This work contains code that is a reimplementation of a Jaccard Similarity pipeline that is (C) Nvidia and licensed under Apache 2.0. The reimplementation has been performed by hand and is entirely contained within CuGraph.chpl From cuGraph (https://github.com/rapidsai/cugraph) revision 3f13ffcdf (Feb. 17, 2021)
- From
cpp/src/utilities/graph_utils.cuh
- The original
fill
wrapper function definition was reimplemented as a Chapel promoted scalar-to-array assignment
- The original
- From
cpp/src/link_predicition/jaccard.cu
- the "row_sum" kernel has been reimplemented as a Chapel GPU
forall
without support for vertex weights - the 3D
jaccard_is
kernel has been reimplemented as a 1D Chapel GPUforall
with 1D->3D logical reindexing - the 1D
jaccard_jw
kernel has been reimplemented as a Chapel GPUforall
- the "row_sum" kernel has been reimplemented as a Chapel GPU