Michael Turner authored 8 years ago and Dave Moxey committed 7 years ago

(cherry picked from commit 0fe0983b)

(cherry picked from commit 4cf3f60c)

Michael Turner authored 8 years ago and Dave Moxey committed 7 years ago

(cherry picked from commit 0a7d1468)

Fix documentation issues.

This MR just fixes a few minor errors in the doxygen docs and adds an extra pdflatex run to the user guide to ensure references are properly included.

See merge request !749

Switch to MPI_lang_... variables

This MR switches the CMake configuration to use `MPI_<lang>_...` variables instead of the existing `MPI_...` variables. Currently Nektar++ uses `MPI_LIBRARY` and `MPI_EXTRA_LIBRARY`, which have been deprecated for some time. In recent versions of CMake, at least on OS X, `MPI_EXTRA_LIBRARY` is no longer set, meaning that Nektar++ fails to configure if MPI is enabled. A similar fix has also been applied to gsmpi which has the same issue.

See merge request !725

Fix: Dont read the same files again for every variable

This branch adds a check to prevent EquationSystem from reading the same fld/pts files again for each variable. This gives a huge speedup in some cases.

I also split the EquationSystem::EvaluateFunction method into three new methods: one for expressions, one for fld files and one for pts files to make the code more readable.

However, i think its still a mess with lots of duplicate code and even more things that would make much more sense in some other class. Even worse is that the Forcing base class holds a complete copy of this code.  I would like to move the whole Function logic from Forcing and EquationSystem into a new class that lives in SolverUtils in the future. Should we discuss that here or do you want me to write to the mailing list?

See merge request !670

fix/3d-boundary-layers

Recent merge of feature/mesh_spacing broke a number of parts of the mesh generation, mostly do with orientation.

The CAD orientation routines have now been rewritten in this branch and the relevant code updated.

3 more regression tests, which didnt work prior to this fix, have been added

A small amount of wall time optimisation has been done to the OCC cad interface and the reduction of some L0 asserts to L1.

See merge request !744

removed redundant m_loadedFields conditional. The case that we try to load a non-transient file again is now handled by the
m_loadedFldFields and m_loadedPtsFields logic.

tidy/warnings

This gets rid of all the compile warnings for the buildbot "full" equivalent on ubuntu 16.04, mac and the compute nodes.

See merge request !743

Fix/PetSc_mkl

this fixes blas/lapack detection of the PetSc configuration for systems which have no native blas/lapack installed

When native blas/lapack is enabled, we use this for PetSc (just like before, now we just do it explicitly and not by accident) , when MKL is enabled we use MKL and for all other cases, PetSc will download its own blas/lapack and compile it itself. This might be extended to ACML and other blas/lapack providers in the future but i can only test native and MKL on my systems.

See merge request !740

feature/thirdpartylibs

Add thirdparty installs to fortran compiled libraries.

Done:
blas/lapack

See merge request !732

Fix/fc interpolator

Provide a fix for the interpolator in FieldUtils which currently using the wrong offset to place interpolated points back into the new field. This is mainly a problem when quad elements are listed before tri elements since this automatically leads to the incorrect offset. This would have not been a problem if tris went before quads. This MR also therefore adds in a regression test for this case.

See merge request !746

Cell model translator for cardiac solver.

This adds a python utility for generating cell model code from CellML models.

See merge request !723

Fix/test fail dump output

This MR fixes the issue of the tester not producing output when a test executable segfaults or other system error occurs.

It also adds an option to generate a hostfile which explicitly specifies the maximum number of required slots when running MPI executables, since newer versions of OpenMPI now stop jobs requesting more than the number of available CPU cores by default (which causes our buildbot to fail).

See merge request !745