Safe NVIDIA FleX bindings for Rust
This is a collection of crates which provide safe idiomatic NVIDIA FleX bindings for Rust (nvflex), as well as raw bindings to the C++ FleX library (nvflex-sys).
Provides safe idiomatic Rust bindings over NVIDIA FleX's C++ API.
By default, structs don't implement Send/Sync, but you can enable that functionality by enabling the unsafe_send_sync feature. Note that FleX is NOT thread safe, meaning only one thread should ever use the API at a time.
FleX is not open source, so you have to bundle the dynamic libraries with your executable. When building, this crate will copy the required libraries into the target/debug or target/release directory.
Vector<T> is basically NvFlexVector<T> from NvFlexExt.h, rewritten in Rust. It exists even if the ext feature is disabled.
To enable NVIDIA FleX extensions (NvFlexExt), enable the ext feature.
- On Windows: D3D11/D3D12/CUDA
- On Linux: only CUDA is supported
- On Android: only CUDA is supported (you need a Tegra GPU)
We somehow need to make the Solver aware that the Library is destroyed
and it shouldn't call NvFlexDestroySolver() when it's dropped if NvFlexShutdown() already destroyed it.
This means that right now, if you create a solver, it will only be destroyed when Library goes
out of scope.
nvflex-sys provides raw NVIDIA FleX bindings for Rust, generated with bindgen.
Note that some functions (NvFlexMakePhaseWithChannels, NvFlexMakePhase, NvFlexMakeShapeFlagsWithChannels and NvFlexMakeShapeFlags) are implemented in Rust, because they are defined inside the NvFlex.h header.
- Proper comment formatting
BSL-1.0.
- Note: NVIDIA FleX is licensed under the Nvidia Source Code License