Re: [GHC] #14941: Switching direct type family application to EqPred (~) prevents inlining in code using vector (10x slowdown)

19 Mar 2018

      #14941: Switching direct type family application to EqPred (~) prevents inlining in
code using vector (10x slowdown)
-------------------------------------+-------------------------------------
        Reporter:  nh2               |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.2.2
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Runtime           |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by mpickering):

 I copied the core of `selectVectorDestructive2` from the good and bad
 examples and diffed them. They looked basically the same apart from lots
 and lots of code like the following:

 In bad:

 {{{
 1# ->
                                   case $wlvl_rnHf
                                          @ VUM.MVector
                                          @ Int
                                          @ IO
 Data.Vector.Unboxed.Base.$fMVectorMVectorInt
 ((Data.Vector.Primitive.Mutable.MVector
                                              @ ghc-
 prim-0.5.2.0:GHC.Prim.RealWorld
                                              @ Int
                                              ww1_smP2
                                              ww2_smP3
                                              ww3_smP4)
                                           `cast` (Sym
 (Data.Vector.Unboxed.Base.N:R:MVectorsInt[0]
                                                            <ghc-
 prim-0.5.2.0:GHC.Prim.RealWorld>_N) ; Sym
 (Data.Vector.Unboxed.Base.DR:MVectorsInt0[0]
 (Control.Monad.Primitive.D:R:PrimStateIO[0]))
                                                   ::
 (Data.Vector.Primitive.Mutable.MVector
                                                         ghc-
 prim-0.5.2.0:GHC.Prim.RealWorld
                                                         Int :: *)
                                                      ~R# (VUM.MVector
 (PrimState IO) Int :: *)))
                                          ww4_smP8
                                   of wild_00 {
                                   }
 }}}

 In good:

 {{{
 1# -> case $wlvl_rnRh ww2_smWB ww3_smWC ww4_smWG of wild_00 { }
 }}}

 So it looks like on every call the `MVector` is being constructed rather
 than created one and shared? Anyone know immediately why this is such a
 problem?

-- 
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14941#comment:3
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler