Intel® Moderncode for Parallel Architectures
Support for developing parallel programming applications on Intel® Architecture.

VTune Loop duplicates

Paulius_V_
Beginner
201 Views

Hello all.  Can anyone explain why my vtune profiling run shows the same loop twice?

vtune.PNG

0 Kudos
3 Replies
TimP
Black Belt
201 Views
You may have multiple versions and remainder loops. opt-report4 should clarify.
Paulius_V_
Beginner
201 Views

That would make sense, thank you. What I find confusing, however, is that the number of clockticks spent is significant for both of them and the CPI is nearly double for one version. In case of multiversioned and peel/remainder loops it would make more sense of the bulk of the time would be spent in a single version. What decides which version will be used?

 

TimP
Black Belt
201 Views
If you look in opt-report and find different assumptions about alignments, that might shed light. You may have aligned and unligned on alternate executions. #pragma vector unaligned might give you a single version. If you have a remainder loop reporting high cpi it may not mean much unless time spent there is significant. If you run advisor it may flag any remainder taking significant time and produce suggestions regardless of whether it is already efficient.
Reply