回声消除(AEC)模块的编译与测试

2019-09-13 12:01:36 广州普斯格智能科技有限公司 66

整体过程同webRTC降噪(NS)、自动增益控制(AGC)模块的编译与测试一样,只是在aec_rdft.c文件中可能会报错,局部变量ip跟全局变量冲突的问题,可以将局部变量重新命名一下,就可以通过编译了。

修改后的代码如下


#include "webrtc/modules/audio_processing/aec/aec_rdft.h"

 

#include

 

#include "webrtc/system_wrappers/interface/cpu_features_wrapper.h"

#include "webrtc/typedefs.h"

 

// constants shared by all paths (C, SSE2).

float rdft_w[64];

// constants used by the C path.

float rdft_wk3ri_first[32];

float rdft_wk3ri_second[32];

// constants used by SSE2 but initialized in C path.

ALIGN16_BEG float ALIGN16_END rdft_wk1r[32];

ALIGN16_BEG float ALIGN16_END rdft_wk2r[32];

ALIGN16_BEG float ALIGN16_END rdft_wk3r[32];

ALIGN16_BEG float ALIGN16_END rdft_wk1i[32];

ALIGN16_BEG float ALIGN16_END rdft_wk2i[32];

ALIGN16_BEG float ALIGN16_END rdft_wk3i[32];

ALIGN16_BEG float ALIGN16_END cftmdl_wk1r[4];

 

static int ip[16];

 

static void bitrv2_32(int* ip1, float* a) {

  const int n = 32;

  int j, j1, k, k1, m, m2;

  float xr, xi, yr, yi;

 

  ip1[0] = 0;

  {

    int l = n;

    m = 1;

    while ((m << 3) < l) {

      l >>= 1;

      for (j = 0; j < m; j++) {

        ip1[m + j] = ip1[j] + l;

      }

      m <<= 1;

    }

  }

  m2 = 2 * m;

  for (k = 0; k < m; k++) {

    for (j = 0; j < k; j++) {

      j1 = 2 * j + ip1[k];

      k1 = 2 * k + ip1[j];

      xr = a[j1];

      xi = a[j1 + 1];

      yr = a[k1];

      yi = a[k1 + 1];

      a[j1] = yr;

      a[j1 + 1] = yi;

      a[k1] = xr;

      a[k1 + 1] = xi;

      j1 += m2;

      k1 += 2 * m2;

      xr = a[j1];

      xi = a[j1 + 1];

      yr = a[k1];

      yi = a[k1 + 1];

      a[j1] = yr;

      a[j1 + 1] = yi;

      a[k1] = xr;

      a[k1 + 1] = xi;

      j1 += m2;

      k1 -= m2;

      xr = a[j1];

      xi = a[j1 + 1];

      yr = a[k1];

      yi = a[k1 + 1];

      a[j1] = yr;

      a[j1 + 1] = yi;

      a[k1] = xr;

      a[k1 + 1] = xi;

      j1 += m2;

      k1 += 2 * m2;

      xr = a[j1];

      xi = a[j1 + 1];

      yr = a[k1];

      yi = a[k1 + 1];

      a[j1] = yr;

      a[j1 + 1] = yi;

      a[k1] = xr;

      a[k1 + 1] = xi;

    }

    j1 = 2 * k + m2 + ip1[k];

    k1 = j1 + m2;

    xr = a[j1];

    xi = a[j1 + 1];

    yr = a[k1];

    yi = a[k1 + 1];

    a[j1] = yr;

    a[j1 + 1] = yi;

    a[k1] = xr;

    a[k1 + 1] = xi;

  }

}

static void bitrv2_128(float* a) {

  /*

      Following things have been attempted but are no faster:

      (a) Storing the swap indexes in a LUT (index calculations are done

          for 'free' while waiting on memory/L1).

      (b) Consolidate the load/store of two consecutive floats by a 64 bit

          integer (execution is memory/L1 bound).

      (c) Do a mix of floats and 64 bit integer to maximize register

          utilization (execution is memory/L1 bound).

      (d) Replacing ip[i] by ((k<<31)>>25) + ((k >> 1)<<5).

      (e) Hard-coding of the offsets to completely eliminates index

          calculations.

  */

  unsigned int j, j1, k, k1;

  float xr, xi, yr, yi;

  static const int ip2[4] = {0, 64, 32, 96};

  for (k = 0; k < 4; k++) {

    for (j = 0; j < k; j++) {

      j1 = 2 * j + ip2[k];

      k1 = 2 * k + ip2[j];

      xr = a[j1 + 0];

      xi = a[j1 + 1];

      yr = a[k1 + 0];

      yi = a[k1 + 1];

      a[j1 + 0] = yr;

      a[j1 + 1] = yi;

      a[k1 + 0] = xr;

      a[k1 + 1] = xi;

      j1 += 8;

      k1 += 16;

      xr = a[j1 + 0];

      xi = a[j1 + 1];

      yr = a[k1 + 0];

      yi = a[k1 + 1];

      a[j1 + 0] = yr;

      a[j1 + 1] = yi;

      a[k1 + 0] = xr;

      a[k1 + 1] = xi;

      j1 += 8;

      k1 -= 8;

      xr = a[j1 + 0];

      xi = a[j1 + 1];

      yr = a[k1 + 0];

      yi = a[k1 + 1];

      a[j1 + 0] = yr;

      a[j1 + 1] = yi;

      a[k1 + 0] = xr;

      a[k1 + 1] = xi;

      j1 += 8;

      k1 += 16;

      xr = a[j1 + 0];

      xi = a[j1 + 1];

      yr = a[k1 + 0];

      yi = a[k1 + 1];

      a[j1 + 0] = yr;

      a[j1 + 1] = yi;

      a[k1 + 0] = xr;

      a[k1 + 1] = xi;

    }

    j1 = 2 * k + 8 + ip2[k];

    k1 = j1 + 8;

    xr = a[j1 + 0];

    xi = a[j1 + 1];

    yr = a[k1 + 0];

    yi = a[k1 + 1];

    a[j1 + 0] = yr;

    a[j1 + 1] = yi;

    a[k1 + 0] = xr;

    a[k1 + 1] = xi;

  }

}

static void makewt_32(void) {

  const int nw = 32;

  int j, nwh;

  float delta, x, y;

  ip[0] = nw;

  ip[1] = 1;

  nwh = nw >> 1;

  delta = atanf(1.0f) / nwh;

  rdft_w[0] = 1;

  rdft_w[1] = 0;

  rdft_w[nwh] = cosf(delta * nwh);

  rdft_w[nwh + 1] = rdft_w[nwh];

  for (j = 2; j < nwh; j += 2) {

    x = cosf(delta * j);

    y = sinf(delta * j);

    rdft_w[j] = x;

    rdft_w[j + 1] = y;

    rdft_w[nw - j] = y;

    rdft_w[nw - j + 1] = x;

  }

  bitrv2_32(ip + 2, rdft_w);

  // pre-calculate constants used by cft1st_128 and cftmdl_128...

  cftmdl_wk1r[0] = rdft_w[2];

  cftmdl_wk1r[1] = rdft_w[2];

  cftmdl_wk1r[2] = rdft_w[2];

  cftmdl_wk1r[3] = -rdft_w[2];

  {

    int k1;

    for (k1 = 0, j = 0; j < 128; j += 16, k1 += 2) {

      const int k2 = 2 * k1;

      const float wk2r = rdft_w[k1 + 0];

      const float wk2i = rdft_w[k1 + 1];

      float wk1r, wk1i;

      // ... scalar version.

      wk1r = rdft_w[k2 + 0];

      wk1i = rdft_w[k2 + 1];

      rdft_wk3ri_first[k1 + 0] = wk1r - 2 * wk2i * wk1i;

      rdft_wk3ri_first[k1 + 1] = 2 * wk2i * wk1r - wk1i;

      wk1r = rdft_w[k2 + 2];

      wk1i = rdft_w[k2 + 3];

      rdft_wk3ri_second[k1 + 0] = wk1r - 2 * wk2r * wk1i;

      rdft_wk3ri_second[k1 + 1] = 2 * wk2r * wk1r - wk1i;

      // ... vector version.

