VecMem 1.6.0
Release including the following update(s) since v1.5.0:
- Made
vecmem::device_atomic_ref<T>::compare_exchange_strong
work correctly with CUDA; - Made
vecmem::device_atomic_ref
generally behave atomically even on the host with GCC and Clang;- MSVC support, and some further code cleanup to come in the future;
- Added a concept for atomic references;
- Added (thread-safe) bulk reservation / extension operations to
vecmem::device_vector
; - Updated the CI build matrix to test for C++20 and C++23 (host) builds explicitly;
- At the same time introduced a (hopefully) temporary workaround for a GitHub runners issue with MSVC;
- Turned on the usage of
-fvisibility-ms-compat
for the main libraries when using a compiler that supports this flag; - Started using doxygen-awesome-css for generating the project's documentation.