c++ - 在 SWIFT 中使用 C++ FFT 代码

Question

我有 C++ FFT 代码(粘贴在下面)。

在 C++ 中，当我在 main() 中输入时，RL_Input = {1,2,-3,4}, IM_Input = {-4,3,2,1}，我得到的答案是 RL_Output = {4, 6, -8, 2}, IM_Output = {2, -4, -6, -8}。

我想从 SWIFT 调用此 C++ 代码。所以，在 SWIFT 中，我想做如下事情:

   let (RL_Output, IM_Output) = Some_Swift_Function([1,2,-3,4], [-4,3,2,1]) // INPUT RL & IM
   print(RL_Output)
   print(IM_Output)

   // RL_Output = [4, 6, -8, 2]  //Answer (REAL)
   // IM_Output = [2, -4, -6, -8] //Answer (IMAG)

如何使用我拥有的 C++ 代码(如下所示)执行上述操作？

    //FftRealPairTest.cpp
    #include <algorithm>
    #include <cmath>
    #include <cstdlib>
    #include <iomanip>
    #include <iostream>
    #include <random>
    #include <vector>
    #include "FftRealPair.hpp"

    using std::cout;
    using std::endl;
    using std::vector;

    int main() {
        vector<double> inputreal({1,2,-3,4});

        vector<double> inputimag({-4,3,2,1});

        vector<double> actualoutreal(inputreal);

        vector<double> actualoutimag(inputimag);

        Fft::transform(actualoutreal, actualoutimag);

        std::cout << "REAL:" << std::endl;
        for (int i = 0; i < inputimag.size(); ++i)
        {
            std::cout << actualoutreal[i] << std::endl;
        }


        std::cout << "IMAG" << std::endl;
        for (int i = 0; i < inputimag.size(); ++i)
        {
            std::cout << actualoutimag[i] << std::endl;
        }
        
    }


    /////////////////////////////////////////////////

    //FftRealPair.cpp
    /*
     * Free FFT and convolution (C++)
     *
     * Copyright (c) 2017 Project Nayuki. (MIT License)
     * https://www.nayuki.io/page/free-small-fft-in-multiple-languages
     *
     * Permission is hereby granted, free of charge, to any person obtaining a copy of
     * this software and associated documentation files (the "Software"), to deal in
     * the Software without restriction, including without limitation the rights to
     * use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
     * the Software, and to permit persons to whom the Software is furnished to do so,
     * subject to the following conditions:
     * - The above copyright notice and this permission notice shall be included in
     *   all copies or substantial portions of the Software.
     * - The Software is provided "as is", without warranty of any kind, express or
     *   implied, including but not limited to the warranties of merchantability,
     *   fitness for a particular purpose and noninfringement. In no event shall the
     *   authors or copyright holders be liable for any claim, damages or other
     *   liability, whether in an action of contract, tort or otherwise, arising from,
     *   out of or in connection with the Software or the use or other dealings in the
     *   Software.
     */

    #include <algorithm>
    #include <cmath>
    #include <cstddef>
    #include <cstdint>
    #include "FftRealPair.hpp"

    using std::size_t;
    using std::vector;


    // Private function prototypes
    static size_t reverseBits(size_t x, int n);


    void Fft::transform(vector<double> &real, vector<double> &imag) {
        size_t n = real.size();
        if (n != imag.size())
            throw "Mismatched lengths";
        if (n == 0)
            return;
        else if ((n & (n - 1)) == 0)  // Is power of 2
            transformRadix2(real, imag);
        else  // More complicated algorithm for arbitrary sizes
            transformBluestein(real, imag);
    }


    void Fft::inverseTransform(vector<double> &real, vector<double> &imag) {
        transform(imag, real);
    }


    void Fft::transformRadix2(vector<double> &real, vector<double> &imag) {
        // Length variables
        size_t n = real.size();
        if (n != imag.size())
            throw "Mismatched lengths";
        int levels = 0;  // Compute levels = floor(log2(n))
        for (size_t temp = n; temp > 1U; temp >>= 1)
            levels++;
        if (static_cast<size_t>(1U) << levels != n)
            throw "Length is not a power of 2";

        // Trignometric tables
        vector<double> cosTable(n / 2);
        vector<double> sinTable(n / 2);
        for (size_t i = 0; i < n / 2; i++) {
            cosTable[i] = std::cos(2 * M_PI * i / n);
            sinTable[i] = std::sin(2 * M_PI * i / n);
        }

        // Bit-reversed addressing permutation
        for (size_t i = 0; i < n; i++) {
            size_t j = reverseBits(i, levels);
            if (j > i) {
                std::swap(real[i], real[j]);
                std::swap(imag[i], imag[j]);
            }
        }

        // Cooley-Tukey decimation-in-time radix-2 FFT
        for (size_t size = 2; size <= n; size *= 2) {
            size_t halfsize = size / 2;
            size_t tablestep = n / size;
            for (size_t i = 0; i < n; i += size) {
                for (size_t j = i, k = 0; j < i + halfsize; j++, k += tablestep) {
                    size_t l = j + halfsize;
                    double tpre =  real[l] * cosTable[k] + imag[l] * sinTable[k];
                    double tpim = -real[l] * sinTable[k] + imag[l] * cosTable[k];
                    real[l] = real[j] - tpre;
                    imag[l] = imag[j] - tpim;
                    real[j] += tpre;
                    imag[j] += tpim;
                }
            }
            if (size == n)  // Prevent overflow in 'size *= 2'
                break;
        }
    }


    void Fft::transformBluestein(vector<double> &real, vector<double> &imag) {
        // Find a power-of-2 convolution length m such that m >= n * 2 + 1
        size_t n = real.size();
        if (n != imag.size())
            throw "Mismatched lengths";
        size_t m = 1;
        while (m / 2 <= n) {
            if (m > SIZE_MAX / 2)
                throw "Vector too large";
            m *= 2;
        }

        // Trignometric tables
        vector<double> cosTable(n), sinTable(n);
        for (size_t i = 0; i < n; i++) {
            unsigned long long temp = static_cast<unsigned long long>(i) * i;
            temp %= static_cast<unsigned long long>(n) * 2;
            double angle = M_PI * temp / n;
            // Less accurate alternative if long long is unavailable: double angle = M_PI * i * i / n;
            cosTable[i] = std::cos(angle);
            sinTable[i] = std::sin(angle);
        }

        // Temporary vectors and preprocessing
        vector<double> areal(m), aimag(m);
        for (size_t i = 0; i < n; i++) {
            areal[i] =  real[i] * cosTable[i] + imag[i] * sinTable[i];
            aimag[i] = -real[i] * sinTable[i] + imag[i] * cosTable[i];
        }
        vector<double> breal(m), bimag(m);
        breal[0] = cosTable[0];
        bimag[0] = sinTable[0];
        for (size_t i = 1; i < n; i++) {
            breal[i] = breal[m - i] = cosTable[i];
            bimag[i] = bimag[m - i] = sinTable[i];
        }

        // Convolution
        vector<double> creal(m), cimag(m);
        convolve(areal, aimag, breal, bimag, creal, cimag);

        // Postprocessing
        for (size_t i = 0; i < n; i++) {
            real[i] =  creal[i] * cosTable[i] + cimag[i] * sinTable[i];
            imag[i] = -creal[i] * sinTable[i] + cimag[i] * cosTable[i];
        }
    }


    void Fft::convolve(const vector<double> &x, const vector<double> &y, vector<double> &out) {
        size_t n = x.size();
        if (n != y.size() || n != out.size())
            throw "Mismatched lengths";
        vector<double> outimag(n);
        convolve(x, vector<double>(n), y, vector<double>(n), out, outimag);
    }


    void Fft::convolve(
                       const vector<double> &xreal, const vector<double> &ximag,
                       const vector<double> &yreal, const vector<double> &yimag,
                       vector<double> &outreal, vector<double> &outimag) {

        size_t n = xreal.size();
        if (n != ximag.size() || n != yreal.size() || n != yimag.size()
            || n != outreal.size() || n != outimag.size())
            throw "Mismatched lengths";

        vector<double> xr(xreal);
        vector<double> xi(ximag);
        vector<double> yr(yreal);
        vector<double> yi(yimag);
        transform(xr, xi);
        transform(yr, yi);
        
        for (size_t i = 0; i < n; i++) {
            double temp = xr[i] * yr[i] - xi[i] * yi[i];
            xi[i] = xi[i] * yr[i] + xr[i] * yi[i];
            xr[i] = temp;
        }
        inverseTransform(xr, xi);
        
        for (size_t i = 0; i < n; i++) {  // Scaling (because this FFT implementation omits it)
            outreal[i] = xr[i] / n;
            outimag[i] = xi[i] / n;
        }
    }


    static size_t reverseBits(size_t x, int n) {
        size_t result = 0;
        for (int i = 0; i < n; i++, x >>= 1)
            result = (result << 1) | (x & 1U);
        return result;
    }



    ////////////////////////////////////////////////////


    //FftRealPair.hpp

    /*
     * Free FFT and convolution (C++)
     *
     * Copyright (c) 2017 Project Nayuki. (MIT License)
     * https://www.nayuki.io/page/free-small-fft-in-multiple-languages
     *
     * Permission is hereby granted, free of charge, to any person obtaining a copy of
     * this software and associated documentation files (the "Software"), to deal in
     * the Software without restriction, including without limitation the rights to
     * use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
     * the Software, and to permit persons to whom the Software is furnished to do so,
     * subject to the following conditions:
     * - The above copyright notice and this permission notice shall be included in
     *   all copies or substantial portions of the Software.
     * - The Software is provided "as is", without warranty of any kind, express or
     *   implied, including but not limited to the warranties of merchantability,
     *   fitness for a particular purpose and noninfringement. In no event shall the
     *   authors or copyright holders be liable for any claim, damages or other
     *   liability, whether in an action of contract, tort or otherwise, arising from,
     *   out of or in connection with the Software or the use or other dealings in the
     *   Software.
     */

    #pragma once

    #include <vector>


    namespace Fft {

        /*
         * Computes the discrete Fourier transform (DFT) of the given complex vector, storing the result back into the vector.
         * The vector can have any length. This is a wrapper function.
         */
        void transform(std::vector<double> &real, std::vector<double> &imag);


        /*
         * Computes the inverse discrete Fourier transform (IDFT) of the given complex vector, storing the result back into the vector.
         * The vector can have any length. This is a wrapper function. This transform does not perform scaling, so the inverse is not a true inverse.
         */
        void inverseTransform(std::vector<double> &real, std::vector<double> &imag);





        /*
         * Computes the discrete Fourier transform (DFT) of the given complex vector, storing the result back into the vector.
         * The vector's length must be a power of 2. Uses the Cooley-Tukey decimation-in-time radix-2 algorithm.
         */
        void transformRadix2(std::vector<double> &real, std::vector<double> &imag);


        /*
         * Computes the discrete Fourier transform (DFT) of the given complex vector, storing the result back into the vector.
         * The vector can have any length. This requires the convolution function, which in turn requires the radix-2 FFT function.
         * Uses Bluestein's chirp z-transform algorithm.
         */
        void transformBluestein(std::vector<double> &real, std::vector<double> &imag);


        /*
         * Computes the circular convolution of the given real vectors. Each vector's length must be the same.
         */
        void convolve(const std::vector<double> &x, const std::vector<double> &y, std::vector<double> &out);


        /*
         * Computes the circular convolution of the given complex vectors. Each vector's length must be the same.
         */
        void convolve(
                      const std::vector<double> &xreal, const std::vector<double> &ximag,
                      const std::vector<double> &yreal, const std::vector<double> &yimag,
                      std::vector<double> &outreal, std::vector<double> &outimag);
        
    }

Answer 1

如果您想从 Swift 调用该 C++ 代码，则需要通过 Objective-C++ 进行桥接。在 SO 上进行简单搜索将显示有关如何执行此操作的大量帖子。

在这种情况下，我们希望尽量减少数据复制，因为我们将 C++/Objective-C++/Swift 粘合在一起，以减少对性能的负面影响。 Swift 中 Double 的 Array 将其数据存储在连续存储中，因为 Double 不是一个类。 Array 的withUnsafeMutableBufferPointer 方法似乎是一个很有前途的解决方案。不过，我会小心，并首先在一个简单的测试程序上测试这种方法。如果时间允许，我会在接下来的几天里想出一些办法。

请参阅位于 https://developer.apple.com/documentation/swift/array 的 Array 文档.另一个非常有用的资源是 https://developer.apple.com/library/content/documentation/Swift/Conceptual/BuildingCocoaApps/index.html#//apple_ref/doc/uid/TP40014216-CH2-ID0 ，您可能已经在搜索此主题时看到了它。

一般要注意的一件事是，如果调整 vector 的大小，C++ 中的 vector 存储可以在内存中重新定位，例如，在这种情况下，我们将不得不制作额外的数据拷贝从 Swift 桥接到 C++ 时。然而，这段 C++ 代码似乎没有做任何会移动底层存储的事情。

2017 年 10 月 25 日更新:使用 withUnsafeMutableBufferPointer 将需要在内存中复制数组，因为创建一个直接就地使用给定的 vector 是有问题的缓冲。参见 How to cheaply assign C-style array to std::vector? .

但是，由于有一个 C 版本的库可用，这就变得小菜一碟了:

• 将 fft.h 和 fft.c 添加到您的 Xcode 项目。
• 在桥接 header 中导入 fft.h。

然后你可以像这样在 Swift 中使用 C 代码:

var dReal : [Double] = [1,2,-3,4]
var dImg : [Double] = [-4,3,2,1]

dReal.withUnsafeMutableBufferPointer { (real : inout UnsafeMutableBufferPointer<Double> ) in
    dImg.withUnsafeMutableBufferPointer { (img : inout UnsafeMutableBufferPointer<Double>) in
        if (real.count == img.count) {
            Fft_transform(real.baseAddress, img.baseAddress, real.count)
        }
    }
}

当然，这只是一个简单的示例。你可以让它更漂亮，添加错误处理等。