[eigen] Problem with fft and custom allocator

[Matthias Pospiech <matthias.pospiech@xxxxxx>]

Hello,
I am using the FFT of eigen, however with a custom allocator.
The custom one is based on the code from Nicolai M. Josuttis
http://www.josuttis.com/libbook/memory/myalloc.hpp.html
with fftw_malloc and fftw_free instead of new and delete.
The class is called fftalloc.

This simple code however fails:

	std::vector<complex<double>,fftalloc<complex<double> > > data(N);
	std::vector<complex<double>,fftalloc<complex<double> > > dataFourier(N);

	FFT<double> fft;
	fft.fwd(dataFourier, data);

It gives me template errors (mingw 4.92 32 bit), because no match was found
basically because 

cannot convert 'dataFourier' (type 
'std::vector<std::complex<double>, fftwcpp::fftalloc<std::complex<double> > >') 
to type 
'std::vector<std::complex<double>, std::allocator<std::complex<double> > >&'


The template code provided by EIGEN::FFT is 
    template <typename _Input>
    inline
    void fwd( std::vector<Complex> & dst, const std::vector<_Input
> & src)
    {
        if ( NumTraits<_Input>::IsComplex == 0 && HasFlag(HalfSpectrum) )
            dst.resize( (src.size()>>1)+1); // half the bins + Nyquist bin
        else
            dst.resize(src.size());
        fwd(&dst[0],&src[0],src.size());
    }

A solutiuon witha copy is not allowed, because of speed and alignment restriction of FFTW.

How would a template look like that allows the use of this custom allocator
together with a vector and a real or complex input?

full error message

.\..\fftwcpp\src\main.cpp:37:34: error: no matching function for call to 'FFT<double>::fwd(std::vector<std::complex<double>, fftwcpp::fftalloc<std::complex<double> > >&, std::vector<std::complex<double>, fftwcpp::fftalloc<std::complex<double> > >&)'

fft.fwd(dataFourier, data);

^

..\..\fftwcpp\src\main.cpp:37:34: note: candidates are:

In file included from ..\..\fftwcpp\src\main.cpp:3:0:

..\..\fftwcpp\src\fftwcpp.h:152:10: note: void FFT<T_Scalar, T_Impl>::fwd(FFT<T_Scalar, T_Impl>::Complex*, const Scalar*, FFT<T_Scalar, T_Impl>::Index) [with T_Scalar = double; T_Impl = default_fft_impl<double>; FFT<T_Scalar, T_Impl>::Complex = std::complex<double>; FFT<T_Scalar, T_Impl>::Scalar = double; FFT<T_Scalar, T_Impl>::Index = int]

void fwd( Complex * dst, const Scalar * src, Index nfft)

^

..\..\fftwcpp\src\fftwcpp.h:152:10: note: candidate expects 3 arguments, 2 provided

..\..\fftwcpp\src\fftwcpp.h:161:10: note: void FFT<T_Scalar, T_Impl>::fwd(FFT<T_Scalar, T_Impl>::Complex*, const Complex*, FFT<T_Scalar, T_Impl>::Index) [with T_Scalar = double; T_Impl = default_fft_impl<double>; FFT<T_Scalar, T_Impl>::Complex = std::complex<double>; FFT<T_Scalar, T_Impl>::Index = int]

void fwd( Complex * dst, const Complex * src, Index nfft)

^

..\..\fftwcpp\src\fftwcpp.h:161:10: note: candidate expects 3 arguments, 2 provided

..\..\fftwcpp\src\fftwcpp.h:177:10: note: template<class _Input> void FFT<T_Scalar, T_Impl>::fwd(std::vector<typename T_Impl::Complex>&, const std::vector<_Input>&) [with _Input = _Input; T_Scalar = double; T_Impl = default_fft_impl<double>]

void fwd( std::vector<Complex> & dst, const std::vector<_Input> & src)

^

..\..\fftwcpp\src\fftwcpp.h:177:10: note: template argument deduction/substitution failed:

..\..\fftwcpp\src\main.cpp:37:34: note: cannot convert 'dataFourier' (type 'std::vector<std::complex<double>, fftwcpp::fftalloc<std::complex<double> > >') to type 'std::vector<std::complex<double>, std::allocator<std::complex<double> > >&'

fft.fwd(dataFourier, data);

BTW: I plan to upload the code to github once it compiles and is in principle useable.

Matthias