[gem5-users] Re: Traces of instructions only.
There is no stupid question around here, ask away :-) On Wed, May 6, 2020 at 10:33 PM Øivind Harket Bakke via gem5-users wrote: > > Dear Santilli, > > that fixed everything except my embarrassment. > Thank you very much, really appriciated! > > Best regards, > Øivind > ___ > gem5-users mailing list -- gem5-users@gem5.org > To unsubscribe send an email to gem5-users-le...@gem5.org > %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s ___ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
[gem5-users] Re: Traces of instructions only.
Dear Santilli, that fixed everything except my embarrassment. Thank you very much, really appriciated! Best regards, Øivind ___ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
[gem5-users] Re: Traces of instructions only.
Maybe did you mean to use -ExecMicro to remove microops rather than -ExecMacro? E.g. one STP generates 4 microops as shown at: https://cirosantilli.com/linux-kernel-module-cheat/#gem5-execall-trace-format On Wed, May 6, 2020 at 8:54 PM Øivind Harket Bakke via gem5-users wrote: > > Hi all, > > I'm currently trying to export traces from a simulation, but I'm only > interested in the instructions that are being simulated. I compare the stats > file (# of simulated instructions) to the traces and they don't match in > size. I've tried to figure out which are the excessive lines in order filter > them out. So my question is as follows: How would I proceed to get traces of > the instructions only? > > Here's the command I run: > ./build/ARM/gem5.opt --debug-flags=Exec,-ExecMacro --debug-file=trace.out -d > configs/Forschung/arm/1G-500M/ configs/Forschung/arm/starter_se.py > --cpu="hpi_nL" --cpu-freq="1GHz" tests/test-progs/hello/bin/arm/linux/hello > > For clearification, the config file is a copy of the one in arm examples, > just adding a HPI model without L2 caches.(hpi_nL) > I've checked out the --debug-help option, but as far as I could tell, there > were no flags to remove from Exec or other flags to enable in order to reach > my goal. > > > Thanks in advance, > Øivind Harket Bakke > ___ > gem5-users mailing list -- gem5-users@gem5.org > To unsubscribe send an email to gem5-users-le...@gem5.org > %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s ___ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
[gem5-users] Re: Traces of instructions only.
Hi Jason, I read through the documentation you linked me and I've actually looked at it before. Only problem is that I need an in-order cpu to represent the hardware that we use. I'll take a closer look at it and see if I can make it work, or atleast run some tests with this to learn more about it and maybe implement atleast parts of it later. Thank you very much. Best regards, Øivind ___ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
[gem5-users] Re: Running simple program with atomic memory at different frequencies. Tick count in the traces are confusing.
Hi Jason, thank you very much for your fast reply. I thought about what you wrote and it makes sense. I thought by having only a cpu and memory, I could remove the timing of the memory to simulate the CPU alone. Would have been somewhat a dirty solution anyway. Thanks again. Best regards, Øivind ___ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
[gem5-users] Re: Traces of instructions only.
Hi Øivind, I would check out the documentation on the TraceCPU: http://www.gem5.org/documentation/general_docs/cpu_models/TraceCPU. I think that page describes how to capture the traces. Cheers, Jaosn On Wed, May 6, 2020 at 12:53 PM Øivind Harket Bakke via gem5-users < gem5-users@gem5.org> wrote: > Hi all, > > I'm currently trying to export traces from a simulation, but I'm only > interested in the instructions that are being simulated. I compare the > stats file (# of simulated instructions) to the traces and they don't match > in size. I've tried to figure out which are the excessive lines in order > filter them out. So my question is as follows: How would I proceed to get > traces of the instructions only? > > Here's the command I run: > ./build/ARM/gem5.opt --debug-flags=Exec,-ExecMacro --debug-file=trace.out > -d configs/Forschung/arm/1G-500M/ configs/Forschung/arm/starter_se.py > --cpu="hpi_nL" --cpu-freq="1GHz" tests/test-progs/hello/bin/arm/linux/hello > > For clearification, the config file is a copy of the one in arm examples, > just adding a HPI model without L2 caches.(hpi_nL) > I've checked out the --debug-help option, but as far as I could tell, > there were no flags to remove from Exec or other flags to enable in order > to reach my goal. > > > Thanks in advance, > Øivind Harket Bakke > ___ > gem5-users mailing list -- gem5-users@gem5.org > To unsubscribe send an email to gem5-users-le...@gem5.org > %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s ___ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
[gem5-users] Re: Running simple program with atomic memory at different frequencies. Tick count in the traces are confusing.
