[parsec-users] streamcluster hangs

Christian Bienia cbienia at CS.Princeton.EDU
Wed Oct 21 08:48:46 EDT 2009


Hi Ramkumar,

 

Yes, there's plenty of opportunity for vector instructions. Some programs
were already manually vectorized, just have a more detailed look. I also
noticed lots of vector instructions in the compiler-generated binaries but
didn't bother to analyze this in more detail.

 

Best,

Chris

 

 

From: parsec-users-bounces at lists.cs.princeton.edu
[mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of ramkumar
jayaseelan
Sent: Wednesday, October 21, 2009 3:06 AM
To: PARSEC Users
Subject: Re: [parsec-users] streamcluster hangs

 

Hi Chis,
Thanks Seems like the smaller download is working fine. I had a generic
question about the suite. Looking through some of the code in the suite it
seems like they will be very friendly to vectorization and any MM/ SSE
extensions. Have you guys seen a lot of vector instructions being generated
in the code with your studies?
Thanks
Ramkumar



2009/10/20 Christian Bienia <cbienia at cs.princeton.edu>

Hi Ramkumar,

 

Some users have reported that older browsers have trouble downloading the
PARSEC archive because it's larger than 2 GB. Have you tried downloading the
version of the suite that is broken up into three pieces? It contains the
same data, you should be able to download that if you have issues with the
large size of the unified archive.

 

- Chris

 

 

From: parsec-users-bounces at lists.cs.princeton.edu
[mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of ramkumar
jayaseelan
Sent: Tuesday, October 20, 2009 2:54 AM
To: PARSEC Users
Subject: Re: [parsec-users] streamcluster hangs

 

Hi All,
I am  not able to download the parsec suite with the large inputs (total
size is around 3 GB).  The download keeps breaking at around 1 GB. Wondering
if there is some limit on the server. Have any of you hit this before. Any
tips regd the download will be of great help
Thanks
Ramkumar

On Sat, Oct 17, 2009 at 12:16 PM, Amittai Aviram <amittai.aviram at yale.edu>
wrote:

Hi!

I am finding that streamcluster consistently hangs on my machine right after
printing the output line giving the number of points read.  E.g., for the
default input size (small--10 points), and 2 threads, I get the following:

[After building with -c gcc-pthreads, apparently successfully ...]

afa9 at beacon:~/parsec-2.0-dthreads> parsecmgmt -a run -p streamcluster -c
gcc-pthreads -n 2
[PARSEC] Benchmarks to run:  streamcluster

[PARSEC] [========== Running benchmark streamcluster ==========]
[PARSEC] Deleting old run directory.
[PARSEC] Setting up run directory.
[PARSEC] No archive for input 'test' available, skipping input setup.
[PARSEC] Running 'time
/export/home/afa9/parsec-2.0-dthreads/pkgs/kernels/streamcluster/inst/i386-l
inux.gcc-pthreads/bin/streamcluster 2 5 1 10 10 5 none output.txt 2':
[PARSEC] [---------- Beginning of output ----------]
PARSEC Benchmark Suite Version 2.0
read 10 points
^C <-------------------------HANGS, HAD TO KILL

I've gotten the same behavior on 16 and 64 threads and on the "simlarge"
input.  If I do not enter a thread number, so that it defaults to sequential
execution, then the program does terminate.

I'm running the benchmarks under 32-bit Open Suse 11 (Linux) on an Intel
(i686) machine.

Any ideas?  Did I miss or forget something from the tutorial or other
documentation?  Thanks!



Amittai Aviram
Graduate Student in Computer Science
Yale University
646 483 2639
amittai.aviram at yale.edu
http://www.amittai.com

_______________________________________________
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

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20091021/1166208b/attachment.html>


More information about the parsec-users mailing list