Re: [Gems-users] Question about DIRECTORY latency in directory protocols

["Mike Marty" <mike.marty@xxxxxxxxx>]

That latency assumes a perfect directory cache, or a directory implemented in SRAM

--Mike

On Thu, Apr 3, 2008 at 6:05 PM, Niket <niketa@xxxxxxxxxxxxx> wrote:

Hi all,

I had a question regarding the latency that the directory protocols
assume for looking up on-chip sharers. For instance, while sending out
invalidates to all L1  shares on a GETX event, the latency assumed is
L2_REQUEST_LATENCY which is of the order of 4 cycles. Since this
invalidation request requires a sharer list lookup, shouldn't this
latency be more ? Something like DIRECTORY_LATENCY which is of the order
of 80 cycles ?

 action(tt_issueSharedInvalidateIntL1copiesRequest, "\t",
desc="invalidate all L1 S copies") {
   enqueue(L1RequestIntraChipL2Network_out, RequestMsg,
latency="L2_REQUEST_LATENCY") {
     out_msg.Address := address;
     out_msg.Type := CoherenceRequestType:INV_S;
     out_msg.RequestorMachId := machineID;
     out_msg.Destination := L2cacheMemory[address].Sharers;
     out_msg.MessageSize := MessageSizeType:Control;
   }
 }

Cheers,
Niket
_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.