|
1 | 1 | @doc raw""" |
2 | 2 | PrecomputedNeighborhoodSearch{NDIMS}(; search_radius = 0.0, n_points = 0, |
3 | | - periodic_box = nothing, update_strategy = nothing) |
| 3 | + periodic_box = nothing, update_strategy = nothing, |
| 4 | + update_neighborhood_search = GridNeighborhoodSearch{NDIMS}(), |
| 5 | + backend = DynamicVectorOfVectors{Int32}, |
| 6 | + max_neighbors = max_neighbors(NDIMS)) |
4 | 7 |
|
5 | 8 | Neighborhood search with precomputed neighbor lists. A list of all neighbors is computed |
6 | 9 | for each point during initialization and update. |
@@ -30,6 +33,9 @@ to strip the internal neighborhood search, which is not needed anymore. |
30 | 33 | - `update_neighborhood_search = GridNeighborhoodSearch{NDIMS}(; periodic_box, update_strategy)`: |
31 | 34 | The neighborhood search used to compute the neighbor lists. |
32 | 35 | By default, a [`GridNeighborhoodSearch`](@ref) is used. |
| 36 | + If the precomputed NHS is to be used on the GPU, make sure to |
| 37 | + either freeze it after initialization and never update it again, |
| 38 | + or pass a GPU-compatible neighborhood search here. |
33 | 39 | - `backend = DynamicVectorOfVectors{Int32}`: Type of the data structure to store |
34 | 40 | the neighbor lists. Can be |
35 | 41 | - `Vector{Vector{Int32}}`: Scattered memory, but very memory-efficient. |
|
0 commit comments