GitLab

Fix/petsc finalize

This MR fixes the issue on latest OpenMPI (as found on Fedora 21) which produces errors when using MPI that MPI_Attr_get() function is called after MPI_Finalize(). This is because we were previously not calling PetscFinalize(). The PetscInitialize and PetscFinalize calls have been moved to CommMpi to ensure they are called at the right time and in the right order.

Dave should probably review this, since he implemented the PETSc stuff.

See merge request !456

Fix/mesh convert star input

This fix sorts out a problem where if you have a very deformed prism the orientation of the quadrilateral faces can get confused with regard to the edges. Instead of using a normal orientation method to decide how to orientate edges I have used a search based on vertex node ids which has corrected the current issue. There are some other elements that use normal orientation but currently I do not know how we might change these.

Jean and Dave are good to review this fix.

See merge request !455

APE allow for vaiable speed of sound

This MR turns the base density rho0 into a field to allow for spatially varying rho0 and changes the APE implementation to match the original APE1/4 LHS which allows for correct treatment of rho0 and p0 gradients. All test and example cases as well as the documentation were adjusted in order to reflect these changes. Existing constant p0/rho0 cases are not affected.

See merge request !438

Replaced the check for how to orientate the faces of a Quadrilateral base face in the Prism. Previously thi was using a check on the normal to a set of two edges. However for highly deformed prisms this may have been failing (or indeed could have had a bug in it anyway). The new check uses the vertex ids to identify if other quad faces in the prism share two vertices with the edge 1 of the base quad. If so then we know the orientation and if not the vertex ordering if flipped.

There is still a check in the code for the orientation of the 0 and 1 vertices based on a normal calculation. Currently this is not causing any problems.

Fix/cardiac stimulus constructors

This MR tidies up the constructors of the Stimulus classes.

See merge request !446

Fix/empty bc exp

This MR fixes a segmentation fault when requesting coordinates for an expansion with no elements.

See merge request !445

Fix consistency issue with FFTW and MVM approaches

This MR fixes inconsistencies between the FFTW and MVM discrete Fourier transforms which were leading to post-processing issues. Additionally, a small bug has been fixed which was leading to the incorrect application of body force terms and in third-party compilation.

See merge request !444

Fix/hobcs

See merge request !447

Updated documention to make normals bold and add a comment to the Kovasnay example to highlight that the use of VALUE is not typically known

Fix/mkl

this fixes the detection of the MKL paths

as the build slaves dont use MKL, running this through buildbot is pointless but i tested this on three systems (openSUSE 13.2 with intel 2013, debian with intel 2013 and SUSE enterprise with Intel 2015).

See merge request !453

Fix/cardiac cell restart

This MR fixes a bug which prevented cell model restart files generated in parallel from being imported correctly. It also optimises the loading of cell model restart files so each file is only read once, rather than once per variable.

See merge request !450

* check the $MKLROOT env variable in addition to $MKL_HOME
* dont use relative paths to find the libdir. this breaks when $MKLROOT is a symlink
* the architecture dependent subdirs are called intel64 and ia32 on most systems
* mark the variables as advanced

Fix curved element normals

This MR fixes some issues with curved element normals, mostly relating to the DG routine `ExpList2D::v_GetNormals`. Additionally, the routines in the 3D element expansions for curvilinear normal generation use an incorrect formulation, where the face Jacobian is not extracted properly for the purposes of projection onto the polynomial space. Both of these issues are corrected in this MR.

See merge request !443

Fix/meshpartition field io

This MR deals with two issues. First, in some situations MeshPartition writes two coordinates with no spaces between them, making the xml file invalid. The second issue is related to metadata not being imported for results from parallel simulations, and not being available in FieldConvert.

See merge request !449