Merge commit '801c39e1e3058fc4ba822bfb5d8612d777111e32'
[ffmpeg.git] / libavcodec / mips / aacdec_mips.c
1 /*
2  * Copyright (c) 2012
3  *      MIPS Technologies, Inc., California.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14  *    contributors may be used to endorse or promote products derived from
15  *    this software without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * Authors:  Darko Laus      (darko@mips.com)
30  *           Djordje Pesut   (djordje@mips.com)
31  *           Mirjana Vulin   (mvulin@mips.com)
32  *
33  * This file is part of FFmpeg.
34  *
35  * FFmpeg is free software; you can redistribute it and/or
36  * modify it under the terms of the GNU Lesser General Public
37  * License as published by the Free Software Foundation; either
38  * version 2.1 of the License, or (at your option) any later version.
39  *
40  * FFmpeg is distributed in the hope that it will be useful,
41  * but WITHOUT ANY WARRANTY; without even the implied warranty of
42  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
43  * Lesser General Public License for more details.
44  *
45  * You should have received a copy of the GNU Lesser General Public
46  * License along with FFmpeg; if not, write to the Free Software
47  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
48  */
49
50 /**
51  * @file
52  * Reference: libavcodec/aacdec.c
53  */
54
55 #include "libavcodec/aac.h"
56 #include "aacdec_mips.h"
57 #include "libavcodec/aactab.h"
58 #include "libavcodec/sinewin.h"
59
60 #if HAVE_INLINE_ASM
61 static av_always_inline int lcg_random(unsigned previous_val)
62 {
63     union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
64     return v.s;
65 }
66
67 static void imdct_and_windowing_mips(AACContext *ac, SingleChannelElement *sce)
68 {
69     IndividualChannelStream *ics = &sce->ics;
70     float *in    = sce->coeffs;
71     float *out   = sce->ret;
72     float *saved = sce->saved;
73     const float *swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
74     const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
75     const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
76     float *buf  = ac->buf_mdct;
77     int i;
78
79     if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
80         for (i = 0; i < 1024; i += 128)
81             ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
82     } else
83         ac->mdct.imdct_half(&ac->mdct, buf, in);
84
85     /* window overlapping
86      * NOTE: To simplify the overlapping code, all 'meaningless' short to long
87      * and long to short transitions are considered to be short to short
88      * transitions. This leaves just two cases (long to long and short to short)
89      * with a little special sauce for EIGHT_SHORT_SEQUENCE.
90      */
91     if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
92             (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
93         ac->fdsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, 512);
94     } else {
95         {
96             float *buf1 = saved;
97             float *buf2 = out;
98             int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
99             int loop_end;
100
101             /* loop unrolled 8 times */
102             __asm__ volatile (
103                 ".set push                               \n\t"
104                 ".set noreorder                          \n\t"
105                 "addiu   %[loop_end], %[src],      1792  \n\t"
106             "1:                                          \n\t"
107                 "lw      %[temp0],    0(%[src])          \n\t"
108                 "lw      %[temp1],    4(%[src])          \n\t"
109                 "lw      %[temp2],    8(%[src])          \n\t"
110                 "lw      %[temp3],    12(%[src])         \n\t"
111                 "lw      %[temp4],    16(%[src])         \n\t"
112                 "lw      %[temp5],    20(%[src])         \n\t"
113                 "lw      %[temp6],    24(%[src])         \n\t"
114                 "lw      %[temp7],    28(%[src])         \n\t"
115                 "addiu   %[src],      %[src],      32    \n\t"
116                 "sw      %[temp0],    0(%[dst])          \n\t"
117                 "sw      %[temp1],    4(%[dst])          \n\t"
118                 "sw      %[temp2],    8(%[dst])          \n\t"
119                 "sw      %[temp3],    12(%[dst])         \n\t"
120                 "sw      %[temp4],    16(%[dst])         \n\t"
121                 "sw      %[temp5],    20(%[dst])         \n\t"
122                 "sw      %[temp6],    24(%[dst])         \n\t"
123                 "sw      %[temp7],    28(%[dst])         \n\t"
124                 "bne     %[src],      %[loop_end], 1b    \n\t"
125                 " addiu  %[dst],      %[dst],      32    \n\t"
126                 ".set pop                                \n\t"
127
128                 : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
129                   [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
130                   [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
131                   [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
132                   [loop_end]"=&r"(loop_end), [src]"+r"(buf1),
133                   [dst]"+r"(buf2)
134                 :
135                 : "memory"
136             );
137         }
138
139         if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
140             {
141                 float wi;
142                 float wj;
143                 int i;
144                 float temp0, temp1, temp2, temp3;
145                 float *dst0 = out + 448 + 0*128;
146                 float *dst1 = dst0 + 64 + 63;
147                 float *dst2 = saved + 63;
148                 float *win0 = (float*)swindow;
149                 float *win1 = win0 + 64 + 63;
150                 float *win0_prev = (float*)swindow_prev;
151                 float *win1_prev = win0_prev + 64 + 63;
152                 float *src0_prev = saved + 448;
153                 float *src1_prev = buf + 0*128 + 63;
154                 float *src0 = buf + 0*128 + 64;
155                 float *src1 = buf + 1*128 + 63;
156
157                 for(i = 0; i < 64; i++)
158                 {
159                     temp0 = src0_prev[0];
160                     temp1 = src1_prev[0];
161                     wi = *win0_prev;
162                     wj = *win1_prev;
163                     temp2 = src0[0];
164                     temp3 = src1[0];
165                     dst0[0] = temp0 * wj - temp1 * wi;
166                     dst1[0] = temp0 * wi + temp1 * wj;
167
168                     wi = *win0;
169                     wj = *win1;
170
171                     temp0 = src0[128];
172                     temp1 = src1[128];
173                     dst0[128] = temp2 * wj - temp3 * wi;
174                     dst1[128] = temp2 * wi + temp3 * wj;
175
176                     temp2 = src0[256];
177                     temp3 = src1[256];
178                     dst0[256] = temp0 * wj - temp1 * wi;
179                     dst1[256] = temp0 * wi + temp1 * wj;
180                     dst0[384] = temp2 * wj - temp3 * wi;
181                     dst1[384] = temp2 * wi + temp3 * wj;
182
183                     temp0 = src0[384];
184                     temp1 = src1[384];
185                     dst0[512] = temp0 * wj - temp1 * wi;
186                     dst2[0] = temp0 * wi + temp1 * wj;
187
188                     src0++;
189                     src1--;
190                     src0_prev++;
191                     src1_prev--;
192                     win0++;
193                     win1--;
194                     win0_prev++;
195                     win1_prev--;
196                     dst0++;
197                     dst1--;
198                     dst2--;
199                 }
200             }
201         } else {
202             ac->fdsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, 64);
203             {
204                 float *buf1 = buf + 64;
205                 float *buf2 = out + 576;
206                 int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
207                 int loop_end;
208
209                 /* loop unrolled 8 times */
210                 __asm__ volatile (
211                     ".set push                               \n\t"
212                     ".set noreorder                          \n\t"
213                     "addiu   %[loop_end], %[src],      1792  \n\t"
214                 "1:                                          \n\t"
215                     "lw      %[temp0],    0(%[src])          \n\t"
216                     "lw      %[temp1],    4(%[src])          \n\t"
217                     "lw      %[temp2],    8(%[src])          \n\t"
218                     "lw      %[temp3],    12(%[src])         \n\t"
219                     "lw      %[temp4],    16(%[src])         \n\t"
220                     "lw      %[temp5],    20(%[src])         \n\t"
221                     "lw      %[temp6],    24(%[src])         \n\t"
222                     "lw      %[temp7],    28(%[src])         \n\t"
223                     "addiu   %[src],      %[src],      32    \n\t"
224                     "sw      %[temp0],    0(%[dst])          \n\t"
225                     "sw      %[temp1],    4(%[dst])          \n\t"
226                     "sw      %[temp2],    8(%[dst])          \n\t"
227                     "sw      %[temp3],    12(%[dst])         \n\t"
228                     "sw      %[temp4],    16(%[dst])         \n\t"
229                     "sw      %[temp5],    20(%[dst])         \n\t"
230                     "sw      %[temp6],    24(%[dst])         \n\t"
231                     "sw      %[temp7],    28(%[dst])         \n\t"
232                     "bne     %[src],      %[loop_end], 1b    \n\t"
233                     " addiu  %[dst],      %[dst],      32    \n\t"
234                     ".set pop                                \n\t"
235
236                     : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
237                       [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
238                       [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
239                       [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
240                       [loop_end]"=&r"(loop_end), [src]"+r"(buf1),
241                       [dst]"+r"(buf2)
242                     :
243                     : "memory"
244                 );
245             }
246         }
247     }
248
249     // buffer update
250     if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
251         ac->fdsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 64);
252         ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
253         ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
254         {
255             float *buf1 = buf + 7*128 + 64;
256             float *buf2 = saved + 448;
257             int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
258             int loop_end;
259
260             /* loop unrolled 8 times */
261             __asm__ volatile (
262                 ".set push                                \n\t"
263                 ".set noreorder                           \n\t"
264                 "addiu   %[loop_end], %[src],       256   \n\t"
265             "1:                                           \n\t"
266                 "lw      %[temp0],    0(%[src])           \n\t"
267                 "lw      %[temp1],    4(%[src])           \n\t"
268                 "lw      %[temp2],    8(%[src])           \n\t"
269                 "lw      %[temp3],    12(%[src])          \n\t"
270                 "lw      %[temp4],    16(%[src])          \n\t"
271                 "lw      %[temp5],    20(%[src])          \n\t"
272                 "lw      %[temp6],    24(%[src])          \n\t"
273                 "lw      %[temp7],    28(%[src])          \n\t"
274                 "addiu   %[src],      %[src],       32    \n\t"
275                 "sw      %[temp0],    0(%[dst])           \n\t"
276                 "sw      %[temp1],    4(%[dst])           \n\t"
277                 "sw      %[temp2],    8(%[dst])           \n\t"
278                 "sw      %[temp3],    12(%[dst])          \n\t"
279                 "sw      %[temp4],    16(%[dst])          \n\t"
280                 "sw      %[temp5],    20(%[dst])          \n\t"
281                 "sw      %[temp6],    24(%[dst])          \n\t"
282                 "sw      %[temp7],    28(%[dst])          \n\t"
283                 "bne     %[src],      %[loop_end],  1b    \n\t"
284                 " addiu  %[dst],      %[dst],       32    \n\t"
285                 ".set pop                                 \n\t"
286
287                 : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
288                   [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
289                   [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
290                   [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
291                   [loop_end]"=&r"(loop_end), [src]"+r"(buf1),
292                   [dst]"+r"(buf2)
293                 :
294                 : "memory"
295             );
296         }
297     } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
298         float *buf1 = buf + 512;
299         float *buf2 = saved;
300         int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
301         int loop_end;
302
303         /* loop unrolled 8 times */
304         __asm__ volatile (
305             ".set push                                \n\t"
306             ".set noreorder                           \n\t"
307             "addiu   %[loop_end], %[src],       1792  \n\t"
308         "1:                                           \n\t"
309             "lw      %[temp0],    0(%[src])           \n\t"
310             "lw      %[temp1],    4(%[src])           \n\t"
311             "lw      %[temp2],    8(%[src])           \n\t"
312             "lw      %[temp3],    12(%[src])          \n\t"
313             "lw      %[temp4],    16(%[src])          \n\t"
314             "lw      %[temp5],    20(%[src])          \n\t"
315             "lw      %[temp6],    24(%[src])          \n\t"
316             "lw      %[temp7],    28(%[src])          \n\t"
317             "addiu   %[src],      %[src],       32    \n\t"
318             "sw      %[temp0],    0(%[dst])           \n\t"
319             "sw      %[temp1],    4(%[dst])           \n\t"
320             "sw      %[temp2],    8(%[dst])           \n\t"
321             "sw      %[temp3],    12(%[dst])          \n\t"
322             "sw      %[temp4],    16(%[dst])          \n\t"
323             "sw      %[temp5],    20(%[dst])          \n\t"
324             "sw      %[temp6],    24(%[dst])          \n\t"
325             "sw      %[temp7],    28(%[dst])          \n\t"
326             "bne     %[src],      %[loop_end],  1b    \n\t"
327             " addiu  %[dst],      %[dst],       32    \n\t"
328             ".set pop                                 \n\t"
329
330             : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
331               [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
332               [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
333               [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
334               [loop_end]"=&r"(loop_end), [src]"+r"(buf1),
335               [dst]"+r"(buf2)
336             :
337             : "memory"
338         );
339         {
340             float *buf1 = buf + 7*128 + 64;
341             float *buf2 = saved + 448;
342             int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
343             int loop_end;
344
345             /* loop unrolled 8 times */
346             __asm__ volatile (
347                 ".set push                                 \n\t"
348                 ".set noreorder                            \n\t"
349                 "addiu   %[loop_end], %[src],        256   \n\t"
350             "1:                                            \n\t"
351                 "lw      %[temp0],    0(%[src])            \n\t"
352                 "lw      %[temp1],    4(%[src])            \n\t"
353                 "lw      %[temp2],    8(%[src])            \n\t"
354                 "lw      %[temp3],    12(%[src])           \n\t"
355                 "lw      %[temp4],    16(%[src])           \n\t"
356                 "lw      %[temp5],    20(%[src])           \n\t"
357                 "lw      %[temp6],    24(%[src])           \n\t"
358                 "lw      %[temp7],    28(%[src])           \n\t"
359                 "addiu   %[src],      %[src],        32    \n\t"
360                 "sw      %[temp0],    0(%[dst])            \n\t"
361                 "sw      %[temp1],    4(%[dst])            \n\t"
362                 "sw      %[temp2],    8(%[dst])            \n\t"
363                 "sw      %[temp3],    12(%[dst])           \n\t"
364                 "sw      %[temp4],    16(%[dst])           \n\t"
365                 "sw      %[temp5],    20(%[dst])           \n\t"
366                 "sw      %[temp6],    24(%[dst])           \n\t"
367                 "sw      %[temp7],    28(%[dst])           \n\t"
368                 "bne     %[src],      %[loop_end],   1b    \n\t"
369                 " addiu  %[dst],      %[dst],        32    \n\t"
370                 ".set pop                                  \n\t"
371
372                 : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
373                   [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
374                   [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
375                   [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
376                   [loop_end]"=&r"(loop_end), [src]"+r"(buf1),
377                   [dst]"+r"(buf2)
378                 :
379                 : "memory"
380             );
381         }
382     } else { // LONG_STOP or ONLY_LONG
383         float *buf1 = buf + 512;
384         float *buf2 = saved;
385         int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
386         int loop_end;
387
388         /* loop unrolled 8 times */
389         __asm__ volatile (
390             ".set push                                 \n\t"
391             ".set noreorder                            \n\t"
392             "addiu   %[loop_end], %[src],        2048  \n\t"
393         "1:                                            \n\t"
394             "lw      %[temp0],    0(%[src])            \n\t"
395             "lw      %[temp1],    4(%[src])            \n\t"
396             "lw      %[temp2],    8(%[src])            \n\t"
397             "lw      %[temp3],    12(%[src])           \n\t"
398             "lw      %[temp4],    16(%[src])           \n\t"
399             "lw      %[temp5],    20(%[src])           \n\t"
400             "lw      %[temp6],    24(%[src])           \n\t"
401             "lw      %[temp7],    28(%[src])           \n\t"
402             "addiu   %[src],      %[src],        32    \n\t"
403             "sw      %[temp0],    0(%[dst])            \n\t"
404             "sw      %[temp1],    4(%[dst])            \n\t"
405             "sw      %[temp2],    8(%[dst])            \n\t"
406             "sw      %[temp3],    12(%[dst])           \n\t"
407             "sw      %[temp4],    16(%[dst])           \n\t"
408             "sw      %[temp5],    20(%[dst])           \n\t"
409             "sw      %[temp6],    24(%[dst])           \n\t"
410             "sw      %[temp7],    28(%[dst])           \n\t"
411             "bne     %[src],      %[loop_end],   1b    \n\t"
412             " addiu  %[dst],      %[dst],        32    \n\t"
413             ".set pop                                  \n\t"
414
415             : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
416               [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
417               [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
418               [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
419               [loop_end]"=&r"(loop_end), [src]"+r"(buf1),
420               [dst]"+r"(buf2)
421             :
422             : "memory"
423         );
424     }
425 }
426
427 static void apply_ltp_mips(AACContext *ac, SingleChannelElement *sce)
428 {
429     const LongTermPrediction *ltp = &sce->ics.ltp;
430     const uint16_t *offsets = sce->ics.swb_offset;
431     int i, sfb;
432     int j, k;
433
434     if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
435         float *predTime = sce->ret;
436         float *predFreq = ac->buf_mdct;
437         float *p_predTime;
438         int16_t num_samples = 2048;
439
440         if (ltp->lag < 1024)
441             num_samples = ltp->lag + 1024;
442             j = (2048 - num_samples) >> 2;
443             k = (2048 - num_samples) & 3;
444             p_predTime = &predTime[num_samples];
445
446         for (i = 0; i < num_samples; i++)
