Skip to content

Avoid calling qpg_iScalar in PGet0

YoshioNishimoto requested to merge YoshioNishimoto/OpenMolcas:avoid_qpg_iscalar into master

Unless I completely misunderstand something, I found that calling qpg_iScalar in PGet0 (specifically, PGet1_RI2, PGet1_RI3, and PGet2_RI2)degrades the performance of gradient calculations with the RI/CD approximation significantly. I suggest these calls are done only once outside loops. The following is an example of the performance improvement (CD-UHF/cc-pVTZ for benzene cation with six processes):

with qpg_iScalar call:

--- Module alaska spent 1 minute 43 seconds ---

without qpg_iScalar call:

--- Module alaska spent 26 seconds ---

This small change improves the performance a lot (more than I could imagine).

  • Documentation updated or not needed.
  • Test created or not needed.
  • Attribution in source files and CONTRIBUTORS.md updated if necessary.

Merge request reports