GitLab

Advance tutorials submodule for v4.3.2.

This MR just advances the submodule commit ID for the tutorials.

See merge request !643

Fix instances of MeshConvert in user guide.

This MR includes some minor fixes to the user guide.

See merge request !640

Fix/parallel assert

This MR fixes a bug in the iterative solver whereby the assertion on exceeding the maximum number of iterations is only effected on the root process. This also updates the NekError class to print messages on the root process only, rather than on all processes.

See merge request !636

Modify FC extract checks to allow use without fld

This MR just changes one of the error checks in the extract module of FieldConvert, since it is only required to have an .fld input when using the fldtoboundary option. Without this option, we can use only an .xml file as input, for example to calculate the normals to the surface.

See merge request !638

Fix wss module with partitioned domains

This fixes a bug which prevents the wss module from working in parallel. This happens because in parallel the index from the boundary expansions do not necessarily correspond to the boundary region id.

See merge request !635

FieldConvert module for combining average fields

This MR introduces a new FieldConvert module that combines fields obtained through `FilterAverageFields` or `FilterReynoldsStresses` (the Reynolds stresses are added correctly, what is not possible using only `addfld`).

The number of samples in each field is automatically considered and the total number of samples is stored in the output file, allowing for multiple files from simulations with different numsteps to be combined easily. 

See merge request !620

Fix condition for output with IO_CheckSteps=1

This fixes a small bug in UnsteadySystem which caused the first checkpoint to be skipped when using IO_CheckSteps=1.

See merge request !639

Switch to Fourier points in FieldConvert with homogeneous expansions

This MR switches the point distribution of homogeneous expansions in FieldConvert to FourierEvenlySpaced.

The main reason for that is a performance issue recently reported when the number of Fourier modes is large. This was caused by the expansion creating a PolyEvenlySpaced distribution with many points, which scales very poorly due to the derivative matrix (for 1000 points this takes around 2 hours). This cost is also independent of the number of mpi processes...

This is also related to issue #10.

See merge request !627

Add option to use SVV only in the spectral/hp expanions

This MR adds a new option `SpectralVanishingViscositySpectralHP` that activates SVV in the spectral/hp expansion, but not in the homogeneous direction (the option `SpectralVanishingViscosityHomo1D` was already available).

The behaviour when using the option `SpectralVanishingViscosity` remains the same.

See merge request !623

Fix minor performance issue in TimeIntegration

This fixes a small mistake in TimeIntegrationScheme, which in some cases lead to a duplicate calculation (the speed up from this fix is relatively low).

See merge request !632

Fix FilterCheckpoint

This fixes a bug with parsing the `OutputFrequency` parameter of FilterCheckpoint.

I also made checkpoint 0 correspond to the initial condition, so that it is consistent with the behaviour from IO_CheckSteps.

See merge request !633

Spencer Sherwin authored May 26, 2016 and Chris Cantwell committed May 27, 2016

First attempt to put a root only error message but believe this will not work if an ASSERT is called before MPI_Init