home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!pipex!bnr.co.uk!uknet!mcsun!fuug!funic!nntp.hut.fi!vipunen.hut.fi!jmunkki
- From: jmunkki@vipunen.hut.fi (Juri Munkki)
- Newsgroups: comp.sys.mac.programmer
- Subject: Re: Curves in QuickDraw GX
- Message-ID: <1993Jan1.183451.22522@nntp.hut.fi>
- Date: 1 Jan 93 18:34:51 GMT
- References: <1i08akINNj9b@agate.berkeley.edu> <1i1urmINNt8g@mirror.digex.com>
- Sender: usenet@nntp.hut.fi (Usenet pseudouser id)
- Reply-To: jmunkki@vipunen.hut.fi (Juri Munkki)
- Organization: Helsinki University of Technology
- Lines: 35
- Nntp-Posting-Host: vipunen.hut.fi
-
- In article <1i1urmINNt8g@mirror.digex.com> sunilg@access.digex.com (Sunil Gupta) writes:
- >werner@dewey.soe.berkeley.edu (John Werner) writes:
- >: Also, hand-coding something in assembler for a RISC chip is MUCH
- >: harder than on a traditional processor. There's a smaller instruction
- >: set, more registers, and you have to worry about pipelining. A good
- >: optimizing compiler can usually do a better job than most humans can
- >: do (at least in a reasonable amount of time).
- >
- >Your comment is true for RISC processors, but the complaint was being
- >made about the CISC implementation. My small (4 person) programming shop
- >maintains our code for a number of platforms. For CISC implementations
- >we usually tune critical sections in assembly and usually let the
- >compiler worry about that for RISC implementations. If we can do that,
- >I'm sure that a megabucks company can spare a few programmers for a
- >project that would greatly enhance their products.
-
- I agree totally. In addition for QD GX, they used fixed point math. No
- current languages have built-in support for fixed point math, so a high
- level implementation will not benefit fully from the integer speed of
- fixed point.
-
- Floating point is more suitable for high level languages and is probably
- just as fast as fixed point on the PowerPC architecture.
-
- Then again, floating point would have been a very bad idea on the 68K
- machines.
-
- It seems to me that the best solution would have been to write two versions:
- first a floating point/high level language implementation to debug the
- algorithms and then a fixed point/assembly language implementation to
- make it run ok on the current Macs.
-
- --
- Juri Munkki Windsurf: fast sailing
- jmunkki@hut.fi Macintosh: fast software
-