Started by timer00003012231113686953.77500990ada9ad7479ac8d906b522adee4c1bc58de14fac4ada9ad7479ac8d906b522adee4c1bc58de14fac4refs/remotes/origin/developada9ad7479ac8d906b522adee4c1bc58de14fac4ada9ad7479ac8d906b522adee4c1bc58de14fac4refs/remotes/origin/developada9ad7479ac8d906b522adee4c1bc58de14fac4ada9ad7479ac8d906b522adee4c1bc58de14fac4refs/remotes/origin/develophttps://github.com/kokkos/kokkos/00199testReportfalse#99030122313012038Kokkos_nightly #990990false99058545SUCCESS1791865907367https://cloud1.cees.ornl.gov/jenkins-ci/job/Kokkos_nightly/990/- simd/unit_tests/include/TestSIMD_MemoryPermute.hppsimd/src/Kokkos_SIMD_Scalar.hpp388f94fb84ddb6a8b730a50a181073f2c434cfda1791566334000https://cloud1.cees.ornl.gov/jenkins-ci/user/thomas.rothethomas.rothethomas.rothe@hydrotec.desimd: Skip masked-off lane in scalar masked scatter
The masked unchecked_scatter_to for simd_abi::scalar wrote a
value-initialized element when the mask was false; the masked
partial_scatter_to forwards to it. Skip the store, as the other ABIs
do, and add a test that fails without the fix.
Fixes #9648
Signed-off-by: Thomas Rothe <thomas.rothe@hydrotec.de>
2026-10-09 19:18:54 +0200388f94fb84ddb6a8b730a50a181073f2c434cfdasimd: Skip masked-off lane in scalar masked scattereditsimd/src/Kokkos_SIMD_Scalar.hppeditsimd/unit_tests/include/TestSIMD_MemoryPermute.hpp
- simd/unit_tests/include/TestSIMD_MemoryPermute.hpp0f250f4d8210c67d712e6f75792f6008b401ab6f1791821298000https://cloud1.cees.ornl.gov/jenkins-ci/user/thomas.rothethomas.rothethomas.rothe@hydrotec.desimd: Use the plain array in the masked-off scatter test
std::span::operator[] is host-only and triggers nvcc warnings in
host/device code.
Signed-off-by: Thomas Rothe <thomas.rothe@hydrotec.de>
2026-10-12 18:08:18 +02000f250f4d8210c67d712e6f75792f6008b401ab6fsimd: Use the plain array in the masked-off scatter testeditsimd/unit_tests/include/TestSIMD_MemoryPermute.hpp
- containers/unit_tests/TestDynRankView_ViewCustomization.hppcontainers/src/Kokkos_DynRankView.hppada9ad7479ac8d906b522adee4c1bc58de14fac41791837284000https://cloud1.cees.ornl.gov/jenkins-ci/user/noreplynoreplynoreply@github.comFix DynRankView rank when wrapping a pointer with an AccessorArg (#9651)
For customized value types, the DynRankView constructor that takes a
pointer always subtracted one from the rank computed from the layout,
assuming the last extent is the accessor argument. When the accessor
argument is passed as an AccessorArg_t instead, e.g.
view_wrap(ptr, AccessorArg_t{n}), the layout holds only the real
extents, so rank() came out one too small while extent(i) was correct.
The allocating constructor already handles this with !has_accessor_arg;
apply the same condition to the pointer constructor.
This affects unmanaged Sacado Fad DynRankViews created with
Sacado::createViewWithType or createDynRankViewWithType, which pass the
Fad size as an AccessorArg_t. Code that dispatches on rank(), such as
Intrepid2's RealSpaceTools and CellTools, then picks the wrong subviews
and silently computes wrong results.
Add a test for this case to view_customization_extra_int_arg.
Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com
Signed-off-by: Mauro Perego <mperego@sandia.gov>
2026-10-12 14:34:44 -0600ada9ad7479ac8d906b522adee4c1bc58de14fac4Fix DynRankView rank when wrapping a pointer with an AccessorArg (#9651)editcontainers/src/Kokkos_DynRankView.hppeditcontainers/unit_tests/TestDynRankView_ViewCustomization.hpp
githttps://cloud1.cees.ornl.gov/jenkins-ci/user/thomas.rothethomas.rothehttps://cloud1.cees.ornl.gov/jenkins-ci/user/noreplynoreplyfalse991https://cloud1.cees.ornl.gov/jenkins-ci/job/Kokkos_nightly/991/989https://cloud1.cees.ornl.gov/jenkins-ci/job/Kokkos_nightly/989/