Vladimir and I (and others) were looking at what the issue was and working on a solution, and trying to make sure we understood what worked and what did not. Here is what we figured out as of this morning -
Yes, this appears to be an issue with the IPP product. Thank you for calling our attention to it, we plan to make a fix available shortly. Fortunately, the issue only affects the value returned by a single function in IPP (ippGetNumCoresOnDie).
The issue does not affect the proper usage of multiple cores on AMD processors by IPP. The determination of the number of cores for the parallel functions in IPP uses the OpenMP infrastructure which does determine the core count properly.
This function, which exists for applications to query, is returning the wrong number. We certainly will fix this!
We have figured out the problem and will have a posting shortly with a fixed version of the library that will resolve the issue.
We will post a follow-up message with a link to the fixed version when it is available. The fix will, or course, also be in future versions of IPP.
Thanks again for letting us know about this problem.