Deeploy.DeeployTypes

Classes

CodeGenVerbosity(tilingProfiling)

Encapsulates verbosity options for downstream configuration

CodeSnippet(template, operatorRepresentation)

A dataclass to hold a NodeTemplate and its associated OperatorRepresentation; used to generate code

CodeTransformation(passes)

Wrapper object to run multiple CodeTransformations sequentially

CodeTransformationPass()

Pass Object to update code generation; may either modify an executionBlock's existing code snippets or add new code snippets to an executionBlock

ConstantBuffer([name, shape, values])

Class to represent compile-time constant tensors (weights, biases, other parameters) within Deeploy.

DeploymentEngine(name, Mapping[, initCode, ...])

Deeploy abstraction to represent a compute engine without a complete host system, like an accelerator

DeploymentPlatform(engines, variableBuffer, ...)

Deeploy abstraction for a complete system, including at least a host core capable of memory allocation

ExecutionBlock([operatorCodeSnippet])

Deeploy abstraction to represent a operator whose kernel has been determined.

GlobalDefinition(name, definition)

Helper class to hoist arbitrary C code into the global program scope; used to perform small amounts of global initialization, declare global synchronization objects, and similar.

NetworkContainer(graph, platform, ...)

Deeploy abstraction for containing the information needed to describe a complete neural network to be deployed

NetworkContext(variableBuffer, ...[, ...])

The global context of the compiler.

NetworkDeployer(graph, deploymentPlatform, ...)

Deeploy abstraction to contain an entire network and all necessary information to deploy it

NetworkOptimizationPass()

Pass to update the NetworkContext and Neural Network Graph in one go

NetworkOptimizer(passes)

Wrapper class to run multiple NetworkOptimizationPasses sequentially

NodeBinding(typeChecker, template, ...)

Deeploy's class to bind individual NodeTypeChecker objects to NodeTemplate and associate a CodeTransformation.

NodeMapper(parser, bindings)

Deeploy class to link a NodeParser and several NodeBindings

NodeParser()

Deeploy's core Parser class.

NodeTemplate(templateStr)

This class wraps a Mako.Template with additional functionality for hoisting transient buffers and adding expressions to the parsers' node representation

NodeTypeChecker(input_types, output_types)

Implements type checking according to user-defined rules to assign Deeploy-types to the Python-typed input graph

ONNXLayer(maps)

Deeploy abstraction to represent one operator in an ONNX graph

StructBuffer(name, structDict)

Class to represent Struct object needed by the generated C Code

TopologyOptimizationPass()

Abstract pass object which modifies an ONNX graph

TopologyOptimizer(passes)

Wrapper object to apply multiple TopologyOptimizationPasses sequentially

TransientBuffer([name, size])

Class to represent memory space required by kernels that is not covered by input and output tensors, e.g. im2col buffers in convolutions.

VariableBuffer([name, shape])

This class represents memory locations containing variable tensor data that is not transient, i.e. intermediate results or input- and output buffers.