Hariprasad Kannan
Hariprasad Kannan
It helps to type `y` (i.e., `yes`) as the answer to "Make breakpoint pending on future shared library load? (y or [n])" that follows "No source file named ...".
Yes, there are more cuSPARSE based dependencies in gpu_spmv.cu (Hybrid matrix) and elsewhere (graphblas backend).
Regarding [TRUST platform](https://cea-trust-platform.github.io/) at CEA (France), the compatible compilers are listed [here](https://github.com/cea-trust-platform/trust-code/releases/tag/v1.9.6). There are compatible versions of GCC (11.2.0, 12.0.1 etc) and Clang (14.0.0, 16.0.0 etc). Also, they support nvc++...
With [8060](https://github.com/kokkos/kokkos/pull/8060) it should be possible to add lesser checks for `parallel_reduce`.
> It also feels that we have a variety of error messages related to wrong usage (e.g. [1]) and it is becoming difficult to maintain and test them. How do...
> It will add a lot of calls to is_initialized ... and afaict this is not cached ... so I would expect it to add a to the launch overhead....
An approach based on std::variant has been committed, in order to pass the tag (ParallelForTag, ParallelScanTag, ParallelReduceTag). Another possible approach could be to pass the tag as a template parameter....
In `core/src/View/Kokkos_ViewDataAnalysis.hpp` there is a struct template called `ViewDataAnalysis` with a template parameter called `ArrayLayout` that is used as a tag in other places (e.g., `core/src/View/Kokkos_ViewDataAnalysis.hpp`, `core/unit_test/TestViewMapping_a.hpp`).
There is this `extents_type` missing error involving `Kokkos::LayoutStride` that is appearing in a few CI tests [linux / CI (x64, ubuntu:latest, g++-12, RelWithDebInfo, OPENMP, 17)](https://github.com/kokkos/kokkos/actions/runs/16759166296/job/47451006419?pr=7401#logs) and [CUDA-11.7-NVCC](https://cloud1.cees.ornl.gov/jenkins-ci/job/Kokkos/job/PR-7401/18/pipeline-overview/?selected-node=389). When I tested...
> Note: This seems to be a compiler specific issue. In my case, clang16 compiles fine, GCC12 fails. I observe the same behavior as what @janciesko is observing. The backends...