Nans for gll nodes
We observe nans for the recently implemented gll nodes as soon as we work with amr on supermucng. We start with a series of tests:
| Basis | Mesh | AMR | Arch | Opt | MPI | Works |
|---|---|---|---|---|---|---|
| GLL | 25 | 1 | hsw | vect | No | yes |
| GLL | 7 | 2 | hsw | vect | No | yes |
| GL | 25 | 1 | skx | vect | Yes | yes |
| GLL | 25 | 1 | skx | vect | Yes | yes |
| GL | 25 | 2 | skx | vect | Yes | nans |
| GLL | 25 | 2 | skx | vect | Yes | nans |
| GL | 25 | 2 | skx | gen | Yes | nans |
| GLL | 25 | 2 | skx | gen | Yes | nans |
| GLL | 79 | 1 | skx | gen | Yes | yes |
| GLL | 79 | 1 | skx | gen | Yes | yes |
| GLL | 239 | 1 | skx | gen | Yes | yes |
As GL as well as GLL nodes seem to fail repeat the GL 2 AMR level test on the master branch:
| Basis | Mesh | AMR | Arch | Opt | MPI | ExaSeis | Works |
|---|---|---|---|---|---|---|---|
| GL | 25 | 2 | skx | gen | Yes | master | nans |
Which shows the same behaviour for default old cases.
Edited by Leonhard Rannabauer