ponderous 'make world' times post GCC 3.3...

[Thomas Moestl]

On Wed, 2003/08/13 at 18:12:09 +0200, Harti Brandt wrote:
> On Wed, 13 Aug 2003, Steve Ames wrote:
> 
> SA>
> SA>----- Original Message -----
> SA>From: "Harti Brandt" <brandt at fokus.fraunhofer.de>
> SA>> I have actually two machines and two kernels: one machine runs with
> SA>> WITNESS and DIAGNOSTICS and the other without that. On both machines
> SA>> the current kernel is three times slower than the old kernel. Of course
> SA>> the debugging kernels are even slower than the non-debugging ones. But,
> SA>> the three times ratio between old and new kernels is the same on both
> SA>> machines.
> SA>
> SA>Ah. Well that blows my theory :) However... in January my system took around
> SA>4 hours to do 'make world' and it still does with today's code (if I don't
> SA>have
> SA>debug options turned on). So I'm not sure there is a global 3x increase.
> 
> Well, you're the 2nd one who reports that nothing has changed. Several
> people however report, that they see the 3x increase. And I see it myself,
> no matter how often I look at this. Today I tried to build the vinum
> module:
> 
> gcc-3.3.1	yesterday's kernel	132 + 85 + 20
> gcc-3.2.2	yesterday's kernel	131 + 85 + 22
> gcc-3.2.2	kernel from June 1st	 52 + 27 +  7
> 
> Exactly the same config and the same world (except for gcc).
> 
> This is an Ultra10. Perhaps it depends on the sparc model?

FWIW, I'm not seeing this either on my Blade 100; with kernels from
Jun 1 and Aug 11, both built with gcc 3.3.1, a complete kernel build
takes the same time down to a few seconds. I'm running world builds
right now.

It would be very interesting if you could narrow the change down to a
shorter time span by binary search.

	- Thomas

-- 
Thomas Moestl <t.moestl at tu-bs.de>	http://www.tu-bs.de/~y0015675/
              <tmm at FreeBSD.org>		http://people.FreeBSD.org/~tmm/
PGP fingerprint: 1C97 A604 2BD0 E492 51D0  9C0F 1FE6 4F1D 419C 776C