Re: I486_CPU or I586_CPU in kernel config
On May 30, 2011, at 4:53 PM, Warren Block wrote: On Mon, 30 May 2011, Adam Vande More wrote: Perhaps this is the one you meant? http://lists.freebsd.org/pipermail/freebsd-questions/2009-January/190568.html That's the one! Thanks! Actually the two threads touch on the same subject, and it seems removal of those options is still desirable on newer CPU's. sys/i386/i386/support.s is mentioned, but doesn't seem to have anything explicitly specific for 586. There are some i686 entries. It depends on which version of FreeBSD we're talking about. At the time of my post, FreeBSD 7.0 had just come out and FreeBSD 6.3/6.4 were widely used. If you check CVSweb or SVN: http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/i386/i386/support.s svn://svn.freebsd.org/base/stable/7/sys/i386/i386/support.s ...you'll see a half-dozen entries which are conditionalized against I586_CPU: % grep -A 1 I586_CPU support.s #if defined(I586_CPU) defined(DEV_NPX) kernel_fpu_lock: -- #if defined(I586_CPU) defined(DEV_NPX) ENTRY(i586_bzero) -- #endif /* I586_CPU defined(DEV_NPX) */ -- #if defined(I586_CPU) defined(DEV_NPX) ENTRY(i586_bcopy) -- #endif /* I586_CPU defined(DEV_NPX) */ -- #if defined(I586_CPU) defined(DEV_NPX) ALIGN_TEXT -- #if defined(I586_CPU) defined(DEV_NPX) ENTRY(i586_copyout) -- #endif /* I586_CPU defined(DEV_NPX) */ -- #if defined(I586_CPU) defined(DEV_NPX) ALIGN_TEXT -- #if defined(I586_CPU) defined(DEV_NPX) ALIGN_TEXT -- #if defined(I586_CPU) defined(DEV_NPX) ENTRY(i586_copyin) -- #endif /* I586_CPU defined(DEV_NPX) */ #if defined(I586_CPU) defined(DEV_NPX) /* fastmove(src, dst, len) -- #endif /* I586_CPU defined(DEV_NPX) */ It appears that this file was significantly re-written under FreeBSD-8. Anyway, actually comparing and gathering some data is the most useful result, and I'm happy to see the numbers from Warren's test. Regards, -- -Chuck ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: I486_CPU or I586_CPU in kernel config
On Mon, 30 May 2011, Warren Block wrote: On Mon, 30 May 2011, Adam Vande More wrote: Perhaps this is the one you meant? http://lists.freebsd.org/pipermail/freebsd-questions/2009-January/190568.html That's the one! Thanks! Actually the two threads touch on the same subject, and it seems removal of those options is still desirable on newer CPU's. sys/i386/i386/support.s is mentioned, but doesn't seem to have anything explicitly specific for 586. There are some i686 entries. A test for cpu_class==CPUCLASS_586 in /sys/i386/isa/npx.c is mentioned in the thread, but that check isn't in the current code. A little empirical testing: Times for buildworld after a fresh reboot, /usr/obj/usr deleted, GENERIC included, running ccache: default (486/586 included) 9:05.84 nocpu I486, nocpu I586_CPU 9.27.88 nocpu I486_CPU 8.53.86 So maybe a 6% increase by removing 486 but leaving 586... These were not rigorous benchmarks, it might just be measurement noise. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: I486_CPU or I586_CPU in kernel config
Warren Block wrote: On Mon, 30 May 2011, Adam Vande More wrote: Perhaps this is the one you meant? http://lists.freebsd.org/pipermail/freebsd-questions/2009-January/190568.html That's the one! Thanks! Actually the two threads touch on the same subject, and it seems removal of those options is still desirable on newer CPU's. sys/i386/i386/support.s is mentioned, but doesn't seem to have anything explicitly specific for 586. There are some i686 entries. A test for cpu_class==CPUCLASS_586 in /sys/i386/isa/npx.c is mentioned in the thread, but that check isn't in the current code. Removed: http://svnweb.FreeBSD.org/base?view=revisionrevision=209460 There was a discussion about the implications for performance, although I don't remember when it took place, and whether it was on the mailing lists or in a PR audit trail. If you're interested you can track down the details. b. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
I486_CPU or I586_CPU in kernel config
