1 /*
2 * libmad - MPEG audio decoder library
3 * Copyright (C) 2000-2004 Underbit Technologies, Inc.
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 *
19 * $Id: synth.c,v 1.25 2004/01/23 09:41:33 rob Exp $
20 */
21
22 # include "global.h"
23
24 # include "fixed.h"
25 # include "frame.h"
26 # include "synth.h"
27
28 /*
29 * NAME: synth->init()
30 * DESCRIPTION: initialize synth struct
31 */
mad_synth_init(struct mad_synth * synth)32 void mad_synth_init(struct mad_synth *synth)
33 {
34 mad_synth_mute(synth);
35
36 synth->phase = 0;
37
38 synth->pcm.samplerate = 0;
39 synth->pcm.channels = 0;
40 synth->pcm.length = 0;
41 }
42
43 /*
44 * NAME: synth->mute()
45 * DESCRIPTION: zero all polyphase filterbank values, resetting synthesis
46 */
mad_synth_mute(struct mad_synth * synth)47 void mad_synth_mute(struct mad_synth *synth)
48 {
49 unsigned int ch, s, v;
50
51 for (ch = 0; ch < 2; ++ch) {
52 for (s = 0; s < 16; ++s) {
53 for (v = 0; v < 8; ++v) {
54 synth->filter[ch][0][0][s][v] = synth->filter[ch][0][1][s][v] =
55 synth->filter[ch][1][0][s][v] = synth->filter[ch][1][1][s][v] = 0;
56 }
57 }
58 }
59 }
60
61 /*
62 * An optional optimization called here the Subband Synthesis Optimization
63 * (SSO) improves the performance of subband synthesis at the expense of
64 * accuracy.
65 *
66 * The idea is to simplify 32x32->64-bit multiplication to 32x32->32 such
67 * that extra scaling and rounding are not necessary. This often allows the
68 * compiler to use faster 32-bit multiply-accumulate instructions instead of
69 * explicit 64-bit multiply, shift, and add instructions.
70 *
71 * SSO works like this: a full 32x32->64-bit multiply of two mad_fixed_t
72 * values requires the result to be right-shifted 28 bits to be properly
73 * scaled to the same fixed-point format. Right shifts can be applied at any
74 * time to either operand or to the result, so the optimization involves
75 * careful placement of these shifts to minimize the loss of accuracy.
76 *
77 * First, a 14-bit shift is applied with rounding at compile-time to the D[]
78 * table of coefficients for the subband synthesis window. This only loses 2
79 * bits of accuracy because the lower 12 bits are always zero. A second
80 * 12-bit shift occurs after the DCT calculation. This loses 12 bits of
81 * accuracy. Finally, a third 2-bit shift occurs just before the sample is
82 * saved in the PCM buffer. 14 + 12 + 2 == 28 bits.
83 */
84
85 /* FPM_DEFAULT without OPT_SSO will actually lose accuracy and performance */
86
87 /* second SSO shift, with rounding */
88
89 # define SHIFT(x) (((x) + (1L << 11)) >> 12)
90
91 /* possible DCT speed optimization */
92
93 # if defined(MAD_F_MLX)
94 # define OPT_DCTO
95 # define MUL(x, y) \
96 ({ mad_fixed64hi_t hi; \
97 mad_fixed64lo_t lo; \
98 MAD_F_MLX(hi, lo, (x), (y)); \
99 hi << (32 - MAD_F_SCALEBITS - 3); \
100 })
101 # else
102 # undef OPT_DCTO
103 # define MUL(x, y) mad_f_mul((x), (y))
104 # endif
105
106 /*
107 * NAME: dct32()
108 * DESCRIPTION: perform fast in[32]->out[32] DCT
109 */
110 static
dct32(mad_fixed_t const in[32],unsigned int slot,mad_fixed_t lo[16][8],mad_fixed_t hi[16][8])111 void dct32(mad_fixed_t const in[32], unsigned int slot,
112 mad_fixed_t lo[16][8], mad_fixed_t hi[16][8])
113 {
114 mad_fixed_t t0, t1, t2, t3, t4, t5, t6, t7;
115 mad_fixed_t t8, t9, t10, t11, t12, t13, t14, t15;
116 mad_fixed_t t16, t17, t18, t19, t20, t21, t22, t23;
117 mad_fixed_t t24, t25, t26, t27, t28, t29, t30, t31;
118 mad_fixed_t t32, t33, t34, t35, t36, t37, t38, t39;
119 mad_fixed_t t40, t41, t42, t43, t44, t45, t46, t47;
120 mad_fixed_t t48, t49, t50, t51, t52, t53, t54, t55;
121 mad_fixed_t t56, t57, t58, t59, t60, t61, t62, t63;
122 mad_fixed_t t64, t65, t66, t67, t68, t69, t70, t71;
123 mad_fixed_t t72, t73, t74, t75, t76, t77, t78, t79;
124 mad_fixed_t t80, t81, t82, t83, t84, t85, t86, t87;
125 mad_fixed_t t88, t89, t90, t91, t92, t93, t94, t95;
126 mad_fixed_t t96, t97, t98, t99, t100, t101, t102, t103;
127 mad_fixed_t t104, t105, t106, t107, t108, t109, t110, t111;
128 mad_fixed_t t112, t113, t114, t115, t116, t117, t118, t119;
129 mad_fixed_t t120, t121, t122, t123, t124, t125, t126, t127;
130 mad_fixed_t t128, t129, t130, t131, t132, t133, t134, t135;
131 mad_fixed_t t136, t137, t138, t139, t140, t141, t142, t143;
132 mad_fixed_t t144, t145, t146, t147, t148, t149, t150, t151;
133 mad_fixed_t t152, t153, t154, t155, t156, t157, t158, t159;
134 mad_fixed_t t160, t161, t162, t163, t164, t165, t166, t167;
135 mad_fixed_t t168, t169, t170, t171, t172, t173, t174, t175;
136 mad_fixed_t t176;
137
138 /* costab[i] = cos(PI / (2 * 32) * i) */
139
140 # if defined(OPT_DCTO)
141 # define costab1 MAD_F(0x7fd8878e)
142 # define costab2 MAD_F(0x7f62368f)
143 # define costab3 MAD_F(0x7e9d55fc)
144 # define costab4 MAD_F(0x7d8a5f40)
145 # define costab5 MAD_F(0x7c29fbee)
146 # define costab6 MAD_F(0x7a7d055b)
147 # define costab7 MAD_F(0x78848414)
148 # define costab8 MAD_F(0x7641af3d)
149 # define costab9 MAD_F(0x73b5ebd1)
150 # define costab10 MAD_F(0x70e2cbc6)
151 # define costab11 MAD_F(0x6dca0d14)
152 # define costab12 MAD_F(0x6a6d98a4)
153 # define costab13 MAD_F(0x66cf8120)
154 # define costab14 MAD_F(0x62f201ac)
155 # define costab15 MAD_F(0x5ed77c8a)
156 # define costab16 MAD_F(0x5a82799a)
157 # define costab17 MAD_F(0x55f5a4d2)
158 # define costab18 MAD_F(0x5133cc94)
159 # define costab19 MAD_F(0x4c3fdff4)
160 # define costab20 MAD_F(0x471cece7)
161 # define costab21 MAD_F(0x41ce1e65)
162 # define costab22 MAD_F(0x3c56ba70)
163 # define costab23 MAD_F(0x36ba2014)
164 # define costab24 MAD_F(0x30fbc54d)
165 # define costab25 MAD_F(0x2b1f34eb)
166 # define costab26 MAD_F(0x25280c5e)
167 # define costab27 MAD_F(0x1f19f97b)
168 # define costab28 MAD_F(0x18f8b83c)
169 # define costab29 MAD_F(0x12c8106f)
170 # define costab30 MAD_F(0x0c8bd35e)
171 # define costab31 MAD_F(0x0647d97c)
172 # else
173 # define costab1 MAD_F(0x0ffb10f2) /* 0.998795456 */
174 # define costab2 MAD_F(0x0fec46d2) /* 0.995184727 */
175 # define costab3 MAD_F(0x0fd3aac0) /* 0.989176510 */
176 # define costab4 MAD_F(0x0fb14be8) /* 0.980785280 */
177 # define costab5 MAD_F(0x0f853f7e) /* 0.970031253 */
178 # define costab6 MAD_F(0x0f4fa0ab) /* 0.956940336 */
179 # define costab7 MAD_F(0x0f109082) /* 0.941544065 */
180 # define costab8 MAD_F(0x0ec835e8) /* 0.923879533 */
181 # define costab9 MAD_F(0x0e76bd7a) /* 0.903989293 */
182 # define costab10 MAD_F(0x0e1c5979) /* 0.881921264 */
183 # define costab11 MAD_F(0x0db941a3) /* 0.857728610 */
184 # define costab12 MAD_F(0x0d4db315) /* 0.831469612 */
185 # define costab13 MAD_F(0x0cd9f024) /* 0.803207531 */
186 # define costab14 MAD_F(0x0c5e4036) /* 0.773010453 */
187 # define costab15 MAD_F(0x0bdaef91) /* 0.740951125 */
188 # define costab16 MAD_F(0x0b504f33) /* 0.707106781 */
189 # define costab17 MAD_F(0x0abeb49a) /* 0.671558955 */
190 # define costab18 MAD_F(0x0a267993) /* 0.634393284 */
191 # define costab19 MAD_F(0x0987fbfe) /* 0.595699304 */
192 # define costab20 MAD_F(0x08e39d9d) /* 0.555570233 */
193 # define costab21 MAD_F(0x0839c3cd) /* 0.514102744 */
194 # define costab22 MAD_F(0x078ad74e) /* 0.471396737 */
195 # define costab23 MAD_F(0x06d74402) /* 0.427555093 */
196 # define costab24 MAD_F(0x061f78aa) /* 0.382683432 */
197 # define costab25 MAD_F(0x0563e69d) /* 0.336889853 */
198 # define costab26 MAD_F(0x04a5018c) /* 0.290284677 */
199 # define costab27 MAD_F(0x03e33f2f) /* 0.242980180 */
200 # define costab28 MAD_F(0x031f1708) /* 0.195090322 */
201 # define costab29 MAD_F(0x0259020e) /* 0.146730474 */
202 # define costab30 MAD_F(0x01917a6c) /* 0.098017140 */
203 # define costab31 MAD_F(0x00c8fb30) /* 0.049067674 */
204 # endif
205
206 t0 = in[0] + in[31]; t16 = MUL(in[0] - in[31], costab1);
207 t1 = in[15] + in[16]; t17 = MUL(in[15] - in[16], costab31);
208
209 t41 = t16 + t17;
210 t59 = MUL(t16 - t17, costab2);
211 t33 = t0 + t1;
212 t50 = MUL(t0 - t1, costab2);
213
214 t2 = in[7] + in[24]; t18 = MUL(in[7] - in[24], costab15);
215 t3 = in[8] + in[23]; t19 = MUL(in[8] - in[23], costab17);
216
217 t42 = t18 + t19;
218 t60 = MUL(t18 - t19, costab30);
219 t34 = t2 + t3;
220 t51 = MUL(t2 - t3, costab30);
221
222 t4 = in[3] + in[28]; t20 = MUL(in[3] - in[28], costab7);
223 t5 = in[12] + in[19]; t21 = MUL(in[12] - in[19], costab25);
224
225 t43 = t20 + t21;
226 t61 = MUL(t20 - t21, costab14);
227 t35 = t4 + t5;
228 t52 = MUL(t4 - t5, costab14);
229
230 t6 = in[4] + in[27]; t22 = MUL(in[4] - in[27], costab9);
231 t7 = in[11] + in[20]; t23 = MUL(in[11] - in[20], costab23);
232
233 t44 = t22 + t23;
234 t62 = MUL(t22 - t23, costab18);
235 t36 = t6 + t7;
236 t53 = MUL(t6 - t7, costab18);
237
238 t8 = in[1] + in[30]; t24 = MUL(in[1] - in[30], costab3);
239 t9 = in[14] + in[17]; t25 = MUL(in[14] - in[17], costab29);
240
241 t45 = t24 + t25;
242 t63 = MUL(t24 - t25, costab6);
243 t37 = t8 + t9;
244 t54 = MUL(t8 - t9, costab6);
245
246 t10 = in[6] + in[25]; t26 = MUL(in[6] - in[25], costab13);
247 t11 = in[9] + in[22]; t27 = MUL(in[9] - in[22], costab19);
248
249 t46 = t26 + t27;
250 t64 = MUL(t26 - t27, costab26);
251 t38 = t10 + t11;
252 t55 = MUL(t10 - t11, costab26);
253
254 t12 = in[2] + in[29]; t28 = MUL(in[2] - in[29], costab5);
255 t13 = in[13] + in[18]; t29 = MUL(in[13] - in[18], costab27);
256
257 t47 = t28 + t29;
258 t65 = MUL(t28 - t29, costab10);
259 t39 = t12 + t13;
260 t56 = MUL(t12 - t13, costab10);
261
262 t14 = in[5] + in[26]; t30 = MUL(in[5] - in[26], costab11);
263 t15 = in[10] + in[21]; t31 = MUL(in[10] - in[21], costab21);
264
265 t48 = t30 + t31;
266 t66 = MUL(t30 - t31, costab22);
267 t40 = t14 + t15;
268 t57 = MUL(t14 - t15, costab22);
269
270 t69 = t33 + t34; t89 = MUL(t33 - t34, costab4);
271 t70 = t35 + t36; t90 = MUL(t35 - t36, costab28);
272 t71 = t37 + t38; t91 = MUL(t37 - t38, costab12);
273 t72 = t39 + t40; t92 = MUL(t39 - t40, costab20);
274 t73 = t41 + t42; t94 = MUL(t41 - t42, costab4);
275 t74 = t43 + t44; t95 = MUL(t43 - t44, costab28);
276 t75 = t45 + t46; t96 = MUL(t45 - t46, costab12);
277 t76 = t47 + t48; t97 = MUL(t47 - t48, costab20);
278
279 t78 = t50 + t51; t100 = MUL(t50 - t51, costab4);
280 t79 = t52 + t53; t101 = MUL(t52 - t53, costab28);
281 t80 = t54 + t55; t102 = MUL(t54 - t55, costab12);
282 t81 = t56 + t57; t103 = MUL(t56 - t57, costab20);
283
284 t83 = t59 + t60; t106 = MUL(t59 - t60, costab4);
285 t84 = t61 + t62; t107 = MUL(t61 - t62, costab28);
286 t85 = t63 + t64; t108 = MUL(t63 - t64, costab12);
287 t86 = t65 + t66; t109 = MUL(t65 - t66, costab20);
288
289 t113 = t69 + t70;
290 t114 = t71 + t72;
291
292 /* 0 */ hi[15][slot] = SHIFT(t113 + t114);
293 /* 16 */ lo[ 0][slot] = SHIFT(MUL(t113 - t114, costab16));
294
295 t115 = t73 + t74;
296 t116 = t75 + t76;
297
298 t32 = t115 + t116;
299
300 /* 1 */ hi[14][slot] = SHIFT(t32);
301
302 t118 = t78 + t79;
303 t119 = t80 + t81;
304
305 t58 = t118 + t119;
306
307 /* 2 */ hi[13][slot] = SHIFT(t58);
308
309 t121 = t83 + t84;
310 t122 = t85 + t86;
311
312 t67 = t121 + t122;
313
314 t49 = (t67 * 2) - t32;
315
316 /* 3 */ hi[12][slot] = SHIFT(t49);
317
318 t125 = t89 + t90;
319 t126 = t91 + t92;
320
321 t93 = t125 + t126;
322
323 /* 4 */ hi[11][slot] = SHIFT(t93);
324
325 t128 = t94 + t95;
326 t129 = t96 + t97;
327
328 t98 = t128 + t129;
329
330 t68 = (t98 * 2) - t49;
331
332 /* 5 */ hi[10][slot] = SHIFT(t68);
333
334 t132 = t100 + t101;
335 t133 = t102 + t103;
336
337 t104 = t132 + t133;
338
339 t82 = (t104 * 2) - t58;
340
341 /* 6 */ hi[ 9][slot] = SHIFT(t82);
342
343 t136 = t106 + t107;
344 t137 = t108 + t109;
345
346 t110 = t136 + t137;
347
348 t87 = (t110 * 2) - t67;
349
350 t77 = (t87 * 2) - t68;
351
352 /* 7 */ hi[ 8][slot] = SHIFT(t77);
353
354 t141 = MUL(t69 - t70, costab8);
355 t142 = MUL(t71 - t72, costab24);
356 t143 = t141 + t142;
357
358 /* 8 */ hi[ 7][slot] = SHIFT(t143);
359 /* 24 */ lo[ 8][slot] =
360 SHIFT((MUL(t141 - t142, costab16) * 2) - t143);
361
362 t144 = MUL(t73 - t74, costab8);
363 t145 = MUL(t75 - t76, costab24);
364 t146 = t144 + t145;
365
366 t88 = (t146 * 2) - t77;
367
368 /* 9 */ hi[ 6][slot] = SHIFT(t88);
369
370 t148 = MUL(t78 - t79, costab8);
371 t149 = MUL(t80 - t81, costab24);
372 t150 = t148 + t149;
373
374 t105 = (t150 * 2) - t82;
375
376 /* 10 */ hi[ 5][slot] = SHIFT(t105);
377
378 t152 = MUL(t83 - t84, costab8);
379 t153 = MUL(t85 - t86, costab24);
380 t154 = t152 + t153;
381
382 t111 = (t154 * 2) - t87;
383
384 t99 = (t111 * 2) - t88;
385
386 /* 11 */ hi[ 4][slot] = SHIFT(t99);
387
388 t157 = MUL(t89 - t90, costab8);
389 t158 = MUL(t91 - t92, costab24);
390 t159 = t157 + t158;
391
392 t127 = (t159 * 2) - t93;
393
394 /* 12 */ hi[ 3][slot] = SHIFT(t127);
395
396 t160 = (MUL(t125 - t126, costab16) * 2) - t127;
397
398 /* 20 */ lo[ 4][slot] = SHIFT(t160);
399 /* 28 */ lo[12][slot] =
400 SHIFT((((MUL(t157 - t158, costab16) * 2) - t159) * 2) - t160);
401
402 t161 = MUL(t94 - t95, costab8);
403 t162 = MUL(t96 - t97, costab24);
404 t163 = t161 + t162;
405
406 t130 = (t163 * 2) - t98;
407
408 t112 = (t130 * 2) - t99;
409
410 /* 13 */ hi[ 2][slot] = SHIFT(t112);
411
412 t164 = (MUL(t128 - t129, costab16) * 2) - t130;
413
414 t166 = MUL(t100 - t101, costab8);
415 t167 = MUL(t102 - t103, costab24);
416 t168 = t166 + t167;
417
418 t134 = (t168 * 2) - t104;
419
420 t120 = (t134 * 2) - t105;
421
422 /* 14 */ hi[ 1][slot] = SHIFT(t120);
423
424 t135 = (MUL(t118 - t119, costab16) * 2) - t120;
425
426 /* 18 */ lo[ 2][slot] = SHIFT(t135);
427
428 t169 = (MUL(t132 - t133, costab16) * 2) - t134;
429
430 t151 = (t169 * 2) - t135;
431
432 /* 22 */ lo[ 6][slot] = SHIFT(t151);
433
434 t170 = (((MUL(t148 - t149, costab16) * 2) - t150) * 2) - t151;
435
436 /* 26 */ lo[10][slot] = SHIFT(t170);
437 /* 30 */ lo[14][slot] =
438 SHIFT((((((MUL(t166 - t167, costab16) * 2) -
439 t168) * 2) - t169) * 2) - t170);
440
441 t171 = MUL(t106 - t107, costab8);
442 t172 = MUL(t108 - t109, costab24);
443 t173 = t171 + t172;
444
445 t138 = (t173 * 2) - t110;
446
447 t123 = (t138 * 2) - t111;
448
449 t139 = (MUL(t121 - t122, costab16) * 2) - t123;
450
451 t117 = (t123 * 2) - t112;
452
453 /* 15 */ hi[ 0][slot] = SHIFT(t117);
454
455 t124 = (MUL(t115 - t116, costab16) * 2) - t117;
456
457 /* 17 */ lo[ 1][slot] = SHIFT(t124);
458
459 t131 = (t139 * 2) - t124;
460
461 /* 19 */ lo[ 3][slot] = SHIFT(t131);
462
463 t140 = (t164 * 2) - t131;
464
465 /* 21 */ lo[ 5][slot] = SHIFT(t140);
466
467 t174 = (MUL(t136 - t137, costab16) * 2) - t138;
468
469 t155 = (t174 * 2) - t139;
470
471 t147 = (t155 * 2) - t140;
472
473 /* 23 */ lo[ 7][slot] = SHIFT(t147);
474
475 t156 = (((MUL(t144 - t145, costab16) * 2) - t146) * 2) - t147;
476
477 /* 25 */ lo[ 9][slot] = SHIFT(t156);
478
479 t175 = (((MUL(t152 - t153, costab16) * 2) - t154) * 2) - t155;
480
481 t165 = (t175 * 2) - t156;
482
483 /* 27 */ lo[11][slot] = SHIFT(t165);
484
485 t176 = (((((MUL(t161 - t162, costab16) * 2) -
486 t163) * 2) - t164) * 2) - t165;
487
488 /* 29 */ lo[13][slot] = SHIFT(t176);
489 /* 31 */ lo[15][slot] =
490 SHIFT((((((((MUL(t171 - t172, costab16) * 2) -
491 t173) * 2) - t174) * 2) - t175) * 2) - t176);
492
493 /*
494 * Totals:
495 * 80 multiplies
496 * 80 additions
497 * 119 subtractions
498 * 49 shifts (not counting SSO)
499 */
500 }
501
502 # undef MUL
503 # undef SHIFT
504
505 /* third SSO shift and/or D[] optimization preshift */
506
507 # if MAD_F_FRACBITS != 28
508 # error "MAD_F_FRACBITS must be 28 to use OPT_SSO"
