Re: Optimization: Conclusions from Evolutionary Analysis

[Jan Hubicka <jh at suse dot cz>]

> Jan Hubicka wrote:
> >I would like to see -ftracer enabled for -O3, in fact I am quite
> >surprised it is not.  We do that for a while in SuSE compilers and it
> >works fine.
> >
> >It is not good -O2 candidate as it increase code size too much.  It may
> >be, however, nice to enable it on -O2 only when -fbranch-probabilities
> >is present.
> 
> Everything I've seen suggests that -ftracer would be an effective 
> addition to -O3. -freduce-all-givs seems valuable in specific situation 
> on the P3, but I'm not certain it is generally applicable.
> 
> I'm trying to figure out the best hueristics for weighing both code size 
> and speed.
> 
> >This is quite surprising, as I can measure pretty clean benefits on SPEC
> >benchmark running Opteron for instance.  You need to use -mfpmath=sse
> >-msse2 (or -march=sse2_enabled_CPU) to get double precision arithmetics
> >in SSE.  I also reproduced similar results for Pentium4 in the past.
> >Only CPU that does not seem to preffer SSE operations appears to be
> >PentiumM in my notebook (the hardware implementation is probably quite
> >poor)
> 
> Being stuck with a Pentium 4 (Northwood), I haven't been able to see how 
> the SSE options affect code on the Opteron. -msse is implied by 
> -march=pentium4; I've checked this, both in the GCC source code and by 
> comparing compiles with and without -msse.
Interesting, it looks like we do something terribly wrong for P4.
P4 tunning needs work :( Perhaps I can look into it during stage3.
> 
> >How does this compare to -fomit-frame-pointer?
> 
> As others have pointed out, -fomit-frame-pointer is not enabled for the 
> P4, so no comparison can be made with -momit-leaf-frame-pointer.
> 
> >The drawback is again the code size, so perhaps it can be -O3 only.
> 
> Or, perhaps, we could consider a -O4, meaning "I don't give a darn about 
> code size, just optimize like heck."

This is -O3 already (inlining is the case).

Honza
> 
> -- 
> Scott Robert Ladd
> Coyote Gulch Productions (http://www.coyotegulch.com)
> Software Invention for High-Performance Computing
> In development: Alex, a database for common folk