I have a very powerful machine (80 cores).
When I run VTune, after the collection is over, VTune then spends a lot of time resolving symbols (even for short collection runs).
It is using only 2 cores out of 80 available.
Is it possible to configure VTune to use more cores and speed up the post-processing?
Thank you for posting in Intel Communities. We would like to let you know that there's no such option in VTune to parallelize the finalization which speeds up the post-processing. As of now, you can explore the finalization mode options(Full, Fast, Deferred, None)
Finalization is the process by which Intel® VTune™ Profiler converts the collected data to a database, resolving symbol information, and pre-computes data to make further analysis more efficient and responsive. VTune Profiler finalizes data automatically when data collection completes.
VTune Profiler provides three basic finalization modes:
- Full mode is used to perform the finalization on unchanged sampling data on the target system. This mode takes the most time and resources to complete, but produces the most accurate results.
- Fast (default) mode is used to perform the finalization on the target system using algorithmically reduced sampling data. This greatly reduces the finalization time with a negligible impact on accuracy in most cases.
- Deferred mode is used to collect the sampling data and calculate the binary checksums to perform the finalization on another machine. After data collection completes, you can finalize and open the analysis result on the host system. This mode may be useful for profiling applications on targets with limited computational resources, such as IoT devices, and finalizing the result later on the host machine.
- None option is used to skip finalization entirely and to not calculate the binary checksums. You can also finalize this result later, however, you may encounter certain limitations. For example, if the binaries on the target system have changed or have become unavailable since the sampling data collection, binary resolution may produce an inaccurate or missing result for the affected binary.
Please refer this documentation on how to select the finalization-mode.
Regarding your suggestions on adding more parallelism to VTune's post processing, we have provided your feedback to the relevant team. At this moment there is no visibility when it will be implemented and available for use. Hence, we are closing this case.