Oh, boy!  Who would have thought simple optimizations would improve so much 
speed!  Maybe it's my opinion, but before I've accidently had "-g" option 
which generated kernel size of 14Megs, kdemedia player would be choppy 
everytime I open something while it's playing.  Now I recompiled the kernel 
without "-g" option and kdemedia player runs very smooth, not a glitch I hear 
even when opening other programs while it's playing.

It seems that NetBSD 1.52 kernel gets compiled without error under both "-O2" 
or "-O3" settings.  And, both "-O2" and "-O3" creates kernel of same size.  
But, I have to admit, my kdemedia player works just as fine under "-O2" or 
"-O3".   But, definitely having "-g" makes noticeable performance degrade 
from kdemedia player (hears short interrupts when opening files!) and my 
kernel was huge, 14Meg!"

Thumbs up:
CFLAGS+=-O2 -march=i386 -mcpu=i686
CXXFLAGS=-O2 -march=i386 -mcpu=i686
COPTS+=-O2 -mcpu=i686 -fexpensive-optimizations -ffast-math -msoft-float \
		         -fmemoize-lookups -fthread-jumps -m486 -fomit-frame-pointer \
			 -finline-functions

Thumbs up:
CFLAGS+=-O3 -march=i386 -mcpu=i686
CXXFLAGS=-O3 -march=i386 -mcpu=i686
COPTS+=-O3 -mcpu=i686 -fexpensive-optimizations -ffast-math -msoft-float \
		         -fmemoize-lookups -fthread-jumps -m486 -fomit-frame-pointer \
			 -finline-functions

Thumbs down:
CFLAGS+=-O3 -march=i386 -mcpu=i686
CXXFLAGS=-O3 -march=i386 -mcpu=i686
COPTS+=-g -O3 -mcpu=i686 -fexpensive-optimizations -ffast-math -msoft-float \
		         -fmemoize-lookups -fthread-jumps -m486 -fomit-frame-pointer \
			 -finline-functions

Infact, I've noticed that having "-g" option disables the optimizations as 
Steven Grunza pointed out earlier.  And, there's no noticeable performance 
difference between "-O2" and "-O3".  Both are fast, but I suppose "-O2" would 
be on the safer measure for most purposes as Matthias Buelow pointed out 
previously.


Thanks for all your info.


Yours sincerely,
Sung N. Cho,
Monday, Jan. 28, 2002.

On Monday 28 January 2002 07:21 pm, Matthias Buelow wrote:
> Sung N. Cho writes:
> >Here's my new, modified mk.conf:
> >CFLAGS+=-O2 -march=i386 -mcpu=i686
> >CXXFLAGS=-O2 -march=i386 -mcpu=i686
> >COPTS+=-O2 -mcpu=i686 -fexpensive-optimizations -ffast-math -msoft-float \
> >		         -fmemoize-lookups -fthread-jumps -m486 -fomit-frame-pointer \
> >			 -finline-functions
>
> I personally would leave the optimizations at -O2, which does about
> all of the above except inline-functions (which might be one of the
> things which could cause problems, or bloat the code).  Dunno about
> the soft-float thing or how architecture-specific options weigh in.
> -m486 or possibly -m686 should be relatively safe but don't quote me
> on that.  In any case, trust me on that that you wouldn't notice
> any difference in system performance.  You're free to experiment and
> share your results with the mailing list, however :)
>
>   --mkb