We try to migrate a project from openmp and CUDA to dpc++ on remote Linux.
Firstly, we used a sparse matrix multiplication function to replace the old CUDA kernel. Then we profiled it by Vtune and we can see the function name.
However, we replace a piece of OpenMP code by parallel_for(dpc++), we found we cannot see the function name and there is only library name.
链接已复制
3 回复数
Hi,
Thank you for posting in Intel Communities.
For dpc++ you can use any debug parameters while compiling -gline-tables-only and -fdebug-info-for-profiling , so it will show the information
Please refer the documentation if needed:
(Or) If program is taking negligible amount of time to run, also cause this kind of issue.
Thanks
Shyam Sundar
