Date: Mon, 15 Jan 2007 12:53:58 -0800 From: Chuck Swiger <cswiger@mac.com> To: daniel.tourde@spray.se Cc: freebsd-questions@freebsd.org Subject: Re: FreeBSD 6.2, rebuild. NO-MMX, NO-SSE? Message-ID: <05A46B30-963A-4C95-B040-E1E2AB4D49A0@mac.com> In-Reply-To: <200701152052.05269.daniel.tourde@spray.se> References: <200701152052.05269.daniel.tourde@spray.se>
next in thread | previous in thread | raw e-mail | index | archive | help
On Jan 15, 2007, at 11:52 AM, Daniel Tourde wrote: > The machine (Inspiron 9400) is fast but I saw at certain moments > something > like "NO-MMX, NO-SSE" (some flags or variables) during the compilation > process. I thought then "How come? What a pity not to use these > instructions". > > Can someone tell me what it was and if it is really supposed to be > like this? You'll find the compiler being passed "-mno-mmx -mno-3dnow -mno-sse - mno-sse2" throughout the compilation of the kernel and bootstrap loader, and yes, it is really supposed to be like that. > My roots are in Gentoo Linux where it is possible to get the > maximum out of a > processor when building a system from scratch by using properly > certain C and > C++ flags. Yes, people writing Linux have spent all kinds of effort writing various fancy block copies which use MMX or SSE or whatnot, and having the system perform benchmarks at boot to pick the fastest, but they are tuning this for microbenchmarks without really considering the effects of prefetching lots of data when it may already be cached or by potentially flushing away other useful data in the caches (especially if you're doing DMA and want the disk drive or NIC to end up with the data and not your CPU's caches). There's tons of information and benchmarks here in this thread: http://lists.freebsd.org/pipermail/freebsd-performance/2006- December/002375.html -- -Chuck
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?05A46B30-963A-4C95-B040-E1E2AB4D49A0>