rtoss - Blame information for rev 26

Subversion Repositories:
Rev:
Rev Author Line No. Line
14 roytam 1 /*
2  * libmad - MPEG audio decoder library
26 roytam 3  * Copyright (C) 2000-2004 Underbit Technologies, Inc.
14 roytam 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  *
26 roytam 19  * $Id: synth.c,v 1.25 2004/01/23 09:41:33 rob Exp $
14 roytam 20  */
21  
22 # ifdef HAVE_CONFIG_H
23 #  include "config.h"
24 # endif
25  
26 # include "global.h"
27  
28 # include "fixed.h"
29 # include "frame.h"
30 # include "synth.h"
31  
32 /*
33  * NAME:        synth->init()
34  * DESCRIPTION: initialize synth struct
35  */
36 void mad_synth_init(struct mad_synth *synth)
37 {
38   mad_synth_mute(synth);
39  
40   synth->phase = 0;
41  
42   synth->pcm.samplerate = 0;
43   synth->pcm.channels   = 0;
44   synth->pcm.length     = 0;
45 }
46  
47 /*
48  * NAME:        synth->mute()
49  * DESCRIPTION: zero all polyphase filterbank values, resetting synthesis
50  */
51 void mad_synth_mute(struct mad_synth *synth)
52 {
53   unsigned int ch, s, v;
54  
55   for (ch = 0; ch < 2; ++ch) {
56     for (s = 0; s < 16; ++s) {
57       for (v = 0; v < 8; ++v) {
58         synth->filter[ch][0][0][s][v] = synth->filter[ch][0][1][s][v] =
59         synth->filter[ch][1][0][s][v] = synth->filter[ch][1][1][s][v] = 0;
60       }
61     }
62   }
63 }
64  
65 /*
66  * An optional optimization called here the Subband Synthesis Optimization
67  * (SSO) improves the performance of subband synthesis at the expense of
68  * accuracy.
69  *
70  * The idea is to simplify 32x32->64-bit multiplication to 32x32->32 such
71  * that extra scaling and rounding are not necessary. This often allows the
72  * compiler to use faster 32-bit multiply-accumulate instructions instead of
73  * explicit 64-bit multiply, shift, and add instructions.
74  *
75  * SSO works like this: a full 32x32->64-bit multiply of two mad_fixed_t
76  * values requires the result to be right-shifted 28 bits to be properly
77  * scaled to the same fixed-point format. Right shifts can be applied at any
78  * time to either operand or to the result, so the optimization involves
79  * careful placement of these shifts to minimize the loss of accuracy.
80  *
81  * First, a 14-bit shift is applied with rounding at compile-time to the D[]
82  * table of coefficients for the subband synthesis window. This only loses 2
83  * bits of accuracy because the lower 12 bits are always zero. A second
84  * 12-bit shift occurs after the DCT calculation. This loses 12 bits of
85  * accuracy. Finally, a third 2-bit shift occurs just before the sample is
86  * saved in the PCM buffer. 14 + 12 + 2 == 28 bits.
87  */
88  
89 /* FPM_DEFAULT without OPT_SSO will actually lose accuracy and performance */
90  
91 # if defined(FPM_DEFAULT) && !defined(OPT_SSO)
92 #  define OPT_SSO
93 # endif
94  
95 /* second SSO shift, with rounding */
96  
97 # if defined(OPT_SSO)
98 #  define SHIFT(x)  (((x) + (1L << 11)) >> 12)
99 # else
100 #  define SHIFT(x)  (x)
101 # endif
102  
103 /* possible DCT speed optimization */
104  
105 # if defined(OPT_SPEED) && defined(MAD_F_MLX)
106 #  define OPT_DCTO
107 #  define MUL(x, y)  \
108     ({ mad_fixed64hi_t hi;  \
109        mad_fixed64lo_t lo;  \
110        MAD_F_MLX(hi, lo, (x), (y));  \
111        hi << (32 - MAD_F_SCALEBITS - 3);  \
112     })
113 # else
114 #  undef OPT_DCTO
115 #  define MUL(x, y)  mad_f_mul((x), (y))
116 # endif
117  
118 /*
119  * NAME:        dct32()
120  * DESCRIPTION: perform fast in[32]->out[32] DCT
121  */
122 static
123 void dct32(mad_fixed_t const in[32], unsigned int slot,
124            mad_fixed_t lo[16][8], mad_fixed_t hi[16][8])
125 {
126   mad_fixed_t t0,   t1,   t2,   t3,   t4,   t5,   t6,   t7;
127   mad_fixed_t t8,   t9,   t10,  t11,  t12,  t13,  t14,  t15;
128   mad_fixed_t t16,  t17,  t18,  t19,  t20,  t21,  t22,  t23;
129   mad_fixed_t t24,  t25,  t26,  t27,  t28,  t29,  t30,  t31;
130   mad_fixed_t t32,  t33,  t34,  t35,  t36,  t37,  t38,  t39;
131   mad_fixed_t t40,  t41,  t42,  t43,  t44,  t45,  t46,  t47;
132   mad_fixed_t t48,  t49,  t50,  t51,  t52,  t53,  t54,  t55;
133   mad_fixed_t t56,  t57,  t58,  t59,  t60,  t61,  t62,  t63;
134   mad_fixed_t t64,  t65,  t66,  t67,  t68,  t69,  t70,  t71;
135   mad_fixed_t t72,  t73,  t74,  t75,  t76,  t77,  t78,  t79;
136   mad_fixed_t t80,  t81,  t82,  t83,  t84,  t85,  t86,  t87;
137   mad_fixed_t t88,  t89,  t90,  t91,  t92,  t93,  t94,  t95;
138   mad_fixed_t t96,  t97,  t98,  t99,  t100, t101, t102, t103;
139   mad_fixed_t t104, t105, t106, t107, t108, t109, t110, t111;
140   mad_fixed_t t112, t113, t114, t115, t116, t117, t118, t119;
141   mad_fixed_t t120, t121, t122, t123, t124, t125, t126, t127;
142   mad_fixed_t t128, t129, t130, t131, t132, t133, t134, t135;
143   mad_fixed_t t136, t137, t138, t139, t140, t141, t142, t143;
144   mad_fixed_t t144, t145, t146, t147, t148, t149, t150, t151;
145   mad_fixed_t t152, t153, t154, t155, t156, t157, t158, t159;
146   mad_fixed_t t160, t161, t162, t163, t164, t165, t166, t167;
147   mad_fixed_t t168, t169, t170, t171, t172, t173, t174, t175;
148   mad_fixed_t t176;
149  
150   /* costab[i] = cos(PI / (2 * 32) * i) */
151  
152 # if defined(OPT_DCTO)
153 #  define costab1       MAD_F(0x7fd8878e)
154 #  define costab2       MAD_F(0x7f62368f)
155 #  define costab3       MAD_F(0x7e9d55fc)
156 #  define costab4       MAD_F(0x7d8a5f40)
157 #  define costab5       MAD_F(0x7c29fbee)
158 #  define costab6       MAD_F(0x7a7d055b)
159 #  define costab7       MAD_F(0x78848414)
160 #  define costab8       MAD_F(0x7641af3d)
161 #  define costab9       MAD_F(0x73b5ebd1)
162 #  define costab10      MAD_F(0x70e2cbc6)
163 #  define costab11      MAD_F(0x6dca0d14)
164 #  define costab12      MAD_F(0x6a6d98a4)
165 #  define costab13      MAD_F(0x66cf8120)
166 #  define costab14      MAD_F(0x62f201ac)
167 #  define costab15      MAD_F(0x5ed77c8a)
168 #  define costab16      MAD_F(0x5a82799a)
169 #  define costab17      MAD_F(0x55f5a4d2)
170 #  define costab18      MAD_F(0x5133cc94)
171 #  define costab19      MAD_F(0x4c3fdff4)
172 #  define costab20      MAD_F(0x471cece7)
173 #  define costab21      MAD_F(0x41ce1e65)
174 #  define costab22      MAD_F(0x3c56ba70)
175 #  define costab23      MAD_F(0x36ba2014)
176 #  define costab24      MAD_F(0x30fbc54d)
177 #  define costab25      MAD_F(0x2b1f34eb)
178 #  define costab26      MAD_F(0x25280c5e)
179 #  define costab27      MAD_F(0x1f19f97b)
180 #  define costab28      MAD_F(0x18f8b83c)
181 #  define costab29      MAD_F(0x12c8106f)
182 #  define costab30      MAD_F(0x0c8bd35e)
183 #  define costab31      MAD_F(0x0647d97c)
184 # else
185 #  define costab1       MAD_F(0x0ffb10f2)  /* 0.998795456 */
186 #  define costab2       MAD_F(0x0fec46d2)  /* 0.995184727 */
187 #  define costab3       MAD_F(0x0fd3aac0)  /* 0.989176510 */
188 #  define costab4       MAD_F(0x0fb14be8)  /* 0.980785280 */
189 #  define costab5       MAD_F(0x0f853f7e)  /* 0.970031253 */
190 #  define costab6       MAD_F(0x0f4fa0ab)  /* 0.956940336 */
191 #  define costab7       MAD_F(0x0f109082)  /* 0.941544065 */
