Opened 4 years ago
Last modified 20 months ago
#48 new enhancement
Further efficiency savings in integral code
| Reported by: | tlatham | Owned by: | tlatham |
|---|---|---|---|
| Priority: | major | Milestone: | v4r0 |
| Version: | Keywords: | ||
| Cc: |
Description
As mentioned in comment:27:ticket:11 there are still some large savings that could be made in the recalculation of the integrals. These should be investigated and implemented if feasible.
Change History (4)
comment:1 Changed 3 years ago by dcraik
comment:2 Changed 2 years ago by tlatham
- Milestone changed from v3r1 to v4r0
comment:3 Changed 2 years ago by tlatham
Have separated the issue of narrow resonances in m12 into a new ticket #65.
comment:4 Changed 20 months ago by tlatham
Caching the full kinematics will require moving the per-event information data members currently in LauKinematics into a POD object in order to minimise memory usage. LauKinematics can then own one of those objects (or even a pointer to one in order to avoid copying?). Should also try and reduce the memory consumed by other parts of the code - ticket #14 should help with this. So these tickets should probably be addressed together.
In addition to caching the full kinematics, it would help if Laura could automatically choose the axes of the integration DP (independently of the efficiency/background (S)DP) to avoid putting narrow resonances on the diagonal whenever possible (with a method to manually force a particular DP if desired by the user).
For the D* in B0->D0Kpi with default integration settings, changing the order of the daughters can currently lead to a ~5-10x change in CPU time.