I don't recall the details, but there's some issue with data accesses and instruction fetches sharing the same port to memory with O3 that leads to a livelock (or deadlock?) situation... something like you need to do an ifetch to make forward progress, but every time the port is free you re-issue a speculative data access instead, or maybe it's the other way around. We have run into this before--I think the same problem occurs with a unified L1--but since it goes away once you use split caches, and it's unrealistic not to have split caches anyway, we never considered it worth fixing.
Steve On Sat, Nov 22, 2014 at 9:27 AM, Thom Popovici via gem5-users < [email protected]> wrote: > Hi, all! > > I want to remove the caches from the full system configuration. I want > multiple processors that are directly connected to the membus and the > memory. As far as I saw the O3CPU demands caches, why? I managed to modify > the function request_caches from True to False. I managed to instantiate > all the objects without errors but Gem5 freezes up at the very beginning: > > Entering event queue @ 0. Starting simulation... > > Any ideas. > > Thanks so much > > _______________________________________________ > gem5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >
_______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
