falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
74789 Commits
32 Branches
310MB
Tree: 933309a6ca
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '933309a6ca'
${ noResults }
FFmpeg/libavcodec/mips/aaccoder_mips.c

2527 lines
104KB
Raw Blame History 

  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.  * Author:  Stanislav Ocovaj (socovaj@mips.com)

  30.  *          Szabolcs Pal     (sabolc@mips.com)

  31.  *

  32.  * AAC coefficients encoder optimized for MIPS floating-point architecture

  33.  *

  34.  * This file is part of FFmpeg.

  35.  *

  36.  * FFmpeg is free software; you can redistribute it and/or

  37.  * modify it under the terms of the GNU Lesser General Public

  38.  * License as published by the Free Software Foundation; either

  39.  * version 2.1 of the License, or (at your option) any later version.

  40.  *

  41.  * FFmpeg is distributed in the hope that it will be useful,

  42.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  43.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  44.  * Lesser General Public License for more details.

  45.  *

  46.  * You should have received a copy of the GNU Lesser General Public

  47.  * License along with FFmpeg; if not, write to the Free Software

  48.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  49.  */

  50. 

  51. /**

  52.  * @file

  53.  * Reference: libavcodec/aaccoder.c

  54.  */

  55. 

  56. #include "libavutil/libm.h"

  57. 

  58. #include <float.h>

  59. #include "libavutil/mathematics.h"

  60. #include "libavcodec/avcodec.h"

  61. #include "libavcodec/put_bits.h"

  62. #include "libavcodec/aac.h"

  63. #include "libavcodec/aacenc.h"

  64. #include "libavcodec/aactab.h"

  65. 

  66. #if HAVE_INLINE_ASM

  67. typedef struct BandCodingPath {

  68.     int prev_idx;

  69.     float cost;

  70.     int run;

  71. } BandCodingPath;

  72. 

  73. static const uint8_t run_value_bits_long[64] = {

  74.      5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,

  75.      5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5,  5, 10,

  76.     10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,

  77.     10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 15

  78. };

  79. 

  80. static const uint8_t run_value_bits_short[16] = {

  81.     3, 3, 3, 3, 3, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 9

  82. };

  83. 

  84. static const uint8_t * const run_value_bits[2] = {

  85.     run_value_bits_long, run_value_bits_short

  86. };

  87. 

  88. static const uint8_t uquad_sign_bits[81] = {

  89.     0, 1, 1, 1, 2, 2, 1, 2, 2,

  90.     1, 2, 2, 2, 3, 3, 2, 3, 3,

  91.     1, 2, 2, 2, 3, 3, 2, 3, 3,

  92.     1, 2, 2, 2, 3, 3, 2, 3, 3,

  93.     2, 3, 3, 3, 4, 4, 3, 4, 4,

  94.     2, 3, 3, 3, 4, 4, 3, 4, 4,

  95.     1, 2, 2, 2, 3, 3, 2, 3, 3,

  96.     2, 3, 3, 3, 4, 4, 3, 4, 4,

  97.     2, 3, 3, 3, 4, 4, 3, 4, 4

  98. };

  99. 

  100. static const uint8_t upair7_sign_bits[64] = {

  101.     0, 1, 1, 1, 1, 1, 1, 1,

  102.     1, 2, 2, 2, 2, 2, 2, 2,

  103.     1, 2, 2, 2, 2, 2, 2, 2,

  104.     1, 2, 2, 2, 2, 2, 2, 2,

  105.     1, 2, 2, 2, 2, 2, 2, 2,

  106.     1, 2, 2, 2, 2, 2, 2, 2,

  107.     1, 2, 2, 2, 2, 2, 2, 2,

  108.     1, 2, 2, 2, 2, 2, 2, 2,

  109. };

  110. 

  111. static const uint8_t upair12_sign_bits[169] = {

  112.     0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,

  113.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  114.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  115.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  116.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  117.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  118.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  119.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  120.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  121.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  122.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  123.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  124.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2

  125. };

  126. 

  127. static const uint8_t esc_sign_bits[289] = {

  128.     0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,

  129.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  130.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  131.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  132.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  133.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  134.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  135.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  136.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  137.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  138.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  139.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  140.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  141.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  142.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  143.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  144.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2

  145. };

  146. 

  147. #define ROUND_STANDARD 0.4054f

  148. #define ROUND_TO_ZERO 0.1054f

  149. 

  150. static void abs_pow34_v(float *out, const float *in, const int size) {

  151. #ifndef USE_REALLY_FULL_SEARCH

  152.     int i;

  153.     float a, b, c, d;

  154.     float ax, bx, cx, dx;

  155. 

  156.     for (i = 0; i < size; i += 4) {

  157.         a = fabsf(in[i  ]);

  158.         b = fabsf(in[i+1]);

  159.         c = fabsf(in[i+2]);

  160.         d = fabsf(in[i+3]);

  161. 

  162.         ax = sqrtf(a);

  163.         bx = sqrtf(b);

  164.         cx = sqrtf(c);

  165.         dx = sqrtf(d);

  166. 

  167.         a = a * ax;

  168.         b = b * bx;

  169.         c = c * cx;

  170.         d = d * dx;

  171. 

  172.         out[i  ] = sqrtf(a);

  173.         out[i+1] = sqrtf(b);

  174.         out[i+2] = sqrtf(c);

  175.         out[i+3] = sqrtf(d);

  176.     }

  177. #endif /* USE_REALLY_FULL_SEARCH */

  178. }

  179. 

  180. static float find_max_val(int group_len, int swb_size, const float *scaled) {

  181.     float maxval = 0.0f;

  182.     int w2, i;

  183.     for (w2 = 0; w2 < group_len; w2++) {

  184.         for (i = 0; i < swb_size; i++) {

  185.             maxval = FFMAX(maxval, scaled[w2*128+i]);

  186.         }

  187.     }

  188.     return maxval;

  189. }

  190. 

  191. static int find_min_book(float maxval, int sf) {

  192.     float Q = ff_aac_pow2sf_tab[POW_SF2_ZERO - sf + SCALE_ONE_POS - SCALE_DIV_512];

  193.     float Q34 = sqrtf(Q * sqrtf(Q));

  194.     int qmaxval, cb;

  195.     qmaxval = maxval * Q34 + 0.4054f;

  196.     if      (qmaxval ==  0) cb = 0;

  197.     else if (qmaxval ==  1) cb = 1;

  198.     else if (qmaxval ==  2) cb = 3;

  199.     else if (qmaxval <=  4) cb = 5;

  200.     else if (qmaxval <=  7) cb = 7;

  201.     else if (qmaxval <= 12) cb = 9;

  202.     else                    cb = 11;

  203.     return cb;

  204. }

  205. 

  206. /**

  207.  * Functions developed from template function and optimized for quantizing and encoding band

  208.  */

  209. static void quantize_and_encode_band_cost_SQUAD_mips(struct AACEncContext *s,

  210.                                                      PutBitContext *pb, const float *in, float *out,

  211.                                                      const float *scaled, int size, int scale_idx,

  212.                                                      int cb, const float lambda, const float uplim,

  213.                                                      int *bits, const float ROUNDING)

  214. {

  215.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  216.     int i;

  217.     int qc1, qc2, qc3, qc4;

  218. 

  219.     uint8_t  *p_bits  = (uint8_t  *)ff_aac_spectral_bits[cb-1];

  220.     uint16_t *p_codes = (uint16_t *)ff_aac_spectral_codes[cb-1];

  221. 

  222.     abs_pow34_v(s->scoefs, in, size);

  223.     scaled = s->scoefs;

  224.     for (i = 0; i < size; i += 4) {

  225.         int curidx;

  226.         int *in_int = (int *)&in[i];

  227.         int t0, t1, t2, t3, t4, t5, t6, t7;

  228. 

  229.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  230.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  231.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  232.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  233. 

  234.         __asm__ volatile (

  235.             ".set push                      \n\t"

  236.             ".set noreorder                 \n\t"

  237. 

  238.             "slt    %[qc1], $zero,  %[qc1]  \n\t"

  239.             "slt    %[qc2], $zero,  %[qc2]  \n\t"

  240.             "slt    %[qc3], $zero,  %[qc3]  \n\t"

  241.             "slt    %[qc4], $zero,  %[qc4]  \n\t"

  242.             "lw     %[t0],  0(%[in_int])    \n\t"

  243.             "lw     %[t1],  4(%[in_int])    \n\t"

  244.             "lw     %[t2],  8(%[in_int])    \n\t"

  245.             "lw     %[t3],  12(%[in_int])   \n\t"

  246.             "srl    %[t0],  %[t0],  31      \n\t"

  247.             "srl    %[t1],  %[t1],  31      \n\t"

  248.             "srl    %[t2],  %[t2],  31      \n\t"

  249.             "srl    %[t3],  %[t3],  31      \n\t"

  250.             "subu   %[t4],  $zero,  %[qc1]  \n\t"

  251.             "subu   %[t5],  $zero,  %[qc2]  \n\t"

  252.             "subu   %[t6],  $zero,  %[qc3]  \n\t"

  253.             "subu   %[t7],  $zero,  %[qc4]  \n\t"

  254.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  255.             "movn   %[qc2], %[t5],  %[t1]   \n\t"

  256.             "movn   %[qc3], %[t6],  %[t2]   \n\t"

  257.             "movn   %[qc4], %[t7],  %[t3]   \n\t"

  258. 

  259.             ".set pop                       \n\t"

  260. 

  261.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  262.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  263.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  264.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  265.             : [in_int]"r"(in_int)

  266.             : "memory"

  267.         );

  268. 

  269.         curidx = qc1;

  270.         curidx *= 3;

  271.         curidx += qc2;

  272.         curidx *= 3;

  273.         curidx += qc3;

  274.         curidx *= 3;

  275.         curidx += qc4;

  276.         curidx += 40;

  277. 

  278.         put_bits(pb, p_bits[curidx], p_codes[curidx]);

  279.     }

  280. }

  281. 

  282. static void quantize_and_encode_band_cost_UQUAD_mips(struct AACEncContext *s,

  283.                                                      PutBitContext *pb, const float *in, float *out,

  284.                                                      const float *scaled, int size, int scale_idx,

  285.                                                      int cb, const float lambda, const float uplim,

  286.                                                      int *bits, const float ROUNDING)

  287. {

  288.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  289.     int i;

  290.     int qc1, qc2, qc3, qc4;

  291. 

  292.     uint8_t  *p_bits  = (uint8_t  *)ff_aac_spectral_bits[cb-1];

  293.     uint16_t *p_codes = (uint16_t *)ff_aac_spectral_codes[cb-1];

  294. 

  295.     abs_pow34_v(s->scoefs, in, size);

  296.     scaled = s->scoefs;

  297.     for (i = 0; i < size; i += 4) {

  298.         int curidx, sign, count;

  299.         int *in_int = (int *)&in[i];

  300.         uint8_t v_bits;

  301.         unsigned int v_codes;

  302.         int t0, t1, t2, t3, t4;

  303. 

  304.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  305.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  306.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  307.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  308. 

  309.         __asm__ volatile (

  310.             ".set push                              \n\t"

  311.             ".set noreorder                         \n\t"

  312. 

  313.             "ori    %[t4],      $zero,      2       \n\t"

  314.             "ori    %[sign],    $zero,      0       \n\t"

  315.             "slt    %[t0],      %[t4],      %[qc1]  \n\t"

  316.             "slt    %[t1],      %[t4],      %[qc2]  \n\t"

  317.             "slt    %[t2],      %[t4],      %[qc3]  \n\t"

  318.             "slt    %[t3],      %[t4],      %[qc4]  \n\t"

  319.             "movn   %[qc1],     %[t4],      %[t0]   \n\t"

  320.             "movn   %[qc2],     %[t4],      %[t1]   \n\t"

  321.             "movn   %[qc3],     %[t4],      %[t2]   \n\t"

  322.             "movn   %[qc4],     %[t4],      %[t3]   \n\t"

  323.             "lw     %[t0],      0(%[in_int])        \n\t"

  324.             "lw     %[t1],      4(%[in_int])        \n\t"

  325.             "lw     %[t2],      8(%[in_int])        \n\t"

  326.             "lw     %[t3],      12(%[in_int])       \n\t"

  327.             "slt    %[t0],      %[t0],      $zero   \n\t"

  328.             "movn   %[sign],    %[t0],      %[qc1]  \n\t"

  329.             "slt    %[t1],      %[t1],      $zero   \n\t"

  330.             "slt    %[t2],      %[t2],      $zero   \n\t"

  331.             "slt    %[t3],      %[t3],      $zero   \n\t"

  332.             "sll    %[t0],      %[sign],    1       \n\t"

  333.             "or     %[t0],      %[t0],      %[t1]   \n\t"

  334.             "movn   %[sign],    %[t0],      %[qc2]  \n\t"

  335.             "slt    %[t4],      $zero,      %[qc1]  \n\t"

  336.             "slt    %[t1],      $zero,      %[qc2]  \n\t"

  337.             "slt    %[count],   $zero,      %[qc3]  \n\t"

  338.             "sll    %[t0],      %[sign],    1       \n\t"

  339.             "or     %[t0],      %[t0],      %[t2]   \n\t"

  340.             "movn   %[sign],    %[t0],      %[qc3]  \n\t"

  341.             "slt    %[t2],      $zero,      %[qc4]  \n\t"

  342.             "addu   %[count],   %[count],   %[t4]   \n\t"

  343.             "addu   %[count],   %[count],   %[t1]   \n\t"

  344.             "sll    %[t0],      %[sign],    1       \n\t"

  345.             "or     %[t0],      %[t0],      %[t3]   \n\t"

  346.             "movn   %[sign],    %[t0],      %[qc4]  \n\t"

  347.             "addu   %[count],   %[count],   %[t2]   \n\t"

  348. 

  349.             ".set pop                               \n\t"

  350. 

  351.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  352.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  353.               [sign]"=&r"(sign), [count]"=&r"(count),

  354.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  355.               [t4]"=&r"(t4)

  356.             : [in_int]"r"(in_int)

  357.             : "memory"

  358.         );

  359. 

  360.         curidx = qc1;

  361.         curidx *= 3;

  362.         curidx += qc2;

  363.         curidx *= 3;

  364.         curidx += qc3;

  365.         curidx *= 3;

  366.         curidx += qc4;

  367. 

  368.         v_codes = (p_codes[curidx] << count) | (sign & ((1 << count) - 1));

  369.         v_bits  = p_bits[curidx] + count;

  370.         put_bits(pb, v_bits, v_codes);

  371.     }

  372. }

  373. 

  374. static void quantize_and_encode_band_cost_SPAIR_mips(struct AACEncContext *s,

  375.                                                      PutBitContext *pb, const float *in, float *out,

  376.                                                      const float *scaled, int size, int scale_idx,

  377.                                                      int cb, const float lambda, const float uplim,

  378.                                                      int *bits, const float ROUNDING)

  379. {

  380.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  381.     int i;

  382.     int qc1, qc2, qc3, qc4;

  383. 

  384.     uint8_t  *p_bits  = (uint8_t  *)ff_aac_spectral_bits[cb-1];

  385.     uint16_t *p_codes = (uint16_t *)ff_aac_spectral_codes[cb-1];

  386. 

  387.     abs_pow34_v(s->scoefs, in, size);

  388.     scaled = s->scoefs;

  389.     for (i = 0; i < size; i += 4) {

  390.         int curidx, curidx2;

  391.         int *in_int = (int *)&in[i];

  392.         uint8_t v_bits;

  393.         unsigned int v_codes;

  394.         int t0, t1, t2, t3, t4, t5, t6, t7;

  395. 

  396.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  397.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  398.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  399.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  400. 

  401.         __asm__ volatile (

  402.             ".set push                      \n\t"

  403.             ".set noreorder                 \n\t"

  404. 

  405.             "ori    %[t4],  $zero,  4       \n\t"

  406.             "slt    %[t0],  %[t4],  %[qc1]  \n\t"

  407.             "slt    %[t1],  %[t4],  %[qc2]  \n\t"

  408.             "slt    %[t2],  %[t4],  %[qc3]  \n\t"

  409.             "slt    %[t3],  %[t4],  %[qc4]  \n\t"

  410.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  411.             "movn   %[qc2], %[t4],  %[t1]   \n\t"

  412.             "movn   %[qc3], %[t4],  %[t2]   \n\t"

  413.             "movn   %[qc4], %[t4],  %[t3]   \n\t"

  414.             "lw     %[t0],  0(%[in_int])    \n\t"

  415.             "lw     %[t1],  4(%[in_int])    \n\t"

  416.             "lw     %[t2],  8(%[in_int])    \n\t"

  417.             "lw     %[t3],  12(%[in_int])   \n\t"

  418.             "srl    %[t0],  %[t0],  31      \n\t"

  419.             "srl    %[t1],  %[t1],  31      \n\t"

  420.             "srl    %[t2],  %[t2],  31      \n\t"

  421.             "srl    %[t3],  %[t3],  31      \n\t"

  422.             "subu   %[t4],  $zero,  %[qc1]  \n\t"

  423.             "subu   %[t5],  $zero,  %[qc2]  \n\t"

  424.             "subu   %[t6],  $zero,  %[qc3]  \n\t"

  425.             "subu   %[t7],  $zero,  %[qc4]  \n\t"

  426.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  427.             "movn   %[qc2], %[t5],  %[t1]   \n\t"

  428.             "movn   %[qc3], %[t6],  %[t2]   \n\t"

  429.             "movn   %[qc4], %[t7],  %[t3]   \n\t"

  430. 

  431.             ".set pop                       \n\t"

  432. 

  433.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  434.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  435.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  436.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  437.             : [in_int]"r"(in_int)

  438.             : "memory"

  439.         );

  440. 

  441.         curidx = 9 * qc1;

  442.         curidx += qc2 + 40;

  443. 

  444.         curidx2 = 9 * qc3;

  445.         curidx2 += qc4 + 40;

  446. 

  447.         v_codes = (p_codes[curidx] << p_bits[curidx2]) | (p_codes[curidx2]);

  448.         v_bits  = p_bits[curidx] + p_bits[curidx2];

  449.         put_bits(pb, v_bits, v_codes);

  450.     }

  451. }

  452. 

  453. static void quantize_and_encode_band_cost_UPAIR7_mips(struct AACEncContext *s,

  454.                                                       PutBitContext *pb, const float *in, float *out,

  455.                                                       const float *scaled, int size, int scale_idx,

  456.                                                       int cb, const float lambda, const float uplim,

  457.                                                       int *bits, const float ROUNDING)

  458. {

  459.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  460.     int i;

  461.     int qc1, qc2, qc3, qc4;

  462. 

  463.     uint8_t  *p_bits  = (uint8_t*) ff_aac_spectral_bits[cb-1];

  464.     uint16_t *p_codes = (uint16_t*)ff_aac_spectral_codes[cb-1];

  465. 

  466.     abs_pow34_v(s->scoefs, in, size);

  467.     scaled = s->scoefs;

  468.     for (i = 0; i < size; i += 4) {

  469.         int curidx, sign1, count1, sign2, count2;

  470.         int *in_int = (int *)&in[i];

  471.         uint8_t v_bits;

  472.         unsigned int v_codes;

  473.         int t0, t1, t2, t3, t4;

  474. 

  475.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  476.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  477.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  478.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  479. 

  480.         __asm__ volatile (

  481.             ".set push                              \n\t"

  482.             ".set noreorder                         \n\t"

  483. 

  484.             "ori    %[t4],      $zero,      7       \n\t"

  485.             "ori    %[sign1],   $zero,      0       \n\t"

  486.             "ori    %[sign2],   $zero,      0       \n\t"

  487.             "slt    %[t0],      %[t4],      %[qc1]  \n\t"

  488.             "slt    %[t1],      %[t4],      %[qc2]  \n\t"

  489.             "slt    %[t2],      %[t4],      %[qc3]  \n\t"

  490.             "slt    %[t3],      %[t4],      %[qc4]  \n\t"

  491.             "movn   %[qc1],     %[t4],      %[t0]   \n\t"

  492.             "movn   %[qc2],     %[t4],      %[t1]   \n\t"

  493.             "movn   %[qc3],     %[t4],      %[t2]   \n\t"

  494.             "movn   %[qc4],     %[t4],      %[t3]   \n\t"

  495.             "lw     %[t0],      0(%[in_int])        \n\t"

  496.             "lw     %[t1],      4(%[in_int])        \n\t"

  497.             "lw     %[t2],      8(%[in_int])        \n\t"

  498.             "lw     %[t3],      12(%[in_int])       \n\t"

  499.             "slt    %[t0],      %[t0],      $zero   \n\t"

  500.             "movn   %[sign1],   %[t0],      %[qc1]  \n\t"

  501.             "slt    %[t2],      %[t2],      $zero   \n\t"

  502.             "movn   %[sign2],   %[t2],      %[qc3]  \n\t"

  503.             "slt    %[t1],      %[t1],      $zero   \n\t"

  504.             "sll    %[t0],      %[sign1],   1       \n\t"

  505.             "or     %[t0],      %[t0],      %[t1]   \n\t"

  506.             "movn   %[sign1],   %[t0],      %[qc2]  \n\t"

  507.             "slt    %[t3],      %[t3],      $zero   \n\t"

  508.             "sll    %[t0],      %[sign2],   1       \n\t"

  509.             "or     %[t0],      %[t0],      %[t3]   \n\t"

  510.             "movn   %[sign2],   %[t0],      %[qc4]  \n\t"

  511.             "slt    %[count1],  $zero,      %[qc1]  \n\t"

  512.             "slt    %[t1],      $zero,      %[qc2]  \n\t"

  513.             "slt    %[count2],  $zero,      %[qc3]  \n\t"

  514.             "slt    %[t2],      $zero,      %[qc4]  \n\t"

  515.             "addu   %[count1],  %[count1],  %[t1]   \n\t"

  516.             "addu   %[count2],  %[count2],  %[t2]   \n\t"

  517. 

  518.             ".set pop                               \n\t"

  519. 

  520.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  521.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  522.               [sign1]"=&r"(sign1), [count1]"=&r"(count1),

  523.               [sign2]"=&r"(sign2), [count2]"=&r"(count2),

  524.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  525.               [t4]"=&r"(t4)

  526.             : [in_int]"r"(in_int)

  527.             : "t0", "t1", "t2", "t3", "t4",

  528.               "memory"

  529.         );

  530. 

  531.         curidx  = 8 * qc1;

  532.         curidx += qc2;

  533. 

  534.         v_codes = (p_codes[curidx] << count1) | sign1;

  535.         v_bits  = p_bits[curidx] + count1;

  536.         put_bits(pb, v_bits, v_codes);

  537. 

  538.         curidx  = 8 * qc3;

  539.         curidx += qc4;

  540. 

  541.         v_codes = (p_codes[curidx] << count2) | sign2;

  542.         v_bits  = p_bits[curidx] + count2;

  543.         put_bits(pb, v_bits, v_codes);

  544.     }

  545. }

  546. 

  547. static void quantize_and_encode_band_cost_UPAIR12_mips(struct AACEncContext *s,

  548.                                                        PutBitContext *pb, const float *in, float *out,

  549.                                                        const float *scaled, int size, int scale_idx,

  550.                                                        int cb, const float lambda, const float uplim,

  551.                                                        int *bits, const float ROUNDING)

  552. {

  553.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  554.     int i;

  555.     int qc1, qc2, qc3, qc4;

  556. 

  557.     uint8_t  *p_bits  = (uint8_t*) ff_aac_spectral_bits[cb-1];

  558.     uint16_t *p_codes = (uint16_t*)ff_aac_spectral_codes[cb-1];

  559. 

  560.     abs_pow34_v(s->scoefs, in, size);

  561.     scaled = s->scoefs;

  562.     for (i = 0; i < size; i += 4) {

  563.         int curidx, sign1, count1, sign2, count2;

  564.         int *in_int = (int *)&in[i];

  565.         uint8_t v_bits;

  566.         unsigned int v_codes;

  567.         int t0, t1, t2, t3, t4;

  568. 

  569.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  570.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  571.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  572.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  573. 

  574.         __asm__ volatile (

  575.             ".set push                              \n\t"

  576.             ".set noreorder                         \n\t"

  577. 

  578.             "ori    %[t4],      $zero,      12      \n\t"

  579.             "ori    %[sign1],   $zero,      0       \n\t"

  580.             "ori    %[sign2],   $zero,      0       \n\t"

  581.             "slt    %[t0],      %[t4],      %[qc1]  \n\t"

  582.             "slt    %[t1],      %[t4],      %[qc2]  \n\t"

  583.             "slt    %[t2],      %[t4],      %[qc3]  \n\t"

  584.             "slt    %[t3],      %[t4],      %[qc4]  \n\t"

  585.             "movn   %[qc1],     %[t4],      %[t0]   \n\t"

  586.             "movn   %[qc2],     %[t4],      %[t1]   \n\t"

  587.             "movn   %[qc3],     %[t4],      %[t2]   \n\t"

  588.             "movn   %[qc4],     %[t4],      %[t3]   \n\t"

  589.             "lw     %[t0],      0(%[in_int])        \n\t"

  590.             "lw     %[t1],      4(%[in_int])        \n\t"

  591.             "lw     %[t2],      8(%[in_int])        \n\t"

  592.             "lw     %[t3],      12(%[in_int])       \n\t"

  593.             "slt    %[t0],      %[t0],      $zero   \n\t"

  594.             "movn   %[sign1],   %[t0],      %[qc1]  \n\t"

  595.             "slt    %[t2],      %[t2],      $zero   \n\t"

  596.             "movn   %[sign2],   %[t2],      %[qc3]  \n\t"

  597.             "slt    %[t1],      %[t1],      $zero   \n\t"

  598.             "sll    %[t0],      %[sign1],   1       \n\t"

  599.             "or     %[t0],      %[t0],      %[t1]   \n\t"

  600.             "movn   %[sign1],   %[t0],      %[qc2]  \n\t"

  601.             "slt    %[t3],      %[t3],      $zero   \n\t"

  602.             "sll    %[t0],      %[sign2],   1       \n\t"

  603.             "or     %[t0],      %[t0],      %[t3]   \n\t"

  604.             "movn   %[sign2],   %[t0],      %[qc4]  \n\t"

  605.             "slt    %[count1],  $zero,      %[qc1]  \n\t"

  606.             "slt    %[t1],      $zero,      %[qc2]  \n\t"

  607.             "slt    %[count2],  $zero,      %[qc3]  \n\t"

  608.             "slt    %[t2],      $zero,      %[qc4]  \n\t"

  609.             "addu   %[count1],  %[count1],  %[t1]   \n\t"

  610.             "addu   %[count2],  %[count2],  %[t2]   \n\t"

  611. 

  612.             ".set pop                               \n\t"

  613. 

  614.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  615.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  616.               [sign1]"=&r"(sign1), [count1]"=&r"(count1),

  617.               [sign2]"=&r"(sign2), [count2]"=&r"(count2),

  618.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  619.               [t4]"=&r"(t4)

  620.             : [in_int]"r"(in_int)

  621.             : "memory"

  622.         );

  623. 

  624.         curidx  = 13 * qc1;

  625.         curidx += qc2;

  626. 

  627.         v_codes = (p_codes[curidx] << count1) | sign1;

  628.         v_bits  = p_bits[curidx] + count1;

  629.         put_bits(pb, v_bits, v_codes);

  630. 

  631.         curidx  = 13 * qc3;

  632.         curidx += qc4;

  633. 

  634.         v_codes = (p_codes[curidx] << count2) | sign2;

  635.         v_bits  = p_bits[curidx] + count2;

  636.         put_bits(pb, v_bits, v_codes);

  637.     }

  638. }

  639. 

  640. static void quantize_and_encode_band_cost_ESC_mips(struct AACEncContext *s,

  641.                                                    PutBitContext *pb, const float *in, float *out,

  642.                                                    const float *scaled, int size, int scale_idx,

  643.                                                    int cb, const float lambda, const float uplim,

  644.                                                    int *bits, const float ROUNDING)

  645. {

  646.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  647.     int i;

  648.     int qc1, qc2, qc3, qc4;

  649. 

  650.     uint8_t  *p_bits    = (uint8_t* )ff_aac_spectral_bits[cb-1];

  651.     uint16_t *p_codes   = (uint16_t*)ff_aac_spectral_codes[cb-1];

  652.     float    *p_vectors = (float*   )ff_aac_codebook_vectors[cb-1];

  653. 

  654.     abs_pow34_v(s->scoefs, in, size);

  655.     scaled = s->scoefs;

  656. 

  657.     if (cb < 11) {

  658.         for (i = 0; i < size; i += 4) {

  659.             int curidx, curidx2, sign1, count1, sign2, count2;

  660.             int *in_int = (int *)&in[i];

  661.             uint8_t v_bits;

  662.             unsigned int v_codes;

  663.             int t0, t1, t2, t3, t4;

  664. 

  665.             qc1 = scaled[i  ] * Q34 + ROUNDING;

  666.             qc2 = scaled[i+1] * Q34 + ROUNDING;

  667.             qc3 = scaled[i+2] * Q34 + ROUNDING;

  668.             qc4 = scaled[i+3] * Q34 + ROUNDING;

  669. 

  670.             __asm__ volatile (

  671.                 ".set push                                  \n\t"

  672.                 ".set noreorder                             \n\t"

  673. 

  674.                 "ori        %[t4],      $zero,      16      \n\t"

  675.                 "ori        %[sign1],   $zero,      0       \n\t"

  676.                 "ori        %[sign2],   $zero,      0       \n\t"

  677.                 "slt        %[t0],      %[t4],      %[qc1]  \n\t"

  678.                 "slt        %[t1],      %[t4],      %[qc2]  \n\t"

  679.                 "slt        %[t2],      %[t4],      %[qc3]  \n\t"

  680.                 "slt        %[t3],      %[t4],      %[qc4]  \n\t"

  681.                 "movn       %[qc1],     %[t4],      %[t0]   \n\t"

  682.                 "movn       %[qc2],     %[t4],      %[t1]   \n\t"

  683.                 "movn       %[qc3],     %[t4],      %[t2]   \n\t"

  684.                 "movn       %[qc4],     %[t4],      %[t3]   \n\t"

  685.                 "lw         %[t0],      0(%[in_int])        \n\t"

  686.                 "lw         %[t1],      4(%[in_int])        \n\t"

  687.                 "lw         %[t2],      8(%[in_int])        \n\t"

  688.                 "lw         %[t3],      12(%[in_int])       \n\t"

  689.                 "slt        %[t0],      %[t0],      $zero   \n\t"

  690.                 "movn       %[sign1],   %[t0],      %[qc1]  \n\t"

  691.                 "slt        %[t2],      %[t2],      $zero   \n\t"

  692.                 "movn       %[sign2],   %[t2],      %[qc3]  \n\t"

  693.                 "slt        %[t1],      %[t1],      $zero   \n\t"

  694.                 "sll        %[t0],      %[sign1],   1       \n\t"

  695.                 "or         %[t0],      %[t0],      %[t1]   \n\t"

  696.                 "movn       %[sign1],   %[t0],      %[qc2]  \n\t"

  697.                 "slt        %[t3],      %[t3],      $zero   \n\t"

  698.                 "sll        %[t0],      %[sign2],   1       \n\t"

  699.                 "or         %[t0],      %[t0],      %[t3]   \n\t"

  700.                 "movn       %[sign2],   %[t0],      %[qc4]  \n\t"

  701.                 "slt        %[count1],  $zero,      %[qc1]  \n\t"

  702.                 "slt        %[t1],      $zero,      %[qc2]  \n\t"

  703.                 "slt        %[count2],  $zero,      %[qc3]  \n\t"

  704.                 "slt        %[t2],      $zero,      %[qc4]  \n\t"

  705.                 "addu       %[count1],  %[count1],  %[t1]   \n\t"

  706.                 "addu       %[count2],  %[count2],  %[t2]   \n\t"

  707. 

  708.                 ".set pop                                   \n\t"

  709. 

  710.                 : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  711.                   [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  712.                   [sign1]"=&r"(sign1), [count1]"=&r"(count1),

  713.                   [sign2]"=&r"(sign2), [count2]"=&r"(count2),

  714.                   [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  715.                   [t4]"=&r"(t4)

  716.                 : [in_int]"r"(in_int)

  717.                 : "memory"

  718.             );

  719. 

  720.             curidx = 17 * qc1;

  721.             curidx += qc2;

  722.             curidx2 = 17 * qc3;

  723.             curidx2 += qc4;

  724. 

  725.             v_codes = (p_codes[curidx] << count1) | sign1;

  726.             v_bits  = p_bits[curidx] + count1;

  727.             put_bits(pb, v_bits, v_codes);

  728. 

  729.             v_codes = (p_codes[curidx2] << count2) | sign2;

  730.             v_bits  = p_bits[curidx2] + count2;

  731.             put_bits(pb, v_bits, v_codes);

  732.         }

  733.     } else {

  734.         for (i = 0; i < size; i += 4) {

  735.             int curidx, curidx2, sign1, count1, sign2, count2;

  736.             int *in_int = (int *)&in[i];

  737.             uint8_t v_bits;

  738.             unsigned int v_codes;

  739.             int c1, c2, c3, c4;

  740.             int t0, t1, t2, t3, t4;

  741. 

  742.             qc1 = scaled[i  ] * Q34 + ROUNDING;

  743.             qc2 = scaled[i+1] * Q34 + ROUNDING;

  744.             qc3 = scaled[i+2] * Q34 + ROUNDING;

  745.             qc4 = scaled[i+3] * Q34 + ROUNDING;

  746. 

  747.             __asm__ volatile (

  748.                 ".set push                                  \n\t"

  749.                 ".set noreorder                             \n\t"

  750. 

  751.                 "ori        %[t4],      $zero,      16      \n\t"

  752.                 "ori        %[sign1],   $zero,      0       \n\t"

  753.                 "ori        %[sign2],   $zero,      0       \n\t"

  754.                 "shll_s.w   %[c1],      %[qc1],     18      \n\t"

  755.                 "shll_s.w   %[c2],      %[qc2],     18      \n\t"

  756.                 "shll_s.w   %[c3],      %[qc3],     18      \n\t"

  757.                 "shll_s.w   %[c4],      %[qc4],     18      \n\t"

  758.                 "srl        %[c1],      %[c1],      18      \n\t"

  759.                 "srl        %[c2],      %[c2],      18      \n\t"

  760.                 "srl        %[c3],      %[c3],      18      \n\t"

  761.                 "srl        %[c4],      %[c4],      18      \n\t"

  762.                 "slt        %[t0],      %[t4],      %[qc1]  \n\t"

  763.                 "slt        %[t1],      %[t4],      %[qc2]  \n\t"

  764.                 "slt        %[t2],      %[t4],      %[qc3]  \n\t"

  765.                 "slt        %[t3],      %[t4],      %[qc4]  \n\t"

  766.                 "movn       %[qc1],     %[t4],      %[t0]   \n\t"

  767.                 "movn       %[qc2],     %[t4],      %[t1]   \n\t"

  768.                 "movn       %[qc3],     %[t4],      %[t2]   \n\t"

  769.                 "movn       %[qc4],     %[t4],      %[t3]   \n\t"

  770.                 "lw         %[t0],      0(%[in_int])        \n\t"

  771.                 "lw         %[t1],      4(%[in_int])        \n\t"

  772.                 "lw         %[t2],      8(%[in_int])        \n\t"

  773.                 "lw         %[t3],      12(%[in_int])       \n\t"

  774.                 "slt        %[t0],      %[t0],      $zero   \n\t"

  775.                 "movn       %[sign1],   %[t0],      %[qc1]  \n\t"

  776.                 "slt        %[t2],      %[t2],      $zero   \n\t"

  777.                 "movn       %[sign2],   %[t2],      %[qc3]  \n\t"

  778.                 "slt        %[t1],      %[t1],      $zero   \n\t"

  779.                 "sll        %[t0],      %[sign1],   1       \n\t"

  780.                 "or         %[t0],      %[t0],      %[t1]   \n\t"

  781.                 "movn       %[sign1],   %[t0],      %[qc2]  \n\t"

  782.                 "slt        %[t3],      %[t3],      $zero   \n\t"

  783.                 "sll        %[t0],      %[sign2],   1       \n\t"

  784.                 "or         %[t0],      %[t0],      %[t3]   \n\t"

  785.                 "movn       %[sign2],   %[t0],      %[qc4]  \n\t"

  786.                 "slt        %[count1],  $zero,      %[qc1]  \n\t"

  787.                 "slt        %[t1],      $zero,      %[qc2]  \n\t"

  788.                 "slt        %[count2],  $zero,      %[qc3]  \n\t"

  789.                 "slt        %[t2],      $zero,      %[qc4]  \n\t"

  790.                 "addu       %[count1],  %[count1],  %[t1]   \n\t"

  791.                 "addu       %[count2],  %[count2],  %[t2]   \n\t"

  792. 

  793.                 ".set pop                                   \n\t"

  794. 

  795.                 : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  796.                   [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  797.                   [sign1]"=&r"(sign1), [count1]"=&r"(count1),

  798.                   [sign2]"=&r"(sign2), [count2]"=&r"(count2),

  799.                   [c1]"=&r"(c1), [c2]"=&r"(c2),

  800.                   [c3]"=&r"(c3), [c4]"=&r"(c4),

  801.                   [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  802.                   [t4]"=&r"(t4)

  803.                 : [in_int]"r"(in_int)

  804.                 : "memory"

  805.             );

  806. 

  807.             curidx = 17 * qc1;

  808.             curidx += qc2;

  809. 

  810.             curidx2 = 17 * qc3;

  811.             curidx2 += qc4;

  812. 

  813.             v_codes = (p_codes[curidx] << count1) | sign1;

  814.             v_bits  = p_bits[curidx] + count1;

  815.             put_bits(pb, v_bits, v_codes);

  816. 

  817.             if (p_vectors[curidx*2  ] == 64.0f) {

  818.                 int len = av_log2(c1);

  819.                 v_codes = (((1 << (len - 3)) - 2) << len) | (c1 & ((1 << len) - 1));

  820.                 put_bits(pb, len * 2 - 3, v_codes);

  821.             }

  822.             if (p_vectors[curidx*2+1] == 64.0f) {

  823.                 int len = av_log2(c2);

  824.                 v_codes = (((1 << (len - 3)) - 2) << len) | (c2 & ((1 << len) - 1));

  825.                 put_bits(pb, len*2-3, v_codes);

  826.             }

  827. 

  828.             v_codes = (p_codes[curidx2] << count2) | sign2;

  829.             v_bits  = p_bits[curidx2] + count2;

  830.             put_bits(pb, v_bits, v_codes);

  831. 

  832.             if (p_vectors[curidx2*2  ] == 64.0f) {

  833.                 int len = av_log2(c3);

  834.                 v_codes = (((1 << (len - 3)) - 2) << len) | (c3 & ((1 << len) - 1));

  835.                 put_bits(pb, len* 2 - 3, v_codes);

  836.             }

  837.             if (p_vectors[curidx2*2+1] == 64.0f) {

  838.                 int len = av_log2(c4);

  839.                 v_codes = (((1 << (len - 3)) - 2) << len) | (c4 & ((1 << len) - 1));

  840.                 put_bits(pb, len * 2 - 3, v_codes);

  841.             }

  842.         }

  843.     }

  844. }

  845. 

  846. static void (*const quantize_and_encode_band_cost_arr[])(struct AACEncContext *s,

  847.                                                          PutBitContext *pb, const float *in, float *out,

  848.                                                          const float *scaled, int size, int scale_idx,

  849.                                                          int cb, const float lambda, const float uplim,

  850.                                                          int *bits, const float ROUNDING) = {

  851.     NULL,

  852.     quantize_and_encode_band_cost_SQUAD_mips,

  853.     quantize_and_encode_band_cost_SQUAD_mips,

  854.     quantize_and_encode_band_cost_UQUAD_mips,

  855.     quantize_and_encode_band_cost_UQUAD_mips,

  856.     quantize_and_encode_band_cost_SPAIR_mips,

  857.     quantize_and_encode_band_cost_SPAIR_mips,

  858.     quantize_and_encode_band_cost_UPAIR7_mips,

  859.     quantize_and_encode_band_cost_UPAIR7_mips,

  860.     quantize_and_encode_band_cost_UPAIR12_mips,

  861.     quantize_and_encode_band_cost_UPAIR12_mips,

  862.     quantize_and_encode_band_cost_ESC_mips,

  863. };

  864. 

  865. #define quantize_and_encode_band_cost(                                       \

  866.                                 s, pb, in, out, scaled, size, scale_idx, cb, \

  867.                                 lambda, uplim, bits, ROUNDING)               \

  868.     quantize_and_encode_band_cost_arr[cb](                                   \

  869.                                 s, pb, in, out, scaled, size, scale_idx, cb, \

  870.                                 lambda, uplim, bits, ROUNDING)

  871. 

  872. static void quantize_and_encode_band_mips(struct AACEncContext *s, PutBitContext *pb,

  873.                                           const float *in, float *out, int size, int scale_idx,

  874.                                           int cb, const float lambda, int rtz)

  875. {

  876.     quantize_and_encode_band_cost(s, pb, in, out, NULL, size, scale_idx, cb, lambda,

  877.                                   INFINITY, NULL, (rtz) ? ROUND_TO_ZERO : ROUND_STANDARD);

  878. }

  879. 

  880. /**

  881.  * Functions developed from template function and optimized for getting the number of bits

  882.  */

  883. static float get_band_numbits_ZERO_mips(struct AACEncContext *s,

  884.                                         PutBitContext *pb, const float *in,

  885.                                         const float *scaled, int size, int scale_idx,

  886.                                         int cb, const float lambda, const float uplim,

  887.                                         int *bits)

  888. {

  889.     return 0;

  890. }

  891. 

  892. static float get_band_numbits_SQUAD_mips(struct AACEncContext *s,

  893.                                          PutBitContext *pb, const float *in,

  894.                                          const float *scaled, int size, int scale_idx,

  895.                                          int cb, const float lambda, const float uplim,

  896.                                          int *bits)

  897. {

  898.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  899.     int i;

  900.     int qc1, qc2, qc3, qc4;

  901.     int curbits = 0;

  902. 

  903.     uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];

  904. 

  905.     for (i = 0; i < size; i += 4) {

  906.         int curidx;

  907.         int *in_int = (int *)&in[i];

  908.         int t0, t1, t2, t3, t4, t5, t6, t7;

  909. 

  910.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  911.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  912.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  913.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  914. 

  915.         __asm__ volatile (

  916.             ".set push                      \n\t"

  917.             ".set noreorder                 \n\t"

  918. 

  919.             "slt    %[qc1], $zero,  %[qc1]  \n\t"

  920.             "slt    %[qc2], $zero,  %[qc2]  \n\t"

  921.             "slt    %[qc3], $zero,  %[qc3]  \n\t"

  922.             "slt    %[qc4], $zero,  %[qc4]  \n\t"

  923.             "lw     %[t0],  0(%[in_int])    \n\t"

  924.             "lw     %[t1],  4(%[in_int])    \n\t"

  925.             "lw     %[t2],  8(%[in_int])    \n\t"

  926.             "lw     %[t3],  12(%[in_int])   \n\t"

  927.             "srl    %[t0],  %[t0],  31      \n\t"

  928.             "srl    %[t1],  %[t1],  31      \n\t"

  929.             "srl    %[t2],  %[t2],  31      \n\t"

  930.             "srl    %[t3],  %[t3],  31      \n\t"

  931.             "subu   %[t4],  $zero,  %[qc1]  \n\t"

  932.             "subu   %[t5],  $zero,  %[qc2]  \n\t"

  933.             "subu   %[t6],  $zero,  %[qc3]  \n\t"

  934.             "subu   %[t7],  $zero,  %[qc4]  \n\t"

  935.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  936.             "movn   %[qc2], %[t5],  %[t1]   \n\t"

  937.             "movn   %[qc3], %[t6],  %[t2]   \n\t"

  938.             "movn   %[qc4], %[t7],  %[t3]   \n\t"

  939. 

  940.             ".set pop                       \n\t"

  941. 

  942.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  943.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  944.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  945.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  946.             : [in_int]"r"(in_int)

  947.             : "memory"

  948.         );

  949. 

  950.         curidx = qc1;

  951.         curidx *= 3;

  952.         curidx += qc2;

  953.         curidx *= 3;

  954.         curidx += qc3;

  955.         curidx *= 3;

  956.         curidx += qc4;

  957.         curidx += 40;

  958. 

  959.         curbits += p_bits[curidx];

  960.     }

  961.     return curbits;

  962. }

  963. 

  964. static float get_band_numbits_UQUAD_mips(struct AACEncContext *s,

  965.                                          PutBitContext *pb, const float *in,

  966.                                          const float *scaled, int size, int scale_idx,

  967.                                          int cb, const float lambda, const float uplim,

  968.                                          int *bits)

  969. {

  970.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  971.     int i;

  972.     int curbits = 0;

  973.     int qc1, qc2, qc3, qc4;

  974. 

  975.     uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];

  976. 

  977.     for (i = 0; i < size; i += 4) {

  978.         int curidx;

  979.         int t0, t1, t2, t3, t4;

  980. 

  981.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  982.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  983.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  984.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  985. 

  986.         __asm__ volatile (

  987.             ".set push                      \n\t"

  988.             ".set noreorder                 \n\t"

  989. 

  990.             "ori    %[t4],  $zero,  2       \n\t"

  991.             "slt    %[t0],  %[t4],  %[qc1]  \n\t"

  992.             "slt    %[t1],  %[t4],  %[qc2]  \n\t"

  993.             "slt    %[t2],  %[t4],  %[qc3]  \n\t"

  994.             "slt    %[t3],  %[t4],  %[qc4]  \n\t"

  995.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  996.             "movn   %[qc2], %[t4],  %[t1]   \n\t"

  997.             "movn   %[qc3], %[t4],  %[t2]   \n\t"

  998.             "movn   %[qc4], %[t4],  %[t3]   \n\t"

  999. 

  1000.             ".set pop                       \n\t"

  1001. 

  1002.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1003.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1004.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1005.               [t4]"=&r"(t4)

  1006.         );

  1007. 

  1008.         curidx = qc1;

  1009.         curidx *= 3;

  1010.         curidx += qc2;

  1011.         curidx *= 3;

  1012.         curidx += qc3;

  1013.         curidx *= 3;

  1014.         curidx += qc4;

  1015. 

  1016.         curbits += p_bits[curidx];

  1017.         curbits += uquad_sign_bits[curidx];

  1018.     }

  1019.     return curbits;

  1020. }

  1021. 

  1022. static float get_band_numbits_SPAIR_mips(struct AACEncContext *s,

  1023.                                          PutBitContext *pb, const float *in,

  1024.                                          const float *scaled, int size, int scale_idx,

  1025.                                          int cb, const float lambda, const float uplim,

  1026.                                          int *bits)

  1027. {

  1028.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1029.     int i;

  1030.     int qc1, qc2, qc3, qc4;

  1031.     int curbits = 0;

  1032. 

  1033.     uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1];

  1034. 

  1035.     for (i = 0; i < size; i += 4) {

  1036.         int curidx, curidx2;

  1037.         int *in_int = (int *)&in[i];

  1038.         int t0, t1, t2, t3, t4, t5, t6, t7;

  1039. 

  1040.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1041.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1042.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1043.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1044. 

  1045.         __asm__ volatile (

  1046.             ".set push                      \n\t"

  1047.             ".set noreorder                 \n\t"

  1048. 

  1049.             "ori    %[t4],  $zero,  4       \n\t"

  1050.             "slt    %[t0],  %[t4],  %[qc1]  \n\t"

  1051.             "slt    %[t1],  %[t4],  %[qc2]  \n\t"

  1052.             "slt    %[t2],  %[t4],  %[qc3]  \n\t"

  1053.             "slt    %[t3],  %[t4],  %[qc4]  \n\t"

  1054.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1055.             "movn   %[qc2], %[t4],  %[t1]   \n\t"

  1056.             "movn   %[qc3], %[t4],  %[t2]   \n\t"

  1057.             "movn   %[qc4], %[t4],  %[t3]   \n\t"

  1058.             "lw     %[t0],  0(%[in_int])    \n\t"

  1059.             "lw     %[t1],  4(%[in_int])    \n\t"

  1060.             "lw     %[t2],  8(%[in_int])    \n\t"

  1061.             "lw     %[t3],  12(%[in_int])   \n\t"

  1062.             "srl    %[t0],  %[t0],  31      \n\t"

  1063.             "srl    %[t1],  %[t1],  31      \n\t"

  1064.             "srl    %[t2],  %[t2],  31      \n\t"

  1065.             "srl    %[t3],  %[t3],  31      \n\t"

  1066.             "subu   %[t4],  $zero,  %[qc1]  \n\t"

  1067.             "subu   %[t5],  $zero,  %[qc2]  \n\t"

  1068.             "subu   %[t6],  $zero,  %[qc3]  \n\t"

  1069.             "subu   %[t7],  $zero,  %[qc4]  \n\t"

  1070.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1071.             "movn   %[qc2], %[t5],  %[t1]   \n\t"

  1072.             "movn   %[qc3], %[t6],  %[t2]   \n\t"

  1073.             "movn   %[qc4], %[t7],  %[t3]   \n\t"

  1074. 

  1075.             ".set pop                       \n\t"

  1076. 

  1077.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1078.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1079.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1080.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  1081.             : [in_int]"r"(in_int)

  1082.             : "memory"

  1083.         );

  1084. 

  1085.         curidx  = 9 * qc1;

  1086.         curidx += qc2 + 40;

  1087. 

  1088.         curidx2  = 9 * qc3;

  1089.         curidx2 += qc4 + 40;

  1090. 

  1091.         curbits += p_bits[curidx] + p_bits[curidx2];

  1092.     }

  1093.     return curbits;

  1094. }

  1095. 

  1096. static float get_band_numbits_UPAIR7_mips(struct AACEncContext *s,

  1097.                                           PutBitContext *pb, const float *in,

  1098.                                           const float *scaled, int size, int scale_idx,

  1099.                                           int cb, const float lambda, const float uplim,

  1100.                                           int *bits)

  1101. {

  1102.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1103.     int i;

  1104.     int qc1, qc2, qc3, qc4;

  1105.     int curbits = 0;

  1106. 

  1107.     uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1108. 

  1109.     for (i = 0; i < size; i += 4) {

  1110.         int curidx, curidx2;

  1111.         int t0, t1, t2, t3, t4;

  1112. 

  1113.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1114.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1115.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1116.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1117. 

  1118.         __asm__ volatile (

  1119.             ".set push                      \n\t"

  1120.             ".set noreorder                 \n\t"

  1121. 

  1122.             "ori    %[t4],  $zero,  7       \n\t"

  1123.             "slt    %[t0],  %[t4],  %[qc1]  \n\t"

  1124.             "slt    %[t1],  %[t4],  %[qc2]  \n\t"

  1125.             "slt    %[t2],  %[t4],  %[qc3]  \n\t"

  1126.             "slt    %[t3],  %[t4],  %[qc4]  \n\t"

  1127.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1128.             "movn   %[qc2], %[t4],  %[t1]   \n\t"

  1129.             "movn   %[qc3], %[t4],  %[t2]   \n\t"

  1130.             "movn   %[qc4], %[t4],  %[t3]   \n\t"

  1131. 

  1132.             ".set pop                       \n\t"

  1133. 

  1134.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1135.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1136.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1137.               [t4]"=&r"(t4)

  1138.         );

  1139. 

  1140.         curidx  = 8 * qc1;

  1141.         curidx += qc2;

  1142. 

  1143.         curidx2  = 8 * qc3;

  1144.         curidx2 += qc4;

  1145. 

  1146.         curbits += p_bits[curidx] +

  1147.                    upair7_sign_bits[curidx] +

  1148.                    p_bits[curidx2] +

  1149.                    upair7_sign_bits[curidx2];

  1150.     }

  1151.     return curbits;

  1152. }

  1153. 

  1154. static float get_band_numbits_UPAIR12_mips(struct AACEncContext *s,

  1155.                                            PutBitContext *pb, const float *in,

  1156.                                            const float *scaled, int size, int scale_idx,

  1157.                                            int cb, const float lambda, const float uplim,

  1158.                                            int *bits)

  1159. {

  1160.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1161.     int i;

  1162.     int qc1, qc2, qc3, qc4;

  1163.     int curbits = 0;

  1164. 

  1165.     uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1166. 

  1167.     for (i = 0; i < size; i += 4) {

  1168.         int curidx, curidx2;

  1169.         int t0, t1, t2, t3, t4;

  1170. 

  1171.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1172.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1173.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1174.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1175. 

  1176.         __asm__ volatile (

  1177.             ".set push                      \n\t"

  1178.             ".set noreorder                 \n\t"

  1179. 

  1180.             "ori    %[t4],  $zero,  12      \n\t"

  1181.             "slt    %[t0],  %[t4],  %[qc1]  \n\t"

  1182.             "slt    %[t1],  %[t4],  %[qc2]  \n\t"

  1183.             "slt    %[t2],  %[t4],  %[qc3]  \n\t"

  1184.             "slt    %[t3],  %[t4],  %[qc4]  \n\t"

  1185.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1186.             "movn   %[qc2], %[t4],  %[t1]   \n\t"

  1187.             "movn   %[qc3], %[t4],  %[t2]   \n\t"

  1188.             "movn   %[qc4], %[t4],  %[t3]   \n\t"

  1189. 

  1190.             ".set pop                       \n\t"

  1191. 

  1192.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1193.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1194.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1195.               [t4]"=&r"(t4)

  1196.         );

  1197. 

  1198.         curidx  = 13 * qc1;

  1199.         curidx += qc2;

  1200. 

  1201.         curidx2  = 13 * qc3;

  1202.         curidx2 += qc4;

  1203. 

  1204.         curbits += p_bits[curidx] +

  1205.                    p_bits[curidx2] +

  1206.                    upair12_sign_bits[curidx] +

  1207.                    upair12_sign_bits[curidx2];

  1208.     }

  1209.     return curbits;

  1210. }

  1211. 

  1212. static float get_band_numbits_ESC_mips(struct AACEncContext *s,

  1213.                                        PutBitContext *pb, const float *in,

  1214.                                        const float *scaled, int size, int scale_idx,

  1215.                                        int cb, const float lambda, const float uplim,

  1216.                                        int *bits)

  1217. {

  1218.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1219.     int i;

  1220.     int qc1, qc2, qc3, qc4;

  1221.     int curbits = 0;

  1222. 

  1223.     uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1];

  1224. 

  1225.     for (i = 0; i < size; i += 4) {

  1226.         int curidx, curidx2;

  1227.         int cond0, cond1, cond2, cond3;

  1228.         int c1, c2, c3, c4;

  1229.         int t4, t5;

  1230. 

  1231.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1232.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1233.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1234.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1235. 

  1236.         __asm__ volatile (

  1237.             ".set push                                  \n\t"

  1238.             ".set noreorder                             \n\t"

  1239. 

  1240.             "ori        %[t4],      $zero,  15          \n\t"

  1241.             "ori        %[t5],      $zero,  16          \n\t"

  1242.             "shll_s.w   %[c1],      %[qc1], 18          \n\t"

  1243.             "shll_s.w   %[c2],      %[qc2], 18          \n\t"

  1244.             "shll_s.w   %[c3],      %[qc3], 18          \n\t"

  1245.             "shll_s.w   %[c4],      %[qc4], 18          \n\t"

  1246.             "srl        %[c1],      %[c1],  18          \n\t"

  1247.             "srl        %[c2],      %[c2],  18          \n\t"

  1248.             "srl        %[c3],      %[c3],  18          \n\t"

  1249.             "srl        %[c4],      %[c4],  18          \n\t"

  1250.             "slt        %[cond0],   %[t4],  %[qc1]      \n\t"

  1251.             "slt        %[cond1],   %[t4],  %[qc2]      \n\t"

  1252.             "slt        %[cond2],   %[t4],  %[qc3]      \n\t"

  1253.             "slt        %[cond3],   %[t4],  %[qc4]      \n\t"

  1254.             "movn       %[qc1],     %[t5],  %[cond0]    \n\t"

  1255.             "movn       %[qc2],     %[t5],  %[cond1]    \n\t"

  1256.             "movn       %[qc3],     %[t5],  %[cond2]    \n\t"

  1257.             "movn       %[qc4],     %[t5],  %[cond3]    \n\t"

  1258.             "ori        %[t5],      $zero,  31          \n\t"

  1259.             "clz        %[c1],      %[c1]               \n\t"

  1260.             "clz        %[c2],      %[c2]               \n\t"

  1261.             "clz        %[c3],      %[c3]               \n\t"

  1262.             "clz        %[c4],      %[c4]               \n\t"

  1263.             "subu       %[c1],      %[t5],  %[c1]       \n\t"

  1264.             "subu       %[c2],      %[t5],  %[c2]       \n\t"

  1265.             "subu       %[c3],      %[t5],  %[c3]       \n\t"

  1266.             "subu       %[c4],      %[t5],  %[c4]       \n\t"

  1267.             "sll        %[c1],      %[c1],  1           \n\t"

  1268.             "sll        %[c2],      %[c2],  1           \n\t"

  1269.             "sll        %[c3],      %[c3],  1           \n\t"

  1270.             "sll        %[c4],      %[c4],  1           \n\t"

  1271.             "addiu      %[c1],      %[c1],  -3          \n\t"

  1272.             "addiu      %[c2],      %[c2],  -3          \n\t"

  1273.             "addiu      %[c3],      %[c3],  -3          \n\t"

  1274.             "addiu      %[c4],      %[c4],  -3          \n\t"

  1275.             "subu       %[cond0],   $zero,  %[cond0]    \n\t"

  1276.             "subu       %[cond1],   $zero,  %[cond1]    \n\t"

  1277.             "subu       %[cond2],   $zero,  %[cond2]    \n\t"

  1278.             "subu       %[cond3],   $zero,  %[cond3]    \n\t"

  1279.             "and        %[c1],      %[c1],  %[cond0]    \n\t"

  1280.             "and        %[c2],      %[c2],  %[cond1]    \n\t"

  1281.             "and        %[c3],      %[c3],  %[cond2]    \n\t"

  1282.             "and        %[c4],      %[c4],  %[cond3]    \n\t"

  1283. 

  1284.             ".set pop                                   \n\t"

  1285. 

  1286.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1287.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1288.               [cond0]"=&r"(cond0), [cond1]"=&r"(cond1),

  1289.               [cond2]"=&r"(cond2), [cond3]"=&r"(cond3),

  1290.               [c1]"=&r"(c1), [c2]"=&r"(c2),

  1291.               [c3]"=&r"(c3), [c4]"=&r"(c4),

  1292.               [t4]"=&r"(t4), [t5]"=&r"(t5)

  1293.         );

  1294. 

  1295.         curidx = 17 * qc1;

  1296.         curidx += qc2;

  1297. 

  1298.         curidx2 = 17 * qc3;

  1299.         curidx2 += qc4;

  1300. 

  1301.         curbits += p_bits[curidx];

  1302.         curbits += esc_sign_bits[curidx];

  1303.         curbits += p_bits[curidx2];

  1304.         curbits += esc_sign_bits[curidx2];

  1305. 

  1306.         curbits += c1;

  1307.         curbits += c2;

  1308.         curbits += c3;

  1309.         curbits += c4;

  1310.     }

  1311.     return curbits;

  1312. }

  1313. 

  1314. static float (*const get_band_numbits_arr[])(struct AACEncContext *s,

  1315.                                              PutBitContext *pb, const float *in,

  1316.                                              const float *scaled, int size, int scale_idx,

  1317.                                              int cb, const float lambda, const float uplim,

  1318.                                              int *bits) = {

  1319.     get_band_numbits_ZERO_mips,

  1320.     get_band_numbits_SQUAD_mips,

  1321.     get_band_numbits_SQUAD_mips,

  1322.     get_band_numbits_UQUAD_mips,

  1323.     get_band_numbits_UQUAD_mips,

  1324.     get_band_numbits_SPAIR_mips,

  1325.     get_band_numbits_SPAIR_mips,

  1326.     get_band_numbits_UPAIR7_mips,

  1327.     get_band_numbits_UPAIR7_mips,

  1328.     get_band_numbits_UPAIR12_mips,

  1329.     get_band_numbits_UPAIR12_mips,

  1330.     get_band_numbits_ESC_mips,

  1331. };

  1332. 

  1333. #define get_band_numbits(                                  \

  1334.                                 s, pb, in, scaled, size, scale_idx, cb, \

  1335.                                 lambda, uplim, bits)                    \

  1336.     get_band_numbits_arr[cb](                              \

  1337.                                 s, pb, in, scaled, size, scale_idx, cb, \

  1338.                                 lambda, uplim, bits)

  1339. 

  1340. static float quantize_band_cost_bits(struct AACEncContext *s, const float *in,

  1341.                                      const float *scaled, int size, int scale_idx,

  1342.                                      int cb, const float lambda, const float uplim,

  1343.                                      int *bits)

  1344. {

  1345.     return get_band_numbits(s, NULL, in, scaled, size, scale_idx, cb, lambda, uplim, bits);

  1346. }

  1347. 

  1348. /**

  1349.  * Functions developed from template function and optimized for getting the band cost

  1350.  */

  1351. #if HAVE_MIPSFPU

  1352. static float get_band_cost_ZERO_mips(struct AACEncContext *s,

  1353.                                      PutBitContext *pb, const float *in,

  1354.                                      const float *scaled, int size, int scale_idx,

  1355.                                      int cb, const float lambda, const float uplim,

  1356.                                      int *bits)

  1357. {

  1358.     int i;

  1359.     float cost = 0;

  1360. 

  1361.     for (i = 0; i < size; i += 4) {

  1362.         cost += in[i  ] * in[i  ];

  1363.         cost += in[i+1] * in[i+1];

  1364.         cost += in[i+2] * in[i+2];

  1365.         cost += in[i+3] * in[i+3];

  1366.     }

  1367.     if (bits)

  1368.         *bits = 0;

  1369.     return cost * lambda;

  1370. }

  1371. 

  1372. static float get_band_cost_SQUAD_mips(struct AACEncContext *s,

  1373.                                       PutBitContext *pb, const float *in,

  1374.                                       const float *scaled, int size, int scale_idx,

  1375.                                       int cb, const float lambda, const float uplim,

  1376.                                       int *bits)

  1377. {

  1378.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1379.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1380.     int i;

  1381.     float cost = 0;

  1382.     int qc1, qc2, qc3, qc4;

  1383.     int curbits = 0;

  1384. 

  1385.     uint8_t *p_bits  = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1386.     float   *p_codes = (float   *)ff_aac_codebook_vectors[cb-1];

  1387. 

  1388.     for (i = 0; i < size; i += 4) {

  1389.         const float *vec;

  1390.         int curidx;

  1391.         int   *in_int = (int   *)&in[i];

  1392.         float *in_pos = (float *)&in[i];

  1393.         float di0, di1, di2, di3;

  1394.         int t0, t1, t2, t3, t4, t5, t6, t7;

  1395. 

  1396.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1397.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1398.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1399.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1400. 

  1401.         __asm__ volatile (

  1402.             ".set push                                  \n\t"

  1403.             ".set noreorder                             \n\t"

  1404. 

  1405.             "slt        %[qc1], $zero,  %[qc1]          \n\t"

  1406.             "slt        %[qc2], $zero,  %[qc2]          \n\t"

  1407.             "slt        %[qc3], $zero,  %[qc3]          \n\t"

  1408.             "slt        %[qc4], $zero,  %[qc4]          \n\t"

  1409.             "lw         %[t0],  0(%[in_int])            \n\t"

  1410.             "lw         %[t1],  4(%[in_int])            \n\t"

  1411.             "lw         %[t2],  8(%[in_int])            \n\t"

  1412.             "lw         %[t3],  12(%[in_int])           \n\t"

  1413.             "srl        %[t0],  %[t0],  31              \n\t"

  1414.             "srl        %[t1],  %[t1],  31              \n\t"

  1415.             "srl        %[t2],  %[t2],  31              \n\t"

  1416.             "srl        %[t3],  %[t3],  31              \n\t"

  1417.             "subu       %[t4],  $zero,  %[qc1]          \n\t"

  1418.             "subu       %[t5],  $zero,  %[qc2]          \n\t"

  1419.             "subu       %[t6],  $zero,  %[qc3]          \n\t"

  1420.             "subu       %[t7],  $zero,  %[qc4]          \n\t"

  1421.             "movn       %[qc1], %[t4],  %[t0]           \n\t"

  1422.             "movn       %[qc2], %[t5],  %[t1]           \n\t"

  1423.             "movn       %[qc3], %[t6],  %[t2]           \n\t"

  1424.             "movn       %[qc4], %[t7],  %[t3]           \n\t"

  1425. 

  1426.             ".set pop                                   \n\t"

  1427. 

  1428.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1429.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1430.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1431.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  1432.             : [in_int]"r"(in_int)

  1433.             : "memory"

  1434.         );

  1435. 

  1436.         curidx = qc1;

  1437.         curidx *= 3;

  1438.         curidx += qc2;

  1439.         curidx *= 3;

  1440.         curidx += qc3;

  1441.         curidx *= 3;

  1442.         curidx += qc4;

  1443.         curidx += 40;

  1444. 

  1445.         curbits += p_bits[curidx];

  1446.         vec     = &p_codes[curidx*4];

  1447. 

  1448.         __asm__ volatile (

  1449.             ".set push                                  \n\t"

  1450.             ".set noreorder                             \n\t"

  1451. 

  1452.             "lwc1       $f0,    0(%[in_pos])            \n\t"

  1453.             "lwc1       $f1,    0(%[vec])               \n\t"

  1454.             "lwc1       $f2,    4(%[in_pos])            \n\t"

  1455.             "lwc1       $f3,    4(%[vec])               \n\t"

  1456.             "lwc1       $f4,    8(%[in_pos])            \n\t"

  1457.             "lwc1       $f5,    8(%[vec])               \n\t"

  1458.             "lwc1       $f6,    12(%[in_pos])           \n\t"

  1459.             "lwc1       $f7,    12(%[vec])              \n\t"

  1460.             "nmsub.s    %[di0], $f0,    $f1,    %[IQ]   \n\t"

  1461.             "nmsub.s    %[di1], $f2,    $f3,    %[IQ]   \n\t"

  1462.             "nmsub.s    %[di2], $f4,    $f5,    %[IQ]   \n\t"

  1463.             "nmsub.s    %[di3], $f6,    $f7,    %[IQ]   \n\t"

  1464. 

  1465.             ".set pop                                   \n\t"

  1466. 

  1467.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  1468.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  1469.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  1470.               [IQ]"f"(IQ)

  1471.             : "$f0", "$f1", "$f2", "$f3",

  1472.               "$f4", "$f5", "$f6", "$f7",

  1473.               "memory"

  1474.         );

  1475. 

  1476.         cost += di0 * di0 + di1 * di1

  1477.                 + di2 * di2 + di3 * di3;

  1478.     }

  1479. 

  1480.     if (bits)

  1481.         *bits = curbits;

  1482.     return cost * lambda + curbits;

  1483. }

  1484. 

  1485. static float get_band_cost_UQUAD_mips(struct AACEncContext *s,

  1486.                                       PutBitContext *pb, const float *in,

  1487.                                       const float *scaled, int size, int scale_idx,

  1488.                                       int cb, const float lambda, const float uplim,

  1489.                                       int *bits)

  1490. {

  1491.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1492.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1493.     int i;

  1494.     float cost = 0;

  1495.     int curbits = 0;

  1496.     int qc1, qc2, qc3, qc4;

  1497. 

  1498.     uint8_t *p_bits  = (uint8_t*)ff_aac_spectral_bits[cb-1];

  1499.     float   *p_codes = (float  *)ff_aac_codebook_vectors[cb-1];

  1500. 

  1501.     for (i = 0; i < size; i += 4) {

  1502.         const float *vec;

  1503.         int curidx;

  1504.         float *in_pos = (float *)&in[i];

  1505.         float di0, di1, di2, di3;

  1506.         int t0, t1, t2, t3, t4;

  1507. 

  1508.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1509.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1510.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1511.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1512. 

  1513.         __asm__ volatile (

  1514.             ".set push                                  \n\t"

  1515.             ".set noreorder                             \n\t"

  1516. 

  1517.             "ori        %[t4],  $zero,  2               \n\t"

  1518.             "slt        %[t0],  %[t4],  %[qc1]          \n\t"

  1519.             "slt        %[t1],  %[t4],  %[qc2]          \n\t"

  1520.             "slt        %[t2],  %[t4],  %[qc3]          \n\t"

  1521.             "slt        %[t3],  %[t4],  %[qc4]          \n\t"

  1522.             "movn       %[qc1], %[t4],  %[t0]           \n\t"

  1523.             "movn       %[qc2], %[t4],  %[t1]           \n\t"

  1524.             "movn       %[qc3], %[t4],  %[t2]           \n\t"

  1525.             "movn       %[qc4], %[t4],  %[t3]           \n\t"

  1526. 

  1527.             ".set pop                                   \n\t"

  1528. 

  1529.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1530.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1531.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1532.               [t4]"=&r"(t4)

  1533.         );

  1534. 

  1535.         curidx = qc1;

  1536.         curidx *= 3;

  1537.         curidx += qc2;

  1538.         curidx *= 3;

  1539.         curidx += qc3;

  1540.         curidx *= 3;

  1541.         curidx += qc4;

  1542. 

  1543.         curbits += p_bits[curidx];

  1544.         curbits += uquad_sign_bits[curidx];

  1545.         vec     = &p_codes[curidx*4];

  1546. 

  1547.         __asm__ volatile (

  1548.             ".set push                                  \n\t"

  1549.             ".set noreorder                             \n\t"

  1550. 

  1551.             "lwc1       %[di0], 0(%[in_pos])            \n\t"

  1552.             "lwc1       %[di1], 4(%[in_pos])            \n\t"

  1553.             "lwc1       %[di2], 8(%[in_pos])            \n\t"

  1554.             "lwc1       %[di3], 12(%[in_pos])           \n\t"

  1555.             "abs.s      %[di0], %[di0]                  \n\t"

  1556.             "abs.s      %[di1], %[di1]                  \n\t"

  1557.             "abs.s      %[di2], %[di2]                  \n\t"

  1558.             "abs.s      %[di3], %[di3]                  \n\t"

  1559.             "lwc1       $f0,    0(%[vec])               \n\t"

  1560.             "lwc1       $f1,    4(%[vec])               \n\t"

  1561.             "lwc1       $f2,    8(%[vec])               \n\t"

  1562.             "lwc1       $f3,    12(%[vec])              \n\t"

  1563.             "nmsub.s    %[di0], %[di0], $f0,    %[IQ]   \n\t"

  1564.             "nmsub.s    %[di1], %[di1], $f1,    %[IQ]   \n\t"

  1565.             "nmsub.s    %[di2], %[di2], $f2,    %[IQ]   \n\t"

  1566.             "nmsub.s    %[di3], %[di3], $f3,    %[IQ]   \n\t"

  1567. 

  1568.             ".set pop                                   \n\t"

  1569. 

  1570.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  1571.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  1572.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  1573.               [IQ]"f"(IQ)

  1574.             : "$f0", "$f1", "$f2", "$f3",

  1575.               "memory"

  1576.         );

  1577. 

  1578.         cost += di0 * di0 + di1 * di1

  1579.                 + di2 * di2 + di3 * di3;

  1580.     }

  1581. 

  1582.     if (bits)

  1583.         *bits = curbits;

  1584.     return cost * lambda + curbits;

  1585. }

  1586. 

  1587. static float get_band_cost_SPAIR_mips(struct AACEncContext *s,

  1588.                                       PutBitContext *pb, const float *in,

  1589.                                       const float *scaled, int size, int scale_idx,

  1590.                                       int cb, const float lambda, const float uplim,

  1591.                                       int *bits)

  1592. {

  1593.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1594.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1595.     int i;

  1596.     float cost = 0;

  1597.     int qc1, qc2, qc3, qc4;

  1598.     int curbits = 0;

  1599. 

  1600.     uint8_t *p_bits  = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1601.     float   *p_codes = (float   *)ff_aac_codebook_vectors[cb-1];

  1602. 

  1603.     for (i = 0; i < size; i += 4) {

  1604.         const float *vec, *vec2;

  1605.         int curidx, curidx2;

  1606.         int   *in_int = (int   *)&in[i];

  1607.         float *in_pos = (float *)&in[i];

  1608.         float di0, di1, di2, di3;

  1609.         int t0, t1, t2, t3, t4, t5, t6, t7;

  1610. 

  1611.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1612.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1613.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1614.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1615. 

  1616.         __asm__ volatile (

  1617.             ".set push                                  \n\t"

  1618.             ".set noreorder                             \n\t"

  1619. 

  1620.             "ori        %[t4],  $zero,  4               \n\t"

  1621.             "slt        %[t0],  %[t4],  %[qc1]          \n\t"

  1622.             "slt        %[t1],  %[t4],  %[qc2]          \n\t"

  1623.             "slt        %[t2],  %[t4],  %[qc3]          \n\t"

  1624.             "slt        %[t3],  %[t4],  %[qc4]          \n\t"

  1625.             "movn       %[qc1], %[t4],  %[t0]           \n\t"

  1626.             "movn       %[qc2], %[t4],  %[t1]           \n\t"

  1627.             "movn       %[qc3], %[t4],  %[t2]           \n\t"

  1628.             "movn       %[qc4], %[t4],  %[t3]           \n\t"

  1629.             "lw         %[t0],  0(%[in_int])            \n\t"

  1630.             "lw         %[t1],  4(%[in_int])            \n\t"

  1631.             "lw         %[t2],  8(%[in_int])            \n\t"

  1632.             "lw         %[t3],  12(%[in_int])           \n\t"

  1633.             "srl        %[t0],  %[t0],  31              \n\t"

  1634.             "srl        %[t1],  %[t1],  31              \n\t"

  1635.             "srl        %[t2],  %[t2],  31              \n\t"

  1636.             "srl        %[t3],  %[t3],  31              \n\t"

  1637.             "subu       %[t4],  $zero,  %[qc1]          \n\t"

  1638.             "subu       %[t5],  $zero,  %[qc2]          \n\t"

  1639.             "subu       %[t6],  $zero,  %[qc3]          \n\t"

  1640.             "subu       %[t7],  $zero,  %[qc4]          \n\t"

  1641.             "movn       %[qc1], %[t4],  %[t0]           \n\t"

  1642.             "movn       %[qc2], %[t5],  %[t1]           \n\t"

  1643.             "movn       %[qc3], %[t6],  %[t2]           \n\t"

  1644.             "movn       %[qc4], %[t7],  %[t3]           \n\t"

  1645. 

  1646.             ".set pop                                   \n\t"

  1647. 

  1648.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1649.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1650.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1651.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  1652.             : [in_int]"r"(in_int)

  1653.             : "memory"

  1654.         );

  1655. 

  1656.         curidx = 9 * qc1;

  1657.         curidx += qc2 + 40;

  1658. 

  1659.         curidx2 = 9 * qc3;

  1660.         curidx2 += qc4 + 40;

  1661. 

  1662.         curbits += p_bits[curidx];

  1663.         curbits += p_bits[curidx2];

  1664. 

  1665.         vec     = &p_codes[curidx*2];

  1666.         vec2    = &p_codes[curidx2*2];

  1667. 

  1668.         __asm__ volatile (

  1669.             ".set push                                  \n\t"

  1670.             ".set noreorder                             \n\t"

  1671. 

  1672.             "lwc1       $f0,    0(%[in_pos])            \n\t"

  1673.             "lwc1       $f1,    0(%[vec])               \n\t"

  1674.             "lwc1       $f2,    4(%[in_pos])            \n\t"

  1675.             "lwc1       $f3,    4(%[vec])               \n\t"

  1676.             "lwc1       $f4,    8(%[in_pos])            \n\t"

  1677.             "lwc1       $f5,    0(%[vec2])              \n\t"

  1678.             "lwc1       $f6,    12(%[in_pos])           \n\t"

  1679.             "lwc1       $f7,    4(%[vec2])              \n\t"

  1680.             "nmsub.s    %[di0], $f0,    $f1,    %[IQ]   \n\t"

  1681.             "nmsub.s    %[di1], $f2,    $f3,    %[IQ]   \n\t"

  1682.             "nmsub.s    %[di2], $f4,    $f5,    %[IQ]   \n\t"

  1683.             "nmsub.s    %[di3], $f6,    $f7,    %[IQ]   \n\t"

  1684. 

  1685.             ".set pop                                   \n\t"

  1686. 

  1687.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  1688.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  1689.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  1690.               [vec2]"r"(vec2), [IQ]"f"(IQ)

  1691.             : "$f0", "$f1", "$f2", "$f3",

  1692.               "$f4", "$f5", "$f6", "$f7",

  1693.               "memory"

  1694.         );

  1695. 

  1696.         cost += di0 * di0 + di1 * di1

  1697.                 + di2 * di2 + di3 * di3;

  1698.     }

  1699. 

  1700.     if (bits)

  1701.         *bits = curbits;

  1702.     return cost * lambda + curbits;

  1703. }

  1704. 

  1705. static float get_band_cost_UPAIR7_mips(struct AACEncContext *s,

  1706.                                        PutBitContext *pb, const float *in,

  1707.                                        const float *scaled, int size, int scale_idx,

  1708.                                        int cb, const float lambda, const float uplim,

  1709.                                        int *bits)

  1710. {

  1711.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1712.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1713.     int i;

  1714.     float cost = 0;

  1715.     int qc1, qc2, qc3, qc4;

  1716.     int curbits = 0;

  1717. 

  1718.     uint8_t *p_bits  = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1719.     float   *p_codes = (float   *)ff_aac_codebook_vectors[cb-1];

  1720. 

  1721.     for (i = 0; i < size; i += 4) {

  1722.         const float *vec, *vec2;

  1723.         int curidx, curidx2, sign1, count1, sign2, count2;

  1724.         int   *in_int = (int   *)&in[i];

  1725.         float *in_pos = (float *)&in[i];

  1726.         float di0, di1, di2, di3;

  1727.         int t0, t1, t2, t3, t4;

  1728. 

  1729.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1730.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1731.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1732.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1733. 

  1734.         __asm__ volatile (

  1735.             ".set push                                          \n\t"

  1736.             ".set noreorder                                     \n\t"

  1737. 

  1738.             "ori        %[t4],      $zero,      7               \n\t"

  1739.             "ori        %[sign1],   $zero,      0               \n\t"

  1740.             "ori        %[sign2],   $zero,      0               \n\t"

  1741.             "slt        %[t0],      %[t4],      %[qc1]          \n\t"

  1742.             "slt        %[t1],      %[t4],      %[qc2]          \n\t"

  1743.             "slt        %[t2],      %[t4],      %[qc3]          \n\t"

  1744.             "slt        %[t3],      %[t4],      %[qc4]          \n\t"

  1745.             "movn       %[qc1],     %[t4],      %[t0]           \n\t"

  1746.             "movn       %[qc2],     %[t4],      %[t1]           \n\t"

  1747.             "movn       %[qc3],     %[t4],      %[t2]           \n\t"

  1748.             "movn       %[qc4],     %[t4],      %[t3]           \n\t"

  1749.             "lw         %[t0],      0(%[in_int])                \n\t"

  1750.             "lw         %[t1],      4(%[in_int])                \n\t"

  1751.             "lw         %[t2],      8(%[in_int])                \n\t"

  1752.             "lw         %[t3],      12(%[in_int])               \n\t"

  1753.             "slt        %[t0],      %[t0],      $zero           \n\t"

  1754.             "movn       %[sign1],   %[t0],      %[qc1]          \n\t"

  1755.             "slt        %[t2],      %[t2],      $zero           \n\t"

  1756.             "movn       %[sign2],   %[t2],      %[qc3]          \n\t"

  1757.             "slt        %[t1],      %[t1],      $zero           \n\t"

  1758.             "sll        %[t0],      %[sign1],   1               \n\t"

  1759.             "or         %[t0],      %[t0],      %[t1]           \n\t"

  1760.             "movn       %[sign1],   %[t0],      %[qc2]          \n\t"

  1761.             "slt        %[t3],      %[t3],      $zero           \n\t"

  1762.             "sll        %[t0],      %[sign2],   1               \n\t"

  1763.             "or         %[t0],      %[t0],      %[t3]           \n\t"

  1764.             "movn       %[sign2],   %[t0],      %[qc4]          \n\t"

  1765.             "slt        %[count1],  $zero,      %[qc1]          \n\t"

  1766.             "slt        %[t1],      $zero,      %[qc2]          \n\t"

  1767.             "slt        %[count2],  $zero,      %[qc3]          \n\t"

  1768.             "slt        %[t2],      $zero,      %[qc4]          \n\t"

  1769.             "addu       %[count1],  %[count1],  %[t1]           \n\t"

  1770.             "addu       %[count2],  %[count2],  %[t2]           \n\t"

  1771. 

  1772.             ".set pop                                           \n\t"

  1773. 

  1774.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1775.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1776.               [sign1]"=&r"(sign1), [count1]"=&r"(count1),

  1777.               [sign2]"=&r"(sign2), [count2]"=&r"(count2),

  1778.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1779.               [t4]"=&r"(t4)

  1780.             : [in_int]"r"(in_int)

  1781.             : "memory"

  1782.         );

  1783. 

  1784.         curidx = 8 * qc1;

  1785.         curidx += qc2;

  1786. 

  1787.         curidx2 = 8 * qc3;

  1788.         curidx2 += qc4;

  1789. 

  1790.         curbits += p_bits[curidx];

  1791.         curbits += upair7_sign_bits[curidx];

  1792.         vec     = &p_codes[curidx*2];

  1793. 

  1794.         curbits += p_bits[curidx2];

  1795.         curbits += upair7_sign_bits[curidx2];

  1796.         vec2    = &p_codes[curidx2*2];

  1797. 

  1798.         __asm__ volatile (

  1799.             ".set push                                          \n\t"

  1800.             ".set noreorder                                     \n\t"

  1801. 

  1802.             "lwc1       %[di0],     0(%[in_pos])                \n\t"

  1803.             "lwc1       %[di1],     4(%[in_pos])                \n\t"

  1804.             "lwc1       %[di2],     8(%[in_pos])                \n\t"

  1805.             "lwc1       %[di3],     12(%[in_pos])               \n\t"

  1806.             "abs.s      %[di0],     %[di0]                      \n\t"

  1807.             "abs.s      %[di1],     %[di1]                      \n\t"

  1808.             "abs.s      %[di2],     %[di2]                      \n\t"

  1809.             "abs.s      %[di3],     %[di3]                      \n\t"

  1810.             "lwc1       $f0,        0(%[vec])                   \n\t"

  1811.             "lwc1       $f1,        4(%[vec])                   \n\t"

  1812.             "lwc1       $f2,        0(%[vec2])                  \n\t"

  1813.             "lwc1       $f3,        4(%[vec2])                  \n\t"

  1814.             "nmsub.s    %[di0],     %[di0],     $f0,    %[IQ]   \n\t"

  1815.             "nmsub.s    %[di1],     %[di1],     $f1,    %[IQ]   \n\t"

  1816.             "nmsub.s    %[di2],     %[di2],     $f2,    %[IQ]   \n\t"

  1817.             "nmsub.s    %[di3],     %[di3],     $f3,    %[IQ]   \n\t"

  1818. 

  1819.             ".set pop                                           \n\t"

  1820. 

  1821.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  1822.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  1823.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  1824.               [vec2]"r"(vec2), [IQ]"f"(IQ)

  1825.             : "$f0", "$f1", "$f2", "$f3",

  1826.               "memory"

  1827.         );

  1828. 

  1829.         cost += di0 * di0 + di1 * di1

  1830.                 + di2 * di2 + di3 * di3;

  1831.     }

  1832. 

  1833.     if (bits)

  1834.         *bits = curbits;

  1835.     return cost * lambda + curbits;

  1836. }

  1837. 

  1838. static float get_band_cost_UPAIR12_mips(struct AACEncContext *s,

  1839.                                         PutBitContext *pb, const float *in,

  1840.                                         const float *scaled, int size, int scale_idx,

  1841.                                         int cb, const float lambda, const float uplim,

  1842.                                         int *bits)

  1843. {

  1844.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1845.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1846.     int i;

  1847.     float cost = 0;

  1848.     int qc1, qc2, qc3, qc4;

  1849.     int curbits = 0;

  1850. 

  1851.     uint8_t *p_bits  = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1852.     float   *p_codes = (float   *)ff_aac_codebook_vectors[cb-1];

  1853. 

  1854.     for (i = 0; i < size; i += 4) {

  1855.         const float *vec, *vec2;

  1856.         int curidx, curidx2;

  1857.         int sign1, count1, sign2, count2;

  1858.         int   *in_int = (int   *)&in[i];

  1859.         float *in_pos = (float *)&in[i];

  1860.         float di0, di1, di2, di3;

  1861.         int t0, t1, t2, t3, t4;

  1862. 

  1863.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1864.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1865.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1866.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1867. 

  1868.         __asm__ volatile (

  1869.             ".set push                                          \n\t"

  1870.             ".set noreorder                                     \n\t"

  1871. 

  1872.             "ori        %[t4],      $zero,      12              \n\t"

  1873.             "ori        %[sign1],   $zero,      0               \n\t"

  1874.             "ori        %[sign2],   $zero,      0               \n\t"

  1875.             "slt        %[t0],      %[t4],      %[qc1]          \n\t"

  1876.             "slt        %[t1],      %[t4],      %[qc2]          \n\t"

  1877.             "slt        %[t2],      %[t4],      %[qc3]          \n\t"

  1878.             "slt        %[t3],      %[t4],      %[qc4]          \n\t"

  1879.             "movn       %[qc1],     %[t4],      %[t0]           \n\t"

  1880.             "movn       %[qc2],     %[t4],      %[t1]           \n\t"

  1881.             "movn       %[qc3],     %[t4],      %[t2]           \n\t"

  1882.             "movn       %[qc4],     %[t4],      %[t3]           \n\t"

  1883.             "lw         %[t0],      0(%[in_int])                \n\t"

  1884.             "lw         %[t1],      4(%[in_int])                \n\t"

  1885.             "lw         %[t2],      8(%[in_int])                \n\t"

  1886.             "lw         %[t3],      12(%[in_int])               \n\t"

  1887.             "slt        %[t0],      %[t0],      $zero           \n\t"

  1888.             "movn       %[sign1],   %[t0],      %[qc1]          \n\t"

  1889.             "slt        %[t2],      %[t2],      $zero           \n\t"

  1890.             "movn       %[sign2],   %[t2],      %[qc3]          \n\t"

  1891.             "slt        %[t1],      %[t1],      $zero           \n\t"

  1892.             "sll        %[t0],      %[sign1],   1               \n\t"

  1893.             "or         %[t0],      %[t0],      %[t1]           \n\t"

  1894.             "movn       %[sign1],   %[t0],      %[qc2]          \n\t"

  1895.             "slt        %[t3],      %[t3],      $zero           \n\t"

  1896.             "sll        %[t0],      %[sign2],   1               \n\t"

  1897.             "or         %[t0],      %[t0],      %[t3]           \n\t"

  1898.             "movn       %[sign2],   %[t0],      %[qc4]          \n\t"

  1899.             "slt        %[count1],  $zero,      %[qc1]          \n\t"

  1900.             "slt        %[t1],      $zero,      %[qc2]          \n\t"

  1901.             "slt        %[count2],  $zero,      %[qc3]          \n\t"

  1902.             "slt        %[t2],      $zero,      %[qc4]          \n\t"

  1903.             "addu       %[count1],  %[count1],  %[t1]           \n\t"

  1904.             "addu       %[count2],  %[count2],  %[t2]           \n\t"

  1905. 

  1906.             ".set pop                                           \n\t"

  1907. 

  1908.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1909.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1910.               [sign1]"=&r"(sign1), [count1]"=&r"(count1),

  1911.               [sign2]"=&r"(sign2), [count2]"=&r"(count2),

  1912.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1913.               [t4]"=&r"(t4)

  1914.             : [in_int]"r"(in_int)

  1915.             : "memory"

  1916.         );

  1917. 

  1918.         curidx = 13 * qc1;

  1919.         curidx += qc2;

  1920. 

  1921.         curidx2 = 13 * qc3;

  1922.         curidx2 += qc4;

  1923. 

  1924.         curbits += p_bits[curidx];

  1925.         curbits += p_bits[curidx2];

  1926.         curbits += upair12_sign_bits[curidx];

  1927.         curbits += upair12_sign_bits[curidx2];

  1928.         vec     = &p_codes[curidx*2];

  1929.         vec2    = &p_codes[curidx2*2];

  1930. 

  1931.         __asm__ volatile (

  1932.             ".set push                                          \n\t"

  1933.             ".set noreorder                                     \n\t"

  1934. 

  1935.             "lwc1       %[di0],     0(%[in_pos])                \n\t"

  1936.             "lwc1       %[di1],     4(%[in_pos])                \n\t"

  1937.             "lwc1       %[di2],     8(%[in_pos])                \n\t"

  1938.             "lwc1       %[di3],     12(%[in_pos])               \n\t"

  1939.             "abs.s      %[di0],     %[di0]                      \n\t"

  1940.             "abs.s      %[di1],     %[di1]                      \n\t"

  1941.             "abs.s      %[di2],     %[di2]                      \n\t"

  1942.             "abs.s      %[di3],     %[di3]                      \n\t"

  1943.             "lwc1       $f0,        0(%[vec])                   \n\t"

  1944.             "lwc1       $f1,        4(%[vec])                   \n\t"

  1945.             "lwc1       $f2,        0(%[vec2])                  \n\t"

  1946.             "lwc1       $f3,        4(%[vec2])                  \n\t"

  1947.             "nmsub.s    %[di0],     %[di0],     $f0,    %[IQ]   \n\t"

  1948.             "nmsub.s    %[di1],     %[di1],     $f1,    %[IQ]   \n\t"

  1949.             "nmsub.s    %[di2],     %[di2],     $f2,    %[IQ]   \n\t"

  1950.             "nmsub.s    %[di3],     %[di3],     $f3,    %[IQ]   \n\t"

  1951. 

  1952.             ".set pop                                           \n\t"

  1953. 

  1954.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  1955.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  1956.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  1957.               [vec2]"r"(vec2), [IQ]"f"(IQ)

  1958.             : "$f0", "$f1", "$f2", "$f3",

  1959.               "memory"

  1960.         );

  1961. 

  1962.         cost += di0 * di0 + di1 * di1

  1963.                 + di2 * di2 + di3 * di3;

  1964.     }

  1965. 

  1966.     if (bits)

  1967.         *bits = curbits;

  1968.     return cost * lambda + curbits;

  1969. }

  1970. 

  1971. static float get_band_cost_ESC_mips(struct AACEncContext *s,

  1972.                                     PutBitContext *pb, const float *in,

  1973.                                     const float *scaled, int size, int scale_idx,

  1974.                                     int cb, const float lambda, const float uplim,

  1975.                                     int *bits)

  1976. {

  1977.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1978.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1979.     const float CLIPPED_ESCAPE = 165140.0f * IQ;

  1980.     int i;

  1981.     float cost = 0;

  1982.     int qc1, qc2, qc3, qc4;

  1983.     int curbits = 0;

  1984. 

  1985.     uint8_t *p_bits  = (uint8_t*)ff_aac_spectral_bits[cb-1];

  1986.     float   *p_codes = (float*  )ff_aac_codebook_vectors[cb-1];

  1987. 

  1988.     for (i = 0; i < size; i += 4) {

  1989.         const float *vec, *vec2;

  1990.         int curidx, curidx2;

  1991.         float t1, t2, t3, t4;

  1992.         float di1, di2, di3, di4;

  1993.         int cond0, cond1, cond2, cond3;

  1994.         int c1, c2, c3, c4;

  1995.         int t6, t7;

  1996. 

  1997.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1998.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1999.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  2000.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  2001. 

  2002.         __asm__ volatile (

  2003.             ".set push                                  \n\t"

  2004.             ".set noreorder                             \n\t"

  2005. 

  2006.             "ori        %[t6],      $zero,  15          \n\t"

  2007.             "ori        %[t7],      $zero,  16          \n\t"

  2008.             "shll_s.w   %[c1],      %[qc1], 18          \n\t"

  2009.             "shll_s.w   %[c2],      %[qc2], 18          \n\t"

  2010.             "shll_s.w   %[c3],      %[qc3], 18          \n\t"

  2011.             "shll_s.w   %[c4],      %[qc4], 18          \n\t"

  2012.             "srl        %[c1],      %[c1],  18          \n\t"

  2013.             "srl        %[c2],      %[c2],  18          \n\t"

  2014.             "srl        %[c3],      %[c3],  18          \n\t"

  2015.             "srl        %[c4],      %[c4],  18          \n\t"

  2016.             "slt        %[cond0],   %[t6],  %[qc1]      \n\t"

  2017.             "slt        %[cond1],   %[t6],  %[qc2]      \n\t"

  2018.             "slt        %[cond2],   %[t6],  %[qc3]      \n\t"

  2019.             "slt        %[cond3],   %[t6],  %[qc4]      \n\t"

  2020.             "movn       %[qc1],     %[t7],  %[cond0]    \n\t"

  2021.             "movn       %[qc2],     %[t7],  %[cond1]    \n\t"

  2022.             "movn       %[qc3],     %[t7],  %[cond2]    \n\t"

  2023.             "movn       %[qc4],     %[t7],  %[cond3]    \n\t"

  2024. 

  2025.             ".set pop                                   \n\t"

  2026. 

  2027.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  2028.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  2029.               [cond0]"=&r"(cond0), [cond1]"=&r"(cond1),

  2030.               [cond2]"=&r"(cond2), [cond3]"=&r"(cond3),

  2031.               [c1]"=&r"(c1), [c2]"=&r"(c2),

  2032.               [c3]"=&r"(c3), [c4]"=&r"(c4),

  2033.               [t6]"=&r"(t6), [t7]"=&r"(t7)

  2034.         );

  2035. 

  2036.         curidx = 17 * qc1;

  2037.         curidx += qc2;

  2038. 

  2039.         curidx2 = 17 * qc3;

  2040.         curidx2 += qc4;

  2041. 

  2042.         curbits += p_bits[curidx];

  2043.         curbits += esc_sign_bits[curidx];

  2044.         vec     = &p_codes[curidx*2];

  2045. 

