Date: Thu, 02 Apr 1998 13:38:17 +1000 From: John Saunders <john.saunders@scitec.com.au> To: "Gregory P. Smith" <greg@nas.nasa.gov>, freebsd-current@FreeBSD.ORG Subject: Re: current-digest V4 #79 Message-ID: <35230829.78DAA808@scitec.com.au> References: <199804020201.SAA03945@ryouko.nas.nasa.gov>
next in thread | previous in thread | raw e-mail | index | archive | help
Gregory P. Smith wrote:
> I think that would violate the Kernel's "never use floating point" rule
> because MMX uses the FP registers. (and on Intel's MMX implementation
> supposedly has a "large" context switch time between MMX and floating
> point mode).
How does the kernel's "never use floating point" rule co-exist with
the bcopy/bzero/copying/copyout speedups that use the FPU? I've noticed
that the GENERIC config disables the bcopy speedup by using flags 0x1
but retains the bzero and copyin/out speedups. I've changed my flags
to 0 to enable all speedups and haven't noticed anything bad.
You are right that the context switch is likely to be a killer,
although it depends on exactly how much the IP checksum can be speeded
up. When the numbers are added up it could be a big win for a server.
I guess the only way to know is to count those clocks.
Cheers.
-- +------------------------------------------------------------+
. | John Saunders mailto:John.Saunders@scitec.com.au (Work) |
,--_|\ | mailto:john@nlc.net.au (Home) |
/ Oz \ | http://www.nlc.net.au/~john/ |
\_,--\_/ | SCITEC LIMITED Phone +61 2 9428 9563 Fax +61 2 9428 9933 |
v | "By the time you make ends meet, they move the ends." |
+------------------------------------------------------------+
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?35230829.78DAA808>