Date: Wed, 21 Mar 2001 05:28:52 -0800 From: Brian Raiter <breadbox@muppetlabs.com> To: David Malone <dwmalone@maths.tcd.ie> Cc: Matthew Emmerton <matt@gsicomp.on.ca.emmerton.org>, Aaron Smith <aaron@mutex.org>, freebsd-hackers@FreeBSD.ORG, jon@csua.berkeley.edu Subject: Re: gzip's custom i386 asm should be disabled Message-ID: <200103211328.FAA07031@eidolon.muppetlabs.com> In-Reply-To: <20010321114832.A31809@walton.maths.tcd.ie> References: <20010320174630.B82004@gelatinous.com> <017d01c0b1ab$df4be1b0$1200a8c0@gsicomp.on.ca> <20010321114832.A31809@walton.maths.tcd.ie>
next in thread | previous in thread | raw e-mail | index | archive | help
>>> Since I would imagine a large percentage of FreeBSD users run on >>> i686 cores, it'd be great to get this pretty significant speed >>> increase into our tree. >> >> I sure hope I'm not the only one with a "lab" of 4 FreeBSD machines >> that are all 486s or 586s. > > You may find that the 686 assembly is as fast on a 386/486/586 as > the old assembly is. Maybe you could test it and let the list know? Possibly, but my guess would be no. My 586 patch will probably outdo my 686 patch when run on a 486. The latter uses "movzx" in place of a simple "and" in order to avoid the partial register stall, which will be slower on the 486. In fact, the C compiler, if optimizing specifically for the 486, might be better than either of my patches. The gzip encoder is much more memory-bound than CPU-bound, so it's really hard to squeeze any speed out of the algorithm. Code that's targeted for the wrong Intel chip will probably lose out. (Back in 1998 when I created those patches, I wanted to make a third one specifically for the 486, but by then I didn't have access to a working 486 machine. I had some ideas written down on paper, but I didn't feel comfortable proceeding without any way to test them out.) b To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200103211328.FAA07031>