Date: | Mon, 09 Oct 2006 16:09:26 -0500 |
---|---|
From: | Dan Gibson <degibson@xxxxxxxx> |
Subject: | Re: [Gems-users] simulating perfect L2 for CMP |
PERFECT_MEMORY_SYSTEM = true implies that *all* memory requests will
observe the same latency. That latency is PERFECT_MEMORY_SYSTEM_LATENCY. IE
PERFECT_MEMORY_SYSTEM: true PERFECT_MEMORY_SYSTEM_LATENCY: 0 would imply that ALL requests are completed in zero time. To set the main memory latency to zero: MEMORY_RESPONSE_LATENCY_MINUS_2: -2and change the definition of memoryResponseLatency() in RubyConfig.h to simply return MEMORY_RESPONSE_LATENCY_MINUS_2 + 2. Regards, Dan Thomas Yeh wrote: Thx for the suggestion. I need the protocol to be simulated, so I'll have to run opal and ruby.I assume that setting the memory latency to zero involves setting PERFECT_MEMORY_SYSTEM to true and PERFECT_MEMORY_SYSTEM_LATENCY to 0.PERFECT_MEMORY_SYSTEM: false PERFECT_MEMORY_SYSTEM_LATENCY: 0 On Mon, 9 Oct 2006, Dan Gibson wrote:I recommend you set the memory latency to zero. My other comments follow: Thomas Yeh wrote:I am currently simulating with the MOESI_CMP_directory protocol, and would like to run the simulations with a perfect L2. What is the best way to do this? 1) Running opal without ruby and specifing L2_IDEAL?Running without ruby doesn't simulate any cache coherence. Unless this is what you're intending, running without Ruby is probably a bad option.2) Run both opal and ruby, but specify L2 is always a hit locally?Yes, you could do that.Is the L1 always 1 cycle when running opal w/o ruby? I don't see a parameter for setting the L1 latency.Opal w/o Ruby uses its own cache model, based on a uniprocessor hierarchy. See Opal's cache.[Ch].If I run without ruby, how do I compare the performance? Since there won't be a ruby_cycle metric.Opal will still produce several metrics. Pick your favourite and use that instead.thx, TomRegards, Dan_______________________________________________ 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.-- http://www.cs.wisc.edu/~gibson [esc]:wq! _______________________________________________ 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._______________________________________________ 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. -- http://www.cs.wisc.edu/~gibson [esc]:wq! |
Previous by Date: | Re: [Gems-users] simulating perfect L2 for CMP, Thomas Yeh |
---|---|
Next by Date: | Re: [Gems-users] Problem setting up runs with SNUCA and DNUCA config, Mishali Naik |
Previous by Thread: | Re: [Gems-users] simulating perfect L2 for CMP, Thomas Yeh |
Next by Thread: | Re: [Gems-users] separate user execution from os effects, Dan Gibson |
Indexes: | [Date] [Thread] |