Path: utzoo!attcan!uunet!husc6!uwvax!oddjob!ncar!noao!nud!tom
From: tom@nud.UUCP (Tom Armistead)
Newsgroups: comp.arch
Subject: Re: RISC machines and scoreboarding
Message-ID: <1099@nud.UUCP>
Date: 24 Jun 88 17:37:21 GMT
References: <1082@nud.UUCP> <22147@amdcad.AMD.COM> <1095@nud.UUCP> <22196@amdcad.AMD.COM>
Reply-To: tom@nud.UUCP (Tom Armistead)
Organization: Motorola Microcomputer Division, Tempe, Az.
Lines: 28

In article <22196@amdcad.AMD.COM> tim@delirun.amd.com (Tim Olson) writes:
>|     The scoreboard method can allow register forwarding also.

>Can you elaborate a bit?  Forwarding requires that you also know which
>functional unit(s) to forward from -- don't you need other hardware than
>the scoreboard bits for this?
>	-- Tim Olson

    Yes other hardware is required.  My posting was intended to elaborate that
scoreboarding doesn't *preclude* register forwarding - not that scoreboarding
itself is sufficient to implement the feature.

   Why do you say that forwarding requires knowledge of which functional 
unit will be supplying the fed forward result?  The instruction unit knows
which register(s) it is waiting on and can enable feed forward circuitry
so that when the result(s) is available, it is fed to the instruction unit
simultaneously with the result is being written to the register file.  With
this method, the feed forward logic can be independent of the functional units.
The instruction unit doesn't know, and doesn't care, what is supplying the
result but just knows that it needs that particular result for the next
instruction and snags it.

   Forwarding could also be implemented by connecting the input of the
instruction unit to the output of the particular functional unit that
will provide the result but this would require a seperate feed forward
path for each functional unit.  This would also work but as you say,
requires knowledge of which functional unit the result(s) is expected
from. 
-- 
Just a few more bits in the stream.

The Sneek