?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Initial 800x480 cabbiev2 port, based on 480x800x16 one
[kugel-rb.git] / apps / codecs / libmad / synth.c
blobebcf3d5015ee5be7af19ea7b1d1dbde41c920f3b
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 memset(synth->filter, 0, sizeof(synth->filter));
59 }
60
61 #if 0 /* dct32 asm implementation is slower on current arm systems */
62 /* #ifdef FPM_ARM */
63
64 void dct32(mad_fixed_t const in[32], unsigned int slot,
65 mad_fixed_t lo[16][8], mad_fixed_t hi[16][8]);
66
67 #else
68
69 /*
70 * An optional optimization called here the Subband Synthesis Optimization
71 * (SSO) improves the performance of subband synthesis at the expense of
72 * accuracy.
73 *
74 * The idea is to simplify 32x32->64-bit multiplication to 32x32->32 such
75 * that extra scaling and rounding are not necessary. This often allows the
76 * compiler to use faster 32-bit multiply-accumulate instructions instead of
77 * explicit 64-bit multiply, shift, and add instructions.
78 *
79 * SSO works like this: a full 32x32->64-bit multiply of two mad_fixed_t
80 * values requires the result to be right-shifted 28 bits to be properly
81 * scaled to the same fixed-point format. Right shifts can be applied at any
82 * time to either operand or to the result, so the optimization involves
83 * careful placement of these shifts to minimize the loss of accuracy.
84 *
85 * First, a 14-bit shift is applied with rounding at compile-time to the D[]
86 * table of coefficients for the subband synthesis window. This only loses 2
87 * bits of accuracy because the lower 12 bits are always zero. A second
88 * 12-bit shift occurs after the DCT calculation. This loses 12 bits of
89 * accuracy. Finally, a third 2-bit shift occurs just before the sample is
90 * saved in the PCM buffer. 14 + 12 + 2 == 28 bits.
91 */
92
93 /* FPM_DEFAULT without OPT_SSO will actually not work. */
94 # if defined(FPM_DEFAULT) && !defined(OPT_SSO)
95 # define OPT_SSO
96 # endif
97
98 /* second SSO shift, with rounding */
99 # if defined(OPT_SSO)
100 # define SHIFT(x) (((x) + (1L << 11)) >> 12)
101 # else
102 # define SHIFT(x) (x)
103 # endif
104
105 /* possible DCT speed optimization */
106 # if defined(FPM_COLDFIRE_EMAC)
107 /* This is a Coldfire version of the OPT_SPEED optimisation below, but in the
108 case of Coldfire it does not loose additional precision. */
109 # define MUL(x, y) \
110 ({ \
111 mad_fixed64hi_t hi; \
112 asm volatile("mac.l %[a], %[b], %%acc0\n\t" \
113 "movclr.l %%acc0, %[hi]" \
114 : [hi] "=r" (hi) \
115 : [a] "r" ((x)), [b] "r" ((y))); \
116 hi; \
117 })
118 # elif defined(FPM_ARM)
119 /* This is an ARM version of the OPT_SPEED optimisation below. This
120 implementation will loose 1 bit of accuracy. */
121 # define MUL(x, y) \
122 ({ \
123 mad_fixed64lo_t lo; \
124 mad_fixed64hi_t hi; \
125 asm volatile ( \
126 "smull %[lo], %[hi], %[a], %[b]\n\t" /* hi = result */ \
127 "mov %[hi], %[hi], lsl #1" /* hi <<= 1 */ \
128 : [lo]"=&r"(lo), [hi]"=&r"(hi) \
129 : [a]"r"(x), [b]"r"(y)); \
130 hi; \
131 })
132 # elif defined(OPT_SPEED) && defined(MAD_F_MLX)
133 # define MUL(x, y) \
134 ({ mad_fixed64hi_t hi; \
135 mad_fixed64lo_t lo; \
136 MAD_F_MLX(hi, lo, (x), (y)); \
137 hi << (32 - MAD_F_SCALEBITS - 3); \
138 })
139 # else
140 # define MUL(x, y) mad_f_mul((x), (y>>3))
141 # endif
142
143 /*
144 * NAME: dct32()
145 * DESCRIPTION: perform fast in[32]->out[32] DCT
146 */
147 static
148 void dct32(mad_fixed_t const in[32], unsigned int slot,
149 mad_fixed_t lo[16][8], mad_fixed_t hi[16][8])
150 {
151 mad_fixed_t t0, t1, t2, t3, t4, t5, t6, t7;
152 mad_fixed_t t8, t9, t10, t11, t12, t13, t14, t15;
153 mad_fixed_t t16, t17, t18, t19, t20, t21, t22, t23;
154 mad_fixed_t t24, t25, t26, t27, t28, t29, t30, t31;
155 mad_fixed_t t32, t33, t34, t35, t36, t37, t38, t39;
156 mad_fixed_t t40, t41, t42, t43, t44, t45, t46, t47;
157 mad_fixed_t t48, t49, t50, t51, t52, t53, t54, t55;
158 mad_fixed_t t56, t57, t58, t59, t60, t61, t62, t63;
159 mad_fixed_t t64, t65, t66, t67, t68, t69, t70, t71;
160 mad_fixed_t t72, t73, t74, t75, t76, t77, t78, t79;
161 mad_fixed_t t80, t81, t82, t83, t84, t85, t86, t87;
162 mad_fixed_t t88, t89, t90, t91, t92, t93, t94, t95;
163 mad_fixed_t t96, t97, t98, t99, t100, t101, t102, t103;
164 mad_fixed_t t104, t105, t106, t107, t108, t109, t110, t111;
165 mad_fixed_t t112, t113, t114, t115, t116, t117, t118, t119;
166 mad_fixed_t t120, t121, t122, t123, t124, t125, t126, t127;
167 mad_fixed_t t128, t129, t130, t131, t132, t133, t134, t135;
168 mad_fixed_t t136, t137, t138, t139, t140, t141, t142, t143;
169 mad_fixed_t t144, t145, t146, t147, t148, t149, t150, t151;
170 mad_fixed_t t152, t153, t154, t155, t156, t157, t158, t159;
171 mad_fixed_t t160, t161, t162, t163, t164, t165, t166, t167;
172 mad_fixed_t t168, t169, t170, t171, t172, t173, t174, t175;
173 mad_fixed_t t176;
174
175 /* costab[i] = cos(PI / (2 * 32) * i) */
176 #define costab1 MAD_F(0x7fd8878e) /* 0.998795456 */
177 #define costab2 MAD_F(0x7f62368f) /* 0.995184727 */
178 #define costab3 MAD_F(0x7e9d55fc) /* 0.989176510 */
179 #define costab4 MAD_F(0x7d8a5f40) /* 0.980785280 */
180 #define costab5 MAD_F(0x7c29fbee) /* 0.970031253 */
181 #define costab6 MAD_F(0x7a7d055b) /* 0.956940336 */
182 #define costab7 MAD_F(0x78848414) /* 0.941544065 */
183 #define costab8 MAD_F(0x7641af3d) /* 0.923879533 */
184 #define costab9 MAD_F(0x73b5ebd1) /* 0.903989293 */
185 #define costab10 MAD_F(0x70e2cbc6) /* 0.881921264 */
186 #define costab11 MAD_F(0x6dca0d14) /* 0.857728610 */
187 #define costab12 MAD_F(0x6a5d98a4) /* 0.831469612 */
188 #define costab13 MAD_F(0x66cf8120) /* 0.803207531 */
189 #define costab14 MAD_F(0x62f201ac) /* 0.773010453 */
190 #define costab15 MAD_F(0x5ed77c8a) /* 0.740951125 */
191 #define costab16 MAD_F(0x5a82799a) /* 0.707106781 */
192 #define costab17 MAD_F(0x55f5a4d2) /* 0.671558955 */
193 #define costab18 MAD_F(0x5133cc94) /* 0.634393284 */
194 #define costab19 MAD_F(0x4c3fdff4) /* 0.595699304 */
195 #define costab20 MAD_F(0x471cece7) /* 0.555570233 */
196 #define costab21 MAD_F(0x41ce1e65) /* 0.514102744 */
197 #define costab22 MAD_F(0x3c56ba70) /* 0.471396737 */
198 #define costab23 MAD_F(0x36ba2014) /* 0.427555093 */
199 #define costab24 MAD_F(0x30fbc54d) /* 0.382683432 */
200 #define costab25 MAD_F(0x2b1f34eb) /* 0.336889853 */
201 #define costab26 MAD_F(0x25280c5e) /* 0.290284677 */
202 #define costab27 MAD_F(0x1f19f97b) /* 0.242980180 */
203 #define costab28 MAD_F(0x18f8b83c) /* 0.195090322 */
204 #define costab29 MAD_F(0x12c8106f) /* 0.146730474 */
205 #define costab30 MAD_F(0x0c8bd35e) /* 0.098017140 */
206 #define costab31 MAD_F(0x0647d97c) /* 0.049067674 */
207
208 t0 = in[0] + in[31]; t16 = MUL(in[0] - in[31], costab1);
209 t1 = in[15] + in[16]; t17 = MUL(in[15] - in[16], costab31);
210
211 t41 = t16 + t17;
212 t59 = MUL(t16 - t17, costab2);
213 t33 = t0 + t1;
214 t50 = MUL(t0 - t1, costab2);
215
216 t2 = in[7] + in[24]; t18 = MUL(in[7] - in[24], costab15);
217 t3 = in[8] + in[23]; t19 = MUL(in[8] - in[23], costab17);
218
219 t42 = t18 + t19;
220 t60 = MUL(t18 - t19, costab30);
221 t34 = t2 + t3;
222 t51 = MUL(t2 - t3, costab30);
223
224 t4 = in[3] + in[28]; t20 = MUL(in[3] - in[28], costab7);
225 t5 = in[12] + in[19]; t21 = MUL(in[12] - in[19], costab25);
226
227 t43 = t20 + t21;
228 t61 = MUL(t20 - t21, costab14);
229 t35 = t4 + t5;
230 t52 = MUL(t4 - t5, costab14);
231
232 t6 = in[4] + in[27]; t22 = MUL(in[4] - in[27], costab9);
233 t7 = in[11] + in[20]; t23 = MUL(in[11] - in[20], costab23);
234
235 t44 = t22 + t23;
236 t62 = MUL(t22 - t23, costab18);
237 t36 = t6 + t7;
238 t53 = MUL(t6 - t7, costab18);
239
240 t8 = in[1] + in[30]; t24 = MUL(in[1] - in[30], costab3);
241 t9 = in[14] + in[17]; t25 = MUL(in[14] - in[17], costab29);
242
243 t45 = t24 + t25;
244 t63 = MUL(t24 - t25, costab6);
245 t37 = t8 + t9;
246 t54 = MUL(t8 - t9, costab6);
247
248 t10 = in[6] + in[25]; t26 = MUL(in[6] - in[25], costab13);
249 t11 = in[9] + in[22]; t27 = MUL(in[9] - in[22], costab19);
250
251 t46 = t26 + t27;
252 t64 = MUL(t26 - t27, costab26);
253 t38 = t10 + t11;
254 t55 = MUL(t10 - t11, costab26);
255
256 t12 = in[2] + in[29]; t28 = MUL(in[2] - in[29], costab5);
257 t13 = in[13] + in[18]; t29 = MUL(in[13] - in[18], costab27);
258
259 t47 = t28 + t29;
260 t65 = MUL(t28 - t29, costab10);
261 t39 = t12 + t13;
262 t56 = MUL(t12 - t13, costab10);
263
264 t14 = in[5] + in[26]; t30 = MUL(in[5] - in[26], costab11);
265 t15 = in[10] + in[21]; t31 = MUL(in[10] - in[21], costab21);
266
267 t48 = t30 + t31;
268 t66 = MUL(t30 - t31, costab22);
269 t40 = t14 + t15;
270 t57 = MUL(t14 - t15, costab22);
271
272 t69 = t33 + t34; t89 = MUL(t33 - t34, costab4);
273 t70 = t35 + t36; t90 = MUL(t35 - t36, costab28);
274 t71 = t37 + t38; t91 = MUL(t37 - t38, costab12);
275 t72 = t39 + t40; t92 = MUL(t39 - t40, costab20);
276 t73 = t41 + t42; t94 = MUL(t41 - t42, costab4);
277 t74 = t43 + t44; t95 = MUL(t43 - t44, costab28);
278 t75 = t45 + t46; t96 = MUL(t45 - t46, costab12);
279 t76 = t47 + t48; t97 = MUL(t47 - t48, costab20);
280
281 t78 = t50 + t51; t100 = MUL(t50 - t51, costab4);
282 t79 = t52 + t53; t101 = MUL(t52 - t53, costab28);
283 t80 = t54 + t55; t102 = MUL(t54 - t55, costab12);
284 t81 = t56 + t57; t103 = MUL(t56 - t57, costab20);
285
286 t83 = t59 + t60; t106 = MUL(t59 - t60, costab4);
287 t84 = t61 + t62; t107 = MUL(t61 - t62, costab28);
288 t85 = t63 + t64; t108 = MUL(t63 - t64, costab12);
289 t86 = t65 + t66; t109 = MUL(t65 - t66, costab20);
290
291 t113 = t69 + t70;
292 t114 = t71 + t72;
293
294 /* 0 */ hi[15][slot] = SHIFT(t113 + t114);
295 /* 16 */ lo[ 0][slot] = SHIFT(MUL(t113 - t114, costab16));
296
297 t115 = t73 + t74;
298 t116 = t75 + t76;
299
300 t32 = t115 + t116;
301
302 /* 1 */ hi[14][slot] = SHIFT(t32);
303
304 t118 = t78 + t79;
305 t119 = t80 + t81;
306
307 t58 = t118 + t119;
308
309 /* 2 */ hi[13][slot] = SHIFT(t58);
310
311 t121 = t83 + t84;
312 t122 = t85 + t86;
313
314 t67 = t121 + t122;
315
316 t49 = (t67 * 2) - t32;
317
318 /* 3 */ hi[12][slot] = SHIFT(t49);
319
320 t125 = t89 + t90;
321 t126 = t91 + t92;
322
323 t93 = t125 + t126;
324
325 /* 4 */ hi[11][slot] = SHIFT(t93);
326
327 t128 = t94 + t95;
328 t129 = t96 + t97;
329
330 t98 = t128 + t129;
331
332 t68 = (t98 * 2) - t49;
333
334 /* 5 */ hi[10][slot] = SHIFT(t68);
335
336 t132 = t100 + t101;
337 t133 = t102 + t103;
338
339 t104 = t132 + t133;
340
341 t82 = (t104 * 2) - t58;
342
343 /* 6 */ hi[ 9][slot] = SHIFT(t82);
344
345 t136 = t106 + t107;
346 t137 = t108 + t109;
347
348 t110 = t136 + t137;
349
350 t87 = (t110 * 2) - t67;
351
352 t77 = (t87 * 2) - t68;
353
354 /* 7 */ hi[ 8][slot] = SHIFT(t77);
355
356 t141 = MUL(t69 - t70, costab8);
357 t142 = MUL(t71 - t72, costab24);
358 t143 = t141 + t142;
359
360 /* 8 */ hi[ 7][slot] = SHIFT(t143);
361 /* 24 */ lo[ 8][slot] =
362 SHIFT((MUL(t141 - t142, costab16) * 2) - t143);
363
364 t144 = MUL(t73 - t74, costab8);
365 t145 = MUL(t75 - t76, costab24);
366 t146 = t144 + t145;
367
368 t88 = (t146 * 2) - t77;
369
370 /* 9 */ hi[ 6][slot] = SHIFT(t88);
371
372 t148 = MUL(t78 - t79, costab8);
373 t149 = MUL(t80 - t81, costab24);
374 t150 = t148 + t149;
375
376 t105 = (t150 * 2) - t82;
377
378 /* 10 */ hi[ 5][slot] = SHIFT(t105);
379
380 t152 = MUL(t83 - t84, costab8);
381 t153 = MUL(t85 - t86, costab24);
382 t154 = t152 + t153;
383
384 t111 = (t154 * 2) - t87;
385
386 t99 = (t111 * 2) - t88;
387
388 /* 11 */ hi[ 4][slot] = SHIFT(t99);
389
390 t157 = MUL(t89 - t90, costab8);
391 t158 = MUL(t91 - t92, costab24);
392 t159 = t157 + t158;
393
394 t127 = (t159 * 2) - t93;
395
396 /* 12 */ hi[ 3][slot] = SHIFT(t127);
397
398 t160 = (MUL(t125 - t126, costab16) * 2) - t127;
399
400 /* 20 */ lo[ 4][slot] = SHIFT(t160);
401 /* 28 */ lo[12][slot] =
402 SHIFT((((MUL(t157 - t158, costab16) * 2) - t159) * 2) - t160);
403
404 t161 = MUL(t94 - t95, costab8);
405 t162 = MUL(t96 - t97, costab24);
406 t163 = t161 + t162;
407
408 t130 = (t163 * 2) - t98;
409
410 t112 = (t130 * 2) - t99;
411
412 /* 13 */ hi[ 2][slot] = SHIFT(t112);
413
414 t164 = (MUL(t128 - t129, costab16) * 2) - t130;
415
416 t166 = MUL(t100 - t101, costab8);
417 t167 = MUL(t102 - t103, costab24);
418 t168 = t166 + t167;
419
420 t134 = (t168 * 2) - t104;
421
422 t120 = (t134 * 2) - t105;
423
424 /* 14 */ hi[ 1][slot] = SHIFT(t120);
425
426 t135 = (MUL(t118 - t119, costab16) * 2) - t120;
427
428 /* 18 */ lo[ 2][slot] = SHIFT(t135);
429
430 t169 = (MUL(t132 - t133, costab16) * 2) - t134;
431
432 t151 = (t169 * 2) - t135;
433
434 /* 22 */ lo[ 6][slot] = SHIFT(t151);
435
436 t170 = (((MUL(t148 - t149, costab16) * 2) - t150) * 2) - t151;
437
438 /* 26 */ lo[10][slot] = SHIFT(t170);
439 /* 30 */ lo[14][slot] =
440 SHIFT((((((MUL(t166 - t167, costab16) * 2) -
441 t168) * 2) - t169) * 2) - t170);
442
443 t171 = MUL(t106 - t107, costab8);
444 t172 = MUL(t108 - t109, costab24);
445 t173 = t171 + t172;
446
447 t138 = (t173 * 2) - t110;
448
449 t123 = (t138 * 2) - t111;
450
451 t139 = (MUL(t121 - t122, costab16) * 2) - t123;
452
453 t117 = (t123 * 2) - t112;
454
455 /* 15 */ hi[ 0][slot] = SHIFT(t117);
456
457 t124 = (MUL(t115 - t116, costab16) * 2) - t117;
458
459 /* 17 */ lo[ 1][slot] = SHIFT(t124);
460
461 t131 = (t139 * 2) - t124;
462
463 /* 19 */ lo[ 3][slot] = SHIFT(t131);
464
465 t140 = (t164 * 2) - t131;
466
467 /* 21 */ lo[ 5][slot] = SHIFT(t140);
468
469 t174 = (MUL(t136 - t137, costab16) * 2) - t138;
470
471 t155 = (t174 * 2) - t139;
472
473 t147 = (t155 * 2) - t140;
474
475 /* 23 */ lo[ 7][slot] = SHIFT(t147);
476
477 t156 = (((MUL(t144 - t145, costab16) * 2) - t146) * 2) - t147;
478
479 /* 25 */ lo[ 9][slot] = SHIFT(t156);
480
481 t175 = (((MUL(t152 - t153, costab16) * 2) - t154) * 2) - t155;
482
483 t165 = (t175 * 2) - t156;
484
485 /* 27 */ lo[11][slot] = SHIFT(t165);
486
487 t176 = (((((MUL(t161 - t162, costab16) * 2) -
488 t163) * 2) - t164) * 2) - t165;
489
490 /* 29 */ lo[13][slot] = SHIFT(t176);
491 /* 31 */ lo[15][slot] =
492 SHIFT((((((((MUL(t171 - t172, costab16) * 2) -
493 t173) * 2) - t174) * 2) - t175) * 2) - t176);
494
495 /*
496 * Totals:
497 * 80 multiplies
498 * 80 additions
499 * 119 subtractions
500 * 49 shifts (not counting SSO)
501 */
502 }
503
504 # undef MUL
505 # undef SHIFT
506
507 #endif
508
509 /* third SSO shift and/or D[] optimization preshift */
510
511 # if defined(OPT_SSO)
512 # if MAD_F_FRACBITS != 28
513 # error "MAD_F_FRACBITS must be 28 to use OPT_SSO"
514 # endif
515 # define ML0(hi, lo, x, y) ((lo) = (x) * (y))
516 # define MLA(hi, lo, x, y) ((lo) += (x) * (y))
517 # define MLN(hi, lo) ((lo) = -(lo))
518 # define MLZ(hi, lo) ((void) (hi), (mad_fixed_t) (lo))
519 # define SHIFT(x) ((x) >> 2)
520 # define PRESHIFT(x) ((MAD_F(x) + (1L << 13)) >> 14)
521 # else
522 # define ML0(hi, lo, x, y) MAD_F_ML0((hi), (lo), (x), (y))
523 # define MLA(hi, lo, x, y) MAD_F_MLA((hi), (lo), (x), (y))
524 # define MLN(hi, lo) MAD_F_MLN((hi), (lo))
525 # define MLZ(hi, lo) MAD_F_MLZ((hi), (lo))
526 # define SHIFT(x) (x)
527 # if defined(MAD_F_SCALEBITS)
528 # undef MAD_F_SCALEBITS
529 # define MAD_F_SCALEBITS (MAD_F_FRACBITS - 12)
530 # define PRESHIFT(x) (MAD_F(x) >> 12)
531 # else
532 # define PRESHIFT(x) MAD_F(x)
533 # endif
534 # endif
535
536 static
537 mad_fixed_t const D[17][32] ICONST_ATTR MEM_ALIGN_ATTR = {
538 # include "D.dat"
539 };
540
541 /*
542 * NAME: synth->full()
543 * DESCRIPTION: perform full frequency PCM synthesis
544 */
545 /* optimised version of synth_full */
546 # ifdef FPM_COLDFIRE_EMAC
547
548 #define SYNTH_EMAC1(res, f1, pD) \
549 asm volatile( \
550 "movem.l (%0), %%d0-%%d7 \n\t" \
551 "move.l (%1), %%a5 \n\t" \
552 "mac.l %%d0, %%a5, 56(%1), %%a5, %%acc0\n\t" \
553 "mac.l %%d1, %%a5, 48(%1), %%a5, %%acc0\n\t" \
554 "mac.l %%d2, %%a5, 40(%1), %%a5, %%acc0\n\t" \
555 "mac.l %%d3, %%a5, 32(%1), %%a5, %%acc0\n\t" \
556 "mac.l %%d4, %%a5, 24(%1), %%a5, %%acc0\n\t" \
557 "mac.l %%d5, %%a5, 16(%1), %%a5, %%acc0\n\t" \
558 "mac.l %%d6, %%a5, 8(%1), %%a5, %%acc0\n\t" \
559 "mac.l %%d7, %%a5, %%acc0\n\t" \
560 : \
561 : "a" (*f1), "a" (*pD) \
562 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5"); \
563 asm volatile ( \
564 "movclr.l %%acc0, %0 \n\t" \
565 : "=d" (res));
566
567 #define SYNTH_EMAC2(res, f1, f2, pD) \
568 asm volatile( \
569 "movem.l (%0), %%d0-%%d7 \n\t" \
570 "move.l 4(%1), %%a5 \n\t" \
571 "msac.l %%d0, %%a5, 60(%1), %%a5, %%acc0\n\t" \
572 "msac.l %%d1, %%a5, 52(%1), %%a5, %%acc0\n\t" \
573 "msac.l %%d2, %%a5, 44(%1), %%a5, %%acc0\n\t" \
574 "msac.l %%d3, %%a5, 36(%1), %%a5, %%acc0\n\t" \
575 "msac.l %%d4, %%a5, 28(%1), %%a5, %%acc0\n\t" \
576 "msac.l %%d5, %%a5, 20(%1), %%a5, %%acc0\n\t" \
577 "msac.l %%d6, %%a5, 12(%1), %%a5, %%acc0\n\t" \
578 "msac.l %%d7, %%a5, (%1), %%a5, %%acc0\n\t" \
579 "movem.l (%2), %%d0-%%d7 \n\t" \
580 "mac.l %%d0, %%a5, 56(%1), %%a5, %%acc0\n\t" \
581 "mac.l %%d1, %%a5, 48(%1), %%a5, %%acc0\n\t" \
582 "mac.l %%d2, %%a5, 40(%1), %%a5, %%acc0\n\t" \
583 "mac.l %%d3, %%a5, 32(%1), %%a5, %%acc0\n\t" \
584 "mac.l %%d4, %%a5, 24(%1), %%a5, %%acc0\n\t" \
585 "mac.l %%d5, %%a5, 16(%1), %%a5, %%acc0\n\t" \
586 "mac.l %%d6, %%a5, 8(%1), %%a5, %%acc0\n\t" \
587 "mac.l %%d7, %%a5, %%acc0\n\t" \
588 : \
589 : "a" (*f1), "a" (*pD), "a" (*f2) \
590 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5", "memory"); \
591 asm volatile ( \
592 "movclr.l %%acc0, %0 \n\t" \
593 : "=d" (res));
594
595 #define SYNTH_EMAC_ODD_SBSAMPLE(f1, f2, pD1, pD2, res1, res2) \
596 asm volatile ( \
597 "movem.l (%0), %%d0-%%d7 \n\t" \
598 "move.l 4(%2), %%a5 \n\t" \
599 "msac.l %%d0, %%a5, 60(%2), %%a5, %%acc0\n\t" \
600 "msac.l %%d1, %%a5, 52(%2), %%a5, %%acc0\n\t" \
601 "msac.l %%d2, %%a5, 44(%2), %%a5, %%acc0\n\t" \
602 "msac.l %%d3, %%a5, 36(%2), %%a5, %%acc0\n\t" \
603 "msac.l %%d4, %%a5, 28(%2), %%a5, %%acc0\n\t" \
604 "msac.l %%d5, %%a5, 20(%2), %%a5, %%acc0\n\t" \
605 "msac.l %%d6, %%a5, 12(%2), %%a5, %%acc0\n\t" \
606 "msac.l %%d7, %%a5, 112(%3), %%a5, %%acc0\n\t" \
607 "mac.l %%d7, %%a5, 104(%3), %%a5, %%acc1\n\t" \
608 "mac.l %%d6, %%a5, 96(%3), %%a5, %%acc1\n\t" \
609 "mac.l %%d5, %%a5, 88(%3), %%a5, %%acc1\n\t" \
610 "mac.l %%d4, %%a5, 80(%3), %%a5, %%acc1\n\t" \
611 "mac.l %%d3, %%a5, 72(%3), %%a5, %%acc1\n\t" \
612 "mac.l %%d2, %%a5, 64(%3), %%a5, %%acc1\n\t" \
613 "mac.l %%d1, %%a5, 120(%3), %%a5, %%acc1\n\t" \
614 "mac.l %%d0, %%a5, 8(%2), %%a5, %%acc1\n\t" \
615 "movem.l (%1), %%d0-%%d7 \n\t" \
616 "mac.l %%d7, %%a5, 16(%2), %%a5, %%acc0\n\t" \
617 "mac.l %%d6, %%a5, 24(%2), %%a5, %%acc0\n\t" \
618 "mac.l %%d5, %%a5, 32(%2), %%a5, %%acc0\n\t" \
619 "mac.l %%d4, %%a5, 40(%2), %%a5, %%acc0\n\t" \
620 "mac.l %%d3, %%a5, 48(%2), %%a5, %%acc0\n\t" \
621 "mac.l %%d2, %%a5, 56(%2), %%a5, %%acc0\n\t" \
622 "mac.l %%d1, %%a5, (%2), %%a5, %%acc0\n\t" \
623 "mac.l %%d0, %%a5, 60(%3), %%a5, %%acc0\n\t" \
624 "mac.l %%d0, %%a5, 68(%3), %%a5, %%acc1\n\t" \
625 "mac.l %%d1, %%a5, 76(%3), %%a5, %%acc1\n\t" \
626 "mac.l %%d2, %%a5, 84(%3), %%a5, %%acc1\n\t" \
627 "mac.l %%d3, %%a5, 92(%3), %%a5, %%acc1\n\t" \
628 "mac.l %%d4, %%a5, 100(%3), %%a5, %%acc1\n\t" \
629 "mac.l %%d5, %%a5, 108(%3), %%a5, %%acc1\n\t" \
630 "mac.l %%d6, %%a5, 116(%3), %%a5, %%acc1\n\t" \
631 "mac.l %%d7, %%a5, %%acc1\n\t" \
632 : \
633 : "a" (*f1), "a" (*f2), "a" (*pD1), "a" (*pD2) \
634 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5", "memory"); \
635 asm volatile( \
636 "movclr.l %%acc0, %0\n\t" \
637 "movclr.l %%acc1, %1\n\t" \
638 : "=d" (res1), "=d" (res2) );
639
640 #define SYNTH_EMAC_EVEN_SBSAMPLE(f1, f2, pD1, pD2, res1, res2) \
641 asm volatile ( \
642 "movem.l (%0), %%d0-%%d7 \n\t" \
643 "move.l (%2), %%a5 \n\t" \
644 "msac.l %%d0, %%a5, 56(%2), %%a5, %%acc0\n\t" \
645 "msac.l %%d1, %%a5, 48(%2), %%a5, %%acc0\n\t" \
646 "msac.l %%d2, %%a5, 40(%2), %%a5, %%acc0\n\t" \
647 "msac.l %%d3, %%a5, 32(%2), %%a5, %%acc0\n\t" \
648 "msac.l %%d4, %%a5, 24(%2), %%a5, %%acc0\n\t" \
649 "msac.l %%d5, %%a5, 16(%2), %%a5, %%acc0\n\t" \
650 "msac.l %%d6, %%a5, 8(%2), %%a5, %%acc0\n\t" \
651 "msac.l %%d7, %%a5, 116(%3), %%a5, %%acc0\n\t" \
652 "mac.l %%d7, %%a5, 108(%3), %%a5, %%acc1\n\t" \
653 "mac.l %%d6, %%a5, 100(%3), %%a5, %%acc1\n\t" \
654 "mac.l %%d5, %%a5, 92(%3), %%a5, %%acc1\n\t" \
655 "mac.l %%d4, %%a5, 84(%3), %%a5, %%acc1\n\t" \
656 "mac.l %%d3, %%a5, 76(%3), %%a5, %%acc1\n\t" \
657 "mac.l %%d2, %%a5, 68(%3), %%a5, %%acc1\n\t" \
658 "mac.l %%d1, %%a5, 60(%3), %%a5, %%acc1\n\t" \
659 "mac.l %%d0, %%a5, 12(%2), %%a5, %%acc1\n\t" \
660 "movem.l (%1), %%d0-%%d7 \n\t" \
661 "mac.l %%d7, %%a5, 20(%2), %%a5, %%acc0\n\t" \
662 "mac.l %%d6, %%a5, 28(%2), %%a5, %%acc0\n\t" \
663 "mac.l %%d5, %%a5, 36(%2), %%a5, %%acc0\n\t" \
664 "mac.l %%d4, %%a5, 44(%2), %%a5, %%acc0\n\t" \
665 "mac.l %%d3, %%a5, 52(%2), %%a5, %%acc0\n\t" \
666 "mac.l %%d2, %%a5, 60(%2), %%a5, %%acc0\n\t" \
667 "mac.l %%d1, %%a5, 4(%2), %%a5, %%acc0\n\t" \
668 "mac.l %%d0, %%a5, 120(%3), %%a5, %%acc0\n\t" \
669 "mac.l %%d0, %%a5, 64(%3), %%a5, %%acc1\n\t" \
670 "mac.l %%d1, %%a5, 72(%3), %%a5, %%acc1\n\t" \
671 "mac.l %%d2, %%a5, 80(%3), %%a5, %%acc1\n\t" \
672 "mac.l %%d3, %%a5, 88(%3), %%a5, %%acc1\n\t" \
673 "mac.l %%d4, %%a5, 96(%3), %%a5, %%acc1\n\t" \
674 "mac.l %%d5, %%a5, 104(%3), %%a5, %%acc1\n\t" \
675 "mac.l %%d6, %%a5, 112(%3), %%a5, %%acc1\n\t" \
676 "mac.l %%d7, %%a5, %%acc1\n\t" \
677 : \
678 : "a" (*f1), "a" (*f2), "a" (*pD1), "a" (*pD2) \
679 : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "a5", "memory"); \
680 asm volatile( \
681 "movclr.l %%acc0, %0\n\t" \
682 "movclr.l %%acc1, %1\n\t" \
683 : "=d" (res1), "=d" (res2) );
684
685 static
686 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
687 unsigned int nch, unsigned int ns)
688 {
689 int sb;
690 unsigned int phase, ch, s, p;
691 mad_fixed_t *pcm, (*filter)[2][2][16][8];
692 mad_fixed_t (*sbsample)[36][32];
693 mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
694 mad_fixed_t const (*D0ptr)[32];
695 mad_fixed_t const (*D1ptr)[32];
696 mad_fixed64hi_t hi0, hi1;
697
698 for (ch = 0; ch < nch; ++ch) {
699 sbsample = &(*frame->sbsample_prev)[ch];
700 filter = &synth->filter[ch];
701 phase = synth->phase;
702 pcm = synth->pcm.samples[ch];
703
704 for (s = 0; s < ns; ++s) {
705 dct32((*sbsample)[s], phase >> 1,
706 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
707
708 p = (phase - 1) & 0xf;
709
710 /* calculate 32 samples */
711 fe = &(*filter)[0][ phase & 1][0];
712 fx = &(*filter)[0][~phase & 1][0];
713 fo = &(*filter)[1][~phase & 1][0];
714
715 D0ptr = (void*)&D[0][ p];
716 D1ptr = (void*)&D[0][-p];
717
718 if(s & 1)
719 {
720 SYNTH_EMAC2(hi0, fx, fe, D0ptr);
721 pcm[0] = hi0 << 3; /* shift result to libmad's fixed point format */
722 pcm += 16;
723
724 for (sb = 15; sb; sb--, fo++) {
725 ++fe;
726 ++D0ptr;
727 ++D1ptr;
728
729 /* D[32 - sb][i] == -D[sb][31 - i] */
730 SYNTH_EMAC_ODD_SBSAMPLE(fo, fe, D0ptr, D1ptr, hi0, hi1);
731 pcm[-sb] = hi0 << 3;
732 pcm[ sb] = hi1 << 3;
733 }
734
735 ++D0ptr;
736 SYNTH_EMAC1(hi0, fo, D0ptr+1);
737 pcm[0] = -(hi0 << 3);
738 }
739 else
740 {
741 SYNTH_EMAC2(hi0, fe, fx, D0ptr);
742 pcm[0] = -(hi0 << 3); /* shift result to libmad's fixed point format */
743 pcm += 16;
744
745 for (sb = 15; sb; sb--, fo++) {
746 ++fe;
747 ++D0ptr;
748 ++D1ptr;
749
750 /* D[32 - sb][i] == -D[sb][31 - i] */
751 SYNTH_EMAC_EVEN_SBSAMPLE(fo, fe, D0ptr, D1ptr, hi0, hi1);
752 pcm[-sb] = hi0 << 3;
753 pcm[ sb] = hi1 << 3;
754 }
755
756 ++D0ptr;
757 SYNTH_EMAC1(hi0, fo, D0ptr);
758 pcm[0] = -(hi0 << 3);
759 }
760 pcm += 16;
761 phase = (phase + 1) % 16;
762 }
763 }
764 }
765
766 #elif defined(FPM_ARM)
767
768 #define PROD_O(hi, lo, f, ptr) \
769 ({ \
770 mad_fixed_t *__p = (f); \
771 asm("ldmia %2!, {r0, r1, r2, r3}\n\t" \
772 "ldr r4, [%3, #0]\n\t" \
773 "ldr r12, [%3, #56]\n\t" \
774 "smull %0, %1, r0, r4\n\t" \
775 "ldr r4, [%3, #48]\n\t" \
776 "smlal %0, %1, r1, r12\n\t" \
777 "ldr r12, [%3, #40]\n\t" \
778 "smlal %0, %1, r2, r4\n\t" \
779 "smlal %0, %1, r3, r12\n\t" \
780 "ldmia %2, {r0, r1, r2, r3}\n\t" \
781 "ldr r4, [%3, #32]\n\t" \
782 "ldr r12, [%3, #24]\n\t" \
783 "smlal %0, %1, r0, r4\n\t" \
784 "ldr r4, [%3, #16]\n\t" \
785 "smlal %0, %1, r1, r12\n\t" \
786 "ldr r12, [%3, #8]\n\t" \
787 "smlal %0, %1, r2, r4\n\t" \
788 "smlal %0, %1, r3, r12\n\t" \
789 : "=&r" (lo), "=&r" (hi), "+r" (__p) \
790 : "r" (ptr) \
791 : "r0", "r1", "r2", "r3", "r4", "r12"); \
792 })
793
794 #define PROD_A(hi, lo, f, ptr) \
795 ({ \
796 mad_fixed_t *__p = (f); \
797 asm("ldmia %2!, {r0, r1, r2, r3}\n\t" \
798 "ldr r4, [%3, #0]\n\t" \
799 "ldr r12, [%3, #56]\n\t" \
800 "smlal %0, %1, r0, r4\n\t" \
801 "ldr r4, [%3, #48]\n\t" \
802 "smlal %0, %1, r1, r12\n\t" \
803 "ldr r12, [%3, #40]\n\t" \
804 "smlal %0, %1, r2, r4\n\t" \
805 "smlal %0, %1, r3, r12\n\t" \
806 "ldmia %2, {r0, r1, r2, r3}\n\t" \
807 "ldr r4, [%3, #32]\n\t" \
808 "ldr r12, [%3, #24]\n\t" \
809 "smlal %0, %1, r0, r4\n\t" \
810 "ldr r4, [%3, #16]\n\t" \
811 "smlal %0, %1, r1, r12\n\t" \
812 "ldr r12, [%3, #8]\n\t" \
813 "smlal %0, %1, r2, r4\n\t" \
814 "smlal %0, %1, r3, r12\n\t" \
815 : "+r" (lo), "+r" (hi), "+r" (__p) \
816 : "r" (ptr) \
817 : "r0", "r1", "r2", "r3", "r4", "r12"); \
818 })
819
820 void synth_full_odd_sbsample (mad_fixed_t *pcm,
821 mad_fixed_t (*fo)[8],
822 mad_fixed_t (*fe)[8],
823 mad_fixed_t const (*D0ptr)[32],
824 mad_fixed_t const (*D1ptr)[32]);
825 void synth_full_even_sbsample(mad_fixed_t *pcm,
826 mad_fixed_t (*fo)[8],
827 mad_fixed_t (*fe)[8],
828 mad_fixed_t const (*D0ptr)[32],
829 mad_fixed_t const (*D1ptr)[32]);
830
831 static
832 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
833 unsigned int nch, unsigned int ns) ICODE_ATTR_MPA_SYNTH;
834 static
835 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
836 unsigned int nch, unsigned int ns)
837 {
838 int p;
839 unsigned int phase, ch, s;
840 mad_fixed_t *pcm, (*filter)[2][2][16][8];
841 mad_fixed_t (*sbsample)[36][32];
842 mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
843 mad_fixed_t const (*D0ptr)[32], *ptr;
844 mad_fixed_t const (*D1ptr)[32];
845 mad_fixed64hi_t hi;
846 mad_fixed64lo_t lo;
847
848 for (ch = 0; ch < nch; ++ch) {
849 sbsample = &(*frame->sbsample_prev)[ch];
850 filter = &synth->filter[ch];
851 phase = synth->phase;
852 pcm = synth->pcm.samples[ch];
853
854 for (s = 0; s < ns; ++s) {
855 dct32((*sbsample)[s], phase >> 1,
856 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
857
858 p = (phase - 1) & 0xf;
859
860 /* calculate 32 samples */
861 fe = &(*filter)[0][ phase & 1][0];
862 fx = &(*filter)[0][~phase & 1][0];
863 fo = &(*filter)[1][~phase & 1][0];
864
865 D0ptr = (void*)&D[0][ p];
866 D1ptr = (void*)&D[0][-p];
867
868 if(s & 1)
869 {
870 ptr = *D0ptr;
871 PROD_O(hi, lo, *fx, ptr+1);
872 MLN(hi, lo);
873 PROD_A(hi, lo, *fe, ptr);
874 pcm[0] = SHIFT(MLZ(hi, lo));
875 pcm += 16;
876
877 synth_full_odd_sbsample(pcm, fo, fe, D0ptr, D1ptr);
878 D0ptr += 15;
879 D1ptr += 15;
880 fo += 15;
881 fe += 15;
882
883 ptr = *(D0ptr + 1);
884 PROD_O(hi, lo, *fo, ptr+1);
885 pcm[0] = SHIFT(-MLZ(hi, lo));
886 }
887 else
888 {
889 ptr = *D0ptr;
890 PROD_O(hi, lo, *fx, ptr);
891 MLN(hi, lo);
892 PROD_A(hi, lo, *fe, ptr+1);
893 pcm[0] = SHIFT(MLZ(hi, lo));
894 pcm += 16;
895
896 synth_full_even_sbsample(pcm, fo, fe, D0ptr, D1ptr);
897 D0ptr += 15;
898 D1ptr += 15;
899 fo += 15;
900 fe += 15;
901
902 ptr = *(D0ptr + 1);
903 PROD_O(hi, lo, *fo, ptr);
904 pcm[0] = SHIFT(-MLZ(hi, lo));
905 }
906
907 pcm += 16;
908 phase = (phase + 1) % 16;
909 }
910 }
911 }
912
913 # else /* not FPM_COLDFIRE_EMAC and not FPM_ARM */
914
915 #define PROD_O(hi, lo, f, ptr, offset) \
916 ML0(hi, lo, (*f)[0], ptr[ 0+offset]); \
917 MLA(hi, lo, (*f)[1], ptr[14+offset]); \
918 MLA(hi, lo, (*f)[2], ptr[12+offset]); \
919 MLA(hi, lo, (*f)[3], ptr[10+offset]); \
920 MLA(hi, lo, (*f)[4], ptr[ 8+offset]); \
921 MLA(hi, lo, (*f)[5], ptr[ 6+offset]); \
922 MLA(hi, lo, (*f)[6], ptr[ 4+offset]); \
923 MLA(hi, lo, (*f)[7], ptr[ 2+offset]);
924
925 #define PROD_A(hi, lo, f, ptr, offset) \
926 MLA(hi, lo, (*f)[0], ptr[ 0+offset]); \
927 MLA(hi, lo, (*f)[1], ptr[14+offset]); \
928 MLA(hi, lo, (*f)[2], ptr[12+offset]); \
929 MLA(hi, lo, (*f)[3], ptr[10+offset]); \
930 MLA(hi, lo, (*f)[4], ptr[ 8+offset]); \
931 MLA(hi, lo, (*f)[5], ptr[ 6+offset]); \
932 MLA(hi, lo, (*f)[6], ptr[ 4+offset]); \
933 MLA(hi, lo, (*f)[7], ptr[ 2+offset]);
934
935 #define PROD_SB(hi, lo, ptr, offset, first_idx, last_idx) \
936 ML0(hi, lo, (*fe)[0], ptr[first_idx]); \
937 MLA(hi, lo, (*fe)[1], ptr[16+offset]); \
938 MLA(hi, lo, (*fe)[2], ptr[18+offset]); \
939 MLA(hi, lo, (*fe)[3], ptr[20+offset]); \
940 MLA(hi, lo, (*fe)[4], ptr[22+offset]); \
941 MLA(hi, lo, (*fe)[5], ptr[24+offset]); \
942 MLA(hi, lo, (*fe)[6], ptr[26+offset]); \
943 MLA(hi, lo, (*fe)[7], ptr[28+offset]); \
944 MLA(hi, lo, (*fo)[7], ptr[29-offset]); \
945 MLA(hi, lo, (*fo)[6], ptr[27-offset]); \
946 MLA(hi, lo, (*fo)[5], ptr[25-offset]); \
947 MLA(hi, lo, (*fo)[4], ptr[23-offset]); \
948 MLA(hi, lo, (*fo)[3], ptr[21-offset]); \
949 MLA(hi, lo, (*fo)[2], ptr[19-offset]); \
950 MLA(hi, lo, (*fo)[1], ptr[17-offset]); \
951 MLA(hi, lo, (*fo)[0], ptr[last_idx ]);
952
953 static
954 void synth_full(struct mad_synth *synth, struct mad_frame const *frame,
955 unsigned int nch, unsigned int ns)
956 {
957 int p, sb;
958 unsigned int phase, ch, s;
959 mad_fixed_t *pcm, (*filter)[2][2][16][8];
960 mad_fixed_t (*sbsample)[36][32];
961 mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
962 mad_fixed_t const (*D0ptr)[32], *ptr;
963 mad_fixed_t const (*D1ptr)[32];
964 mad_fixed64hi_t hi;
965 mad_fixed64lo_t lo;
966
967 for (ch = 0; ch < nch; ++ch) {
968 sbsample = &(*frame->sbsample_prev)[ch];
969 filter = &synth->filter[ch];
970 phase = synth->phase;
971 pcm = synth->pcm.samples[ch];
972
973 for (s = 0; s < ns; ++s) {
974 dct32((*sbsample)[s], phase >> 1,
975 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
976
977 p = (phase - 1) & 0xf;
978
979 /* calculate 32 samples */
980 fe = &(*filter)[0][ phase & 1][0];
981 fx = &(*filter)[0][~phase & 1][0];
982 fo = &(*filter)[1][~phase & 1][0];
983
984 D0ptr = (void*)&D[0][ p];
985 D1ptr = (void*)&D[0][-p];
986
987 if(s & 1)
988 {
989 ptr = *D0ptr;
990 PROD_O(hi, lo, fx, ptr, 1)
991 MLN(hi, lo);
992 PROD_A(hi, lo, fe, ptr, 0)
993 pcm[0] = SHIFT(MLZ(hi, lo));
994 pcm += 16;
995
996 for (sb = 15; sb; sb--, fo++)
997 {
998 ++fe;
999 ++D0ptr;
1000 ++D1ptr;
1001
1002 /* D[32 - sb][i] == -D[sb][31 - i] */
1003 ptr = *D0ptr;
1004 PROD_O(hi, lo, fo, ptr, 1)
1005 MLN(hi, lo);
1006 PROD_A(hi, lo, fe, ptr, 0)
1007 pcm[-sb] = SHIFT(MLZ(hi, lo));
1008
1009 ptr = *D1ptr;
1010 PROD_SB(hi, lo, ptr, 1, 15, 30)
1011 pcm[sb] = SHIFT(MLZ(hi, lo));
1012 }
1013
1014 ptr = *(D0ptr + 1);
1015 PROD_O(hi, lo, fo, ptr, 1)
1016 pcm[0] = SHIFT(-MLZ(hi, lo));
1017 }
1018 else
1019 {
1020 ptr = *D0ptr;
1021 PROD_O(hi, lo, fx, ptr, 0)
1022 MLN(hi, lo);
1023 PROD_A(hi, lo, fe, ptr, 1)
1024 pcm[0] = SHIFT(MLZ(hi, lo));
1025 pcm += 16;
1026
1027 for (sb = 15; sb; sb--, fo++)
1028 {
1029 ++fe;
1030 ++D0ptr;
1031 ++D1ptr;
1032
1033 /* D[32 - sb][i] == -D[sb][31 - i] */
1034 ptr = *D0ptr;
1035 PROD_O(hi, lo, fo, ptr, 0)
1036 MLN(hi, lo);
1037 PROD_A(hi, lo, fe, ptr, 1)
1038 pcm[-sb] = SHIFT(MLZ(hi, lo));
1039
1040 ptr = *D1ptr;
1041 PROD_SB(hi, lo, ptr, 0, 30, 15)
1042 pcm[sb] = SHIFT(MLZ(hi, lo));
1043 }
1044
1045 ptr = *(D0ptr + 1);
1046 PROD_O(hi, lo, fo, ptr, 0)
1047 pcm[0] = SHIFT(-MLZ(hi, lo));
1048 }
1049
1050 pcm += 16;
1051 phase = (phase + 1) % 16;
1052 }
1053 }
1054 }
1055 # endif /* FPM_COLDFIRE_EMAC, FPM_ARM */
1056
1057 #if 0 /* rockbox: unused */
1058 /*
1059 * NAME: synth->half()
1060 * DESCRIPTION: perform half frequency PCM synthesis
1061 */
1062 static
1063 void synth_half(struct mad_synth *synth, struct mad_frame const *frame,
1064 unsigned int nch, unsigned int ns)
1065 {
1066 unsigned int phase, ch, s, sb, pe, po;
1067 mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8];
1068 mad_fixed_t (*sbsample)[36][32];
1069 register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8];
1070 register mad_fixed_t const (*Dptr)[32], *ptr;
1071 register mad_fixed64hi_t hi;
1072 register mad_fixed64lo_t lo;
1073
1074 for (ch = 0; ch < nch; ++ch) {
1075 sbsample = &(*frame->sbsample_prev)[ch];
1076 filter = &synth->filter[ch];
1077 phase = synth->phase;
1078 pcm1 = synth->pcm.samples[ch];
1079
1080 for (s = 0; s < ns; ++s) {
1081 dct32((*sbsample)[s], phase >> 1,
1082 (*filter)[0][phase & 1], (*filter)[1][phase & 1]);
1083
1084 pe = phase & ~1;
1085 po = ((phase - 1) & 0xf) | 1;
1086
1087 /* calculate 16 samples */
1088
1089 fe = &(*filter)[0][ phase & 1][0];
1090 fx = &(*filter)[0][~phase & 1][0];
1091 fo = &(*filter)[1][~phase & 1][0];
1092
1093 Dptr = &D[0];
1094
1095 ptr = *Dptr + po;
1096 ML0(hi, lo, (*fx)[0], ptr[ 0]);
1097 MLA(hi, lo, (*fx)[1], ptr[14]);
1098 MLA(hi, lo, (*fx)[2], ptr[12]);
1099 MLA(hi, lo, (*fx)[3], ptr[10]);
1100 MLA(hi, lo, (*fx)[4], ptr[ 8]);
1101 MLA(hi, lo, (*fx)[5], ptr[ 6]);
1102 MLA(hi, lo, (*fx)[6], ptr[ 4]);
1103 MLA(hi, lo, (*fx)[7], ptr[ 2]);
1104 MLN(hi, lo);
1105
1106 ptr = *Dptr + pe;
1107 MLA(hi, lo, (*fe)[0], ptr[ 0]);
1108 MLA(hi, lo, (*fe)[1], ptr[14]);
1109 MLA(hi, lo, (*fe)[2], ptr[12]);
1110 MLA(hi, lo, (*fe)[3], ptr[10]);
1111 MLA(hi, lo, (*fe)[4], ptr[ 8]);
1112 MLA(hi, lo, (*fe)[5], ptr[ 6]);
1113 MLA(hi, lo, (*fe)[6], ptr[ 4]);
1114 MLA(hi, lo, (*fe)[7], ptr[ 2]);
1115
1116 *pcm1++ = SHIFT(MLZ(hi, lo));
1117
1118 pcm2 = pcm1 + 14;
1119
1120 for (sb = 1; sb < 16; ++sb) {
1121 ++fe;
1122 ++Dptr;
1123
1124 /* D[32 - sb][i] == -D[sb][31 - i] */
1125
1126 if (!(sb & 1)) {
1127 ptr = *Dptr + po;
1128 ML0(hi, lo, (*fo)[0], ptr[ 0]);
1129 MLA(hi, lo, (*fo)[1], ptr[14]);
1130 MLA(hi, lo, (*fo)[2], ptr[12]);
1131 MLA(hi, lo, (*fo)[3], ptr[10]);
1132 MLA(hi, lo, (*fo)[4], ptr[ 8]);
1133 MLA(hi, lo, (*fo)[5], ptr[ 6]);
1134 MLA(hi, lo, (*fo)[6], ptr[ 4]);
1135 MLA(hi, lo, (*fo)[7], ptr[ 2]);
1136 MLN(hi, lo);
1137
1138 ptr = *Dptr + pe;
1139 MLA(hi, lo, (*fe)[7], ptr[ 2]);
1140 MLA(hi, lo, (*fe)[6], ptr[ 4]);
1141 MLA(hi, lo, (*fe)[5], ptr[ 6]);
1142 MLA(hi, lo, (*fe)[4], ptr[ 8]);
1143 MLA(hi, lo, (*fe)[3], ptr[10]);
1144 MLA(hi, lo, (*fe)[2], ptr[12]);
1145 MLA(hi, lo, (*fe)[1], ptr[14]);
1146 MLA(hi, lo, (*fe)[0], ptr[ 0]);
1147
1148 *pcm1++ = SHIFT(MLZ(hi, lo));
1149
1150 ptr = *Dptr - po;
1151 ML0(hi, lo, (*fo)[7], ptr[31 - 2]);
1152 MLA(hi, lo, (*fo)[6], ptr[31 - 4]);
1153 MLA(hi, lo, (*fo)[5], ptr[31 - 6]);
1154 MLA(hi, lo, (*fo)[4], ptr[31 - 8]);
1155 MLA(hi, lo, (*fo)[3], ptr[31 - 10]);
1156 MLA(hi, lo, (*fo)[2], ptr[31 - 12]);
1157 MLA(hi, lo, (*fo)[1], ptr[31 - 14]);
1158 MLA(hi, lo, (*fo)[0], ptr[31 - 16]);
1159
1160 ptr = *Dptr - pe;
1161 MLA(hi, lo, (*fe)[0], ptr[31 - 16]);
1162 MLA(hi, lo, (*fe)[1], ptr[31 - 14]);
1163 MLA(hi, lo, (*fe)[2], ptr[31 - 12]);
1164 MLA(hi, lo, (*fe)[3], ptr[31 - 10]);
1165 MLA(hi, lo, (*fe)[4], ptr[31 - 8]);
1166 MLA(hi, lo, (*fe)[5], ptr[31 - 6]);
1167 MLA(hi, lo, (*fe)[6], ptr[31 - 4]);
1168 MLA(hi, lo, (*fe)[7], ptr[31 - 2]);
1169
1170 *pcm2-- = SHIFT(MLZ(hi, lo));
1171 }
1172
1173 ++fo;
1174 }
1175
1176 ++Dptr;
1177
1178 ptr = *Dptr + po;
1179 ML0(hi, lo, (*fo)[0], ptr[ 0]);
1180 MLA(hi, lo, (*fo)[1], ptr[14]);
1181 MLA(hi, lo, (*fo)[2], ptr[12]);
1182 MLA(hi, lo, (*fo)[3], ptr[10]);
1183 MLA(hi, lo, (*fo)[4], ptr[ 8]);
1184 MLA(hi, lo, (*fo)[5], ptr[ 6]);
1185 MLA(hi, lo, (*fo)[6], ptr[ 4]);
1186 MLA(hi, lo, (*fo)[7], ptr[ 2]);
1187
1188 *pcm1 = SHIFT(-MLZ(hi, lo));
1189 pcm1 += 8;
1190
1191 phase = (phase + 1) % 16;
1192 }
1193 }
1194 }
1195 #endif /* unused */
1196
1197 /*
1198 * NAME: synth->frame()
1199 * DESCRIPTION: perform PCM synthesis of frame subband samples
1200 */
1201 void mad_synth_frame(struct mad_synth *synth, struct mad_frame const *frame)
1202 {
1203 unsigned int nch, ns;
1204 #if 0 /* rockbox: unused */
1205 void (*synth_frame)(struct mad_synth *, struct mad_frame const *,
1206 unsigned int, unsigned int);
1207 #endif
1208
1209 nch = MAD_NCHANNELS(&frame->header);
1210 ns = MAD_NSBSAMPLES(&frame->header);
1211
1212 synth->pcm.samplerate = frame->header.samplerate;
1213 synth->pcm.channels = nch;
1214 synth->pcm.length = 32 * ns;
1215
1216 #if 0 /* rockbox: unused */
1217 synth_frame = synth_full;
1218
1219 if (frame->options & MAD_OPTION_HALFSAMPLERATE) {
1220 synth->pcm.samplerate /= 2;
1221 synth->pcm.length /= 2;
1222
1223 synth_frame = synth_half;
1224 }
1225
1226 synth_frame(synth, frame, nch, ns);
1227 #else
1228 synth_full(synth, frame, nch, ns);
1229 #endif
1230
1231 synth->phase = (synth->phase + ns) % 16;
1232 }
My public Rockbox git repo