      rdft_wk1r[k2 + 0] = rdft_w[k2 + 0];

      rdft_wk1r[k2 + 1] = rdft_w[k2 + 0];

      rdft_wk1r[k2 + 2] = rdft_w[k2 + 2];

      rdft_wk1r[k2 + 3] = rdft_w[k2 + 2];

      rdft_wk2r[k2 + 0] = rdft_w[k1 + 0];

      rdft_wk2r[k2 + 1] = rdft_w[k1 + 0];

      rdft_wk2r[k2 + 2] = -rdft_w[k1 + 1];

      rdft_wk2r[k2 + 3] = -rdft_w[k1 + 1];

      rdft_wk3r[k2 + 0] = rdft_wk3ri_first[k1 + 0];

      rdft_wk3r[k2 + 1] = rdft_wk3ri_first[k1 + 0];

      rdft_wk3r[k2 + 2] = rdft_wk3ri_second[k1 + 0];

      rdft_wk3r[k2 + 3] = rdft_wk3ri_second[k1 + 0];

      rdft_wk1i[k2 + 0] = -rdft_w[k2 + 1];

      rdft_wk1i[k2 + 1] = rdft_w[k2 + 1];

      rdft_wk1i[k2 + 2] = -rdft_w[k2 + 3];

      rdft_wk1i[k2 + 3] = rdft_w[k2 + 3];

      rdft_wk2i[k2 + 0] = -rdft_w[k1 + 1];

      rdft_wk2i[k2 + 1] = rdft_w[k1 + 1];

      rdft_wk2i[k2 + 2] = -rdft_w[k1 + 0];

      rdft_wk2i[k2 + 3] = rdft_w[k1 + 0];

      rdft_wk3i[k2 + 0] = -rdft_wk3ri_first[k1 + 1];

      rdft_wk3i[k2 + 1] = rdft_wk3ri_first[k1 + 1];

      rdft_wk3i[k2 + 2] = -rdft_wk3ri_second[k1 + 1];

      rdft_wk3i[k2 + 3] = rdft_wk3ri_second[k1 + 1];

    }

  }

}

static void makect_32(void) {

  float* c = rdft_w + 32;

  const int nc = 32;

  int j, nch;

  float delta;

  ip[1] = nc;

  nch = nc >> 1;

  delta = atanf(1.0f) / nch;

  c[0] = cosf(delta * nch);

  c[nch] = 0.5f * c[0];

  for (j = 1; j < nch; j++) {

    c[j] = 0.5f * cosf(delta * j);

    c[nc - j] = 0.5f * sinf(delta * j);

  }

}

static void cft1st_128_C(float* a) {

  const int n = 128;

  int j, k1, k2;

  float wk1r, wk1i, wk2r, wk2i, wk3r, wk3i;

  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;

  x0r = a[0] + a[2];

  x0i = a[1] + a[3];

  x1r = a[0] - a[2];

  x1i = a[1] - a[3];

  x2r = a[4] + a[6];

  x2i = a[5] + a[7];

  x3r = a[4] - a[6];

  x3i = a[5] - a[7];

  a[0] = x0r + x2r;

  a[1] = x0i + x2i;

  a[4] = x0r - x2r;

  a[5] = x0i - x2i;

  a[2] = x1r - x3i;

  a[3] = x1i + x3r;

  a[6] = x1r + x3i;

  a[7] = x1i - x3r;

  wk1r = rdft_w[2];

  x0r = a[8] + a[10];

  x0i = a[9] + a[11];

  x1r = a[8] - a[10];

  x1i = a[9] - a[11];

  x2r = a[12] + a[14];

  x2i = a[13] + a[15];

  x3r = a[12] - a[14];

  x3i = a[13] - a[15];

  a[8] = x0r + x2r;

  a[9] = x0i + x2i;

  a[12] = x2i - x0i;

  a[13] = x0r - x2r;

  x0r = x1r - x3i;

  x0i = x1i + x3r;

  a[10] = wk1r * (x0r - x0i);

  a[11] = wk1r * (x0r + x0i);

  x0r = x3i + x1r;

  x0i = x3r - x1i;

  a[14] = wk1r * (x0i - x0r);

  a[15] = wk1r * (x0i + x0r);

  k1 = 0;

  for (j = 16; j < n; j += 16) {

    k1 += 2;

    k2 = 2 * k1;

    wk2r = rdft_w[k1 + 0];

    wk2i = rdft_w[k1 + 1];

    wk1r = rdft_w[k2 + 0];

    wk1i = rdft_w[k2 + 1];

    wk3r = rdft_wk3ri_first[k1 + 0];

    wk3i = rdft_wk3ri_first[k1 + 1];

    x0r = a[j + 0] + a[j + 2];

    x0i = a[j + 1] + a[j + 3];

    x1r = a[j + 0] - a[j + 2];

    x1i = a[j + 1] - a[j + 3];

    x2r = a[j + 4] + a[j + 6];

    x2i = a[j + 5] + a[j + 7];

    x3r = a[j + 4] - a[j + 6];

    x3i = a[j + 5] - a[j + 7];

    a[j + 0] = x0r + x2r;

    a[j + 1] = x0i + x2i;

    x0r -= x2r;

    x0i -= x2i;

    a[j + 4] = wk2r * x0r - wk2i * x0i;

    a[j + 5] = wk2r * x0i + wk2i * x0r;

    x0r = x1r - x3i;

    x0i = x1i + x3r;

    a[j + 2] = wk1r * x0r - wk1i * x0i;

    a[j + 3] = wk1r * x0i + wk1i * x0r;

    x0r = x1r + x3i;

    x0i = x1i - x3r;

    a[j + 6] = wk3r * x0r - wk3i * x0i;

    a[j + 7] = wk3r * x0i + wk3i * x0r;

    wk1r = rdft_w[k2 + 2];

    wk1i = rdft_w[k2 + 3];

    wk3r = rdft_wk3ri_second[k1 + 0];

    wk3i = rdft_wk3ri_second[k1 + 1];

    x0r = a[j + 8] + a[j + 10];

    x0i = a[j + 9] + a[j + 11];

    x1r = a[j + 8] - a[j + 10];

    x1i = a[j + 9] - a[j + 11];

    x2r = a[j + 12] + a[j + 14];

    x2i = a[j + 13] + a[j + 15];

    x3r = a[j + 12] - a[j + 14];

    x3i = a[j + 13] - a[j + 15];

    a[j + 8] = x0r + x2r;

    a[j + 9] = x0i + x2i;

    x0r -= x2r;

    x0i -= x2i;

    a[j + 12] = -wk2i * x0r - wk2r * x0i;

    a[j + 13] = -wk2i * x0i + wk2r * x0r;

    x0r = x1r - x3i;

    x0i = x1i + x3r;

    a[j + 10] = wk1r * x0r - wk1i * x0i;

    a[j + 11] = wk1r * x0i + wk1i * x0r;

    x0r = x1r + x3i;

    x0i = x1i - x3r;

    a[j + 14] = wk3r * x0r - wk3i * x0i;

    a[j + 15] = wk3r * x0i + wk3i * x0r;

  }

}

static void cftmdl_128_C(float* a) {

  const int l = 8;

  const int n = 128;

  const int m = 32;

  int j0, j1, j2, j3, k, k1, k2, m2;

  float wk1r, wk1i, wk2r, wk2i, wk3r, wk3i;

  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;

  for (j0 = 0; j0 < l; j0 += 2) {

    j1 = j0 + 8;

    j2 = j0 + 16;

    j3 = j0 + 24;

    x0r = a[j0 + 0] + a[j1 + 0];

    x0i = a[j0 + 1] + a[j1 + 1];

    x1r = a[j0 + 0] - a[j1 + 0];

    x1i = a[j0 + 1] - a[j1 + 1];

    x2r = a[j2 + 0] + a[j3 + 0];

    x2i = a[j2 + 1] + a[j3 + 1];

    x3r = a[j2 + 0] - a[j3 + 0];

    x3i = a[j2 + 1] - a[j3 + 1];

    a[j0 + 0] = x0r + x2r;

    a[j0 + 1] = x0i + x2i;

    a[j2 + 0] = x0r - x2r;

    a[j2 + 1] = x0i - x2i;

    a[j1 + 0] = x1r - x3i;

    a[j1 + 1] = x1i + x3r;

    a[j3 + 0] = x1r + x3i;

    a[j3 + 1] = x1i - x3r;

  }

  wk1r = rdft_w[2];

  for (j0 = m; j0 < l + m; j0 += 2) {

    j1 = j0 + 8;

    j2 = j0 + 16;

    j3 = j0 + 24;

    x0r = a[j0 + 0] + a[j1 + 0];

    x0i = a[j0 + 1] + a[j1 + 1];

    x1r = a[j0 + 0] - a[j1 + 0];

    x1i = a[j0 + 1] - a[j1 + 1];

    x2r = a[j2 + 0] + a[j3 + 0];

    x2i = a[j2 + 1] + a[j3 + 1];

    x3r = a[j2 + 0] - a[j3 + 0];

    x3i = a[j2 + 1] - a[j3 + 1];

    a[j0 + 0] = x0r + x2r;

    a[j0 + 1] = x0i + x2i;

    a[j2 + 0] = x2i - x0i;

    a[j2 + 1] = x0r - x2r;

    x0r = x1r - x3i;

    x0i = x1i + x3r;

    a[j1 + 0] = wk1r * (x0r - x0i);

    a[j1 + 1] = wk1r * (x0r + x0i);

    x0r = x3i + x1r;

    x0i = x3r - x1i;

    a[j3 + 0] = wk1r * (x0i - x0r);

    a[j3 + 1] = wk1r * (x0i + x0r);

  }

  k1 = 0;

  m2 = 2 * m;

  for (k = m2; k < n; k += m2) {

    k1 += 2;

    k2 = 2 * k1;

    wk2r = rdft_w[k1 + 0];

    wk2i = rdft_w[k1 + 1];

    wk1r = rdft_w[k2 + 0];

    wk1i = rdft_w[k2 + 1];

    wk3r = rdft_wk3ri_first[k1 + 0];

    wk3i = rdft_wk3ri_first[k1 + 1];

    for (j0 = k; j0 < l + k; j0 += 2) {

      j1 = j0 + 8;

      j2 = j0 + 16;

      j3 = j0 + 24;

      x0r = a[j0 + 0] + a[j1 + 0];

      x0i = a[j0 + 1] + a[j1 + 1];

      x1r = a[j0 + 0] - a[j1 + 0];

      x1i = a[j0 + 1] - a[j1 + 1];

      x2r = a[j2 + 0] + a[j3 + 0];

      x2i = a[j2 + 1] + a[j3 + 1];

      x3r = a[j2 + 0] - a[j3 + 0];

      x3i = a[j2 + 1] - a[j3 + 1];

      a[j0 + 0] = x0r + x2r;

      a[j0 + 1] = x0i + x2i;

      x0r -= x2r;

      x0i -= x2i;

      a[j2 + 0] = wk2r * x0r - wk2i * x0i;

      a[j2 + 1] = wk2r * x0i + wk2i * x0r;

      x0r = x1r - x3i;

      x0i = x1i + x3r;

      a[j1 + 0] = wk1r * x0r - wk1i * x0i;

      a[j1 + 1] = wk1r * x0i + wk1i * x0r;

      x0r = x1r + x3i;

      x0i = x1i - x3r;

      a[j3 + 0] = wk3r * x0r - wk3i * x0i;

      a[j3 + 1] = wk3r * x0i + wk3i * x0r;

    }

    wk1r = rdft_w[k2 + 2];

    wk1i = rdft_w[k2 + 3];

    wk3r = rdft_wk3ri_second[k1 + 0];

    wk3i = rdft_wk3ri_second[k1 + 1];

    for (j0 = k + m; j0 < l + (k + m); j0 += 2) {

      j1 = j0 + 8;

      j2 = j0 + 16;

      j3 = j0 + 24;

      x0r = a[j0 + 0] + a[j1 + 0];

      x0i = a[j0 + 1] + a[j1 + 1];

      x1r = a[j0 + 0] - a[j1 + 0];

      x1i = a[j0 + 1] - a[j1 + 1];

      x2r = a[j2 + 0] + a[j3 + 0];

      x2i = a[j2 + 1] + a[j3 + 1];

      x3r = a[j2 + 0] - a[j3 + 0];

      x3i = a[j2 + 1] - a[j3 + 1];

      a[j0 + 0] = x0r + x2r;

      a[j0 + 1] = x0i + x2i;

      x0r -= x2r;

      x0i -= x2i;

      a[j2 + 0] = -wk2i * x0r - wk2r * x0i;

      a[j2 + 1] = -wk2i * x0i + wk2r * x0r;

      x0r = x1r - x3i;

      x0i = x1i + x3r;

      a[j1 + 0] = wk1r * x0r - wk1i * x0i;

      a[j1 + 1] = wk1r * x0i + wk1i * x0r;

      x0r = x1r + x3i;

      x0i = x1i - x3r;

      a[j3 + 0] = wk3r * x0r - wk3i * x0i;

      a[j3 + 1] = wk3r * x0i + wk3i * x0r;

    }

  }

}

static void cftfsub_128(float* a) {

  int j, j1, j2, j3, l;

  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;

  cft1st_128(a);

  cftmdl_128(a);

  l = 32;

  for (j = 0; j < l; j += 2) {

    j1 = j + l;

    j2 = j1 + l;

    j3 = j2 + l;

    x0r = a[j] + a[j1];

    x0i = a[j + 1] + a[j1 + 1];

    x1r = a[j] - a[j1];

    x1i = a[j + 1] - a[j1 + 1];

    x2r = a[j2] + a[j3];

    x2i = a[j2 + 1] + a[j3 + 1];

    x3r = a[j2] - a[j3];

    x3i = a[j2 + 1] - a[j3 + 1];

    a[j] = x0r + x2r;

    a[j + 1] = x0i + x2i;

    a[j2] = x0r - x2r;

    a[j2 + 1] = x0i - x2i;

    a[j1] = x1r - x3i;

    a[j1 + 1] = x1i + x3r;

    a[j3] = x1r + x3i;

    a[j3 + 1] = x1i - x3r;

  }

}

static void cftbsub_128(float* a) {

  int j, j1, j2, j3, l;

  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;

  cft1st_128(a);

  cftmdl_128(a);

  l = 32;

  for (j = 0; j < l; j += 2) {

    j1 = j + l;

    j2 = j1 + l;

    j3 = j2 + l;

    x0r = a[j] + a[j1];

    x0i = -a[j + 1] - a[j1 + 1];

    x1r = a[j] - a[j1];

    x1i = -a[j + 1] + a[j1 + 1];

    x2r = a[j2] + a[j3];

    x2i = a[j2 + 1] + a[j3 + 1];

    x3r = a[j2] - a[j3];

    x3i = a[j2 + 1] - a[j3 + 1];

    a[j] = x0r + x2r;

    a[j + 1] = x0i - x2i;

    a[j2] = x0r - x2r;

    a[j2 + 1] = x0i + x2i;

    a[j1] = x1r - x3i;

    a[j1 + 1] = x1i - x3r;

    a[j3] = x1r + x3i;

    a[j3 + 1] = x1i + x3r;

  }

}

static void rftfsub_128_C(float* a) {

  const float* c = rdft_w + 32;

  int j1, j2, k1, k2;

  float wkr, wki, xr, xi, yr, yi;

  for (j1 = 1, j2 = 2; j2 < 64; j1 += 1, j2 += 2) {

    k2 = 128 - j2;

    k1 = 32 - j1;

    wkr = 0.5f - c[k1];

    wki = c[j1];

    xr = a[j2 + 0] - a[k2 + 0];

    xi = a[j2 + 1] + a[k2 + 1];

    yr = wkr * xr - wki * xi;

    yi = wkr * xi + wki * xr;

    a[j2 + 0] -= yr;

    a[j2 + 1] -= yi;

    a[k2 + 0] += yr;

    a[k2 + 1] -= yi;

  }

}

static void rftbsub_128_C(float* a) {

  const float* c = rdft_w + 32;

  int j1, j2, k1, k2;

  float wkr, wki, xr, xi, yr, yi;

  a[1] = -a[1];

  for (j1 = 1, j2 = 2; j2 < 64; j1 += 1, j2 += 2) {

    k2 = 128 - j2;

    k1 = 32 - j1;

    wkr = 0.5f - c[k1];

    wki = c[j1];

    xr = a[j2 + 0] - a[k2 + 0];

    xi = a[j2 + 1] + a[k2 + 1];

    yr = wkr * xr + wki * xi;

    yi = wkr * xi - wki * xr;

    a[j2 + 0] = a[j2 + 0] - yr;

    a[j2 + 1] = yi - a[j2 + 1];

    a[k2 + 0] = yr + a[k2 + 0];

    a[k2 + 1] = yi - a[k2 + 1];

  }

  a[65] = -a[65];

}

void aec_rdft_forward_128(float* a) {

  float xi;

  bitrv2_128(a);

  cftfsub_128(a);

  rftfsub_128(a);

  xi = a[0] - a[1];

  a[0] += a[1];

  a[1] = xi;

}

void aec_rdft_inverse_128(float* a) {

  a[1] = 0.5f * (a[0] - a[1]);

  a[0] -= a[1];

  rftbsub_128(a);

  bitrv2_128(a);

  cftbsub_128(a);

}

// code path selection

rft_sub_128_t cft1st_128;

rft_sub_128_t cftmdl_128;

rft_sub_128_t rftfsub_128;

rft_sub_128_t rftbsub_128;

void aec_rdft_init(void) {

  cft1st_128 = cft1st_128_C;

  cftmdl_128 = cftmdl_128_C;

  rftfsub_128 = rftfsub_128_C;

  rftbsub_128 = rftbsub_128_C;

#if defined(WEBRTC_ARCH_X86_FAMILY)

  if (WebRtc_GetCPUInfo(kSSE2)) {

    aec_rdft_init_sse2();

  }

#endif

  // init library constants.

  makewt_32();

  makect_32();

}