sdfg::passes::LibNodeExpander Class Reference

#include <sdfg/passes/expansion/lib_node_expander.h>

Inheritance diagram for sdfg::passes::LibNodeExpander:

Classes
struct	ExpandOutcome
class	AccessNodeExpand
class	ExpandContext

Public Types
enum class	InputUse {   Skip = 0 , Scalar , IndirectRead , IndirectWrite ,   IndirectReadWrite }

Public Member Functions
virtual	~LibNodeExpander ()=default
virtual const LibNodeExpander *	for_lib_node (const data_flow::LibraryNode &node) const
	Prefiltering of libnodes.
virtual const data_flow::LibraryNodeCode *	matched_code () const
	If this expander dispatches on a single, fixed library node code, expose it here.
virtual ExpandOutcome	handle_expand (ExpandContext &context, structured_control_flow::Block &block, data_flow::LibraryNode &node) const =0

Member Enumeration Documentation

InputUse

enum class sdfg::passes::LibNodeExpander::InputUse

strong

Enumerator
Skip
Scalar
IndirectRead
IndirectWrite
IndirectReadWrite

Constructor & Destructor Documentation

~LibNodeExpander()

virtual sdfg::passes::LibNodeExpander::~LibNodeExpander ( )

virtualdefault

Member Function Documentation

for_lib_node()

virtual const LibNodeExpander * sdfg::passes::LibNodeExpander::for_lib_node ( const data_flow::LibraryNode & node ) const

inlinevirtual

Prefiltering of libnodes.

Allows that an expander sub-divides based on types, codes or impl-types or sth. more elaborate, because we get ready to transform

Parameters

node

Returns

nullptr if not applicable

Reimplemented in sdfg::gpu::GpuMmaExpander, sdfg::passes::CodeLibNodeExpander< T >, sdfg::passes::CodeLibNodeExpander< math::tensor::ConvNode >, sdfg::passes::CodeLibNodeExpander< math::tensor::MatMulNode >, sdfg::passes::CodeLibNodeExpander< MatMulNode >, sdfg::passes::CodeLibNodeExpander< OzakiMatMulNode >, sdfg::passes::SwitchExpander, sdfg::passes::TypedLibNodeExpander< T >, sdfg::passes::TypedLibNodeExpander< math::MathNode >, sdfg::passes::TypedLibNodeExpander< math::tensor::ConvNode >, sdfg::passes::TypedLibNodeExpander< math::tensor::MatMulNode >, sdfg::passes::TypedLibNodeExpander< MatMulNode >, and sdfg::passes::TypedLibNodeExpander< OzakiMatMulNode >.

handle_expand()

virtual ExpandOutcome sdfg::passes::LibNodeExpander::handle_expand ( ExpandContext & context, structured_control_flow::Block & block, data_flow::LibraryNode & node ) const

pure virtual

Implemented in sdfg::passes::SwitchExpander, and sdfg::passes::TypedLibNodeExpander< T >.

matched_code()

virtual const data_flow::LibraryNodeCode * sdfg::passes::LibNodeExpander::matched_code ( ) const

inlinevirtual

If this expander dispatches on a single, fixed library node code, expose it here.

This lets aggregating expanders (e.g. SwitchExpander) build an efficient code -> expander map and jump straight to the right expander, instead of testing every expander linearly via for_lib_node.

Returns

the matched code, or nullptr if this expander does not dispatch purely by code.

Reimplemented in sdfg::passes::CodeLibNodeExpander< T >, sdfg::passes::CodeLibNodeExpander< math::tensor::ConvNode >, sdfg::passes::CodeLibNodeExpander< math::tensor::MatMulNode >, sdfg::passes::CodeLibNodeExpander< MatMulNode >, and sdfg::passes::CodeLibNodeExpander< OzakiMatMulNode >.

---

The documentation for this class was generated from the following file:

• sdfg/include/sdfg/passes/expansion/lib_node_expander.h