Struct async_tensorrt::builder::Builder

pub struct Builder { /* private fields */ }

Builds an engine from a network definition.

TensorRT documentation

Implementations

impl Builder

pub async fn new() -> Self

Create a new Builder.

TensorRT documentation

pub fn add_optimization_profile(&mut self) -> Result<(), Error>

Create a new optimization profile.

Note that the official TensorRT documentation states:

“If the network has any dynamic input tensors, the appropriate calls to setDimensions() must be made.”

But this part of the API has not been implemented yet in async-tensorrt.

This function is still useful to allocate the optimization profile in the Builder, which may or may not actually affect the building process later.

TensorRT documentation

pub fn with_optimization_profile(self) -> Result<Self, Error>

Create a new optimization profile.

Note that the official TensorRT documentation states:

“If the network has any dynamic input tensors, the appropriate calls to setDimensions() must be made.”

But this part of the API has not been implemented yet in async-tensorrt.

This function is still useful to allocate the optimization profile in the Builder, which may or may not actually affect the building process later.

TensorRT documentation

pub async fn config(&mut self) -> BuilderConfig

Create a builder configuration object.

TensorRT documentation

Return value

A BuilderConfig that can later be passed to build_serialized_network.

pub fn network_definition( &mut self, flags: NetworkDefinitionCreationFlags ) -> NetworkDefinition

Create a network definition object.

TensorRT documentation

Arguments

• flags - Flags for specifying network properties.

pub async fn build_serialized_network( &mut self, network_definition: &mut NetworkDefinition, config: BuilderConfig ) -> Result<HostBuffer, Error>

Builds and serializes a network for the provided crate::ffi::network::NetworkDefinition and BuilderConfig.

TensorRT documentation

Arguments

• network_definition - Network definition.
• config - Builder configuration.

pub fn platform_has_fast_int8(&self) -> bool

Determine whether the platform has fast native INT8.

TensorRT documentation

pub fn platform_has_fast_fp16(&self) -> bool

Determine whether the platform has fast native FP16.

TensorRT documentation