Undetectable data races and deadlocks

[Ami_Marowka]

Dear All,

I would like to know the situations that inhibit Inspector XE to detect data races and deadlock for sure. For example, Inspector XE does not detect data races in region of code that is not executed.

I would like to know about other cases as well.

Moreover, I would like to know if there are possibilities (and if answer is yes, please give examples) that Inspector will report on data races and deadlock that were not occurred.

Thanks in advance.

Ami

[David_A_Intel1]

The best place to find this type of information is the Release Notes. For example, this is an excerpt from the Release Notes:

• • The Intel Inspector XE does not detect deadlocks or potential deadlocks created with:
    • Some types of locks via Intel C/C++ parallel extension (__critical) provided by the Intel Compiler Professional Edition 11.0.
    • Some types of locks in Intel TBB (spin_mutex, spin_rw_mutex)
    • Non-exclusive ownership synchronization objects involved, for example, condition variables, semaphores and events and reader/writer locks.
  • The Intel Inspector XE may not detect threading issues on data accessed in the C runtime library (like memmove and memcpy).
  • The Intel Inspector XE does not detect inter-process data races or deadlock/potential deadlocks.
  • The Intel Inspector XE may report false positives for analyzed applications using customized synchronization primitives. Use of _itt_notify to annotate your source code can reduce these false positives.

[David_A_Intel1]

BTW, check out this article I just published, based on your question.