[GHC] #14737: Improve performance of Simplify.simplCast

[GHC]

#14737: Improve performance of Simplify.simplCast
-------------------------------------+-------------------------------------
        Reporter:  tdammers          |                Owner:  (none)
            Type:  bug               |               Status:  patch
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.2.2
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #11735 #14683     |  Differential Rev(s):  Phab:D4385
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by tdammers):

 > However, that may still leave simplCast at the top... does it? If it was
 taking 78% of 20sec before, that's 16 sec. So reducing that to near-zero
 (which it should be) would take use to 5s, not 12s.

 Correct; `isReflexiveCo` accounted for about half the execution time spent
 in `simplCast`. With this change, `isReflexiveCo` drops to 1.1% of overall
 execution time, and `simplCast` to about 61%. So this is a big
 improvement, but we're not done yet. The remaining main culprit is
 `pushCoTyArg`.

 > What happens if you leave out the call to isReflexiveCo altogether?
 It'll get done in the next round anyway, by optCoercion.

 I'll try that, but considering that `isReflexiveCo` is no longer critical,
 I don't expect it to make much of a difference.

 For reference, here's the current profiler output:

 {{{
         Wed Mar 28 20:44 2018 Time and Allocation Profiling Report
 (Final)

            ghc-stage2 +RTS -p -RTS -B/home/tobias/well-
 typed/devel/ghc/T14737/inplace/lib ./cases/Grammar.hs -o ./a -fforce-
 recomp

         total time  =       12.35 secs   (12354 ticks @ 1000 us, 1
 processor)
         total alloc = 14,410,284,936 bytes  (excludes profiling overheads)

 COST CENTRE     MODULE     SRC
 %time %alloc

 pushCoTyArg     Simplify   compiler/simplCore/Simplify.hs:1229:63-83
 54.6   55.5
 tc_rn_src_decls TcRnDriver
 compiler/typecheck/TcRnDriver.hs:(494,4)-(556,7)     13.1   14.3
 CoreTidy        HscMain    compiler/main/HscMain.hs:1253:27-67
 4.8    3.9
 SimplTopBinds   SimplCore  compiler/simplCore/SimplCore.hs:770:39-74
 4.2    3.2
 coercionKind    Coercion   compiler/types/Coercion.hs:1701:3-7
 2.8    6.0
 simplCast       Simplify
 compiler/simplCore/Simplify.hs:(1213,5)-(1215,37)     2.5    2.5
 zonkTopDecls    TcRnDriver
 compiler/typecheck/TcRnDriver.hs:(445,16)-(446,43)    2.3    2.2
 deSugar         HscMain    compiler/main/HscMain.hs:511:7-44
 1.5    1.3
 isReflexiveCo   Simplify   compiler/simplCore/Simplify.hs:1260:40-55
 1.1    1.0
 Parser          HscMain    compiler/main/HscMain.hs:(316,5)-(384,20)
 1.1    1.6
 }}}

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14737#comment:9>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler