#4741 closed defect (fixed)
analyzeInitialSystem scales as O(N^3) when NF is used
| Reported by: | Francesco Casella | Owned by: | Per Östlund |
|---|---|---|---|
| Priority: | high | Milestone: | 2.0.0 |
| Component: | New Instantiation | Version: | |
| Keywords: | Cc: |
Description
Please consider the ScalableTestSuite.Thermal.Advection.ScaledExperiments.AdvectionReaction_N_XXX models in the ScalableTestSuite.
The times spent by the analyzeInitialSystem function of the back-end when the old FE is used are roughly linear with N and have a marginal impact on the total build time. If you check the times spent when the NF is used, you get these values:
| N | Time (s) |
|---|---|
| 400 | 0.08 |
| 800 | 0.61 |
| 1600 | 4.74 |
| 3200 | 40.50 |
It seems that for some reason if the backend is handed the model by the NF it scales very badly. Please check.
Note:
See TracTickets
for help on using tickets.
Fixed in 58a87d01. The issue was that the scalarization module accidentally paired type attributes up with the wrong expression. So e.g.
Real x[1](start = 0.0, fixed = true)would becomeReal x[1](start = true, fixed = 0.0)after scalarization, which obviously caused a few issues.