auto parallelization vs openmp

bustaf
Total Points:
2,510
Status Points:
2,010
Brown Belt
July 1, 2009 8:13 PM PDT
Rate
 
#3 Reply to #2
Quoting - tim18

The performance should be comparable as auto parallelization is base on OpenMP implementation. We may be able to anylyze the issue if you provide sample test case and compiler options.

This assumes that the auto parallelizer recognizes the opportunity to parallelize.  You didn't say whether you turned on -par-report and -openmp-report to see if the same loops report parallelization.  In my test cases, the current compilers, with normal options, are quite reluctant to auto-parallelize.  Auto-parallel never was advocated as a replacement for OpenMP.
The 10.0 compilers were much more aggressive with -parallel than the recent ones, leading to justified complaints about the priorities.

Hi majeti_deepak and all
I think that confusing about parallel and OpenMp
1] when you compile with -parallel 
you see your functions() messages compiler  as attribute share affinity several processor ??
You must having several physical processor in machine for
this option can work.
With 2 cores only one processor  -parrallel not operate can just slow machine.
Also with wrong flag type processor  (SSS?) - parallel can also not work.
If parallel working correctly you having very very better result
that without (benefit is better  potential loop vectorized (with var local  in function) also charge is really decreased from the server (make test with Apache or great database, is easy  you observe benefit).
About relation affinity processor and OpenMp (thread) i think
explain documentation not exist you are obliged read the sources, without probably several confusing (trhead, pid, core processors,  physical processors) .
Is difficult to controlling or to share  manually specific affinity processor in
same time OpenMp function. Intel have call councils that programmer must use parallel programming, I answer  is actually
same you driving your car without rules or without the wheel.
For finish  if you using SCTP (one socket for several address)in you programming is better clear or easy to use MPICH2
with several computer (you have exactly processor that you want is working)
Here, is the night, can not started an big machine (noise PSU)
Tomorrow i start an big machine and i add  Icc messages compiler parallel options for that  you can see?
Best regards


Intel Software Network Forums Statistics

8491 users have contributed to 31629 threads and 100769 posts to date.
In the past 24 hours, we have 28 new thread(s) 129 new posts(s), and 184 new user(s).

In the past 3 days, the most popular thread for everyone has been Implicite multithreading ??? The most posts were made to Crash when loading skeleton The post with the most views is Dear Steve, excuse me for a d

Please welcome our newest member shadowwolf99