a9e64c4b FloatPointReturnType is float if 'T' is < 32bytes instead of being double.
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Kenneth Moreland <kmorel@sandia.gov>
Acked-by: Thomas Otahal <tjotaha@sandia.gov>
Merge-request: !1048
In trying to give error diagnostics with template definitions of invalid
types, the user encounters some pretty confusing error messages at
first. There is no way to get the compiler to give exactly the
diagnostics we want in a nice readable error message, so we are putting
some verbose instructions as comments in the code. However, a user might
not know to look at the source code since the error happens deep in an
unfamiliar (and complicated) class. Thus, add (yet another) error at the
front that gives a (hopefully) clear indication to look at the source
code for help in understanding the errors.
When one of the parameters to DispatcherBase::Invoke is incorrect, you
get an error in a strange place. It is deep in a call stack using some
heavily templated types rather than where the Invoke is actually called.
Formerly, the code was structured to give a very obfuscated error
message. Try to make this easier on users by first adding helpful hints
in the code around where the error is to explain why the error occured
and helpful advice on how to track down the problem. Second, reconfigure
the static_assert so that the compiler gives a readable error message.
Third, add some auxiliary error messages that provide additional
information about the types and classes involved in the error.
Previously FloatPointReturnType would always be double for types that
are not float, which caused Int8/Int16 types to promote to double instead
of float.
1. Add option to copy user supplied array in make_ArrayHandle.
2. Replace Field constructors that take user supplied arrays with make_Field.
3. Replace CoordinateSystem constructors that take user supplied arrays with
make_CoordinateSystem.
If a global static array is declared with VTKM_EXEC_CONSTANT and the code
is compiled by nvcc (for multibackend code) then the array is only accesible
on the GPU. If for some reason a worklet fails on the cuda backend and it is
re-executed on any of the CPU backends, it will continue to fail.
We couldn't find a simple way to declare the array once and have it available
on both CPU and GPU. The approach we are using here is to declare the arrays
as static inside some "Get" function which is marked as VTKM_EXEC_CONT.
The speed improvement fix regressed support for non scalar types, this
correct that issue.
The issue was found while trying to bump the VTK-m version inside VTK.
32148cdb A first pass at improving the compile time of MarchingCubes
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Matt Larsen <mlarsen@cs.uoregon.edu>
Merge-request: !989
8fabece1 Use median point from cluster as representative vertex.
c7bf0c95 Compute PointIdMap while reducing cluster ids.
5dee7c6a Select input point from cluster rather than averaging.
28e76ddb Update vertex clustering benchmarking code.
e3c9e7bb Optimize cell map computation.
d7669650 Use requested grid in VertexClustering worklet.
0472dc11 Fix warning on Cuda.
3f4e17e2 Add field mapping to VertexClustering.
...
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Robert Maynard <robert.maynard@kitware.com>
Merge-request: !960
This is a bit counterintuitive, but choosing a random point from each
cluster rather than averaging them gives a better visual result. The
averages poorly represent an surface that runs through the grid block and
tends to bias the output points towards the center of each block, creating
very noticeable grid artifacts that look blocky.
This is to match the default behavior of vtkQuadricClustering. If we
want to add this functionality back, it should go into the filter as
an option that adjusts nDivisions before calling the worklet.
Rob noticed a degridation in performance in some worklet tests when
ArrayCopy was added. I hypothesize that this slowdown is doing the array
copy with TBB instead of serial in the serial tests. (There have been
some checks in the existing code to suggest that some operations in TBB
can be slower than serial.) This change forces the array copy to be on
the device for which we are testing.
1147edb1 MarchingCubes now uses Gradient fast paths when possible.
d7d5da4f More changes to Neighborhood code to make it more easy to use.
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Sujin Philip <sujin.philip@kitware.com>
Merge-request: !916
This ensures that the order of the values presented to the
WorkletReduceByKey functor is consistent.
After this change, the key array used to build the worklet::Keys object
is no longer modified. The sorted keys can be obtained by using permuting
the input keys with Keys::GetSortedValuesMap().