You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We are replacing custom atomic implementations in libcudf with cuda::atomic_ref (rapidsai/cudf#13575). The current atomic_ref would work fine with 4-byte or 8-byte integers and floating points but doesn't support time types like durations or timestamps from cuda/std/chrono. Also, it would be great to have atomic_ref support 1-byte and 2-byte types like bool, int8_t and int16_t. The functions we are mostly interested in are fetch_min, fetch_max and fetch_add.
We are replacing custom atomic implementations in
libcudf
withcuda::atomic_ref
(rapidsai/cudf#13575). The currentatomic_ref
would work fine with 4-byte or 8-byte integers and floating points but doesn't support time types like durations or timestamps fromcuda/std/chrono
. Also, it would be great to haveatomic_ref
support 1-byte and 2-byte types likebool
,int8_t
andint16_t
. The functions we are mostly interested in arefetch_min
,fetch_max
andfetch_add
.For your reference, here is the list of time types used in
libcudf
:https://github.com/rapidsai/cudf/blob/2ca46306aad848f5102314374cef1c026856569e/cpp/include/cudf/wrappers/durations.hpp
https://github.com/rapidsai/cudf/blob/2ca46306aad848f5102314374cef1c026856569e/cpp/include/cudf/wrappers/timestamps.hpp
The text was updated successfully, but these errors were encountered: