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
Facilitate easier examination of user data by allowing to save primitives and queries and subsequently share with us. Immediate use cases: NimbleSM (to study OpenMP/MPI scaling), CabanaMD (to play with the problem).
The filenames automatically postfixed with .arborx. For distributed runs, they are postfixed with .<rank>.arborx.
Some ideas
We want to be able to save both primitives and predicates data.
For primitives, we save Kokkos::View<T*,MemorySpace> (either provided by a user, or constructed through AccessTraits). The metadata should include:
[optional] little endian/big endian (if saving in binary mode)
Size (number of primitives)
[optional] Data type Point/Box (we could deduce it from the sizes of the file, though)
Data
For predicates, we save Geometry (and k for the nearest). We can limit ourselves with Point and Box for the geometry for now, but think about providing a way to treat custom geometry as well in the future. In addition, possibly treat attachments. Callback will probably be harder.
The text was updated successfully, but these errors were encountered:
Purpose
Facilitate easier examination of user data by allowing to save primitives and queries and subsequently share with us. Immediate use cases: NimbleSM (to study OpenMP/MPI scaling), CabanaMD (to play with the problem).
API
Something along the lines of
The filenames automatically postfixed with
.arborx
. For distributed runs, they are postfixed with.<rank>.arborx
.Some ideas
We want to be able to save both primitives and predicates data.
For primitives, we save
Kokkos::View<T*,MemorySpace>
(either provided by a user, or constructed throughAccessTraits
). The metadata should include:For predicates, we save
Geometry
(andk
for the nearest). We can limit ourselves withPoint
andBox
for the geometry for now, but think about providing a way to treat custom geometry as well in the future. In addition, possibly treat attachments. Callback will probably be harder.The text was updated successfully, but these errors were encountered: