?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Rework of libfaad in several areas. Allow removal of malloc with a new define FAAD_ST...
[kugel-rb.git] / apps / codecs / libfaad / sbr_qmf.c
blob5f8203e5b143d5bb0a8db0c0e6aee1058172acdd
1 /*
2 ** FAAD2 - Freeware Advanced Audio (AAC) Decoder including SBR decoding
3 ** Copyright (C) 2003-2004 M. Bakker, Ahead Software AG, http://www.nero.com
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 ** Any non-GPL usage of this software or parts of this software is strictly
20 ** forbidden.
21 **
22 ** Commercial non-GPL licensing of this software is possible.
23 ** For more info contact Ahead Software through Mpeg4AAClicense@nero.com.
24 **
25 ** $Id$
26 **/
27
28 #include "common.h"
29 #include "structs.h"
30
31 #ifdef SBR_DEC
32
33
34 #include <stdlib.h>
35 #include <string.h>
36 #include "sbr_dct.h"
37 #include "sbr_qmf.h"
38 #include "sbr_qmf_c.h"
39 #include "sbr_syntax.h"
40
41 #ifdef FIXED_POINT
42 #define FAAD_SYNTHESIS_SCALE(X) ((X)>>1)
43 #define FAAD_ANALYSIS_SCALE1(X) ((X)>>4)
44 #define FAAD_ANALYSIS_SCALE2(X) ((X))
45 #define FAAD_ANALYSIS_SCALE3(X) ((X))
46 #else
47 #define FAAD_SYNTHESIS_SCALE(X) ((X)/64.0f)
48 #define FAAD_ANALYSIS_SCALE1(X) ((X))
49 #define FAAD_ANALYSIS_SCALE2(X) (2.0f*(X))
50 #define FAAD_ANALYSIS_SCALE3(X) ((X)/32.0f)
51 #endif
52
53
54 void sbr_qmf_analysis_32(sbr_info *sbr, qmfa_info *qmfa, const real_t *input,
55 qmf_t X[MAX_NTSR][64], uint8_t offset, uint8_t kx)
56 {
57 real_t u[64] MEM_ALIGN_ATTR;
58 #ifndef SBR_LOW_POWER
59 real_t real[32] MEM_ALIGN_ATTR;
60 real_t imag[32] MEM_ALIGN_ATTR;
61 #else
62 real_t y[32] MEM_ALIGN_ATTR;
63 #endif
64 qmf_t *pX;
65 uint32_t in = 0;
66 uint32_t l, idx0, idx1;
67
68 /* qmf subsample l */
69 for (l = 0; l < sbr->numTimeSlotsRate; l++)
70 {
71 int32_t n;
72
73 /* shift input buffer x */
74 /* input buffer is not shifted anymore, x is implemented as double ringbuffer */
75 //memmove(qmfa->x + 32, qmfa->x, (320-32)*sizeof(real_t));
76
77 /* add new samples to input buffer x */
78 idx0 = qmfa->x_index + 31; idx1 = idx0 + 320;
79 for (n = 0; n < 32; n+=4)
80 {
81 qmfa->x[idx0--] = qmfa->x[idx1--] = (input[in++]);
82 qmfa->x[idx0--] = qmfa->x[idx1--] = (input[in++]);
83 qmfa->x[idx0--] = qmfa->x[idx1--] = (input[in++]);
84 qmfa->x[idx0--] = qmfa->x[idx1--] = (input[in++]);
85 }
86
87 /* window and summation to create array u */
88 for (n = 0; n < 32; n++)
89 {
90 idx0 = qmfa->x_index + n; idx1 = n * 20;
91 u[n] = FAAD_ANALYSIS_SCALE1(
92 MUL_F(qmfa->x[idx0 ], qmf_c[idx1 ]) +
93 MUL_F(qmfa->x[idx0 + 64], qmf_c[idx1 + 2]) +
94 MUL_F(qmfa->x[idx0 + 128], qmf_c[idx1 + 4]) +
95 MUL_F(qmfa->x[idx0 + 192], qmf_c[idx1 + 6]) +
96 MUL_F(qmfa->x[idx0 + 256], qmf_c[idx1 + 8]));
97 }
98 for (n = 32; n < 64; n++)
99 {
100 idx0 = qmfa->x_index + n; idx1 = n * 20 - 639;
101 u[n] = FAAD_ANALYSIS_SCALE1(
102 MUL_F(qmfa->x[idx0 ], qmf_c[idx1 ]) +
103 MUL_F(qmfa->x[idx0 + 64], qmf_c[idx1 + 2]) +
104 MUL_F(qmfa->x[idx0 + 128], qmf_c[idx1 + 4]) +
105 MUL_F(qmfa->x[idx0 + 192], qmf_c[idx1 + 6]) +
106 MUL_F(qmfa->x[idx0 + 256], qmf_c[idx1 + 8]));
107 }
108
109 /* update ringbuffer index */
110 qmfa->x_index -= 32;
111 if (qmfa->x_index < 0)
112 qmfa->x_index = (320-32);
113
114 /* calculate 32 subband samples by introducing X */
115 #ifdef SBR_LOW_POWER
116 y[0] = u[48];
117 for (n = 1; n < 16; n++)
118 y[n] = u[n+48] + u[48-n];
119 for (n = 16; n < 32; n++)
120 y[n] = -u[n-16] + u[48-n];
121
122 DCT3_32_unscaled(u, y);
123
124 for (n = 0; n < 32; n++)
125 {
126 if (n < kx)
127 {
128 QMF_RE(X[l + offset][n]) = FAAD_ANALYSIS_SCALE2(u[n]);
129 } else {
130 QMF_RE(X[l + offset][n]) = 0;
131 }
132 }
133 #else /* #ifdef SBR_LOW_POWER */
134
135 // Reordering of data moved from DCT_IV to here
136 idx0 = 30; idx1 = 63;
137 imag[31] = u[ 1]; real[ 0] = u[ 0];
138 for (n = 1; n < 31; n+=3)
139 {
140 imag[idx0--] = u[n+1]; real[n ] = -u[idx1--];
141 imag[idx0--] = u[n+2]; real[n+1] = -u[idx1--];
142 imag[idx0--] = u[n+3]; real[n+2] = -u[idx1--];
143 }
144 imag[ 0] = u[32]; real[31] = -u[33];
145
146 // dct4_kernel is DCT_IV without reordering which is done before and after FFT
147 dct4_kernel(real, imag);
148
149 // Reordering of data moved from DCT_IV to here
150 /* Step 1: Calculate all non-zero pairs */
151 pX = X[l + offset];
152 for (n = 0; n < kx/2; n++) {
153 idx0 = 2*n; idx1 = idx0 + 1;
154 QMF_RE(pX[idx0]) = FAAD_ANALYSIS_SCALE2( real[n ]);
155 QMF_IM(pX[idx0]) = FAAD_ANALYSIS_SCALE2( imag[n ]);
156 QMF_RE(pX[idx1]) = FAAD_ANALYSIS_SCALE2(-imag[31-n]);
157 QMF_IM(pX[idx1]) = FAAD_ANALYSIS_SCALE2(-real[31-n]);
158 }
159 /* Step 2: Calculate a single pair with half zero'ed */
160 if (kx&1) {
161 idx0 = 2*n; idx1 = idx0 + 1;
162 QMF_RE(pX[idx0]) = FAAD_ANALYSIS_SCALE2( real[n]);
163 QMF_IM(pX[idx0]) = FAAD_ANALYSIS_SCALE2( imag[n]);
164 QMF_RE(pX[idx1]) = QMF_IM(pX[idx1]) = 0;
165 n++;
166 }
167 /* Step 3: All other are zero'ed */
168 for (; n < 16; n++) {
169 idx0 = 2*n; idx1 = idx0 + 1;
170 QMF_RE(pX[idx0]) = QMF_IM(pX[idx0]) = 0;
171 QMF_RE(pX[idx1]) = QMF_IM(pX[idx1]) = 0;
172 }
173 #endif /* #ifdef SBR_LOW_POWER */
174 }
175 }
176
177 #ifdef SBR_LOW_POWER
178
179 void sbr_qmf_synthesis_32(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSR][64],
180 real_t *output)
181 {
182 real_t x[16] MEM_ALIGN_ATTR;
183 real_t y[16] MEM_ALIGN_ATTR;
184 int16_t n, k, out = 0;
185 uint8_t l;
186
187 /* qmf subsample l */
188 for (l = 0; l < sbr->numTimeSlotsRate; l++)
189 {
190 /* shift buffers */
191 /* we are not shifting v, it is a double ringbuffer */
192 //memmove(qmfs->v + 64, qmfs->v, (640-64)*sizeof(real_t));
193
194 /* calculate 64 samples */
195 for (k = 0; k < 16; k++)
196 {
197 y[k] = FAAD_ANALYSIS_SCALE3((QMF_RE(X[l][k]) - QMF_RE(X[l][31-k])));
198 x[k] = FAAD_ANALYSIS_SCALE3((QMF_RE(X[l][k]) + QMF_RE(X[l][31-k])));
199 }
200
201 /* even n samples */
202 DCT2_16_unscaled(x, x);
203 /* odd n samples */
204 DCT4_16(y, y);
205
206 for (n = 8; n < 24; n++)
207 {
208 qmfs->v[qmfs->v_index + n*2 ] = qmfs->v[qmfs->v_index + 640 + n*2 ] = x[n-8];
209 qmfs->v[qmfs->v_index + n*2+1] = qmfs->v[qmfs->v_index + 640 + n*2+1] = y[n-8];
210 }
211 for (n = 0; n < 16; n++)
212 {
213 qmfs->v[qmfs->v_index + n] = qmfs->v[qmfs->v_index + 640 + n] = qmfs->v[qmfs->v_index + 32-n];
214 }
215 qmfs->v[qmfs->v_index + 48] = qmfs->v[qmfs->v_index + 640 + 48] = 0;
216 for (n = 1; n < 16; n++)
217 {
218 qmfs->v[qmfs->v_index + 48+n] = qmfs->v[qmfs->v_index + 640 + 48+n] = -qmfs->v[qmfs->v_index + 48-n];
219 }
220
221 /* calculate 32 output samples and window */
222 for (k = 0; k < 32; k++)
223 {
224 output[out++] = MUL_F(qmfs->v[qmfs->v_index + k], qmf_c[ 2*k*10]) +
225 MUL_F(qmfs->v[qmfs->v_index + 96 + k], qmf_c[1 + 2*k*10]) +
226 MUL_F(qmfs->v[qmfs->v_index + 128 + k], qmf_c[2 + 2*k*10]) +
227 MUL_F(qmfs->v[qmfs->v_index + 224 + k], qmf_c[3 + 2*k*10]) +
228 MUL_F(qmfs->v[qmfs->v_index + 256 + k], qmf_c[4 + 2*k*10]) +
229 MUL_F(qmfs->v[qmfs->v_index + 352 + k], qmf_c[5 + 2*k*10]) +
230 MUL_F(qmfs->v[qmfs->v_index + 384 + k], qmf_c[6 + 2*k*10]) +
231 MUL_F(qmfs->v[qmfs->v_index + 480 + k], qmf_c[7 + 2*k*10]) +
232 MUL_F(qmfs->v[qmfs->v_index + 512 + k], qmf_c[8 + 2*k*10]) +
233 MUL_F(qmfs->v[qmfs->v_index + 608 + k], qmf_c[9 + 2*k*10]);
234 }
235
236 /* update the ringbuffer index */
237 qmfs->v_index -= 64;
238 if (qmfs->v_index < 0)
239 qmfs->v_index = (640-64);
240 }
241 }
242
243 void sbr_qmf_synthesis_64(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSR][64],
244 real_t *output)
245 {
246 real_t x[64] MEM_ALIGN_ATTR;
247 real_t y[64] MEM_ALIGN_ATTR;
248 int16_t n, k, out = 0;
249 uint8_t l;
250
251
252 /* qmf subsample l */
253 for (l = 0; l < sbr->numTimeSlotsRate; l++)
254 {
255 /* shift buffers */
256 /* we are not shifting v, it is a double ringbuffer */
257 //memmove(qmfs->v + 128, qmfs->v, (1280-128)*sizeof(real_t));
258
259 /* calculate 128 samples */
260 for (k = 0; k < 32; k++)
261 {
262 y[k] = FAAD_ANALYSIS_SCALE3((QMF_RE(X[l][k]) - QMF_RE(X[l][63-k])));
263 x[k] = FAAD_ANALYSIS_SCALE3((QMF_RE(X[l][k]) + QMF_RE(X[l][63-k])));
264 }
265
266 /* even n samples */
267 DCT2_32_unscaled(x, x);
268 /* odd n samples */
269 DCT4_32(y, y);
270
271 for (n = 16; n < 48; n++)
272 {
273 qmfs->v[qmfs->v_index + n*2] = qmfs->v[qmfs->v_index + 1280 + n*2 ] = x[n-16];
274 qmfs->v[qmfs->v_index + n*2+1] = qmfs->v[qmfs->v_index + 1280 + n*2+1] = y[n-16];
275 }
276 for (n = 0; n < 32; n++)
277 {
278 qmfs->v[qmfs->v_index + n] = qmfs->v[qmfs->v_index + 1280 + n] = qmfs->v[qmfs->v_index + 64-n];
279 }
280 qmfs->v[qmfs->v_index + 96] = qmfs->v[qmfs->v_index + 1280 + 96] = 0;
281 for (n = 1; n < 32; n++)
282 {
283 qmfs->v[qmfs->v_index + 96+n] = qmfs->v[qmfs->v_index + 1280 + 96+n] = -qmfs->v[qmfs->v_index + 96-n];
284 }
285
286 /* calculate 64 output samples and window */
287 for (k = 0; k < 64; k++)
288 {
289 output[out++] = MUL_F(qmfs->v[qmfs->v_index + k], qmf_c[ k*10]) +
290 MUL_F(qmfs->v[qmfs->v_index + 192 + k], qmf_c[1 + k*10]) +
291 MUL_F(qmfs->v[qmfs->v_index + 256 + k], qmf_c[2 + k*10]) +
292 MUL_F(qmfs->v[qmfs->v_index + 256 + 192 + k], qmf_c[3 + k*10]) +
293 MUL_F(qmfs->v[qmfs->v_index + 512 + k], qmf_c[4 + k*10]) +
294 MUL_F(qmfs->v[qmfs->v_index + 512 + 192 + k], qmf_c[5 + k*10]) +
295 MUL_F(qmfs->v[qmfs->v_index + 768 + k], qmf_c[6 + k*10]) +
296 MUL_F(qmfs->v[qmfs->v_index + 768 + 192 + k], qmf_c[7 + k*10]) +
297 MUL_F(qmfs->v[qmfs->v_index + 1024 + k], qmf_c[8 + k*10]) +
298 MUL_F(qmfs->v[qmfs->v_index + 1024 + 192 + k], qmf_c[9 + k*10]);
299 }
300
301 /* update the ringbuffer index */
302 qmfs->v_index -= 128;
303 if (qmfs->v_index < 0)
304 qmfs->v_index = (1280-128);
305 }
306 }
307 #else /* #ifdef SBR_LOW_POWER */
308
309 static const complex_t qmf32_pre_twiddle[] =
310 {
311 { FRAC_CONST(0.999924701839145), FRAC_CONST(-0.012271538285720) },
312 { FRAC_CONST(0.999322384588350), FRAC_CONST(-0.036807222941359) },
313 { FRAC_CONST(0.998118112900149), FRAC_CONST(-0.061320736302209) },
314 { FRAC_CONST(0.996312612182778), FRAC_CONST(-0.085797312344440) },
315 { FRAC_CONST(0.993906970002356), FRAC_CONST(-0.110222207293883) },
316 { FRAC_CONST(0.990902635427780), FRAC_CONST(-0.134580708507126) },
317 { FRAC_CONST(0.987301418157858), FRAC_CONST(-0.158858143333861) },
318 { FRAC_CONST(0.983105487431216), FRAC_CONST(-0.183039887955141) },
319 { FRAC_CONST(0.978317370719628), FRAC_CONST(-0.207111376192219) },
320 { FRAC_CONST(0.972939952205560), FRAC_CONST(-0.231058108280671) },
321 { FRAC_CONST(0.966976471044852), FRAC_CONST(-0.254865659604515) },
322 { FRAC_CONST(0.960430519415566), FRAC_CONST(-0.278519689385053) },
323 { FRAC_CONST(0.953306040354194), FRAC_CONST(-0.302005949319228) },
324 { FRAC_CONST(0.945607325380521), FRAC_CONST(-0.325310292162263) },
325 { FRAC_CONST(0.937339011912575), FRAC_CONST(-0.348418680249435) },
326 { FRAC_CONST(0.928506080473216), FRAC_CONST(-0.371317193951838) },
327 { FRAC_CONST(0.919113851690058), FRAC_CONST(-0.393992040061048) },
328 { FRAC_CONST(0.909167983090522), FRAC_CONST(-0.416429560097637) },
329 { FRAC_CONST(0.898674465693954), FRAC_CONST(-0.438616238538528) },
330 { FRAC_CONST(0.887639620402854), FRAC_CONST(-0.460538710958240) },
331 { FRAC_CONST(0.876070094195407), FRAC_CONST(-0.482183772079123) },
332 { FRAC_CONST(0.863972856121587), FRAC_CONST(-0.503538383725718) },
333 { FRAC_CONST(0.851355193105265), FRAC_CONST(-0.524589682678469) },
334 { FRAC_CONST(0.838224705554838), FRAC_CONST(-0.545324988422046) },
335 { FRAC_CONST(0.824589302785025), FRAC_CONST(-0.565731810783613) },
336 { FRAC_CONST(0.810457198252595), FRAC_CONST(-0.585797857456439) },
337 { FRAC_CONST(0.795836904608884), FRAC_CONST(-0.605511041404326) },
338 { FRAC_CONST(0.780737228572094), FRAC_CONST(-0.624859488142386) },
339 { FRAC_CONST(0.765167265622459), FRAC_CONST(-0.643831542889791) },
340 { FRAC_CONST(0.749136394523459), FRAC_CONST(-0.662415777590172) },
341 { FRAC_CONST(0.732654271672413), FRAC_CONST(-0.680600997795453) },
342 { FRAC_CONST(0.715730825283819), FRAC_CONST(-0.698376249408973) }
343 };
344
345 #define FAAD_CMPLX_PRETWIDDLE_SUB(k) \
346 (MUL_F(QMF_RE(X[l][k]), RE(qmf32_pre_twiddle[k])) - \
347 MUL_F(QMF_IM(X[l][k]), IM(qmf32_pre_twiddle[k])))
348
349 #define FAAD_CMPLX_PRETWIDDLE_ADD(k) \
350 (MUL_F(QMF_IM(X[l][k]), RE(qmf32_pre_twiddle[k])) + \
351 MUL_F(QMF_RE(X[l][k]), IM(qmf32_pre_twiddle[k])))
352
353 void sbr_qmf_synthesis_32(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSR][64],
354 real_t *output)
355 {
356 real_t x1[32] MEM_ALIGN_ATTR;
357 real_t x2[32] MEM_ALIGN_ATTR;
358 int32_t n, k, idx0, idx1, out = 0;
359 uint32_t l;
360
361 /* qmf subsample l */
362 for (l = 0; l < sbr->numTimeSlotsRate; l++)
363 {
364 /* shift buffer v */
365 /* buffer is not shifted, we are using a ringbuffer */
366 //memmove(qmfs->v + 64, qmfs->v, (640-64)*sizeof(real_t));
367
368 /* calculate 64 samples */
369 /* complex pre-twiddle */
370 for (k = 0; k < 32;)
371 {
372 x1[k] = FAAD_CMPLX_PRETWIDDLE_SUB(k); x2[k] = FAAD_CMPLX_PRETWIDDLE_ADD(k); k++;
373 x1[k] = FAAD_CMPLX_PRETWIDDLE_SUB(k); x2[k] = FAAD_CMPLX_PRETWIDDLE_ADD(k); k++;
374 x1[k] = FAAD_CMPLX_PRETWIDDLE_SUB(k); x2[k] = FAAD_CMPLX_PRETWIDDLE_ADD(k); k++;
375 x1[k] = FAAD_CMPLX_PRETWIDDLE_SUB(k); x2[k] = FAAD_CMPLX_PRETWIDDLE_ADD(k); k++;
376 }
377
378 /* transform */
379 DCT4_32(x1, x1);
380 DST4_32(x2, x2);
381
382 idx0 = qmfs->v_index;
383 idx1 = qmfs->v_index + 63;
384 for (n = 0; n < 32; n+=2)
385 {
386 qmfs->v[idx0] = qmfs->v[idx0 + 640] = -x1[n ] + x2[n ]; idx0++;
387 qmfs->v[idx1] = qmfs->v[idx1 + 640] = x1[n ] + x2[n ]; idx1--;
388 qmfs->v[idx0] = qmfs->v[idx0 + 640] = -x1[n+1] + x2[n+1]; idx0++;
389 qmfs->v[idx1] = qmfs->v[idx1 + 640] = x1[n+1] + x2[n+1]; idx1--;
390 }
391
392 /* calculate 32 output samples and window */
393 for (k = 0; k < 32; k++)
394 {
395 idx0 = qmfs->v_index + k; idx1 = 2*k*10;
396 output[out++] = FAAD_SYNTHESIS_SCALE(
397 MUL_F(qmfs->v[idx0 ], qmf_c[idx1 ]) +
398 MUL_F(qmfs->v[idx0 + 96], qmf_c[idx1+1]) +
399 MUL_F(qmfs->v[idx0 + 128], qmf_c[idx1+2]) +
400 MUL_F(qmfs->v[idx0 + 224], qmf_c[idx1+3]) +
401 MUL_F(qmfs->v[idx0 + 256], qmf_c[idx1+4]) +
402 MUL_F(qmfs->v[idx0 + 352], qmf_c[idx1+5]) +
403 MUL_F(qmfs->v[idx0 + 384], qmf_c[idx1+6]) +
404 MUL_F(qmfs->v[idx0 + 480], qmf_c[idx1+7]) +
405 MUL_F(qmfs->v[idx0 + 512], qmf_c[idx1+8]) +
406 MUL_F(qmfs->v[idx0 + 608], qmf_c[idx1+9]));
407 }
408
409 /* update ringbuffer index */
410 qmfs->v_index -= 64;
411 if (qmfs->v_index < 0)
412 qmfs->v_index = (640 - 64);
413 }
414 }
415
416 void sbr_qmf_synthesis_64(sbr_info *sbr, qmfs_info *qmfs, qmf_t X[MAX_NTSR][64],
417 real_t *output)
418 {
419 real_t real1[32] MEM_ALIGN_ATTR;
420 real_t imag1[32] MEM_ALIGN_ATTR;
421 real_t real2[32] MEM_ALIGN_ATTR;
422 real_t imag2[32] MEM_ALIGN_ATTR;
423 qmf_t *pX;
424 real_t *p_buf_1, *p_buf_3;
425 int32_t n, k, idx0, idx1, out = 0;
426 uint32_t l;
427
428 /* qmf subsample l */
429 for (l = 0; l < sbr->numTimeSlotsRate; l++)
430 {
431 /* shift buffer v */
432 /* buffer is not shifted, we use double ringbuffer */
433 //memmove(qmfs->v + 128, qmfs->v, (1280-128)*sizeof(real_t));
434
435 /* calculate 128 samples */
436 pX = X[l];
437 for (k = 0; k < 32; k++)
438 {
439 idx0 = 2*k; idx1 = idx0+1;
440 real1[ k] = QMF_RE(pX[idx0]); imag2[ k] = QMF_IM(pX[idx0]);
441 imag1[31-k] = QMF_RE(pX[idx1]); real2[31-k] = QMF_IM(pX[idx1]);
442 }
443
444 // dct4_kernel is DCT_IV without reordering which is done before and after FFT
445 dct4_kernel(real1, imag1);
446 dct4_kernel(real2, imag2);
447
448 p_buf_1 = qmfs->v + qmfs->v_index;
449 p_buf_3 = p_buf_1 + 1280;
450
451 idx0 = 0; idx1 = 127;
452 for (n = 0; n < 32; n++)
453 {
454 p_buf_1[idx0] = p_buf_3[idx0] = real2[ n] - real1[ n]; idx0++;
455 p_buf_1[idx1] = p_buf_3[idx1] = real2[ n] + real1[ n]; idx1--;
456 p_buf_1[idx0] = p_buf_3[idx0] = imag2[31-n] + imag1[31-n]; idx0++;
457 p_buf_1[idx1] = p_buf_3[idx1] = imag2[31-n] - imag1[31-n]; idx1--;
458 }
459
460 p_buf_1 = qmfs->v + qmfs->v_index;
461
462 /* calculate 64 output samples and window */
463 #ifdef CPU_ARM
464 const real_t *qtab = qmf_c;
465 real_t *pbuf = p_buf_1;
466 for (k = 0; k < 64; k++, pbuf++)
467 {
468 real_t *pout = &output[out++];
469 asm volatile (
470 "ldmia %[qtab]!, { r0-r3 } \n\t"
471 "ldr r4, [%[pbuf]] \n\t"
472 "ldr r7, [%[pbuf], #192*4] \n\t"
473 "smull r5, r6, r4, r0 \n\t"
474 "ldr r4, [%[pbuf], #256*4] \n\t"
475 "smlal r5, r6, r7, r1 \n\t"
476 "ldr r7, [%[pbuf], #448*4] \n\t"
477 "smlal r5, r6, r4, r2 \n\t"
478 "ldr r4, [%[pbuf], #512*4] \n\t"
479 "smlal r5, r6, r7, r3 \n\t"
480
481 "ldmia %[qtab]!, { r0-r3 } \n\t"
482 "ldr r7, [%[pbuf], #704*4] \n\t"
483 "smlal r5, r6, r4, r0 \n\t"
484 "ldr r4, [%[pbuf], #768*4] \n\t"
485 "smlal r5, r6, r7, r1 \n\t"
486 "ldr r7, [%[pbuf], #960*4] \n\t"
487 "smlal r5, r6, r4, r2 \n\t"
488 "mov r2, #1024*4 \n\t"
489
490 "ldmia %[qtab]!, { r0-r1 } \n\t"
491 "ldr r4, [%[pbuf], r2] \n\t"
492 "smlal r5, r6, r7, r3 \n\t"
493 "mov r2, #1216*4 \n\t"
494 "ldr r7, [%[pbuf], r2] \n\t"
495 "smlal r5, r6, r4, r0 \n\t"
496 "smlal r5, r6, r7, r1 \n\t"
497
498 "str r6, [%[pout]] \n"
499 : [qtab] "+r" (qtab)
500 : [pbuf] "r" (pbuf), [pout] "r" (pout)
501 : "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7", "memory");
502 }
503 #elif defined CPU_COLDFIRE
504 const real_t *qtab = qmf_c;
505 real_t *pbuf = p_buf_1;
506 for (k = 0; k < 64; k++, pbuf++)
507 {
508 real_t *pout = &output[out++];
509 asm volatile (
510 "move.l (%[pbuf]), %%d5 \n"
511
512 "movem.l (%[qtab]), %%d0-%%d4 \n"
513 "mac.l %%d0, %%d5, (192*4, %[pbuf]), %%d5, %%acc0 \n"
514 "mac.l %%d1, %%d5, (256*4, %[pbuf]), %%d5, %%acc0 \n"
515 "mac.l %%d2, %%d5, (448*4, %[pbuf]), %%d5, %%acc0 \n"
516 "mac.l %%d3, %%d5, (512*4, %[pbuf]), %%d5, %%acc0 \n"
517 "mac.l %%d4, %%d5, (704*4, %[pbuf]), %%d5, %%acc0 \n"
518 "lea.l (20, %[qtab]), %[qtab] \n"
519
520 "movem.l (%[qtab]), %%d0-%%d4 \n"
521 "mac.l %%d0, %%d5, (768*4, %[pbuf]), %%d5, %%acc0 \n"
522 "mac.l %%d1, %%d5, (960*4, %[pbuf]), %%d5, %%acc0 \n"
523 "mac.l %%d2, %%d5, (1024*4, %[pbuf]), %%d5, %%acc0 \n"
524 "mac.l %%d3, %%d5, (1216*4, %[pbuf]), %%d5, %%acc0 \n"
525 "mac.l %%d4, %%d5, %%acc0 \n"
526 "lea.l (20, %[qtab]), %[qtab] \n"
527
528 "movclr.l %%acc0, %%d0 \n"
529 "move.l %%d0, (%[pout]) \n"
530 : [qtab] "+a" (qtab)
531 : [pbuf] "a" (pbuf),
532 [pout] "a" (pout)
533 : "d0", "d1", "d2", "d3", "d4", "d5", "memory");
534 }
535 #else
536 for (k = 0; k < 64; k++)
537 {
538 idx0 = k*10;
539 output[out++] = FAAD_SYNTHESIS_SCALE(
540 MUL_F(p_buf_1[k ], qmf_c[idx0 ]) +
541 MUL_F(p_buf_1[k+ 192 ], qmf_c[idx0+1]) +
542 MUL_F(p_buf_1[k+ 256 ], qmf_c[idx0+2]) +
543 MUL_F(p_buf_1[k+ 256+192], qmf_c[idx0+3]) +
544 MUL_F(p_buf_1[k+ 512 ], qmf_c[idx0+4]) +
545 MUL_F(p_buf_1[k+ 512+192], qmf_c[idx0+5]) +
546 MUL_F(p_buf_1[k+ 768 ], qmf_c[idx0+6]) +
547 MUL_F(p_buf_1[k+ 768+192], qmf_c[idx0+7]) +
548 MUL_F(p_buf_1[k+1024 ], qmf_c[idx0+8]) +
549 MUL_F(p_buf_1[k+1024+192], qmf_c[idx0+9]));
550 }
551 #endif
552
553 /* update ringbuffer index */
554 qmfs->v_index -= 128;
555 if (qmfs->v_index < 0)
556 qmfs->v_index = (1280 - 128);
557 }
558 }
559 #endif /* #ifdef SBR_LOW_POWER */
560
561 #endif /* #ifdef SBR_DEC */
My public Rockbox git repo