#include <PerfMonitor.h>

Public Member Functions
	PerfMonitor (const Scop &S, llvm::Module *M)
	Create a new performance monitor.

void	initialize ()
	Initialize the performance monitor.

void	insertRegionStart (llvm::Instruction *InsertBefore)
	Mark the beginning of a timing region.

void	insertRegionEnd (llvm::Instruction *InsertBefore)
	Mark the end of a timing region.

Private Member Functions
llvm::Function *	insertInitFunction (llvm::Function *FinalReporting)

void	addToGlobalConstructors (llvm::Function *Fn)
	Add Function `to` list of global constructors.

void	addGlobalVariables ()
	Add global variables to module.

void	addScopCounter ()
	Add per-scop tracking to module.

llvm::Function *	getRDTSCP ()
	Get a reference to the intrinsic "{ i64, i32 } @llvm.x86.rdtscp()".

llvm::Function *	getAtExit ()
	Get a reference to "int atexit(void (*function)(void))" function.

llvm::Function *	insertFinalReporting ()
	Create function "__polly_perf_final_reporting".

void	AppendScopReporting ()
	Append Scop reporting data to "__polly_perf_final_reporting".

Private Attributes
llvm::Module *	M

PollyIRBuilder	Builder

const Scop &	S

bool	Supported
	Indicates if performance profiling is supported on this architecture.

llvm::Value *	CyclesTotalStartPtr
	The cycle counter at the beginning of the program execution.

llvm::Value *	CyclesInCurrentScopPtr
	The total number of cycles spent in the current scop S.

llvm::Value *	TripCountForCurrentScopPtr
	The total number of times the current scop S is executed.

llvm::Value *	CyclesInScopsPtr
	The total number of cycles spent within scops.

llvm::Value *	CyclesInScopStartPtr
	The value of the cycle counter at the beginning of the last scop.

llvm::Value *	AlreadyInitializedPtr
	A global variable, that keeps track if the performance monitor initialization has already been run.

Detailed Description

Definition at line 16 of file PerfMonitor.h.

Constructor & Destructor Documentation

◆ PerfMonitor()

PerfMonitor::PerfMonitor	(	const Scop &	S,
		llvm::Module *	M
	)

Create a new performance monitor.

Parameters

S	The scop for which to generate fine-grained performance monitoring information.
M	The module for which to generate the performance monitor.

Definition at line 64 of file PerfMonitor.cpp.

References M, and Supported.

Member Function Documentation

◆ addGlobalVariables()

void PerfMonitor::addGlobalVariables ( )

private

Add global variables to module.

Insert a set of global variables that are used to track performance, into the module (or obtain references to them if they already exist).

Definition at line 102 of file PerfMonitor.cpp.

References AlreadyInitializedPtr, Builder, CyclesInScopsPtr, CyclesInScopStartPtr, CyclesTotalStartPtr, M, and TryRegisterGlobal().

Referenced by initialize().

◆ addScopCounter()

void PerfMonitor::addScopCounter ( )

private

Add per-scop tracking to module.

Insert the global variable which is used to track the number of cycles this scop runs.

Definition at line 93 of file PerfMonitor.cpp.

References Builder, CyclesInCurrentScopPtr, GetScopUniqueVarname(), M, TripCountForCurrentScopPtr, and TryRegisterGlobal().

Referenced by initialize().

◆ addToGlobalConstructors()

void PerfMonitor::addToGlobalConstructors ( llvm::Function * Fn )

private

Add Function to list of global constructors.

If no global constructors are available in this current module, insert a new list of global constructors containing Fn as only global constructor. Otherwise, append Fn to the list of global constructors.

All functions listed as global constructors are executed before the main() function is called.

Parameters

Fn	Function to add to global constructors

Definition at line 35 of file PerfMonitor.cpp.

References polly::Array, Builder, M, polly::Value, and X().

Referenced by initialize().

◆ AppendScopReporting()

void PerfMonitor::AppendScopReporting ( )

private

Append Scop reporting data to "__polly_perf_final_reporting".

This function appends the current scop (S)'s information to the final printing function.

Definition at line 167 of file PerfMonitor.cpp.

References assert, Builder, polly::RuntimeDebugBuilder::createCPUPrinter(), CyclesInCurrentScopPtr, FinalStartBB, ReturnFromFinal, Supported, TripCountForCurrentScopPtr, and polly::Value.

Referenced by initialize().

◆ getAtExit()

Function * PerfMonitor::getAtExit ( )

private

Get a reference to "int atexit(void (*function)(void))" function.

This function allows to register function pointers that must be executed when the program is terminated.

Returns: A reference to @atexit().

Definition at line 21 of file PerfMonitor.cpp.

References Builder, Function, and M.

Referenced by insertInitFunction().

◆ getRDTSCP()

Function * PerfMonitor::getRDTSCP ( )

private

Get a reference to the intrinsic "{ i64, i32 } @llvm.x86.rdtscp()".

The rdtscp function returns the current value of the processor's time-stamp counter as well as the current CPU identifier. On modern x86 systems, the returned value is independent of the dynamic clock frequency and consistent across multiple cores. It can consequently be used to get accurate and low-overhead timing information. Even though the counter is wrapping, it can be reliably used even for measuring longer time intervals, as on a 1 GHz processor the counter only wraps every 545 years.

The RDTSCP instruction is "pseudo" serializing:

"“The RDTSCP instruction waits until all previous instructions have been executed before reading the counter. However, subsequent instructions may begin execution before the read operation is performed.”

To ensure that no later instructions are scheduled before the RDTSCP instruction it is often recommended to schedule a cpuid call after the RDTSCP instruction. We do not do this yet, trading some imprecision in our timing for a reduced overhead in our timing.

Returns: A reference to the declaration of @llvm.x86.rdtscp.

Definition at line 60 of file PerfMonitor.cpp.

References Function, and M.

Referenced by insertFinalReporting(), insertInitFunction(), insertRegionEnd(), and insertRegionStart().

◆ initialize()

void PerfMonitor::initialize ( )

Initialize the performance monitor.

Ensure that all global variables, functions, and callbacks needed to manage the performance monitor are initialized and registered.

Definition at line 199 of file PerfMonitor.cpp.

References addGlobalVariables(), addScopCounter(), addToGlobalConstructors(), AppendScopReporting(), FinalReporting, Function, insertFinalReporting(), and insertInitFunction().

Referenced by generateCode().

◆ insertFinalReporting()

Function * PerfMonitor::insertFinalReporting ( )

private

Create function "__polly_perf_final_reporting".

This function finalizes the performance measurements and prints the results to stdout. It is expected to be registered with 'atexit()'.

Definition at line 122 of file PerfMonitor.cpp.

References Builder, polly::RuntimeDebugBuilder::createCPUPrinter(), CyclesInScopsPtr, CyclesTotalStartPtr, FinalReportingFunctionName, FinalStartBB, Function, getRDTSCP(), M, ReturnFromFinal, Supported, and polly::Value.

Referenced by initialize().

◆ insertInitFunction()

Function * PerfMonitor::insertInitFunction ( llvm::Function * FinalReporting )

private

Definition at line 215 of file PerfMonitor.cpp.

References AlreadyInitializedPtr, Builder, CyclesTotalStartPtr, FinalReporting, Function, getAtExit(), getRDTSCP(), InitFunctionName, M, Supported, and polly::Value.

Referenced by initialize().

◆ insertRegionEnd()

void PerfMonitor::insertRegionEnd ( llvm::Instruction * InsertBefore )

Mark the end of a timing region.

Parameters

InsertBefore The instruction before which the timing region starts.

Definition at line 276 of file PerfMonitor.cpp.

References Builder, CyclesInCurrentScopPtr, CyclesInScopsPtr, CyclesInScopStartPtr, Function, getRDTSCP(), Supported, TripCountForCurrentScopPtr, and polly::Value.

Referenced by generateCode().

◆ insertRegionStart()

void PerfMonitor::insertRegionStart ( llvm::Instruction * InsertBefore )

Mark the beginning of a timing region.

Parameters

InsertBefore The instruction before which the timing region starts.

Definition at line 265 of file PerfMonitor.cpp.

References Builder, CyclesInScopStartPtr, Function, getRDTSCP(), Supported, and polly::Value.

Referenced by generateCode().

Member Data Documentation

◆ AlreadyInitializedPtr

llvm::Value* polly::PerfMonitor::AlreadyInitializedPtr

private

A global variable, that keeps track if the performance monitor initialization has already been run.

Definition at line 68 of file PerfMonitor.h.

Referenced by addGlobalVariables(), and insertInitFunction().

◆ Builder

PollyIRBuilder polly::PerfMonitor::Builder

private

Definition at line 43 of file PerfMonitor.h.

Referenced by addGlobalVariables(), addScopCounter(), addToGlobalConstructors(), AppendScopReporting(), getAtExit(), insertFinalReporting(), insertInitFunction(), insertRegionEnd(), and insertRegionStart().

◆ CyclesInCurrentScopPtr

llvm::Value* polly::PerfMonitor::CyclesInCurrentScopPtr

private

The total number of cycles spent in the current scop S.

Definition at line 55 of file PerfMonitor.h.

Referenced by addScopCounter(), AppendScopReporting(), and insertRegionEnd().

◆ CyclesInScopsPtr

llvm::Value* polly::PerfMonitor::CyclesInScopsPtr

private

The total number of cycles spent within scops.

Definition at line 61 of file PerfMonitor.h.

Referenced by addGlobalVariables(), insertFinalReporting(), and insertRegionEnd().

◆ CyclesInScopStartPtr

llvm::Value* polly::PerfMonitor::CyclesInScopStartPtr

private

The value of the cycle counter at the beginning of the last scop.

Definition at line 64 of file PerfMonitor.h.

Referenced by addGlobalVariables(), insertRegionEnd(), and insertRegionStart().

◆ CyclesTotalStartPtr

llvm::Value* polly::PerfMonitor::CyclesTotalStartPtr

private

The cycle counter at the beginning of the program execution.

Definition at line 52 of file PerfMonitor.h.

Referenced by addGlobalVariables(), insertFinalReporting(), and insertInitFunction().

◆ M

llvm::Module* polly::PerfMonitor::M

private

Definition at line 42 of file PerfMonitor.h.

Referenced by addGlobalVariables(), addScopCounter(), addToGlobalConstructors(), getAtExit(), getRDTSCP(), insertFinalReporting(), insertInitFunction(), and PerfMonitor().

◆ S

const Scop& polly::PerfMonitor::S

private

Definition at line 46 of file PerfMonitor.h.

◆ Supported

bool polly::PerfMonitor::Supported

private

Indicates if performance profiling is supported on this architecture.

Definition at line 49 of file PerfMonitor.h.

Referenced by AppendScopReporting(), insertFinalReporting(), insertInitFunction(), insertRegionEnd(), insertRegionStart(), and PerfMonitor().

◆ TripCountForCurrentScopPtr

llvm::Value* polly::PerfMonitor::TripCountForCurrentScopPtr

private

The total number of times the current scop S is executed.

Definition at line 58 of file PerfMonitor.h.

Referenced by addScopCounter(), AppendScopReporting(), and insertRegionEnd().

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

Public Member Functions

Private Member Functions

Private Attributes

Detailed Description

Constructor & Destructor Documentation

◆ PerfMonitor()

Member Function Documentation

◆ addGlobalVariables()

◆ addScopCounter()

◆ addToGlobalConstructors()

◆ AppendScopReporting()

◆ getAtExit()

◆ getRDTSCP()

◆ initialize()

◆ insertFinalReporting()

◆ insertInitFunction()

◆ insertRegionEnd()

◆ insertRegionStart()

Member Data Documentation

◆ AlreadyInitializedPtr

◆ Builder

◆ CyclesInCurrentScopPtr

◆ CyclesInScopsPtr

◆ CyclesInScopStartPtr

◆ CyclesTotalStartPtr

◆ M

◆ S

◆ Supported

◆ TripCountForCurrentScopPtr