Depth-aware nested GPU offload pass. More...

#include <sdfg/passes/offloading/gpu_nested_offload_pass.h>

Inheritance diagram for sdfg::passes::GPUNestedOffloadPass:

Public Member Functions
	GPUNestedOffloadPass (const std::vector< structured_control_flow::StructuredLoop * > &loops, GPUTarget target)
	~GPUNestedOffloadPass () override=default
bool	run_pass (builder::StructuredSDFGBuilder &builder, analysis::AnalysisManager &analysis_manager) override
std::string	name () override
Public Member Functions inherited from sdfg::passes::Pass
	Pass ()=default
	Pass (const Options &options)
virtual	~Pass ()=default
virtual std::vector< OptionSpec >	options ()
bool	run (builder::SDFGBuilder &builder, bool create_report=false)
bool	run (builder::StructuredSDFGBuilder &builder, analysis::AnalysisManager &analysis_manager, bool create_report=false)
virtual bool	run_pass (builder::SDFGBuilder &builder)
virtual void	invalidates (analysis::AnalysisManager &analysis_manager, bool applied)

Additional Inherited Members
Protected Member Functions inherited from sdfg::passes::Pass
template<class T>
T	option (const OptionKey< T > &key, T fallback=T{}) const

Detailed Description

Depth-aware nested GPU offload pass.

For each already grid-offloaded outer loop, inspects the maximum nesting depth of its loop nest and offloads the nested loops via GPUOffloadNestedLoop:

depth 2: X_BLOCK
depth 3: X_BLOCK, then WARP
depth >= 4: Y_GRID, X_BLOCK, Y_BLOCK

Grid/block sizes derive from each loop's iteration count: a known integer count is used directly for grid dimensions and capped at the default for block dimensions; a symbolic count falls back to the default.

Constructor & Destructor Documentation

◆ GPUNestedOffloadPass()

sdfg::passes::GPUNestedOffloadPass::GPUNestedOffloadPass	(	const std::vector< structured_control_flow::StructuredLoop * > &	loops,
		GPUTarget	target )

◆ ~GPUNestedOffloadPass()

sdfg::passes::GPUNestedOffloadPass::~GPUNestedOffloadPass ( )

overridedefault

Member Function Documentation

◆ name()

std::string sdfg::passes::GPUNestedOffloadPass::name ( )

inlineoverridevirtual

Implements sdfg::passes::Pass.

◆ run_pass()

bool sdfg::passes::GPUNestedOffloadPass::run_pass	(	builder::StructuredSDFGBuilder &	builder,
		analysis::AnalysisManager &	analysis_manager )

overridevirtual

Reimplemented from sdfg::passes::Pass.

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

opt/include/sdfg/passes/offloading/gpu_nested_offload_pass.h

Public Member Functions