v0.6.2

[waltsims]

Highlights

Binary refresh — Blackwell (sm_120) support on Linux CUDA

Pinned to upstream v1.4.1 C++ binaries. Blackwell support is Linux CUDA only: Windows CUDA stays on v1.3.0 pending DLL packaging, and macOS has no CUDA path.

• Linux CUDA + OMP — v1.4.1 adds sm_120 / Blackwell support. Rebuilt with static CUDA + cuFFT + FFTW + libstdc++ on ubuntu-22.04 (glibc 2.35 floor).
• macOS OMP — v1.4.1 includes the fast-math fix that prevents NaNs in absorbing-media simulations on Apple Silicon, plus the libhdf5.320.dylib ABI refresh from v1.4.0 ([BUG] hdf5 version mismatch on fresh install #661).
• Windows OMP + CUDA — both still pinned to v1.3.0; v1.4.x DLL bundling not yet ready. Windows Blackwell support is not in this release.
• Intel Mac guard — kwave/__init__.py emits a RuntimeWarning and skips the darwin OMP download on darwin x86_64 (v1.4.x darwin is arm64-only). Universal2 coverage tracked for v0.6.5.
• Chmod fix — download_binaries now sets the exec bit on Linux/macOS (Set exec bit on Linux/macOS binaries after download #741).

Solver & API fixes

• alpha_mode is now respected by the C++ backend, and unsupported combinations are rejected instead of silently ignored ([BUG] Numerical instabilities when medium.alpha_coeff is close to 1 #664, Fix #664: respect alpha_mode + reject silent C++ ignore #712).
• C++ source flags now write signal length instead of a boolean (Fix cpp source flags in new API #698).
• voxel_plot: swapped x and z axes ([BUG]Voxel Plot switched axes #501, Fix #501: voxel_plot swapped x and z axes #734).
• t_array setter no longer raises ValueError when passed a numpy array (fix: fix t_array setter ValueError when passing numpy array #732).
• binary_path is no longer dropped by options_to_kwargs() (Fix binary_path silently dropped by options_to_kwargs() #726).
• Added dtype parameter to kspaceFirstOrder() ([ENHANCE] Add a parameter to kspaceFirstOrder() to control the data type that variables are cast to before computation #695, Add dtype parameter to kspaceFirstOrder() (#695) #716).
• Added AlphaMode enum (Add AlphaMode enum for stringly-typed alpha_mode field #705).

Internals

• step() is branch-free; lambdas are bound at setup (Make step() branch-free by binding lambdas at setup #715).
• _setup_physics_operators split into four single-purpose methods (Refactor _setup_physics_operators into four single-purpose methods #713).
• matlab_find replaced with NumPy in mapgen (Replace matlab_find with numpy functions in mapgen #609).
• kWaveMedium type annotations and test coverage (Fix kWaveMedium type annotations (rebased #636 + Greptile fixes) #736).
• Pinned sphinx<9 to fix Read the Docs (Pin sphinx<9 in docs extras to fix Read the Docs build #714).
• gaussian cleanup using np.pi (Use np.pi and remove redundant variable in gaussian function #717).
• Added CLAUDE.md (Add CLAUDE.md for Claude Code onboarding #696).

Closes

#501, #622, #656, #661, #664, #695, #738, #740

Out of scope

Darwin x86_64 (Intel Mac) coverage — tracked for v0.6.5.

Thanks

• @aconesac — original Blackwell report (#656) and Linux CUDA v1.4.1 validation on Blackwell hardware (#739).
• @elma16 — macOS OMP fast-math fix (k-wave-omp-darwin#4).

---

What's Changed

• Add CLAUDE.md for Claude Code onboarding by @waltsims in Add CLAUDE.md for Claude Code onboarding #696
• Fix cpp source flags in new API by @waltsims in Fix cpp source flags in new API #698
• Bump requests from 2.33.0 to 2.33.1 by @dependabot[bot] in Bump requests from 2.33.0 to 2.33.1 #699
• Bump jaxtyping from 0.3.2 to 0.3.7 by @dependabot[bot] in Bump jaxtyping from 0.3.2 to 0.3.7 #703
• Bump deepdiff from 8.6.2 to 9.0.0 by @dependabot[bot] in Bump deepdiff from 8.6.2 to 9.0.0 #702
• Bump pytest from 9.0.2 to 9.0.3 by @dependabot[bot] in Bump pytest from 9.0.2 to 9.0.3 #711
• Add AlphaMode enum for stringly-typed alpha_mode field by @waltsims in Add AlphaMode enum for stringly-typed alpha_mode field #705
• Bump pillow from 12.1.1 to 12.2.0 by @dependabot[bot] in Bump pillow from 12.1.1 to 12.2.0 #709
• Bump gdown from 5.2.0 to 5.2.2 by @dependabot[bot] in Bump gdown from 5.2.0 to 5.2.2 #710
• Bump coverage from 7.10.6 to 7.13.5 by @dependabot[bot] in Bump coverage from 7.10.6 to 7.13.5 #707
• Bump furo from 2024.8.6 to 2025.12.19 by @dependabot[bot] in Bump furo from 2024.8.6 to 2025.12.19 #706
• Bump matplotlib from 3.10.7 to 3.10.8 by @dependabot[bot] in Bump matplotlib from 3.10.7 to 3.10.8 #708
• Pin sphinx<9 in docs extras to fix Read the Docs build by @waltsims in Pin sphinx<9 in docs extras to fix Read the Docs build #714
• Fix [BUG] Numerical instabilities when medium.alpha_coeff is close to 1 #664: respect alpha_mode + reject silent C++ ignore by @waltsims in Fix #664: respect alpha_mode + reject silent C++ ignore #712
• Refactor _setup_physics_operators into four single-purpose methods by @waltsims in Refactor _setup_physics_operators into four single-purpose methods #713
• Make step() branch-free by binding lambdas at setup by @waltsims in Make step() branch-free by binding lambdas at setup #715
• Bump gdown from 5.2.2 to 6.0.0 by @dependabot[bot] in Bump gdown from 5.2.2 to 6.0.0 #720
• Bump h5py from 3.15.1 to 3.16.0 by @dependabot[bot] in Bump h5py from 3.15.1 to 3.16.0 #719
• Bump pre-commit from 4.5.1 to 4.6.0 by @dependabot[bot] in Bump pre-commit from 4.5.1 to 4.6.0 #718
• Use np.pi and remove redundant variable in gaussian function by @maayanmatsliah-tech in Use np.pi and remove redundant variable in gaussian function #717
• Bump urllib3 from 2.6.3 to 2.7.0 by @dependabot[bot] in Bump urllib3 from 2.6.3 to 2.7.0 #724
• Bump coverage from 7.13.5 to 7.14.0 by @dependabot[bot] in Bump coverage from 7.13.5 to 7.14.0 #729
• Bump requests from 2.33.1 to 2.34.0 by @dependabot[bot] in Bump requests from 2.33.1 to 2.34.0 #728
• Bump matplotlib from 3.10.8 to 3.10.9 by @dependabot[bot] in Bump matplotlib from 3.10.8 to 3.10.9 #727
• Bump mistune from 3.2.0 to 3.2.1 by @dependabot[bot] in Bump mistune from 3.2.0 to 3.2.1 #723
• Fix binary_path silently dropped by options_to_kwargs() by @Copilot in Fix binary_path silently dropped by options_to_kwargs() #726
• Bump sphinx-toolbox 3.8.0 -> 4.1.2 and sphinx-tabs 3.4.7 -> 3.4.5 by @waltsims in Bump sphinx-toolbox 3.8.0 -> 4.1.2 and sphinx-tabs 3.4.7 -> 3.4.5 #733
• Fix [BUG]Voxel Plot switched axes #501: voxel_plot swapped x and z axes by @waltsims in Fix #501: voxel_plot swapped x and z axes #734
• fix: fix t_array setter ValueError when passing numpy array by @wangwangbobo in fix: fix t_array setter ValueError when passing numpy array #732
• Replace matlab_find with numpy functions in mapgen by @faridyagubbayli in Replace matlab_find with numpy functions in mapgen #609
• Add dtype parameter to kspaceFirstOrder() ([ENHANCE] Add a parameter to kspaceFirstOrder() to control the data type that variables are cast to before computation #695) by @waltsims in Add dtype parameter to kspaceFirstOrder() (#695) #716
• Fix kWaveMedium type annotations (rebased fix: correct type annotations in kWaveMedium class #636 + Greptile fixes) by @waltsims in Fix kWaveMedium type annotations (rebased #636 + Greptile fixes) #736
• Set exec bit on Linux/macOS binaries after download by @waltsims in Set exec bit on Linux/macOS binaries after download #741
• Bump BINARY_VERSION to v1.4.1 (sm_120 / Blackwell) by @waltsims in Bump BINARY_VERSION to v1.4.1 (sm_120 / Blackwell) #739

New Contributors

• @maayanmatsliah-tech made their first contribution in Use np.pi and remove redundant variable in gaussian function #717
• @wangwangbobo made their first contribution in fix: fix t_array setter ValueError when passing numpy array #732

Full Changelog: v0.6.1...v0.6.2

---

This discussion was created from the release v0.6.2.