aboutsummaryrefslogtreecommitdiff
path: root/test_pffft.c
blob: 99a4e11053b9be4fdb0f6d5043412530d2367a70 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
/*
  Copyright (c) 2013 Julien Pommier.

  Small test & bench for PFFFT, comparing its performance with the scalar FFTPACK, FFTW, and Apple vDSP

  How to build: 

  on linux, with fftw3:
  gcc -o test_pffft -DHAVE_FFTW -msse -mfpmath=sse -O3 -Wall -W pffft.c test_pffft.c fftpack.c -L/usr/local/lib -I/usr/local/include/ -lfftw3f -lm

  on macos, without fftw3:
  clang -o test_pffft -DHAVE_VECLIB -O3 -Wall -W pffft.c test_pffft.c fftpack.c -L/usr/local/lib -I/usr/local/include/ -framework Accelerate

  on macos, with fftw3:
  clang -o test_pffft -DHAVE_FFTW -DHAVE_VECLIB -O3 -Wall -W pffft.c test_pffft.c fftpack.c -L/usr/local/lib -I/usr/local/include/ -lfftw3f -framework Accelerate

  as alternative: replace clang by gcc.

  on windows, with visual c++:
  cl /Ox -D_USE_MATH_DEFINES /arch:SSE test_pffft.c pffft.c fftpack.c
  
  build without SIMD instructions:
  gcc -o test_pffft -DPFFFT_SIMD_DISABLE -O3 -Wall -W pffft.c test_pffft.c fftpack.c -lm

 */

#include "pffft.h"

#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <assert.h>
#include <string.h>


/* EXPECTED_DYN_RANGE in dB:
 * single precision float has 24 bits mantissa
 * => 24 Bits * 6 dB = 144 dB
 * allow a few dB tolerance (even 144 dB looks good on my PC)
 */
#define EXPECTED_DYN_RANGE  140.0

/* maximum allowed phase error in degree */
#define DEG_ERR_LIMIT   1E-4

/* maximum allowed magnitude error in amplitude (of 1.0 or 1.1) */
#define MAG_ERR_LIMIT  1E-6


#define PRINT_SPEC  0

#define PWR2LOG(PWR)  ( (PWR) < 1E-30 ? 10.0*log10(1E-30) : 10.0*log10(PWR) )



int test(int N, int cplx, int useOrdered) {
  int Nfloat = (cplx ? N*2 : N);
  float *X = pffft_aligned_malloc((unsigned)Nfloat * sizeof(float));
  float *Y = pffft_aligned_malloc((unsigned)Nfloat * sizeof(float));
  float *Z = pffft_aligned_malloc((unsigned)Nfloat * sizeof(float));
  float *W = pffft_aligned_malloc((unsigned)Nfloat * sizeof(float));
  int k, j, m, iter, kmaxOther, retError = 0;
  double freq, dPhi, phi, phi0;
  double pwr, pwrCar, pwrOther, err, errSum, mag, expextedMag;
  float amp = 1.0F;

  assert( pffft_is_power_of_two(N) );

  PFFFT_Setup *s = pffft_new_setup(N, cplx ? PFFFT_COMPLEX : PFFFT_REAL);
  assert(s);
  if (!s) {
    printf("Error setting up PFFFT!\n");
    return 1;
  }

  for ( k = m = 0; k < (cplx? N : (1 + N/2) ); k += N/16, ++m )
  {
    amp = ( (m % 3) == 0 ) ? 1.0F : 1.1F;
    freq = (k < N/2) ? ((double)k / N) : ((double)(k-N) / N);
    dPhi = 2.0 * M_PI * freq;
    if ( dPhi < 0.0 )
      dPhi += 2.0 * M_PI;

    iter = -1;
    while (1)
    {
      ++iter;

      if (iter)
        printf("bin %d: dphi = %f for freq %f\n", k, dPhi, freq);

      /* generate cosine carrier as time signal - start at defined phase phi0 */
      phi = phi0 = (m % 4) * 0.125 * M_PI;  /* have phi0 < 90 deg to be normalized */
      for ( j = 0; j < N; ++j )
      {
        if (cplx) {
          X[2*j] = amp * (float)cos(phi);  /* real part */
          X[2*j+1] = amp * (float)sin(phi);  /* imag part */
        }
        else
          X[j] = amp * (float)cos(phi);  /* only real part */

        /* phase increment .. stay normalized - cos()/sin() might degrade! */
        phi += dPhi;
        if ( phi >= M_PI )
          phi -= 2.0 * M_PI;
      }

      /* forward transform from X --> Y  .. using work buffer W */
      if ( useOrdered )
        pffft_transform_ordered(s, X, Y, W, PFFFT_FORWARD );
      else
      {
        pffft_transform(s, X, Z, W, PFFFT_FORWARD );  /* temporarily use Z for reordering */
        pffft_zreorder(s, Z, Y, PFFFT_FORWARD );
      }

      pwrOther = -1.0;
      pwrCar = 0;


      /* for positive frequencies: 0 to 0.5 * samplerate */
      /* and also for negative frequencies: -0.5 * samplerate to 0 */
      for ( j = 0; j < ( cplx ? N : (1 + N/2) ); ++j )
      {
        if (!cplx && !j)  /* special treatment for DC for real input */
          pwr = Y[j]*Y[j];
        else if (!cplx && j == N/2)  /* treat 0.5 * samplerate */
          pwr = Y[1] * Y[1];  /* despite j (for freq calculation) we have index 1 */
        else
          pwr = Y[2*j] * Y[2*j] + Y[2*j+1] * Y[2*j+1];
        if (iter || PRINT_SPEC)
          printf("%s fft %d:  pwr[j = %d] = %g == %f dB\n", (cplx ? "cplx":"real"), N, j, pwr, PWR2LOG(pwr) );
        if (k == j)
          pwrCar = pwr;
        else if ( pwr > pwrOther ) {
          pwrOther = pwr;
          kmaxOther = j;
        }
      }

      if ( PWR2LOG(pwrCar) - PWR2LOG(pwrOther) < EXPECTED_DYN_RANGE ) {
        printf("%s fft %d amp %f iter %d:\n", (cplx ? "cplx":"real"), N, amp, iter);
        printf("  carrier power  at bin %d: %g == %f dB\n", k, pwrCar, PWR2LOG(pwrCar) );
        printf("  carrier mag || at bin %d: %g\n", k, sqrt(pwrCar) );
        printf("  max other pwr  at bin %d: %g == %f dB\n", kmaxOther, pwrOther, PWR2LOG(pwrOther) );
        printf("  dynamic range: %f dB\n\n", PWR2LOG(pwrCar) - PWR2LOG(pwrOther) );
        retError = 1;
        if ( iter == 0 )
          continue;
      }

      if ( k > 0 && k != N/2 )
      {
        phi = atan2( Y[2*k+1], Y[2*k] );
        if ( fabs( phi - phi0) > DEG_ERR_LIMIT * M_PI / 180.0 )
        {
        retError = 1;
        printf("%s fft %d  bin %d amp %f : phase mismatch! phase = %f deg   expected = %f deg\n",
            (cplx ? "cplx":"real"), N, k, amp, phi * 180.0 / M_PI, phi0 * 180.0 / M_PI );
        }
      }

      expextedMag = cplx ? amp : ( (k == 0 || k == N/2) ? amp : (amp/2) );
      mag = sqrt(pwrCar) / N;
      if ( fabs(mag - expextedMag) > MAG_ERR_LIMIT )
      {
        retError = 1;
        printf("%s fft %d  bin %d amp %f : mag = %g   expected = %g\n", (cplx ? "cplx":"real"), N, k, amp, mag, expextedMag );
      }


      /* now convert spectrum back */
      pffft_transform_ordered(s, Y, Z, W, PFFFT_BACKWARD);

      errSum = 0.0;
      for ( j = 0; j < (cplx ? (2*N) : N); ++j )
      {
        /* scale back */
        Z[j] /= N;
        /* square sum errors over real (and imag parts) */
        err = (X[j]-Z[j]) * (X[j]-Z[j]);
        errSum += err;
      }

      if ( errSum > N * 1E-7 )
      {
        retError = 1;
        printf("%s fft %d  bin %d : inverse FFT doesn't match original signal! errSum = %g ; mean err = %g\n", (cplx ? "cplx":"real"), N, k, errSum, errSum / N);
      }

      break;
    }

  }
  pffft_destroy_setup(s);
  pffft_aligned_free(X);
  pffft_aligned_free(Y);
  pffft_aligned_free(Z);
  pffft_aligned_free(W);

  return retError;
}



int main(int argc, char **argv)
{
  int N, result, resN, resAll, k, resNextPw2, resIsPw2, resFFT;

  int inp_power_of_two[] = { 1, 2, 3, 4, 5, 6, 7, 8,  9, 511, 512,  513 };
  int ref_power_of_two[] = { 1, 2, 4, 4, 8, 8, 8, 8, 16, 512, 512, 1024 };

  resNextPw2 = 0;
  resIsPw2 = 0;
  for ( k = 0; k < (sizeof(inp_power_of_two)/sizeof(inp_power_of_two[0])); ++k) {
    N = pffft_next_power_of_two(inp_power_of_two[k]);
    if (N != ref_power_of_two[k]) {
      resNextPw2 = 1;
      printf("pffft_next_power_of_two(%d) does deliver %d, which is not reference result %d!\n",
        inp_power_of_two[k], N, ref_power_of_two[k] );
    }

    result = pffft_is_power_of_two(inp_power_of_two[k]);
    if (inp_power_of_two[k] == ref_power_of_two[k]) {
      if (!result) {
        resIsPw2 = 1;
        printf("pffft_is_power_of_two(%d) delivers false; expected true!\n", inp_power_of_two[k]);
      }
    } else {
      if (result) {
        resIsPw2 = 1;
        printf("pffft_is_power_of_two(%d) delivers true; expected false!\n", inp_power_of_two[k]);
      }
    }
  }
  if (!resNextPw2)
    printf("tests for pffft_next_power_of_two() succeeded successfully.\n");
  if (!resIsPw2)
    printf("tests for pffft_is_power_of_two() succeeded successfully.\n");

  resFFT = 0;
  for ( N = 32; N <= 65536; N *= 2 )
  {
    result = test(N, 1 /* cplx fft */, 1 /* useOrdered */);
    resN = result;
    resFFT |= result;

    result = test(N, 0 /* cplx fft */, 1 /* useOrdered */);
    resN |= result;
    resFFT |= result;

    result = test(N, 1 /* cplx fft */, 0 /* useOrdered */);
    resN |= result;
    resFFT |= result;

    result = test(N, 0 /* cplx fft */, 0 /* useOrdered */);
    resN |= result;
    resFFT |= result;

    if (!resN)
      printf("tests for size %d succeeded successfully.\n", N);
  }

  if (!resFFT)
    printf("all pffft transform tests (FORWARD/BACKWARD, REAL/COMPLEX) succeeded successfully.\n");

  resAll = resNextPw2 | resIsPw2 | resFFT;
  if (!resAll)
    printf("all tests succeeded successfully.\n");
  else
    printf("there are failed tests!\n");

  return resAll;
}