Concurrent multisource backwardpass#5206

Merged

rapids-bot[bot] merged 98 commits intorapidsai:branch-25.10from

HowardHuang1:HH-multisource-backwardpass

Sep 8, 2025

Commits on Jun 30, 2025

cpp implementation of brandes algo
HowardHuang1
committed

Commits on Jul 1, 2025

loop through frontier levels instead of all vertices --> small ~1.7x speedup
HowardHuang1
committed

Commits on Jul 3, 2025

per_v_transform_reduce_outgoing_e() API has restriction that it takes input array of size vertex_list and outputs array of same size vertex_list + use scatter to map array of size vertex_list back …
HowardHuang1
committed
clean up debug statements
HowardHuang1
committed
add test files for cugraph, nx-cugraph, and pylibcugraph
HowardHuang1
committed

Commits on Jul 4, 2025

Commits on Jul 7, 2025

Commits on Jul 8, 2025

avoid scanning entire vertex range in every loop instead perform edge and delta updates on frontier vertices only
HowardHuang1
committed

Commits on Jul 9, 2025

Commits on Jul 10, 2025

combine 3 thrust calls into 1
HowardHuang1
committed

Commits on Jul 11, 2025

Commits on Jul 15, 2025

clean up timers and debug statements
HowardHuang1
committed

Commits on Jul 18, 2025

Commits on Jul 19, 2025

Merge remote-tracking branch 'upstream/branch-25.08' into HH-betweenness-centrality
HowardHuang1
committed

Commits on Jul 22, 2025

Commits on Jul 24, 2025

Commits on Jul 25, 2025

implement multisource backward pass
HowardHuang1
committed

Commits on Jul 29, 2025

add debug statements
HowardHuang1
committed

Commits on Jul 30, 2025

replace functor with lambda multisource backward pass implementation
HowardHuang1
committed

Commits on Aug 4, 2025

this code doesn't build due to errors with per_v_transform_reduce_outgoing_e not supporting vertex frontier and tags
HowardHuang1
committed

Commits on Aug 6, 2025

Commits on Aug 7, 2025

Commits on Aug 8, 2025

eliminated bottleneck of nested loops in parallel frontier creation which resulted in massive speedup
HowardHuang1
committed

Commits on Aug 9, 2025

pre-compute distance levels and do O(1) lookup this gave massive speedup
HowardHuang1
committed

Commits on Aug 11, 2025

add in time synchronization
HowardHuang1
committed

Commits on Aug 12, 2025

Commits on Aug 13, 2025

Commits on Aug 14, 2025

Commits on Aug 15, 2025

Commits on Aug 16, 2025

modify buffer in place rather than copy to temporaries
HowardHuang1
committed

Commits on Aug 18, 2025

modify buffer in place for BFS forward pass
HowardHuang1
committed

Commits on Aug 19, 2025

Commits on Aug 20, 2025

Commits on Aug 21, 2025

Merge remote-tracking branch 'upstream/branch-25.10' into HH-multisource-backwardpass
HowardHuang1
committed

Commits on Aug 23, 2025

move code to minimize scope of variable
HowardHuang1
committed

Commits on Aug 25, 2025

change sorted=true to sorted=false which affects correctness of split between high degree and low degree vertices. This slows down performance from 80x to 68x
HowardHuang1
committed

Commits on Aug 26, 2025

added CUB sort which changed performance from 68x to 73x
HowardHuang1
committed

Commits on Aug 27, 2025

CUB segmented sort implementation with cast from int64 to int32 workaround
HowardHuang1
committed

Commits on Aug 28, 2025

Commits on Aug 29, 2025

Commits on Aug 30, 2025

directly use distance_level_offsets with correct shift offset values
HowardHuang1
committed

Commits on Sep 2, 2025

Commits on Sep 3, 2025

Commits on Sep 5, 2025