baracuda-tensorrt-sys

Raw FFI bindings + dynamic loader for NVIDIA TensorRT — the high-performance inference runtime that consumes pre-built engine blobs and executes them on GPU.

Symbols resolve lazily via libloading; no link-time dependency on libnvinfer.so / nvinfer64_*.dll.

Most users want baracuda-tensorrt — that crate exposes typed Runtime, Engine, ExecutionContext, Logger, and tensor-binding APIs in idiomatic Rust.

Scope: runtime side only

TensorRT's builder (network construction, optimization passes, plan serialization) is C++-only by NVIDIA's design — there is no stable C ABI for it. Use trtexec or the Python bindings to produce engine blobs, then load them through this crate at inference time.

The runtime side has a clean C ABI (libnvinfer's _C symbols), which is what this crate wraps:

• Runtime create / destroy.
• Engine deserialize / inspect (IO names, shapes, dtypes, optimization profiles) / serialize back to bytes.
• ExecutionContext create with allocation strategy, set input shape, set tensor address, enqueueV3.
• Logger callback registration.

Part of the baracuda workspace.

License

Dual MIT / Apache-2.0.