triton_cpp v1.2.1
Header-only C++ wrapper for NVIDIA Triton Inference Server clients
Loading...
Searching...
No Matches
types.hpp File Reference
#include <algorithm>
#include <cstddef>
#include <cstdint>
#include <map>
#include <memory>
#include <stdexcept>
#include <string>
#include <type_traits>
#include <variant>
#include <vector>
#include <Eigen/Dense>
#include <eigen3/unsupported/Eigen/CXX11/Tensor>
#include <model_config.pb.h>
#include "triton_cpp/utils.hpp"

Go to the source code of this file.

Classes

struct  triton_cpp::InputData
 Couple a Triton input descriptor with its backing storage. More...
 
struct  triton_cpp::InputOutputMetaData
 Shape, datatype, and byte-size metadata for one model tensor. More...
 

Namespaces

namespace  triton_cpp
 

Typedefs

template<typename T >
using triton_cpp::VectorType
 Helper type template for Eigen::Vector compatible with Triton.
 
template<typename T >
using triton_cpp::MatrixType
 Helper type template for Eigen::Matrix compatible with Triton.
 
template<typename T , int rank>
using triton_cpp::TensorType
 Helper type template for Eigen::Tensor compatible with Triton.
 
using triton_cpp::TritonDataType
 Type alias for all possible C++ scalar types that the triton server supports.
 
using triton_cpp::ModelOutput = std::map<std::string, std::shared_ptr<triton::client::InferRequestedOutput>>
 Map model output names to Triton requested-output objects.
 

Functions

TritonDataType triton_cpp::getZero (inference::DataType type)
 Get a zero value in the correct C++ type, given a Triton datatype.
 
std::size_t triton_cpp::getAlignment (inference::DataType type)
 Return the natural C++ alignment for a Triton datatype.
 
std::size_t triton_cpp::getSharedMemoryAlignment (inference::DataType type)
 Return the alignment used when packing tensors into shared memory.
 
std::size_t triton_cpp::alignUp (std::size_t offset, std::size_t alignment)
 Round an offset up to an alignment boundary.