Benchmarking for automatic parallelization project

Hi all,
I am looking appropriate Benchmarking for the assessment of automatic parallelization project. What Benchmarking do you suggest me?

regards
neda

I am looking appropriate Benchmarking for the assessment of automatic
parallelization project. What Benchmarking do you suggest me?

  Do you mean auto-parallelization or auto-vectorization? If it's the latter, I
sugguest you find scientific or graphic applications. I don't think there is
auto-parallelization benchmark, however.

Regards,
chenwj

> I am looking appropriate Benchmarking for the assessment of automatic
> parallelization project. What Benchmarking do you suggest me?

  Do you mean auto-parallelization or auto-vectorization? If it's the latter, I
sugguest you find scientific or graphic applications.

For auto-vectorization, look at:
http://polaris.cs.uiuc.edu/~garzaran/doc/pact11.pdf
http://polaris.cs.uiuc.edu/~maleki1/TSVC.tar.gz

If you're looking for auto-parallelization, I can ask around.

-Hal

Hi Hal,

For auto-vectorization, look at:
http://polaris.cs.uiuc.edu/~garzaran/doc/pact11.pdf
http://polaris.cs.uiuc.edu/~maleki1/TSVC.tar.gz

If you're looking for auto-parallelization, I can ask around.

  Thanks for the sharing. If that doesn't bother you too much, I would like to
know auto-parallelization benchmarks. :stuck_out_tongue:

  Just curious. I was thinking that auto-parallelization in compiler optimization
is not mature/practical, that's why I thought there is no auto-parallelization
benchmark. Do you know any project on this? Thanks.

Regards,
chenwj

Don't take this the wrong way, but ...
Auto-parallelization techniques in compilers have been mature since
the late 70's.

There are plenty of benchmarks and test suites

The typical ones are:

Parallel Loops (by Cray/Oak Ridge)
The Perfect Benchmarks
Livermore loops
LINPACK
NAS Parallel Benchmarks
etc

You just have to realize that compiler parallelization happened in
Fortran *many many* years before it became common for C/C++/etc

Yes, auto-parallelization is not practical, but it solves something. So it should have benchmarks.

Other people have tried to answer you directly, but I think a
broader response may also be helpful.

There is no such thing as a benchmark for the assessment of
automatic parallelization. Automatic parallelization is too
diverse a field to be coherently represented by a single benchmark,
or even a single family of benchmarks.

I suggest the following questions:

Do you have an application or area of applications you want to study?
Do you have a platform or architecture family you want to study?
Are you interested in a particular programming language, or family?
Are you studying parallelization as part of a broader inquery?

Dan