Thanks for this. I don't think it's working quite right though. When I call
'-kill', it does close the QEMU window but my stats file is incomplete
(<4MB) and ptlstats can't extract the final snapshot.
Maybe you can advise me here; I'm following the PTLsim suggestion and trying
to skip over "boring" parts of my application. I would like to wrap
important kernels between calls to ptlcall_switch_to_sim()
and ptlcall_switch_to_native(). When the application runs to completion I
would like the stop the simulation entirely and flush my log and
stats. Since ptlcall_switch_to_native() wasn't functioning correctly I
replaced it with ptlcall_single_flush("-stop") hoping for a similar effect.
Maybe this is where I'm making a mistake?
Kind regards
Tim
On 7 January 2011 22:01, avadh patel <[email protected]> wrote:
>
> On Fri, Jan 7, 2011 at 11:27 AM, Timothy Hayes <[email protected]> wrote:
>
>> Thanks for the swift reply.
>>
>> Yes, the '-stop' causes it to return to emulation mode. The '-kill' is
>> thus given in emulation mode too. I was hoping I could wrap functions of
>> interest in '-run' & '-stop' statements and then flush the stats/logfile at
>> the end of the program's execution by using '-kill'.
>>
>> I have attached a patch that will check '-kill' flag and kill the
> simulation process even if its in emulation mode.
> I haven't tested it because currently I dont have access to any setup on
> which I can test it. Let me know if this works then I can patch to master
> branch.
>
>
>> By the way, is there a way to do the above without using '-kill'? It would
>> be nice to run several benchmarks & recalling simconfig without having to
>> reboot the MARSS environment each time.
>>
>> Currently there is no method to do that but way we work is we create
> checkpoint for each benchmark and then use a script to run each benchmark
> one after another.
> You can find this script to run checkpoints here:
> https://github.com/downloads/avadhpatel/marss/run_bench.py
> Or you can click on 'Downloads' on github and it will show up.
>
> - Avadh
>
> Kind regards
>> Tim
>>
>>
>> On 7 January 2011 18:40, avadh patel <[email protected]> wrote:
>>
>>> I guess I found the issue. When '-kill' is issued if Marss is not in
>>> simulation mode it will no kill the whole process. The fix is to kill the
>>> process when we saw '-kill' signal in 'plt_machine_configure'.
>>>
>>> Please can you answer the following two questions so I can clearly set
>>> the scenario as you mentioned and fix this issue in master branch.
>>>
>>> 1. When you give '-stop' does it switch to emulation mode and keep
>>> running or it stops the emulation also?
>>> 2. When '-kill' is given is it running in emulation mode or simulation?
>>>
>>> Thanks,
>>> Avadh
>>>
>>> On Fri, Jan 7, 2011 at 9:41 AM, Timothy Hayes <[email protected]> wrote:
>>>
>>>> Hello
>>>>
>>>> I've started using MARSS for a research project, my thanks to all of you
>>>> involved for a great piece of software.
>>>>
>>>> I have a question regarding switching between simulated and native
>>>> modes. I would like to execute my program in full and wrap the significant
>>>> kernels in 'ptlcall_switch_to_sim()' and 'ptlcall_switch_to_native()'. The
>>>> former works fine but when I call the latter, MARSS informs:
>>>> MARSSx86::Command received : -native
>>>> Warning: invalid option '-native'
>>>>
>>>> I've tried using 'ptlcall_single_flush("-stop")' in lieu of -native, but
>>>> when I try to do 'ptlcall_kill()' at the end of the program's execution,
>>>> MARSS informs:
>>>> MARSSx86::Command received : -kill
>>>> Warning: only one action (from -run, -stop, -kill) can be specified at
>>>> once
>>>>
>>>> This is problematic as the final snapshot isn't written to my stats
>>>> file. Can you advise me the typical way users circumvent this issue?
>>>>
>>>> Kind regards
>>>> Tim
>>>>
>>>> p.s. this naturally leads me to ask if I can run multiple benchmarks
>>>> with different logfiles/stats without having to kill and restart the SUT
>>>> each time?
>>>>
>>>>
>>>> _______________________________________________
>>>> 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
