This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: -O4 needed




  In message <199809110841.KAA02439@portofix.ida.liu.se>you write:

  > 100000 elements
  >    cc -fast -xO5:   195 ms
  >    gcc -O3 -fstrength-reduce -fthread-jumps -fcse-follow-jumps
  > -fcse-skip-blocks -frerun-cse-after-loop -frerun-loop-opt -fgcse
  > -fexpensive-optimizations -fregmove -fschedule-insns -fschedule-insns2
  > -ffunction-sections -fcaller-saves -funroll-loops -fmove-all-movables
  > -freduce-all-givs -fstrict-aliasing
  >                 :   202 ms
  > 
  > Hmm... much better. 20% better than just -O3, and almost as fast as
  > Suns compiler.
  > 
  > My question is: why not just put all those extra optimizations flags
  > (or almost all) that not are used in -O3 in a new -O4 flag. I think
  > that 20% i a very significant improvement.
Aside from -fstrict-aliasing the other options you're using that
are not enabled by default are just as likely to be generating worse
code then they are to generate better code.  THus it does not make
much sense to enable them at this time.

-fstrict-aliasing will be enabled by default in future releases once
we have had a chance to test it more.

jeff


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]