Feature #2336
Make reflectometry computation as fast as in Refnx
| Status: | Resolved | Start date: | 04 Apr 2019 | |
|---|---|---|---|---|
| Priority: | Low | Due date: | ||
| Assignee: | - | % Done: | 0% | |
| Category: | Reflectometry | |||
| Target version: | - |
Description
During our last group meeting, we decided to carry out several performance tests:
- Time scaling with the number of layers
- Time taken to run a simple simulation
- Time scaling with the simulation resolution
The results are not satisfactory as can be seen from the following notebook:
As this issue is not a priority for the moment (is it?), I suggest to keep further investigations and possible fixes on the backlog.
However, looking ahead for a possible fix, I did some experimenting on reusing an already performing piece of code and embedding it inside a python API --which could also be a GUI for that matter:
Related issues
History
#1 Updated by wuttke over 1 year ago
- Tracker changed from Testing to Feature
- Subject changed from BornAgain performance is not optimal to Make reflectometry computation as fast as in Refnx
- Status changed from New to Backlog
- Priority changed from Normal to Low
#3 Updated by dmitry 9 months ago
- Related to Refactoring #2343: Improve performance in reflectometry engine added
#4 Updated by dmitry 9 months ago
- Related to Refactoring #2410: Speeding up computations on samples with a large number of layers added
#5 Updated by wuttke about 2 months ago
- Status changed from Backlog to Sprint
#6 Updated by wuttke about 2 months ago
- Status changed from Sprint to Resolved
The fundamental problem with n log n from the bijection is solved by Randolf's new algorithm. Everything else will be reinvestigated as need arises.