OMP abort

[ooloan]

Hello, I have been experiencing the following error intermittently:

OMP abort: currently configured to support a maximum of 32 threads. Please specify a larger value via the KMP_ALL_THREADS environment variable to raise this limit.

The platform is WinXP, VC7.1, MKL7.2.1, Dual Zeon 2.8Ghz and I am using Win32 threads. My application never has more that 32 concurrent threads (maximum 4), but does create more than 32 in the lifetime of the application. Each threads calls into the MKL.

Does KMP_ALL_THREADS apply to the maximum number of concurrent threads or to the total number used? This page http://www.centerspace.net/support.php?page=1 suggests the latter. However, if this is the case I would always expect my application to terminate at thread # 33; however it frequently completes, having created over 1000 threads in the lifetime of the application. On other occasions, it terminates randomly.

As expected, I can consistently get the application to terminate by changing KMP_ALL_THREADS to less than the number of concurrent threads.

I hope someone can shed some light on this.

Yours,

Owen.

[auchich]

For everyone stumbling across this:

There was an article @ http://www.intel.com/support/performancetools/libraries/mkl/sb/CS-022854.htm

that said "fixed in Intel MKL8.0.2".

[Todd_R_Intel]

Thanks Auchich. Looks like the above link is no longer live, but I can confirm that this limitation to the threading library was fixed in a later version of MKL. I'll see if I can dig up the original page to see if it answers any of your questions.
-Todd

[Bin]

Hi Todd,

What does the fix mean here?

We encounter the same problem in MKL 10.0.3. We threads our application with Win32 thread. For each thread, it is possible to invoke functions in MKL using MKL threading. Then, the situationis nested parallelism.One of our test shows a console dialog pops up which says that

"OMP warning: Cannot form a team with 8 threads, using 7 instead. Consider setting KMP_ALL_THREADS to a number larger than 8."

Moreover, the worst thing is that it quits our application.

Bin WANG

[Todd_R_Intel]

Well this error message looks slightly different than the one that I saw previously. I'll have to look into this.
-Todd

[Bin]

Any update?

[Todd_R_Intel]

No update on this yet. I hope to get some answers from our threading library team soon. -Todd