On Tue, Jul 27, 2010 at 3:18 PM, Hz Xes <[email protected]> wrote:
> Hi
>
>
>
> Firstly, thanks for this tremendous work of architecture simulator for
> ISA-x86.
>
Thanks. This work is an incremental work based on two great applications
QEMU and PTLsim and we do appreciate their efforts.
>
>
> I am new for Marss86, and I have questions on checkpoint and
> fast-forwarding.
>
>
>
> (1)
>
> I used provided python scripts (“create_checkpoint.py”) to create
> checkpoint for parsec benchmarks (via provided parsec disk image). And I
> could do it and run benchmarks starting from the checkpoint. My question is
> that: since parsec benchmark set has several input set such as test,
> simsmall, simdev, simmedium, simlarge and native, what is the input set used
> for creating the checkpoints by the “create_checkpoint.py” script? When I
> start simulation from the created checkpoint by provided
> “create_checkpoint.py” script, what is the input set used for the
> simulation? I read the scripts, but I can not figure it out. And does the
> checkpoint start from beginning of benchmark or somewhere after the
> beginning point?
>
For parsec 'create-checkpoint.py', I dont remember which input set is used,
but its either 'simsmall' or 'simmedium'. Other input set 'simlarge' and
'native' are too big for the simulator to complete in decent time. The
script contains a command line that is executed in VM to start the
benchmark. The format of the command executed is like following:
~/create_checkpoint bench_name; bench_executable parameters ; ~/stop_sim
Here 'create_checkpoint' is an executable in the disk image that triggers
the 'ptlcall' to create a checkpoint with benchmark name. So when you start
our simulations from checkpoints, it will continue executing other part of
the command line and will start the benchmark simulation.
> (2)
>
> How could I create a checkpoint after benchmark runs a certain mount of
> instructions (such as 2 billion instructions in native x86 host)? With such
> checkpoint, simulation could start the experiments directly from the region
> of interest.
>
Currently this feature is not implemented yet. One possible way is to start
the checkpoint without '-run' option, so it will run in the emulation mode.
And then you can use '-startrip IPADDRESS' option to switch to simulation
mode at specific RIP address of the benchmark.
> (3)
>
> When launching the simulation, how could I do the fast-forwarding a certain
> mount of instructions (such as 2 billion instructions in native x86 host),
> then do the detail architecture simulation with hooked PTLsim?
>
This feature is not implemented yet. If you want to implement it, here is a
quick pointer:
QEMU has a counter that counts how many instructions are emulated, you can
use this to setup a configuration option to switch to simulation mode when
specific amount of instructions are emulated.
- Avadh
> Thanks a lot!
>
> HZ
>
>
> _______________________________________________
> 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
