[GHC] #1261: optLevelFlags is not monotonic

[GHC]

#1261: optLevelFlags is not monotonic
----------------------------------------------+-----------------------------
 Reporter:  Stefan O'Rear <stefanor at cox.net>  |          Owner:         
     Type:  bug                               |         Status:  new    
 Priority:  normal                            |      Milestone:         
Component:  Compiler                          |        Version:  6.7    
 Severity:  normal                            |     Resolution:         
 Keywords:                                    |     Difficulty:  Unknown
 Testcase:                                    |   Architecture:  x86    
       Os:  Linux                             |  
----------------------------------------------+-----------------------------
Changes (by Stefan O'Rear <stefanor at cox.net>):

  * summary:  15x performance regression between HEAD and 6.6 =>
              optLevelFlags is not monotonic

Comment:

 By suggestion of Don Stewart, I tried again with -O2 (vs. -O3), and the
 regression vanished.  Nor is it a backfiring optimization, for -ddump-
 simpl-stats (his idea again) shows -O3 does less in every way.  It appears
 from cursory examination of the driver that the option flags are decoded
 in such a way that they are not monotonic, which is confusingly counter to
 gcc etc tradition.

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/1261>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
-------------- next part --------------
_______________________________________________
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs at haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs