v1.4.1

Auto-published from kspacefirstorder-unified@9cd41a45.

v1.4.0 — REVOKED (see v1.4.1)

⚠️ REVOKED — do not use this release.

This release has packaging or correctness regressions that make it unusable on a significant portion of installs:

• Linux binaries (CUDA + OMP): dynamically linked CUDA + FFTW + libstdc++ and required glibc 2.38+, breaking Ubuntu 22.04 LTS / Colab / RHEL 9. See kspacefirstorder-unified#15.
• macOS OMP binary: the CMake build path passed -ffast-math, causing NaNs in absorbing-media simulations on Apple Silicon. See k-wave-omp-darwin#4.
• Windows OMP binary: switched compiler/OpenMP/FFT stack (Intel → MSVC + VCOMP + FFTW) without bundling the new runtime DLLs (VCOMP140.dll, VCRUNTIME140_1.dll, fftw3f.dll). See kspacefirstorder-unified#14.

Use v1.4.1 instead — see the latest release on this repository.

The artifact below is preserved for audit purposes.

v1.3.1

update compilation to support newer NVIDIA compute architectures.

• Now supports Maxwell (sm_50) to Hopper (sm_90a)

v1.3.0

Mirror of version 1.3 release of kspaceFirstOrder-CUDA-linux