add C implementation of MST #5044
Conversation
cpp/src/c_api/legacy_mst.cpp
Outdated
| namespace c_api { | ||
|
|
||
| struct cugraph_layout_result_t { | ||
| cugraph_type_erased_device_array_t* vertices_{nullptr}; |
There was a problem hiding this comment.
I'd suggest using cugraph_induced_subgraph_result_t as the return type... as that has a COO of edges.
We should probably think of a better name, but perhaps a FIXME on that for now. I have a branch where I've been working on C API updates in graph creation that I think will create a better name for this structure.
|
|
||
| const size_t num_edges = result_legacy_coo_graph->view().number_of_edges; | ||
|
|
||
| rmm::device_uvector<vertex_t> result_src(num_edges, handle_.get_stream()); |
There was a problem hiding this comment.
Could you add a FIXME here.
These copies could be unnecessary. src_indices, dst_indices and edge_data within the result_legacy_coo_graph are already rmm::device_buffer instances. If we made a new constructor for cugraph_type_erased_host_array_t that took an rmm::device_buffer we could skip these copies.
We can do that optimization later.
cpp/tests/c_api/legacy_mst_test.c
Outdated
| @@ -0,0 +1,227 @@ | |||
| /* | |||
| * Copyright (c) 2023-2024, NVIDIA CORPORATION. | |||
cpp/tests/traversal/ms_bfs_test.cu
Outdated
| d_distances.end(), | ||
| static_cast<vertex_t>(0)); | ||
| ASSERT_TRUE(ref_sum > 0); | ||
|
|
There was a problem hiding this comment.
Why is this update in this PR? Seems out of place.
Is this just a matter of merging the MSBFS PR first?
ChuckHastings
added
improvement
|
|
||
| def minimum_spanning_tree(ResourceHandle resource_handle, | ||
| _GPUGraph graph, | ||
| bool_t do_expensive_check): |
There was a problem hiding this comment.
Can this have a default value of False?
| [0 1 2 3 4 5 1 1 1 4] | ||
| >>> destinations | ||
| [1 0 1 1 1 4 2 3 4 5] | ||
| >>> edge_weights | ||
| [0.1 0.1 3.1 2.1 1.1 3.2 3.1 2.1 1.1 3.2] | ||
| >>> subgraph_offsets | ||
| [0 10] |
There was a problem hiding this comment.
Is this the actual output of running the example code? These don't look like device array reprs.
| else: | ||
| cupy_edge_weights = None | ||
|
|
||
| # FIXME: Should we keep the offsets array or just drop it from the final |
There was a problem hiding this comment.
Can we resolve this in this PR?
|
/merge |
This PR adds the C, PLC and python API for Minimum Spanning Tree
closes #4882