On Sat, Jul 06, 2002 at 12:12:11PM +0900, Bang Jun-Young wrote:
> Arrgh. There was a stupid mistake. I forgot to add -O2 to COPTS in
> /etc/mk.conf. Sorry for the wrong results. 

Yes - I built a kernel without O2, built quickly but had 7Mb of
code instead of 5Mb...

> I did the same tests again, this time with proper COPTS. To eliminate
> cache effects, I rebooted the machines between the tests.
> 
> 		Athlon XP 1700  Pentium 166  Pentium 166  Pentium 166 
> 		    w/256MB	   w/64MB       w/32MB	     w/16MB	
> -O2 (default)	    6:15          1:12:18      1:13:13      6:34:40
> -pipe -O2	    6:04          1:12:55      1:13:32     manually
> 							  aborted at
> 				 	                   10:58:44		
> Improvement	     3%	            -1%          -0%       -100%???
> 
> Now, it is clear that -pipe doesn't boost up kernel compile on a
> slow machine. You can benefit from it with a fast machine with enough
> memory.

But it doesn't go completely 'tits up' until you get down to 16Mb.

On my (now sluggish and old) athlon 700 with 256Mb,
'rm *.o;time make' reported:
		real	user	sys
1st build	792	556	41
2nd build	702	555	39
-pipe 1st	712	625	48
-pipe 2nd	677	687	54

(all running on a console window, fresh from reboot)
I'm not actually sure I how much I trust these figures!  Note that in
the last case the 'user' time exceeds the 'real' time!
(I've might look at what these three values mean!)
> 
> A better method to speed up build is to simply not use any optimization
> options like -O2.

That would give you a kernel that was SO big the system would never
build another one in a finite time!


	David

-- 
David Laight: david@l8s.co.uk