Struct TensorRtBackend 

pub struct TensorRtBackend {
    pub inference_metrics: InferenceMetrics,
    pub precision_policy: PrecisionPolicy,
    pub batch_config: BatchConfig,
    /* private fields */
}

Stub TensorRT/CUDA ORT inference backend.

Fields

inference_metrics: InferenceMetrics

Atomic inference latency and frame count metrics.

precision_policy: PrecisionPolicy

Precision policy used when building the TensorRT EP session.

batch_config: BatchConfig

Batch configuration.

Implementations

impl TensorRtBackend

pub fn new( model_path: PathBuf, ctx: Arc<GpuContext>, device_id: i32, ring_size: usize, downstream_capacity: usize, ) -> Self

pub fn with_precision( _model_path: PathBuf, _ctx: Arc<GpuContext>, _device_id: i32, _ring_size: usize, _downstream_capacity: usize, precision_policy: PrecisionPolicy, batch_config: BatchConfig, ) -> Self

pub async fn ring_metrics(&self) -> Option<RingMetricsSnapshot>

pub fn selected_provider(&self) -> Option<&str>

Trait Implementations

impl UpscaleBackend for TensorRtBackend

fn initialize<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Warm up the backend: load the model, allocate buffers, build engine plans.

fn process<'life0, 'async_trait>( &'life0 self, _input: GpuTexture, ) -> Pin<Box<dyn Future<Output = Result<GpuTexture>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Run a single upscale pass on the given GPU texture, returning the upscaled output.

fn shutdown<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Flush any pending work, synchronize streams, and release GPU resources.

fn metadata(&self) -> Result<&ModelMetadata>

Return the model metadata extracted during initialize.