Warning, /eic-opticks/third_party/optix-toolkit/README.md is written in an unsupported language. File is not indexed.
0001 Vendored NVIDIA OptiX Toolkit headers.
0002
0003 Source repository: <https://github.com/NVIDIA/optix-toolkit>
0004
0005 Pinned upstream release:
0006 <https://github.com/NVIDIA/optix-toolkit/releases/tag/v1.3.0>
0007
0008 Pinned upstream tag:
0009 `v1.3.0`
0010
0011 Compatibility backport:
0012
0013 `OptiXToolkit/ShaderUtil/vec_math.h` includes a local backport of upstream
0014 commit `7acd88bc272a23067a36dc396ca835db145882fd`, which guards CUDA 13-only
0015 `_16a` vector helpers behind `CUDART_VERSION >= 13000`. Without this guard,
0016 CUDA 12 builds fail because types such as `double4_16a` are not defined.
0017
0018 Backport reference:
0019 <https://github.com/NVIDIA/optix-toolkit/commit/7acd88bc272a23067a36dc396ca835db145882fd>
0020
0021 Backport commit:
0022 `7acd88bc272a23067a36dc396ca835db145882fd`
0023
0024 Vendored files:
0025
0026 | Local file | Upstream file |
0027 | --- | --- |
0028 | `OptiXToolkit/ShaderUtil/vec_math.h` | `ShaderUtil/include/OptiXToolkit/ShaderUtil/vec_math.h` |
0029 | `OptiXToolkit/ShaderUtil/Preprocessor.h` | `ShaderUtil/include/OptiXToolkit/ShaderUtil/Preprocessor.h` |
0030 | `LICENSE.txt` | `LICENSE.txt` |
0031
0032 `Preprocessor.h` and `LICENSE.txt` match the pinned upstream release
0033 byte-for-byte. `vec_math.h` is otherwise from the pinned upstream release plus
0034 the CUDA 12 compatibility guard described above.
0035
0036 License:
0037
0038 The upstream repository is published under `BSD-3-Clause`. The vendored
0039 headers include SPDX license identifiers, and the upstream `LICENSE.txt` is
0040 included here to preserve the BSD-3 license conditions and disclaimer required
0041 for redistribution.
0042
0043 Simphony compatibility:
0044
0045 Project-local compatibility code is kept in `sysrap/scuda.h`. That header
0046 includes `OptiXToolkit/ShaderUtil/vec_math.h` and retains only the local API
0047 surface that is not provided by the upstream OptiX Toolkit header. For CUDA 12,
0048 it also restores the legacy narrowing overloads `make_float2(float3)`,
0049 `make_float2(float4)`, and `make_float3(float4)` because the compatibility
0050 backport guards those overloads together with adjacent CUDA 13-only `_16a`
0051 helpers.
