tensorflow Namespace Reference

Namespaces
	tfcompile
	Namespace of tfcompile.

Classes
class	XlaCompiler
class	XlaOpRegistry

Functions
Status	ConvertGraphDefToXla (const GraphDef &graph_def, const tf2xla::Config &config, xla::Client *client, xla::Computation *computation)
	Convert tensorflow graph_def to XLA UserComputation. More...
Status	ValidateConfig (const tf2xla::Config &config)
Status	AddPlaceholdersForFeeds (const tf2xla::Config &config, const OpRegistryInterface *op_registry, std::unordered_map< string, string > *feed_remapping, GraphDef *graph_def)
	Generate corresponding placeholder for each feed tensor. Traverse each inputs of each nodes. Replace feeds to corresponding placeholders. More...
Status	PruneGraphDefInto (const tf2xla::Config &config, const GraphDef &in, GraphDef *out)
	Remove source nodes of feed tensor. Because they are removed at tensorflow::AddPlaceholdersForFeeds More...
Status	AddDefaultAttrsToGraphDef (GraphDef *graph_def, const OpRegistryInterface &op_registry, int node_offset)
	Assign each NodeDefs of GraphDef default values. More...
Status	ConvertGraphDefToGraph (const GraphConstructorOptions &opts, const GraphDef &gdef, Graph *g)
	Convert GraphDef defined protobuf to type Graph More...

Detailed Description

Namespace of tensorflow

namespace of tensorflow

Function Documentation

AddDefaultAttrsToGraphDef()

Status tensorflow::AddDefaultAttrsToGraphDef	(	GraphDef *	graph_def,
		const OpRegistryInterface &	op_registry,
		int	node_offset
	)

Assign each NodeDefs of GraphDef default values.

Google Docs:

Adds default attributes to NodeDefs in 'graph_def' starting from the 'node_offset' node in 'graph_def'.

Default attributes are defined by 'op_registry'.

Returns OK on success, an error if 'graph_def' has a NodeDef that cannot be found in 'op_registry'.

REQUIRES: 'graph_def' and 'op_registry' are not nullptr.

Here is the caller graph for this function:

AddPlaceholdersForFeeds()

Status tensorflow::AddPlaceholdersForFeeds	(	const tf2xla::Config &	config,
		const OpRegistryInterface *	op_registry,
		std::unordered_map< string, string > *	feed_remapping,
		GraphDef *	graph_def
	)

Generate corresponding placeholder for each feed tensor. Traverse each inputs of each nodes. Replace feeds to corresponding placeholders.

Modifies <graph_def> to include placeholders for each fed tensor, and update references to the fed tensors to refer to the placeholders. The existing nodes referenced by the feeds are not removed or modified (except where their input edges are modified by the replacement of other feeds).

Here is the call graph for this function:

Here is the caller graph for this function:

ConvertGraphDefToGraph()

Status tensorflow::ConvertGraphDefToGraph	(	const GraphConstructorOptions &	opts,
		const GraphDef &	gdef,
		Graph *	g
	)

Convert GraphDef defined protobuf to type Graph

As brief

Here is the caller graph for this function:

ConvertGraphDefToXla()

Status tensorflow::ConvertGraphDefToXla	(	const GraphDef &	graph_def,
		const tf2xla::Config &	config,
		xla::Client *	client,
		xla::Computation *	computation
	)

Convert tensorflow graph_def to XLA UserComputation.

1. Call tensorflow::anonymous_namespace{tf2xla.cc}::InitGraph
2. Call tensorflow::anonymous_namespace{tf2xla.cc}::ConvertGraphToXla

Here is the call graph for this function:

Here is the caller graph for this function:

PruneGraphDefInto()

Status tensorflow::PruneGraphDefInto	(	const tf2xla::Config &	config,
		const GraphDef &	in,
		GraphDef *	out
	)

Remove source nodes of feed tensor. Because they are removed at tensorflow::AddPlaceholdersForFeeds

Returns in <out> a copy of <in>, pruned to only include fetches from <config>.

Here is the caller graph for this function:

ValidateConfig()

Status tensorflow::ValidateConfig

(

const tf2xla::Config &

config

)

Check settings of input config(feed & fetch). For example, if id, shape are legal, duplicated names...

Here is the caller graph for this function: