irf@kuling.UUCP (Bo Thide) (11/06/88)
I just came across a report written June this year by the Gartner Group. It says: "We believe that future computer architectures will lie somewhere in between current RISC and CISC implementations... To design an architecture, a vendor must analyze years of instruction traces for existing applications to determine the impact of each instruction in the architecture on real applications. Then for each instruction, the vendor must weigh the costs of including that instruction in the architecture or omitting it. The actual financial cost of adding the instruction, as well as the performance cost, must be considered. Hewlett-Packard used this method in designing its Precision Architecture, and it may account for why HP-PA has demonstrated the best architecture that we have seen. It may also be why HP-PA has 125 instructions instead of only 50. HP-PA may be the first reduced complexity computer." Who else has done the same thing and come up with an RCC (Reduced Complexity Computer)? What if "existing applications" change - would that mean that future RCCs/RISCs might have a different instruction set than HP-PA? ^ Bo Thide'-------------------------------------------------------------- | | Swedish Institute of Space Physics, S-755 91 Uppsala, Sweden |I| [In Swedish: Institutet f|r RymdFysik, Uppsalaavdelningen (IRFU)] |R| Phone: (+46) 18-403000. Telex: 76036 (IRFUPP S). Fax: (+46) 18-403100 /|F|\ INTERNET: bt@irfu.se UUCP: ...!enea!kuling!irfu!bt IP: 192.36.174.1 ~~U~~ -----------------------------------------------------------------sm5dfw
albaugh@dms.UUCP (Mike Albaugh) (11/09/88)
From article <891@kuling.UUCP>, by irf@kuling.UUCP (Bo Thide): > > I just came across a report written June this year by the Gartner Group. > It says: > > "We believe that future computer architectures will lie somewhere in between > current RISC and CISC implementations... To design an architecture, a vendor > must analyze years of instruction traces for existing applications... I would like to inject the note that the _choice_ of instruction traces is very important. As an example, consider the Motorola 6809. It contains quite a few uh, interesting, instructions that seem to be sort of "macros" for some quite common code sequences on the 6800. The funny thing is, those code sequences were commonly used by 6800 programmers to get around holes in the 6800 instruction set _that_don't_exist_ in the 6809. Also, apparently Motorola never used register-register moves in their own code (or wherever they got their traces), while everyone else I know did, and was rudely surprised at the _very_slow_ performance of formerly optimal code. The latter "feature" may also have been the result of using static, rather than dynamic, instruction counts, but Moto salesman claimed not. | Mike Albaugh (albaugh@dms.UUCP || {...decwrl!turtlevax!}weitek!dms!albaugh) | Atari Games Corp (Arcade Games, no relation to the makers of the ST) | 675 Sycamore Dr. Milpitas, CA 95035 voice: (408)434-1709 | The opinions expressed are my own (Boy, are they ever)