home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!decwrl!sun-barr!news2me.ebay.sun.com!exodus.Eng.Sun.COM!nonsequitur!narad
- From: narad@nonsequitur.Eng.Sun.COM (Chuck Narad)
- Newsgroups: comp.arch
- Subject: Re: CISC Microcode (was Re: RISC Mainframe)
- Message-ID: <l7961rINNnni@exodus.Eng.Sun.COM>
- Date: 28 Jul 92 00:43:39 GMT
- References: <DOCONNOR.92Jul23141720@potato.sedona.intel.com>
- Reply-To: narad@nonsequitur.Eng.Sun.COM
- Organization: Sun Microsystems
- Lines: 30
- NNTP-Posting-Host: nonsequitur
-
- davsmith@nic.cerf.net (David Smith) writes:
- > All CPUs I have seen to date (not every CPU by any means - if you know
- > of counter examples, please post) cannot do asynchronous address
- > generation. When they request a word of memory they want it *NOW* or
- > within a cycle or two and will block until it arrives.
- >
-
- In a previous life I worked on a machine called the Culler-7 at (now
- defunct) Culler Scientific Systems in Santa Barbara. It was basically
- an asymmetric machine with a RISC-like engine that did address
- computation, scaler computation, program flow control, and data
- scheduling (the A machine); and a CISC, microcoded FP engine with a
- different register set, PF ALU and MPY units, a vector memory with
- address stride engines, and a few other whizzy and interesting features
- (the XY machine). These days that structure is called VLIW; back then
- we just called it wonderful :-)
-
- The A and XY machines shared an instruction stream, but A was
- single-cycle and XY was multi-cycle at the mercy of the u-Code. A
- scheduled data reads and writes to the main memory system, and both A
- and XY could either consume read data or source write data. The data
- read FIFO was 3 deep (and the memory pipeline was 3 clocks plus
- potential bus arbitration) so it was possible to schedule data way in
- advance of the time it was used, if desired. The compiler was able to
- take advantage of this both for streaming vector data and for
- preplanning much on the non-vectorized data accesses.
-
- chuck/
-
- disclaimer: speaking for myself, not for my emplyoer
-