v3.34.0

LinearSolve v3.34.0

Diff since v3.33.0

Merged pull requests:

• Cleanup pardiso test (#741) (@j-fu)
• Fix missing imports in CUDA extension (#742) (@ChrisRackauckas-Claude)

v3.33.0

LinearSolve v3.33.0

Diff since v3.32.0

Merged pull requests:

• Use verbosity system (#622) (@jClugstor)

v3.32.0

LinearSolve v3.32.0

Diff since v3.31.0

Merged pull requests:

• Add BLISLUFactorization, CudaOffloadLUFactorization, and MetalLUFactorization to default solver choices (#733) (@ChrisRackauckas-Claude)

v3.31.0

LinearSolve v3.31.0

Diff since v3.30.0

Merged pull requests:

• Comprehensive docstring improvements for core LinearSolve types and functions (#728) (@ChrisRackauckas-Claude)
• Display NaN values in LinearSolveAutotune telemetry output (#729) (@ChrisRackauckas-Claude)
• Add complete autotune preference integration with availability checking (#730) (@ChrisRackauckas-Claude)
• Update LinearSolveAutotune preferences integration for dual preference system (#731) (@ChrisRackauckas-Claude)

v3.30.0

LinearSolve v3.30.0

Diff since v3.29.0

Merged pull requests:

• CompatHelper: bump compat for ForwardDiff in [weakdeps] to 1, (keep existing compat) (#726) (@github-actions[bot])

v3.29.0

LinearSolve v3.29.0

Diff since v3.28.0

Merged pull requests:

• Add FastLapackInterface.jl's LU to LinearSolveAutotune (#714) (@ChrisRackauckas-Claude)
• Update compat bounds to drop support for packages older than 3 months (#715) (@ChrisRackauckas-Claude)
• Add maxtime parameter to LinearSolveAutotune for timeout handling (#716) (@ChrisRackauckas-Claude)
• Fix hanging issue in autotune timeout mechanism (#717) (@ChrisRackauckas-Claude)
• Improve timeout handling: skip timed-out algorithms and cap big benchmarks at 15000 (#718) (@ChrisRackauckas-Claude)
• Revert timeout handling in autotune to use maxtime blocking (#719) (@ChrisRackauckas-Claude)
• CompatHelper: bump compat for AMDGPU in [weakdeps] to 2, (keep existing compat) (#720) (@github-actions[bot])
• Pure-Julia Sparse Cholesky (#721) (@samuelsonric)
• Improve constructor. (#722) (@samuelsonric)
• Add warning messages when algorithms are skipped in autotune benchmarking (#723) (@ChrisRackauckas-Claude)
• Get rid of deepcopy, put Dual types in type system (#724) (@jClugstor)

Closed issues:

• Use of deepcopy doesn't trim well (#648)

v3.28.0

LinearSolve v3.28.0

Diff since v3.27.1

Merged pull requests:

• Fix Pardiso.jl tests (#701) (@ChrisRackauckas)
• Improve error messages in LinearSolveAutotune telemetry (#703) (@ChrisRackauckas)
• Fix gh auth status false negative detection in LinearSolveAutotune (#704) (@ChrisRackauckas)
• Add MKL preference management to autotune system (#706) (@ChrisRackauckas)
• Add detailed CPU information from Sys.cpu_info() to telemetry (#707) (@ChrisRackauckas)
• Add AMDGPUOffloadFactorization algorithm support (#708) (@ChrisRackauckas)
• Refactor CudaOffloadFactorization into LU and QR variants (#709) (@ChrisRackauckas)
• Update jet.jl tests (#710) (@ChrisRackauckas)
• Fix broken note admonitions (#711) (@klamike)
• Add AMDGPU methods to documentation (#712) (@ChrisRackauckas)
• Add GPU type detection to system information (#713) (@ChrisRackauckas-Claude)

v3.27.1

LinearSolve v3.27.1

Diff since v3.27.0

Merged pull requests:

• Improve JET tests to cover all solvers (#694) (@ChrisRackauckas-Claude)
• Fix BLIS 64 (#699) (@ChrisRackauckas)

v3.27.0

LinearSolve v3.27.0

Diff since v3.26.0

Merged pull requests:

• LinearSolveAutotune works for me with these changes (#692) (@chriselrod)
• bump some minimum versions (#693) (@ChrisRackauckas)
• Add allocation tests for caching interface (#695) (@ChrisRackauckas-Claude)
• Build docs with LinearSolveAutotune (#696) (@ChrisRackauckas)
• Update benchmarking.jl definition of big (#697) (@ChrisRackauckas)

v3.26.0

LinearSolve v3.26.0

Diff since v3.25.1

Merged pull requests:

• Make GenericLUFactorization bypass overloads when used on DualLinear Problems (#685) (@jClugstor)
• Improve LinearSolveAutotune UI/UX (#686) (@ChrisRackauckas-Claude)
• Add gh_cli_jll as fallback for LinearSolveAutotune when gh CLI is not installed (#687) (@ChrisRackauckas-Claude)
• Update autotuning documentation to reflect current interface (#688) (@ChrisRackauckas-Claude)
• Add automatic gh auth login prompt when sharing benchmark results (#689) (@ChrisRackauckas-Claude)
• Add MIT license to LinearSolveAutotune and update auth docs (#690) (@ChrisRackauckas-Claude)

Closed issues:

• ForwardDiff.hessian produces stack overflow for ODEProblem on 3.18 (#627)
• test out ways to upload autotune plots (#681)
• Regression when using ForwardDiff.gradient and ForwardDiff.Hessian with 3.19 (#684)