[parsec-users] some questions about ferret
jim at quickthreadprogramming.com
Mon Dec 23 08:44:49 EST 2013
For some applications, like ferret, the processing can be setup in stages,
also called a pipeline, whereby processing can be performed concurrently in
multiple stages. Some parallel programming paradigms are task model (TBB)
while others are thread model (pThreads). When using thread model paradigms,
you want to tune the thread loads such that no stage has threads waiting.
Pipelining can also reduce the buffering requirements.
From: parsec-users-bounces at lists.cs.princeton.edu
[mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of Benzhi Cao
Sent: Sunday, December 22, 2013 3:06 AM
To: parsec-users at lists.cs.princeton.edu
Subject: [parsec-users] some questions about ferret
Hi, recently I use the parsec benchmark suite to do performance testing, I
know the ferret benchmark is based on ferret toolkit. And I have some
questions about ferret , explains as follows:
There are six stages in ferret,including load, Image segmentation, Feature
extraction, Indexing , Ranking and out. And in the source code, you declares
different variables for the number of threads for different stages like:
NTHREAD_SEG, NTHREAD_EXTRACT, NTHREAD_VEC, NTHREAD_RANK. So my question is:
1)Does ferret often be used in a way that the number of threads for
different stages are different? 2)If it was true, can you give me an example
and explain why?
Thanks a lot, any help would be appreciated~
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the parsec-users