how to specify custom synchronization primitives?

[joewhaley]

Hello,

I am trying out Intel Parallel Inspector to look for data races in our C++ codebase. The tool identifies a number of false "races", most likely because it does not seem to understand the synchronization primitives. The code uses Boost C++ synchronization, specifically boost::mutex, boost::shared_lock, boost::unique_lock, and boost::upgrade_lock. It also has some custom rwlocks.

How can I specify the lock functions so Parallel Studio won't complain about data accesses that are correctly protected by the locks? I read something about ITT_NOTIFY() but I couldn't find any documentation or examples.

[David_A_Intel1]

Hello joe!

You have correctly analyzed the issue! By default, the Parallel Inspector recognizes all native threading primitives. Our previous generation of products enabled support for custom threading implementations via _itt_notify APIs. For backward-compatibility, Intel Parallel Inspector supports these legacy APIs.

Regarding documentation of the _itt_notify APIs, they are part of the Thread Checker and Thread Profiler products and are documented in those products. Please see the Thread Checker documentation for details.