The previous implementation declared a static array and looked up
values in functions. This leaves the compiler to manage to data.
That can have potential advantages such as putting data in
faster constant data. But it could also cause problems if the
compiler has problems managing all the data.
The switch to ArrayHandles simplifies the implementation for
the compiler by managing these static arrays on the host and
using the existing array management to make sure everything
is correct.
The implementation of `Vec` made it difficult to use within a `constexpr`
function. For example, the following would not compile.
``` cpp
constexpr vtkm::UInt8 OctFlip(vtkm::UInt8 value)
{
vtkm::Vec<vtkm::UInt8, 8> flips = { 0, 6, 2, 6, 1, 5, 3, 7 };
return flips[value];
}
```
The problem was that `flips` could not be initialized in a `constexpr`
function because the constructer was not `constexpr`.
This change fixes this problem by _removing_ the constructor that takes a
`std::initializer_list`. The problem with `std::initializer_list` is that
it cannot be used to directly initialize an array and the loop to do the
initialization cannot be used in a `constexpr`.
With the removal of this constructor, construction with a braced list
should go to the constructor that takes a variadic number of parameters.
Note that it is possible that some initialization of `vtkm::Vec` could
break. You can no longer construct an `std::initializer_list` and pass it
to a `vtkm::Vec` constructor. This seems like a less likely case and can be
gotten around by just copying the data yourself. It is also possible that
putting a braced list in parentheses will stop working. This can be fixed
by just deleting the parentheses.
124c58eb4 Minor fixes for distributed contour tree example
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Kenneth Moreland <morelandkd@ornl.gov>
Merge-request: !3217
The `ExtractGeometry` filter was outputing datasets containing
`CellSetPermutation` as the representation for the cells. Although this is
technically correct and a very fast implementation, it is essentially
useless. The problem is that any downstream processing will have to know
that the data has a `CellSetPermutation`. None do (because the permutation
can be on any other cell set type, which creates an explosion of possible
cell types).
Like was done with `Threshold` a while ago, this problem is fixed by deep
copying the result into a `CellSetExplicit`. This behavior is consistent
with VTK.
The Intel compiler by default turns on an optimization that assumes that
all floating point values are finite. This breaks any ligitimate uses of
non-finite values including checking values with functions like `isnan`
and `isinf`. Turn off this feature for the intel compiler.
Using `${OpenMP_CXX_LIBRARIES}` results in paths to certain libraries
being hard-coded in VTKmTargets.cmake, which can lead to issues in some
cases (e.g.: when VTK is built by the CI system for MSYS2).
b7ef24678 fix assumption that ANARIMapperVolume only consumed point data
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Kenneth Moreland <morelandkd@ornl.gov>
Merge-request: !3207
The `CellMeasures` and `MeshQuality` filters had constructors that took the
metric that the filter should generate. However, this is different than the
iterface of the rest of the filters. To make the interface more consistent,
these filters now have a default (no argument) constructor, and the metric
to compute is selected via a method. This makes it more clear what is being
done.
In addition, the documentation for these two classes is updated.
59c545c15 add test and fix cuda long compiling issue
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Abhishek Yenpure <abhi.yenpure@kitware.com>
Merge-request: !3188
The `reproduce_ci_env.py` script reads in the yaml for the GitLab
runners and replicates the runs using Docker. Through a quirk of the
implemenation of MR !3191, some of the scripts, which are expected to be
a list of command strings, get defined as a list of lists. This is fixed
by flattening this list of commands.
Doxygen was having trouble with `@copydoc` when copying documentation of
templated functions. The doxygen comments in `Math.h` is restructured to
not need `@copydoc`, and the documentation is generated correctly.
