Once I actually add a 'dist_fast_inline_caller', that indirection disappears in the inlined code, just as it does for dist_fast itself.

   dist_fast_inlined_caller :: UArr Double -> UArr Double -> Bool
   dist_fast_inlined_caller p1 p2 = dist_fast_inlined p1 p2 > 2

However, in the simpl output for 'dist_fast_inline_caller', the
'sumU' and 'zipWithU' still don't seem to be fused - Don?

All the 'seq's and so on should be unnecessary, and even so, I still get
the expected fusion:

As I said, I don't get the fusion if I just add the function above to the original Dist.hs, export it and compile the module with '-c -O2 -ddump-simpl':

Dist.dist_fast_inlined_caller =
 \ (w1_s1nb :: Data.Array.Vector.UArr.UArr GHC.Types.Double)
   (w2_s1nc :: Data.Array.Vector.UArr.UArr GHC.Types.Double) ->
   case (Dist.$wzipWithU Dist.lvl2 w1_s1nb w2_s1nc)
        `cast` (trans
                  Data.Array.Vector.UArr.TFCo:R56:UArr
                  Data.Array.Vector.UArr.NTCo:R56:UArr
                :: Data.Array.Vector.UArr.UArr GHC.Types.Double
                     ~
                   Data.Array.Vector.Prim.BUArr.BUArr GHC.Types.Double)
   of _
   { Data.Array.Vector.Prim.BUArr.BUArr ipv_s1lb
                                        ipv1_s1lc
                                        ipv2_s1ld ->
   letrec {
     $wfold_s1nN :: GHC.Prim.Double#
                    -> GHC.Prim.Int#
                    -> GHC.Prim.Double#
     LclId
     [Arity 2
      Str: DmdType LL]
     $wfold_s1nN =
       \ (ww_s1mZ :: GHC.Prim.Double#) (ww1_s1n3 :: GHC.Prim.Int#) ->
         case GHC.Prim.==# ww1_s1n3 ipv1_s1lc of _ {
           GHC.Bool.False ->
             $wfold_s1nN
               (GHC.Prim.+##
                  ww_s1mZ
                  (GHC.Prim.indexDoubleArray#
                     ipv2_s1ld (GHC.Prim.+# ipv_s1lb ww1_s1n3)))
               (GHC.Prim.+# ww1_s1n3 1);
           GHC.Bool.True -> ww_s1mZ
         }; } in
   case $wfold_s1nN 0.0 0 of ww_s1n7 { __DEFAULT ->
   GHC.Prim.>## (GHC.Prim.sqrtDouble# ww_s1n7) 2.0
   }
   }

Claus


_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe

Reply via email to