Skip to content

Fix output offsets in threaded complete reductions #85

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
lkdvos merged 2 commits into from
Oct 7, 2026
Merged

Fix output offsets in threaded complete reductions #85

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
fe4cf1f
fix small bug for reduction into offset singleton
lkdvos Oct 7, 2026
e53d582
Test complete reductions into offset views
lkdvos Oct 7, 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
8 changes: 5 additions & 3 deletions src/mapreduce.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -209,22 +209,24 @@ function _mapreduce_block!(
T = eltype(arrays[1])
spacing = isbitstype(T) ? max(1, div(64, sizeof(T))) : 1 # to avoid false sharing
threadedout = similar(arrays[1], spacing * get_num_threads())
a = arrays[1][ParentIndex(1)]
iout = ParentIndex(offsets[1] + 1)
a = arrays[1][iout]

@Jutho Jutho Oct 7, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

This was an actual bug? Interesting that it hadn't caused havoc yet.

@lkdvos lkdvos Oct 7, 2026

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

Yeah I'm also unclear where this came from, I can confirm it was a bug and failed before, but I also didn't bump into it myself 🤷

if initop !== nothing
a = initop(a)
end
_init_reduction!(threadedout, f, op, a)

newarrays = (threadedout, Base.tail(arrays)...)
newoffsets = (offset(threadedout), Base.tail(offsets)...)
_mapreduce_threaded!(
f, op, nothing, dims, blocks, strides, offsets, costs,
f, op, nothing, dims, blocks, strides, newoffsets, costs,
newarrays, get_num_threads(), spacing, 1
)

for i in 1:get_num_threads()
a = op(a, threadedout[(i - 1) * spacing + 1])
end
arrays[1][ParentIndex(1)] = a
arrays[1][iout] = a
else
costs = costs .* .!(iszero.(strides[1]))
# make cost of dimensions with zero stride in output array (reduction dimensions),
Expand Down
18 changes: 18 additions & 0 deletions test/othertests.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -161,6 +161,24 @@ end

R3 = rand(T, (5, 5, 5))
@test prod(exp, StridedView(R3)) ≈ exp(sum(StridedView(R3)))

# Large enough to exercise threaded reduction, with offsets in both views.
n = 2 * Strided.MINTHREADLENGTH
input = sview(StridedView(fill(T(2), n + 20)), 11:(n + 10))
for initop in (nothing, zero, x -> 3x)
storage = StridedView(fill(T(99), 12))
output = sview(storage, 5:5)
fill!(output, T(7))
if initop === nothing
Base.mapreducedim!(identity, +, output, input)
else
Strided._mapreducedim!(identity, +, initop, (n,), (output, input))
end
initial = initop === nothing ? T(7) : initop(T(7))
expected = fill(T(99), 12)
expected[5] = initial + T(2) * n
@test Array(storage) == expected
end
end
end

Expand Down
Loading