Skip to content

Add propagate w_mul_xj CUDA sparse support using matrix mul #610

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jul 16, 2025
Merged

Add propagate w_mul_xj CUDA sparse support using matrix mul #610

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion GNNlib/ext/GNNlibCUDAExt.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ end
## W_MUL_XJ

## avoid the fast path on gpu until we have better cuda support
function GNNlib.propagate(::typeof(w_mul_xj), g::GNNGraph{<:Union{COO_T, SPARSE_T}}, ::typeof(+),
function GNNlib.propagate(::typeof(w_mul_xj), g::GNNGraph{COO_T}, ::typeof(+),
xi, xj::AnyCuMatrix, e::Nothing)
propagate((xi, xj, e) -> w_mul_xj(xi, xj, e), g, +, xi, xj, e)
end
Expand Down
2 changes: 1 addition & 1 deletion GNNlib/src/msgpass.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -233,7 +233,7 @@ end
# for weighted convolution
function propagate(::typeof(w_mul_xj), g::GNNGraph, ::typeof(+), xi, xj::AbstractMatrix,
e::Nothing)
A = adjacency_matrix(g, weighted = true)
A = adjacency_matrix(g, eltype(xj); weighted = true)
return xj * A
end

Expand Down
1 change: 0 additions & 1 deletion GraphNeuralNetworks/perf/Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,4 +7,3 @@ GNNlib = "a6a84749-d869-43f8-aacc-be26a1996e48"
GraphNeuralNetworks = "cffab07f-9bc2-4db1-8861-388f63bf7694"
Graphs = "86223c79-3864-5bf0-83f7-82e725a168b6"
JLD2 = "033835bb-8acc-5ee8-8aae-3f567f8a3819"
Graphs = "093fc24a-ae57-5d10-9952-331d41423f4d"
17 changes: 17 additions & 0 deletions GraphNeuralNetworks/perf/sparse_propagate_cuda.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,23 @@ function prop_copy_xj(graph_type, sp_p, n, feat_size)
return nothing
end

function prop_w_mul_xj(graph_type, sp_p, n, feat_size)
A = sprand(n, n, sp_p)
b = rand(1, n)
B = rand(feat_size, n)
g = GNNGraph(A,
ndata = (; b = b, B = B),
edata = (; A = reshape(A.nzval, 1, :)),
graph_type = graph_type) |> dev
printstyled("propagate w_mul_xj for graph type: $graph_type", "\n", color=:yellow)
CUDA.@sync propagate(w_mul_xj, g, +; xj = g.ndata.B) # run once to compile before benchmarking
@btime CUDA.@sync propagate($w_mul_xj, $g, +; xj = $g.ndata.B) # using spmm for :sparse
printstyled("gather/scatter propagate w_mul_xj for graph type: $graph_type", "\n", color=:yellow)
CUDA.@sync propagate((xi, xj, e) -> w_mul_xj(xi, xj, e), g, +; xj = g.ndata.B) # run once to compile before benchmarking
@btime CUDA.@sync propagate((xi, xj, e) -> w_mul_xj(xi, xj, e), $g, +; xj = $g.ndata.B) # using gather/scatter
return nothing
end

seed!(0)
dev = gpu_device()
println("Device: ", dev)
Expand Down
Loading