Re: Re: [f-cpu] F-CPU invited at the Libre Software Meeting

[whygee@club-internet.fr]

hi !

>De: Juergen Goeritz
>.... and that leads me to another idea.
>
><imagination on>
>
>Has anybody ever thought about data triggered software
>concepts?
>
>If a single processor today has to do a thing twice, it
>must run the programm sequentially two times, one after
>another with different data. In this case instruction
>caches will only bring additional power (beside burst
>loading) if the code completely fits into the cache.
>Otherwise it may be faster without cache, since reload
>always adds penalty.
>
>The processors usually run a program that has to fetch
>data, modify it and store it somewhere. This is a code
>(program) focused solution.
>
>The other way I can imagine is a data focused solution.
>The availability of data triggers the execution of a
>code part for that data to modify it. That could lead
>to a completely event triggered software scheme, i.e.
>data triggers software, which generates data, which
>triggers software a.s.o.
>
>Since this is a very basic mechanism it could be run
>on a single processor and on multiple processors. Or
>on the nearest processor available...
>It's not the same as task switching because it takes
>place on a much lower level. And it may obsolete task
>switching concepts like they exist today.
>
><imagination off>
>
>comments?

one usually calls this : a dataflow machine.

on a lower level, it's simply a TTA execution core
(like the previous F-CPU core idea that never
succeeded).

Look also at the CRAY MTA.

>JG
YG

*************************************************************
To unsubscribe, send an e-mail to majordomo@seul.org with
unsubscribe f-cpu       in the body. http://f-cpu.seul.org/