gem5-users@gem5.org

The gem5 Users mailing list

View all threads

Questions on GPU Viper + directory

AM
Anoop Mysore
Tue, Nov 7, 2023 2:29 AM

I'm looking into MOESI_AMD_Base-dir.sm
https://github.com/gem5/gem5/blob/71973b386ee7375137c6e27865ab07db8036f616/src/mem/ruby/protocol/MOESI_AMD_Base-dir.sm#L1042C54-L1042C59
The condition being checked here determines if the TBE will be written-back
to memory or not
The condition should be: tbe.wtData || tbe.atomicData || tbe.Dirty == *true*, which would mean dirty data will be written back to memory --
right?

Also, CtoD event is never triggered. So state BP, action ic_probeInvCore
and c_sendResponseCtoD are currently useless. Is there an upcoming need for
this event? (I see this on the first commit 7 years ago)

I'm looking into MOESI_AMD_Base-dir.sm https://github.com/gem5/gem5/blob/71973b386ee7375137c6e27865ab07db8036f616/src/mem/ruby/protocol/MOESI_AMD_Base-dir.sm#L1042C54-L1042C59 The condition being checked here determines if the TBE will be written-back to memory or not The condition should be: `tbe.wtData || tbe.atomicData || tbe.Dirty == *true*`, which would mean dirty data will be written back to memory -- right? Also, CtoD event is never triggered. So state BP, action ic_probeInvCore and c_sendResponseCtoD are currently useless. Is there an upcoming need for this event? (I see this on the first commit 7 years ago)