AMREX_ALWAYS_ASSERT that works on device?

[BenWibking]

Is there a standard way to abort the code from a GPU kernel? I have a cell-by-cell iterative solve that should trigger a code abort if it fails to converge but I can't find a macro that will do this from the device side.

[WeiqunZhang]

Yes, AMREX_ALWAYS_ASSERT works on device if NDEBUG is not defined. This is the same behavior as assert in CUDA. https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#assertion

[BenWibking]

Ah... CMake was adding -DNDEBUG to the compile flags when set to a Release build. Thanks.