Some time back, there was a post on one of the mailing lists that suggested it was better to leave either I486_CPU or I586_CPU enabled in a kernel config even for much newer processors. For performance reasons, AFAIR. Naturally I didn't save that post or a link to it. Can anyone find that message, or explain why it would be good to keep either of those cpu options in a kernel that will only run on much newer CPUs? ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: I486_CPU or I586_CPU in kernel config
On Mon, May 30, 2011 at 3:35 PM, Adam Vande More amvandem...@gmail.comwrote: On Mon, May 30, 2011 at 9:50 AM, Warren Block wbl...@wonkity.com wrote: Some time back, there was a post on one of the mailing lists that suggested it was better to leave either I486_CPU or I586_CPU enabled in a kernel config even for much newer processors. For performance reasons, AFAIR. Naturally I didn't save that post or a link to it. Can anyone find that message, or explain why it would be good to keep either of those cpu options in a kernel that will only run on much newer CPUs? Um, I don't recall seeing that and have removed them automatically for a long time. Here is one that suggests keeping I586_CPU with results that seem less than conclusive. http://lists.freebsd.org/pipermail/freebsd-stable/2005-December/020702.html Perhaps this is the one you meant? http://lists.freebsd.org/pipermail/freebsd-questions/2009-January/190568.html Actually the two threads touch on the same subject, and it seems removal of those options is still desirable on newer CPU's. -- Adam Vande More ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: I486_CPU or I586_CPU in kernel config
On Mon, May 30, 2011 at 9:50 AM, Warren Block wbl...@wonkity.com wrote: Some time back, there was a post on one of the mailing lists that suggested it was better to leave either I486_CPU or I586_CPU enabled in a kernel config even for much newer processors. For performance reasons, AFAIR. Naturally I didn't save that post or a link to it. Can anyone find that message, or explain why it would be good to keep either of those cpu options in a kernel that will only run on much newer CPUs? Um, I don't recall seeing that and have removed them automatically for a long time. Here is one that suggests keeping I586_CPU with results that seem less than conclusive. http://lists.freebsd.org/pipermail/freebsd-stable/2005-December/020702.html -- Adam Vande More ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: I486_CPU or I586_CPU in kernel config
On Mon, 30 May 2011, Adam Vande More wrote: Perhaps this is the one you meant? http://lists.freebsd.org/pipermail/freebsd-questions/2009-January/190568.html That's the one! Thanks! Actually the two threads touch on the same subject, and it seems removal of those options is still desirable on newer CPU's. sys/i386/i386/support.s is mentioned, but doesn't seem to have anything explicitly specific for 586. There are some i686 entries. A test for cpu_class==CPUCLASS_586 in /sys/i386/isa/npx.c is mentioned in the thread, but that check isn't in the current code. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: I486_CPU or I586_CPU in kernel config
On Mon, May 30, 2011 at 6:53 PM, Warren Block wbl...@wonkity.com wrote: sys/i386/i386/support.s is mentioned, but doesn't seem to have anything explicitly specific for 586. There are some i686 entries. A test for cpu_class==CPUCLASS_586 in /sys/i386/isa/npx.c is mentioned in the thread, but that check isn't in the current code. Yeah, there doesn't seem to be concrete answer that's within my ability to diagnose. If you have some time and the burning desire, you could try some benchmark runs on each kernel to see if 686 is as preformant as 686/586. The first thread I posted seemed indicate a fairly noticeable slowdown when both were enabled, but there were perhaps problems with test methodology and of course that is quite old. Also seems like a good candidate question for hackers@ -- Adam Vande More ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
