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
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
|
////////////////////////////////////////////////////////////////////////////////
///
/// General FIR digital filter routines with MMX optimization.
///
/// Notes : MMX optimized functions reside in a separate, platform-specific file,
/// e.g. 'mmx_win.cpp' or 'mmx_gcc.cpp'
///
/// This source file contains OpenMP optimizations that allow speeding up the
/// corss-correlation algorithm by executing it in several threads / CPU cores
/// in parallel. See the following article link for more detailed discussion
/// about SoundTouch OpenMP optimizations:
/// http://www.softwarecoven.com/parallel-computing-in-embedded-mobile-devices
///
/// Author : Copyright (c) Olli Parviainen
/// Author e-mail : oparviai 'at' iki.fi
/// SoundTouch WWW: http://www.surina.net/soundtouch
///
////////////////////////////////////////////////////////////////////////////////
//
// License :
//
// SoundTouch audio processing library
// Copyright (c) Olli Parviainen
//
// This library is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 2.1 of the License, or (at your option) any later version.
//
// This library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
// Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License along with this library; if not, write to the Free Software
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
//
////////////////////////////////////////////////////////////////////////////////
#include <memory.h>
#include <assert.h>
#include <math.h>
#include <stdlib.h>
#include "FIRFilter.h"
#include "cpu_detect.h"
using namespace soundtouch;
/*****************************************************************************
*
* Implementation of the class 'FIRFilter'
*
*****************************************************************************/
FIRFilter::FIRFilter()
{
resultDivFactor = 0;
resultDivider = 0;
length = 0;
lengthDiv8 = 0;
filterCoeffs = NULL;
}
FIRFilter::~FIRFilter()
{
delete[] filterCoeffs;
}
// Usual C-version of the filter routine for stereo sound
uint FIRFilter::evaluateFilterStereo(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples) const
{
int j, end;
#ifdef SOUNDTOUCH_FLOAT_SAMPLES
// when using floating point samples, use a scaler instead of a divider
// because division is much slower operation than multiplying.
double dScaler = 1.0 / (double)resultDivider;
#endif
assert(length != 0);
assert(src != NULL);
assert(dest != NULL);
assert(filterCoeffs != NULL);
end = 2 * (numSamples - length);
#pragma omp parallel for
for (j = 0; j < end; j += 2)
{
const SAMPLETYPE *ptr;
LONG_SAMPLETYPE suml, sumr;
uint i;
suml = sumr = 0;
ptr = src + j;
for (i = 0; i < length; i += 4)
{
// loop is unrolled by factor of 4 here for efficiency
suml += ptr[2 * i + 0] * filterCoeffs[i + 0] +
ptr[2 * i + 2] * filterCoeffs[i + 1] +
ptr[2 * i + 4] * filterCoeffs[i + 2] +
ptr[2 * i + 6] * filterCoeffs[i + 3];
sumr += ptr[2 * i + 1] * filterCoeffs[i + 0] +
ptr[2 * i + 3] * filterCoeffs[i + 1] +
ptr[2 * i + 5] * filterCoeffs[i + 2] +
ptr[2 * i + 7] * filterCoeffs[i + 3];
}
#ifdef SOUNDTOUCH_INTEGER_SAMPLES
suml >>= resultDivFactor;
sumr >>= resultDivFactor;
// saturate to 16 bit integer limits
suml = (suml < -32768) ? -32768 : (suml > 32767) ? 32767 : suml;
// saturate to 16 bit integer limits
sumr = (sumr < -32768) ? -32768 : (sumr > 32767) ? 32767 : sumr;
#else
suml *= dScaler;
sumr *= dScaler;
#endif // SOUNDTOUCH_INTEGER_SAMPLES
dest[j] = (SAMPLETYPE)suml;
dest[j + 1] = (SAMPLETYPE)sumr;
}
return numSamples - length;
}
// Usual C-version of the filter routine for mono sound
uint FIRFilter::evaluateFilterMono(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples) const
{
int j, end;
#ifdef SOUNDTOUCH_FLOAT_SAMPLES
// when using floating point samples, use a scaler instead of a divider
// because division is much slower operation than multiplying.
double dScaler = 1.0 / (double)resultDivider;
#endif
assert(length != 0);
end = numSamples - length;
#pragma omp parallel for
for (j = 0; j < end; j ++)
{
const SAMPLETYPE *pSrc = src + j;
LONG_SAMPLETYPE sum;
uint i;
sum = 0;
for (i = 0; i < length; i += 4)
{
// loop is unrolled by factor of 4 here for efficiency
sum += pSrc[i + 0] * filterCoeffs[i + 0] +
pSrc[i + 1] * filterCoeffs[i + 1] +
pSrc[i + 2] * filterCoeffs[i + 2] +
pSrc[i + 3] * filterCoeffs[i + 3];
}
#ifdef SOUNDTOUCH_INTEGER_SAMPLES
sum >>= resultDivFactor;
// saturate to 16 bit integer limits
sum = (sum < -32768) ? -32768 : (sum > 32767) ? 32767 : sum;
#else
sum *= dScaler;
#endif // SOUNDTOUCH_INTEGER_SAMPLES
dest[j] = (SAMPLETYPE)sum;
}
return end;
}
uint FIRFilter::evaluateFilterMulti(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels)
{
int j, end;
#ifdef SOUNDTOUCH_FLOAT_SAMPLES
// when using floating point samples, use a scaler instead of a divider
// because division is much slower operation than multiplying.
double dScaler = 1.0 / (double)resultDivider;
#endif
assert(length != 0);
assert(src != NULL);
assert(dest != NULL);
assert(filterCoeffs != NULL);
assert(numChannels < 16);
end = numChannels * (numSamples - length);
#pragma omp parallel for
for (j = 0; j < end; j += numChannels)
{
const SAMPLETYPE *ptr;
LONG_SAMPLETYPE sums[16];
uint c, i;
for (c = 0; c < numChannels; c ++)
{
sums[c] = 0;
}
ptr = src + j;
for (i = 0; i < length; i ++)
{
SAMPLETYPE coef=filterCoeffs[i];
for (c = 0; c < numChannels; c ++)
{
sums[c] += ptr[0] * coef;
ptr ++;
}
}
for (c = 0; c < numChannels; c ++)
{
#ifdef SOUNDTOUCH_INTEGER_SAMPLES
sums[c] >>= resultDivFactor;
#else
sums[c] *= dScaler;
#endif // SOUNDTOUCH_INTEGER_SAMPLES
dest[j+c] = (SAMPLETYPE)sums[c];
}
}
return numSamples - length;
}
// Set filter coeffiecients and length.
//
// Throws an exception if filter length isn't divisible by 8
void FIRFilter::setCoefficients(const SAMPLETYPE *coeffs, uint newLength, uint uResultDivFactor)
{
assert(newLength > 0);
if (newLength % 8) ST_THROW_RT_ERROR("FIR filter length not divisible by 8");
lengthDiv8 = newLength / 8;
length = lengthDiv8 * 8;
assert(length == newLength);
resultDivFactor = uResultDivFactor;
resultDivider = (SAMPLETYPE)::pow(2.0, (int)resultDivFactor);
delete[] filterCoeffs;
filterCoeffs = new SAMPLETYPE[length];
memcpy(filterCoeffs, coeffs, length * sizeof(SAMPLETYPE));
}
uint FIRFilter::getLength() const
{
return length;
}
// Applies the filter to the given sequence of samples.
//
// Note : The amount of outputted samples is by value of 'filter_length'
// smaller than the amount of input samples.
uint FIRFilter::evaluate(SAMPLETYPE *dest, const SAMPLETYPE *src, uint numSamples, uint numChannels)
{
assert
|