Matthias Maier [Fri, 23 Jun 2023 04:23:16 +0000 (23:23 -0500)]
Test base/hdf5_04.mpirun={1,4}: do not compare test output
We have been adding output variants for this test for years, but the
test output is simply not stable. I see different output depending on
debug or release configuration and chosen optimization level.
Let us stop here and only check whether the test runs correctly without
throwing an exception.
Matthias Maier [Fri, 23 Jun 2023 04:11:34 +0000 (23:11 -0500)]
Test base/hdf5_03.mpirun={1,4}: do not compare test output
We have been adding output variants for this test for years, but the
test output is simply not stable. I see different output depending on
debug or release configuration and chosen optimization level.
Let us stop here and only check whether the test runs correctly without
throwing an exception.
Matthias Maier [Thu, 22 Jun 2023 01:41:53 +0000 (20:41 -0500)]
CMake: avoid adding ${DEAL_II_BUNDLED_INCLUDE_DIRS} to bundled targets
These include files should only be added as system include directories
to dependend targets and not to the bundled targets themselves.
While this commit looks purely aesthetic, this commit actually works
around an obscure compiler bug with ICC 19 that otherwises errors out
with when trying to compile our bundled tbb library.
This part has been commited and merged accidentally (partly due to our
CI not testing any symengine artifacts).
Unfortunately, the anticipated replacement of `begin_raw()` by
`unrolled_to_component_indices(i)` does not work as intended. It is
neither implemented for Tensor of rank 0, nor for SymmetricTensor of
rank 4.
Matthias Maier [Tue, 20 Jun 2023 22:00:47 +0000 (17:00 -0500)]
Test sharedtria/communicate_active_fe_indices_01b: do not compare test output
The parallel distribution is computed with metis, which is not stable.
Simply skip comparing the test output which listed the number of local
degrees of freedom of a distributed, shared triangulation.
Matthias Maier [Tue, 20 Jun 2023 21:50:39 +0000 (16:50 -0500)]
Test scalapack/scalapack_05: add output variant
It is a bit disconcerting that the computed recirpocal condition number
of the matrix changes. But given the fact that the manual computation
(first number) and the scalapack-internal function return vastly
differing numbers already tell me that the test setup simply isn't
robust.
Let us add an output variant for the scalapack_05 test. I get almost the
same number on three different configuration so I am hopeful that we can
get away with an output variant for the time being.
Matthias Maier [Tue, 20 Jun 2023 21:06:18 +0000 (16:06 -0500)]
Test rpe/distributed_compute_intersection_locations_{02,04} do not compare output
Let us also modify the following two tests with all their mpirun
variants. I again see slight differences between gcc and clang versions,
and between libc++ and libstdc++.
Matthias Maier [Tue, 20 Jun 2023 20:57:34 +0000 (15:57 -0500)]
Test rpe/convert_intersections_to_rpe_data_02: do not compare output
This test outputs a mesh with differing results between standard
libraries (gnu libstdc++ versus clang/llvm libc++).
Instead of adding another output variant for libc++, let us simply
disable the final comparison and run the test only relying on cgal and
our library to throw an error.
Matthias Maier [Tue, 20 Jun 2023 20:51:27 +0000 (15:51 -0500)]
Test rpe/convert_intersections_to_rpe_data_02: do not compare output
This test outputs a mesh with differing results between cgal library
versions and underlying standard library. I see at least three different
results on the regression tester.
Instead of adding three different variants per mpirun, let us simply
disable the final comparison and run the test only relying on cgal and
our library to throw an error.
Matthias Maier [Tue, 20 Jun 2023 19:45:45 +0000 (14:45 -0500)]
Test opencascade/stl_write: do not compare output
The output of this test is a mesh where depending on the opencascade
version in question, toolchain and hardware a slightly different
interpolation is found. Let us not compare the final mesh and simply
rely on the routine to throw an error if it cannot create and write the
mesh.
Matthias Maier [Tue, 20 Jun 2023 19:42:52 +0000 (14:42 -0500)]
Test opencascade/stl_refinement: do not compare output
The output of this test is a mesh where depending on the opencascade
version in question, toolchain and hardware a slightly different
interpolation is found. Let us not compare the final mesh and simply
rely on the routine to throw an error if it cannot create and refine the
mesh.
projects / dealii.git / log