[parsec-users] Errors running Parsec benchmarks in M5

Christian Bienia cbienia at CS.Princeton.EDU
Tue Aug 18 02:04:09 EDT 2009

Hi Ef,


That looks very strange. The first error inside streamcluster happens inside
the pthreads library. I don't think it's a bug in the program. There is
however always the chance that it's some sort of race condition and you just
happen to discover it.


The second bug in fluidanimate happens when the program writes its output to
disk. That's after the ROI has ended. You can check whether the error also
happens if you disable disk output simply by removing the last argument for
fluidanimate in the *.runconf file in the parsec/ directory of the package.
It specifies the output file and if nothing is given then disk output is
disabled. Since it's not part of the ROI it won't affect the characteristics
as long as our guidelines are followed and only the ROI is measured.


If you find out more please let me know. I'm willing to help debugging any
issues that are found in the workloads themselves.


- Chris



From: parsec-users-bounces at lists.cs.princeton.edu
[mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of ef
Sent: Tuesday, August 18, 2009 1:56 AM
To: PARSEC Users
Subject: [parsec-users] Errors running Parsec benchmarks in M5



At the moment I am having difficulties on the following two benchmarks:

For streamcluster I ran into the following issue:
PARSEC Benchmark Suite Version 2.1
read 4096 points
streamcluster: pthread_mutex_lock.c:82: __pthread_mutex_lock: Assertion
`mutex->__data.__owner == 0' failed.

Any ideas on this? I noticed a bug here:

I am not sure its revelant as I am running GCC 4.3.2.

For Fluid Animate I am getting the following error:
PARSEC Benchmark Suite Version 2.1
Loading file "/parsec/install/inputs/fluidanimate/in_35K.fluid"...
Number of cells: 36900
Number of particles: 35402 (1102 skipped)
Saving file "/parsec/install/inputs/fluidanimate/out.fluid"...
Segmentation fault

I am not to sure on this error, M5 does not allow real writes to the disk
image, however it offers a "fake layer" COW where writes can be temporarily
written too, I am not to sure this error is related to that, I need to do
some further debugging but I don't think the ROI region is being executed.
as it takes around 3 hours to get the segfault which is to fast. 

Any Ideas on this?


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20090818/b438b5e9/attachment.htm>

More information about the parsec-users mailing list