[dependabot]

Updates the requirements on GPUArrays and JLArrays to permit the latest version.
Updates GPUArrays to 12.1.0

Release notes

Sourced from GPUArrays's releases.

v12.1.0

GPUArrays v12.1.0

Diff since v12.0.3

Merged pull requests:

• Define alias detection for every GPU array through a location hook (#811) (@​maleadt)

Changelog

Sourced from GPUArrays's changelog.

GPUArrays.jl release history

This document lists the noteworthy changes in GPUArrays.jl releases that back-ends act on, newest first, aimed at the authors of back-end packages. For the complete list of merged pull requests, see the GitHub releases.

v12.1

GPUArrays defines Base.dataids and Base.mightalias for every AbstractGPUArray. A back-end that implements the new GPUArrays.memory_location hook (the coordinate of an array's allocation, normally its address, and its byte offset into it) gets precise alias detection: disjoint views of one buffer don't alias, empty arrays alias nothing on every Julia version, and SubArrays of two such arrays are compared without converting their parents to pointers. Arrays wrapped in a SubArray or other wrapper are still compared by allocation, which is conservative. Back-ends should implement the hook and delete their own dataids and mightalias methods, which would keep winning dispatch.

v12.0

Base's reductions, sorting, scans, findall, logical indexing and reverse are now implemented once for every AnyGPUArray, on top of AcceleratedKernels.jl (AK) 0.5. A back-end gets all of them by defining nothing beyond the array interface and a KernelAbstractions back-end, and they follow Base's semantics: stable sorting by default, init applied once, no neutral element needed, Base's result types, empty results and errors.

Breaking changes for back-ends:

• The GPUArrays.mapreducedim! hook is removed and GPUArrays no longer picks neutral elements. Every Base reduction goes to AK directly. Back-ends must delete their methods for the hook, together with the reduction kernels only it used.
• GPUArrays.default_rng and the deprecated RNG(state::AbstractGPUArray) and seed!(rng, ::Vector{UInt32}) methods are removed. Construct the RNG with RNG{AT}() and seed it with an integer.
• KernelAbstractions 0.9.43 or later and Adapt 4.7.3 or later are required.

What back-ends should delete:

Methods narrower than GPUArrays' keep winning dispatch, so a back-end that keeps its own findall, to_index/to_indices, _accumulate!/accumulate, sort!/sortperm!/ partialsort!, or reverse! bypasses GPUArrays' Base conformance, and some of those methods are ambiguous with GPUArrays' (e.g. findall(f, ::MyArray) against findall(::Fix2{typeof(in)}, ::AnyGPUArray)). Delete them.

A vendor sort or scan (e.g. one built on MPSGraph) can stay as a fast path, as long as it falls back to GPUArrays' method for the cases it doesn't handle and accepts AK's algorithm objects (alg=AK.MergeSort() etc.). There is no extension point for vendor reductions; GPUArrays._ak_mapreducedim! is internal.

... (truncated)

Commits

• 6b164fa Define alias detection for every GPU array through a location hook (#811)
• 7ce1d67 Bump versions.
• 70cf21c Use byte-range overlap for dense JLArray alias detection (#803)
• 6ef94ae Sweep reduction shapes for one real and one complex element type (#810)
• 2f930aa Merge pull request #808 from JuliaGPU/tb/gpuarrays-12.0.2
• dbba903 Bump version to 12.0.2
• 7aba731 Check a single logical mask's shape on Julia 1.10
• bba15c2 Merge pull request #806 from JuliaGPU/tb/gpuarrays-12.0.1
• 348c99d Bump version to 12.0.1
• 19746f9 Don't overflow abs2 in the adjoint reduction test
• Additional commits viewable in compare view

Updates JLArrays to 0.4.2

Commits

• See full diff in compare view