KernelAbstractions 0.10 - #625
Closed
christiangnrd wants to merge 6 commits into
Closed
christiangnrd wants to merge 6 commits into
christiangnrd wants to merge 6 commits into
Conversation
christiangnrd
force-pushed
the
intrinsics
branch
5 times, most recently
from
September 7, 2026 23:34
b17c721 to
0b25323
Compare
christiangnrd
force-pushed
the
intrinsics
branch
2 times, most recently
from
September 11, 2026 21:00
e0191ff to
62848d0
Compare
christiangnrd
force-pushed
the
intrinsics
branch
6 times, most recently
from
September 26, 2026 19:34
d3b0f28 to
eb64fcf
Compare
In preparation for KernelAbstractions becoming a weak dependency.
The `sub_group_size` compiler keyword sets the `intel_reqd_sub_group_size` metadata of the kernel, defaulting to 32.
Implement KernelInterface next to the KernelAbstractions back end, which moves to oneAPIKernelsOld.jl until the port to KernelAbstractions 0.10. The per-dimension launch limits (`max_work_group_dims`, `max_num_groups`) come from the Level Zero compute properties. Every auto-sized launch queries them, so they are cached per device. The KernelInterface test suite runs as the `kernelinterface` test. Its events test is skipped when every submission synchronizes (the Aurora LTS workaround, ONEAPI_SYNC_EACH_SUBMISSION), as the ordering it checks is not observable then. Co-authored-by: Tim Besard <tim.besard@gmail.com>
Needed to make JuliaGPU/KernelAbstractions.jl#659 pass CI. We should eventually add RNG support to oneAPI by using a similar approach to OpenCL.jl, but lets skip these tests for now
KernelAbstractions 0.10 builds on KernelInterface, so the KernelInterface back end becomes `oneAPIBackend` and the old KernelAbstractions back end is removed. What remains specific to KernelAbstractions (launching `@kernel` functions, `__validindex`, scratchpads, adapting to the CPU) moves into a package extension. Adapting to the back end (`adapt(oneAPIBackend(), x)`) is now defined next to the back end, as KernelAbstractions 0.10 requires of every back end. Since KernelAbstractions 0.10, an ndrange may contain ranges, so the preliminary and the tuned workgroup size, as well as the number of items used by the prefer_blocks heuristic, are computed from the number of indices along each dimension rather than from the raw ndrange. An empty ndrange returns before compiling. The KernelAbstractions tests that only apply to the CPU back end are skipped. Co-authored-by: Tim Besard <tim.besard@gmail.com>
KernelAbstractions 0.10 is unreleased, and the released AcceleratedKernels does not support it yet.
Member
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.