[GHC] #10678: integer-gmp's runS seems unnecessarily expensive

[GHC]

#10678: integer-gmp's runS seems unnecessarily expensive
-------------------------------------+-------------------------------------
        Reporter:  rwbarton          |                   Owner:
            Type:  bug               |                  Status:  new
        Priority:  normal            |               Milestone:
       Component:  Compiler          |                 Version:  7.10.1
  (CodeGen)                          |
      Resolution:                    |                Keywords:
Operating System:  Unknown/Multiple  |            Architecture:
 Type of failure:  Runtime           |  Unknown/Multiple
  performance bug                    |               Test Case:
      Blocked By:                    |                Blocking:
 Related Tickets:                    |  Differential Revisions:
-------------------------------------+-------------------------------------
Changes (by hvr):

 * cc: bgamari (added)


Comment:

 Interesting, there's some (albeit not perfect) overlap between the
 improved nofib cases with the regressed nofib cases noted in
 CompilerPerformance#Comparinginteger-gmp0.5and1.0

 E.g.

 {{{
      bernouilli          -0.3%     -8.2%     -2.4%     -2.4%      0.0%
           fasta          -0.3%     -5.1%     -0.1%     -0.1%      0.0%
           kahan          -0.2%    -11.4%     +0.3%     +0.3%      0.0%
       primetest          -0.3%    -18.5%     -3.5%     -3.6%      0.0%
             rsa          -0.3%    -19.8%     0.060     0.060      0.0%
             scs          -0.4%     -3.2%     -2.0%     -2.0%    +14.3%
          symalg          -0.3%     -2.1%     0.024     0.024      0.0%
       wave4main          -0.2%     -4.1%     -2.1%     -2.1%     -5.3%
 }}}

 vs
 ([https://gist.githubusercontent.com/bgamari/5de75ac998a346b70ce8/raw/ffc2bd4f7df2fe036e2b33869011b3b2b19e666f
 /nofib-integer-gmp-comparison.txt source])

 {{{
         Program           Size    Allocs   Runtime   Elapsed  TotalMem
      bernouilli         -68.7%    +15.3%     0.092     0.092      0.0%
           fasta         -76.4%     -0.1%     -0.6%     -0.7%      0.0%
           kahan         -74.0%    +39.9%     0.131     0.131      0.0%
       primetest         -67.5%    +49.9%     0.050     0.050      0.0%
             rsa         -67.9%    +53.4%     0.012     0.012      0.0%
             scs         -60.2%     +6.8%     +1.2%     +1.2%      0.0%
          symalg         -63.1%     +9.5%     0.005     0.005      0.0%
       wave4main         -69.5%     +0.0%     0.109     0.109      0.0%
 }}}

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10678#comment:8>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler