tenzr::ExecutionResult¶

Result of executing a graph, holding device-resident buffers plus lazily materialized host storage for its outputs.

#include <execution_engine.h>

Public Functions¶

Public Attributes¶

Public Functions Documentation¶

function EnsureHostBuffer¶

void EnsureHostBuffer() const

Materialize the pinned host buffer from the device buffer when not already staged.

Lazily copy the device output into a pooled host buffer.

function ToVector¶

template <typenameT >
inline std::vector< T > ToVector() const

Template Parameters:

• T Requested element type; must match the result dtype.

Return: Host vector containing the result elements.

Copy the result into a host std::vector.

function AsArray¶

Array AsArray(
    std::string name =""
) const

Parameters:

• name Symbolic name for the resulting Array.
• name Name assigned to the produced variable.

Return:

• An Array view of the result.
• Array referencing the execution result.

Wrap the result as an Array bound to an optional name. Wrap the result in an Array variable node carrying the device and/or host buffer.

function AsTensor¶

inline Array AsTensor(
    std::string name =""
) const

Alias for as_array.

function operator Array¶

inline operator Array() const

Public Attributes Documentation¶

variable device_buffer¶

std::shared_ptr< DeviceBuffer > device_buffer;

Device-resident buffer holding the result.

variable output_buffer¶

std::shared_ptr< PinnedBuffer > output_buffer;

variable shape¶

Shape shape;

variable dtype¶

DType dtype;

variable api¶

constPJRT_Api * api {nullptr};

variable device_buffers¶

std::vector< std::shared_ptr< DeviceBuffer > > device_buffers;

All device-resident outputs; index 0 mirrors device_buffer.

variable shapes¶

std::vector< Shape > shapes;

Shapes of every output; mirrors the executable's out_shapes.

variable dtypes¶

std::vector< DType > dtypes;

DTypes of every output.