Skip to content

Fix for GPU Diag issue #1679 #1684

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
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix for GPU Diag issue #1679 #1684

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
9457b17
Outer works generically for CPU and GPU ( I also verified that Comple…
kmp5VT Nov 15, 2025
dba57de
Merge branch 'main' of github.com:kmp5VT/ITensors.jl
kmp5VT Nov 15, 2025
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: 3 additions & 4 deletions NDTensors/src/diag/tensoralgebra/outer.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,9 @@
function outer!(
R::DenseTensor{<:Number, NR}, T1::DiagTensor{<:Number, N1}, T2::DiagTensor{<:Number, N2}
) where {NR, N1, N2}
for i1 in 1:diaglength(T1), i2 in 1:diaglength(T2)
indsR = CartesianIndex{NR}(ntuple(r -> r ≤ N1 ? i1 : i2, Val(NR)))
R[indsR] = getdiagindex(T1, i1) * getdiagindex(T2, i2)
end
t1 = T1.storage.data
t2 = T2.storage.data
array(R) .= t1 .* t2'
Copy link
Member

@mtfishman mtfishman Nov 15, 2025 •
edited
Loading

Choose a reason for hiding this comment

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

It looks like there are a few issues with this proposal:

  1. t2' will complex conjugate the elements, while outer[!] isn't supposed to conjugate the elements. Maybe transpose could be used as an alternative.
  2. t1 * transpose(t2) when t1 and t2 are vectors will output a matrix whose length this the product of the lengths of the diagonals of T1 and T2, I don't understand how the shapes of this operation work out since R in general is a tensor whose length is the product of the total lengths of T1 and T2 (not just the lengths of their diagonals). I think maybe you want to do something like t1 * transpose(t2) and reshape it into a vector (or relatedly, something like kron(t1, t2), which has an in-place counterpart kron!) and use that to write into the diagonal values of R.

It looks like you'll need to investigate this more to make sure it really does the correct operation.

return R
end

Expand Down
5 changes: 5 additions & 0 deletions NDTensors/test/test_diag.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,6 +123,11 @@ end

## Test dot on GPU
@test dot(t, A) ≈ dot(dev(array(t)), array(A)) rtol = sqrt(eps(elt))

NDTensors.outer!(A, t,t);
for i in NDTensors.cpu(A)
@test i == one(elt)
end
end
nothing
end
Loading