- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
Hi,
I am trying to use the experimental support for OpenMP Applications in the Flow Graph Analyzer to visualize my openMP tasks. I am able to generate output in a directory _fga_YYYMMDD_HHMMSS as described in the user guide, but when I try converting this output to the GraphML and TraceML formats with fgt2xml, I always get the warning "WARNING: 1 tasks do not have completions with task_end - ignoring tasks for thread(i) !" for all threads I been running, and while there are .graphml and .traceml files, in FGA I only see omp0::0 and if I double click it, I see omp0::n0::n1 and omp0::n0::n2 as for the empty parallel region in the example in the user guide.
This is true for complex programs but also for this simple hello world example where I definitely would expect to see the tasks in FGA:
int main()
{
#pragma omp parallel
{
int s;
#pragma omp single
{
#pragma omp task
{
std::cout << "Hello World!" << std::endl;
}
#pragma omp task shared (s) depend (out: s)
{
s = 42;
std::cout << s << std::endl;
}
#pragma omp task shared (s) depend (in: s)
{
s += 2;
std::cout << s << std::endl;
}
}
}
}
I do see in the per-thread view that the tasks are executed by different threads, but in the Graph Canvas they do not show up.
I would be very grateful for any hints on how I can get this to work.
링크가 복사됨
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
Thanks for reporting the issue. We will reproduce it locally and then give an update.
- 신규로 표시
- 북마크
- 구독
- 소거
- RSS 피드 구독
- 강조
- 인쇄
- 부적절한 컨텐트 신고
This appears to be a duplicate of another forum post (https://community.intel.com/t5/Analyzers/Visualizing-openMP-tasks-with-FGA/m-p/1587463#M24538). I will close this from Intel's perspective and track this issue on the forum post.
