?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
lcd-m6sp.c: remove \r
[kugel-rb.git] / apps / codecs / libmad / synth.c
blob5ae9811ead356fbf18b66543566ba57e80d90b9c
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 * $Id$
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 #if defined(CPU_COLDFIRE)
46 /* init the emac unit here, since this function should always be called
47 before using libmad */
48 coldfire_set_macsr(EMAC_FRACTIONAL | EMAC_SATURATE | EMAC_ROUND);
49 #endif
50 }
51
52 /*
53 * NAME: synth->mute()
54 * DESCRIPTION: zero all polyphase filterbank values, resetting synthesis
55 */
56 void mad_synth_mute(struct mad_synth *synth)
57 {
58 unsigned int ch, s, v;
59
60 for (ch = 0; ch < 2; ++ch) {
61 for (s = 0; s < 16; ++s) {
62 for (v = 0; v < 8; ++v) {
63 synth->filter[ch][0][0][s][v] = synth->filter[ch][0][1][s][v] =
64 synth->filter[ch][1][0][s][v] = synth->filter[ch][1][1][s][v] = 0;
65 }
66 }
67 }
68 }
69
70 #if 0 /* dct32 asm implementation is slower on current arm systems */
71 /* #ifdef FPM_ARM */
72
73 void dct32(mad_fixed_t const in[32], unsigned int slot,
74 mad_fixed_t lo[16][8], mad_fixed_t hi[16][8]);
75
76 #else
77
78 /*
79 * An optional optimization called here the Subband Synthesis Optimization
80 * (SSO) improves the performance of subband synthesis at the expense of
81 * accuracy.
82 *
83 * The idea is to simplify 32x32->64-bit multiplication to 32x32->32 such
84 * that extra scaling and rounding are not necessary. This often allows the
85 * compiler to use faster 32-bit multiply-accumulate instructions instead of
86 * explicit 64-bit multiply, shift, and add instructions.
87 *
88 * SSO works like this: a full 32x32->64-bit multiply of two mad_fixed_t
89 * values requires the result to be right-shifted 28 bits to be properly
90 * scaled to the same fixed-point format. Right shifts can be applied at any
91 * time to either operand or to the result, so the optimization involves
92 * careful placement of these shifts to minimize the loss of accuracy.
93 *
94 * First, a 14-bit shift is applied with rounding at compile-time to the D[]
95 * table of coefficients for the subband synthesis window. This only loses 2
96 * bits of accuracy because the lower 12 bits are always zero. A second
97 * 12-bit shift occurs after the DCT calculation. This loses 12 bits of
98 * accuracy. Finally, a third 2-bit shift occurs just before the sample is
99 * saved in the PCM buffer. 14 + 12 + 2 == 28 bits.
100 */
101
102 /* FPM_DEFAULT without OPT_SSO will actually lose accuracy and performance */
103
104 # if defined(FPM_DEFAULT) && !defined(OPT_SSO)
105 # define OPT_SSO
106 # endif
107
108 /* second SSO shift, with rounding */
109
110 # if defined(OPT_SSO)
111 # define SHIFT(x) (((x) + (1L << 11)) >> 12)
112 # else
113 # define SHIFT(x) (x)
114 # endif
115
116 /* possible DCT speed optimization */
117
118 /* This is a Coldfire version of the OPT_SPEED optimisation below, but in the
119 case of Coldfire it doesn't lose any more precision than we would ordinarily
120 lose, */
121 # ifdef FPM_COLDFIRE_EMAC
122 # define OPT_DCTO
123 # define MUL(x, y) \
124 ({ \
125 mad_fixed64hi_t hi; \
126 asm volatile("mac.l %[a], %[b], %%acc0\n\t" \
127 "movclr.l %%acc0, %[hi]" \
128 : [hi] "=r" (hi) \
129 : [a] "r" ((x)), [b] "r" ((y))); \
130 hi; \
131 })
132 # elif defined(OPT_SPEED) && defined(MAD_F_MLX)
133 # define OPT_DCTO
134 # define MUL(x, y) \
135 ({ mad_fixed64hi_t hi; \
136 mad_fixed64lo_t lo; \
137 MAD_F_MLX(hi, lo, (x), (y)); \
138 hi << (32 - MAD_F_SCALEBITS - 3); \
139 })
140 # else
141 # undef OPT_DCTO
142 # define MUL(x, y) mad_f_mul((x), (y))
143 # endif
144
145 /*
146 * NAME: dct32()
147 * DESCRIPTION: perform fast in[32]->out[32] DCT
148 */
149 static
150 void dct32(mad_fixed_t const in[32], unsigned int slot,
151 mad_fixed_t lo[16][8], mad_fixed_t hi[16][8])
152 {
153 mad_fixed_t t0, t1, t2, t3, t4, t5, t6, t7;
154 mad_fixed_t t8, t9, t10, t11, t12, t13, t14, t15;
155 mad_fixed_t t16, t17, t18, t19, t20, t21, t22, t23;
156 mad_fixed_t t24, t25, t26, t27, t28, t29, t30, t31;
157 mad_fixed_t t32, t33, t34, t35, t36, t37, t38, t39;
158 mad_fixed_t t40, t41, t42, t43, t44, t45, t46, t47;
159 mad_fixed_t t48, t49, t50, t51, t52, t53, t54, t55;
160 mad_fixed_t t56, t57, t58, t59, t60, t61, t62, t63;
161 mad_fixed_t t64, t65, t66, t67, t68, t69, t70, t71;
162 mad_fixed_t t72, t73, t74, t75, t76, t77, t78, t79;
163 mad_fixed_t t80, t81, t82, t83, t84, t85, t86, t87;
164 mad_fixed_t t88, t89, t90, t91, t92, t93, t94, t95;
165 mad_fixed_t t96, t97, t98, t99, t100, t101, t102, t103;
166 mad_fixed_t t104, t105, t106, t107, t108, t109, t110, t111;
167 mad_fixed_t t112, t113, t114, t115, t116, t117, t118, t119;
168 mad_fixed_t t120, t121, t122, t123, t124, t125, t126, t127;
169 mad_fixed_t t128, t129, t130, t131, t132, t133, t134, t135;
170 mad_fixed_t t136, t137, t138, t139, t140, t141, t142, t143;
171 mad_fixed_t t144, t145, t146, t147, t148, t149, t150, t151;
172 mad_fixed_t t152, t153, t154, t155, t156, t157, t158, t159;
173 mad_fixed_t t160, t161, t162, t163, t164, t165, t166, t167;
174 mad_fixed_t t168, t169, t170, t171, t172, t173, t174, t175;
175 mad_fixed_t t176;
176
177 /* costab[i] = cos(PI / (2 * 32) * i) */
178
179 # if defined(OPT_DCTO)
180 # define costab1 MAD_F(0x7fd8878e)
181 # define costab2 MAD_F(0x7f62368f)
182 # define costab3 MAD_F(0x7e9d55fc)
183 # define costab4 MAD_F(0x7d8a5f40)
184 # define costab5 MAD_F(0x7c29fbee)
185 # define costab6 MAD_F(0x7a7d055b)
186 # define costab7 MAD_F(0x78848414)
187 # define costab8 MAD_F(0x7641af3d)
188 # define costab9 MAD_F(0x73b5ebd1)
189 # define costab10 MAD_F(0x70e2cbc6)
190 # define costab11 MAD_F(0x6dca0d14)
191 # define costab12 MAD_F(0x6a5d98a4)
192 # define costab13 MAD_F(0x66cf8120)
193 # define costab14 MAD_F(0x62f201ac)
194 # define costab15 MAD_F(0x5ed77c8a)
195 # define costab16 MAD_F(0x5a82799a)
196 # define costab17 MAD_F(0x55f5a4d2)
197 # define costab18 MAD_F(0x5133cc94)
198 # define costab19 MAD_F(0x4c3fdff4)
199 # define costab20 MAD_F(0x471cece7)
200 # define costab21 MAD_F(0x41ce1e65)
201 # define costab22 MAD_F(0x3c56ba70)
202 # define costab23 MAD_F(0x36ba2014)
203 # define costab24 MAD_F(0x30fbc54d)
204 # define costab25 MAD_F(0x2b1f34eb)
205 # define costab26 MAD_F(0x25280c5e)
206 # define costab27 MAD_F(0x1f19f97b)
207 # define costab28 MAD_F(0x18f8b83c)
208 # define costab29 MAD_F(0x12c8106f)
209 # define costab30 MAD_F(0x0c8bd35e)
210 # define costab31 MAD_F(0x0647d97c)
211 # else
212 # define costab1 MAD_F(0x0ffb10f2) /* 0.998795456 */
213 # define costab2 MAD_F(0x0fec46d2) /* 0.995184727 */
214 # define costab3 MAD_F(0x0fd3aac0) /* 0.989176510 */
215 # define costab4 MAD_F(0x0fb14be8) /* 0.980785280 */
216 # define costab5 MAD_F(0x0f853f7e) /* 0.970031253 */
217 # define costab6 MAD_F(0x0f4fa0ab) /* 0.956940336 */
218 # define costab7 MAD_F(0x0f109082) /* 0.941544065 */
219 # define costab8 MAD_F(0x0ec835e8) /* 0.923879533 */
220 # define costab9 MAD_F(0x0e76bd7a) /* 0.903989293 */
221 # define costab10 MAD_F(0x0e1c5979) /* 0.881921264 */
222 # define costab11 MAD_F(0x0db941a3) /* 0.857728610 */
223 # define costab12 MAD_F(0x0d4db315) /* 0.831469612 */
224 # define costab13 MAD_F(0x0cd9f024) /* 0.803207531 */
225 # define costab14 MAD_F(0x0c5e4036) /* 0.773010453 */
226 # define costab15 MAD_F(0x0bdaef91) /* 0.740951125 */
227 # define costab16 MAD_F(0x0b504f33) /* 0.707106781 */
228 # define costab17 MAD_F(0x0abeb49a) /* 0.671558955 */
229 # define costab18 MAD_F(0x0a267993) /* 0.634393284 */
230 # define costab19 MAD_F(0x0987fbfe) /* 0.595699304 */
231 # define costab20 MAD_F(0x08e39d9d) /* 0.555570233 */
232 # define costab21 MAD_F(0x0839c3cd) /* 0.514102744 */
233 # define costab22 MAD_F(0x078ad74e) /* 0.471396737 */
234 # define costab23 MAD_F(0x06d74402) /* 0.427555093 */
235 # define costab24 MAD_F(0x061f78aa) /* 0.382683432 */
236 # define costab25 MAD_F(0x0563e69d) /* 0.336889853 */
237 # define costab26 MAD_F(0x04a5018c) /* 0.290284677 */
238 # define costab27 MAD_F(0x03e33f2f) /* 0.242980180 */
239 # define costab28 MAD_F(0x031f1708) /* 0.195090322 */
240 # define costab29 MAD_F(0x0259020e) /* 0.146730474 */
241 # define costab30 MAD_F(0x01917a5c) /* 0.098017140 */
242 # define costab31 MAD_F(0x00c8fb30) /* 0.049067674 */
243 # endif
244
245 t0 = in[0] + in[31]; t16 = MUL(in[0] - in[31], costab1);
246 t1 = in[15] + in[16]; t17 = MUL(in[15] - in[16], costab31);
247
248 t41 = t16 + t17;
249 t59 = MUL(t16 - t17, costab2);
250 t33 = t0 + t1;
251 t50 = MUL(t0 - t1, costab2);
252
253 t2 = in[7] + in[24]; t18 = MUL(in[7] - in[24], costab15);
254 t3 = in[8] + in[23]; t19 = MUL(in[8] - in[23], costab17);
255
256 t42 = t18 + t19;
257 t60 = MUL(t18 - t19, costab30);
258 t34 = t2 + t3;
259 t51 = MUL(t2 - t3, costab30);
260
261 t4 = in[3] + in[28]; t20 = MUL(in[3] - in[28], costab7);
262 t5 = in[12] + in[19]; t21 = MUL(in[12] - in[19], costab25);
263
264 t43 = t20 + t21;
265 t61 = MUL(t20 - t21, costab14);
266 t35 = t4 + t5;
267 t52 = MUL(t4 - t5, costab14);
268
269 t6 = in[4] + in[27]; t22 = MUL(in[4] - in[27], costab9);
270 t7 = in[11] + in[20]; t23 = MUL(in[11] - in[20], costab23);
271
272 t44 = t22 + t23;
273 t62 = MUL(t22 - t23, costab18);
274 t36 = t6 + t7;
275 t53 = MUL(t6 - t7, costab18);
276
277 t8 = in[1] + in[30]; t24 = MUL(in[1] - in[30], costab3);
278 t9 = in[14] + in[17]; t25 = MUL(in[14] - in[17], costab29);
279
280 t45 = t24 + t25;
281 t63 = MUL(t24 - t25, costab6);
282 t37 = t8 + t9;
283 t54 = MUL(t8 - t9, costab6);
284
285 t10 = in[6] + in[25]; t26 = MUL(in[6] - in[25], costab13);
286 t11 = in[9] + in[22]; t27 = MUL(in[9] - in[22], costab19);
287
288 t46 = t26 + t27;
289 t64 = MUL(t26 - t27, costab26);
290 t38 = t10 + t11;
291 t55 = MUL(t10 - t11, costab26);
292
293 t12 = in[2] + in[29]; t28 = MUL(in[2] - in[29], costab5);
294 t13 = in[13] + in[18]; t29 = MUL(in[13] - in[18], costab27);
295
296 t47 = t28 + t29;
297 t65 = MUL(t28 - t29, costab10);
298 t39 = t12 + t13;
299 t56 = MUL(t12 - t13, costab10);
300
301 t14 = in[5] + in[26]; t30 = MUL(in[5] - in[26], costab11);
302 t15 = in[10] + in[21]; t31 = MUL(in[10] - in[21], costab21);
303
304 t48 = t30 + t31;
305 t66 = MUL(t30 - t31, costab22);
306 t40 = t14 + t15;
307 t57 = MUL(t14 - t15, costab22);
308
309 t69 = t33 + t34; t89 = MUL(t33 - t34, costab4);
310 t70 = t35 + t36; t90 = MUL(t35 - t36, costab28);
311 t71 = t37 + t38; t91 = MUL(t37 - t38, costab12);
312 t72 = t39 + t40; t92 = MUL(t39 - t40, costab20);
313 t73 = t41 + t42; t94 = MUL(t41 - t42, costab4);
314 t74 = t43 + t44; t95 = MUL(t43 - t44, costab28);
315 t75 = t45 + t46; t96 = MUL(t45 - t46, costab12);
316 t76 = t47 + t48; t97 = MUL(t47 - t48, costab20);
317
318 t78 = t50 + t51; t100 = MUL(t50 - t51, costab4);
319 t79 = t52 + t53; t101 = MUL(t52 - t53, costab28);
320 t80 = t54 + t55; t102 = MUL(t54 - t55, costab12);
321 t81 = t56 + t57; t103 = MUL(t56 - t57, costab20);
322
323 t83 = t59 + t60; t106 = MUL(t59 - t60, costab4);
324 t84 = t61 + t62; t107 = MUL(t61 - t62, costab28);
325 t85 = t63 + t64; t108 = MUL(t63 - t64, costab12);
326 t86 = t65 + t66; t109 = MUL(t65 - t66, costab20);
327
328 t113 = t69 + t70;
329 t114 = t71 + t72;
330
331 /* 0 */ hi[15][slot] = SHIFT(t113 + t114);
332 /* 16 */ lo[ 0][slot] = SHIFT(MUL(t113 - t114, costab16));
333
334 t115 = t73 + t74;
335 t116 = t75 + t76;
336
337 t32 = t115 + t116;
338
339 /* 1 */ hi[14][slot] = SHIFT(t32);
340
341 t118 = t78 + t79;
342 t119 = t80 + t81;
343
344 t58 = t118 + t119;
345
346 /* 2 */ hi[13][slot] = SHIFT(t58);
347
348 t121 = t83 + t84;
349 t122 = t85 + t86;
350
351 t67 = t121 + t122;
352
353 t49 = (t67 * 2) - t32;
354
355 /* 3 */ hi[12][slot] = SHIFT(t49);
356
357 t125 = t89 + t90;
358 t126 = t91 + t92;
359
360 t93 = t125 + t126;
361
362 /* 4 */ hi[11][slot] = SHIFT(t93);
363
364 t128 = t94 + t95;
365 t129 = t96 + t97;
366
367 t98 = t128 + t129;
368
369 t68 = (t98 * 2) - t49;
370
371 /* 5 */ hi[10][slot] = SHIFT(t68);
372
373 t132 = t100 + t101;
374 t133 = t102 + t103;
375
376 t104 = t132 + t133;
377
378 t82 = (t104 * 2) - t58;
379
380 /* 6 */ hi[ 9][slot] = SHIFT(t82);
381
382 t136 = t106 + t107;
383 t137 = t108 + t109;
384
385 t110 = t136 + t137;
386
387 t87 = (t110 * 2) - t67;
388
389 t77 = (t87 * 2) - t68;
390
391 /* 7 */ hi[ 8][slot] = SHIFT(t77);
392
393 t141 = MUL(t69 - t70, costab8);
394 t142 = MUL(t71 - t72, costab24);
395 t143 = t141 + t142;
396
397 /* 8 */ hi[ 7][slot] = SHIFT(t143);
398 /* 24 */ lo[ 8][slot] =
399 SHIFT((MUL(t141 - t142, costab16) * 2) - t143);
400
401 t144 = MUL(t73 - t74, costab8);
402 t145 = MUL(t75 - t76, costab24);
403 t146 = t144 + t145;
404
405 t88 = (t146 * 2) - t77;
406
407 /* 9 */ hi[ 6][slot] = SHIFT(t88);
408
409 t148 = MUL(t78 - t79, costab8);
410 t149 = MUL(t80 - t81, costab24);
411 t150 = t148 + t149;
412
413 t105 = (t150 * 2) - t82;
414
415 /* 10 */ hi[ 5][slot] = SHIFT(t105);
416
417 t152 = MUL(t83 - t84, costab8);
418 t153 = MUL(t85 - t86, costab24);
419 t154 = t152 + t153;
420
421 t111 = (t154 * 2) - t87;
422
423 t99 = (t111 * 2) - t88;
424
425 /* 11 */ hi[ 4][slot] = SHIFT(t99);
426
427 t157 = MUL(t89 - t90, costab8);
428 t158 = MUL(t91 - t92, costab24);
429 t159 = t157 + t158;
430
431 t127 = (t159 * 2) - t93;
432
433 /* 12 */ hi[ 3][slot] = SHIFT(t127);
434
435 t160 = (MUL(t125 - t126, costab16) * 2) - t127;
436
437 /* 20 */ lo[ 4][slot] = SHIFT(t160);
438 /* 28 */ lo[12][slot] =
439 SHIFT((((MUL(t157 - t158, costab16) * 2) - t159) * 2) - t160);
440
441 t161 = MUL(t94 - t95, costab8);
442 t162 = MUL(t96 - t97, costab24);
443 t163 = t161 + t162;
444
445 t130 = (t163 * 2) - t98;
446
447 t112 = (t130 * 2) - t99;
448
449 /* 13 */ hi[ 2][slot] = SHIFT(t112);
450
451 t164 = (MUL(t128 - t129, costab16) * 2) - t130;
452
453 t166 = MUL(t100 - t101, costab8);
454 t167 = MUL(t102 - t103, costab24);
455 t168 = t166 + t167;
456
457 t134 = (t168 * 2) - t104;
458
459 t120 = (t134 * 2) - t105;
460
461 /* 14 */ hi[ 1][slot] = SHIFT(t120);
462
463 t135 = (MUL(t118 - t119, costab16) * 2) - t120;
464
465 /* 18 */ lo[ 2][slot] = SHIFT(t135);
466
467 t169 = (MUL(t132 - t133, costab16) * 2) - t134;
468
469 t151 = (t169 * 2) - t135;
470
471 /* 22 */ lo[ 6][slot] = SHIFT(t151);
472
473 t170 = (((MUL(t148 - t149, costab16) * 2) - t150) * 2) - t151;
474
475 /* 26 */ lo[10][slot] = SHIFT(t170);
476 /* 30 */ lo[14][slot] =
477 SHIFT((((((MUL(t166 - t167, costab16) * 2) -
478 t168) * 2) - t169) * 2) - t170);
479
480 t171 = MUL(t106 - t107, costab8);
481 t172 = MUL(t108 - t109, costab24);
482 t173 = t171 + t172;
483
484 t138 = (t173 * 2) - t110;
485
486 t123 = (t138 * 2) - t111;
487
488 t139 = (MUL(t121 - t122, costab16) * 2) - t123;
489
490 t117 = (t123 * 2) - t112;
491
492 /* 15 */ hi[ 0][slot] = SHIFT(t117);
493
494 t124 = (MUL(t115 - t116, costab16) * 2) - t117;
495
496 /* 17 */ lo[ 1][slot] = SHIFT(t124);
497
498 t131 = (t139 * 2) - t124;
499
500 /* 19 */ lo[ 3][slot] = SHIFT(t131);
501
502 t140 = (t164 * 2) - t131;
503
504 /* 21 */ lo[ 5][slot] = SHIFT(t140);
505
506 t174 = (MUL(t136 - t137, costab16) * 2) - t138;
507
508 t155 = (t174 * 2) - t139;
509
510 t147 = (t155 * 2) - t140;
511
512 /* 23 */ lo[ 7][slot] = SHIFT(t147);
513
514 t156 = (((MUL(t144 - t145, costab16) * 2) - t146) * 2) - t147;
515
516 /* 25 */ lo[ 9][slot] = SHIFT(t156);
517
518 t175 = (((MUL(t152 - t153, costab16) * 2) - t154) * 2) - t155;
519
520 t165 = (t175 * 2) - t156;
521
522 /* 27 */ lo[11][slot] = SHIFT(t165);
523
524 t176 = (((((MUL(t161 - t162, costab16) * 2) -
525 t163) * 2) - t164) * 2) - t165;
526
527 /* 29 */ lo[13][slot] = SHIFT(t176);
528 /* 31 */ lo[15][slot] =
529 SHIFT((((((((MUL(t171 - t172, costab16) * 2) -
530 t173) * 2) - t174) * 2) - t175) * 2) - t176);
531
532 /*
533 * Totals:
534 * 80 multiplies
535 * 80 additions
536 * 119 subtractions
537 * 49 shifts (not counting SSO)
538 */
539 }
540
541 # undef MUL
542 # undef SHIFT
543
544 #endif
545
546 /* third SSO shift and/or D[] optimization preshift */
547
548 # if defined(OPT_SSO)
549 # if MAD_F_FRACBITS != 28
550 # error "MAD_F_FRACBITS must be 28 to use OPT_SSO"
551 # endif
552 # define ML0(hi, lo, x, y) ((lo) = (x) * (y))
553 # define MLA(hi, lo, x, y) ((lo) += (x) * (y))
554 # define MLN(hi, lo) ((lo) = -(lo))
555 # define MLZ(hi, lo) ((void) (hi), (mad_fixed_t) (lo))
556 # define SHIFT(x) ((x) >> 2)
557 # define PRESHIFT(x) ((MAD_F(x) + (1L << 13)) >> 14)
558 # else
559 # define ML0(hi, lo, x, y) MAD_F_ML0((hi), (lo), (x), (y))
560 # define MLA(hi, lo, x, y) MAD_F_MLA((hi), (lo), (x), (y))
561 # define MLN(hi, lo) MAD_F_MLN((hi), (lo))
562 # define MLZ(hi, lo) MAD_F_MLZ((hi), (lo))
563 # define SHIFT(x) (x)
564 # if defined(MAD_F_SCALEBITS)
565 # undef MAD_F_SCALEBITS
566 # define MAD_F_SCALEBITS (MAD_F_FRACBITS - 12)
567 # define PRESHIFT(x) (MAD_F(x) >> 12)
568 # else
569 # define PRESHIFT(x) MAD_F(x)
570 # endif
571 # endif
572
573 static
574 mad_fixed_t const D[17][32] ICONST_ATTR = {
575 # include "D.dat"
576 };
577
578 # if defined(ASO_SYNTH)
579 void synth_full(struct mad_synth *, struct mad_frame const *,
580 unsigned int, unsigned int);
581 # else
582 /*
583 * NAME: synth->full()
584 * DESCRIPTION: perform full frequency PCM synthesis
585 */
586
587 /* optimised version of synth_full */
588 # ifdef FPM_COLDFIRE_EMAC
589 static
590 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
591 unsigned int nch, unsigned int ns)
592 {
593 int sb;
594 unsigned int phase, ch, s, p;
595 mad_fixed_t *pcm, (*filter)[2][2][16][8];
596 mad_fixed_t (*sbsample)[36][32];
597 mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
598 mad_fixed_t const (*D0ptr)[32];
599 mad_fixed_t const (*D1ptr)[32];
600 mad_fixed64hi_t hi0, hi1;
601
602 for (ch = 0; ch < nch; ++ch) {
603 sbsample = &(*frame->sbsample_prev)[ch];
604 filter = &synth->filter[ch];
605 phase = synth->phase;
606 pcm = synth->pcm.samples[ch];
607
608 for (s = 0; s < ns; ++s) {
609 dct32((*sbsample)[s], phase >> 1,
610 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
611
612 p = (phase - 1) & 0xf;
613
614 /* calculate 32 samples */
615 fe = &(*filter)[0][ phase & 1][0];
616 fx = &(*filter)[0][~phase & 1][0];
617 fo = &(*filter)[1][~phase & 1][0];
618
619 D0ptr = (void*)&D[0][ p];
620 D1ptr = (void*)&D[0][-p];
621
622 if(s & 1)
623 {
624 asm volatile(
625 "movem.l (%1), %%d0-%%d7\n\t"
626 "move.l 4(%2), %%a5\n\t"
627 "msac.l %%d0, %%a5, 60(%2), %%a5, %%acc0\n\t"
628 "msac.l %%d1, %%a5, 52(%2), %%a5, %%acc0\n\t"
629 "msac.l %%d2, %%a5, 44(%2), %%a5, %%acc0\n\t"
630 "msac.l %%d3, %%a5, 36(%2), %%a5, %%acc0\n\t"
631 "msac.l %%d4, %%a5, 28(%2), %%a5, %%acc0\n\t"
632 "msac.l %%d5, %%a5, 20(%2), %%a5, %%acc0\n\t"
633 "msac.l %%d6, %%a5, 12(%2), %%a5, %%acc0\n\t"
634 "msac.l %%d7, %%a5, (%2), %%a5, %%acc0\n\t"
635
636 "movem.l (%3), %%d0-%%d7\n\t"
637 "mac.l %%d0, %%a5, 56(%2), %%a5, %%acc0\n\t"
638 "mac.l %%d1, %%a5, 48(%2), %%a5, %%acc0\n\t"
639 "mac.l %%d2, %%a5, 40(%2), %%a5, %%acc0\n\t"
640 "mac.l %%d3, %%a5, 32(%2), %%a5, %%acc0\n\t"
641 "mac.l %%d4, %%a5, 24(%2), %%a5, %%acc0\n\t"
642 "mac.l %%d5, %%a5, 16(%2), %%a5, %%acc0\n\t"
643 "mac.l %%d6, %%a5, 8(%2), %%a5, %%acc0\n\t"
644 "mac.l %%d7, %%a5, %%acc0\n\t"
645 "movclr.l %%acc0, %0\n\t"
646 : "=r" (hi0) : "a" (*fx), "a" (*D0ptr), "a" (*fe)
647 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5");
648
649 pcm[0] = hi0 << 3; /* shift result to libmad's fixed point format */
650 pcm += 16;
651
652 for (sb = 15; sb; sb--, fo++) {
653 ++fe;
654 ++D0ptr;
655 ++D1ptr;
656
657 /* D[32 - sb][i] == -D[sb][31 - i] */
658 asm volatile (
659 "movem.l (%0), %%d0-%%d7\n\t"
660 "move.l 4(%2), %%a5\n\t"
661 "msac.l %%d0, %%a5, 60(%2), %%a5, %%acc0\n\t"
662 "msac.l %%d1, %%a5, 52(%2), %%a5, %%acc0\n\t"
663 "msac.l %%d2, %%a5, 44(%2), %%a5, %%acc0\n\t"
664 "msac.l %%d3, %%a5, 36(%2), %%a5, %%acc0\n\t"
665 "msac.l %%d4, %%a5, 28(%2), %%a5, %%acc0\n\t"
666 "msac.l %%d5, %%a5, 20(%2), %%a5, %%acc0\n\t"
667 "msac.l %%d6, %%a5, 12(%2), %%a5, %%acc0\n\t"
668 "msac.l %%d7, %%a5, 112(%3), %%a5, %%acc0\n\t"
669 "mac.l %%d7, %%a5, 104(%3), %%a5, %%acc1\n\t"
670 "mac.l %%d6, %%a5, 96(%3), %%a5, %%acc1\n\t"
671 "mac.l %%d5, %%a5, 88(%3), %%a5, %%acc1\n\t"
672 "mac.l %%d4, %%a5, 80(%3), %%a5, %%acc1\n\t"
673 "mac.l %%d3, %%a5, 72(%3), %%a5, %%acc1\n\t"
674 "mac.l %%d2, %%a5, 64(%3), %%a5, %%acc1\n\t"
675 "mac.l %%d1, %%a5, 120(%3), %%a5, %%acc1\n\t"
676 "mac.l %%d0, %%a5, 8(%2), %%a5, %%acc1\n\t"
677 "movem.l (%1), %%d0-%%d7\n\t"
678 "mac.l %%d7, %%a5, 16(%2), %%a5, %%acc0\n\t"
679 "mac.l %%d6, %%a5, 24(%2), %%a5, %%acc0\n\t"
680 "mac.l %%d5, %%a5, 32(%2), %%a5, %%acc0\n\t"
681 "mac.l %%d4, %%a5, 40(%2), %%a5, %%acc0\n\t"
682 "mac.l %%d3, %%a5, 48(%2), %%a5, %%acc0\n\t"
683 "mac.l %%d2, %%a5, 56(%2), %%a5, %%acc0\n\t"
684 "mac.l %%d1, %%a5, (%2), %%a5, %%acc0\n\t"
685 "mac.l %%d0, %%a5, 60(%3), %%a5, %%acc0\n\t"
686 "mac.l %%d0, %%a5, 68(%3), %%a5, %%acc1\n\t"
687 "mac.l %%d1, %%a5, 76(%3), %%a5, %%acc1\n\t"
688 "mac.l %%d2, %%a5, 84(%3), %%a5, %%acc1\n\t"
689 "mac.l %%d3, %%a5, 92(%3), %%a5, %%acc1\n\t"
690 "mac.l %%d4, %%a5, 100(%3), %%a5, %%acc1\n\t"
691 "mac.l %%d5, %%a5, 108(%3), %%a5, %%acc1\n\t"
692 "mac.l %%d6, %%a5, 116(%3), %%a5, %%acc1\n\t"
693 "mac.l %%d7, %%a5, %%acc1\n\t"
694 : : "a" (*fo), "a" (*fe), "a" (*D0ptr), "a" (*D1ptr)
695 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5");
696
697 asm volatile(
698 "movclr.l %%acc0, %0\n\t"
699 "movclr.l %%acc1, %1\n\t" : "=d" (hi0), "=d" (hi1) );
700
701 pcm[-sb] = hi0 << 3;
702 pcm[ sb] = hi1 << 3;
703 }
704
705 ++D0ptr;
706 asm volatile(
707 "movem.l (%1), %%d0-%%d7\n\t"
708 "move.l 4(%2), %%a5\n\t"
709 "mac.l %%d0, %%a5, 60(%2), %%a5, %%acc0\n\t"
710 "mac.l %%d1, %%a5, 52(%2), %%a5, %%acc0\n\t"
711 "mac.l %%d2, %%a5, 44(%2), %%a5, %%acc0\n\t"
712 "mac.l %%d3, %%a5, 36(%2), %%a5, %%acc0\n\t"
713 "mac.l %%d4, %%a5, 28(%2), %%a5, %%acc0\n\t"
714 "mac.l %%d5, %%a5, 20(%2), %%a5, %%acc0\n\t"
715 "mac.l %%d6, %%a5, 12(%2), %%a5, %%acc0\n\t"
716 "mac.l %%d7, %%a5, %%acc0\n\t"
717 "movclr.l %%acc0, %0\n\t"
718 : "=r" (hi0) : "a" (*fo), "a" (*D0ptr)
719 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5");
720
721 pcm[0] = -(hi0 << 3);
722 }
723 else
724 {
725 asm volatile(
726 "movem.l (%1), %%d0-%%d7\n\t"
727 "move.l (%2), %%a5\n\t"
728 "msac.l %%d0, %%a5, 56(%2), %%a5, %%acc0\n\t"
729 "msac.l %%d1, %%a5, 48(%2), %%a5, %%acc0\n\t"
730 "msac.l %%d2, %%a5, 40(%2), %%a5, %%acc0\n\t"
731 "msac.l %%d3, %%a5, 32(%2), %%a5, %%acc0\n\t"
732 "msac.l %%d4, %%a5, 24(%2), %%a5, %%acc0\n\t"
733 "msac.l %%d5, %%a5, 16(%2), %%a5, %%acc0\n\t"
734 "msac.l %%d6, %%a5, 8(%2), %%a5, %%acc0\n\t"
735 "msac.l %%d7, %%a5, 4(%2), %%a5, %%acc0\n\t"
736
737 "movem.l (%3), %%d0-%%d7\n\t"
738 "mac.l %%d0, %%a5, 60(%2), %%a5, %%acc0\n\t"
739 "mac.l %%d1, %%a5, 52(%2), %%a5, %%acc0\n\t"
740 "mac.l %%d2, %%a5, 44(%2), %%a5, %%acc0\n\t"
741 "mac.l %%d3, %%a5, 36(%2), %%a5, %%acc0\n\t"
742 "mac.l %%d4, %%a5, 28(%2), %%a5, %%acc0\n\t"
743 "mac.l %%d5, %%a5, 20(%2), %%a5, %%acc0\n\t"
744 "mac.l %%d6, %%a5, 12(%2), %%a5, %%acc0\n\t"
745 "mac.l %%d7, %%a5, %%acc0\n\t"
746 "movclr.l %%acc0, %0\n\t"
747 : "=r" (hi0) : "a" (*fx), "a" (*D0ptr), "a" (*fe)
748 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5");
749
750 pcm[0] = hi0 << 3; /* shift result to libmad's fixed point format */
751 pcm += 16;
752
753 for (sb = 15; sb; sb--, fo++) {
754 ++fe;
755 ++D0ptr;
756 ++D1ptr;
757
758 /* D[32 - sb][i] == -D[sb][31 - i] */
759 asm volatile (
760 "movem.l (%0), %%d0-%%d7\n\t"
761 "move.l (%2), %%a5\n\t"
762 "msac.l %%d0, %%a5, 56(%2), %%a5, %%acc0\n\t"
763 "msac.l %%d1, %%a5, 48(%2), %%a5, %%acc0\n\t"
764 "msac.l %%d2, %%a5, 40(%2), %%a5, %%acc0\n\t"
765 "msac.l %%d3, %%a5, 32(%2), %%a5, %%acc0\n\t"
766 "msac.l %%d4, %%a5, 24(%2), %%a5, %%acc0\n\t"
767 "msac.l %%d5, %%a5, 16(%2), %%a5, %%acc0\n\t"
768 "msac.l %%d6, %%a5, 8(%2), %%a5, %%acc0\n\t"
769 "msac.l %%d7, %%a5, 116(%3), %%a5, %%acc0\n\t"
770 "mac.l %%d7, %%a5, 108(%3), %%a5, %%acc1\n\t"
771 "mac.l %%d6, %%a5, 100(%3), %%a5, %%acc1\n\t"
772 "mac.l %%d5, %%a5, 92(%3), %%a5, %%acc1\n\t"
773 "mac.l %%d4, %%a5, 84(%3), %%a5, %%acc1\n\t"
774 "mac.l %%d3, %%a5, 76(%3), %%a5, %%acc1\n\t"
775 "mac.l %%d2, %%a5, 68(%3), %%a5, %%acc1\n\t"
776 "mac.l %%d1, %%a5, 60(%3), %%a5, %%acc1\n\t"
777 "mac.l %%d0, %%a5, 12(%2), %%a5, %%acc1\n\t"
778 "movem.l (%1), %%d0-%%d7\n\t"
779 "mac.l %%d7, %%a5, 20(%2), %%a5, %%acc0\n\t"
780 "mac.l %%d6, %%a5, 28(%2), %%a5, %%acc0\n\t"
781 "mac.l %%d5, %%a5, 36(%2), %%a5, %%acc0\n\t"
782 "mac.l %%d4, %%a5, 44(%2), %%a5, %%acc0\n\t"
783 "mac.l %%d3, %%a5, 52(%2), %%a5, %%acc0\n\t"
784 "mac.l %%d2, %%a5, 60(%2), %%a5, %%acc0\n\t"
785 "mac.l %%d1, %%a5, 4(%2), %%a5, %%acc0\n\t"
786 "mac.l %%d0, %%a5, 120(%3), %%a5, %%acc0\n\t"
787 "mac.l %%d0, %%a5, 64(%3), %%a5, %%acc1\n\t"
788 "mac.l %%d1, %%a5, 72(%3), %%a5, %%acc1\n\t"
789 "mac.l %%d2, %%a5, 80(%3), %%a5, %%acc1\n\t"
790 "mac.l %%d3, %%a5, 88(%3), %%a5, %%acc1\n\t"
791 "mac.l %%d4, %%a5, 96(%3), %%a5, %%acc1\n\t"
792 "mac.l %%d5, %%a5, 104(%3), %%a5, %%acc1\n\t"
793 "mac.l %%d6, %%a5, 112(%3), %%a5, %%acc1\n\t"
794 "mac.l %%d7, %%a5, %%acc1\n\t"
795 : : "a" (*fo), "a" (*fe), "a" (*D0ptr), "a" (*D1ptr)
796 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5");
797
798 asm volatile(
799 "movclr.l %%acc0, %0\n\t"
800 "movclr.l %%acc1, %1\n\t" : "=d" (hi0), "=d" (hi1) );
801
802 pcm[-sb] = hi0 << 3;
803 pcm[ sb] = hi1 << 3;
804 }
805
806 ++D0ptr;
807 asm volatile(
808 "movem.l (%1), %%d0-%%d7\n\t"
809 "move.l (%2), %%a5\n\t"
810 "mac.l %%d0, %%a5, 56(%2), %%a5, %%acc0\n\t"
811 "mac.l %%d1, %%a5, 48(%2), %%a5, %%acc0\n\t"
812 "mac.l %%d2, %%a5, 40(%2), %%a5, %%acc0\n\t"
813 "mac.l %%d3, %%a5, 32(%2), %%a5, %%acc0\n\t"
814 "mac.l %%d4, %%a5, 24(%2), %%a5, %%acc0\n\t"
815 "mac.l %%d5, %%a5, 16(%2), %%a5, %%acc0\n\t"
816 "mac.l %%d6, %%a5, 8(%2), %%a5, %%acc0\n\t"
817 "mac.l %%d7, %%a5, %%acc0\n\t"
818 "movclr.l %%acc0, %0\n\t"
819 : "=r" (hi0) : "a" (*fo), "a" (*D0ptr)
820 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5");
821
822 pcm[0] = -(hi0 << 3);
823 }
824 pcm += 16;
825 phase = (phase + 1) % 16;
826 }
827 }
828 }
829
830 #elif defined(FPM_ARM)
831
832 #define PROD_O(hi, lo, f, ptr) \
833 ({ \
834 mad_fixed_t *__p = (f); \
835 asm volatile ( \
836 "ldmia %2!, {r0, r1, r2, r3} \n\t" \
837 "ldr r4, [%3, #0] \n\t" \
838 "smull %0, %1, r0, r4 \n\t" \
839 "ldr r4, [%3, #56] \n\t" \
840 "smlal %0, %1, r1, r4 \n\t" \
841 "ldr r4, [%3, #48] \n\t" \
842 "smlal %0, %1, r2, r4 \n\t" \
843 "ldr r4, [%3, #40] \n\t" \
844 "smlal %0, %1, r3, r4 \n\t" \
845 "ldmia %2, {r0, r1, r2, r3} \n\t" \
846 "ldr r4, [%3, #32] \n\t" \
847 "smlal %0, %1, r0, r4 \n\t" \
848 "ldr r4, [%3, #24] \n\t" \
849 "smlal %0, %1, r1, r4 \n\t" \
850 "ldr r4, [%3, #16] \n\t" \
851 "smlal %0, %1, r2, r4 \n\t" \
852 "ldr r4, [%3, #8] \n\t" \
853 "smlal %0, %1, r3, r4 \n\t" \
854 : "=&r" (lo), "=&r" (hi), "+r" (__p) \
855 : "r" (ptr) \
856 : "r0", "r1", "r2", "r3", "r4", "memory"); \
857 })
858
859 #define PROD_A(hi, lo, f, ptr) \
860 ({ \
861 mad_fixed_t *__p = (f); \
862 asm volatile ( \
863 "ldmia %2!, {r0, r1, r2, r3} \n\t" \
864 "ldr r4, [%3, #0] \n\t" \
865 "smlal %0, %1, r0, r4 \n\t" \
866 "ldr r4, [%3, #56] \n\t" \
867 "smlal %0, %1, r1, r4 \n\t" \
868 "ldr r4, [%3, #48] \n\t" \
869 "smlal %0, %1, r2, r4 \n\t" \
870 "ldr r4, [%3, #40] \n\t" \
871 "smlal %0, %1, r3, r4 \n\t" \
872 "ldmia %2, {r0, r1, r2, r3} \n\t" \
873 "ldr r4, [%3, #32] \n\t" \
874 "smlal %0, %1, r0, r4 \n\t" \
875 "ldr r4, [%3, #24] \n\t" \
876 "smlal %0, %1, r1, r4 \n\t" \
877 "ldr r4, [%3, #16] \n\t" \
878 "smlal %0, %1, r2, r4 \n\t" \
879 "ldr r4, [%3, #8] \n\t" \
880 "smlal %0, %1, r3, r4 \n\t" \
881 : "+r" (lo), "+r" (hi), "+r" (__p) \
882 : "r" (ptr) \
883 : "r0", "r1", "r2", "r3", "r4", "memory"); \
884 })
885
886 void synth_full_odd_sbsample (mad_fixed_t *pcm,
887 mad_fixed_t (*fo)[8],
888 mad_fixed_t (*fe)[8],
889 mad_fixed_t const (*D0ptr)[32],
890 mad_fixed_t const (*D1ptr)[32]);
891 void synth_full_even_sbsample(mad_fixed_t *pcm,
892 mad_fixed_t (*fo)[8],
893 mad_fixed_t (*fe)[8],
894 mad_fixed_t const (*D0ptr)[32],
895 mad_fixed_t const (*D1ptr)[32]);
896
897 static
898 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
899 unsigned int nch, unsigned int ns) ICODE_ATTR_MPA_SYNTH;
900 static
901 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
902 unsigned int nch, unsigned int ns)
903 {
904 int p;
905 unsigned int phase, ch, s;
906 mad_fixed_t *pcm, (*filter)[2][2][16][8];
907 mad_fixed_t (*sbsample)[36][32];
908 mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
909 mad_fixed_t const (*D0ptr)[32], *ptr;
910 mad_fixed_t const (*D1ptr)[32];
911 mad_fixed64hi_t hi;
912 mad_fixed64lo_t lo;
913
914 for (ch = 0; ch < nch; ++ch) {
915 sbsample = &(*frame->sbsample_prev)[ch];
916 filter = &synth->filter[ch];
917 phase = synth->phase;
918 pcm = synth->pcm.samples[ch];
919
920 for (s = 0; s < ns; ++s) {
921 dct32((*sbsample)[s], phase >> 1,
922 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
923
924 p = (phase - 1) & 0xf;
925
926 /* calculate 32 samples */
927 fe = &(*filter)[0][ phase & 1][0];
928 fx = &(*filter)[0][~phase & 1][0];
929 fo = &(*filter)[1][~phase & 1][0];
930
931 D0ptr = (void*)&D[0][ p];
932 D1ptr = (void*)&D[0][-p];
933
934 if(s & 1)
935 {
936 ptr = *D0ptr;
937 PROD_O(hi, lo, *fx, ptr+1);
938 MLN(hi, lo);
939 PROD_A(hi, lo, *fe, ptr);
940 pcm[0] = SHIFT(MLZ(hi, lo));
941 pcm += 16;
942
943 synth_full_odd_sbsample(pcm, fo, fe, D0ptr, D1ptr);
944 D0ptr += 15;
945 D1ptr += 15;
946 fo += 15;
947 fe += 15;
948
949 ptr = *(D0ptr + 1);
950 PROD_O(hi, lo, *fo, ptr+1);
951 pcm[0] = SHIFT(-MLZ(hi, lo));
952 }
953 else
954 {
955 ptr = *D0ptr;
956 PROD_O(hi, lo, *fx, ptr);
957 MLN(hi, lo);
958 PROD_A(hi, lo, *fe, ptr+1);
959 pcm[0] = SHIFT(MLZ(hi, lo));
960 pcm += 16;
961
962 synth_full_even_sbsample(pcm, fo, fe, D0ptr, D1ptr);
963 D0ptr += 15;
964 D1ptr += 15;
965 fo += 15;
966 fe += 15;
967
968 ptr = *(D0ptr + 1);
969 PROD_O(hi, lo, *fo, ptr);
970 pcm[0] = SHIFT(-MLZ(hi, lo));
971 }
972
973 pcm += 16;
974 phase = (phase + 1) % 16;
975 }
976 }
977 }
978
979 # else /* not FPM_COLDFIRE_EMAC and not FPM_ARM */
980
981 static
982 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
983 unsigned int nch, unsigned int ns)
984 {
985 int p, sb;
986 unsigned int phase, ch, s;
987 mad_fixed_t *pcm, (*filter)[2][2][16][8];
988 mad_fixed_t (*sbsample)[36][32];
989 mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
990 mad_fixed_t const (*D0ptr)[32], *ptr;
991 mad_fixed_t const (*D1ptr)[32];
992 mad_fixed64hi_t hi;
993 mad_fixed64lo_t lo;
994
995 for (ch = 0; ch < nch; ++ch) {
996 sbsample = &(*frame->sbsample_prev)[ch];
997 filter = &synth->filter[ch];
998 phase = synth->phase;
999 pcm = synth->pcm.samples[ch];
1000
1001 for (s = 0; s < ns; ++s) {
1002 dct32((*sbsample)[s], phase >> 1,
1003 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
1004
1005 p = (phase - 1) & 0xf;
1006
1007 /* calculate 32 samples */
1008 fe = &(*filter)[0][ phase & 1][0];
1009 fx = &(*filter)[0][~phase & 1][0];
1010 fo = &(*filter)[1][~phase & 1][0];
1011
1012 D0ptr = (void*)&D[0][ p];
1013 D1ptr = (void*)&D[0][-p];
1014
1015 if(s & 1)
1016 {
1017 ptr = *D0ptr;
1018 ML0(hi, lo, (*fx)[0], ptr[ 1]);
1019 MLA(hi, lo, (*fx)[1], ptr[15]);
1020 MLA(hi, lo, (*fx)[2], ptr[13]);
1021 MLA(hi, lo, (*fx)[3], ptr[11]);
1022 MLA(hi, lo, (*fx)[4], ptr[ 9]);
1023 MLA(hi, lo, (*fx)[5], ptr[ 7]);
1024 MLA(hi, lo, (*fx)[6], ptr[ 5]);
1025 MLA(hi, lo, (*fx)[7], ptr[ 3]);
1026 MLN(hi, lo);
1027 MLA(hi, lo, (*fe)[0], ptr[ 0]);
1028 MLA(hi, lo, (*fe)[1], ptr[14]);
1029 MLA(hi, lo, (*fe)[2], ptr[12]);
1030 MLA(hi, lo, (*fe)[3], ptr[10]);
1031 MLA(hi, lo, (*fe)[4], ptr[ 8]);
1032 MLA(hi, lo, (*fe)[5], ptr[ 6]);
1033 MLA(hi, lo, (*fe)[6], ptr[ 4]);
1034 MLA(hi, lo, (*fe)[7], ptr[ 2]);
1035 pcm[0] = SHIFT(MLZ(hi, lo));
1036 pcm += 16;
1037
1038 for (sb = 15; sb; sb--, fo++)
1039 {
1040 ++fe;
1041 ++D0ptr;
1042 ++D1ptr;
1043
1044 /* D[32 - sb][i] == -D[sb][31 - i] */
1045 ptr = *D0ptr;
1046 ML0(hi, lo, (*fo)[0], ptr[ 1]);
1047 MLA(hi, lo, (*fo)[1], ptr[15]);
1048 MLA(hi, lo, (*fo)[2], ptr[13]);
1049 MLA(hi, lo, (*fo)[3], ptr[11]);
1050 MLA(hi, lo, (*fo)[4], ptr[ 9]);
1051 MLA(hi, lo, (*fo)[5], ptr[ 7]);
1052 MLA(hi, lo, (*fo)[6], ptr[ 5]);
1053 MLA(hi, lo, (*fo)[7], ptr[ 3]);
1054 MLN(hi, lo);
1055 MLA(hi, lo, (*fe)[7], ptr[ 2]);
1056 MLA(hi, lo, (*fe)[6], ptr[ 4]);
1057 MLA(hi, lo, (*fe)[5], ptr[ 6]);
1058 MLA(hi, lo, (*fe)[4], ptr[ 8]);
1059 MLA(hi, lo, (*fe)[3], ptr[10]);
1060 MLA(hi, lo, (*fe)[2], ptr[12]);
1061 MLA(hi, lo, (*fe)[1], ptr[14]);
1062 MLA(hi, lo, (*fe)[0], ptr[ 0]);
1063 pcm[-sb] = SHIFT(MLZ(hi, lo));
1064
1065 ptr = *D1ptr;
1066 ML0(hi, lo, (*fe)[0], ptr[31 - 16]);
1067 MLA(hi, lo, (*fe)[1], ptr[31 - 14]);
1068 MLA(hi, lo, (*fe)[2], ptr[31 - 12]);
1069 MLA(hi, lo, (*fe)[3], ptr[31 - 10]);
1070 MLA(hi, lo, (*fe)[4], ptr[31 - 8]);
1071 MLA(hi, lo, (*fe)[5], ptr[31 - 6]);
1072 MLA(hi, lo, (*fe)[6], ptr[31 - 4]);
1073 MLA(hi, lo, (*fe)[7], ptr[31 - 2]);
1074 MLA(hi, lo, (*fo)[7], ptr[31 - 3]);
1075 MLA(hi, lo, (*fo)[6], ptr[31 - 5]);
1076 MLA(hi, lo, (*fo)[5], ptr[31 - 7]);
1077 MLA(hi, lo, (*fo)[4], ptr[31 - 9]);
1078 MLA(hi, lo, (*fo)[3], ptr[31 - 11]);
1079 MLA(hi, lo, (*fo)[2], ptr[31 - 13]);
1080 MLA(hi, lo, (*fo)[1], ptr[31 - 15]);
1081 MLA(hi, lo, (*fo)[0], ptr[31 - 1]);
1082 pcm[sb] = SHIFT(MLZ(hi, lo));
1083 }
1084
1085 ptr = *(D0ptr + 1);
1086 ML0(hi, lo, (*fo)[0], ptr[ 1]);
1087 MLA(hi, lo, (*fo)[1], ptr[15]);
1088 MLA(hi, lo, (*fo)[2], ptr[13]);
1089 MLA(hi, lo, (*fo)[3], ptr[11]);
1090 MLA(hi, lo, (*fo)[4], ptr[ 9]);
1091 MLA(hi, lo, (*fo)[5], ptr[ 7]);
1092 MLA(hi, lo, (*fo)[6], ptr[ 5]);
1093 MLA(hi, lo, (*fo)[7], ptr[ 3]);
1094 pcm[0] = SHIFT(-MLZ(hi, lo));
1095 }
1096 else
1097 {
1098 ptr = *D0ptr;
1099 ML0(hi, lo, (*fx)[0], ptr[ 0]);
1100 MLA(hi, lo, (*fx)[1], ptr[14]);
1101 MLA(hi, lo, (*fx)[2], ptr[12]);
1102 MLA(hi, lo, (*fx)[3], ptr[10]);
1103 MLA(hi, lo, (*fx)[4], ptr[ 8]);
1104 MLA(hi, lo, (*fx)[5], ptr[ 6]);
1105 MLA(hi, lo, (*fx)[6], ptr[ 4]);
1106 MLA(hi, lo, (*fx)[7], ptr[ 2]);
1107 MLN(hi, lo);
1108 MLA(hi, lo, (*fe)[0], ptr[ 1]);
1109 MLA(hi, lo, (*fe)[1], ptr[15]);
1110 MLA(hi, lo, (*fe)[2], ptr[13]);
1111 MLA(hi, lo, (*fe)[3], ptr[11]);
1112 MLA(hi, lo, (*fe)[4], ptr[ 9]);
1113 MLA(hi, lo, (*fe)[5], ptr[ 7]);
1114 MLA(hi, lo, (*fe)[6], ptr[ 5]);
1115 MLA(hi, lo, (*fe)[7], ptr[ 3]);
1116 pcm[0] = SHIFT(MLZ(hi, lo));
1117 pcm += 16;
1118
1119 for (sb = 15; sb; sb--, fo++)
1120 {
1121 ++fe;
1122 ++D0ptr;
1123 ++D1ptr;
1124
1125 /* D[32 - sb][i] == -D[sb][31 - i] */
1126 ptr = *D0ptr;
1127 ML0(hi, lo, (*fo)[0], ptr[ 0]);
1128 MLA(hi, lo, (*fo)[1], ptr[14]);
1129 MLA(hi, lo, (*fo)[2], ptr[12]);
1130 MLA(hi, lo, (*fo)[3], ptr[10]);
1131 MLA(hi, lo, (*fo)[4], ptr[ 8]);
1132 MLA(hi, lo, (*fo)[5], ptr[ 6]);
1133 MLA(hi, lo, (*fo)[6], ptr[ 4]);
1134 MLA(hi, lo, (*fo)[7], ptr[ 2]);
1135 MLN(hi, lo);
1136 MLA(hi, lo, (*fe)[7], ptr[ 3]);
1137 MLA(hi, lo, (*fe)[6], ptr[ 5]);
1138 MLA(hi, lo, (*fe)[5], ptr[ 7]);
1139 MLA(hi, lo, (*fe)[4], ptr[ 9]);
1140 MLA(hi, lo, (*fe)[3], ptr[11]);
1141 MLA(hi, lo, (*fe)[2], ptr[13]);
1142 MLA(hi, lo, (*fe)[1], ptr[15]);
1143 MLA(hi, lo, (*fe)[0], ptr[ 1]);
1144 pcm[-sb] = SHIFT(MLZ(hi, lo));
1145
1146 ptr = *D1ptr;
1147 ML0(hi, lo, (*fe)[0], ptr[31 - 1]);
1148 MLA(hi, lo, (*fe)[1], ptr[31 - 15]);
1149 MLA(hi, lo, (*fe)[2], ptr[31 - 13]);
1150 MLA(hi, lo, (*fe)[3], ptr[31 - 11]);
1151 MLA(hi, lo, (*fe)[4], ptr[31 - 9]);
1152 MLA(hi, lo, (*fe)[5], ptr[31 - 7]);
1153 MLA(hi, lo, (*fe)[6], ptr[31 - 5]);
1154 MLA(hi, lo, (*fe)[7], ptr[31 - 3]);
1155 MLA(hi, lo, (*fo)[7], ptr[31 - 2]);
1156 MLA(hi, lo, (*fo)[6], ptr[31 - 4]);
1157 MLA(hi, lo, (*fo)[5], ptr[31 - 6]);
1158 MLA(hi, lo, (*fo)[4], ptr[31 - 8]);
1159 MLA(hi, lo, (*fo)[3], ptr[31 - 10]);
1160 MLA(hi, lo, (*fo)[2], ptr[31 - 12]);
1161 MLA(hi, lo, (*fo)[1], ptr[31 - 14]);
1162 MLA(hi, lo, (*fo)[0], ptr[31 - 16]);
1163 pcm[sb] = SHIFT(MLZ(hi, lo));
1164 }
1165
1166 ptr = *(D0ptr + 1);
1167 ML0(hi, lo, (*fo)[0], ptr[ 0]);
1168 MLA(hi, lo, (*fo)[1], ptr[14]);
1169 MLA(hi, lo, (*fo)[2], ptr[12]);
1170 MLA(hi, lo, (*fo)[3], ptr[10]);
1171 MLA(hi, lo, (*fo)[4], ptr[ 8]);
1172 MLA(hi, lo, (*fo)[5], ptr[ 6]);
1173 MLA(hi, lo, (*fo)[6], ptr[ 4]);
1174 MLA(hi, lo, (*fo)[7], ptr[ 2]);
1175 pcm[0] = SHIFT(-MLZ(hi, lo));
1176 }
1177
1178 pcm += 16;
1179 phase = (phase + 1) % 16;
1180 }
1181 }
1182 }
1183
1184 # endif
1185 # endif
1186
1187 /*
1188 * NAME: synth->half()
1189 * DESCRIPTION: perform half frequency PCM synthesis
1190 */
1191 static
1192 void synth_half(struct mad_synth *synth, struct mad_frame const *frame,
1193 unsigned int nch, unsigned int ns)
1194 {
1195 unsigned int phase, ch, s, sb, pe, po;
1196 mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8];
1197 mad_fixed_t (*sbsample)[36][32];
1198 register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
1199 register mad_fixed_t const (*Dptr)[32], *ptr;
1200 register mad_fixed64hi_t hi;
1201 register mad_fixed64lo_t lo;
1202
1203 for (ch = 0; ch < nch; ++ch) {
1204 sbsample = &(*frame->sbsample_prev)[ch];
1205 filter = &synth->filter[ch];
1206 phase = synth->phase;
1207 pcm1 = synth->pcm.samples[ch];
1208
1209 for (s = 0; s < ns; ++s) {
1210 dct32((*sbsample)[s], phase >> 1,
1211 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
1212
1213 pe = phase & ~1;
1214 po = ((phase - 1) & 0xf) | 1;
1215
1216 /* calculate 16 samples */
1217
1218 fe = &(*filter)[0][ phase & 1][0];
1219 fx = &(*filter)[0][~phase & 1][0];
1220 fo = &(*filter)[1][~phase & 1][0];
1221
1222 Dptr = &D[0];
1223
1224 ptr = *Dptr + po;
1225 ML0(hi, lo, (*fx)[0], ptr[ 0]);
1226 MLA(hi, lo, (*fx)[1], ptr[14]);
1227 MLA(hi, lo, (*fx)[2], ptr[12]);
1228 MLA(hi, lo, (*fx)[3], ptr[10]);
1229 MLA(hi, lo, (*fx)[4], ptr[ 8]);
1230 MLA(hi, lo, (*fx)[5], ptr[ 6]);
1231 MLA(hi, lo, (*fx)[6], ptr[ 4]);
1232 MLA(hi, lo, (*fx)[7], ptr[ 2]);
1233 MLN(hi, lo);
1234
1235 ptr = *Dptr + pe;
1236 MLA(hi, lo, (*fe)[0], ptr[ 0]);
1237 MLA(hi, lo, (*fe)[1], ptr[14]);
1238 MLA(hi, lo, (*fe)[2], ptr[12]);
1239 MLA(hi, lo, (*fe)[3], ptr[10]);
1240 MLA(hi, lo, (*fe)[4], ptr[ 8]);
1241 MLA(hi, lo, (*fe)[5], ptr[ 6]);
1242 MLA(hi, lo, (*fe)[6], ptr[ 4]);
1243 MLA(hi, lo, (*fe)[7], ptr[ 2]);
1244
1245 *pcm1++ = SHIFT(MLZ(hi, lo));
1246
1247 pcm2 = pcm1 + 14;
1248
1249 for (sb = 1; sb < 16; ++sb) {
1250 ++fe;
1251 ++Dptr;
1252
1253 /* D[32 - sb][i] == -D[sb][31 - i] */
1254
1255 if (!(sb & 1)) {
1256 ptr = *Dptr + po;
1257 ML0(hi, lo, (*fo)[0], ptr[ 0]);
1258 MLA(hi, lo, (*fo)[1], ptr[14]);
1259 MLA(hi, lo, (*fo)[2], ptr[12]);
1260 MLA(hi, lo, (*fo)[3], ptr[10]);
1261 MLA(hi, lo, (*fo)[4], ptr[ 8]);
1262 MLA(hi, lo, (*fo)[5], ptr[ 6]);
1263 MLA(hi, lo, (*fo)[6], ptr[ 4]);
1264 MLA(hi, lo, (*fo)[7], ptr[ 2]);
1265 MLN(hi, lo);
1266
1267 ptr = *Dptr + pe;
1268 MLA(hi, lo, (*fe)[7], ptr[ 2]);
1269 MLA(hi, lo, (*fe)[6], ptr[ 4]);
1270 MLA(hi, lo, (*fe)[5], ptr[ 6]);
1271 MLA(hi, lo, (*fe)[4], ptr[ 8]);
1272 MLA(hi, lo, (*fe)[3], ptr[10]);
1273 MLA(hi, lo, (*fe)[2], ptr[12]);
1274 MLA(hi, lo, (*fe)[1], ptr[14]);
1275 MLA(hi, lo, (*fe)[0], ptr[ 0]);
1276
1277 *pcm1++ = SHIFT(MLZ(hi, lo));
1278
1279 ptr = *Dptr - po;
1280 ML0(hi, lo, (*fo)[7], ptr[31 - 2]);
1281 MLA(hi, lo, (*fo)[6], ptr[31 - 4]);
1282 MLA(hi, lo, (*fo)[5], ptr[31 - 6]);
1283 MLA(hi, lo, (*fo)[4], ptr[31 - 8]);
1284 MLA(hi, lo, (*fo)[3], ptr[31 - 10]);
1285 MLA(hi, lo, (*fo)[2], ptr[31 - 12]);
1286 MLA(hi, lo, (*fo)[1], ptr[31 - 14]);
1287 MLA(hi, lo, (*fo)[0], ptr[31 - 16]);
1288
1289 ptr = *Dptr - pe;
1290 MLA(hi, lo, (*fe)[0], ptr[31 - 16]);
1291 MLA(hi, lo, (*fe)[1], ptr[31 - 14]);
1292 MLA(hi, lo, (*fe)[2], ptr[31 - 12]);
1293 MLA(hi, lo, (*fe)[3], ptr[31 - 10]);
1294 MLA(hi, lo, (*fe)[4], ptr[31 - 8]);
1295 MLA(hi, lo, (*fe)[5], ptr[31 - 6]);
1296 MLA(hi, lo, (*fe)[6], ptr[31 - 4]);
1297 MLA(hi, lo, (*fe)[7], ptr[31 - 2]);
1298
1299 *pcm2-- = SHIFT(MLZ(hi, lo));
1300 }
1301
1302 ++fo;
1303 }
1304
1305 ++Dptr;
1306
1307 ptr = *Dptr + po;
1308 ML0(hi, lo, (*fo)[0], ptr[ 0]);
1309 MLA(hi, lo, (*fo)[1], ptr[14]);
1310 MLA(hi, lo, (*fo)[2], ptr[12]);
1311 MLA(hi, lo, (*fo)[3], ptr[10]);
1312 MLA(hi, lo, (*fo)[4], ptr[ 8]);
1313 MLA(hi, lo, (*fo)[5], ptr[ 6]);
1314 MLA(hi, lo, (*fo)[6], ptr[ 4]);
1315 MLA(hi, lo, (*fo)[7], ptr[ 2]);
1316
1317 *pcm1 = SHIFT(-MLZ(hi, lo));
1318 pcm1 += 8;
1319
1320 phase = (phase + 1) % 16;
1321 }
1322 }
1323 }
1324
1325 /*
1326 * NAME: synth->frame()
1327 * DESCRIPTION: perform PCM synthesis of frame subband samples
1328 */
1329 void mad_synth_frame(struct mad_synth *synth, struct mad_frame const *frame)
1330 {
1331 unsigned int nch, ns;
1332 void (*synth_frame)(struct mad_synth *, struct mad_frame const *,
1333 unsigned int, unsigned int);
1334
1335 nch = MAD_NCHANNELS(&frame->header);
1336 ns = MAD_NSBSAMPLES(&frame->header);
1337
1338 synth->pcm.samplerate = frame->header.samplerate;
1339 synth->pcm.channels = nch;
1340 synth->pcm.length = 32 * ns;
1341
1342 synth_frame = synth_full;
1343
1344 if (frame->options & MAD_OPTION_HALFSAMPLERATE) {
1345 synth->pcm.samplerate /= 2;
1346 synth->pcm.length /= 2;
1347
1348 synth_frame = synth_half;
1349 }
1350
1351 synth_frame(synth, frame, nch, ns);
1352
1353 synth->phase = (synth->phase + ns) % 16;
1354 }
My public Rockbox git repo