Well, unfortunately the address stream is going to be very highly application specific. One thing you could try to do is write a small test application that tries to allocate a very big chunk of memory and then just scans though it. That should in theory give you much better coverage of the address space and would be a good indication that everything is working correctly between DRAMSim2 and marss.
Other than that, there's not a whole lot you can do to affect the address stream -- perhaps by rewriting the virtual memory system in Linux :) On Tue, Jun 28, 2011 at 8:53 PM, Zhe Wang <[email protected]> wrote: > Hi, > > I am trying to use marss with DRAMsim2 to run some spec2006 benchmarks. I > have a question about the range of the physical address related to the size > of the memory. > > I set the memory size to 8092 megabytes to run the benchmarks, I expect the > range of the physical address would be 8*1024*1024*1024, that is using the > last 33 bits of the physical address to map the memory resource. But when I > run the marss simulation, I found the range of the physical address is much > smaller which might be within 2*1024*1024*1024. Since when I map the > physical address to the memory resource, I would like to map the highest bit > to the channel, that is map the 33 bit to the channel ID, the problem is I > found the 33 bit always 0 since the range of the physical address in marss > is smaller than the memory address. So does anybody has some idea about > this? Thanks. > > > > Best > zhe > > _______________________________________________ > http://www.marss86.org > Marss86-Devel mailing list > [email protected] > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > >
_______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
