mountaindust
commented
6 months ago Will need to:
- Keep track of all of the time points represented by the data
- Keep track of time points currently loaded
- Implement property getter for fluid that checks availability and loads data on demand
- Implement lazy disc read instead of reading in all the data and then replacing with spline like we are now
While the memory management updates in v0.5.2 helped, the real problem is that spline objects are huge, and creating a spline across all of the temporal fluid velocity data (potentially many time points) is extremely expensive to store. The result is that 3D temporally varying fluid velocities are currently impractical to work with unless the number of time points is small (or the fluid grid is coarse enough).
The most obvious solution is to calculate a spline consisting of only one polynomial - the currently needed one. Since we are using cublic splines, this will be the spline using the two data sets on either side of the current time (four in total). It then gets recalculated every time the calling function advances past a time point with fluid data associated with it.