[GHC] #2273: inlining defeats seq

[GHC]

#2273: inlining defeats seq
---------------------------------+------------------------------------------
    Reporter:  igloo             |        Owner:                  
        Type:  merge             |       Status:  new             
    Priority:  normal            |    Milestone:                  
   Component:  Compiler          |      Version:  6.9             
    Severity:  normal            |   Resolution:                  
    Keywords:                    |   Difficulty:  Unknown         
    Testcase:                    |           Os:  Unknown/Multiple
Architecture:  Unknown/Multiple  |  
---------------------------------+------------------------------------------
Comment (by simonpj):

 Note to self.  I think the Right Solution is simply not to inline things
 that are "cheap", but only things that are "values".  This is easy to
 achieve:
 {{{
 hunk ./compiler/coreSyn/CoreUnfold.lhs 637
 -       yes_or_no = active_inline && is_cheap && consider_safe
 +       yes_or_no = active_inline && is_value && consider_safe
 }}}
 But it's not quite so easy: this change makes allocation go up in a couple
 of programs, and runtime goes up quite a bit.  (We'd need to double-check
 that the runtime figures are right.)
 {{{
 --------------------------------------------------------------------------------
         Program           Size    Allocs   Runtime   Elapsed
 --------------------------------------------------------------------------------
          fulsom          -2.8%    +17.2%    +39.1%    +39.8%
          puzzle          +1.0%     +8.1%    +15.2%    +18.6%
            atom          +0.7%     +7.3%    +55.7%    +64.7%
         circsim          +0.5%     -0.0%    +35.2%    +37.1%
       compress2          +0.7%     -0.0%    +25.8%    +29.9%
            lcss          +0.7%     -0.0%    +42.8%    +48.1%
 --------------------------------------------------------------------------------
             Min          -2.8%     -4.3%     -4.8%    -10.0%
             Max          +1.0%    +17.2%    +55.7%    +64.7%
  Geometric Mean          +0.5%     +0.5%    +18.2%    +21.4%
 }}}
 So, annoyingly, more investigation required. I can't do it today, so I'm
 recording the breadcrumbs here for when I get back to it.

 Simon

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/2273#comment:10>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler