• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* libFLAC - Free Lossless Audio Codec library
2  * Copyright (C) 2000,2001,2002,2003,2004,2005,2006,2007  Josh Coalson
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  *
8  * - Redistributions of source code must retain the above copyright
9  * notice, this list of conditions and the following disclaimer.
10  *
11  * - Redistributions in binary form must reproduce the above copyright
12  * notice, this list of conditions and the following disclaimer in the
13  * documentation and/or other materials provided with the distribution.
14  *
15  * - Neither the name of the Xiph.org Foundation nor the names of its
16  * contributors may be used to endorse or promote products derived from
17  * this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
22  * A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
23  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
24  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
25  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
26  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
27  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
28  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
29  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #if HAVE_CONFIG_H
33 #  include <config.h>
34 #endif
35 
36 #include <math.h>
37 #include <string.h>
38 #include "private/bitmath.h"
39 #include "private/fixed.h"
40 #include "FLAC/assert.h"
41 
42 #ifndef M_LN2
43 /* math.h in VC++ doesn't seem to have this (how Microsoft is that?) */
44 #define M_LN2 0.69314718055994530942
45 #endif
46 
47 #ifdef min
48 #undef min
49 #endif
50 #define min(x,y) ((x) < (y)? (x) : (y))
51 
52 #ifdef local_abs
53 #undef local_abs
54 #endif
55 #define local_abs(x) ((unsigned)((x)<0? -(x) : (x)))
56 
57 #ifdef FLAC__INTEGER_ONLY_LIBRARY
58 /* rbps stands for residual bits per sample
59  *
60  *             (ln(2) * err)
61  * rbps = log  (-----------)
62  *           2 (     n     )
63  */
local__compute_rbps_integerized(FLAC__uint32 err,FLAC__uint32 n)64 static FLAC__fixedpoint local__compute_rbps_integerized(FLAC__uint32 err, FLAC__uint32 n)
65 {
66 	FLAC__uint32 rbps;
67 	unsigned bits; /* the number of bits required to represent a number */
68 	int fracbits; /* the number of bits of rbps that comprise the fractional part */
69 
70 	FLAC__ASSERT(sizeof(rbps) == sizeof(FLAC__fixedpoint));
71 	FLAC__ASSERT(err > 0);
72 	FLAC__ASSERT(n > 0);
73 
74 	FLAC__ASSERT(n <= FLAC__MAX_BLOCK_SIZE);
75 	if(err <= n)
76 		return 0;
77 	/*
78 	 * The above two things tell us 1) n fits in 16 bits; 2) err/n > 1.
79 	 * These allow us later to know we won't lose too much precision in the
80 	 * fixed-point division (err<<fracbits)/n.
81 	 */
82 
83 	fracbits = (8*sizeof(err)) - (FLAC__bitmath_ilog2(err)+1);
84 
85 	err <<= fracbits;
86 	err /= n;
87 	/* err now holds err/n with fracbits fractional bits */
88 
89 	/*
90 	 * Whittle err down to 16 bits max.  16 significant bits is enough for
91 	 * our purposes.
92 	 */
93 	FLAC__ASSERT(err > 0);
94 	bits = FLAC__bitmath_ilog2(err)+1;
95 	if(bits > 16) {
96 		err >>= (bits-16);
97 		fracbits -= (bits-16);
98 	}
99 	rbps = (FLAC__uint32)err;
100 
101 	/* Multiply by fixed-point version of ln(2), with 16 fractional bits */
102 	rbps *= FLAC__FP_LN2;
103 	fracbits += 16;
104 	FLAC__ASSERT(fracbits >= 0);
105 
106 	/* FLAC__fixedpoint_log2 requires fracbits%4 to be 0 */
107 	{
108 		const int f = fracbits & 3;
109 		if(f) {
110 			rbps >>= f;
111 			fracbits -= f;
112 		}
113 	}
114 
115 	rbps = FLAC__fixedpoint_log2(rbps, fracbits, (unsigned)(-1));
116 
117 	if(rbps == 0)
118 		return 0;
119 
120 	/*
121 	 * The return value must have 16 fractional bits.  Since the whole part
122 	 * of the base-2 log of a 32 bit number must fit in 5 bits, and fracbits
123 	 * must be >= -3, these assertion allows us to be able to shift rbps
124 	 * left if necessary to get 16 fracbits without losing any bits of the
125 	 * whole part of rbps.
126 	 *
127 	 * There is a slight chance due to accumulated error that the whole part
128 	 * will require 6 bits, so we use 6 in the assertion.  Really though as
129 	 * long as it fits in 13 bits (32 - (16 - (-3))) we are fine.
130 	 */
131 	FLAC__ASSERT((int)FLAC__bitmath_ilog2(rbps)+1 <= fracbits + 6);
132 	FLAC__ASSERT(fracbits >= -3);
133 
134 	/* now shift the decimal point into place */
135 	if(fracbits < 16)
136 		return rbps << (16-fracbits);
137 	else if(fracbits > 16)
138 		return rbps >> (fracbits-16);
139 	else
140 		return rbps;
141 }
142 
local__compute_rbps_wide_integerized(FLAC__uint64 err,FLAC__uint32 n)143 static FLAC__fixedpoint local__compute_rbps_wide_integerized(FLAC__uint64 err, FLAC__uint32 n)
144 {
145 	FLAC__uint32 rbps;
146 	unsigned bits; /* the number of bits required to represent a number */
147 	int fracbits; /* the number of bits of rbps that comprise the fractional part */
148 
149 	FLAC__ASSERT(sizeof(rbps) == sizeof(FLAC__fixedpoint));
150 	FLAC__ASSERT(err > 0);
151 	FLAC__ASSERT(n > 0);
152 
153 	FLAC__ASSERT(n <= FLAC__MAX_BLOCK_SIZE);
154 	if(err <= n)
155 		return 0;
156 	/*
157 	 * The above two things tell us 1) n fits in 16 bits; 2) err/n > 1.
158 	 * These allow us later to know we won't lose too much precision in the
159 	 * fixed-point division (err<<fracbits)/n.
160 	 */
161 
162 	fracbits = (8*sizeof(err)) - (FLAC__bitmath_ilog2_wide(err)+1);
163 
164 	err <<= fracbits;
165 	err /= n;
166 	/* err now holds err/n with fracbits fractional bits */
167 
168 	/*
169 	 * Whittle err down to 16 bits max.  16 significant bits is enough for
170 	 * our purposes.
171 	 */
172 	FLAC__ASSERT(err > 0);
173 	bits = FLAC__bitmath_ilog2_wide(err)+1;
174 	if(bits > 16) {
175 		err >>= (bits-16);
176 		fracbits -= (bits-16);
177 	}
178 	rbps = (FLAC__uint32)err;
179 
180 	/* Multiply by fixed-point version of ln(2), with 16 fractional bits */
181 	rbps *= FLAC__FP_LN2;
182 	fracbits += 16;
183 	FLAC__ASSERT(fracbits >= 0);
184 
185 	/* FLAC__fixedpoint_log2 requires fracbits%4 to be 0 */
186 	{
187 		const int f = fracbits & 3;
188 		if(f) {
189 			rbps >>= f;
190 			fracbits -= f;
191 		}
192 	}
193 
194 	rbps = FLAC__fixedpoint_log2(rbps, fracbits, (unsigned)(-1));
195 
196 	if(rbps == 0)
197 		return 0;
198 
199 	/*
200 	 * The return value must have 16 fractional bits.  Since the whole part
201 	 * of the base-2 log of a 32 bit number must fit in 5 bits, and fracbits
202 	 * must be >= -3, these assertion allows us to be able to shift rbps
203 	 * left if necessary to get 16 fracbits without losing any bits of the
204 	 * whole part of rbps.
205 	 *
206 	 * There is a slight chance due to accumulated error that the whole part
207 	 * will require 6 bits, so we use 6 in the assertion.  Really though as
208 	 * long as it fits in 13 bits (32 - (16 - (-3))) we are fine.
209 	 */
210 	FLAC__ASSERT((int)FLAC__bitmath_ilog2(rbps)+1 <= fracbits + 6);
211 	FLAC__ASSERT(fracbits >= -3);
212 
213 	/* now shift the decimal point into place */
214 	if(fracbits < 16)
215 		return rbps << (16-fracbits);
216 	else if(fracbits > 16)
217 		return rbps >> (fracbits-16);
218 	else
219 		return rbps;
220 }
221 #endif
222 
223 #ifndef FLAC__INTEGER_ONLY_LIBRARY
FLAC__fixed_compute_best_predictor(const FLAC__int32 data[],unsigned data_len,FLAC__float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER+1])224 unsigned FLAC__fixed_compute_best_predictor(const FLAC__int32 data[], unsigned data_len, FLAC__float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER+1])
225 #else
226 unsigned FLAC__fixed_compute_best_predictor(const FLAC__int32 data[], unsigned data_len, FLAC__fixedpoint residual_bits_per_sample[FLAC__MAX_FIXED_ORDER+1])
227 #endif
228 {
229 	FLAC__int32 last_error_0 = data[-1];
230 	FLAC__int32 last_error_1 = data[-1] - data[-2];
231 	FLAC__int32 last_error_2 = last_error_1 - (data[-2] - data[-3]);
232 	FLAC__int32 last_error_3 = last_error_2 - (data[-2] - 2*data[-3] + data[-4]);
233 	FLAC__int32 error, save;
234 	FLAC__uint32 total_error_0 = 0, total_error_1 = 0, total_error_2 = 0, total_error_3 = 0, total_error_4 = 0;
235 	unsigned i, order;
236 
237 	for(i = 0; i < data_len; i++) {
238 		error  = data[i]     ; total_error_0 += local_abs(error);                      save = error;
239 		error -= last_error_0; total_error_1 += local_abs(error); last_error_0 = save; save = error;
240 		error -= last_error_1; total_error_2 += local_abs(error); last_error_1 = save; save = error;
241 		error -= last_error_2; total_error_3 += local_abs(error); last_error_2 = save; save = error;
242 		error -= last_error_3; total_error_4 += local_abs(error); last_error_3 = save;
243 	}
244 
245 	if(total_error_0 < min(min(min(total_error_1, total_error_2), total_error_3), total_error_4))
246 		order = 0;
247 	else if(total_error_1 < min(min(total_error_2, total_error_3), total_error_4))
248 		order = 1;
249 	else if(total_error_2 < min(total_error_3, total_error_4))
250 		order = 2;
251 	else if(total_error_3 < total_error_4)
252 		order = 3;
253 	else
254 		order = 4;
255 
256 	/* Estimate the expected number of bits per residual signal sample. */
257 	/* 'total_error*' is linearly related to the variance of the residual */
258 	/* signal, so we use it directly to compute E(|x|) */
259 	FLAC__ASSERT(data_len > 0 || total_error_0 == 0);
260 	FLAC__ASSERT(data_len > 0 || total_error_1 == 0);
261 	FLAC__ASSERT(data_len > 0 || total_error_2 == 0);
262 	FLAC__ASSERT(data_len > 0 || total_error_3 == 0);
263 	FLAC__ASSERT(data_len > 0 || total_error_4 == 0);
264 #ifndef FLAC__INTEGER_ONLY_LIBRARY
265 	residual_bits_per_sample[0] = (FLAC__float)((total_error_0 > 0) ? log(M_LN2 * (FLAC__double)total_error_0 / (FLAC__double)data_len) / M_LN2 : 0.0);
266 	residual_bits_per_sample[1] = (FLAC__float)((total_error_1 > 0) ? log(M_LN2 * (FLAC__double)total_error_1 / (FLAC__double)data_len) / M_LN2 : 0.0);
267 	residual_bits_per_sample[2] = (FLAC__float)((total_error_2 > 0) ? log(M_LN2 * (FLAC__double)total_error_2 / (FLAC__double)data_len) / M_LN2 : 0.0);
268 	residual_bits_per_sample[3] = (FLAC__float)((total_error_3 > 0) ? log(M_LN2 * (FLAC__double)total_error_3 / (FLAC__double)data_len) / M_LN2 : 0.0);
269 	residual_bits_per_sample[4] = (FLAC__float)((total_error_4 > 0) ? log(M_LN2 * (FLAC__double)total_error_4 / (FLAC__double)data_len) / M_LN2 : 0.0);
270 #else
271 	residual_bits_per_sample[0] = (total_error_0 > 0) ? local__compute_rbps_integerized(total_error_0, data_len) : 0;
272 	residual_bits_per_sample[1] = (total_error_1 > 0) ? local__compute_rbps_integerized(total_error_1, data_len) : 0;
273 	residual_bits_per_sample[2] = (total_error_2 > 0) ? local__compute_rbps_integerized(total_error_2, data_len) : 0;
274 	residual_bits_per_sample[3] = (total_error_3 > 0) ? local__compute_rbps_integerized(total_error_3, data_len) : 0;
275 	residual_bits_per_sample[4] = (total_error_4 > 0) ? local__compute_rbps_integerized(total_error_4, data_len) : 0;
276 #endif
277 
278 	return order;
279 }
280 
281 #ifndef FLAC__INTEGER_ONLY_LIBRARY
FLAC__fixed_compute_best_predictor_wide(const FLAC__int32 data[],unsigned data_len,FLAC__float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER+1])282 unsigned FLAC__fixed_compute_best_predictor_wide(const FLAC__int32 data[], unsigned data_len, FLAC__float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER+1])
283 #else
284 unsigned FLAC__fixed_compute_best_predictor_wide(const FLAC__int32 data[], unsigned data_len, FLAC__fixedpoint residual_bits_per_sample[FLAC__MAX_FIXED_ORDER+1])
285 #endif
286 {
287 	FLAC__int32 last_error_0 = data[-1];
288 	FLAC__int32 last_error_1 = data[-1] - data[-2];
289 	FLAC__int32 last_error_2 = last_error_1 - (data[-2] - data[-3]);
290 	FLAC__int32 last_error_3 = last_error_2 - (data[-2] - 2*data[-3] + data[-4]);
291 	FLAC__int32 error, save;
292 	/* total_error_* are 64-bits to avoid overflow when encoding
293 	 * erratic signals when the bits-per-sample and blocksize are
294 	 * large.
295 	 */
296 	FLAC__uint64 total_error_0 = 0, total_error_1 = 0, total_error_2 = 0, total_error_3 = 0, total_error_4 = 0;
297 	unsigned i, order;
298 
299 	for(i = 0; i < data_len; i++) {
300 		error  = data[i]     ; total_error_0 += local_abs(error);                      save = error;
301 		error -= last_error_0; total_error_1 += local_abs(error); last_error_0 = save; save = error;
302 		error -= last_error_1; total_error_2 += local_abs(error); last_error_1 = save; save = error;
303 		error -= last_error_2; total_error_3 += local_abs(error); last_error_2 = save; save = error;
304 		error -= last_error_3; total_error_4 += local_abs(error); last_error_3 = save;
305 	}
306 
307 	if(total_error_0 < min(min(min(total_error_1, total_error_2), total_error_3), total_error_4))
308 		order = 0;
309 	else if(total_error_1 < min(min(total_error_2, total_error_3), total_error_4))
310 		order = 1;
