[GHC] #15019: Fix performance regressions from #14737

[GHC]

#15019: Fix performance regressions from #14737
-------------------------------------+-------------------------------------
        Reporter:  tdammers          |                Owner:  tdammers
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler          |              Version:
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #14737            |  Differential Rev(s):  phab:D4568
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by tdammers):

 OK, sure.

 Just check out GHC HEAD, and apply this patch:

 {{{
 diff --git a/compiler/coreSyn/CoreOpt.hs b/compiler/coreSyn/CoreOpt.hs
 index f1ff68d133..6ddab2c94e 100644
 --- a/compiler/coreSyn/CoreOpt.hs
 +++ b/compiler/coreSyn/CoreOpt.hs
 @@ -958,10 +958,14 @@ pushCoTyArg :: CoercionR -> Type -> Maybe (Type,
 Maybe CoercionR)
  -- If the returned coercion is Nothing, then it would have been
 reflexive;
  -- it's faster not to compute it, though.
  pushCoTyArg co ty
 +  | isReflCo co
 +  = Just (ty, Nothing)
 +
    -- The following is inefficient - don't do `eqType` here, the coercion
    -- optimizer will take care of it. See Trac #14737.
 -  -- -- | tyL `eqType` tyR
 -  -- -- = Just (ty, Nothing)
 +  | tyL `eqType` tyR
 +  = pprTrace "eqType fired: " (ppr (co, tyL, tyR)) $
 +    Just (ty, Nothing)

    | isForAllTy tyL
    = ASSERT2( isForAllTy tyR, ppr co $$ ppr ty )
 @@ -993,10 +997,15 @@ pushCoValArg :: CoercionR -> Maybe (Coercion, Maybe
 Coercion)
  -- If the second returned Coercion is actually Nothing, then no cast is
 necessary;
  -- the returned coercion would have been reflexive.
  pushCoValArg co
 +  -- The "easy" case; this doesn't interest us.
 +  | isReflCo co
 +  = Just (mkRepReflCo arg, Nothing)
 +
    -- The following is inefficient - don't do `eqType` here, the coercion
    -- optimizer will take care of it. See Trac #14737.
 -  -- -- | tyL `eqType` tyR
 -  -- -- = Just (mkRepReflCo arg, Nothing)
 +  | tyL `eqType` tyR
 +  = pprTrace "eqType fired: " (ppr (co, tyL, tyR)) $
 +    Just (mkRepReflCo arg, Nothing)

    | isFunTy tyL
    , (co1, co2) <- decomposeFunCo Representational co
 }}}

 Build with the `validate` flavor, and then:

 {{{
 ./inplace/bin/ghc-stage2 -O2 testsuite/tests/perf/compiler/T9020.hs
 -fforce-recomp
 }}}

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15019#comment:8>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler