Skip to content

Homogenize default behavior of higher-dimensional similar to SparseArrays #3091

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

Open
alonsoC1s wants to merge 4 commits into
base: master
Choose a base branch
from
+7 −0
Open

Homogenize default behavior of higher-dimensional similar to SparseArrays #3091

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
dbc61da
Modified similar so when called with dims >= 3 a dense (GPU) array is…
alonsoC1s Apr 10, 2026
eea305a
Fixing missing type parameter
alonsoC1s Apr 10, 2026
14792c6
Merging main
alonsoC1s Apr 10, 2026
14bf466
Fixes overzealous dispatch that ruined linearization for sparse matrices
alonsoC1s Apr 13, 2026
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
7 changes: 7 additions & 0 deletions lib/cusparse/src/array.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -288,18 +288,25 @@ Base.similar(Mat::CuSparseMatrixCOO, T::Type) = CuSparseMatrixCOO(copy(Mat.rowIn
Base.similar(Mat::CuSparseMatrixCSC, T::Type, N::Int, M::Int) = CuSparseMatrixCSC(CUDACore.zeros(Int32, 1), CUDACore.zeros(Int32, 0), CuVector{T}(undef, 0), (N, M))
Base.similar(Mat::CuSparseMatrixCSR, T::Type, N::Int, M::Int) = CuSparseMatrixCSR(CUDACore.zeros(Int32, 1), CUDACore.zeros(Int32, 0), CuVector{T}(undef, 0), (N,M))
Base.similar(Mat::CuSparseMatrixCOO, T::Type, N::Int, M::Int) = CuSparseMatrixCOO(CUDACore.zeros(Int32, 0), CUDACore.zeros(Int32, 0), CuVector{T}(undef, 0), (N,M))
# For dims higher than 3 we default to dense (GPU) arrays as does Base
Base.similar(::CuSparseMatrix, T::Type, dims::Vararg{Int,N}) where N = CuArray{T}(undef, dims)

Base.similar(Mat::CuSparseMatrixCSC{Tv, Ti}, N::Int, M::Int) where {Tv, Ti} = similar(Mat, Tv, N, M)
Base.similar(Mat::CuSparseMatrixCSR{Tv, Ti}, N::Int, M::Int) where {Tv, Ti} = similar(Mat, Tv, N, M)
Base.similar(Mat::CuSparseMatrixCOO{Tv, Ti}, N::Int, M::Int) where {Tv, Ti} = similar(Mat, Tv, N, M)
# For dims higher than 3 we default to dense arrays as does Base
Base.similar(::CuSparseMatrix{Tv}, dims::Vararg{Int, N}) where {N, Tv} = CuArray{Tv}(undef, dims)

Base.similar(Mat::CuSparseMatrixCSC, T::Type, dims::Tuple{Int, Int}) = similar(Mat, T, dims...)
Base.similar(Mat::CuSparseMatrixCSR, T::Type, dims::Tuple{Int, Int}) = similar(Mat, T, dims...)
Base.similar(Mat::CuSparseMatrixCOO, T::Type, dims::Tuple{Int, Int}) = similar(Mat, T, dims...)
# The next one would overwrite `similar` for dims::Tuple{Int}, which breaks linearization
# Base.similar(::CuSparseMatrix, T::Type, dims::NTuple{N, Int}) where N = CuArray{T}(undef, dims)
Copy link
Copy Markdown
Author

@alonsoC1s alonsoC1s Apr 14, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This line works effectively the same as the fallback from SparseArrays. It uses NTuple instead of Dims{N}, but Dims{N} would have the same issue AFAICT


Base.similar(Mat::CuSparseMatrixCSC, dims::Tuple{Int, Int}) = similar(Mat, dims...)
Base.similar(Mat::CuSparseMatrixCSR, dims::Tuple{Int, Int}) = similar(Mat, dims...)
Base.similar(Mat::CuSparseMatrixCOO, dims::Tuple{Int, Int}) = similar(Mat, dims...)
Base.similar(::CuSparseMatrix{Tv}, dims::NTuple{N, Int}) where {N, Tv} = CuArray{Tv}(undef, dims)

Base.similar(Mat::CuSparseArrayCSR) = CuSparseArrayCSR(copy(Mat.rowPtr), copy(Mat.colVal), similar(nonzeros(Mat)), size(Mat))

Expand Down