447             predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
448         for (i = 0; i < j; i++) {
449
450             /* loop unrolled 4 times */
451             __asm__ volatile (
452                 "sw      $0,              0(%[p_predTime])        \n\t"
453                 "sw      $0,              4(%[p_predTime])        \n\t"
454                 "sw      $0,              8(%[p_predTime])        \n\t"
455                 "sw      $0,              12(%[p_predTime])       \n\t"
456                 "addiu   %[p_predTime],   %[p_predTime],     16   \n\t"
457
458                 : [p_predTime]"+r"(p_predTime)
459                 :
460                 : "memory"
461             );
462         }
463         for (i = 0; i < k; i++) {
464
465             __asm__ volatile (
466                 "sw      $0,              0(%[p_predTime])        \n\t"
467                 "addiu   %[p_predTime],   %[p_predTime],     4    \n\t"
468
469                 : [p_predTime]"+r"(p_predTime)
470                 :
471                 : "memory"
472             );
473         }
474
475         ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
476
477         if (sce->tns.present)
478             ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
479
480         for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
481             if (ltp->used[sfb])
482                 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
483                     sce->coeffs[i] += predFreq[i];
484     }
485 }
486
487 #if HAVE_MIPSFPU
488 static void update_ltp_mips(AACContext *ac, SingleChannelElement *sce)
489 {
490     IndividualChannelStream *ics = &sce->ics;
491     float *saved     = sce->saved;
492     float *saved_ltp = sce->coeffs;
493     const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
494     const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
495     int i;
496     int loop_end, loop_end1, loop_end2;
497     float temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7, temp8, temp9, temp10, temp11;
498
499     if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
500         float *buf = saved;
501         float *buf0 = saved_ltp;
502         float *p_saved_ltp = saved_ltp + 576;
503         float *ptr1 = &saved_ltp[512];
504         float *ptr2 = &ac->buf_mdct[1023];
505         float *ptr3 = (float*)&swindow[63];
506         loop_end1 = (int)(p_saved_ltp + 448);
507
508         /* loop unrolled 8 times */
509         __asm__ volatile (
510             ".set push                                     \n\t"
511             ".set noreorder                                \n\t"
512             "addiu   %[loop_end],   %[src],         2048   \n\t"
513         "1:                                                \n\t"
514             "lw      %[temp0],      0(%[src])              \n\t"
515             "lw      %[temp1],      4(%[src])              \n\t"
516             "lw      %[temp2],      8(%[src])              \n\t"
517             "lw      %[temp3],      12(%[src])             \n\t"
518             "lw      %[temp4],      16(%[src])             \n\t"
519             "lw      %[temp5],      20(%[src])             \n\t"
520             "lw      %[temp6],      24(%[src])             \n\t"
521             "lw      %[temp7],      28(%[src])             \n\t"
522             "addiu   %[src],        %[src],         32     \n\t"
523             "sw      %[temp0],      0(%[dst])              \n\t"
524             "sw      %[temp1],      4(%[dst])              \n\t"
525             "sw      %[temp2],      8(%[dst])              \n\t"
526             "sw      %[temp3],      12(%[dst])             \n\t"
527             "sw      %[temp4],      16(%[dst])             \n\t"
528             "sw      %[temp5],      20(%[dst])             \n\t"
529             "sw      %[temp6],      24(%[dst])             \n\t"
530             "sw      %[temp7],      28(%[dst])             \n\t"
531             "bne     %[src],        %[loop_end],    1b     \n\t"
532             " addiu  %[dst],        %[dst],         32     \n\t"
533             ".set pop                                      \n\t"
534
535             : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
536               [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
537               [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
538               [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
539               [loop_end]"=&r"(loop_end), [src]"+r"(buf),
540               [dst]"+r"(buf0)
541             :
542             : "memory"
543         );
544
545         /* loop unrolled 8 times */
546         __asm__ volatile (
547         "1:                                                   \n\t"
548             "sw     $0,              0(%[p_saved_ltp])        \n\t"
549             "sw     $0,              4(%[p_saved_ltp])        \n\t"
550             "sw     $0,              8(%[p_saved_ltp])        \n\t"
551             "sw     $0,              12(%[p_saved_ltp])       \n\t"
552             "sw     $0,              16(%[p_saved_ltp])       \n\t"
553             "sw     $0,              20(%[p_saved_ltp])       \n\t"
554             "sw     $0,              24(%[p_saved_ltp])       \n\t"
555             "sw     $0,              28(%[p_saved_ltp])       \n\t"
556             "addiu  %[p_saved_ltp],  %[p_saved_ltp],     32   \n\t"
557             "bne    %[p_saved_ltp],  %[loop_end1],       1b   \n\t"
558
559             : [p_saved_ltp]"+r"(p_saved_ltp)
560             : [loop_end1]"r"(loop_end1)
561             : "memory"
562         );
563
564         ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     &swindow[64],      64);
565         for (i = 0; i < 16; i++){
566             /* loop unrolled 4 times */
567             __asm__ volatile (
568                 "lwc1    %[temp0],    0(%[ptr2])                \n\t"
569                 "lwc1    %[temp1],    -4(%[ptr2])               \n\t"
570                 "lwc1    %[temp2],    -8(%[ptr2])               \n\t"
571                 "lwc1    %[temp3],    -12(%[ptr2])              \n\t"
572                 "lwc1    %[temp4],    0(%[ptr3])                \n\t"
573                 "lwc1    %[temp5],    -4(%[ptr3])               \n\t"
574                 "lwc1    %[temp6],    -8(%[ptr3])               \n\t"
575                 "lwc1    %[temp7],    -12(%[ptr3])              \n\t"
576                 "mul.s   %[temp8],    %[temp0],     %[temp4]    \n\t"
577                 "mul.s   %[temp9],    %[temp1],     %[temp5]    \n\t"
578                 "mul.s   %[temp10],   %[temp2],     %[temp6]    \n\t"
579                 "mul.s   %[temp11],   %[temp3],     %[temp7]    \n\t"
580                 "swc1    %[temp8],    0(%[ptr1])                \n\t"
581                 "swc1    %[temp9],    4(%[ptr1])                \n\t"
582                 "swc1    %[temp10],   8(%[ptr1])                \n\t"
583                 "swc1    %[temp11],   12(%[ptr1])               \n\t"
584                 "addiu   %[ptr1],     %[ptr1],      16          \n\t"
585                 "addiu   %[ptr2],     %[ptr2],      -16         \n\t"
586                 "addiu   %[ptr3],     %[ptr3],      -16         \n\t"
587
588                 : [temp0]"=&f"(temp0), [temp1]"=&f"(temp1),
589                   [temp2]"=&f"(temp2), [temp3]"=&f"(temp3),
590                   [temp4]"=&f"(temp4), [temp5]"=&f"(temp5),
591                   [temp6]"=&f"(temp6), [temp7]"=&f"(temp7),
592                   [temp8]"=&f"(temp8), [temp9]"=&f"(temp9),
593                   [temp10]"=&f"(temp10), [temp11]"=&f"(temp11),
594                   [ptr1]"+r"(ptr1), [ptr2]"+r"(ptr2), [ptr3]"+r"(ptr3)
595                 :
596                 : "memory"
597             );
598         }
599     } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
600         float *buff0 = saved;
601         float *buff1 = saved_ltp;
602         float *ptr1 = &saved_ltp[512];
603         float *ptr2 = &ac->buf_mdct[1023];
604         float *ptr3 = (float*)&swindow[63];
605         loop_end = (int)(saved + 448);
606
607         /* loop unrolled 8 times */
608         __asm__ volatile (
609             ".set push                                  \n\t"
610             ".set noreorder                             \n\t"
611         "1:                                             \n\t"
612             "lw      %[temp0],    0(%[src])             \n\t"
613             "lw      %[temp1],    4(%[src])             \n\t"
614             "lw      %[temp2],    8(%[src])             \n\t"
615             "lw      %[temp3],    12(%[src])            \n\t"
616             "lw      %[temp4],    16(%[src])            \n\t"
617             "lw      %[temp5],    20(%[src])            \n\t"
618             "lw      %[temp6],    24(%[src])            \n\t"
619             "lw      %[temp7],    28(%[src])            \n\t"
620             "addiu   %[src],      %[src],         32    \n\t"
621             "sw      %[temp0],    0(%[dst])             \n\t"
622             "sw      %[temp1],    4(%[dst])             \n\t"
623             "sw      %[temp2],    8(%[dst])             \n\t"
624             "sw      %[temp3],    12(%[dst])            \n\t"
625             "sw      %[temp4],    16(%[dst])            \n\t"
626             "sw      %[temp5],    20(%[dst])            \n\t"
627             "sw      %[temp6],    24(%[dst])            \n\t"
628             "sw      %[temp7],    28(%[dst])            \n\t"
629             "sw      $0,          2304(%[dst])          \n\t"
630             "sw      $0,          2308(%[dst])          \n\t"
631             "sw      $0,          2312(%[dst])          \n\t"
632             "sw      $0,          2316(%[dst])          \n\t"
633             "sw      $0,          2320(%[dst])          \n\t"
634             "sw      $0,          2324(%[dst])          \n\t"
635             "sw      $0,          2328(%[dst])          \n\t"
636             "sw      $0,          2332(%[dst])          \n\t"
637             "bne     %[src],      %[loop_end],    1b    \n\t"
638             " addiu  %[dst],      %[dst],         32    \n\t"
639             ".set pop                                   \n\t"
640
641             : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
642               [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
643               [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
644               [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
645               [src]"+r"(buff0), [dst]"+r"(buff1)
646             : [loop_end]"r"(loop_end)
647             : "memory"
648         );
649         ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     &swindow[64],      64);
650         for (i = 0; i < 16; i++){
651             /* loop unrolled 8 times */
652             __asm__ volatile (
653                 "lwc1    %[temp0],    0(%[ptr2])                \n\t"
654                 "lwc1    %[temp1],    -4(%[ptr2])               \n\t"
655                 "lwc1    %[temp2],    -8(%[ptr2])               \n\t"
656                 "lwc1    %[temp3],    -12(%[ptr2])              \n\t"
657                 "lwc1    %[temp4],    0(%[ptr3])                \n\t"
658                 "lwc1    %[temp5],    -4(%[ptr3])               \n\t"
659                 "lwc1    %[temp6],    -8(%[ptr3])               \n\t"
660                 "lwc1    %[temp7],    -12(%[ptr3])              \n\t"
661                 "mul.s   %[temp8],    %[temp0],     %[temp4]    \n\t"
662                 "mul.s   %[temp9],    %[temp1],     %[temp5]    \n\t"
663                 "mul.s   %[temp10],   %[temp2],     %[temp6]    \n\t"
664                 "mul.s   %[temp11],   %[temp3],     %[temp7]    \n\t"
665                 "swc1    %[temp8],    0(%[ptr1])                \n\t"
666                 "swc1    %[temp9],    4(%[ptr1])                \n\t"
667                 "swc1    %[temp10],   8(%[ptr1])                \n\t"
668                 "swc1    %[temp11],   12(%[ptr1])               \n\t"
669                 "addiu   %[ptr1],     %[ptr1],      16          \n\t"
670                 "addiu   %[ptr2],     %[ptr2],      -16         \n\t"
671                 "addiu   %[ptr3],     %[ptr3],      -16         \n\t"
672
673                 : [temp0]"=&f"(temp0), [temp1]"=&f"(temp1),
674                   [temp2]"=&f"(temp2), [temp3]"=&f"(temp3),
675                   [temp4]"=&f"(temp4), [temp5]"=&f"(temp5),
676                   [temp6]"=&f"(temp6), [temp7]"=&f"(temp7),
677                   [temp8]"=&f"(temp8), [temp9]"=&f"(temp9),
678                   [temp10]"=&f"(temp10), [temp11]"=&f"(temp11),
679                   [ptr1]"+r"(ptr1), [ptr2]"+r"(ptr2), [ptr3]"+r"(ptr3)
680                 :
681                 : "memory"
682             );
683         }
684     } else { // LONG_STOP or ONLY_LONG
685         float *ptr1, *ptr2, *ptr3;
686         ac->fdsp.vector_fmul_reverse(saved_ltp,       ac->buf_mdct + 512,     &lwindow[512],     512);
687
688         ptr1 = &saved_ltp[512];
689         ptr2 = &ac->buf_mdct[1023];
690         ptr3 = (float*)&lwindow[511];
691
692         for (i = 0; i < 512; i+=4){
693             /* loop unrolled 4 times */
694             __asm__ volatile (
695                 "lwc1    %[temp0],    0(%[ptr2])                \n\t"
696                 "lwc1    %[temp1],    -4(%[ptr2])               \n\t"
697                 "lwc1    %[temp2],    -8(%[ptr2])               \n\t"
698                 "lwc1    %[temp3],    -12(%[ptr2])              \n\t"
699                 "lwc1    %[temp4],    0(%[ptr3])                \n\t"
700                 "lwc1    %[temp5],    -4(%[ptr3])               \n\t"
701                 "lwc1    %[temp6],    -8(%[ptr3])               \n\t"
702                 "lwc1    %[temp7],    -12(%[ptr3])              \n\t"
703                 "mul.s   %[temp8],    %[temp0],     %[temp4]    \n\t"
704                 "mul.s   %[temp9],    %[temp1],     %[temp5]    \n\t"
705                 "mul.s   %[temp10],   %[temp2],     %[temp6]    \n\t"
706                 "mul.s   %[temp11],   %[temp3],     %[temp7]    \n\t"
707                 "swc1    %[temp8],    0(%[ptr1])                \n\t"
708                 "swc1    %[temp9],    4(%[ptr1])                \n\t"
709                 "swc1    %[temp10],   8(%[ptr1])                \n\t"
710                 "swc1    %[temp11],   12(%[ptr1])               \n\t"
711                 "addiu   %[ptr1],     %[ptr1],      16          \n\t"
712                 "addiu   %[ptr2],     %[ptr2],      -16         \n\t"
713                 "addiu   %[ptr3],     %[ptr3],      -16         \n\t"
714
715                 : [temp0]"=&f"(temp0), [temp1]"=&f"(temp1),
716                   [temp2]"=&f"(temp2), [temp3]"=&f"(temp3),
717                   [temp4]"=&f"(temp4), [temp5]"=&f"(temp5),
718                   [temp6]"=&f"(temp6), [temp7]"=&f"(temp7),
719                   [temp8]"=&f"(temp8), [temp9]"=&f"(temp9),
720                   [temp10]"=&f"(temp10), [temp11]"=&f"(temp11),
721                   [ptr1]"+r"(ptr1), [ptr2]"+r"(ptr2),
722                   [ptr3]"+r"(ptr3)
723                 :
724                 : "memory"
725             );
726         }
727     }
728
729     {
730         float *buf1 = sce->ltp_state+1024;
731         float *buf2 = sce->ltp_state;
732         float *buf3 = sce->ret;
733         float *buf4 = sce->ltp_state+1024;
734         float *buf5 = saved_ltp;
735         float *buf6 = sce->ltp_state+2048;
736
737         /* loops unrolled 8 times */
738         __asm__ volatile (
739             ".set push                                    \n\t"
740             ".set noreorder                               \n\t"
741             "addiu   %[loop_end],   %[src],         4096  \n\t"
742             "addiu   %[loop_end1],  %[src1],        4096  \n\t"
743             "addiu   %[loop_end2],  %[src2],        4096  \n\t"
744         "1:                                               \n\t"
745             "lw      %[temp0],      0(%[src])             \n\t"
746             "lw      %[temp1],      4(%[src])             \n\t"
747             "lw      %[temp2],      8(%[src])             \n\t"
748             "lw      %[temp3],      12(%[src])            \n\t"
749             "lw      %[temp4],      16(%[src])            \n\t"
750             "lw      %[temp5],      20(%[src])            \n\t"
751             "lw      %[temp6],      24(%[src])            \n\t"
752             "lw      %[temp7],      28(%[src])            \n\t"
753             "addiu   %[src],        %[src],         32    \n\t"
754             "sw      %[temp0],      0(%[dst])             \n\t"
755             "sw      %[temp1],      4(%[dst])             \n\t"
756             "sw      %[temp2],      8(%[dst])             \n\t"
757             "sw      %[temp3],      12(%[dst])            \n\t"
758             "sw      %[temp4],      16(%[dst])            \n\t"
759             "sw      %[temp5],      20(%[dst])            \n\t"
760             "sw      %[temp6],      24(%[dst])            \n\t"
761             "sw      %[temp7],      28(%[dst])            \n\t"
762             "bne     %[src],        %[loop_end],    1b    \n\t"
763             " addiu  %[dst],        %[dst],         32    \n\t"
764         "2:                                               \n\t"
765             "lw      %[temp0],      0(%[src1])            \n\t"
766             "lw      %[temp1],      4(%[src1])            \n\t"
767             "lw      %[temp2],      8(%[src1])            \n\t"
768             "lw      %[temp3],      12(%[src1])           \n\t"
769             "lw      %[temp4],      16(%[src1])           \n\t"
770             "lw      %[temp5],      20(%[src1])           \n\t"
771             "lw      %[temp6],      24(%[src1])           \n\t"
772             "lw      %[temp7],      28(%[src1])           \n\t"
773             "addiu   %[src1],       %[src1],        32    \n\t"
774             "sw      %[temp0],      0(%[dst1])            \n\t"
775             "sw      %[temp1],      4(%[dst1])            \n\t"
776             "sw      %[temp2],      8(%[dst1])            \n\t"
777             "sw      %[temp3],      12(%[dst1])           \n\t"
778             "sw      %[temp4],      16(%[dst1])           \n\t"
779             "sw      %[temp5],      20(%[dst1])           \n\t"
780             "sw      %[temp6],      24(%[dst1])           \n\t"
781             "sw      %[temp7],      28(%[dst1])           \n\t"
782             "bne     %[src1],       %[loop_end1],   2b    \n\t"
783             " addiu  %[dst1],       %[dst1],        32    \n\t"
784         "3:                                               \n\t"
785             "lw      %[temp0],      0(%[src2])            \n\t"
786             "lw      %[temp1],      4(%[src2])            \n\t"
787             "lw      %[temp2],      8(%[src2])            \n\t"
788             "lw      %[temp3],      12(%[src2])           \n\t"
789             "lw      %[temp4],      16(%[src2])           \n\t"
790             "lw      %[temp5],      20(%[src2])           \n\t"
791             "lw      %[temp6],      24(%[src2])           \n\t"
792             "lw      %[temp7],      28(%[src2])           \n\t"
793             "addiu   %[src2],       %[src2],        32    \n\t"
794             "sw      %[temp0],      0(%[dst2])            \n\t"
795             "sw      %[temp1],      4(%[dst2])            \n\t"
796             "sw      %[temp2],      8(%[dst2])            \n\t"
797             "sw      %[temp3],      12(%[dst2])           \n\t"
798             "sw      %[temp4],      16(%[dst2])           \n\t"
799             "sw      %[temp5],      20(%[dst2])           \n\t"
800             "sw      %[temp6],      24(%[dst2])           \n\t"
801             "sw      %[temp7],      28(%[dst2])           \n\t"
802             "bne     %[src2],       %[loop_end2],   3b    \n\t"
803             " addiu  %[dst2],       %[dst2],        32    \n\t"
804             ".set pop                                     \n\t"
805
806             : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
807               [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
808               [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
809               [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
810               [loop_end]"=&r"(loop_end), [loop_end1]"=&r"(loop_end1),
811               [loop_end2]"=&r"(loop_end2), [src]"+r"(buf1),
812               [dst]"+r"(buf2), [src1]"+r"(buf3), [dst1]"+r"(buf4),
813               [src2]"+r"(buf5), [dst2]"+r"(buf6)
814             :
815             : "memory"
816         );
817     }
818 }
819 #endif /* HAVE_MIPSFPU */
820 #endif /* HAVE_INLINE_ASM */
821
822 void ff_aacdec_init_mips(AACContext *c)
823 {
824 #if HAVE_INLINE_ASM
825     c->imdct_and_windowing         = imdct_and_windowing_mips;
826     c->apply_ltp                   = apply_ltp_mips;
827 #if HAVE_MIPSFPU
828     c->update_ltp                  = update_ltp_mips;
829 #endif /* HAVE_MIPSFPU */
830 #endif /* HAVE_INLINE_ASM */
831 }