192 #  define costab8       MAD_F(0x0ec835e8)  /* 0.923879533 */
193 #  define costab9       MAD_F(0x0e76bd7a)  /* 0.903989293 */
194 #  define costab10      MAD_F(0x0e1c5979)  /* 0.881921264 */
195 #  define costab11      MAD_F(0x0db941a3)  /* 0.857728610 */
196 #  define costab12      MAD_F(0x0d4db315)  /* 0.831469612 */
197 #  define costab13      MAD_F(0x0cd9f024)  /* 0.803207531 */
198 #  define costab14      MAD_F(0x0c5e4036)  /* 0.773010453 */
199 #  define costab15      MAD_F(0x0bdaef91)  /* 0.740951125 */
200 #  define costab16      MAD_F(0x0b504f33)  /* 0.707106781 */
201 #  define costab17      MAD_F(0x0abeb49a)  /* 0.671558955 */
202 #  define costab18      MAD_F(0x0a267993)  /* 0.634393284 */
203 #  define costab19      MAD_F(0x0987fbfe)  /* 0.595699304 */
204 #  define costab20      MAD_F(0x08e39d9d)  /* 0.555570233 */
205 #  define costab21      MAD_F(0x0839c3cd)  /* 0.514102744 */
206 #  define costab22      MAD_F(0x078ad74e)  /* 0.471396737 */
207 #  define costab23      MAD_F(0x06d74402)  /* 0.427555093 */
208 #  define costab24      MAD_F(0x061f78aa)  /* 0.382683432 */
209 #  define costab25      MAD_F(0x0563e69d)  /* 0.336889853 */
210 #  define costab26      MAD_F(0x04a5018c)  /* 0.290284677 */
211 #  define costab27      MAD_F(0x03e33f2f)  /* 0.242980180 */
212 #  define costab28      MAD_F(0x031f1708)  /* 0.195090322 */
213 #  define costab29      MAD_F(0x0259020e)  /* 0.146730474 */
214 #  define costab30      MAD_F(0x01917a6c)  /* 0.098017140 */
215 #  define costab31      MAD_F(0x00c8fb30)  /* 0.049067674 */
216 # endif
217  
218   t0   = in[0]  + in[31];  t16  = MUL(in[0]  - in[31], costab1);
219   t1   = in[15] + in[16];  t17  = MUL(in[15] - in[16], costab31);
220  
221   t41  = t16 + t17;
222   t59  = MUL(t16 - t17, costab2);
223   t33  = t0  + t1;
224   t50  = MUL(t0  - t1,  costab2);
225  
226   t2   = in[7]  + in[24];  t18  = MUL(in[7]  - in[24], costab15);
227   t3   = in[8]  + in[23];  t19  = MUL(in[8]  - in[23], costab17);
228  
229   t42  = t18 + t19;
230   t60  = MUL(t18 - t19, costab30);
231   t34  = t2  + t3;
232   t51  = MUL(t2  - t3,  costab30);
233  
234   t4   = in[3]  + in[28];  t20  = MUL(in[3]  - in[28], costab7);
235   t5   = in[12] + in[19];  t21  = MUL(in[12] - in[19], costab25);
236  
237   t43  = t20 + t21;
238   t61  = MUL(t20 - t21, costab14);
239   t35  = t4  + t5;
240   t52  = MUL(t4  - t5,  costab14);
241  
242   t6   = in[4]  + in[27];  t22  = MUL(in[4]  - in[27], costab9);
243   t7   = in[11] + in[20];  t23  = MUL(in[11] - in[20], costab23);
244  
245   t44  = t22 + t23;
246   t62  = MUL(t22 - t23, costab18);
247   t36  = t6  + t7;
248   t53  = MUL(t6  - t7,  costab18);
249  
250   t8   = in[1]  + in[30];  t24  = MUL(in[1]  - in[30], costab3);
251   t9   = in[14] + in[17];  t25  = MUL(in[14] - in[17], costab29);
252  
253   t45  = t24 + t25;
254   t63  = MUL(t24 - t25, costab6);
255   t37  = t8  + t9;
256   t54  = MUL(t8  - t9,  costab6);
257  
258   t10  = in[6]  + in[25];  t26  = MUL(in[6]  - in[25], costab13);
259   t11  = in[9]  + in[22];  t27  = MUL(in[9]  - in[22], costab19);
260  
261   t46  = t26 + t27;
262   t64  = MUL(t26 - t27, costab26);
263   t38  = t10 + t11;
264   t55  = MUL(t10 - t11, costab26);
265  
266   t12  = in[2]  + in[29];  t28  = MUL(in[2]  - in[29], costab5);
267   t13  = in[13] + in[18];  t29  = MUL(in[13] - in[18], costab27);
268  
269   t47  = t28 + t29;
270   t65  = MUL(t28 - t29, costab10);
271   t39  = t12 + t13;
272   t56  = MUL(t12 - t13, costab10);
273  
274   t14  = in[5]  + in[26];  t30  = MUL(in[5]  - in[26], costab11);
275   t15  = in[10] + in[21];  t31  = MUL(in[10] - in[21], costab21);
276  
277   t48  = t30 + t31;
278   t66  = MUL(t30 - t31, costab22);
279   t40  = t14 + t15;
280   t57  = MUL(t14 - t15, costab22);
281  
282   t69  = t33 + t34;  t89  = MUL(t33 - t34, costab4);
283   t70  = t35 + t36;  t90  = MUL(t35 - t36, costab28);
284   t71  = t37 + t38;  t91  = MUL(t37 - t38, costab12);
285   t72  = t39 + t40;  t92  = MUL(t39 - t40, costab20);
286   t73  = t41 + t42;  t94  = MUL(t41 - t42, costab4);
287   t74  = t43 + t44;  t95  = MUL(t43 - t44, costab28);
288   t75  = t45 + t46;  t96  = MUL(t45 - t46, costab12);
289   t76  = t47 + t48;  t97  = MUL(t47 - t48, costab20);
290  
291   t78  = t50 + t51;  t100 = MUL(t50 - t51, costab4);
292   t79  = t52 + t53;  t101 = MUL(t52 - t53, costab28);
293   t80  = t54 + t55;  t102 = MUL(t54 - t55, costab12);
294   t81  = t56 + t57;  t103 = MUL(t56 - t57, costab20);
295  
296   t83  = t59 + t60;  t106 = MUL(t59 - t60, costab4);
297   t84  = t61 + t62;  t107 = MUL(t61 - t62, costab28);
298   t85  = t63 + t64;  t108 = MUL(t63 - t64, costab12);
299   t86  = t65 + t66;  t109 = MUL(t65 - t66, costab20);
300  
301   t113 = t69  + t70;
302   t114 = t71  + t72;
303  
304   /*  0 */ hi[15][slot] = SHIFT(t113 + t114);
305   /* 16 */ lo[ 0][slot] = SHIFT(MUL(t113 - t114, costab16));
306  
307   t115 = t73  + t74;
308   t116 = t75  + t76;
309  
310   t32  = t115 + t116;
311  
312   /*  1 */ hi[14][slot] = SHIFT(t32);
313  
314   t118 = t78  + t79;
315   t119 = t80  + t81;
316  
317   t58  = t118 + t119;
318  
319   /*  2 */ hi[13][slot] = SHIFT(t58);
320  
321   t121 = t83  + t84;
322   t122 = t85  + t86;
323  
324   t67  = t121 + t122;
325  
326   t49  = (t67 * 2) - t32;
327  
328   /*  3 */ hi[12][slot] = SHIFT(t49);
329  
330   t125 = t89  + t90;
331   t126 = t91  + t92;
332  
333   t93  = t125 + t126;
334  
335   /*  4 */ hi[11][slot] = SHIFT(t93);
336  
337   t128 = t94  + t95;
338   t129 = t96  + t97;
339  
340   t98  = t128 + t129;
341  
342   t68  = (t98 * 2) - t49;
343  
344   /*  5 */ hi[10][slot] = SHIFT(t68);
345  
346   t132 = t100 + t101;
347   t133 = t102 + t103;
348  
349   t104 = t132 + t133;
350  
351   t82  = (t104 * 2) - t58;
352  
353   /*  6 */ hi[ 9][slot] = SHIFT(t82);
354  
355   t136 = t106 + t107;
356   t137 = t108 + t109;
357  
358   t110 = t136 + t137;
359  
360   t87  = (t110 * 2) - t67;
361  
362   t77  = (t87 * 2) - t68;
363  
364   /*  7 */ hi[ 8][slot] = SHIFT(t77);
365  
366   t141 = MUL(t69 - t70, costab8);
367   t142 = MUL(t71 - t72, costab24);
368   t143 = t141 + t142;
369  
370   /*  8 */ hi[ 7][slot] = SHIFT(t143);
371   /* 24 */ lo[ 8][slot] =
372              SHIFT((MUL(t141 - t142, costab16) * 2) - t143);
373  
374   t144 = MUL(t73 - t74, costab8);
375   t145 = MUL(t75 - t76, costab24);
376   t146 = t144 + t145;
377  
378   t88  = (t146 * 2) - t77;
379  
380   /*  9 */ hi[ 6][slot] = SHIFT(t88);
381  
382   t148 = MUL(t78 - t79, costab8);
383   t149 = MUL(t80 - t81, costab24);
384   t150 = t148 + t149;
385  
386   t105 = (t150 * 2) - t82;
387  
388   /* 10 */ hi[ 5][slot] = SHIFT(t105);
389  
390   t152 = MUL(t83 - t84, costab8);
391   t153 = MUL(t85 - t86, costab24);
392   t154 = t152 + t153;
393  
394   t111 = (t154 * 2) - t87;
395  
396   t99  = (t111 * 2) - t88;
397  
398   /* 11 */ hi[ 4][slot] = SHIFT(t99);
399  
400   t157 = MUL(t89 - t90, costab8);
401   t158 = MUL(t91 - t92, costab24);
402   t159 = t157 + t158;
403  
404   t127 = (t159 * 2) - t93;
405  
406   /* 12 */ hi[ 3][slot] = SHIFT(t127);
407  
408   t160 = (MUL(t125 - t126, costab16) * 2) - t127;
409  
410   /* 20 */ lo[ 4][slot] = SHIFT(t160);
411   /* 28 */ lo[12][slot] =
412              SHIFT((((MUL(t157 - t158, costab16) * 2) - t159) * 2) - t160);
413  
414   t161 = MUL(t94 - t95, costab8);
415   t162 = MUL(t96 - t97, costab24);
416   t163 = t161 + t162;
417  
418   t130 = (t163 * 2) - t98;
419  
420   t112 = (t130 * 2) - t99;
421  
422   /* 13 */ hi[ 2][slot] = SHIFT(t112);
423  
424   t164 = (MUL(t128 - t129, costab16) * 2) - t130;
425  
426   t166 = MUL(t100 - t101, costab8);
427   t167 = MUL(t102 - t103, costab24);
428   t168 = t166 + t167;
429  
430   t134 = (t168 * 2) - t104;
431  
432   t120 = (t134 * 2) - t105;
433  
434   /* 14 */ hi[ 1][slot] = SHIFT(t120);
435  
436   t135 = (MUL(t118 - t119, costab16) * 2) - t120;
437  
438   /* 18 */ lo[ 2][slot] = SHIFT(t135);
439  
440   t169 = (MUL(t132 - t133, costab16) * 2) - t134;
441  
442   t151 = (t169 * 2) - t135;
443  
444   /* 22 */ lo[ 6][slot] = SHIFT(t151);
445  
446   t170 = (((MUL(t148 - t149, costab16) * 2) - t150) * 2) - t151;
447  
448   /* 26 */ lo[10][slot] = SHIFT(t170);
449   /* 30 */ lo[14][slot] =
450              SHIFT((((((MUL(t166 - t167, costab16) * 2) -
451                        t168) * 2) - t169) * 2) - t170);
452  
453   t171 = MUL(t106 - t107, costab8);
454   t172 = MUL(t108 - t109, costab24);
455   t173 = t171 + t172;
456  
457   t138 = (t173 * 2) - t110;
458  
459   t123 = (t138 * 2) - t111;
460  
461   t139 = (MUL(t121 - t122, costab16) * 2) - t123;
462  
463   t117 = (t123 * 2) - t112;
464  
465   /* 15 */ hi[ 0][slot] = SHIFT(t117);
466  
467   t124 = (MUL(t115 - t116, costab16) * 2) - t117;
468  
469   /* 17 */ lo[ 1][slot] = SHIFT(t124);
470  
471   t131 = (t139 * 2) - t124;
472  
473   /* 19 */ lo[ 3][slot] = SHIFT(t131);
474  
475   t140 = (t164 * 2) - t131;
476  
477   /* 21 */ lo[ 5][slot] = SHIFT(t140);
478  
479   t174 = (MUL(t136 - t137, costab16) * 2) - t138;
480  
481   t155 = (t174 * 2) - t139;
482  
483   t147 = (t155 * 2) - t140;
484  
485   /* 23 */ lo[ 7][slot] = SHIFT(t147);
486  
487   t156 = (((MUL(t144 - t145, costab16) * 2) - t146) * 2) - t147;
488  
489   /* 25 */ lo[ 9][slot] = SHIFT(t156);
490  
491   t175 = (((MUL(t152 - t153, costab16) * 2) - t154) * 2) - t155;
492  
493   t165 = (t175 * 2) - t156;
494  
495   /* 27 */ lo[11][slot] = SHIFT(t165);
496  
497   t176 = (((((MUL(t161 - t162, costab16) * 2) -
498              t163) * 2) - t164) * 2) - t165;
499  
500   /* 29 */ lo[13][slot] = SHIFT(t176);
501   /* 31 */ lo[15][slot] =
502              SHIFT((((((((MUL(t171 - t172, costab16) * 2) -
503                          t173) * 2) - t174) * 2) - t175) * 2) - t176);
504  
505   /*
506    * Totals:
507    *  80 multiplies
508    *  80 additions
509    * 119 subtractions
510    *  49 shifts (not counting SSO)
511    */
512 }
513  
514 # undef MUL
515 # undef SHIFT
516  
517 /* third SSO shift and/or D[] optimization preshift */
518  
519 # if defined(OPT_SSO)
520 #  if MAD_F_FRACBITS != 28
521 #   error "MAD_F_FRACBITS must be 28 to use OPT_SSO"
522 #  endif
523 #  define ML0(hi, lo, x, y)     ((lo)  = (x) * (y))
524 #  define MLA(hi, lo, x, y)     ((lo) += (x) * (y))
525 #  define MLN(hi, lo)           ((lo)  = -(lo))
526 #  define MLZ(hi, lo)           ((void) (hi), (mad_fixed_t) (lo))
527 #  define SHIFT(x)              ((x) >> 2)
528 #  define PRESHIFT(x)           ((MAD_F(x) + (1L << 13)) >> 14)
529 # else
530 #  define ML0(hi, lo, x, y)     MAD_F_ML0((hi), (lo), (x), (y))
531 #  define MLA(hi, lo, x, y)     MAD_F_MLA((hi), (lo), (x), (y))
532 #  define MLN(hi, lo)           MAD_F_MLN((hi), (lo))
533 #  define MLZ(hi, lo)           MAD_F_MLZ((hi), (lo))
534 #  define SHIFT(x)              (x)
535 #  if defined(MAD_F_SCALEBITS)
536 #   undef  MAD_F_SCALEBITS
537 #   define MAD_F_SCALEBITS      (MAD_F_FRACBITS - 12)
538 #   define PRESHIFT(x)          (MAD_F(x) >> 12)
539 #  else
540 #   define PRESHIFT(x)          MAD_F(x)
541 #  endif
542 # endif
543  
544 static
545 mad_fixed_t const D[17][32] = {
546 # include "D.dat"
547 };
548  
549 # if defined(ASO_SYNTH)
550 void synth_full(struct mad_synth *, struct mad_frame const *,
551                 unsigned int, unsigned int);
552 # else
553 /*
554  * NAME:        synth->full()
555  * DESCRIPTION: perform full frequency PCM synthesis
556  */
557 static
558 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
559                 unsigned int nch, unsigned int ns)
560 {
561   unsigned int phase, ch, s, sb, pe, po;
562   mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8];
563   mad_fixed_t const (*sbsample)[36][32];
564   register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
565   register mad_fixed_t const (*Dptr)[32], *ptr;
566   register mad_fixed64hi_t hi;
567   register mad_fixed64lo_t lo;
568  
569   for (ch = 0; ch < nch; ++ch) {
570     sbsample = &frame->sbsample[ch];
571     filter   = &synth->filter[ch];
572     phase    = synth->phase;
573     pcm1     = synth->pcm.samples[ch];
574  
575     for (s = 0; s < ns; ++s) {
576       dct32((*sbsample)[s], phase >> 1,
577             (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
578  
579       pe = phase & ~1;
580       po = ((phase - 1) & 0xf) | 1;
581  
582       /* calculate 32 samples */
583  
584       fe = &(*filter)[0][ phase & 1][0];
585       fx = &(*filter)[0][~phase & 1][0];
586       fo = &(*filter)[1][~phase & 1][0];
587  
588       Dptr = &D[0];
589  
590       ptr = *Dptr + po;
591       ML0(hi, lo, (*fx)[0], ptr[ 0]);
592       MLA(hi, lo, (*fx)[1], ptr[14]);
593       MLA(hi, lo, (*fx)[2], ptr[12]);
594       MLA(hi, lo, (*fx)[3], ptr[10]);
595       MLA(hi, lo, (*fx)[4], ptr[ 8]);
596       MLA(hi, lo, (*fx)[5], ptr[ 6]);
597       MLA(hi, lo, (*fx)[6], ptr[ 4]);
598       MLA(hi, lo, (*fx)[7], ptr[ 2]);
599       MLN(hi, lo);
600  
601       ptr = *Dptr + pe;
602       MLA(hi, lo, (*fe)[0], ptr[ 0]);
603       MLA(hi, lo, (*fe)[1], ptr[14]);
604       MLA(hi, lo, (*fe)[2], ptr[12]);
605       MLA(hi, lo, (*fe)[3], ptr[10]);
606       MLA(hi, lo, (*fe)[4], ptr[ 8]);
607       MLA(hi, lo, (*fe)[5], ptr[ 6]);
608       MLA(hi, lo, (*fe)[6], ptr[ 4]);
609       MLA(hi, lo, (*fe)[7], ptr[ 2]);
610  
611       *pcm1++ = SHIFT(MLZ(hi, lo));
612  
613       pcm2 = pcm1 + 30;
614  
615       for (sb = 1; sb < 16; ++sb) {
616         ++fe;
617         ++Dptr;
618  
619         /* D[32 - sb][i] == -D[sb][31 - i] */
620  
621         ptr = *Dptr + po;
622         ML0(hi, lo, (*fo)[0], ptr[ 0]);
623         MLA(hi, lo, (*fo)[1], ptr[14]);
624         MLA(hi, lo, (*fo)[2], ptr[12]);
625         MLA(hi, lo, (*fo)[3], ptr[10]);
626         MLA(hi, lo, (*fo)[4], ptr[ 8]);
627         MLA(hi, lo, (*fo)[5], ptr[ 6]);
628         MLA(hi, lo, (*fo)[6], ptr[ 4]);
629         MLA(hi, lo, (*fo)[7], ptr[ 2]);
630         MLN(hi, lo);
631  
632         ptr = *Dptr + pe;
633         MLA(hi, lo, (*fe)[7], ptr[ 2]);
634         MLA(hi, lo, (*fe)[6], ptr[ 4]);
635         MLA(hi, lo, (*fe)[5], ptr[ 6]);
636         MLA(hi, lo, (*fe)[4], ptr[ 8]);
637         MLA(hi, lo, (*fe)[3], ptr[10]);
638         MLA(hi, lo, (*fe)[2], ptr[12]);
639         MLA(hi, lo, (*fe)[1], ptr[14]);
640         MLA(hi, lo, (*fe)[0], ptr[ 0]);
641  
642         *pcm1++ = SHIFT(MLZ(hi, lo));
643  
644         ptr = *Dptr - pe;
645         ML0(hi, lo, (*fe)[0], ptr[31 - 16]);
646         MLA(hi, lo, (*fe)[1], ptr[31 - 14]);
647         MLA(hi, lo, (*fe)[2], ptr[31 - 12]);
648         MLA(hi, lo, (*fe)[3], ptr[31 - 10]);
649         MLA(hi, lo, (*fe)[4], ptr[31 -  8]);
650         MLA(hi, lo, (*fe)[5], ptr[31 -  6]);
651         MLA(hi, lo, (*fe)[6], ptr[31 -  4]);
652         MLA(hi, lo, (*fe)[7], ptr[31 -  2]);
653  
654         ptr = *Dptr - po;
655         MLA(hi, lo, (*fo)[7], ptr[31 -  2]);
656         MLA(hi, lo, (*fo)[6], ptr[31 -  4]);
657         MLA(hi, lo, (*fo)[5], ptr[31 -  6]);
658         MLA(hi, lo, (*fo)[4], ptr[31 -  8]);
659         MLA(hi, lo, (*fo)[3], ptr[31 - 10]);
660         MLA(hi, lo, (*fo)[2], ptr[31 - 12]);
661         MLA(hi, lo, (*fo)[1], ptr[31 - 14]);
662         MLA(hi, lo, (*fo)[0], ptr[31 - 16]);
663  
664         *pcm2-- = SHIFT(MLZ(hi, lo));
665  
666         ++fo;
667       }
668  
669       ++Dptr;
670  
671       ptr = *Dptr + po;
672       ML0(hi, lo, (*fo)[0], ptr[ 0]);
673       MLA(hi, lo, (*fo)[1], ptr[14]);
674       MLA(hi, lo, (*fo)[2], ptr[12]);
675       MLA(hi, lo, (*fo)[3], ptr[10]);
676       MLA(hi, lo, (*fo)[4], ptr[ 8]);
677       MLA(hi, lo, (*fo)[5], ptr[ 6]);
678       MLA(hi, lo, (*fo)[6], ptr[ 4]);
679       MLA(hi, lo, (*fo)[7], ptr[ 2]);
680  
681       *pcm1 = SHIFT(-MLZ(hi, lo));
682       pcm1 += 16;
683  
684       phase = (phase + 1) % 16;
685     }
686   }
687 }
688 # endif
689  
690 /*
691  * NAME:        synth->half()
692  * DESCRIPTION: perform half frequency PCM synthesis
693  */
694 static
695 void synth_half(struct mad_synth *synth, struct mad_frame const *frame,
696                 unsigned int nch, unsigned int ns)
697 {
698   unsigned int phase, ch, s, sb, pe, po;
699   mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8];
700   mad_fixed_t const (*sbsample)[36][32];
701   register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
702   register mad_fixed_t const (*Dptr)[32], *ptr;
703   register mad_fixed64hi_t hi;
704   register mad_fixed64lo_t lo;
705  
706   for (ch = 0; ch < nch; ++ch) {
707     sbsample = &frame->sbsample[ch];
708     filter   = &synth->filter[ch];
709     phase    = synth->phase;
710     pcm1     = synth->pcm.samples[ch];
711  
712     for (s = 0; s < ns; ++s) {
713       dct32((*sbsample)[s], phase >> 1,
714             (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
715  
716       pe = phase & ~1;
717       po = ((phase - 1) & 0xf) | 1;
718  
719       /* calculate 16 samples */
720  
721       fe = &(*filter)[0][ phase & 1][0];
722       fx = &(*filter)[0][~phase & 1][0];
723       fo = &(*filter)[1][~phase & 1][0];
724  
725       Dptr = &D[0];
726  
727       ptr = *Dptr + po;
728       ML0(hi, lo, (*fx)[0], ptr[ 0]);
729       MLA(hi, lo, (*fx)[1], ptr[14]);
730       MLA(hi, lo, (*fx)[2], ptr[12]);
731       MLA(hi, lo, (*fx)[3], ptr[10]);
732       MLA(hi, lo, (*fx)[4], ptr[ 8]);
733       MLA(hi, lo, (*fx)[5], ptr[ 6]);
734       MLA(hi, lo, (*fx)[6], ptr[ 4]);
735       MLA(hi, lo, (*fx)[7], ptr[ 2]);
736       MLN(hi, lo);
737  
738       ptr = *Dptr + pe;
739       MLA(hi, lo, (*fe)[0], ptr[ 0]);
740       MLA(hi, lo, (*fe)[1], ptr[14]);
741       MLA(hi, lo, (*fe)[2], ptr[12]);
742       MLA(hi, lo, (*fe)[3], ptr[10]);
743       MLA(hi, lo, (*fe)[4], ptr[ 8]);
744       MLA(hi, lo, (*fe)[5], ptr[ 6]);
745       MLA(hi, lo, (*fe)[6], ptr[ 4]);
746       MLA(hi, lo, (*fe)[7], ptr[ 2]);
747  
748       *pcm1++ = SHIFT(MLZ(hi, lo));
749  
750       pcm2 = pcm1 + 14;
751  
752       for (sb = 1; sb < 16; ++sb) {
753         ++fe;
754         ++Dptr;
755  
756         /* D[32 - sb][i] == -D[sb][31 - i] */
757  
758         if (!(sb & 1)) {
759           ptr = *Dptr + po;
760           ML0(hi, lo, (*fo)[0], ptr[ 0]);
761           MLA(hi, lo, (*fo)[1], ptr[14]);
762           MLA(hi, lo, (*fo)[2], ptr[12]);
763           MLA(hi, lo, (*fo)[3], ptr[10]);
764           MLA(hi, lo, (*fo)[4], ptr[ 8]);
765           MLA(hi, lo, (*fo)[5], ptr[ 6]);
766           MLA(hi, lo, (*fo)[6], ptr[ 4]);
767           MLA(hi, lo, (*fo)[7], ptr[ 2]);
768           MLN(hi, lo);
769  
770           ptr = *Dptr + pe;
771           MLA(hi, lo, (*fe)[7], ptr[ 2]);
772           MLA(hi, lo, (*fe)[6], ptr[ 4]);
773           MLA(hi, lo, (*fe)[5], ptr[ 6]);
774           MLA(hi, lo, (*fe)[4], ptr[ 8]);
775           MLA(hi, lo, (*fe)[3], ptr[10]);
776           MLA(hi, lo, (*fe)[2], ptr[12]);
777           MLA(hi, lo, (*fe)[1], ptr[14]);
778           MLA(hi, lo, (*fe)[0], ptr[ 0]);
779  
780           *pcm1++ = SHIFT(MLZ(hi, lo));
781  
782           ptr = *Dptr - po;
783           ML0(hi, lo, (*fo)[7], ptr[31 -  2]);
784           MLA(hi, lo, (*fo)[6], ptr[31 -  4]);
785           MLA(hi, lo, (*fo)[5], ptr[31 -  6]);
786           MLA(hi, lo, (*fo)[4], ptr[31 -  8]);
787           MLA(hi, lo, (*fo)[3], ptr[31 - 10]);
788           MLA(hi, lo, (*fo)[2], ptr[31 - 12]);
789           MLA(hi, lo, (*fo)[1], ptr[31 - 14]);
790           MLA(hi, lo, (*fo)[0], ptr[31 - 16]);
791  
792           ptr = *Dptr - pe;
793           MLA(hi, lo, (*fe)[0], ptr[31 - 16]);
794           MLA(hi, lo, (*fe)[1], ptr[31 - 14]);
795           MLA(hi, lo, (*fe)[2], ptr[31 - 12]);
796           MLA(hi, lo, (*fe)[3], ptr[31 - 10]);
797           MLA(hi, lo, (*fe)[4], ptr[31 -  8]);
798           MLA(hi, lo, (*fe)[5], ptr[31 -  6]);
799           MLA(hi, lo, (*fe)[6], ptr[31 -  4]);
800           MLA(hi, lo, (*fe)[7], ptr[31 -  2]);
801  
802           *pcm2-- = SHIFT(MLZ(hi, lo));
803         }
804  
805         ++fo;
806       }
807  
808       ++Dptr;
809  
810       ptr = *Dptr + po;
811       ML0(hi, lo, (*fo)[0], ptr[ 0]);
812       MLA(hi, lo, (*fo)[1], ptr[14]);
813       MLA(hi, lo, (*fo)[2], ptr[12]);
814       MLA(hi, lo, (*fo)[3], ptr[10]);
815       MLA(hi, lo, (*fo)[4], ptr[ 8]);
816       MLA(hi, lo, (*fo)[5], ptr[ 6]);
817       MLA(hi, lo, (*fo)[6], ptr[ 4]);
818       MLA(hi, lo, (*fo)[7], ptr[ 2]);
819  
820       *pcm1 = SHIFT(-MLZ(hi, lo));
821       pcm1 += 8;
822  
823       phase = (phase + 1) % 16;
824     }
825   }
826 }
827  
828 /*
829  * NAME:        synth->frame()
830  * DESCRIPTION: perform PCM synthesis of frame subband samples
831  */
832 void mad_synth_frame(struct mad_synth *synth, struct mad_frame const *frame)
833 {
834   unsigned int nch, ns;
835   void (*synth_frame)(struct mad_synth *, struct mad_frame const *,
836                       unsigned int, unsigned int);
837  
838   nch = MAD_NCHANNELS(&frame->header);
839   ns  = MAD_NSBSAMPLES(&frame->header);
840  
841   synth->pcm.samplerate = frame->header.samplerate;
842   synth->pcm.channels   = nch;
843   synth->pcm.length     = 32 * ns;
844  
845   synth_frame = synth_full;
846  
847   if (frame->options & MAD_OPTION_HALFSAMPLERATE) {
848     synth->pcm.samplerate /= 2;
849     synth->pcm.length     /= 2;
850  
851     synth_frame = synth_half;
852   }
853  
854   synth_frame(synth, frame, nch, ns);
855  
856   synth->phase = (synth->phase + ns) % 16;
857 }