Timo Heister [Mon, 2 Mar 2015 12:35:54 +0000 (07:35 -0500)]
Merge pull request #627 from kronbichler/master
Fix compiler warnings with clang 3.6. Comparing the address of a reference to an object with the zero pointer like in
&dof_handler.get_fe() != 0
will always evaluate to true. Thus, these assertions do not make sense and clang rightfully complains.
Comparing the address of a reference to an object with the zero pointer like in
&dof_handler.get_fe() != 0
will always evaluate to true. Thus, these assertions do not make sense and clang is right to complain.
Ross Kynch [Sun, 1 Mar 2015 19:10:19 +0000 (19:10 +0000)]
Fix issue when using the bundled umfpack - cmake moves all headers to a single directory where this header file includes a relative directory (../../AMD/Include)
David Wells [Sat, 28 Feb 2015 16:09:42 +0000 (11:09 -0500)]
Remove use of `DEAL_II_COMPILER_SUPPORTS_MPI`.
This macro was removed in favor of `DEAL_II_WITH_MPI` in early 2015;
see `doc/news/changes.h` for more information.
Timo Heister [Thu, 26 Feb 2015 14:14:11 +0000 (09:14 -0500)]
Disable SIMD vector operations for certain types
Disable long double SIMD instructions on ICC. This is to work around a
bug that generates wrong code at least up to intel 15 (see tests/lac
/vector-vector, tests/lac/intel-15-bug, and the discussion at
https://github.com/dealii/dealii/issues/598).
Timo Heister [Fri, 27 Feb 2015 21:14:48 +0000 (16:14 -0500)]
Merge pull request #618 from bangerth/remove-exceptions-without-argument-3
Remove assertion without argument.
Specifically, replace ExcNumberNotFinite with a way to provide both this number and a (lengthy) explanation of what may possibly have caused this problem.
Specifically, replace ExcNumberNotFinite with a way to provide both this number and a (lengthy) explanation of what may possibly have caused this problem. I see my students struggle with this error.
In this case, the exception is really an internal error that indicates an inconsistent state of our data structures. There isn't really a good reason to not use ExcMessage with some descriptive text.
In this case, the exception is really an internal error that indicates an inconsistent state of our data structures. There isn't really a good reason to not use ExcMessage with some descriptive text.
In this case, the exception is really an internal error that indicates an inconsistent state of our data structures. There isn't really a good reason to not use ExcMessage with some descriptive text.
Timo Heister [Tue, 24 Feb 2015 14:45:01 +0000 (09:45 -0500)]
remove compress without args from all PETSc sparse matrix types
- compress() without arguments was deprecated and now got remove
- replace redundant LastAction enum in by VectorOperation
- add additional compress() check.
In all classes of table.h the type Use AlignedVector::size_type is now
used for the calculation of indices for the values vector and the
indexing of elements in the values vector.
Note: This commit changes the interface of the n_elements function
in TableBase<N,T>
Ross Kynch [Tue, 14 Oct 2014 13:54:57 +0000 (14:54 +0100)]
Added project_boundary_values_curl_conforming_l2() to VectorTools for
use with FE_Nedelec elements in 2D & 3D.
This includes a test code included in tests/deal.II to check it works correctly.
This new function uses an L2-projection on edges and
faces (only edges in 2D, both for 3D) to compute
the constaints and factors in the residual from the edge
projection when computing the face constraints. The original
function project_boundary_values_curl_conforming() has
problems with non-rectangular faces, this one does not. The
function assumes that the face orientations are standard (e.g. a
sphere may cause problems due to mismatched face orientations).
The test shows that project_boundary_curl_conforming_l2() works on a
distorted cube (non-rectangular faces) for a simple test case
(polynomial function).
Use AlignedVector<T>::size_type as the integer type to carry out the
calculation of the index in the values AlignedVector object for the
various operator() and el() functions of Table<N,T> and its explicit
specialisations.
MFraters [Tue, 24 Feb 2015 10:32:07 +0000 (11:32 +0100)]
empty string in split_string_list fix
When an empty string or a string with only spaces is given to this function, the split_string_list function will ask in line 222 for name[-1], which will result in unpredictable behaviour. I propose to prevent this by adding a condition that the length of the string may not be 0.