Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!tut.cis.ohio-state.edu!ucbvax!decwrl!crltrx!max!jg
From: jg@max.crl.dec.com (Jim Gettys)
Newsgroups: comp.windows.x
Subject: Re: Purdue Speedups
Message-ID: <4548@crltrx.crl.dec.com>
Date: 13 Feb 90 15:28:58 GMT
References: <9002122217.AA27286@xenon.lcs.mit.edu> <222@hal.CSS.GOV>
Sender: news@crltrx.crl.dec.com
Reply-To: jg@max.crl.dec.com (Jim Gettys)
Organization: DEC Cambridge Research Lab
Lines: 19

Keith is being too modest.  He did a super job on the CFB code.  With
the exception of a some
things he's done since R4 came out, most things are about as good as you
can get using
portable C code.  You may be able to tune things a bit better for a
given machine/frame
buffer combination, but that is about it.

It has been fun to watch Keith and Joel McCormack (implementor of the
DECstation cfb server,
who has been known to write MIPS assembly code occasionally, when the optimizer
won't do cross basic block optimizations!) compete with each other on
who has the 
fastest implementation of a given function.  By the consortium's
standards, Joel is
cheating when he wins that way, but by product metrics, Joel isn't
cheating, just
adding value (though Joel hasn't written much assembly code).
				- Jim