Newsgroups: comp.arch
Path: utzoo!henry
From: henry@utzoo.uucp (Henry Spencer)
Subject: Re: hmmm, BitBlt, New Instructions For RISC
Message-ID: <1990Feb23.181713.14334@utzoo.uucp>
Organization: U of Toronto Zoology
References: <7398@pdn.paradyne.com> <1990Feb18.084034.7015@utzoo.uucp> <7404@pdn.paradyne.com> <1990Feb19.174926.7899@utzoo.uucp> <7452@pdn.paradyne.com>
Date: Fri, 23 Feb 90 18:17:13 GMT

In article <7452@pdn.paradyne.com> alan@oz.paradyne.com (Alan Lovejoy) writes:
>>>... It only takes roughly 50 cycles (on an 88k) to setup for a blit...
>>I'd be interested to know exactly what you include under "setup".  Pike's
>>numbers showed a much greater disparity the other way, counting *all*
>>the overhead.
>
>Consulting my source code reveals the following:
>
>Given two pointers to two bitmaps (source and destination) which are structs
>describing...

Where do you get those two pointers?  For character blitting you'll need
font lookup and possibly some other bits of overhead, depending on exactly
how the stuff is being invoked by higher levels.  The relevant measurement
of overhead for character blitting is "end to end", including absolutely
everything.  However, done well, that's probably not going to add a whole
lot more.  That *is* an impressively fast setup you've got there.
-- 
"The N in NFS stands for Not, |     Henry Spencer at U of Toronto Zoology
or Need, or perhaps Nightmare"| uunet!attcan!utzoo!henry henry@zoo.toronto.edu