[GHC] #11564: Possible overzealous unfolding

[GHC]

#11564: Possible overzealous unfolding
-------------------------------------+-------------------------------------
        Reporter:  simonmar          |                Owner:
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:
       Component:  Compiler          |              Version:  8.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):  Phab:D1900
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by simonmar):

 I tried this again, the results are a bit different:

 {{{
 bytes allocated value is too high:
     Expected    T5631(normal) bytes allocated: 1124068664 +/-5%
     Lower bound T5631(normal) bytes allocated: 1067865230
     Upper bound T5631(normal) bytes allocated: 1180272098
     Actual      T5631(normal) bytes allocated: 1384316488
     Deviation   T5631(normal) bytes allocated:       23.2 %

 bytes allocated value is too high:
     Expected    T9020(optasm) bytes allocated: 698401736 +/-10%
     Lower bound T9020(optasm) bytes allocated: 628561562
     Upper bound T9020(optasm) bytes allocated: 768241910
     Actual      T9020(optasm) bytes allocated: 843289480
     Deviation   T9020(optasm) bytes allocated:      20.7 %

 update the test so that GHC doesn't regress again)
     Expected    T7257(normal) bytes allocated: 1654893248 +/-5%
     Lower bound T7257(normal) bytes allocated: 1572148585
     Upper bound T7257(normal) bytes allocated: 1737637911
     Actual      T7257(normal) bytes allocated: 1414893248
     Deviation   T7257(normal) bytes allocated:      -14.5 %
 }}}

 On nofib:

 {{{
 --------------------------------------------------------------------------------
         Program           Size    Allocs   Runtime   Elapsed  TotalMem
 --------------------------------------------------------------------------------
    k-nucleotide          +0.5%   +106.7%     +4.9%     +4.9%      0.0%
 --------------------------------------------------------------------------------
             Min          +0.4%     -0.3%     -8.2%     -8.3%     -1.8%
             Max          +0.7%   +106.7%     +7.9%     +7.8%     +2.6%
  Geometric Mean          +0.6%     +0.7%     +0.0%     +0.0%     -0.0%

 Compile Allocations
         -1 s.d.                -----            -0.1%
         +1 s.d.                -----            +1.4%
         Average                -----            +0.7%

 Compile Times
         -1 s.d.                -----           -10.6%
         +1 s.d.                -----            +8.7%
         Average                -----            -1.4%
 }}}

 To investigate:
 * k-nucleotide
 * compiler perf on T5631 and T9020

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11564#comment:9>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler