[parsec-users] Fluianimate Runtime error

Shaunak Shah shaunaks at umich.edu
Tue Jun 30 18:03:06 EDT 2009


Unfortunately, I do not have access to the machine on which I have Parsec
installed, so I am a little fuzzy about the details. Let me sum up what i
remember. Note that I will be very vague about details.

I used Simics to simulate a solaris/sparc machine. I used the provided
parsecmgmt script to compile the binaries on a local machine. The script
copies the source code (located in ....pkgs/apps/blackscholes/src) to
...pkgs/apps/blackscholes/inst/sparc-solaris.gcc-hooks and uses that copy
for compilation. Once the code is compiled, the executable is moved to a
folder called 'run' (I forget exactly where, but in the blackscholes
subdirectory, perhaps under obj). The input archives are extracted to this
folder and the script runs the executable (with the appropriate command line
options) from this run folder. The reason parsecmgmt does this is because it
does not know whether the benchmark was run earlier or if an extracted
version of the input exists. It simply creates its own resources.

If you look at the parsecmgmt script, you can place echo statements at
strategic locations that will give you the run directory as well as the
command line arguments. Also, the command line arguments to be passed to a
benchmark are in a file somewhere in the benchmark subdirectory.
If you look in the imputs folder in blackscholes, it is empty. Blackscholes
does use external files as input. Almost all the other benchmarks do. This
is why blackscholes ran without complaining.

In order to get fluidanimate to run in simics, I imported the compiled
binary as well as the extracted input (different archives for simsmall,
simmedium, simlarge, etc) into the simulated machine. Then I ran the
fluidanimate binary with the command line arguments (which specify the path
to the input files, the number of threads,  the path to the output log,
etc).

-Shaunak


2009/6/30 Wael Kdouh <waelkdouh at gmail.com>

> Shuanak could you elaborate more on "As things would have it, blackscholes
> does not use an input archive.". As for the extra step you mentioned
> regarding simics I am not sure if it is necessary as parsec is aware of
> simics. Thus I would imagine such a step being listed in the manual. Finally
> could you provide me with the path for the run folder so I can check if the
> binary file is sent to the right place.
>
> 2009/6/30 Shaunak Shah <shaunaks at umich.edu>
>
>  Hi Wael,
>>
>> parsecmgnt creates a run folder into which it copies the compiled binary
>> as well as the extracted input. The benchmark is then executed from that
>> directory. If you are running the benchmark in simics, you probably need
>> to import the extracted input and the binary into simics and provide the
>> appropriate path to the binary in the command line execution.
>>
>> As things would have it, blackscholes does not use an input archive.
>>
>> [Correct me if I'm wrong]
>>
>> Regards,
>> Shaunak
>>
>>  2009/6/30 Wael Kdouh <waelkdouh at gmail.com>
>>>
>>> Chris sorry to shift from the main topic of this thread, but when I
>>>>  installed the benchmark I unzipped the archived input file within the same
>>>> directory. Now I never thought about the installation process of the input
>>>> until you raised the issue. Please let me know if unzipping the archive in
>>>> the first place was the right thing to do.
>>>> You mentioned that parsecmgmt should tell me that it is unpacking the
>>>> archive with the input. For instance when I ran blackscholes  on a 16 core
>>>> machine I got  the following:
>>>>
>>>> [parsec] setting up run directory
>>>> pushd error retrieving current directory: getcwd : cannot access parent
>>>> directories. No such file or directory.
>>>> [parsec] no archive for input 'simlarge' available, skipping input
>>>> setup.
>>>> [parsec] running 'time
>>>> /parsec-2.0/pkgs/apps/blackscholes/inst/sparc-solaris.gcc
>>>> [parsec] [-----------------beginning of output ----------------------]
>>>> Parsec benchmark suite version 2
>>>> [Hooks] parsec hooks version 1.2
>>>> Num of options 65536
>>>> Num of runs 100
>>>> Size of data 2621440
>>>> [Hooks]Entering ROI
>>>> [Hooks]Leaving ROI
>>>> [Hook] Total time spent in Roi: 0.146 sec
>>>>
>>>> real 0.189 s
>>>> user 2.355 s
>>>> sys  0.007 s
>>>>
>>>> [parsec] end of output
>>>> /parsec-2.0/bin/parsecmgmt: line 943: popd: directory stack empty
>>>> [parsec] Done
>>>>
>>>> Chris I thought in the past that the above was a successful run since it
>>>> generated time, but since you mentioned the unarchiving step, which in the
>>>> above example does not seem successful, it made me realize that I may need
>>>> to reconsider installing the input in the right way. Could you please let me
>>>> know if the above run is indeed an unsuccessful run. If this is the case
>>>> could you please provide me with some insight on installing the input.
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> 2009/6/30 Christian Bienia <cbienia at cs.princeton.edu>
>>>>
>>>>>    Wael,
>>>>>
>>>>>
>>>>>
>>>>> It appears that the program failed to open the file with the input
>>>>> data. Have you installed the simulation inputs correctly? Parsecmgmt should
>>>>> tell you that it is unpacking the archive with the input.
>>>>>
>>>>>
>>>>>
>>>>> - Chris
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> *From:* parsec-users-bounces at lists.cs.princeton.edu [mailto:
>>>>> parsec-users-bounces at lists.cs.princeton.edu] *On Behalf Of *Wael Kdouh
>>>>> *Sent:* Tuesday, June 30, 2009 3:55 PM
>>>>> *To:* parsec-users at lists.cs.princeton.edu
>>>>> *Subject:* [parsec-users] Fluianimate Runtime error
>>>>>
>>>>>
>>>>>
>>>>> Hi I am trying to use parsec with Simics version
>>>>> 3.0.3. I am simulating a solaris/sparc machine with GCC 4.2
>>>>> installed. I managed to compile fluidanimate with the gcc-hooks
>>>>> configuration to use the magic break point capability within simics. But
>>>>> when I run it I get the following error:
>>>>>
>>>>>
>>>>>
>>>>> Assertion failed: file, file pthreads.cpp, line 185
>>>>>
>>>>> /parsec-2.0/bin/parsecmgmt: line 910: 15925 Abort (core dumped)
>>>>>
>>>>>
>>>>>
>>>>> Any insight would be highly appreciated.
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> parsec-users mailing list
>>>>> parsec-users at lists.cs.princeton.edu
>>>>> https://lists.cs.princeton.edu/mailman/listinfo/parsec-users
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> Sincerely yours,
>>>>
>>>> Wael Kdouh
>>>>
>>>>
>>>> _______________________________________________
>>>> parsec-users mailing list
>>>> parsec-users at lists.cs.princeton.edu
>>>> https://lists.cs.princeton.edu/mailman/listinfo/parsec-users
>>>>
>>>>
>>>
>>
>> _______________________________________________
>> parsec-users mailing list
>> parsec-users at lists.cs.princeton.edu
>> https://lists.cs.princeton.edu/mailman/listinfo/parsec-users
>>
>>
>
>
> --
> Sincerely yours,
>
> Wael Kdouh
>
>
> _______________________________________________
> parsec-users mailing list
> parsec-users at lists.cs.princeton.edu
> https://lists.cs.princeton.edu/mailman/listinfo/parsec-users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20090630/48ada3ce/attachment.html>


More information about the parsec-users mailing list