v2.6.3

Strided v2.6.3

Diff since v2.6.2

Bug fix for 0-dimensional strided view edge cases.

Merged pull requests:

• fix: bypass kernels for 0-dim inputs (#76) (@lkdvos)

v2.6.2

Strided v2.6.2

Diff since v2.6.1

A performance-focused patch release that reduces compilation time and with small improvements to the mapreduce kernels.

Highlights

• Faster compilation: @nospecialize added to the mapreduce dispatch chain (#66) and blas_mul! de-specialized (#73) to cut compile time.
• Faster reductions: added a unit-stride fast path to _mapreduce_kernel! and specialized the non-reducing kernel by dropping the reduction-only iszero hoist (#69).
• Kernel ordering: improved loop ordering in the mapreduce kernel — "order → fuse → order" (#74).
• Minor internal cleanups (simplified firststride condition, improved comments) and CI maintenance (bump actions/checkout 6 → 7, #68).

Full Changelog

See the commit history for the complete list of changes.

Merged pull requests:

• Add @nospecialize to mapreduce dispatch chain to cut compile time (#66) (@lkdvos)
• Bump actions/checkout from 6 to 7 (#68) (@dependabot[bot])
• Specialize non-reducing kernel: drop reduction-only iszero hoist (#69) (@lkdvos)
• Add unit-stride fast path to _mapreduce_kernel! (#70) (@lkdvos)
• Despecialize blas_mul! for improved compilation time (#73) (@lkdvos)
• order - fuse - order kernel logic (#74) (@lkdvos)
• Release v2.6.2 (#75) (@lkdvos)

Closed issues:

• Correctness error in AMD kernel with conj (#63)
• perf: threading unused for permuted contiguous views (#72)

v2.6.1

Strided v2.6.1

Diff since v2.6.0

Patch release with a small fix for the AMD kernels on Julia versions <v1.11, that failed to properly use conj.

Merged pull requests:

• Bump codecov/codecov-action from 6 to 7 (#61) (@dependabot[bot])
• Update Buildkite pipeline for the new JuliaGPU cluster (#62) (@maleadt)
• Attempt to fix AMD 1.10 conj (#64) (@kshyatt)
• Bump v2.6.1 (#65) (@kshyatt)

v2.6.0

Strided v2.6.0

Diff since v2.5.0

• Add support for using vendor (NVIDIA and AMD) BLAS libraries where possible for mul! calls on StridedViews, falling back to the GPUArrays kernel otherwise
• Add two new extensions to support these vendor libraries

Merged pull requests:

• Fix isblasmatrix for GPUArrays (#59) (@kshyatt)
• Bump v2.6.0 (#60) (@kshyatt)

v2.5.0

Strided v2.5.0

Diff since v2.4.0

We are pleased to announce the release of Strided v2.5.0, which introduces experimental support for GPUArray parent arrays via KernelAbstractions.jl.

This feature enables strided operations on GPU-backed arrays, expanding the range of supported execution backends. As this support is still in an early stage, additional performance improvements and stability refinements are expected in upcoming releases.

Feedback is important at this stage—please report any issues, unexpected behavior, or performance observations to help guide further development.

Merged pull requests:

• GPU support through KernelAbstractions (#51) (@lkdvos)
• Bump julia-actions/setup-julia from 2 to 3 (#58) (@dependabot[bot])

Closed issues:

• GPU and AD compatibility (#7)

v2.4.0

Strided v2.4.0

Diff since v2.3.6

Breaking changes

• Switch from CUDA.jl to CUDACore.jl for reduction in load and precompilation times

Other changes

• Fixes for TagBot
• Switch to Dependabot from CompatHelper

Merged pull requests:

• replace compathelper with dependabot (#55) (@Jutho)
• Switch to CUDACore (#57) (@kshyatt)

v2.3.6

Strided v2.3.6

Diff since v2.3.4

This release has been identified as a backport.
Automated changelogs for backports tend to be wildly incorrect.
Therefore, the list of issues and pull requests is hidden.

v2.3.5

Strided v2.3.5

Diff since v2.3.4

This release has been identified as a backport.
Automated changelogs for backports tend to be wildly incorrect.
Therefore, the list of issues and pull requests is hidden.

v2.3.4

Strided v2.3.4

Diff since v2.3.2

Merged pull requests:

• Bump actions/checkout from 4 to 5 (#38) (@dependabot[bot])
• Bump actions/checkout from 5 to 6 (#39) (@dependabot[bot])
• Extensions to support copy! for CuStridedView (and friends!) (#40) (@kshyatt)
• Update README.md (#41) (@kshyatt)
• Use QKH format checker (#42) (@kshyatt)
• Apply runic formatting (#43) (@kshyatt)
• Create JLArrays extension and tests for GPU mocking (#44) (@kshyatt)
• Update Project.toml (#45) (@kshyatt)
• Add some mul overrides too (#46) (@kshyatt)
• Update Project.toml (#47) (@kshyatt)

v2.3.3

Strided v2.3.3

Diff since v2.3.2

Merged pull requests:

• Bump actions/checkout from 4 to 5 (#38) (@dependabot[bot])
• Bump actions/checkout from 5 to 6 (#39) (@dependabot[bot])
• Extensions to support copy! for CuStridedView (and friends!) (#40) (@kshyatt)
• Update README.md (#41) (@kshyatt)
• Use QKH format checker (#42) (@kshyatt)
• Apply runic formatting (#43) (@kshyatt)
• Create JLArrays extension and tests for GPU mocking (#44) (@kshyatt)
• Update Project.toml (#45) (@kshyatt)