Metadata-Version: 2.4
Name: starwave
Version: 6.0.0
Summary: Scalar, VRZ, VTI CUDA and independent 2D/3D elastic CPU/CUDA propagation for PyTorch
License-Expression: MIT AND LicenseRef-NVIDIA-CUDA-11.8 AND (GPL-3.0-or-later WITH GCC-exception-3.1)
Requires-Python: <3.13,>=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
License-File: LICENSE.NVIDIA-CUDA
License-File: LICENSE.GCC-RUNTIME
License-File: THIRD_PARTY_NOTICES.md
License-File: native/LICENSE.SWEEP
License-File: cpu_validation/reference/LICENSE
License-File: starwave/_deepwave/LICENSE.Deepwave.txt
Requires-Dist: torch<2.6,>=2.5
Requires-Dist: numpy>=1.23
Dynamic: license-file

# StarWave 6.0.0 binary distribution candidate

Public binary version 6.0.0 maps to maintained source 0.1.0.dev13 (V13).
The exact private source commit, source manifest, core/elastic input fingerprints,
and post-link library hashes are bound in the binary receipt. This candidate
is not a publication or GPU-acceptance claim.

V13 integrates elastic 2D/3D full/boundary whole-batch execution without L2
trajectory grouping, universal direct-load 32x8 Scalar2D optimization, and two
Scalar2D PML adjoint corrections. Scalar2D native semantics 2/capabilities 15
and Scalar3D semantics 2/capabilities 15 are enforced; core ABI remains 2.
The final source's elastic CUDA launch-size checks are preserved. Public API,
defaults, source/time/CFL and model-padding-gradient conventions are unchanged.
Both native libraries are freshly built. Restart Python after upgrading.

## Compatibility target

- Linux x86_64, glibc 2.35+; GLIBCXX <=3.4.30 and CXXABI <=1.3.13 requirements
- Python 3.10-3.12, PyTorch 2.5.x and NumPy >=1.23
- CUDA Runtime 11.8.89 statically linked; no CUDA compiler or driver is bundled
- SASS sm70/75/80/86/89/90 and compute80 PTX verified separately in both libraries
- Elastic CPU OpenMP enabled with the exact reviewed isolated libgomp runtime

These remain build/install targets until the exact wheel passes its gates.
CUDA compile/link, host ABI loading, and CPU installation tests do not establish
GPU correctness, multi-GPU scaling, long-running FWI convergence or speed.

## Contents and loading

The wheel contains readable Python API/runtime modules, two precompiled native
libraries, an isolated OpenMP runtime, integrity metadata and required licenses.
Native C/CUDA/C++ source and headers, build programs, private references, source
manifests, tests, notebooks, utils, models, data and results are excluded.
Source distributions are disabled. Import does not compile or load native code.
Python scientific modules are copied byte-for-byte from frozen V13; only package
path and loader-integrity adaptations are applied separately for binary loading.

Use starwave.prepare_native([0]) for scalar/VRZ/VTI or
starwave.prepare_elastic([0]) for elastic GPU work, with explicit visible logical
GPU IDs. starwave.prepare_elastic() loads elastic CPU capability without running
a propagation. Runtime receipts enforce exact module/library hashes and native
ABI/capability/export compatibility. These unsigned hashes detect corruption or
mixed revisions; PyPI publisher attestations establish publication provenance.

Elastic accepts raw lamb, mu and buoyancy; convert vp/vs/rho explicitly with
starwave.common. It retains [y,x]/[z,y,x] axes and all 16/31 returns, full CPU/CUDA
memory modes and CUDA-only boundary memory. The inherited model-gradient sampling
edge case with zero complete groups remains unsupported. Boundary memory does
not guarantee a full survey fits a GPU. Scalar2D PML repairs do not introduce
complete replicated-padding model derivatives. See the maintained V13 source
notes for exact API, limitations and separate numerical acceptance evidence.

## Licenses

StarWave, SWEEP-derived code and Deepwave-derived code retain their original MIT
notices. CUDA Runtime is redistributed under its bundled CUDA 11.8 license.
Bundled libgomp retains GPL-3.0-or-later WITH GCC-exception-3.1, full notices and
exact corresponding-source download directions. No license is inferred for
excluded private datasets, models, notebooks, utils or results.
