        }
        };
      GHC.Types.True -> (# w_s1FK, GHC.Unit.() #)
    }
end Rec }

i.e. we know the form of the first STUArray argument after the first iteration, but we leave it boxed and hence do the case every time around the loop.

Is constructor specialisation not meant to hit this case? It seems so simple that I feel that this problem must come from my misunderstanding rather than a compiler error.

Full command line:

ghc -fforce-recomp -c -dverbose-core2core -O2 -fno-liberate-case STUArray-Rewrite2.hs

(I disabled case liberation because I suspect it of having a bad effect elsewhere, but my understanding is that SpecConstr should still hit this case - IIRC all that CaseLib does that SpecConstr does not is deal with is hoisting case scrutinisation of the free variables of a function)

Note that this is a GHC version *with* the patch "Improve Simplifier and SpecConstr behaviour". I can reproduce this with the GHC 7.0 RC as well.

Trac metadata

Trac field	Value
Version	7.1
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture