blob: 844af99c282a17aa757cf4cde0a7c4166ca45cd4
1 | /* |
2 | * libmad - MPEG audio decoder library |
3 | * Copyright (C) 2000-2004 Underbit Technologies, Inc. |
4 | * |
5 | * This program is free software; you can redistribute it and/or modify |
6 | * it under the terms of the GNU General Public License as published by |
7 | * the Free Software Foundation; either version 2 of the License, or |
8 | * (at your option) any later version. |
9 | * |
10 | * This program is distributed in the hope that it will be useful, |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
13 | * GNU General Public License for more details. |
14 | * |
15 | * You should have received a copy of the GNU General Public License |
16 | * along with this program; if not, write to the Free Software |
17 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
18 | * |
19 | * $Id: synth.c,v 1.25 2004/01/23 09:41:33 rob Exp $ |
20 | */ |
21 | |
22 | # 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]; |
219 | t16 = MUL(in[0] - in[31], costab1); |
220 | t1 = in[15] + in[16]; |
221 | t17 = MUL(in[15] - in[16], costab31); |
222 | |
223 | t41 = t16 + t17; |
224 | t59 = MUL(t16 - t17, costab2); |
225 | t33 = t0 + t1; |
226 | t50 = MUL(t0 - t1, costab2); |
227 | |
228 | t2 = in[7] + in[24]; |
229 | t18 = MUL(in[7] - in[24], costab15); |
230 | t3 = in[8] + in[23]; |
231 | t19 = MUL(in[8] - in[23], costab17); |
232 | |
233 | t42 = t18 + t19; |
234 | t60 = MUL(t18 - t19, costab30); |
235 | t34 = t2 + t3; |
236 | t51 = MUL(t2 - t3, costab30); |
237 | |
238 | t4 = in[3] + in[28]; |
239 | t20 = MUL(in[3] - in[28], costab7); |
240 | t5 = in[12] + in[19]; |
241 | t21 = MUL(in[12] - in[19], costab25); |
242 | |
243 | t43 = t20 + t21; |
244 | t61 = MUL(t20 - t21, costab14); |
245 | t35 = t4 + t5; |
246 | t52 = MUL(t4 - t5, costab14); |
247 | |
248 | t6 = in[4] + in[27]; |
249 | t22 = MUL(in[4] - in[27], costab9); |
250 | t7 = in[11] + in[20]; |
251 | t23 = MUL(in[11] - in[20], costab23); |
252 | |
253 | t44 = t22 + t23; |
254 | t62 = MUL(t22 - t23, costab18); |
255 | t36 = t6 + t7; |
256 | t53 = MUL(t6 - t7, costab18); |
257 | |
258 | t8 = in[1] + in[30]; |
259 | t24 = MUL(in[1] - in[30], costab3); |
260 | t9 = in[14] + in[17]; |
261 | t25 = MUL(in[14] - in[17], costab29); |
262 | |
263 | t45 = t24 + t25; |
264 | t63 = MUL(t24 - t25, costab6); |
265 | t37 = t8 + t9; |
266 | t54 = MUL(t8 - t9, costab6); |
267 | |
268 | t10 = in[6] + in[25]; |
269 | t26 = MUL(in[6] - in[25], costab13); |
270 | t11 = in[9] + in[22]; |
271 | t27 = MUL(in[9] - in[22], costab19); |
272 | |
273 | t46 = t26 + t27; |
274 | t64 = MUL(t26 - t27, costab26); |
275 | t38 = t10 + t11; |
276 | t55 = MUL(t10 - t11, costab26); |
277 | |
278 | t12 = in[2] + in[29]; |
279 | t28 = MUL(in[2] - in[29], costab5); |
280 | t13 = in[13] + in[18]; |
281 | t29 = MUL(in[13] - in[18], costab27); |
282 | |
283 | t47 = t28 + t29; |
284 | t65 = MUL(t28 - t29, costab10); |
285 | t39 = t12 + t13; |
286 | t56 = MUL(t12 - t13, costab10); |
287 | |
288 | t14 = in[5] + in[26]; |
289 | t30 = MUL(in[5] - in[26], costab11); |
290 | t15 = in[10] + in[21]; |
291 | t31 = MUL(in[10] - in[21], costab21); |
292 | |
293 | t48 = t30 + t31; |
294 | t66 = MUL(t30 - t31, costab22); |
295 | t40 = t14 + t15; |
296 | t57 = MUL(t14 - t15, costab22); |
297 | |
298 | t69 = t33 + t34; |
299 | t89 = MUL(t33 - t34, costab4); |
300 | t70 = t35 + t36; |
301 | t90 = MUL(t35 - t36, costab28); |
302 | t71 = t37 + t38; |
303 | t91 = MUL(t37 - t38, costab12); |
304 | t72 = t39 + t40; |
305 | t92 = MUL(t39 - t40, costab20); |
306 | t73 = t41 + t42; |
307 | t94 = MUL(t41 - t42, costab4); |
308 | t74 = t43 + t44; |
309 | t95 = MUL(t43 - t44, costab28); |
310 | t75 = t45 + t46; |
311 | t96 = MUL(t45 - t46, costab12); |
312 | t76 = t47 + t48; |
313 | t97 = MUL(t47 - t48, costab20); |
314 | |
315 | t78 = t50 + t51; |
316 | t100 = MUL(t50 - t51, costab4); |
317 | t79 = t52 + t53; |
318 | t101 = MUL(t52 - t53, costab28); |
319 | t80 = t54 + t55; |
320 | t102 = MUL(t54 - t55, costab12); |
321 | t81 = t56 + t57; |
322 | t103 = MUL(t56 - t57, costab20); |
323 | |
324 | t83 = t59 + t60; |
325 | t106 = MUL(t59 - t60, costab4); |
326 | t84 = t61 + t62; |
327 | t107 = MUL(t61 - t62, costab28); |
328 | t85 = t63 + t64; |
329 | t108 = MUL(t63 - t64, costab12); |
330 | t86 = t65 + t66; |
331 | t109 = MUL(t65 - t66, costab20); |
332 | |
333 | t113 = t69 + t70; |
334 | t114 = t71 + t72; |
335 | |
336 | /* 0 */ |
337 | hi[15][slot] = SHIFT(t113 + t114); |
338 | /* 16 */ |
339 | lo[ 0][slot] = SHIFT(MUL(t113 - t114, costab16)); |
340 | |
341 | t115 = t73 + t74; |
342 | t116 = t75 + t76; |
343 | |
344 | t32 = t115 + t116; |
345 | |
346 | /* 1 */ |
347 | hi[14][slot] = SHIFT(t32); |
348 | |
349 | t118 = t78 + t79; |
350 | t119 = t80 + t81; |
351 | |
352 | t58 = t118 + t119; |
353 | |
354 | /* 2 */ |
355 | hi[13][slot] = SHIFT(t58); |
356 | |
357 | t121 = t83 + t84; |
358 | t122 = t85 + t86; |
359 | |
360 | t67 = t121 + t122; |
361 | |
362 | t49 = (t67 * 2) - t32; |
363 | |
364 | /* 3 */ |
365 | hi[12][slot] = SHIFT(t49); |
366 | |
367 | t125 = t89 + t90; |
368 | t126 = t91 + t92; |
369 | |
370 | t93 = t125 + t126; |
371 | |
372 | /* 4 */ |
373 | hi[11][slot] = SHIFT(t93); |
374 | |
375 | t128 = t94 + t95; |
376 | t129 = t96 + t97; |
377 | |
378 | t98 = t128 + t129; |
379 | |
380 | t68 = (t98 * 2) - t49; |
381 | |
382 | /* 5 */ |
383 | hi[10][slot] = SHIFT(t68); |
384 | |
385 | t132 = t100 + t101; |
386 | t133 = t102 + t103; |
387 | |
388 | t104 = t132 + t133; |
389 | |
390 | t82 = (t104 * 2) - t58; |
391 | |
392 | /* 6 */ |
393 | hi[ 9][slot] = SHIFT(t82); |
394 | |
395 | t136 = t106 + t107; |
396 | t137 = t108 + t109; |
397 | |
398 | t110 = t136 + t137; |
399 | |
400 | t87 = (t110 * 2) - t67; |
401 | |
402 | t77 = (t87 * 2) - t68; |
403 | |
404 | /* 7 */ |
405 | hi[ 8][slot] = SHIFT(t77); |
406 | |
407 | t141 = MUL(t69 - t70, costab8); |
408 | t142 = MUL(t71 - t72, costab24); |
409 | t143 = t141 + t142; |
410 | |
411 | /* 8 */ |
412 | hi[ 7][slot] = SHIFT(t143); |
413 | /* 24 */ |
414 | lo[ 8][slot] = |
415 | SHIFT((MUL(t141 - t142, costab16) * 2) - t143); |
416 | |
417 | t144 = MUL(t73 - t74, costab8); |
418 | t145 = MUL(t75 - t76, costab24); |
419 | t146 = t144 + t145; |
420 | |
421 | t88 = (t146 * 2) - t77; |
422 | |
423 | /* 9 */ |
424 | hi[ 6][slot] = SHIFT(t88); |
425 | |
426 | t148 = MUL(t78 - t79, costab8); |
427 | t149 = MUL(t80 - t81, costab24); |
428 | t150 = t148 + t149; |
429 | |
430 | t105 = (t150 * 2) - t82; |
431 | |
432 | /* 10 */ |
433 | hi[ 5][slot] = SHIFT(t105); |
434 | |
435 | t152 = MUL(t83 - t84, costab8); |
436 | t153 = MUL(t85 - t86, costab24); |
437 | t154 = t152 + t153; |
438 | |
439 | t111 = (t154 * 2) - t87; |
440 | |
441 | t99 = (t111 * 2) - t88; |
442 | |
443 | /* 11 */ |
444 | hi[ 4][slot] = SHIFT(t99); |
445 | |
446 | t157 = MUL(t89 - t90, costab8); |
447 | t158 = MUL(t91 - t92, costab24); |
448 | t159 = t157 + t158; |
449 | |
450 | t127 = (t159 * 2) - t93; |
451 | |
452 | /* 12 */ |
453 | hi[ 3][slot] = SHIFT(t127); |
454 | |
455 | t160 = (MUL(t125 - t126, costab16) * 2) - t127; |
456 | |
457 | /* 20 */ |
458 | lo[ 4][slot] = SHIFT(t160); |
459 | /* 28 */ |
460 | lo[12][slot] = |
461 | SHIFT((((MUL(t157 - t158, costab16) * 2) - t159) * 2) - t160); |
462 | |
463 | t161 = MUL(t94 - t95, costab8); |
464 | t162 = MUL(t96 - t97, costab24); |
465 | t163 = t161 + t162; |
466 | |
467 | t130 = (t163 * 2) - t98; |
468 | |
469 | t112 = (t130 * 2) - t99; |
470 | |
471 | /* 13 */ |
472 | hi[ 2][slot] = SHIFT(t112); |
473 | |
474 | t164 = (MUL(t128 - t129, costab16) * 2) - t130; |
475 | |
476 | t166 = MUL(t100 - t101, costab8); |
477 | t167 = MUL(t102 - t103, costab24); |
478 | t168 = t166 + t167; |
479 | |
480 | t134 = (t168 * 2) - t104; |
481 | |
482 | t120 = (t134 * 2) - t105; |
483 | |
484 | /* 14 */ |
485 | hi[ 1][slot] = SHIFT(t120); |
486 | |
487 | t135 = (MUL(t118 - t119, costab16) * 2) - t120; |
488 | |
489 | /* 18 */ |
490 | lo[ 2][slot] = SHIFT(t135); |
491 | |
492 | t169 = (MUL(t132 - t133, costab16) * 2) - t134; |
493 | |
494 | t151 = (t169 * 2) - t135; |
495 | |
496 | /* 22 */ |
497 | lo[ 6][slot] = SHIFT(t151); |
498 | |
499 | t170 = (((MUL(t148 - t149, costab16) * 2) - t150) * 2) - t151; |
500 | |
501 | /* 26 */ |
502 | lo[10][slot] = SHIFT(t170); |
503 | /* 30 */ |
504 | lo[14][slot] = |
505 | SHIFT((((((MUL(t166 - t167, costab16) * 2) - |
506 | t168) * 2) - t169) * 2) - t170); |
507 | |
508 | t171 = MUL(t106 - t107, costab8); |
509 | t172 = MUL(t108 - t109, costab24); |
510 | t173 = t171 + t172; |
511 | |
512 | t138 = (t173 * 2) - t110; |
513 | |
514 | t123 = (t138 * 2) - t111; |
515 | |
516 | t139 = (MUL(t121 - t122, costab16) * 2) - t123; |
517 | |
518 | t117 = (t123 * 2) - t112; |
519 | |
520 | /* 15 */ |
521 | hi[ 0][slot] = SHIFT(t117); |
522 | |
523 | t124 = (MUL(t115 - t116, costab16) * 2) - t117; |
524 | |
525 | /* 17 */ |
526 | lo[ 1][slot] = SHIFT(t124); |
527 | |
528 | t131 = (t139 * 2) - t124; |
529 | |
530 | /* 19 */ |
531 | lo[ 3][slot] = SHIFT(t131); |
532 | |
533 | t140 = (t164 * 2) - t131; |
534 | |
535 | /* 21 */ |
536 | lo[ 5][slot] = SHIFT(t140); |
537 | |
538 | t174 = (MUL(t136 - t137, costab16) * 2) - t138; |
539 | |
540 | t155 = (t174 * 2) - t139; |
541 | |
542 | t147 = (t155 * 2) - t140; |
543 | |
544 | /* 23 */ |
545 | lo[ 7][slot] = SHIFT(t147); |
546 | |
547 | t156 = (((MUL(t144 - t145, costab16) * 2) - t146) * 2) - t147; |
548 | |
549 | /* 25 */ |
550 | lo[ 9][slot] = SHIFT(t156); |
551 | |
552 | t175 = (((MUL(t152 - t153, costab16) * 2) - t154) * 2) - t155; |
553 | |
554 | t165 = (t175 * 2) - t156; |
555 | |
556 | /* 27 */ |
557 | lo[11][slot] = SHIFT(t165); |
558 | |
559 | t176 = (((((MUL(t161 - t162, costab16) * 2) - |
560 | t163) * 2) - t164) * 2) - t165; |
561 | |
562 | /* 29 */ |
563 | lo[13][slot] = SHIFT(t176); |
564 | /* 31 */ |
565 | lo[15][slot] = |
566 | SHIFT((((((((MUL(t171 - t172, costab16) * 2) - |
567 | t173) * 2) - t174) * 2) - t175) * 2) - t176); |
568 | |
569 | /* |
570 | * Totals: |
571 | * 80 multiplies |
572 | * 80 additions |
573 | * 119 subtractions |
574 | * 49 shifts (not counting SSO) |
575 | */ |
576 | } |
577 | |
578 | # undef MUL |
579 | # undef SHIFT |
580 | |
581 | /* third SSO shift and/or D[] optimization preshift */ |
582 | |
583 | # if defined(OPT_SSO) |
584 | # if MAD_F_FRACBITS != 28 |
585 | # error "MAD_F_FRACBITS must be 28 to use OPT_SSO" |
586 | # endif |
587 | # define ML0(hi, lo, x, y) ((lo) = (x) * (y)) |
588 | # define MLA(hi, lo, x, y) ((lo) += (x) * (y)) |
589 | # define MLN(hi, lo) ((lo) = -(lo)) |
590 | # define MLZ(hi, lo) ((void) (hi), (mad_fixed_t) (lo)) |
591 | # define SHIFT(x) ((x) >> 2) |
592 | # define PRESHIFT(x) ((MAD_F(x) + (1L << 13)) >> 14) |
593 | # else |
594 | # define ML0(hi, lo, x, y) MAD_F_ML0((hi), (lo), (x), (y)) |
595 | # define MLA(hi, lo, x, y) MAD_F_MLA((hi), (lo), (x), (y)) |
596 | # define MLN(hi, lo) MAD_F_MLN((hi), (lo)) |
597 | # define MLZ(hi, lo) MAD_F_MLZ((hi), (lo)) |
598 | # define SHIFT(x) (x) |
599 | # if defined(MAD_F_SCALEBITS) |
600 | # undef MAD_F_SCALEBITS |
601 | # define MAD_F_SCALEBITS (MAD_F_FRACBITS - 12) |
602 | # define PRESHIFT(x) (MAD_F(x) >> 12) |
603 | # else |
604 | # define PRESHIFT(x) MAD_F(x) |
605 | # endif |
606 | # endif |
607 | |
608 | static |
609 | mad_fixed_t const D[17][32] = { |
610 | # include "D.dat" |
611 | }; |
612 | |
613 | # if defined(ASO_SYNTH) |
614 | void synth_full(struct mad_synth *, struct mad_frame const *, |
615 | unsigned int, unsigned int); |
616 | # else |
617 | /* |
618 | * NAME: synth->full() |
619 | * DESCRIPTION: perform full frequency PCM synthesis |
620 | */ |
621 | static |
622 | void synth_full(struct mad_synth *synth, struct mad_frame const *frame, |
623 | unsigned int nch, unsigned int ns) |
624 | { |
625 | unsigned int phase, ch, s, sb, pe, po; |
626 | mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8]; |
627 | mad_fixed_t const(*sbsample)[36][32]; |
628 | register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8]; |
629 | register mad_fixed_t const(*Dptr)[32], *ptr; |
630 | register mad_fixed64hi_t hi; |
631 | register mad_fixed64lo_t lo; |
632 | |
633 | for (ch = 0; ch < nch; ++ch) { |
634 | sbsample = &frame->sbsample[ch]; |
635 | filter = &synth->filter[ch]; |
636 | phase = synth->phase; |
637 | pcm1 = synth->pcm.samples[ch]; |
638 | |
639 | for (s = 0; s < ns; ++s) { |
640 | dct32((*sbsample)[s], phase >> 1, |
641 | (*filter)[0][phase & 1], (*filter)[1][phase & 1]); |
642 | |
643 | pe = phase & ~1; |
644 | po = ((phase - 1) & 0xf) | 1; |
645 | |
646 | /* calculate 32 samples */ |
647 | |
648 | fe = &(*filter)[0][ phase & 1][0]; |
649 | fx = &(*filter)[0][~phase & 1][0]; |
650 | fo = &(*filter)[1][~phase & 1][0]; |
651 | |
652 | Dptr = &D[0]; |
653 | |
654 | ptr = *Dptr + po; |
655 | ML0(hi, lo, (*fx)[0], ptr[ 0]); |
656 | MLA(hi, lo, (*fx)[1], ptr[14]); |
657 | MLA(hi, lo, (*fx)[2], ptr[12]); |
658 | MLA(hi, lo, (*fx)[3], ptr[10]); |
659 | MLA(hi, lo, (*fx)[4], ptr[ 8]); |
660 | MLA(hi, lo, (*fx)[5], ptr[ 6]); |
661 | MLA(hi, lo, (*fx)[6], ptr[ 4]); |
662 | MLA(hi, lo, (*fx)[7], ptr[ 2]); |
663 | MLN(hi, lo); |
664 | |
665 | ptr = *Dptr + pe; |
666 | MLA(hi, lo, (*fe)[0], ptr[ 0]); |
667 | MLA(hi, lo, (*fe)[1], ptr[14]); |
668 | MLA(hi, lo, (*fe)[2], ptr[12]); |
669 | MLA(hi, lo, (*fe)[3], ptr[10]); |
670 | MLA(hi, lo, (*fe)[4], ptr[ 8]); |
671 | MLA(hi, lo, (*fe)[5], ptr[ 6]); |
672 | MLA(hi, lo, (*fe)[6], ptr[ 4]); |
673 | MLA(hi, lo, (*fe)[7], ptr[ 2]); |
674 | |
675 | *pcm1++ = SHIFT(MLZ(hi, lo)); |
676 | |
677 | pcm2 = pcm1 + 30; |
678 | |
679 | for (sb = 1; sb < 16; ++sb) { |
680 | ++fe; |
681 | ++Dptr; |
682 | |
683 | /* D[32 - sb][i] == -D[sb][31 - i] */ |
684 | |
685 | ptr = *Dptr + po; |
686 | ML0(hi, lo, (*fo)[0], ptr[ 0]); |
687 | MLA(hi, lo, (*fo)[1], ptr[14]); |
688 | MLA(hi, lo, (*fo)[2], ptr[12]); |
689 | MLA(hi, lo, (*fo)[3], ptr[10]); |
690 | MLA(hi, lo, (*fo)[4], ptr[ 8]); |
691 | MLA(hi, lo, (*fo)[5], ptr[ 6]); |
692 | MLA(hi, lo, (*fo)[6], ptr[ 4]); |
693 | MLA(hi, lo, (*fo)[7], ptr[ 2]); |
694 | MLN(hi, lo); |
695 | |
696 | ptr = *Dptr + pe; |
697 | MLA(hi, lo, (*fe)[7], ptr[ 2]); |
698 | MLA(hi, lo, (*fe)[6], ptr[ 4]); |
699 | MLA(hi, lo, (*fe)[5], ptr[ 6]); |
700 | MLA(hi, lo, (*fe)[4], ptr[ 8]); |
701 | MLA(hi, lo, (*fe)[3], ptr[10]); |
702 | MLA(hi, lo, (*fe)[2], ptr[12]); |
703 | MLA(hi, lo, (*fe)[1], ptr[14]); |
704 | MLA(hi, lo, (*fe)[0], ptr[ 0]); |
705 | |
706 | *pcm1++ = SHIFT(MLZ(hi, lo)); |
707 | |
708 | ptr = *Dptr - pe; |
709 | ML0(hi, lo, (*fe)[0], ptr[31 - 16]); |
710 | MLA(hi, lo, (*fe)[1], ptr[31 - 14]); |
711 | MLA(hi, lo, (*fe)[2], ptr[31 - 12]); |
712 | MLA(hi, lo, (*fe)[3], ptr[31 - 10]); |
713 | MLA(hi, lo, (*fe)[4], ptr[31 - 8]); |
714 | MLA(hi, lo, (*fe)[5], ptr[31 - 6]); |
715 | MLA(hi, lo, (*fe)[6], ptr[31 - 4]); |
716 | MLA(hi, lo, (*fe)[7], ptr[31 - 2]); |
717 | |
718 | ptr = *Dptr - po; |
719 | MLA(hi, lo, (*fo)[7], ptr[31 - 2]); |
720 | MLA(hi, lo, (*fo)[6], ptr[31 - 4]); |
721 | MLA(hi, lo, (*fo)[5], ptr[31 - 6]); |
722 | MLA(hi, lo, (*fo)[4], ptr[31 - 8]); |
723 | MLA(hi, lo, (*fo)[3], ptr[31 - 10]); |
724 | MLA(hi, lo, (*fo)[2], ptr[31 - 12]); |
725 | MLA(hi, lo, (*fo)[1], ptr[31 - 14]); |
726 | MLA(hi, lo, (*fo)[0], ptr[31 - 16]); |
727 | |
728 | *pcm2-- = SHIFT(MLZ(hi, lo)); |
729 | |
730 | ++fo; |
731 | } |
732 | |
733 | ++Dptr; |
734 | |
735 | ptr = *Dptr + po; |
736 | ML0(hi, lo, (*fo)[0], ptr[ 0]); |
737 | MLA(hi, lo, (*fo)[1], ptr[14]); |
738 | MLA(hi, lo, (*fo)[2], ptr[12]); |
739 | MLA(hi, lo, (*fo)[3], ptr[10]); |
740 | MLA(hi, lo, (*fo)[4], ptr[ 8]); |
741 | MLA(hi, lo, (*fo)[5], ptr[ 6]); |
742 | MLA(hi, lo, (*fo)[6], ptr[ 4]); |
743 | MLA(hi, lo, (*fo)[7], ptr[ 2]); |
744 | |
745 | *pcm1 = SHIFT(-MLZ(hi, lo)); |
746 | pcm1 += 16; |
747 | |
748 | phase = (phase + 1) % 16; |
749 | } |
750 | } |
751 | } |
752 | # endif |
753 | |
754 | /* |
755 | * NAME: synth->half() |
756 | * DESCRIPTION: perform half frequency PCM synthesis |
757 | */ |
758 | static |
759 | void synth_half(struct mad_synth *synth, struct mad_frame const *frame, |
760 | unsigned int nch, unsigned int ns) |
761 | { |
762 | unsigned int phase, ch, s, sb, pe, po; |
763 | mad_fixed_t *pcm1, *pcm2, (*filter)[2][2][16][8]; |
764 | mad_fixed_t const(*sbsample)[36][32]; |
765 | register mad_fixed_t (*fe)[8], (*fx)[8], (*fo)[8]; |
766 | register mad_fixed_t const(*Dptr)[32], *ptr; |
767 | register mad_fixed64hi_t hi; |
768 | register mad_fixed64lo_t lo; |
769 | |
770 | for (ch = 0; ch < nch; ++ch) { |
771 | sbsample = &frame->sbsample[ch]; |
772 | filter = &synth->filter[ch]; |
773 | phase = synth->phase; |
774 | pcm1 = synth->pcm.samples[ch]; |
775 | |
776 | for (s = 0; s < ns; ++s) { |
777 | dct32((*sbsample)[s], phase >> 1, |
778 | (*filter)[0][phase & 1], (*filter)[1][phase & 1]); |
779 | |
780 | pe = phase & ~1; |
781 | po = ((phase - 1) & 0xf) | 1; |
782 | |
783 | /* calculate 16 samples */ |
784 | |
785 | fe = &(*filter)[0][ phase & 1][0]; |
786 | fx = &(*filter)[0][~phase & 1][0]; |
787 | fo = &(*filter)[1][~phase & 1][0]; |
788 | |
789 | Dptr = &D[0]; |
790 | |
791 | ptr = *Dptr + po; |
792 | ML0(hi, lo, (*fx)[0], ptr[ 0]); |
793 | MLA(hi, lo, (*fx)[1], ptr[14]); |
794 | MLA(hi, lo, (*fx)[2], ptr[12]); |
795 | MLA(hi, lo, (*fx)[3], ptr[10]); |
796 | MLA(hi, lo, (*fx)[4], ptr[ 8]); |
797 | MLA(hi, lo, (*fx)[5], ptr[ 6]); |
798 | MLA(hi, lo, (*fx)[6], ptr[ 4]); |
799 | MLA(hi, lo, (*fx)[7], ptr[ 2]); |
800 | MLN(hi, lo); |
801 | |
802 | ptr = *Dptr + pe; |
803 | MLA(hi, lo, (*fe)[0], ptr[ 0]); |
804 | MLA(hi, lo, (*fe)[1], ptr[14]); |
805 | MLA(hi, lo, (*fe)[2], ptr[12]); |
806 | MLA(hi, lo, (*fe)[3], ptr[10]); |
807 | MLA(hi, lo, (*fe)[4], ptr[ 8]); |
808 | MLA(hi, lo, (*fe)[5], ptr[ 6]); |
809 | MLA(hi, lo, (*fe)[6], ptr[ 4]); |
810 | MLA(hi, lo, (*fe)[7], ptr[ 2]); |
811 | |
812 | *pcm1++ = SHIFT(MLZ(hi, lo)); |
813 | |
814 | pcm2 = pcm1 + 14; |
815 | |
816 | for (sb = 1; sb < 16; ++sb) { |
817 | ++fe; |
818 | ++Dptr; |
819 | |
820 | /* D[32 - sb][i] == -D[sb][31 - i] */ |
821 | |
822 | if (!(sb & 1)) { |
823 | ptr = *Dptr + po; |
824 | ML0(hi, lo, (*fo)[0], ptr[ 0]); |
825 | MLA(hi, lo, (*fo)[1], ptr[14]); |
826 | MLA(hi, lo, (*fo)[2], ptr[12]); |
827 | MLA(hi, lo, (*fo)[3], ptr[10]); |
828 | MLA(hi, lo, (*fo)[4], ptr[ 8]); |
829 | MLA(hi, lo, (*fo)[5], ptr[ 6]); |
830 | MLA(hi, lo, (*fo)[6], ptr[ 4]); |
831 | MLA(hi, lo, (*fo)[7], ptr[ 2]); |
832 | MLN(hi, lo); |
833 | |
834 | ptr = *Dptr + pe; |
835 | MLA(hi, lo, (*fe)[7], ptr[ 2]); |
836 | MLA(hi, lo, (*fe)[6], ptr[ 4]); |
837 | MLA(hi, lo, (*fe)[5], ptr[ 6]); |
838 | MLA(hi, lo, (*fe)[4], ptr[ 8]); |
839 | MLA(hi, lo, (*fe)[3], ptr[10]); |
840 | MLA(hi, lo, (*fe)[2], ptr[12]); |
841 | MLA(hi, lo, (*fe)[1], ptr[14]); |
842 | MLA(hi, lo, (*fe)[0], ptr[ 0]); |
843 | |
844 | *pcm1++ = SHIFT(MLZ(hi, lo)); |
845 | |
846 | ptr = *Dptr - po; |
847 | ML0(hi, lo, (*fo)[7], ptr[31 - 2]); |
848 | MLA(hi, lo, (*fo)[6], ptr[31 - 4]); |
849 | MLA(hi, lo, (*fo)[5], ptr[31 - 6]); |
850 | MLA(hi, lo, (*fo)[4], ptr[31 - 8]); |
851 | MLA(hi, lo, (*fo)[3], ptr[31 - 10]); |
852 | MLA(hi, lo, (*fo)[2], ptr[31 - 12]); |
853 | MLA(hi, lo, (*fo)[1], ptr[31 - 14]); |
854 | MLA(hi, lo, (*fo)[0], ptr[31 - 16]); |
855 | |
856 | ptr = *Dptr - pe; |
857 | MLA(hi, lo, (*fe)[0], ptr[31 - 16]); |
858 | MLA(hi, lo, (*fe)[1], ptr[31 - 14]); |
859 | MLA(hi, lo, (*fe)[2], ptr[31 - 12]); |
860 | MLA(hi, lo, (*fe)[3], ptr[31 - 10]); |
861 | MLA(hi, lo, (*fe)[4], ptr[31 - 8]); |
862 | MLA(hi, lo, (*fe)[5], ptr[31 - 6]); |
863 | MLA(hi, lo, (*fe)[6], ptr[31 - 4]); |
864 | MLA(hi, lo, (*fe)[7], ptr[31 - 2]); |
865 | |
866 | *pcm2-- = SHIFT(MLZ(hi, lo)); |
867 | } |
868 | |
869 | ++fo; |
870 | } |
871 | |
872 | ++Dptr; |
873 | |
874 | ptr = *Dptr + po; |
875 | ML0(hi, lo, (*fo)[0], ptr[ 0]); |
876 | MLA(hi, lo, (*fo)[1], ptr[14]); |
877 | MLA(hi, lo, (*fo)[2], ptr[12]); |
878 | MLA(hi, lo, (*fo)[3], ptr[10]); |
879 | MLA(hi, lo, (*fo)[4], ptr[ 8]); |
880 | MLA(hi, lo, (*fo)[5], ptr[ 6]); |
881 | MLA(hi, lo, (*fo)[6], ptr[ 4]); |
882 | MLA(hi, lo, (*fo)[7], ptr[ 2]); |
883 | |
884 | *pcm1 = SHIFT(-MLZ(hi, lo)); |
885 | pcm1 += 8; |
886 | |
887 | phase = (phase + 1) % 16; |
888 | } |
889 | } |
890 | } |
891 | |
892 | /* |
893 | * NAME: synth->frame() |
894 | * DESCRIPTION: perform PCM synthesis of frame subband samples |
895 | */ |
896 | void mad_synth_frame(struct mad_synth *synth, struct mad_frame const *frame) |
897 | { |
898 | unsigned int nch, ns; |
899 | void (*synth_frame)(struct mad_synth *, struct mad_frame const *, |
900 | unsigned int, unsigned int); |
901 | |
902 | nch = MAD_NCHANNELS(&frame->header); |
903 | ns = MAD_NSBSAMPLES(&frame->header); |
904 | |
905 | synth->pcm.samplerate = frame->header.samplerate; |
906 | synth->pcm.channels = nch; |
907 | synth->pcm.length = 32 * ns; |
908 | |
909 | synth_frame = synth_full; |
910 | |
911 | if (frame->options & MAD_OPTION_HALFSAMPLERATE) { |
912 | synth->pcm.samplerate /= 2; |
913 | synth->pcm.length /= 2; |
914 | |
915 | synth_frame = synth_half; |
916 | } |
917 | |
918 | synth_frame(synth, frame, nch, ns); |
919 | |
920 | synth->phase = (synth->phase + ns) % 16; |
921 | } |
922 |