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 "libavutil/attributes.h"
56#include "libavcodec/aac.h"
57#include "aacdec_mips.h"
58#include "libavcodec/aactab.h"
59#include "libavcodec/sinewin.h"
60#include "libavutil/mips/asmdefs.h"
61
62#if HAVE_INLINE_ASM
63#if HAVE_MIPSFPU
64static av_always_inline void float_copy(float *dst, const float *src, int count)
65{
66    // Copy 'count' floats from src to dst
67    const float *loop_end = src + count;
68    int temp[8];
69
70    // count must be a multiple of 8
71    av_assert2(count % 8 == 0);
72
73    // loop unrolled 8 times
74    __asm__ volatile (
75        ".set push                               \n\t"
76        ".set noreorder                          \n\t"
77    "1:                                          \n\t"
78        "lw      %[temp0],    0(%[src])          \n\t"
79        "lw      %[temp1],    4(%[src])          \n\t"
80        "lw      %[temp2],    8(%[src])          \n\t"
81        "lw      %[temp3],    12(%[src])         \n\t"
82        "lw      %[temp4],    16(%[src])         \n\t"
83        "lw      %[temp5],    20(%[src])         \n\t"
84        "lw      %[temp6],    24(%[src])         \n\t"
85        "lw      %[temp7],    28(%[src])         \n\t"
86        PTR_ADDIU "%[src],    %[src],      32    \n\t"
87        "sw      %[temp0],    0(%[dst])          \n\t"
88        "sw      %[temp1],    4(%[dst])          \n\t"
89        "sw      %[temp2],    8(%[dst])          \n\t"
90        "sw      %[temp3],    12(%[dst])         \n\t"
91        "sw      %[temp4],    16(%[dst])         \n\t"
92        "sw      %[temp5],    20(%[dst])         \n\t"
93        "sw      %[temp6],    24(%[dst])         \n\t"
94        "sw      %[temp7],    28(%[dst])         \n\t"
95        "bne     %[src],      %[loop_end], 1b    \n\t"
96        PTR_ADDIU "%[dst],    %[dst],      32    \n\t"
97        ".set pop                                \n\t"
98
99        : [temp0]"=&r"(temp[0]), [temp1]"=&r"(temp[1]),
100          [temp2]"=&r"(temp[2]), [temp3]"=&r"(temp[3]),
101          [temp4]"=&r"(temp[4]), [temp5]"=&r"(temp[5]),
102          [temp6]"=&r"(temp[6]), [temp7]"=&r"(temp[7]),
103          [src]"+r"(src), [dst]"+r"(dst)
104        : [loop_end]"r"(loop_end)
105        : "memory"
106    );
107}
108
109static av_always_inline int lcg_random(unsigned previous_val)
110{
111    union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
112    return v.s;
113}
114
115static void imdct_and_windowing_mips(AACContext *ac, SingleChannelElement *sce)
116{
117    IndividualChannelStream *ics = &sce->ics;
118    float *in    = sce->coeffs;
119    float *out   = sce->ret;
120    float *saved = sce->saved;
121    const float *swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
122    const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
123    const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
124    float *buf  = ac->buf_mdct;
125    int i;
126
127    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
128        for (i = 0; i < 1024; i += 128)
129            ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
130    } else
131        ac->mdct.imdct_half(&ac->mdct, buf, in);
132
133    /* window overlapping
134     * NOTE: To simplify the overlapping code, all 'meaningless' short to long
135     * and long to short transitions are considered to be short to short
136     * transitions. This leaves just two cases (long to long and short to short)
137     * with a little special sauce for EIGHT_SHORT_SEQUENCE.
138     */
139    if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
140            (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
141        ac->fdsp->vector_fmul_window(    out,               saved,            buf,         lwindow_prev, 512);
142    } else {
143        float_copy(out, saved, 448);
144
145        if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
146            {
147                float wi;
148                float wj;
149                int i;
150                float temp0, temp1, temp2, temp3;
151                float *dst0 = out + 448 + 0*128;
152                float *dst1 = dst0 + 64 + 63;
153                float *dst2 = saved + 63;
154                float *win0 = (float*)swindow;
155                float *win1 = win0 + 64 + 63;
156                float *win0_prev = (float*)swindow_prev;
157                float *win1_prev = win0_prev + 64 + 63;
158                float *src0_prev = saved + 448;
159                float *src1_prev = buf + 0*128 + 63;
160                float *src0 = buf + 0*128 + 64;
161                float *src1 = buf + 1*128 + 63;
162
163                for(i = 0; i < 64; i++)
164                {
165                    temp0 = src0_prev[0];
166                    temp1 = src1_prev[0];
167                    wi = *win0_prev;
168                    wj = *win1_prev;
169                    temp2 = src0[0];
170                    temp3 = src1[0];
171                    dst0[0] = temp0 * wj - temp1 * wi;
172                    dst1[0] = temp0 * wi + temp1 * wj;
173
174                    wi = *win0;
175                    wj = *win1;
176
177                    temp0 = src0[128];
178                    temp1 = src1[128];
179                    dst0[128] = temp2 * wj - temp3 * wi;
180                    dst1[128] = temp2 * wi + temp3 * wj;
181
182                    temp2 = src0[256];
183                    temp3 = src1[256];
184                    dst0[256] = temp0 * wj - temp1 * wi;
185                    dst1[256] = temp0 * wi + temp1 * wj;
186                    dst0[384] = temp2 * wj - temp3 * wi;
187                    dst1[384] = temp2 * wi + temp3 * wj;
188
189                    temp0 = src0[384];
190                    temp1 = src1[384];
191                    dst0[512] = temp0 * wj - temp1 * wi;
192                    dst2[0] = temp0 * wi + temp1 * wj;
193
194                    src0++;
195                    src1--;
196                    src0_prev++;
197                    src1_prev--;
198                    win0++;
199                    win1--;
200                    win0_prev++;
201                    win1_prev--;
202                    dst0++;
203                    dst1--;
204                    dst2--;
205                }
206            }
207        } else {
208            ac->fdsp->vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, 64);
209            float_copy(out + 576, buf + 64, 448);
210        }
211    }
212
213    // buffer update
214    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
215        ac->fdsp->vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 64);
216        ac->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
217        ac->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
218        float_copy(saved + 448, buf + 7*128 + 64, 64);
219    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
220        float_copy(saved, buf + 512, 448);
221        float_copy(saved + 448, buf + 7*128 + 64, 64);
222    } else { // LONG_STOP or ONLY_LONG
223        float_copy(saved, buf + 512, 512);
224    }
225}
226
227static void apply_ltp_mips(AACContext *ac, SingleChannelElement *sce)
228{
229    const LongTermPrediction *ltp = &sce->ics.ltp;
230    const uint16_t *offsets = sce->ics.swb_offset;
231    int i, sfb;
232    int j, k;
233
234    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
235        float *predTime = sce->ret;
236        float *predFreq = ac->buf_mdct;
237        float *p_predTime;
238        int16_t num_samples = 2048;
239
240        if (ltp->lag < 1024)
241            num_samples = ltp->lag + 1024;
242        j = (2048 - num_samples) >> 2;
243        k = (2048 - num_samples) & 3;
244        p_predTime = &predTime[num_samples];
245
246        for (i = 0; i < num_samples; i++)
247            predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
248        for (i = 0; i < j; i++) {
249
250            /* loop unrolled 4 times */
251            __asm__ volatile (
252                "sw      $0,              0(%[p_predTime])        \n\t"
253                "sw      $0,              4(%[p_predTime])        \n\t"
254                "sw      $0,              8(%[p_predTime])        \n\t"
255                "sw      $0,              12(%[p_predTime])       \n\t"
256                PTR_ADDIU "%[p_predTime], %[p_predTime],     16   \n\t"
257
258                : [p_predTime]"+r"(p_predTime)
259                :
260                : "memory"
261            );
262        }
263        for (i = 0; i < k; i++) {
264
265            __asm__ volatile (
266                "sw      $0,              0(%[p_predTime])        \n\t"
267                PTR_ADDIU "%[p_predTime], %[p_predTime],     4    \n\t"
268
269                : [p_predTime]"+r"(p_predTime)
270                :
271                : "memory"
272            );
273        }
274
275        ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
276
277        if (sce->tns.present)
278            ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
279
280        for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
281            if (ltp->used[sfb])
282                for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
283                    sce->coeffs[i] += predFreq[i];
284    }
285}
286
287static av_always_inline void fmul_and_reverse(float *dst, const float *src0, const float *src1, int count)
288{
289    /* Multiply 'count' floats in src0 by src1 and store the results in dst in reverse */
290    /* This should be equivalent to a normal fmul, followed by reversing dst */
291
292    // count must be a multiple of 4
293    av_assert2(count % 4 == 0);
294
295    // move src0 and src1 to the last element of their arrays
296    src0 += count - 1;
297    src1 += count - 1;
298
299    for (; count > 0; count -= 4){
300        float temp[12];
301
302        /* loop unrolled 4 times */
303        __asm__ volatile (
304            "lwc1    %[temp0],    0(%[ptr2])                \n\t"
305            "lwc1    %[temp1],    -4(%[ptr2])               \n\t"
306            "lwc1    %[temp2],    -8(%[ptr2])               \n\t"
307            "lwc1    %[temp3],    -12(%[ptr2])              \n\t"
308            "lwc1    %[temp4],    0(%[ptr3])                \n\t"
309            "lwc1    %[temp5],    -4(%[ptr3])               \n\t"
310            "lwc1    %[temp6],    -8(%[ptr3])               \n\t"
311            "lwc1    %[temp7],    -12(%[ptr3])              \n\t"
312            "mul.s   %[temp8],    %[temp0],     %[temp4]    \n\t"
313            "mul.s   %[temp9],    %[temp1],     %[temp5]    \n\t"
314            "mul.s   %[temp10],   %[temp2],     %[temp6]    \n\t"
315            "mul.s   %[temp11],   %[temp3],     %[temp7]    \n\t"
316            "swc1    %[temp8],    0(%[ptr1])                \n\t"
317            "swc1    %[temp9],    4(%[ptr1])                \n\t"
318            "swc1    %[temp10],   8(%[ptr1])                \n\t"
319            "swc1    %[temp11],   12(%[ptr1])               \n\t"
320            PTR_ADDIU "%[ptr1],   %[ptr1],      16          \n\t"
321            PTR_ADDIU "%[ptr2],   %[ptr2],      -16         \n\t"
322            PTR_ADDIU "%[ptr3],   %[ptr3],      -16         \n\t"
323
324            : [temp0]"=&f"(temp[0]), [temp1]"=&f"(temp[1]),
325              [temp2]"=&f"(temp[2]), [temp3]"=&f"(temp[3]),
326              [temp4]"=&f"(temp[4]), [temp5]"=&f"(temp[5]),
327              [temp6]"=&f"(temp[6]), [temp7]"=&f"(temp[7]),
328              [temp8]"=&f"(temp[8]), [temp9]"=&f"(temp[9]),
329              [temp10]"=&f"(temp[10]), [temp11]"=&f"(temp[11]),
330              [ptr1]"+r"(dst), [ptr2]"+r"(src0), [ptr3]"+r"(src1)
331            :
332            : "memory"
333        );
334    }
335}
336
337static void update_ltp_mips(AACContext *ac, SingleChannelElement *sce)
338{
339    IndividualChannelStream *ics = &sce->ics;
340    float *saved     = sce->saved;
341    float *saved_ltp = sce->coeffs;
342    const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
343    const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
344    uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
345
346    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
347        float *p_saved_ltp = saved_ltp + 576;
348        float *loop_end1 = p_saved_ltp + 448;
349
350        float_copy(saved_ltp, saved, 512);
351
352        /* loop unrolled 8 times */
353        __asm__ volatile (
354        "1:                                                   \n\t"
355            "sw     $0,              0(%[p_saved_ltp])        \n\t"
356            "sw     $0,              4(%[p_saved_ltp])        \n\t"
357            "sw     $0,              8(%[p_saved_ltp])        \n\t"
358            "sw     $0,              12(%[p_saved_ltp])       \n\t"
359            "sw     $0,              16(%[p_saved_ltp])       \n\t"
360            "sw     $0,              20(%[p_saved_ltp])       \n\t"
361            "sw     $0,              24(%[p_saved_ltp])       \n\t"
362            "sw     $0,              28(%[p_saved_ltp])       \n\t"
363            PTR_ADDIU "%[p_saved_ltp],%[p_saved_ltp],    32   \n\t"
364            "bne    %[p_saved_ltp],  %[loop_end1],       1b   \n\t"
365
366            : [p_saved_ltp]"+r"(p_saved_ltp)
367            : [loop_end1]"r"(loop_end1)
368            : "memory"
369        );
370
371        ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     &swindow[64],      64);
372        fmul_and_reverse(saved_ltp + 512, ac->buf_mdct + 960, swindow, 64);
373    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
374        float *buff0 = saved;
375        float *buff1 = saved_ltp;
376        float *loop_end = saved + 448;
377
378        /* loop unrolled 8 times */
379        __asm__ volatile (
380            ".set push                                  \n\t"
381            ".set noreorder                             \n\t"
382        "1:                                             \n\t"
383            "lw      %[temp0],    0(%[src])             \n\t"
384            "lw      %[temp1],    4(%[src])             \n\t"
385            "lw      %[temp2],    8(%[src])             \n\t"
386            "lw      %[temp3],    12(%[src])            \n\t"
387            "lw      %[temp4],    16(%[src])            \n\t"
388            "lw      %[temp5],    20(%[src])            \n\t"
389            "lw      %[temp6],    24(%[src])            \n\t"
390            "lw      %[temp7],    28(%[src])            \n\t"
391            PTR_ADDIU "%[src],    %[src],         32    \n\t"
392            "sw      %[temp0],    0(%[dst])             \n\t"
393            "sw      %[temp1],    4(%[dst])             \n\t"
394            "sw      %[temp2],    8(%[dst])             \n\t"
395            "sw      %[temp3],    12(%[dst])            \n\t"
396            "sw      %[temp4],    16(%[dst])            \n\t"
397            "sw      %[temp5],    20(%[dst])            \n\t"
398            "sw      %[temp6],    24(%[dst])            \n\t"
399            "sw      %[temp7],    28(%[dst])            \n\t"
400            "sw      $0,          2304(%[dst])          \n\t"
401            "sw      $0,          2308(%[dst])          \n\t"
402            "sw      $0,          2312(%[dst])          \n\t"
403            "sw      $0,          2316(%[dst])          \n\t"
404            "sw      $0,          2320(%[dst])          \n\t"
405            "sw      $0,          2324(%[dst])          \n\t"
406            "sw      $0,          2328(%[dst])          \n\t"
407            "sw      $0,          2332(%[dst])          \n\t"
408            "bne     %[src],      %[loop_end],    1b    \n\t"
409            PTR_ADDIU "%[dst],    %[dst],         32    \n\t"
410            ".set pop                                   \n\t"
411
412            : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1),
413              [temp2]"=&r"(temp2), [temp3]"=&r"(temp3),
414              [temp4]"=&r"(temp4), [temp5]"=&r"(temp5),
415              [temp6]"=&r"(temp6), [temp7]"=&r"(temp7),
416              [src]"+r"(buff0), [dst]"+r"(buff1)
417            : [loop_end]"r"(loop_end)
418            : "memory"
419        );
420        ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     &swindow[64],      64);
421        fmul_and_reverse(saved_ltp + 512, ac->buf_mdct + 960, swindow, 64);
422    } else { // LONG_STOP or ONLY_LONG
423        ac->fdsp->vector_fmul_reverse(saved_ltp,       ac->buf_mdct + 512,     &lwindow[512],     512);
424        fmul_and_reverse(saved_ltp + 512, ac->buf_mdct + 512, lwindow, 512);
425    }
426
427    float_copy(sce->ltp_state, sce->ltp_state + 1024, 1024);
428    float_copy(sce->ltp_state + 1024, sce->ret, 1024);
429    float_copy(sce->ltp_state + 2048, saved_ltp, 1024);
430}
431#endif /* HAVE_MIPSFPU */
432#endif /* HAVE_INLINE_ASM */
433
434void ff_aacdec_init_mips(AACContext *c)
435{
436#if HAVE_INLINE_ASM
437#if HAVE_MIPSFPU
438    c->imdct_and_windowing         = imdct_and_windowing_mips;
439    c->apply_ltp                   = apply_ltp_mips;
440    c->update_ltp                  = update_ltp_mips;
441#endif /* HAVE_MIPSFPU */
442#endif /* HAVE_INLINE_ASM */
443}
444