Hi Øivind, Atomic mode should not be used for timing. While *some* timing is required even in atomic mode (gem5 is still an event driven simulator), the timing is meaningless. To answer your question as to "why": it's because only the memory system is "atomic" the rest of the simulator is still operating with some timing. Cheers, Jason On Wed, May 6, 2020 at 1:11 PM Øivind Harket Bakke via gem5-users < gem5-users@gem5.org> wrote: > Hi all, > > quick question regarding atomic memory. I'm using the starter_se.py config > script (gem5/configs/example/arm/) and with memory mode being atomic. For > what I understood, using atomic memory mode should provide the data > irregardless of its position in memory(L1, L2 or off-chip mem) within the > next (simulation) clock cycle. When running a simple program and exporting > the traces at a cpu frequency of 1GHz and 500MHz, I assumed the tick-count > to be equal, meaning 1GHz would finish at half the time of the 500MHz. To > my surprice it didn't, leading me to believe there are some other system > dependencies running at a different frequency. Does anyone know what that > might be, or any clue what causes these different runtimes? I tried setting > the system frequency to match the clock frequency aswell, with the same > result. I ended up with a factor of 1.24 more ticks in the 500MHz > simulation. > > Thanks in advance, > Øivind Harket Bakke > ___ > gem5-users mailing list -- gem5-users@gem5.org > To unsubscribe send an email to gem5-users-le...@gem5.org > %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s ___ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
[gem5-users] Running simple program with atomic memory at different frequencies. Tick count in the traces are confusing.
Hi all, quick question regarding atomic memory. I'm using the starter_se.py config script (gem5/configs/example/arm/) and with memory mode being atomic. For what I understood, using atomic memory mode should provide the data irregardless of its position in memory(L1, L2 or off-chip mem) within the next (simulation) clock cycle. When running a simple program and exporting the traces at a cpu frequency of 1GHz and 500MHz, I assumed the tick-count to be equal, meaning 1GHz would finish at half the time of the 500MHz. To my surprice it didn't, leading me to believe there are some other system dependencies running at a different frequency. Does anyone know what that might be, or any clue what causes these different runtimes? I tried setting the system frequency to match the clock frequency aswell, with the same result. I ended up with a factor of 1.24 more ticks in the 500MHz simulation. Thanks in advance, Øivind Harket Bakke ___ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
[gem5-users] Traces of instructions only.
Hi all, I'm currently trying to export traces from a simulation, but I'm only interested in the instructions that are being simulated. I compare the stats file (# of simulated instructions) to the traces and they don't match in size. I've tried to figure out which are the excessive lines in order filter them out. So my question is as follows: How would I proceed to get traces of the instructions only? Here's the command I run: ./build/ARM/gem5.opt --debug-flags=Exec,-ExecMacro --debug-file=trace.out -d configs/Forschung/arm/1G-500M/ configs/Forschung/arm/starter_se.py --cpu="hpi_nL" --cpu-freq="1GHz" tests/test-progs/hello/bin/arm/linux/hello For clearification, the config file is a copy of the one in arm examples, just adding a HPI model without L2 caches.(hpi_nL) I've checked out the --debug-help option, but as far as I could tell, there were no flags to remove from Exec or other flags to enable in order to reach my goal. Thanks in advance, Øivind Harket Bakke ___ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s