Date: Fri, 10 Mar 2000 20:42:31 +0000 (GMT) From: Steve Roome <steve@snuggly.demon.co.uk> To: multimedia@freebsd.org Cc: ports@freebsd.org Subject: gogo + 3dnow Message-ID: <Pine.BSF.4.10.10003102030490.12062-100000@snuggly.demon.co.uk>
next in thread | raw e-mail | index | archive | help
(Firstly, apologies for crossposting, I wasn't sure which group to send this to.) Anyone else used gogo much from the ports collection ? I've just been having a play with it, and it's looking to me like 3dnow, and even the extra new 3dnow stuff in the athlon could come in quite handy for this sort of stuff... On an Athlon 500, (3.4-STABLE cvsupped about last weekend) I get the following results : gogo -test -nopsy -off 3dn -off e3dn -off mmx : 16.61x [I think that's pretty much how it would get compile under FreeBSD without nasm, i.e. with the default tools] gogo -test -nopsy -off 3dn -off e3dn : 17.74x [So with mmx it's not much faster] gogo -test -nopsy -off 3dn : 17.58 [no idea why this is slower, perhaps using e3dn but not 3dn doesn't work properly in the port ?] gogo -test -nopsy -off e3dn : 28.81 [so using 3dnow instructions almost doubles the speed ?] gogo -test -nopsy : 30.52 [and another 6% improvement with the e3dn instructions - granted, I had to patch nasm for these..] Anyway, in comparison to say, gcc-2.7.2.3 and the default assembler this sort of program goes about twice as fast... Is that a good reason to have 3dnow support in the default compiler/assembler, or would it be worth adding these e3dn patches only to the nasm port ? I'm probably missing something really obvious and talking a load of rubbish.. (what a change there) but at first glance this is an astounding difference in performance. Steve To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-multimedia" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.10.10003102030490.12062-100000>