GitLab

Feature/field convert homogeneous plane

This MR creates a module to FieldConvert that extracts one of the planes of a 3DH1D expansion (with the option of getting it in wavespace). This way we can easily obtain slices along the z-direction.

The meanmode module is also modified to use this new module, avoiding duplicate code.

See merge request !542

Add MatShell functionality to PETSc wrapper

This MR extends the PETSc global linear system classes to use MatShell and PCShell objects to wrap Nektar++ matrix-vector multiplication and preconditioners respectively. This therefore allows us to use PETSc's KSP objects, giving access to a wide range of iterative solvers, without explicitly constructing a PETSc sparse matrix.

Other minor additions:

- Generally improve documentation and formatting of the PETSc code
- Adds support for a Neumann boundary condition to linear elastic solver
- Updates PETSc cmake files to latest version

See merge request !537

Fix/compressed io mesh partition

This MR fixes an issue with compressed I/O format caused by a lack of initialisation of the `c.ptid` variable inside `MeshPartition`.

See merge request !547

Modified call to check on COMPRESSED attribute from a ->Attribute call to a ->QueryStringAttribute call in TinyXML

Corrected for missing initialisation of c.ptid and modified the check for COMPRESSED data from a Attribute read to a QueryStringAttribute

Fix/floquet

This MR primarily fixes the Arnoldi algorithm when using the HalfMode configuration. The main problem was a missing sign in the base flow interpolation and the default noise level being far too high and polluting the eigenspace.

This branch also adds a time-dependent component to the body forcing and adds an alias to body forcing called field forcing.

See merge request !536

Conflicts:
	CHANGELOG.md

Fix/3dh2d HelmSolve

This fixes a bug in 3DH2D where HelmSolve is being called with the wrong index for m_lines. This leads to wrong boundary conditions being imposed on the field.

The channel flow test was passing because in that case all boundary conditions are zero. Therefore, a test for the Couette flow (which fails without this fix) was included. Also, minor changes regarding indentation and duplicate code were made. 

See merge request !545

Feature/driver adaptive

This MR introduces a driver which uses an adaptive polynomial order procedure for 2D and 3DH1D expansions. 

I only tested it for the incompressible solver, but I suppose it might work with other solvers as well. Extending it to 3D should be simple, but I did not attempt that since the incompressible solver does not support variable order in 3D yet. 

A bug in ExtractCoeffsToCoeffs is also fixed by this branch.

See merge request !538

Documentation updates

This MR includes a variety of general updates, fixes and improvements to the documentation. Also adds the logo to the user guide and developers guide documentation.

See merge request !523

Fixes for prism refinement from XML input

This MR addresses issue #7 and fixes boundary layer splitting with XML input. Specifically, it fixes prismatic input, since hexahedra _seem_ to work. There are two changes:

* MeshConvert by default reorients elements to align collapsed coordinates. `InputNekpp` uses this, and since the collapsed coordinates are presumed to be aligned by construction, edges are not reprocessed. This therefore breaks element connectivity when virtually any processing is done on the mesh that alters the connectivity.
* Address a bug in `ProcessBL` which causes the geometry to subtly change as elements are refined. This is addressed by constructing a map of the existing elemental geometries before any processing occurs.

See merge request !535

This branch fixes some DG fluxes, particularly LaxFriedrichs.

This fixes the LaxFriedrichs flux for the CompressibleFlowSolver. A slight alteration in Roe's flux has also been done. Both have been test in 2d and 3d cases and seem to be working fine now.

See merge request !546

New functions to work with boundaries

This MR creates new functions in MultiRegions to help in performing calculations on boundaries (pressure boundary condition, post-processing,...). The most important of these is GetBndElmtExpansion (we may need better names for these functions), which allows us to obtain an expansion (of the same dimension as the solution) containing only the elements in contact with one of the boundary condition expansions. Other functions are implemented to moving a phys array from the original expansion to this auxiliary expansion, and from the auxiliary expansion to the boundary condition expansion.

Some advantages of this approach are:
- Reduce the need for expansion specific code in solvers and utilities
- Encapsulate low-level operations (using BoundaryToElmtMap, looping through elements, etc) in Multi Regions
- It is important for Homogeneous expansions. Since we don't have homogeneous elements, the current approach of looping through elements is very hard to implement when we need derivatives in the homogeneous direction.

These new functions were then applied to the wss module of FieldConvert, resulting in a simpler code that also works for 2D and 3DH1D (I may have reversed the orientation of the wss, because this is the way it makes sense to me. I do not mind changing it back if this affects someone). This could also be applied to Extrapolate.cpp (for the pressure boundary conditions), but I suppose it is better to do this in another branch (or leave it as it is).

See merge request !539

Conflicts:
	CHANGELOG.md

Fix formatting in DriverAdaptive, move GenerateSeqString to ParseUtils and replace code elsewhere in the library

Fix/variable p restart

This makes Vmath::Scatr consistent with the way we are dealing with variable P expansions in CG. This bugs was messing up the initial conditions of variable P simulations.

A new test is also included for this. 

See merge request !543