Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!cs.utexas.edu!uwm.edu!genbank!ames!fxgrp!news
From: mikew@fxgrp.fx.com (Mike Wexler)
Newsgroups: comp.arch
Subject: Re: Uses for unusual instructions
Message-ID: <1989Nov27.173555.19673@fxgrp.fx.com>
Date: 27 Nov 89 17:35:55 GMT
References: <1743@l.cc.purdue.edu>
Sender: news@fxgrp.fx.com (News)
Reply-To: mikew@fxgrp.fx.com (Mike Wexler)
Organization: FXD/Telerate, Mountain View, CA
Lines: 29

>cik@l.cc.purdue.edu (Herman Rubin) writes:
>Soem time ago, I posted to this group certain instructions which are cheap
>in hardware and expensive in software.  I wish to discuss these and give
>partial examples; the full examples are too lengthy and would provide little
>insight.
It sounds like some special purpose hardware might help solve your problems
                                                             ^^^^
more quickly. I propose an experiment: 
1. Design a CPU with these instructions added and one without(you might be
   able to start with some kind of publicly available designs from a 
   university project).
2. Modify your codes to take advantage of these instructions. Run your code
   through a good simulator and see how much perfomance you gain.
3. Run some benchmarks or even some real applications that are used by a
   large number of people through both simulations.
4. Tweak the compilers to take advantage of the new instructions.
5. Do step 3 over.

If step 2 show a big performance win, go to a ASIC house and pay them to
build a chip for you.

If steps 3 or 5 show a big performance win, provide the data to people who
make micro-processors and you will be able to convince them to add some new
instructions.

BTW, I understand that the above process is time consuming and expensive,
but you seem to have the most to gain from it also. 

Mike Wexler (mikew@fx.com)


Brought to you by Super Global Mega Corp .com