Thread-safe implementation of fftw3. More...

#include <fftimplementation.h>

Inheritance diagram for FFT_Implementation:

Collaboration diagram for FFT_Implementation:

Public Member Functions
	FFT_Implementation ()
	Constructor, clears member variables and checks type consistency. More...

	~FFT_Implementation ()
	Destructor, deletes plans and local vector copies if existing. More...

void	calculateFFT (const FFTRealVector &in, FFTComplexVector &out)
	Foward FFT, mapping a real vector with size N to a complex one with size N/2+1. More...

void	calculateFFT (const FFTComplexVector &in, FFTRealVector &out)
	Backward FFT, mapping a complex vector with size M to a complex one with size 2M-1. More...

void	optimize (FFTRealVector &in)
	Optimize plan for the forward transfomation. More...

void	optimize (FFTComplexVector &in)
	Optimize plan for the backward transformation. More...

Public Member Functions inherited from FFTAdapter
	FFTAdapter ()

virtual	~FFTAdapter ()

Private Member Functions
void	updatePlan (const FFTRealVector &in, unsigned flags)
	Construct a new plan for the FFT transformation if necessary. More...

void	updatePlan (const FFTComplexVector &in, unsigned flags)
	Construct a new plan for the FFT transformation if necessary. More...

Private Attributes
double *	mRvec1
	Local copy of incoming real data. More...

double *	mRvec2
	Local copy of outgoing real data. More...

fftw_complex *	mCvec1
	Local copy of incoming complex data. More...

fftw_complex *	mCvec2
	Local copy of outgoing complex data. More...

size_t	mNRC
	Size of the FFT real -> complex. More...

size_t	mNCR
	Size of the FFT complex -> real. More...

fftw_plan	mPlanRC
	Plan for FFT real -> complex. More...

fftw_plan	mPlanCR
	Plan for FFT complex -> real. More...

Static Private Attributes
static std::mutex	mPlanMutex
	Static mutex protecting planmaking. More...

Detailed Description

Thread-safe implementation of fftw3.

USAGE: FFT's can be carried out simply by calling the function calculateFFT with the corresponding arguments. If a given instance has to carry out a larger number of FFTs using the same vector sizes it is meaningful to optimize the calculation by initially calling the function optimize with the input vector as the argument.

DETAILS: A Fourier transformation using fftw3 requires two steps. First a so-called 'plan' has to be computed which optimizes the FFT algorithm. Then in a second step the actual Fourier transform is carried out. A plan may be kept for several calculations as long as the location of the vectors and their sizes are not changed. By keeping a plan the computation time is drastically reduced.

As of now, the second step (the FFT) is thread-safe while the first one is not. This means that only one thread is allowed to create a plan at a given time. To this end this implementation class protects all accesses to planmaking by a static mutex. In addition, it keeps local input and output vectors with a constant location.

The class provides two types of FFTs, namely, real to complex and comples to real. The second one is the inverse of the first one.

Since memory allocation should be carried out with the inbuilt allocation function of FFTW3, the implementation copies the vectors into local member vectors by memcpy.

Definition at line 66 of file fftimplementation.h.

Constructor & Destructor Documentation

FFT_Implementation::FFT_Implementation ( )

Constructor, clears member variables and checks type consistency.

Definition at line 48 of file fftimplementation.cpp.

FFT_Implementation::~FFT_Implementation ( )

Destructor, deletes plans and local vector copies if existing.

Definition at line 74 of file fftimplementation.cpp.

Member Function Documentation

void FFT_Implementation::calculateFFT	(	const FFTRealVector &	in,
		FFTComplexVector &	out
	)

virtual

Foward FFT, mapping a real vector with size N to a complex one with size N/2+1.

Parameters

in	: vector of real numbers of size N
out	: vector of complex numbers, will be resized to N/2+1

Implements FFTAdapter.

Definition at line 230 of file fftimplementation.cpp.

Here is the call graph for this function:

void FFT_Implementation::calculateFFT	(	const FFTComplexVector &	in,
		FFTRealVector &	out
	)

virtual

Backward FFT, mapping a complex vector with size M to a complex one with size 2M-1.

Parameters

in	: vector of complex numbers of size M
out	: vector of real numbers, will be resized to size 2*M-1

Implements FFTAdapter.

Definition at line 257 of file fftimplementation.cpp.

Here is the call graph for this function:

void FFT_Implementation::optimize ( FFTRealVector & in )

virtual

Optimize plan for the forward transfomation.

This function should be called at the beginning if the subsequent code performs many FFTs on vectors with varying content, but with the same location and the same size. Note that this function may be time-consuming, but it accelerates subsequent computations significantly.

Parameters

in	: vector of real numbers to be transformed

Implements FFTAdapter.

Definition at line 105 of file fftimplementation.cpp.

Here is the call graph for this function:

void FFT_Implementation::optimize ( FFTComplexVector & in )

virtual

Optimize plan for the backward transformation.

This function should be called at the beginning if the subsequent code performs many FFTs on vectors with varying content, but with the same location and the same size. Note that this function may be time-consuming, but it accelerates subsequent computations significantly.

Parameters

in	: vector of complex numbers to be transformed

Implements FFTAdapter.

Definition at line 126 of file fftimplementation.cpp.

Here is the call graph for this function:

void FFT_Implementation::updatePlan	(	const FFTRealVector &	in,
		unsigned	flags
	)

private

Construct a new plan for the FFT transformation if necessary.

This function checks whether the plan and the local vectors still exist and whether they have the correct size. If so, the function does nothing which means that the existing plan will be reused. Ohterwise, it deletes the plan and the vectors and constructs them again with the given size.

Parameters

in	: vector of real numbers of size N
flags	: fftw3 internal flags controlling planmaking

Definition at line 149 of file fftimplementation.cpp.

void FFT_Implementation::updatePlan	(	const FFTComplexVector &	in,
		unsigned	flags
	)

private

Construct a new plan for the FFT transformation if necessary.

This function checks whether the plan and the local vectors still exist and whether they have the correct size. If so, the function does nothing which means that the existing plan will be reused. Ohterwise, it deletes the plan and the vectors and constructs them again with the given size.

Parameters

in	: vector of complex numbers
flags	: fftw3 internal flags controlling planmaking

Definition at line 192 of file fftimplementation.cpp.

Member Data Documentation

fftw_complex* FFT_Implementation::mCvec1

private

Local copy of incoming complex data.

Definition at line 88 of file fftimplementation.h.

fftw_complex* FFT_Implementation::mCvec2

private

Local copy of outgoing complex data.

Definition at line 89 of file fftimplementation.h.

size_t FFT_Implementation::mNCR

private

Size of the FFT complex -> real.

Definition at line 91 of file fftimplementation.h.

size_t FFT_Implementation::mNRC

private

Size of the FFT real -> complex.

Definition at line 90 of file fftimplementation.h.

fftw_plan FFT_Implementation::mPlanCR

private

Plan for FFT complex -> real.

Definition at line 94 of file fftimplementation.h.

std::mutex FFT_Implementation::mPlanMutex

staticprivate

Static mutex protecting planmaking.

Definition at line 95 of file fftimplementation.h.

fftw_plan FFT_Implementation::mPlanRC

private

Plan for FFT real -> complex.

Definition at line 93 of file fftimplementation.h.

double* FFT_Implementation::mRvec1

private

Local copy of incoming real data.

Definition at line 86 of file fftimplementation.h.

double* FFT_Implementation::mRvec2

private

Local copy of outgoing real data.

Definition at line 87 of file fftimplementation.h.

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

/home/hinrichsen/Dokumente/Aktuell/EPT/Entropy-Piano-Tuner/core/math/fftimplementation.h
/home/hinrichsen/Dokumente/Aktuell/EPT/Entropy-Piano-Tuner/core/math/fftimplementation.cpp

Public Member Functions

Private Member Functions

Private Attributes

Static Private Attributes

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation