[GHC] #14944: Compile speed regression

[GHC]

#14944: Compile speed regression
-------------------------------------+-------------------------------------
        Reporter:  br1               |                Owner:  dfeuer
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.4.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by simonpj):

 Thanks for this data.  Several interesting things here.

 * There's a big jump in compilation cost between 7.8 and 7.10.  (Like a
 factor of 2 in compiler allocation, according to comment:3.   According to
 comment:6, the result of Specialise is much bigger in 7.10 than in 7.8.
 This extra code appears to be discarded at the end, but it persists
 through the compilation pipeline.   '''Questions''':  what are the extra
 specialisations?  Do we need them?  Why are they eventually discarded?
 Could we discard them earlier?

 * As you say in comment:8 it seems that GHC 8.4 has a new, and huge
 (temporary) blow-up in porgram size.  It'd be really good to work out what
 this is.   (You'll presumably need to cut down the program size to make
 this tractable.)

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14944#comment:9>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler