v1.0.3

Amalthea.jl v1.0.3

GPU geometry expansion and portable-installation update for the native
backend.

Highlights

• Added explicit CUDA-resident support across additional radial, modal, and
  free-space Kerr, Raman, PPT, and ADK configurations on RealGrid and
  EnvGrid.
• Corrected oversampled EnvGrid spectral-half handling in modal and
  free-space paths and released the free-space c2c cuFFT plan during teardown.
• Made normal package and release builds CPU-only by default. CUDA is optional
  and enabled explicitly from source with AMALTHEA_CUDA_BUILD=required.
• Added exact OS/architecture prebuilt selection and the first Linux ARM64
  release binary and native ARM64 installation/FFI CI gate.
• Added a cross-platform installation and configuration guide for Linux,
  macOS, Windows, ARM64, source builds, CPU-only hosts, and optional CUDA.

Installation

Amalthea is not registered in Julia General. Install this release directly
from GitHub:

pkg> add https://github.com/vdiego28/Amalthea.jl#v1.0.3

Published CPU-only binaries cover Linux x86_64, Linux AArch64, macOS Apple
Silicon, and Windows x86_64. SHA256SUMS.txt authenticates all four assets.
Other platforms compile from source.

See the repository CHANGELOG
and full commit comparison.

v1.0.2

Full Changelog: v1.0.1...v1.0.2

v1.0.1

What's Changed

• build(deps): bump actions/github-script from 7 to 9 by @dependabot[bot] in #61
• build(deps): bump actions/checkout from 5 to 7 by @dependabot[bot] in #63
• build(deps): bump softprops/action-gh-release from 2 to 3 by @dependabot[bot] in #59
• build(deps): bump actions/download-artifact from 4 to 8 by @dependabot[bot] in #60
• build(deps): bump actions/upload-artifact from 4 to 7 by @dependabot[bot] in #62

Full Changelog: v1.0.0...v1.0.1

Amalthea.jl v1.0.0

Important

Historical correction (2026-08-10): The original release text overstated compatibility and hardware dispatch. Amalthea intends to retain Luna-compatible APIs and tests its Julia/native paths for numerical equivalence, but it is a moving hard fork rather than an unconditional backwards-compatibility guarantee. dispatch.rs was detection-only and did not automatically dispatch propagation through CUDA, Vulkan, and CPU backends; no Vulkan propagation implementation existed. This release also incorrectly said Amalthea was registered in Julia General. It was not and remains installed from GitHub.

---

Amalthea.jl v1.0.0

First stable release, and first release under the new name: this package was
previously developed as Luna-Rust.jl, a performance-focused fork of
Luna.jl. The Julia API remains fully
backwards-compatible with Luna.jl; performance-critical numerical kernels are
offloaded to a native Rust backend (luna-rust), called transparently via
ccall — no Rust knowledge is required to use the package.

Added

• Native-Rust resident stepper (RustNativeStepper / NativeSim): the
  entire RK45 hot loop — field, RK scratch buffers, and FFTW plans — lives in
  Rust for the duration of a solve, eliminating the per-stage Julia
  callback round-trip. Covers mode-averaged, radial, modal, and free-space
  geometries; RealGrid and EnvGrid; Kerr, plasma (PPT/ADK), and Raman
  (including :SiO2 intermediate-broadening) nonlinearities; gas mixtures;
  z-dependent (graded-core, tapered, multi-point gradient) linear operators;
  and shot noise. Falls back to the Julia stepper automatically for any
  configuration outside this scope (NativeIneligible).
• Runtime hardware dispatch: automatic selection of CUDA → Vulkan →
  AVX-512/Apple AMX → AVX2/NEON → portable scalar, including an opt-in
  GPU-resident backend (LUNA_USE_RUST_CUDA_NATIVE=1).
• Per-kernel Rust acceleration (opt-in via LUNA_USE_RUST_* toggles,
  used independently of the resident stepper): PPT ionisation rate,
  time-domain Raman (ADE exponential integrator), Zeisberger/Marcatili
  dispersion, and QDHT batch transforms.
• Python bindings (python/, juliacall-based, pip-installable) with a
  numpy-backed output wrapper and ASCII/Unicode keyword translation
  (lambda0 ↔ λ0, etc.).
• Prebuilt binary releases: tagged releases publish libluna_rust for
  Linux, macOS, and Windows; deps/build.jl downloads the matching prebuilt
  library for the installed version before falling back to a local
  cargo build.
• Cross-platform scan-queue locking (flock on Unix, LockFileEx on
  Windows), validated by CI on all three platforms.

Changed

• Package renamed to Amalthea.jl (previously Luna-Rust.jl, itself a fork
  of Luna.jl); citation metadata updated (Zenodo DOI), new package UUID
  minted, registered as a new package in the Julia General registry.

Fixed

• Polarisation.ellipse angle calculation (was always 0; now
  angle(Q + 1im*U)/2), and other fork-vs-upstream parity fixes — see
  docs/dev/REVIEW.md for the full audit.
• Shell/command-injection hardening in Scans.jl's SSH/Slurm/Condor
  submission paths (Cmd arrays instead of interpolated shell strings).

---

See CHANGELOG.md for the full changelog and
docs/dev/BACKLOG.md / docs/dev/native-port/PORT_LOG.md
for the full phase-by-phase native-port development history.

What's Changed

• Bump julia-actions/cache from 2 to 3 by @dependabot[bot] in #3
• Bump criterion from 0.5.1 to 0.8.2 in /luna-rust by @dependabot[bot] in #4
• 🧪 Add tests for check_spline_args error path by @vdiego28 in #23
• ⚡ Optimize array slicing with @views in Stats.jl by @vdiego28 in #25
• ⚡ Optimize RK45 interpolator array allocations by @vdiego28 in #26
• 🧪 Test error path in Utils.jl at line 131 by @vdiego28 in #27
• 🧪 Add unit tests for scans::FlockLock::new by @vdiego28 in #28
• 🧪 Add unit test for d2n_dω2 in dispersion.rs by @vdiego28 in #29
• ⚡ Optimize QDHT transform matrix population by @vdiego28 in #31
• 🧪 test(dispersion): add analytical derivative tests for SellmeierGas by @vdiego28 in #33
• 🧪 Add comprehensive tests for RectModes.jl by @vdiego28 in #37
• ⚡ Optimize Utils subscript functions to reduce array allocations by @vdiego28 in #39
• ⚡ Optimize string join functions in NonlinearRHS.jl to prevent temporary array allocations by @vdiego28 in #40
• ⚡ Optimize makeinputs using lazy generator by @vdiego28 in #43
• 🧪 Add test coverage for dopri.jl integrator and Butcher Tableau by @vdiego28 in #44
• ⚡ perf: Use generator in tuple construction for scanshape by @vdiego28 in #46
• ⚡ perf: optimize any allocation in doinput_sm by @vdiego28 in #47
• Parallelize pointcalc! with Threads.@threads by @vdiego28 in #48
• 🧹 Refactor sqrt handling in Sellmeier equations (PhysData.jl) by @vdiego28 in #50
• 🧹 Fix actionable TODO for branch-cut adjustment of SiO2 refractive index calculation by @vdiego28 in #21
• 🧪 Add test coverage for fftfreq and rfftfreq in Maths.jl by @vdiego28 in #22
• ⚡ Optimise fwhm and level_xings Memory Allocations by @vdiego28 in #24
• 🧪 Add tests for SellmeierGas dispersion functions by @vdiego28 in #30
• 🧪 [testing improvement] Add tests for translate_kwargs in python API by @vdiego28 in #32
• 🧪 Add unit test for translate_kwargs error path by @vdiego28 in #34
• 🧪 Add test for LunaOutput.getitem error path by @vdiego28 in #35
• 🧪 Add test coverage for Grid.jl by @vdiego28 in #36
• 🧪 add SimpleFibre tests by @vdiego28 in #38
• 🧪 Add test for arrivaltime error path by @vdiego28 in #10
• 🔒 Fix command injection in Scans job execution by @vdiego28 in #41
• ⚡ Use @views to prevent array allocation in Maths.jl by @vdiego28 in #52
• 🔒 Fix Arbitrary Code Execution vulnerability in Scans UnitRange parsing by @vdiego28 in #42
• Bump actions/setup-python from 4 to 6 by @dependabot[bot] in #5
• 🧹 [Code Health] Parallelize pointcalc! loop in NonlinearRHS.jl by @vdiego28 in #8
• ⚡ Fix redundant array allocation in Stats array slicing by @vdiego28 in #9
• ⚡ Hoist idcs allocation out of save loops in Output.jl by @vdiego28 in #19
• ⚡ Optimize dense RK45 output interpolation to eliminate allocations by @vdiego28 in #53
• build(deps): bump actions/checkout from 6 to 7 by @dependabot[bot] in #55
• build(deps): bump windows-sys from 0.52.0 to 0.61.2 in /luna-rust by @dependabot[bot] in #56
• [ImgBot] Optimize images by @imgbot[bot] in #57

New Contributors

• @dependabot[bot] made their first contribution in #3
• @vdiego28 made their first contribution in #23
• @imgbot[bot] made their first contribution in #57

Full Changelog: v0.7.0...v1.0.0

v0.7.0 - Rust-Accelerated Backend and Python Wrapper

Release v0.7.0 — Rust-Accelerated Backend & Python Wrapper

This release introduces major performance and usability improvements to the Luna simulation suite, including a high-performance Rust core, a native Python wrapper, and customizable material properties.

What's New

Rust-Accelerated Backend (luna-rust)

• Migrated core waveguide solvers and numerical integration algorithms to a stateless, cache-aligned Rust codebase.
• Added multithreaded diffraction (QDHT) solver support via rayon.
• Implemented zero-copy C-FFI memory boundaries for seamless in-place data transfer between Julia and Rust.

Native Python API

• Released a pip-installable python package luna-rust under python/.
• Embedded Julia interface calls using juliacall and juliapkg.
• Built a zero-copy NumPy array wrapper LunaOutput to convert simulation output vectors on-the-fly.

Dynamic Material Properties

• Added a global material registry with runtime registration APIs (register_material!) for gases and glasses.
• Enabled loading custom Sellmeier configurations from TOML files without modifying package source code.

Spelled-Out Greek Keyword Aliases

• Supported ASCII keyword arguments (e.g., lambda0, tau_fwhm, phi) interchangeably with Unicode keywords (λ0, τfwhm, ϕ) across all pulse constructors and entrypoints.

Parallelized CI Pipeline

• Restructured the 31-test-set suite into 5 parallel job groups (physics, simulation, io, fields, rust), reducing CI wall-clock validation times from 32 minutes down to ~8 minutes.