Re: [Gems-users] Flow of an IFetch


Date: Wed, 18 Oct 2006 13:40:28 -0500 (CDT)
From: Mike Marty <mikem@xxxxxxxxxxx>
Subject: Re: [Gems-users] Flow of an IFetch
> > Thanks.
> > yes, actually It stalls (not recycle) the the mandatory queue. I guess
> > this is for in order requirement (mandatory queue is a FIFO). right?
>
> MOESI_CMP_directory uses recycles instead so that other mandatory queue
> requests can be processed while one is stalled.
>
> I would think the same thing could (and should) be done for
> MSI_MOSI_CMP_directory.
>

On the other hand, the only time a request is stalled in the mandatory
queue is if something else is already going on for the same block (like a
replacement, or a conflicting request for the same block address).  This
should be very rare, so MLP (memory-level parallelism) is not really
impacted by this implementation decision of using z_stall.

[← Prev in Thread] Current Thread [Next in Thread→]