I don't know all the ins and outs of how stats are accumulated (they're not always counting what you think they are), but do realize that in FS mode cores that aren't running user code are in the kernel idle loop (just like on a real machine). So total ticks should normally be consistent across all the cores since the cores don't magically vanish and reappear based on whether they have something to do. (Though there may be differences if you're simulating from boot and the OS actually enables the extra cores later in the boot process.)
As for 2 vs 4 cores, that's more mysterious, but if your benchmark is too small then you could be getting eaten up by overheads. Steve On Mon, Aug 31, 2009 at 12:42 PM, Aaron Williams<[email protected]> wrote: > Hello All, > > I had a question about the results I am obtaining while running a benchmark > program I wrote using pthreads. The benchmark I am using to test is a > simple operation where each thread created does a dot product of two vectors > with 1 million elements in each. I am seeing a result where the ticks from > USER mode look proper the way I am exspecting them. That is to say that one > core has a higher load as it is where the main thread runs and then each of > the other cores has some small amount of usage on them. Then when I look at > the total number of ticks spent on each core, they are all equal. This > happens because there is a large portion of the time spent in "KERNEL" mode > on the cores with lower utilization. I am not sure why the KERNEL mode is > so high on these other cores. Any insight. > > Also another strange artifact that is occuring is that when I run a two core > version, the total number of ticks on spent on each core is about half that > of the time for the 4 core version. This makes no sense to me as I would > expect the 4 core version to take half the time not twice the time... > > Any insights? > > Thanks in advance. > > -- > Aaron S. Williams > Graduate Student > Arizona State University > Department of Electrical Engineering > [email protected] > > _______________________________________________ > m5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > _______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
