Tobias Ribizel

@yhmtsai I needed to add a new pipeline, since all remaining ones both build and test, so they are restricted to our available GPUs. compute capabilities 3.5 - 5.0 have...

Achieving binary compatibility with different stdlib implementations is hard to impossible if we use standard library types across the ABI boundary. We might be able to achieve this in a...

> Can you give some hints how to detect this case with cmake if it can't be supported? CMake still doesn't have native support for specifying stdlib implementations, so I...

> 0 * NaN should be NaN not zero that makes calculations more fragile, we already do similar things (special cases) for zeros inside our solver kernels

The behavior seems well-documented, e.g. with [GEMV](https://www.netlib.org/lapack/explore-html-3.6.1/d7/d15/group__double__blas__level2_gadd421a107a488d524859b4a64c1901a9.html) > When BETA is supplied as zero then Y need not be set on input. This is also not meant as a performance...

Here is the paper Pratik referred to in today's meeting: [Proposed Consistent Exception Handling for the BLAS and LAPACK](https://arxiv.org/abs/2207.09281)

I would suggest disabling the corresponding tests using `ctest -E benchmark_.*_distributed` in the short term, changing this behavior would require some refactoring of the benchmark code that we can't prioritize...

To make it short: We don't really support 32bit builds, since pure 32 bit CPUs are rather uncommon nowadays

This should have a huge impact, excerpt from the HIP 5.14 debug build log ``` 6534.89 hip/CMakeFiles/ginkgo_hip.dir/solver/batch_bicgstab_kernels.hip.cpp.o ```

Also relevant for investigating #1026 (which is where this idea came up originally)