509 # endif
510 # define ML0(hi, lo, x, y) ((lo) = (x) * (y))
511 # define MLA(hi, lo, x, y) ((lo) += (x) * (y))
512 # define MLN(hi, lo) ((lo) = -(lo))
513 # define MLZ(hi, lo) ((void) (hi), (mad_fixed_t) (lo))
514 # define SHIFT(x) ((x) >> 2)
515 # define PRESHIFT(x) ((MAD_F(x) + (1L << 13)) >> 14)
516
517 static
518 mad_fixed_t const D[17][32] = {
519 # include "D.dat"
520 };
521
522 # if defined(ASO_SYNTH)
523 void synth_full(struct mad_synth *, struct mad_frame const *,
524 unsigned int, unsigned int);
525 # else
526 /*
527 * NAME: synth->full()
528 * DESCRIPTION: perform full frequency PCM synthesis
529 */
530 static
synth_full(struct mad_synth * synth,struct mad_frame const * frame,unsigned int nch,unsigned int ns)531 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
532 unsigned int nch, unsigned int ns)
533 {
534 unsigned int phase, ch, s, sb, pe, po;
535 mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8];
536 mad_fixed_t const (*sbsample)[36][32];
537 register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
538 register mad_fixed_t const (*Dptr)[32], *ptr;
539 register mad_fixed64hi_t hi;
540 register mad_fixed64lo_t lo;
541
542 for (ch = 0; ch < nch; ++ch) {
543 sbsample = &frame->sbsample[ch];
544 filter = &synth->filter[ch];
545 phase = synth->phase;
546 pcm1 = synth->pcm.samples[ch];
547
548 for (s = 0; s < ns; ++s) {
549 dct32((*sbsample)[s], phase >> 1,
550 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
551
552 pe = phase & ~1;
553 po = ((phase - 1) & 0xf) | 1;
554
555 /* calculate 32 samples */
556
557 fe = &(*filter)[0][ phase & 1][0];
558 fx = &(*filter)[0][~phase & 1][0];
559 fo = &(*filter)[1][~phase & 1][0];
560
561 Dptr = &D[0];
562
563 ptr = *Dptr + po;
564 ML0(hi, lo, (*fx)[0], ptr[ 0]);
565 MLA(hi, lo, (*fx)[1], ptr[14]);
566 MLA(hi, lo, (*fx)[2], ptr[12]);
567 MLA(hi, lo, (*fx)[3], ptr[10]);
568 MLA(hi, lo, (*fx)[4], ptr[ 8]);
569 MLA(hi, lo, (*fx)[5], ptr[ 6]);
570 MLA(hi, lo, (*fx)[6], ptr[ 4]);
571 MLA(hi, lo, (*fx)[7], ptr[ 2]);
572 MLN(hi, lo);
573
574 ptr = *Dptr + pe;
575 MLA(hi, lo, (*fe)[0], ptr[ 0]);
576 MLA(hi, lo, (*fe)[1], ptr[14]);
577 MLA(hi, lo, (*fe)[2], ptr[12]);
578 MLA(hi, lo, (*fe)[3], ptr[10]);
579 MLA(hi, lo, (*fe)[4], ptr[ 8]);
580 MLA(hi, lo, (*fe)[5], ptr[ 6]);
581 MLA(hi, lo, (*fe)[6], ptr[ 4]);
582 MLA(hi, lo, (*fe)[7], ptr[ 2]);
583
584 *pcm1++ = SHIFT(MLZ(hi, lo));
585
586 pcm2 = pcm1 + 30;
587
588 for (sb = 1; sb < 16; ++sb) {
589 ++fe;
590 ++Dptr;
591
592 /* D[32 - sb][i] == -D[sb][31 - i] */
593
594 ptr = *Dptr + po;
595 ML0(hi, lo, (*fo)[0], ptr[ 0]);
596 MLA(hi, lo, (*fo)[1], ptr[14]);
597 MLA(hi, lo, (*fo)[2], ptr[12]);
598 MLA(hi, lo, (*fo)[3], ptr[10]);
599 MLA(hi, lo, (*fo)[4], ptr[ 8]);
600 MLA(hi, lo, (*fo)[5], ptr[ 6]);
601 MLA(hi, lo, (*fo)[6], ptr[ 4]);
602 MLA(hi, lo, (*fo)[7], ptr[ 2]);
603 MLN(hi, lo);
604
605 ptr = *Dptr + pe;
606 MLA(hi, lo, (*fe)[7], ptr[ 2]);
607 MLA(hi, lo, (*fe)[6], ptr[ 4]);
608 MLA(hi, lo, (*fe)[5], ptr[ 6]);
609 MLA(hi, lo, (*fe)[4], ptr[ 8]);
610 MLA(hi, lo, (*fe)[3], ptr[10]);
611 MLA(hi, lo, (*fe)[2], ptr[12]);
612 MLA(hi, lo, (*fe)[1], ptr[14]);
613 MLA(hi, lo, (*fe)[0], ptr[ 0]);
614
615 *pcm1++ = SHIFT(MLZ(hi, lo));
616
617 ptr = *Dptr - pe;
618 ML0(hi, lo, (*fe)[0], ptr[31 - 16]);
619 MLA(hi, lo, (*fe)[1], ptr[31 - 14]);
620 MLA(hi, lo, (*fe)[2], ptr[31 - 12]);
621 MLA(hi, lo, (*fe)[3], ptr[31 - 10]);
622 MLA(hi, lo, (*fe)[4], ptr[31 - 8]);
623 MLA(hi, lo, (*fe)[5], ptr[31 - 6]);
624 MLA(hi, lo, (*fe)[6], ptr[31 - 4]);
625 MLA(hi, lo, (*fe)[7], ptr[31 - 2]);
626
627 ptr = *Dptr - po;
628 MLA(hi, lo, (*fo)[7], ptr[31 - 2]);
629 MLA(hi, lo, (*fo)[6], ptr[31 - 4]);
630 MLA(hi, lo, (*fo)[5], ptr[31 - 6]);
631 MLA(hi, lo, (*fo)[4], ptr[31 - 8]);
632 MLA(hi, lo, (*fo)[3], ptr[31 - 10]);
633 MLA(hi, lo, (*fo)[2], ptr[31 - 12]);
634 MLA(hi, lo, (*fo)[1], ptr[31 - 14]);
635 MLA(hi, lo, (*fo)[0], ptr[31 - 16]);
636
637 *pcm2-- = SHIFT(MLZ(hi, lo));
638
639 ++fo;
640 }
641
642 ++Dptr;
643
644 ptr = *Dptr + po;
645 ML0(hi, lo, (*fo)[0], ptr[ 0]);
646 MLA(hi, lo, (*fo)[1], ptr[14]);
647 MLA(hi, lo, (*fo)[2], ptr[12]);
648 MLA(hi, lo, (*fo)[3], ptr[10]);
649 MLA(hi, lo, (*fo)[4], ptr[ 8]);
650 MLA(hi, lo, (*fo)[5], ptr[ 6]);
651 MLA(hi, lo, (*fo)[6], ptr[ 4]);
652 MLA(hi, lo, (*fo)[7], ptr[ 2]);
653
654 *pcm1 = SHIFT(-MLZ(hi, lo));
655 pcm1 += 16;
656
657 phase = (phase + 1) % 16;
658 }
659 }
660 }
661 # endif
662
663 /*
664 * NAME: synth->half()
665 * DESCRIPTION: perform half frequency PCM synthesis
666 */
667 static
synth_half(struct mad_synth * synth,struct mad_frame const * frame,unsigned int nch,unsigned int ns)668 void synth_half(struct mad_synth *synth, struct mad_frame const *frame,
669 unsigned int nch, unsigned int ns)
670 {
671 unsigned int phase, ch, s, sb, pe, po;
672 mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8];
673 mad_fixed_t const (*sbsample)[36][32];
674 register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
675 register mad_fixed_t const (*Dptr)[32], *ptr;
676 register mad_fixed64hi_t hi;
677 register mad_fixed64lo_t lo;
678
679 for (ch = 0; ch < nch; ++ch) {
680 sbsample = &frame->sbsample[ch];
681 filter = &synth->filter[ch];
682 phase = synth->phase;
683 pcm1 = synth->pcm.samples[ch];
684
685 for (s = 0; s < ns; ++s) {
686 dct32((*sbsample)[s], phase >> 1,
687 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
688
689 pe = phase & ~1;
690 po = ((phase - 1) & 0xf) | 1;
691
692 /* calculate 16 samples */
693
694 fe = &(*filter)[0][ phase & 1][0];
695 fx = &(*filter)[0][~phase & 1][0];
696 fo = &(*filter)[1][~phase & 1][0];
697
698 Dptr = &D[0];
699
700 ptr = *Dptr + po;
701 ML0(hi, lo, (*fx)[0], ptr[ 0]);
702 MLA(hi, lo, (*fx)[1], ptr[14]);
703 MLA(hi, lo, (*fx)[2], ptr[12]);
704 MLA(hi, lo, (*fx)[3], ptr[10]);
705 MLA(hi, lo, (*fx)[4], ptr[ 8]);
706 MLA(hi, lo, (*fx)[5], ptr[ 6]);
707 MLA(hi, lo, (*fx)[6], ptr[ 4]);
708 MLA(hi, lo, (*fx)[7], ptr[ 2]);
709 MLN(hi, lo);
710
711 ptr = *Dptr + pe;
712 MLA(hi, lo, (*fe)[0], ptr[ 0]);
713 MLA(hi, lo, (*fe)[1], ptr[14]);
714 MLA(hi, lo, (*fe)[2], ptr[12]);
715 MLA(hi, lo, (*fe)[3], ptr[10]);
716 MLA(hi, lo, (*fe)[4], ptr[ 8]);
717 MLA(hi, lo, (*fe)[5], ptr[ 6]);
718 MLA(hi, lo, (*fe)[6], ptr[ 4]);
719 MLA(hi, lo, (*fe)[7], ptr[ 2]);
720
721 *pcm1++ = SHIFT(MLZ(hi, lo));
722
723 pcm2 = pcm1 + 14;
724
725 for (sb = 1; sb < 16; ++sb) {
726 ++fe;
727 ++Dptr;
728
729 /* D[32 - sb][i] == -D[sb][31 - i] */
730
731 if (!(sb & 1)) {
732 ptr = *Dptr + po;
733 ML0(hi, lo, (*fo)[0], ptr[ 0]);
734 MLA(hi, lo, (*fo)[1], ptr[14]);
735 MLA(hi, lo, (*fo)[2], ptr[12]);
736 MLA(hi, lo, (*fo)[3], ptr[10]);
737 MLA(hi, lo, (*fo)[4], ptr[ 8]);
738 MLA(hi, lo, (*fo)[5], ptr[ 6]);
739 MLA(hi, lo, (*fo)[6], ptr[ 4]);
740 MLA(hi, lo, (*fo)[7], ptr[ 2]);
741 MLN(hi, lo);
742
743 ptr = *Dptr + pe;
744 MLA(hi, lo, (*fe)[7], ptr[ 2]);
745 MLA(hi, lo, (*fe)[6], ptr[ 4]);
746 MLA(hi, lo, (*fe)[5], ptr[ 6]);
747 MLA(hi, lo, (*fe)[4], ptr[ 8]);
748 MLA(hi, lo, (*fe)[3], ptr[10]);
749 MLA(hi, lo, (*fe)[2], ptr[12]);
750 MLA(hi, lo, (*fe)[1], ptr[14]);
751 MLA(hi, lo, (*fe)[0], ptr[ 0]);
752
753 *pcm1++ = SHIFT(MLZ(hi, lo));
754
755 ptr = *Dptr - po;
756 ML0(hi, lo, (*fo)[7], ptr[31 - 2]);
757 MLA(hi, lo, (*fo)[6], ptr[31 - 4]);
758 MLA(hi, lo, (*fo)[5], ptr[31 - 6]);
759 MLA(hi, lo, (*fo)[4], ptr[31 - 8]);
760 MLA(hi, lo, (*fo)[3], ptr[31 - 10]);
761 MLA(hi, lo, (*fo)[2], ptr[31 - 12]);
762 MLA(hi, lo, (*fo)[1], ptr[31 - 14]);
763 MLA(hi, lo, (*fo)[0], ptr[31 - 16]);
764
765 ptr = *Dptr - pe;
766 MLA(hi, lo, (*fe)[0], ptr[31 - 16]);
767 MLA(hi, lo, (*fe)[1], ptr[31 - 14]);
768 MLA(hi, lo, (*fe)[2], ptr[31 - 12]);
769 MLA(hi, lo, (*fe)[3], ptr[31 - 10]);
770 MLA(hi, lo, (*fe)[4], ptr[31 - 8]);
771 MLA(hi, lo, (*fe)[5], ptr[31 - 6]);
772 MLA(hi, lo, (*fe)[6], ptr[31 - 4]);
773 MLA(hi, lo, (*fe)[7], ptr[31 - 2]);
774
775 *pcm2-- = SHIFT(MLZ(hi, lo));
776 }
777
778 ++fo;
779 }
780
781 ++Dptr;
782
783 ptr = *Dptr + po;
784 ML0(hi, lo, (*fo)[0], ptr[ 0]);
785 MLA(hi, lo, (*fo)[1], ptr[14]);
786 MLA(hi, lo, (*fo)[2], ptr[12]);
787 MLA(hi, lo, (*fo)[3], ptr[10]);
788 MLA(hi, lo, (*fo)[4], ptr[ 8]);
789 MLA(hi, lo, (*fo)[5], ptr[ 6]);
790 MLA(hi, lo, (*fo)[6], ptr[ 4]);
791 MLA(hi, lo, (*fo)[7], ptr[ 2]);
792
793 *pcm1 = SHIFT(-MLZ(hi, lo));
794 pcm1 += 8;
795
796 phase = (phase + 1) % 16;
797 }
798 }
799 }
800
801 /*
802 * NAME: synth->frame()
803 * DESCRIPTION: perform PCM synthesis of frame subband samples
804 */
mad_synth_frame(struct mad_synth * synth,struct mad_frame const * frame)805 void mad_synth_frame(struct mad_synth *synth, struct mad_frame const *frame)
806 {
807 unsigned int nch, ns;
808 void (*synth_frame)(struct mad_synth *, struct mad_frame const *,
809 unsigned int, unsigned int);
810
811 nch = MAD_NCHANNELS(&frame->header);
812 ns = MAD_NSBSAMPLES(&frame->header);
813
814 synth->pcm.samplerate = frame->header.samplerate;
815 synth->pcm.channels = nch;
816 synth->pcm.length = 32 * ns;
817
818 synth_frame = synth_full;
819
820 if (frame->options & MAD_OPTION_HALFSAMPLERATE) {
821 synth->pcm.samplerate /= 2;
822 synth->pcm.length /= 2;
823
824 synth_frame = synth_half;
825 }
826
827 synth_frame(synth, frame, nch, ns);
828
829 synth->phase = (synth->phase + ns) % 16;
830 }
831