OpenMP and TBB Task Graphs: Unraveling the Spaghetti with Flow Graph Analyzer

With parallel programming models here to stay, task-based programming has become increasingly common in both mainstream and computationally complex applications. Programming models such as Threading Building Blocks (TBB) flow graph API, OpenMP tasking API, or specialized models—SYCL*, OpenCL* and OpenVX*—give programmers the right level of abstraction to build algorithms that take advantage of task-based programming and realize the performance of the underlying system.

In this talk, we present Intel® Advisor’s Flow Graph Analyzer (FGA)—a powerful feature that uses task graphs to help you visually analyze your parallel applications, either explicitly through TBB flow graph API or implicitly through OpenMP depends clause.

You will:

  • Get an overview of Intel® Advisor FGA
  • See relevant demonstrations, including using the feature to optimize the structure and performance of computational graphs
  • Learn how to capture traces from running applications
  • Understand the scalability exhibited by the graph
  • Calculate your application’s critical paths
  • And more

Intel Advisor is available as part of Intel® Parallel Studio XE and Intel® System Studio suites of development tools. Try them both free now.

Vasanth Tovinkere, Software Engineer, Intel Corporation

Vasanth Tovinkere is a software engineer in Intel’s Developer Products Division. With the company for over 20 years, his work has spanned an impressive gamut—from building research prototypes and products for Wall Street multi-processor “early adopters” to developing automatic semantic event detectors for digital sports technologies (for which he holds a patent). Vasanth is currently responsible for exploring heterogeneous and distributed compute models and new visualization approaches to performance tuning/debugging, and he is the architect of Intel® Advisor Flow Graph Analyzer. Prior to joining Intel, he was involved in the development of automated fuzzy pattern recognition algorithms for NASA’s Mission to Planet Earth Program.

Performance varies by use, configuration, and other factors. Learn more at