Skip to content

Zero the triangle in one kernel on CuArrays and ROCArrays #313

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
kshyatt merged 2 commits into from
Oct 27, 2026
Merged

Zero the triangle in one kernel on CuArrays and ROCArrays #313

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
67d09be
One kernel for uppertriangular!/lowertriangular! on CuArrays and ROCA…
leburgel Oct 26, 2026
12c5a86
Drop the GPU lowertriangular! methods, which no decomposition reaches
leburgel Oct 26, 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 ext/MatrixAlgebraKitAMDGPUExt/MatrixAlgebraKitAMDGPUExt.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,6 +35,13 @@ end

MatrixAlgebraKit.supports_svd_full(::ROCSOLVER, f::Symbol) = f in (:qr_iteration, :jacobi, :divide_and_conquer, :bisection)

# one kernel for the whole triangle instead of one per column
function MatrixAlgebraKit.uppertriangular!(A::StridedROCMatrix)
m, n = size(A)
A .= ifelse.((1:m) .> transpose(1:n), zero(eltype(A)), A)
return A
end

# rocSOLVER's `gesvd*_batched` functions take the batch as an array
# of device pointers, so a group of equally sized matrices
# doesn't need a copy into a 3D ROCArray.
Expand Down
7 changes: 7 additions & 0 deletions ext/MatrixAlgebraKitCUDAExt/MatrixAlgebraKitCUDAExt.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,6 +42,13 @@ end

MatrixAlgebraKit.prefers_ungqr(::CUSOLVER) = true

# one kernel for the whole triangle instead of one per column
function MatrixAlgebraKit.uppertriangular!(A::StridedCuMatrix)
m, n = size(A)
A .= ifelse.((1:m) .> transpose(1:n), zero(eltype(A)), A)
return A
end

MatrixAlgebraKit.supports_svd_full(::CUSOLVER, f::Symbol) = f in (:qr_iteration, :jacobi, :svd_polar)

# `cusolverDnXgesvdjBatched` only accepts matrices up to 32x32
Expand Down
Loading