tree_d_3d.kernels.fmm_traverse.fmm_traverse_pyccel

The Pyccel version (CPU) of the fmm_traverse kernel of the Barnes-Hut and FMM solvers.

fmm_traverse_cuda.cu next to this file is the CUDA version (GPU) and fmm_traverse_numpy.py the version for machines without Pyccel, all with the same arguments; the folder’s __init__.py declares the kernel (see :mod:tree_d_3d.kernels). The flat quadtree is described in :mod:tree_d_3d.flat_tree.

Functions

Name	Description
fmm_traverse	One generation of the dual-tree traversal, which fills mutual M2L and P2P lists.

fmm_traversefunction#

View source

def fmm_traverse(node_center: float[:, :], node_half: float[:], node_start: int64[:], node_end: int64[:], node_child: int64[:], theta: float, frontier: int64[:, :], n_frontier: int, next_frontier: int64[:, :], m2l: int64[:, :], p2p: int64[:, :], counts: int64[:]) -> None

One generation of the dual-tree traversal, which fills mutual M2L and P2P lists.

Each node pair of frontier[:n_frontier] becomes an M2L pair (well separated: (r_a + r_b) < theta d, with r the half-diagonal and d the centre distance), a P2P pair (adjacent leaves) or the pairs of its children, appended to next_frontier. Start from the single pair (0, 0) and repeat with the new frontier until it is empty. counts holds the numbers of M2L, P2P and next frontier pairs, and pairs are appended at those positions; counts[3] becomes -1 if a buffer is too small (pairs that do not fit are counted, not written, so the counts are the sizes needed).