最近用到FFT了。cuda里有GPUFFT的函数，不过是C的。最近找了个封装，还挺好用。具体效率没测过，主要是内存和显存之间来回传递比较费时间吧。代码如下：

CudafyModule km = CudafyTranslator.Cudafy();             GPGPU gpu = CudafyHost.GetDevice(CudafyModes.Target);            gpu.LoadModule(km);             ComplexD[] cm = new ComplexD[4];            cm[0] = new ComplexD(1.5, 1);            cm[1] = new ComplexD(2.5, 2);            cm[2] = new ComplexD(3.5, 3);            cm[3] = new ComplexD(4.5, 4);            ComplexD[] dev_cm = gpu.CopyToDevice(cm);             ComplexD[] ifftData = new ComplexD[4];            ComplexD[] dev_ifftData = gpu.CopyToDevice(ifftData);             Cudafy.Maths.FFT.GPGPUFFT gpuFFT = Cudafy.Maths.FFT.GPGPUFFT.Create(gpu);            Cudafy.Maths.FFT.FFTPlan1D fft_1d = gpuFFT.Plan1D(Cudafy.Maths.FFT.eFFTType.Complex2Complex, Cudafy.Maths.FFT.eDataType.Double, 4, 1);             fft_1d.Execute
    
     (dev_cm, dev_ifftData, true);            gpu.CopyFromDevice(dev_ifftData, ifftData);             gpu.FreeAll();