Implement calls to cuBlas, cuSparse and implement necessary
CUDA kernels to perform a single iteration of the jacobi preconditioner.
Add tests that verify new kernels and the preconditioner in its totality.
The preconditioner is verified on 2x2 and 3x3 blocks, which as of now
are the only supported sizes. 1x1 are not supported because cuSparse
does not support it.
The NLDD solver would always take one global non-linear (Newton)
iteration before starting the local non-linear iterations. This
commit introduces a new command-line parameter,
--nldd-num-initial-newton-iter (NlddNumInitialNewtonIter)
which allows the user to configure this value at runtime. The
default value, 1, preserves the current behaviour.
The connections entered in the EDITNNC keyword are already fully
accounted for in the main 'update()' loop followed by the helper
function applyEditNncToGridTrans_(). Therefore, it is wrong to
include those connections when applying MULTREGT to NNCs. This
commit removes the pertinent connections from the main loop in
helper function applyNncMultreg_().
Commit ddcafa8a91 got this wrong, so pointy hat to [at]bska.
Due to issues with anonymous enums, newer versions of {fmt} (v10) do not seem to handle the implicit conversion from enum value to `int`. Hence we need a cast for `fmt::format`-calls involving `Matrix::block_matrix::rows` (or `cols`) to compile. This seems to only be relevant for one part of the code.
For an isolated example, see https://github.com/kjetilly/fmt_fails_with_enum
We did only break in the loop for rank 0 and not the other ones. Hence
all other processes kept iterating beyond the maximum number of
allowed iterations. This lead to hard to find crashes because of
non-matching MPI communication.
also:
-- fixed but not checked for diffusion
-- bug for zero thermal diffusion (nan in derivatives)
-- added checking of consistency between input and taged phases
when local_well_solver_control_switching_ is off
and incoporating commenst regarding resetting wellStatus_ in StandardWell within
the function iterateWellEqWithSwitching.
when we do the local solve for well equations, control/status will be
updated during the iteration process, such that the converged well gets
correct control/status regarding to the current reservoir state.
various change in the other parts of the code were made to make the
function work as intended.
This commit switches the region set tag matching algorithm to using
unique prefixes. This enables the simulator to recognise that the
region set name
FIPUNI
should match up with the user defined region set 'FIPUNIT'. In the
current master sources, the above summary vector would produce a
diagnostic message saying that the region set 'FIPUNI' (without the
final 'T') does not exist.
To this end, we instruct the ParallelEclipseState to always pass the
six character substring beginning with 'FIP' for FIP-like region
arrays and defer to the rank-0 EclipseState/FieldProps mechanism to
match this prefix with its canonical region set.
