Path: utzoo!utgpu!watmath!clyde!att!rutgers!mailrus!iuvax!pur-ee!hankd From: hankd@pur-ee.UUCP (Hank Dietz) Newsgroups: comp.arch Subject: Memory-Mapped vs. Memory-Controlled (was Re: ZISC computers) Summary: Dumb trivia and the NorthStar FPB Keywords: ZISC Message-ID: <9924@pur-ee.UUCP> Date: 29 Nov 88 21:32:02 GMT References: <22115@sgi.SGI.COM> <278@antares.UUCP> <2958@ima.ima.isc.com> Organization: Purdue University Engineering Computer Network Lines: 17 You don't even have to be memory-mapped to get operands to a functional unit which is out there somewhere. The old NorthStar Horizon FPB -- a BCD floating-point S100 board made of TTL (circa 1974) -- wasn't memory-mapped, but was controlled by a memory reference. First, you'd touch a location, then the FPB would actually watch bus references and would load things into the FPB as it saw them being loaded into the CPU... yes, I know that sounds like the bus loading is wrong, but it did work.... Supposedly, the idea was to save a few T-states over the usual memory-mapped technique: 8080-family CPUs were not good at multi-byte memory-to-memory moves, but could do a sequence of memory-to-register loads fairly fast. Results were read from the FPB in a similar way. Actually, I recall quite a few things being memory-controlled in the above sense, but not memory-mapped. I just thought it might be worth mentioning this alternative, however "kludgey" it might seem.... -hankd@ee.ecn.purdue.edu