Skip to content

CompatHelper: bump compat for AcceleratedKernels to 0.5, (keep existing compat) #655

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

CompatHelper: bump compat for AcceleratedKernels to 0.5, (keep existing compat) #655

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
064df73
CompatHelper: bump compat for AcceleratedKernels to 0.5, (keep existi…
Oct 2, 2026
782a97d
Adapt to the AcceleratedKernels 0.5 host API
michel2323 Oct 2, 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
2 changes: 1 addition & 1 deletion Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,7 +34,7 @@ oneAPI_Support_jll = "b049733a-a71d-5ed3-8eba-7d323ac00b36"

[compat]
AbstractFFTs = "1.5.0"
AcceleratedKernels = "0.3.1, 0.4"
AcceleratedKernels = "0.5"
Adapt = "4"
CEnum = "0.4, 0.5"
ExprTools = "0.1"
Expand Down
26 changes: 15 additions & 11 deletions src/accumulate.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,19 +3,23 @@ import oneAPI: oneArray, oneAPIBackend
import AcceleratedKernels as AK

# Use a smaller block size on Intel GPUs to work around a scan correctness issue
# with the Blelloch parallel prefix sum at larger block sizes (>=128).
# with the parallel prefix sum at larger block sizes (>=128).
const _ACCUMULATE_BLOCK_SIZE = 64

# The scan algorithm for the given `dims`: whole-array scans and scans along a dimension
# use different algorithms, so pick the one AcceleratedKernels' `Auto()` would, with our
# block size.
_scan_alg(dims) = dims === nothing ? AK.ScanPrefixes(block_size = _ACCUMULATE_BLOCK_SIZE) :
AK.SliceScan(block_size = _ACCUMULATE_BLOCK_SIZE)

# Accumulate operations using AcceleratedKernels
Base.accumulate!(op, B::oneArray, A::oneArray; init = zero(eltype(A)),
block_size = _ACCUMULATE_BLOCK_SIZE, kwargs...) =
AK.accumulate!(op, B, A, oneAPIBackend(); init, block_size, kwargs...)
Base.accumulate!(op, B::oneArray, A::oneArray; dims = nothing, alg = _scan_alg(dims), kwargs...) =
AK.accumulate!(op, B, A; dims, alg, kwargs...)

Base.accumulate(op, A::oneArray; init = zero(eltype(A)),
block_size = _ACCUMULATE_BLOCK_SIZE, kwargs...) =
AK.accumulate(op, A, oneAPIBackend(); init, block_size, kwargs...)
Base.accumulate(op, A::oneArray; dims = nothing, alg = _scan_alg(dims), kwargs...) =
AK.accumulate(op, A; dims, alg, kwargs...)

Base.cumsum(src::oneArray; block_size = _ACCUMULATE_BLOCK_SIZE, kwargs...) =
AK.cumsum(src, oneAPIBackend(); block_size, kwargs...)
Base.cumprod(src::oneArray; block_size = _ACCUMULATE_BLOCK_SIZE, kwargs...) =
AK.cumprod(src, oneAPIBackend(); block_size, kwargs...)
Base.cumsum(src::oneArray; dims = nothing, alg = _scan_alg(dims), kwargs...) =
AK.cumsum(src; dims, alg, kwargs...)
Base.cumprod(src::oneArray; dims = nothing, alg = _scan_alg(dims), kwargs...) =
AK.cumprod(src; dims, alg, kwargs...)
Loading