Yes, you should be comparing them with commitedOps not Insts and yes it would be approximate since some instructions are getting squashed, but it's likely in the same ballpark. It's should be pretty easy to duplicate the reporting in the IEW that is creating the stats and have it print a similar instruction mix statistic in the commit stage. If you implement this please send along a patch.
Thanks, Ali On 01.08.2013 11:06, Mahshid Sedghi wrote: > Hi Ali, > Thanks for the hint. But these numbers would be approximate, since part of them won't commit. Is that correct? > About Yongbing's point: I am also observing a huge difference between those numbers. I was thinking maybe we should compare that number with commit.committedOps, rather than commit.committedInsts. Please let us know your comments. > > Thanks. > > On Wed, Jul 31, 2013 at 10:01 AM, huangyongbing <[email protected]> wrote: > >> Hi， >> >> I notice that in the stat.txt results, the number of executed instructions (iew.iewExecutedInsts) is quite larger than the number of instructions committed (commit.committedInsts) for o3 cpu. For example, 2504310485 [3] for iewExecutedInsts V.S. 1758659165 for committedInsts. >> >> So what is the reason? It cannot be caused just because of branch misprediction. >> >> Thanks. >> >> Best regards, >> >> Yongbing Huang >> >> FROM: [email protected] [mailto:[email protected]] ON BEHALF OF Ali Saidi >> SENT: Wednesday, July 31, 2013 6:48 AM >> TO: gem5 users mailing list >> SUBJECT: Re: [gem5-users] Finding the Instruction Mix for a Benchmark >> >> The IEW statistics based on instruction class for the o3 CPU might provide what you want. If you want to print OpClass information out in the simple CPU it should be doable. >> >> Ali >> >> On 26.07.2013 17 [4]:36, Mahshid Sedghi wrote: >> >>> Hi, >>> >>> I need to find out the instruction mix for some benchmarks. In fact, I want to see what is the breakdown of instruction count into integer insts, memory ref, and so on. According to the following thread, gem5 does not report the breakdown, and "int_inst" includes all the instructions that use integer registers, which will include a big part of load instructions as well. >>> >>> http://www.mail-archive.com/[email protected]/msg06828.html [2] >>> >>> Any idea how to get the breakdown numbers? Which part of the source code do I need to modify to implement this? >>> >>> Thanks everyone, >>> >>> Mahshid >>> >>> _______________________________________________ >>> >>> gem5-users mailing list >>> >>> [email protected] >>> >>> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users [1] >> >> _______________________________________________ >> gem5-users mailing list >> [email protected] >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users [1] > > _______________________________________________ > gem5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users [1] Links: ------ [1] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users [2] http://www.mail-archive.com/[email protected]/msg06828.html [3] tel:2504310485 [4] tel:26.07.2013%2017
_______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
