commit patch for compiling on mingw32
[flac.git] / src / libFLAC / lpc.c
1 /* libFLAC - Free Lossless Audio Codec library
2  * Copyright (C) 2000,2001  Josh Coalson
3  *
4  * This library is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Library General Public
6  * License as published by the Free Software Foundation; either
7  * version 2 of the License, or (at your option) any later version.
8  *
9  * This library is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
12  * Library General Public License for more details.
13  *
14  * You should have received a copy of the GNU Library General Public
15  * License along with this library; if not, write to the
16  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
17  * Boston, MA  02111-1307, USA.
18  */
19
20 #include <math.h>
21 #include <stdio.h>
22 #include "FLAC/assert.h"
23 #include "FLAC/format.h"
24 #include "private/lpc.h"
25
26 #ifndef M_LN2
27 /* math.h in VC++ doesn't seem to have this (how Microsoft is that?) */
28 #define M_LN2 0.69314718055994530942
29 #endif
30
31 void FLAC__lpc_compute_autocorrelation(const FLAC__real data[], unsigned data_len, unsigned lag, FLAC__real autoc[])
32 {
33         /* a readable, but slower, version */
34 #if 0
35         FLAC__real d;
36         unsigned i;
37
38         FLAC__ASSERT(lag > 0);
39         FLAC__ASSERT(lag <= data_len);
40
41         while(lag--) {
42                 for(i = lag, d = 0.0; i < data_len; i++)
43                         d += data[i] * data[i - lag];
44                 autoc[lag] = d;
45         }
46 #endif
47
48         /*
49          * this version tends to run faster because of better data locality
50          * ('data_len' is usually much larger than 'lag')
51          */
52         FLAC__real d;
53         unsigned sample, coeff;
54         const unsigned limit = data_len - lag;
55
56         FLAC__ASSERT(lag > 0);
57         FLAC__ASSERT(lag <= data_len);
58
59         for(coeff = 0; coeff < lag; coeff++)
60                 autoc[coeff] = 0.0;
61         for(sample = 0; sample <= limit; sample++) {
62                 d = data[sample];
63                 for(coeff = 0; coeff < lag; coeff++)
64                         autoc[coeff] += d * data[sample+coeff];
65         }
66         for(; sample < data_len; sample++) {
67                 d = data[sample];
68                 for(coeff = 0; coeff < data_len - sample; coeff++)
69                         autoc[coeff] += d * data[sample+coeff];
70         }
71 }
72
73 void FLAC__lpc_compute_lp_coefficients(const FLAC__real autoc[], unsigned max_order, FLAC__real lp_coeff[][FLAC__MAX_LPC_ORDER], FLAC__real error[])
74 {
75         unsigned i, j;
76         double r, err, ref[FLAC__MAX_LPC_ORDER], lpc[FLAC__MAX_LPC_ORDER];
77
78         FLAC__ASSERT(0 < max_order);
79         FLAC__ASSERT(max_order <= FLAC__MAX_LPC_ORDER);
80         FLAC__ASSERT(autoc[0] != 0.0);
81
82         err = autoc[0];
83
84         for(i = 0; i < max_order; i++) {
85                 /* Sum up this iteration's reflection coefficient. */
86                 r = -autoc[i+1];
87                 for(j = 0; j < i; j++)
88                         r -= lpc[j] * autoc[i-j];
89                 ref[i] = (r/=err);
90
91                 /* Update LPC coefficients and total error. */
92                 lpc[i]=r;
93                 for(j = 0; j < (i>>1); j++) {
94                         double tmp = lpc[j];
95                         lpc[j] += r * lpc[i-1-j];
96                         lpc[i-1-j] += r * tmp;
97                 }
98                 if(i & 1)
99                         lpc[j] += lpc[j] * r;
100
101                 err *= (1.0 - r * r);
102
103                 /* save this order */
104                 for(j = 0; j <= i; j++)
105                         lp_coeff[i][j] = (FLAC__real)(-lpc[j]); /* negate FIR filter coeff to get predictor coeff */
106                 error[i] = (FLAC__real)err;
107         }
108 }
109
110 int FLAC__lpc_quantize_coefficients(const FLAC__real lp_coeff[], unsigned order, unsigned precision, unsigned bits_per_sample, FLAC__int32 qlp_coeff[], int *shift)
111 {
112         unsigned i;
113         double d, cmax = -1e32;
114         FLAC__int32 qmax, qmin;
115         const int max_shiftlimit = (1 << (FLAC__SUBFRAME_LPC_QLP_SHIFT_LEN-1)) - 1;
116         const int min_shiftlimit = -max_shiftlimit - 1;
117
118         FLAC__ASSERT(bits_per_sample > 0);
119         FLAC__ASSERT(bits_per_sample <= sizeof(FLAC__int32)*8);
120         FLAC__ASSERT(precision > 0);
121         FLAC__ASSERT(precision >= FLAC__MIN_QLP_COEFF_PRECISION);
122         FLAC__ASSERT(precision + bits_per_sample < sizeof(FLAC__int32)*8);
123 #ifdef NDEBUG
124         (void)bits_per_sample; /* silence compiler warning about unused parameter */
125 #endif
126
127         /* drop one bit for the sign; from here on out we consider only |lp_coeff[i]| */
128         precision--;
129         qmax = 1 << precision;
130         qmin = -qmax;
131         qmax--;
132
133         for(i = 0; i < order; i++) {
134                 if(lp_coeff[i] == 0.0)
135                         continue;
136                 d = fabs(lp_coeff[i]);
137                 if(d > cmax)
138                         cmax = d;
139         }
140 redo_it:
141         if(cmax <= 0.0) {
142                 /* => coefficients are all 0, which means our constant-detect didn't work */
143                 return 2;
144         }
145         else {
146                 int log2cmax;
147
148                 (void)frexp(cmax, &log2cmax);   
149                 log2cmax--;
150                 *shift = (int)precision - log2cmax - 1;
151
152                 if(*shift < min_shiftlimit || *shift > max_shiftlimit) {
153                         return 1;
154                 }
155         }
156
157         if(*shift >= 0) {
158                 for(i = 0; i < order; i++) {
159                         qlp_coeff[i] = (FLAC__int32)floor((double)lp_coeff[i] * (double)(1 << *shift));
160
161                         /* double-check the result */
162                         if(qlp_coeff[i] > qmax || qlp_coeff[i] < qmin) {
163 #ifdef FLAC__OVERFLOW_DETECT
164                                 fprintf(stderr,"FLAC__lpc_quantize_coefficients: compensating for overflow, qlp_coeff[%u]=%d, lp_coeff[%u]=%f, cmax=%f, precision=%u, shift=%d, q=%f, f(q)=%f\n", i, qlp_coeff[i], i, lp_coeff[i], cmax, precision, *shift, (double)lp_coeff[i] * (double)(1 << *shift), floor((double)lp_coeff[i] * (double)(1 << *shift)));
165 #endif
166                                 cmax *= 2.0;
167                                 goto redo_it;
168                         }
169                 }
170         }
171         else { /* (*shift < 0) */
172                 const int nshift = -(*shift);
173 #ifdef DEBUG
174                 fprintf(stderr,"FLAC__lpc_quantize_coefficients: negative shift = %d\n", *shift);
175 #endif
176                 for(i = 0; i < order; i++) {
177                         qlp_coeff[i] = (FLAC__int32)floor((double)lp_coeff[i] / (double)(1 << nshift));
178
179                         /* double-check the result */
180                         if(qlp_coeff[i] > qmax || qlp_coeff[i] < qmin) {
181 #ifdef FLAC__OVERFLOW_DETECT
182                                 fprintf(stderr,"FLAC__lpc_quantize_coefficients: compensating for overflow, qlp_coeff[%u]=%d, lp_coeff[%u]=%f, cmax=%f, precision=%u, shift=%d, q=%f, f(q)=%f\n", i, qlp_coeff[i], i, lp_coeff[i], cmax, precision, *shift, (double)lp_coeff[i] / (double)(1 << nshift), floor((double)lp_coeff[i] / (double)(1 << nshift)));
183 #endif
184                                 cmax *= 2.0;
185                                 goto redo_it;
186                         }
187                 }
188         }
189
190         return 0;
191 }
192
193 void FLAC__lpc_compute_residual_from_qlp_coefficients(const FLAC__int32 data[], unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 residual[])
194 {
195 #ifdef FLAC__OVERFLOW_DETECT
196         FLAC__int64 sumo;
197 #endif
198         unsigned i, j;
199         FLAC__int32 sum;
200         const FLAC__int32 *history;
201
202 #ifdef FLAC__OVERFLOW_DETECT_VERBOSE
203         fprintf(stderr,"FLAC__lpc_compute_residual_from_qlp_coefficients: data_len=%d, order=%u, lpq=%d",data_len,order,lp_quantization);
204         for(i=0;i<order;i++)
205                 fprintf(stderr,", q[%u]=%d",i,qlp_coeff[i]);
206         fprintf(stderr,"\n");
207 #endif
208         FLAC__ASSERT(order > 0);
209
210         for(i = 0; i < data_len; i++) {
211 #ifdef FLAC__OVERFLOW_DETECT
212                 sumo = 0;
213 #endif
214                 sum = 0;
215                 history = data;
216                 for(j = 0; j < order; j++) {
217                         sum += qlp_coeff[j] * (*(--history));
218 #ifdef FLAC__OVERFLOW_DETECT
219                         sumo += (FLAC__int64)qlp_coeff[j] * (FLAC__int64)(*history);
220 #if defined _MSC_VER || defined __MINGW32__ /* don't know how to do 64-bit literals in VC++ */
221                         if(sumo < 0) sumo = -sumo;
222                         if(sumo > 2147483647)
223 #else
224                         if(sumo > 2147483647ll || sumo < -2147483648ll)
225 #endif
226                         {
227                                 fprintf(stderr,"FLAC__lpc_compute_residual_from_qlp_coefficients: OVERFLOW, i=%u, j=%u, c=%d, d=%d, sumo=%lld\n",i,j,qlp_coeff[j],*history,sumo);
228                         }
229 #endif
230                 }
231                 *(residual++) = *(data++) - (sum >> lp_quantization);
232         }
233
234         /* Here's a slower but clearer version:
235         for(i = 0; i < data_len; i++) {
236                 sum = 0;
237                 for(j = 0; j < order; j++)
238                         sum += qlp_coeff[j] * data[i-j-1];
239                 residual[i] = data[i] - (sum >> lp_quantization);
240         }
241         */
242 }
243
244 void FLAC__lpc_restore_signal(const FLAC__int32 residual[], unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 data[])
245 {
246 #ifdef FLAC__OVERFLOW_DETECT
247         FLAC__int64 sumo;
248 #endif
249         unsigned i, j;
250         FLAC__int32 sum;
251         const FLAC__int32 *history;
252
253 #ifdef FLAC__OVERFLOW_DETECT_VERBOSE
254         fprintf(stderr,"FLAC__lpc_restore_signal: data_len=%d, order=%u, lpq=%d",data_len,order,lp_quantization);
255         for(i=0;i<order;i++)
256                 fprintf(stderr,", q[%u]=%d",i,qlp_coeff[i]);
257         fprintf(stderr,"\n");
258 #endif
259         FLAC__ASSERT(order > 0);
260
261         for(i = 0; i < data_len; i++) {
262 #ifdef FLAC__OVERFLOW_DETECT
263                 sumo = 0;
264 #endif
265                 sum = 0;
266                 history = data;
267                 for(j = 0; j < order; j++) {
268                         sum += qlp_coeff[j] * (*(--history));
269 #ifdef FLAC__OVERFLOW_DETECT
270                         sumo += (FLAC__int64)qlp_coeff[j] * (FLAC__int64)(*history);
271 #if defined _MSC_VER || defined __MINGW32__ /* don't know how to do 64-bit literals in VC++ */
272                         if(sumo < 0) sumo = -sumo;
273                         if(sumo > 2147483647)
274 #else
275                         if(sumo > 2147483647ll || sumo < -2147483648ll)
276 #endif
277                         {
278                                 fprintf(stderr,"FLAC__lpc_restore_signal: OVERFLOW, i=%u, j=%u, c=%d, d=%d, sumo=%lld\n",i,j,qlp_coeff[j],*history,sumo);
279                         }
280 #endif
281                 }
282                 *(data++) = *(residual++) + (sum >> lp_quantization);
283         }
284
285         /* Here's a slower but clearer version:
286         for(i = 0; i < data_len; i++) {
287                 sum = 0;
288                 for(j = 0; j < order; j++)
289                         sum += qlp_coeff[j] * data[i-j-1];
290                 data[i] = residual[i] + (sum >> lp_quantization);
291         }
292         */
293 }
294
295 FLAC__real FLAC__lpc_compute_expected_bits_per_residual_sample(FLAC__real lpc_error, unsigned total_samples)
296 {
297         double error_scale;
298
299         FLAC__ASSERT(total_samples > 0);
300
301         error_scale = 0.5 * M_LN2 * M_LN2 / (FLAC__real)total_samples;
302
303         if(lpc_error > 0.0) {
304                 FLAC__real bps = (FLAC__real)((double)0.5 * log(error_scale * lpc_error) / M_LN2);
305                 if(bps >= 0.0)
306                         return bps;
307                 else
308                         return 0.0;
309         }
310         else if(lpc_error < 0.0) { /* error should not be negative but can happen due to inadequate float resolution */
311                 return (FLAC__real)1e32;
312         }
313         else {
314                 return 0.0;
315         }
316 }
317
318 FLAC__real FLAC__lpc_compute_expected_bits_per_residual_sample_with_error_scale(FLAC__real lpc_error, double error_scale)
319 {
320         if(lpc_error > 0.0) {
321                 FLAC__real bps = (FLAC__real)((double)0.5 * log(error_scale * lpc_error) / M_LN2);
322                 if(bps >= 0.0)
323                         return bps;
324                 else
325                         return 0.0;
326         }
327         else if(lpc_error < 0.0) { /* error should not be negative but can happen due to inadequate float resolution */
328                 return (FLAC__real)1e32;
329         }
330         else {
331                 return 0.0;
332         }
333 }
334
335 unsigned FLAC__lpc_compute_best_order(const FLAC__real lpc_error[], unsigned max_order, unsigned total_samples, unsigned bits_per_signal_sample)
336 {
337         unsigned order, best_order;
338         FLAC__real best_bits, tmp_bits;
339         double error_scale;
340
341         FLAC__ASSERT(max_order > 0);
342         FLAC__ASSERT(total_samples > 0);
343
344         error_scale = 0.5 * M_LN2 * M_LN2 / (FLAC__real)total_samples;
345
346         best_order = 0;
347         best_bits = FLAC__lpc_compute_expected_bits_per_residual_sample_with_error_scale(lpc_error[0], error_scale) * (FLAC__real)total_samples;
348
349         for(order = 1; order < max_order; order++) {
350                 tmp_bits = FLAC__lpc_compute_expected_bits_per_residual_sample_with_error_scale(lpc_error[order], error_scale) * (FLAC__real)(total_samples - order) + (FLAC__real)(order * bits_per_signal_sample);
351                 if(tmp_bits < best_bits) {
352                         best_order = order;
353                         best_bits = tmp_bits;
354                 }
355         }
356
357         return best_order+1; /* +1 since index of lpc_error[] is order-1 */
358 }