Date: Mon, 09 Nov 1998 19:21:40 +0800 From: Peter Wemm <peter@netplex.com.au> To: Bruce Evans <bde@zeta.org.au> Cc: narvi@haldjas.folklore.ee, current@FreeBSD.ORG, jc@irbs.com, mike@smith.net.au, smp@FreeBSD.ORG Subject: Re: Dog Sloooow SMP Message-ID: <199811091121.TAA29164@spinner.netplex.com.au> In-Reply-To: Your message of "Mon, 09 Nov 1998 21:22:30 %2B1100." <199811091022.VAA22031@godzilla.zeta.org.au>
next in thread | previous in thread | raw e-mail | index | archive | help
Bruce Evans wrote: > >> FreeBSD should use its own bitmap of capabilities and not test the Intel > >> flags except once to translate them. 32 general flags might even be > >> enough. > >> > > > >How about 64 for the odd case that K7 actually materialises as promised > >and people start putting them in dual motherboards? > > That would be almost twice as slow for CC=gcc. CC=egcs handles 64-bit > bit tests better, especially for the low 32 bits. 32 vs. 64 is almost irrelevant.. There's no limit to the number of 32 bit variables that we can use with flags in them, so there's no reason why we'd use a 64 bit variable in the first place. However.. One thing that bugs me is that we presently can optimize out code and tests for a runtime boost when compiled for a specific cpu. eg: if we support 386 cpus, we test for whether we have an invlpg instruction or not - but if we are not compiling with a 386 option then this code and the test for >= 486 goes away. > Bruce Cheers, -Peter To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-smp" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199811091121.TAA29164>