Struct TrtActor 

pub struct TrtActor { /* private fields */ }

TrtActor — owns nothing across messages besides the FFI runtime/builder handles, all engines/contexts ride the messages.

The actor itself is intentionally minimal: most of the heavy state lives in Arc<TrtEngine> values that the caller threads through. This mirrors DeviceActor’s design where per-context state lives in the ContextActor but engines live with the caller.

Implementations

impl TrtActor

pub fn new() -> Self

pub fn ensure_runtime(&self) -> Result<(), TrtError>

Get-or-create the cached runtime. Without tensorrt-link the inner constructor returns NotLinked.

pub fn execute( &self, engine: &Arc<TrtEngine>, bindings: &[(String, u64)], input_shapes: &[(String, Vec<i32>)], _stream: &Arc<CudaStream>, ) -> Result<(), TrtError>

Phase 4.5++ — synchronous helper that drives the TrtMsg::Execute semantics (creates an IExecutionContext, binds tensor addresses, calls enqueueV3).

Without tensorrt-link this returns TrtError::NotLinked without ever touching libnvinfer. With the feature on, the actor performs the full FFI sequence under the supplied Arc<CudaStream>. The function returns once the launch has been submitted — real GPU completion is observed downstream (the typical caller pairs this with an atomr-accel-cuda completion strategy on the same stream).

pub fn build_from_onnx( &self, _onnx_bytes: &[u8], _config: &IBuilderConfig, ) -> Result<EnginePlan, TrtError>

Phase 4.5++ — synchronous helper that drives the TrtMsg::BuildFromOnnx semantics. Parses an ONNX model and returns a serialised plan blob ready for TrtRuntime::deserialize. Gated on tensorrt-onnx (transitively tensorrt-link).

Trait Implementations

impl Default for TrtActor

fn default() -> Self

Returns the “default value” for a type.