Hello, While designing a new cache coherence protocol on release consistency model, I make an assumption regarding the simulation.
As far as I know, SIMICS relies on sequential consistency model, and it does not support different memory consistency model. At first, I thought that if SIMICS does not take advantages of release consistency model, such as active reordering of memory accesses, I might lose some possible improvement. However, since SIMICS executes instructions and feeds memory accesses to GEMS one by one in order, I guess reordering would not affect the latency of each memory access much. I guess if I want to take reordering of instructions into consideration, I have to deal with OPAL, and it would be more related to the overall performance, such as the execution time, not the performance of the coherence protocol. Therefore, I am assuming that if I use "properly labeled" applications for the simulation and treat different kinds of memory accesses accordingly (e.g. the new protocol needs to take some actions when executing Acquire or Release instructions, and I will pass those instructions to GEMS as well.), I can still evaluate the performance of the new coherence protocol, such as the average latency or the miss rate. Am I safe to make this assumption? I was wondering if it makes sense. Any comment would be highly appreciated. Thank you. Sincerely, Ray
_______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
