Segmentation Fault in FGA

[zeronsix]

Hi!

I'm using the fgt_run.sh script to generate profiling traces of my oneTBB application for Flow Graph Analyzer. The script successfully generates *.graphml and *.traceml files.

However, I get segmentation fault when I try to open my_application.graphml with FGA.

/run_fga.sh: line 24: 4167772 Segmentation fault      (core dumped) ${DIR}/fga &> /dev/null

I use FGA version 2020 1.1.17998 and oneTBB 2021.2.0.

Please find the archive with the collected traces attached.

[JananiC_Intel]

Hi,

Thanks for posting in Intel forums.

We are checking on this. Meanwhile could you share a sample reproducer and your OS details?

Regards,

Janani Chandran

[zeronsix]

I run the application inside nvidia/cuda:11.3.1-cudnn8-devel-ubuntu20.04 Docker container, meanwhile FGT is installed inside the container from intel-oneapi-advisor package from https://apt.repos.intel.com/oneapi repository. oneTBB is built from the source code as a CMake target along with the project itself.

Unfortunately, it is hard to share a sample reproducer without unveiling the key proprietary technologies and approval in our legal department could take ages.

[zeronsix]

I've managed to find the source of the problem. The flow graph was wrapped with a composite node. The output node of the graph is a tbb::flow::function_node<Result, tbb::flow::continue_msg, tbb::flow::queueing>, so the composite node had tbb::flow::composite_node<std::tuple<Input>, std::tuple<tbb::flow::continue_msg>> type. When I moved the output node outside of the composite node, FGA stopped crashing.

[JananiC_Intel]

Hi,

Glad to know that your issue is resolved. Thanks for sharing the solution with us. If you need any additional information, please post a new question as this thread will no longer be monitored by Intel.

Regards,

Janani Chandran