311 	else if(total_error_2 < min(total_error_3, total_error_4))
312 		order = 2;
313 	else if(total_error_3 < total_error_4)
314 		order = 3;
315 	else
316 		order = 4;
317 
318 	/* Estimate the expected number of bits per residual signal sample. */
319 	/* 'total_error*' is linearly related to the variance of the residual */
320 	/* signal, so we use it directly to compute E(|x|) */
321 	FLAC__ASSERT(data_len > 0 || total_error_0 == 0);
322 	FLAC__ASSERT(data_len > 0 || total_error_1 == 0);
323 	FLAC__ASSERT(data_len > 0 || total_error_2 == 0);
324 	FLAC__ASSERT(data_len > 0 || total_error_3 == 0);
325 	FLAC__ASSERT(data_len > 0 || total_error_4 == 0);
326 #ifndef FLAC__INTEGER_ONLY_LIBRARY
327 #if defined _MSC_VER || defined __MINGW32__
328 	/* with MSVC you have to spoon feed it the casting */
329 	residual_bits_per_sample[0] = (FLAC__float)((total_error_0 > 0) ? log(M_LN2 * (FLAC__double)(FLAC__int64)total_error_0 / (FLAC__double)data_len) / M_LN2 : 0.0);
330 	residual_bits_per_sample[1] = (FLAC__float)((total_error_1 > 0) ? log(M_LN2 * (FLAC__double)(FLAC__int64)total_error_1 / (FLAC__double)data_len) / M_LN2 : 0.0);
331 	residual_bits_per_sample[2] = (FLAC__float)((total_error_2 > 0) ? log(M_LN2 * (FLAC__double)(FLAC__int64)total_error_2 / (FLAC__double)data_len) / M_LN2 : 0.0);
332 	residual_bits_per_sample[3] = (FLAC__float)((total_error_3 > 0) ? log(M_LN2 * (FLAC__double)(FLAC__int64)total_error_3 / (FLAC__double)data_len) / M_LN2 : 0.0);
333 	residual_bits_per_sample[4] = (FLAC__float)((total_error_4 > 0) ? log(M_LN2 * (FLAC__double)(FLAC__int64)total_error_4 / (FLAC__double)data_len) / M_LN2 : 0.0);
334 #else
335 	residual_bits_per_sample[0] = (FLAC__float)((total_error_0 > 0) ? log(M_LN2 * (FLAC__double)total_error_0 / (FLAC__double)data_len) / M_LN2 : 0.0);
336 	residual_bits_per_sample[1] = (FLAC__float)((total_error_1 > 0) ? log(M_LN2 * (FLAC__double)total_error_1 / (FLAC__double)data_len) / M_LN2 : 0.0);
337 	residual_bits_per_sample[2] = (FLAC__float)((total_error_2 > 0) ? log(M_LN2 * (FLAC__double)total_error_2 / (FLAC__double)data_len) / M_LN2 : 0.0);
338 	residual_bits_per_sample[3] = (FLAC__float)((total_error_3 > 0) ? log(M_LN2 * (FLAC__double)total_error_3 / (FLAC__double)data_len) / M_LN2 : 0.0);
339 	residual_bits_per_sample[4] = (FLAC__float)((total_error_4 > 0) ? log(M_LN2 * (FLAC__double)total_error_4 / (FLAC__double)data_len) / M_LN2 : 0.0);
340 #endif
341 #else
342 	residual_bits_per_sample[0] = (total_error_0 > 0) ? local__compute_rbps_wide_integerized(total_error_0, data_len) : 0;
343 	residual_bits_per_sample[1] = (total_error_1 > 0) ? local__compute_rbps_wide_integerized(total_error_1, data_len) : 0;
344 	residual_bits_per_sample[2] = (total_error_2 > 0) ? local__compute_rbps_wide_integerized(total_error_2, data_len) : 0;
345 	residual_bits_per_sample[3] = (total_error_3 > 0) ? local__compute_rbps_wide_integerized(total_error_3, data_len) : 0;
346 	residual_bits_per_sample[4] = (total_error_4 > 0) ? local__compute_rbps_wide_integerized(total_error_4, data_len) : 0;
347 #endif
348 
349 	return order;
350 }
351 
