[GHC] #9476: Implement late lambda-lifting

[GHC]

#9476: Implement late lambda-lifting
-------------------------------------+-------------------------------------
        Reporter:  simonpj           |                Owner:  sgraf
            Type:  feature request   |               Status:  closed
        Priority:  normal            |            Milestone:  8.8.1
       Component:  Compiler          |              Version:  7.8.2
      Resolution:  fixed             |             Keywords:  LateLamLift
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Runtime           |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #8763 #13286      |  Differential Rev(s):  Phab:D5224
       Wiki Page:  LateLamLift       |
-------------------------------------+-------------------------------------

Comment (by sgraf):

 I was referring to the `paraffins` measurements in comment:55.

 Deactivating the closure growth checks (thus allowing more bindings to
 lift at the potential cost of more allocations) leads to an overall
 improvement, most significantly visible in `paraffins` with an increase of
 18% in allocations but 11% less executed instructions. The measurements I
 posted are for runtime. There's a reproducible speedup (0.533s vs 0.581s),
 too, which is mostly due to trading time in the GC for time in the
 mutator. That's also apparent by resizing the nursery so that no GC needs
 to happen (the second pair of measurements with `-A600M`).

 I regard this as 'better code', because it's faster in the default case,
 where we don't tune the nursery to fit all allocations, i.e. more mutator
 time is OK as long as total time goes down. In comment:56 I go on to
 wonder why that is the case.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9476#comment:58>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler