home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!sequent!muncher.sequent.com!dafuller
- From: dafuller@sequent.com (David Fuller)
- Subject: Re: CISC Microcode (was Re: RISC Mainframe)
- Message-ID: <1992Jul23.193448.10482@sequent.com>
- Sender: usenet@sequent.com (usenet )
- Nntp-Posting-Host: sequent.sequent.com
- Organization: Sequent Computer Systems Inc.
- References: <Brsx7o.G69@zoo.toronto.edu> <2369@nic.cerf.net> <2105@devnull.mpd.tandem.com>
- Date: Thu, 23 Jul 92 19:34:48 GMT
- Lines: 27
-
- In article <2105@devnull.mpd.tandem.com> rrt@amadeus.UUCP (Robert Teisberg) writes:
- >In article <2369@nic.cerf.net> 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.
- >
- >The CDC 6600 and its descendants did this (why does it seem that every
- >clever idea in machine architecture was invented many years ago by
- >Seymour Cray :-). To load from memory, you stuffed an address in
- >register A1, A2, A3, A4 or A5 and sometime later the contents of that
- >address appeared in the corresponding X register. To store, you
- >stuffed an address in A6 or A7 and eventually the contents of X6 or X7
- >found its way to memory at that address. As I recall (do any other
- >old CDC hands know better?), the only reason the CPU stalled due to
- >memory access was a scoreboard conflict or a memory bank conflict. If
- >your memory access pattern was either sequential and contiguous or
- >random, bank conflicts were rare, since the 6600 used interleaved
- >memory just as davsmith recommends elsewhere in his article.
-
- What I recall:
-
- The PPUs had first access to memory, then the CPUs. The PPUs could hold the
- CPU out of main memory if they tried hard enough.
- --
- Dave Fuller All opinions expressed are my own and not
- Sequent Computer Systems those of Sequent Computer Systems, Inc.
- dafuller@sequent.com
-