FLAC__fixed_compute_residual(const FLAC__int32 data[],unsigned data_len,unsigned order,FLAC__int32 residual[])352 void FLAC__fixed_compute_residual(const FLAC__int32 data[], unsigned data_len, unsigned order, FLAC__int32 residual[])
353 {
354 	const int idata_len = (int)data_len;
355 	int i;
356 
357 	switch(order) {
358 		case 0:
359 			FLAC__ASSERT(sizeof(residual[0]) == sizeof(data[0]));
360 			memcpy(residual, data, sizeof(residual[0])*data_len);
361 			break;
362 		case 1:
363 			for(i = 0; i < idata_len; i++)
364 				residual[i] = data[i] - data[i-1];
365 			break;
366 		case 2:
367 			for(i = 0; i < idata_len; i++)
368 #if 1 /* OPT: may be faster with some compilers on some systems */
369 				residual[i] = data[i] - (data[i-1] << 1) + data[i-2];
370 #else
371 				residual[i] = data[i] - 2*data[i-1] + data[i-2];
372 #endif
373 			break;
374 		case 3:
375 			for(i = 0; i < idata_len; i++)
376 #if 1 /* OPT: may be faster with some compilers on some systems */
377 				residual[i] = data[i] - (((data[i-1]-data[i-2])<<1) + (data[i-1]-data[i-2])) - data[i-3];
378 #else
379 				residual[i] = data[i] - 3*data[i-1] + 3*data[i-2] - data[i-3];
380 #endif
381 			break;
382 		case 4:
383 			for(i = 0; i < idata_len; i++)
384 #if 1 /* OPT: may be faster with some compilers on some systems */
385 				residual[i] = data[i] - ((data[i-1]+data[i-3])<<2) + ((data[i-2]<<2) + (data[i-2]<<1)) + data[i-4];
386 #else
387 				residual[i] = data[i] - 4*data[i-1] + 6*data[i-2] - 4*data[i-3] + data[i-4];
388 #endif
389 			break;
390 		default:
391 			FLAC__ASSERT(0);
392 	}
393 }
394 
FLAC__fixed_restore_signal(const FLAC__int32 residual[],unsigned data_len,unsigned order,FLAC__int32 data[])395 void FLAC__fixed_restore_signal(const FLAC__int32 residual[], unsigned data_len, unsigned order, FLAC__int32 data[])
396 {
397 	int i, idata_len = (int)data_len;
398 
399 	switch(order) {
400 		case 0:
401 			FLAC__ASSERT(sizeof(residual[0]) == sizeof(data[0]));
402 			memcpy(data, residual, sizeof(residual[0])*data_len);
403 			break;
404 		case 1:
405 			for(i = 0; i < idata_len; i++)
406 				data[i] = residual[i] + data[i-1];
407 			break;
408 		case 2:
409 			for(i = 0; i < idata_len; i++)
410 #if 1 /* OPT: may be faster with some compilers on some systems */
411 				data[i] = residual[i] + (data[i-1]<<1) - data[i-2];
412 #else
413 				data[i] = residual[i] + 2*data[i-1] - data[i-2];
414 #endif
415 			break;
416 		case 3:
417 			for(i = 0; i < idata_len; i++)
418 #if 1 /* OPT: may be faster with some compilers on some systems */
419 				data[i] = residual[i] + (((data[i-1]-data[i-2])<<1) + (data[i-1]-data[i-2])) + data[i-3];
420 #else
421 				data[i] = residual[i] + 3*data[i-1] - 3*data[i-2] + data[i-3];
422 #endif
423 			break;
424 		case 4:
425 			for(i = 0; i < idata_len; i++)
426 #if 1 /* OPT: may be faster with some compilers on some systems */
427 				data[i] = residual[i] + ((data[i-1]+data[i-3])<<2) - ((data[i-2]<<2) + (data[i-2]<<1)) - data[i-4];
428 #else
429 				data[i] = residual[i] + 4*data[i-1] - 6*data[i-2] + 4*data[i-3] - data[i-4];
430 #endif
431 			break;
432 		default:
433 			FLAC__ASSERT(0);
434 	}
435 }
436