CCX Crossbar Arbiter

Hi all,

In OpenSPARC micro-spec 3.4.3 "PCX Arbiter Control Flow", Sun proposed a round-robin scheme (toggle priority direction every cycle).

[1] One clock is left-to-right (C0-C7), another clock is right-to-left(C7-C0). So C3 and C4 have the lowest priority, is it right?

[2] Is it a method that used by a lot of design or only SPARC?

[3] Where can I get some reference about this Arbiter scheme? Thanks!

[4] How to profile a Crossbar Arbiter performance? By cycle-accurate model?

Any suggestions are welcome!

Davy

Message was edited by:

Shenli

[608 byte] By [Shenli] at [2007-11-26 9:50:55]

«« Any solution for clearenv() Function
»» V210 RAM configuration

# 1

> [1] One clock is left-to-right (C0-C7), another clock

> is right-to-left(C7-C0). So C3 and C4 have the lowest

> priority, is it right?

In my opinion, I think it is one tradeoff between the performance and the complexity of hardware.

> [2] Is it a method that used by a lot of design or

> only SPARC?

I think it is a special design for SPARC.

Doria at 2007-7-7 1:02:57 >

top of Java-index,Open Source Technologies,OpenSPARC...

# 2

I think the idea was to "avoid persistent bias" if multiple requests arrived in the same cycle. Consistently arbitrating left-to-right (or right-to-left) could probably be considered "left-biased" (or "right-biased").

niagara_archperf at 2007-7-7 1:02:57 >

OpenSPARC

Java programming resources: FAQs, tutorials, compiler and browser download sites, documentation, books lists, IDEs, etc.

CCX Crossbar Arbiter

Open Source Technologies New Topic

Open Source Technologies Hot Topic

Open Source Technologies

All Classified