Re: [Gems-users] Flow of an IFetch

[Mike Marty <mikem@xxxxxxxxxxx>]

> Thanks for the prompt reply.
>
> > The original request message will sit in the
> > mandatory queue until this is done.
> How is it determined that the IFetch can't be processed when the it has
> already issued PUTS (block is already deallocated from L1D)? Any poniter?

When the PUTS is issued, I believe the block will enter a transient state.
Look what transition happens when an Ifetch arrives at a block in a
transient state due to replacement?  It probably "recycles" the mandatory
queue...thereby trying to process the Ifetch again in the future (I think
3 cycles is our default recycle latency)

--Mike