  2046.         curbits += p_bits[curidx2];

  2047.         curbits += esc_sign_bits[curidx2];

  2048.         vec2     = &p_codes[curidx2*2];

  2049. 

  2050.         curbits += (av_log2(c1) * 2 - 3) & (-cond0);

  2051.         curbits += (av_log2(c2) * 2 - 3) & (-cond1);

  2052.         curbits += (av_log2(c3) * 2 - 3) & (-cond2);

  2053.         curbits += (av_log2(c4) * 2 - 3) & (-cond3);

  2054. 

  2055.         t1 = fabsf(in[i  ]);

  2056.         t2 = fabsf(in[i+1]);

  2057.         t3 = fabsf(in[i+2]);

  2058.         t4 = fabsf(in[i+3]);

  2059. 

  2060.         if (cond0) {

  2061.             if (t1 >= CLIPPED_ESCAPE) {

  2062.                 di1 = t1 - CLIPPED_ESCAPE;

  2063.             } else {

  2064.                 di1 = t1 - c1 * cbrtf(c1) * IQ;

  2065.             }

  2066.         } else

  2067.             di1 = t1 - vec[0] * IQ;

  2068. 

  2069.         if (cond1) {

  2070.             if (t2 >= CLIPPED_ESCAPE) {

  2071.                 di2 = t2 - CLIPPED_ESCAPE;

  2072.             } else {

  2073.                 di2 = t2 - c2 * cbrtf(c2) * IQ;

  2074.             }

  2075.         } else

  2076.             di2 = t2 - vec[1] * IQ;

  2077. 

  2078.         if (cond2) {

  2079.             if (t3 >= CLIPPED_ESCAPE) {

  2080.                 di3 = t3 - CLIPPED_ESCAPE;

  2081.             } else {

  2082.                 di3 = t3 - c3 * cbrtf(c3) * IQ;

  2083.             }

  2084.         } else

  2085.             di3 = t3 - vec2[0] * IQ;

  2086. 

  2087.         if (cond3) {

  2088.             if (t4 >= CLIPPED_ESCAPE) {

  2089.                 di4 = t4 - CLIPPED_ESCAPE;

  2090.             } else {

  2091.                 di4 = t4 - c4 * cbrtf(c4) * IQ;

  2092.             }

  2093.         } else

  2094.             di4 = t4 - vec2[1]*IQ;

  2095. 

  2096.         cost += di1 * di1 + di2 * di2

  2097.                 + di3 * di3 + di4 * di4;

  2098.     }

  2099. 

  2100.     if (bits)

  2101.         *bits = curbits;

  2102.     return cost * lambda + curbits;

  2103. }

  2104. 

  2105. static float (*const get_band_cost_arr[])(struct AACEncContext *s,

  2106.                                           PutBitContext *pb, const float *in,

  2107.                                           const float *scaled, int size, int scale_idx,

  2108.                                           int cb, const float lambda, const float uplim,

  2109.                                           int *bits) = {

  2110.     get_band_cost_ZERO_mips,

  2111.     get_band_cost_SQUAD_mips,

  2112.     get_band_cost_SQUAD_mips,

  2113.     get_band_cost_UQUAD_mips,

  2114.     get_band_cost_UQUAD_mips,

  2115.     get_band_cost_SPAIR_mips,

  2116.     get_band_cost_SPAIR_mips,

  2117.     get_band_cost_UPAIR7_mips,

  2118.     get_band_cost_UPAIR7_mips,

  2119.     get_band_cost_UPAIR12_mips,

  2120.     get_band_cost_UPAIR12_mips,

  2121.     get_band_cost_ESC_mips,

  2122. };

  2123. 

  2124. #define get_band_cost(                                  \

  2125.                                 s, pb, in, scaled, size, scale_idx, cb, \

  2126.                                 lambda, uplim, bits)                    \

  2127.     get_band_cost_arr[cb](                              \

  2128.                                 s, pb, in, scaled, size, scale_idx, cb, \

  2129.                                 lambda, uplim, bits)

  2130. 

  2131. static float quantize_band_cost(struct AACEncContext *s, const float *in,

  2132.                                 const float *scaled, int size, int scale_idx,

  2133.                                 int cb, const float lambda, const float uplim,

  2134.                                 int *bits)

  2135. {

  2136.     return get_band_cost(s, NULL, in, scaled, size, scale_idx, cb, lambda, uplim, bits);

  2137. }

  2138. 

  2139. static void search_for_quantizers_twoloop_mips(AVCodecContext *avctx,

  2140.                                                AACEncContext *s,

  2141.                                                SingleChannelElement *sce,

  2142.                                                const float lambda)

  2143. {

  2144.     int start = 0, i, w, w2, g;

  2145.     int destbits = avctx->bit_rate * 1024.0 / avctx->sample_rate / avctx->channels;

  2146.     float dists[128] = { 0 }, uplims[128];

  2147.     float maxvals[128];

  2148.     int fflag, minscaler;

  2149.     int its  = 0;

  2150.     int allz = 0;

  2151.     float minthr = INFINITY;

  2152. 

  2153.     destbits = FFMIN(destbits, 5800);

  2154.     for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {

  2155.         for (g = 0;  g < sce->ics.num_swb; g++) {

  2156.             int nz = 0;

  2157.             float uplim = 0.0f;

  2158.             for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {

  2159.                 FFPsyBand *band = &s->psy.ch[s->cur_channel].psy_bands[(w+w2)*16+g];

  2160.                 uplim += band->threshold;

  2161.                 if (band->energy <= band->threshold || band->threshold == 0.0f) {

  2162.                     sce->zeroes[(w+w2)*16+g] = 1;

  2163.                     continue;

  2164.                 }

  2165.                 nz = 1;

  2166.             }

  2167.             uplims[w*16+g] = uplim *512;

  2168.             sce->zeroes[w*16+g] = !nz;

  2169.             if (nz)

  2170.                 minthr = FFMIN(minthr, uplim);

  2171.             allz |= nz;

  2172.         }

  2173.     }

  2174.     for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {

  2175.         for (g = 0;  g < sce->ics.num_swb; g++) {

  2176.             if (sce->zeroes[w*16+g]) {

  2177.                 sce->sf_idx[w*16+g] = SCALE_ONE_POS;

  2178.                 continue;

  2179.             }

  2180.             sce->sf_idx[w*16+g] = SCALE_ONE_POS + FFMIN(log2f(uplims[w*16+g]/minthr)*4,59);

  2181.         }

  2182.     }

  2183. 

  2184.     if (!allz)

  2185.         return;

  2186.     abs_pow34_v(s->scoefs, sce->coeffs, 1024);

  2187. 

  2188.     for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {

  2189.         start = w*128;

  2190.         for (g = 0;  g < sce->ics.num_swb; g++) {

  2191.             const float *scaled = s->scoefs + start;

  2192.             maxvals[w*16+g] = find_max_val(sce->ics.group_len[w], sce->ics.swb_sizes[g], scaled);

  2193.             start += sce->ics.swb_sizes[g];

  2194.         }

  2195.     }

  2196. 

  2197.     do {

  2198.         int tbits, qstep;

  2199.         minscaler = sce->sf_idx[0];

  2200.         qstep = its ? 1 : 32;

  2201.         do {

  2202.             int prev = -1;

  2203.             tbits = 0;

  2204.             fflag = 0;

  2205. 

  2206.             if (qstep > 1) {

  2207.                 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {

  2208.                     start = w*128;

  2209.                     for (g = 0;  g < sce->ics.num_swb; g++) {

  2210.                         const float *coefs = sce->coeffs + start;

  2211.                         const float *scaled = s->scoefs + start;

  2212.                         int bits = 0;

  2213.                         int cb;

  2214. 

  2215.                         if (sce->zeroes[w*16+g] || sce->sf_idx[w*16+g] >= 218) {

  2216.                             start += sce->ics.swb_sizes[g];

  2217.                             continue;

  2218.                         }

  2219.                         minscaler = FFMIN(minscaler, sce->sf_idx[w*16+g]);

  2220.                         cb = find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]);

  2221.                         for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {

  2222.                             int b;

  2223.                             bits += quantize_band_cost_bits(s, coefs + w2*128,

  2224.                                                             scaled + w2*128,

  2225.                                                             sce->ics.swb_sizes[g],

  2226.                                                             sce->sf_idx[w*16+g],

  2227.                                                             cb,

  2228.                                                             1.0f,

  2229.                                                             INFINITY,

  2230.                                                             &b);

  2231.                         }

  2232.                         if (prev != -1) {

  2233.                             bits += ff_aac_scalefactor_bits[sce->sf_idx[w*16+g] - prev + SCALE_DIFF_ZERO];

  2234.                         }

  2235.                         tbits += bits;

  2236.                         start += sce->ics.swb_sizes[g];

  2237.                         prev = sce->sf_idx[w*16+g];

  2238.                     }

  2239.                 }

  2240.             }

  2241.             else {

  2242.                 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {

  2243.                     start = w*128;

  2244.                     for (g = 0;  g < sce->ics.num_swb; g++) {

  2245.                         const float *coefs = sce->coeffs + start;

  2246.                         const float *scaled = s->scoefs + start;

  2247.                         int bits = 0;

  2248.                         int cb;

  2249.                         float dist = 0.0f;

  2250. 

  2251.                         if (sce->zeroes[w*16+g] || sce->sf_idx[w*16+g] >= 218) {

  2252.                             start += sce->ics.swb_sizes[g];

  2253.                             continue;

  2254.                         }

  2255.                         minscaler = FFMIN(minscaler, sce->sf_idx[w*16+g]);

  2256.                         cb = find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]);

  2257.                         for (w2 = 0; w2 < sce->ics.group_len[w]; w2++) {

  2258.                             int b;

  2259.                             dist += quantize_band_cost(s, coefs + w2*128,

  2260.                                                        scaled + w2*128,

  2261.                                                        sce->ics.swb_sizes[g],

  2262.                                                        sce->sf_idx[w*16+g],

  2263.                                                        cb,

  2264.                                                        1.0f,

  2265.                                                        INFINITY,

  2266.                                                        &b);

  2267.                             bits += b;

  2268.                         }

  2269.                         dists[w*16+g] = dist - bits;

  2270.                         if (prev != -1) {

  2271.                             bits += ff_aac_scalefactor_bits[sce->sf_idx[w*16+g] - prev + SCALE_DIFF_ZERO];

  2272.                         }

  2273.                         tbits += bits;

  2274.                         start += sce->ics.swb_sizes[g];

  2275.                         prev = sce->sf_idx[w*16+g];

  2276.                     }

  2277.                 }

  2278.             }

  2279.             if (tbits > destbits) {

  2280.                 for (i = 0; i < 128; i++)

  2281.                     if (sce->sf_idx[i] < 218 - qstep)

  2282.                         sce->sf_idx[i] += qstep;

  2283.             } else {

  2284.                 for (i = 0; i < 128; i++)

  2285.                     if (sce->sf_idx[i] > 60 - qstep)

  2286.                         sce->sf_idx[i] -= qstep;

  2287.             }

  2288.             qstep >>= 1;

  2289.             if (!qstep && tbits > destbits*1.02 && sce->sf_idx[0] < 217)

  2290.                 qstep = 1;

  2291.         } while (qstep);

  2292. 

  2293.         fflag = 0;

  2294.         minscaler = av_clip(minscaler, 60, 255 - SCALE_MAX_DIFF);

  2295.         for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {

  2296.             for (g = 0; g < sce->ics.num_swb; g++) {

  2297.                 int prevsc = sce->sf_idx[w*16+g];

  2298.                 if (dists[w*16+g] > uplims[w*16+g] && sce->sf_idx[w*16+g] > 60) {

  2299.                     if (find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]-1))

  2300.                         sce->sf_idx[w*16+g]--;

  2301.                     else

  2302.                         sce->sf_idx[w*16+g]-=2;

  2303.                 }

  2304.                 sce->sf_idx[w*16+g] = av_clip(sce->sf_idx[w*16+g], minscaler, minscaler + SCALE_MAX_DIFF);

  2305.                 sce->sf_idx[w*16+g] = FFMIN(sce->sf_idx[w*16+g], 219);

  2306.                 if (sce->sf_idx[w*16+g] != prevsc)

  2307.                     fflag = 1;

  2308.                 sce->band_type[w*16+g] = find_min_book(maxvals[w*16+g], sce->sf_idx[w*16+g]);

  2309.             }

  2310.         }

  2311.         its++;

  2312.     } while (fflag && its < 10);

  2313. }

  2314. 

  2315. static void search_for_ms_mips(AACEncContext *s, ChannelElement *cpe)

  2316. {

  2317.     int start = 0, i, w, w2, g;

  2318.     float M[128], S[128];

  2319.     float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3;

  2320.     SingleChannelElement *sce0 = &cpe->ch[0];

  2321.     SingleChannelElement *sce1 = &cpe->ch[1];

  2322.     if (!cpe->common_window)

  2323.         return;

  2324.     for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {

  2325.         for (g = 0;  g < sce0->ics.num_swb; g++) {

  2326.             if (!cpe->ch[0].zeroes[w*16+g] && !cpe->ch[1].zeroes[w*16+g]) {

  2327.                 float dist1 = 0.0f, dist2 = 0.0f;

  2328.                 for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {

  2329.                     FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];

  2330.                     FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g];

  2331.                     float minthr = FFMIN(band0->threshold, band1->threshold);

  2332.                     float maxthr = FFMAX(band0->threshold, band1->threshold);

  2333.                     for (i = 0; i < sce0->ics.swb_sizes[g]; i+=4) {

  2334.                         M[i  ] = (sce0->coeffs[start+w2*128+i  ]

  2335.                                 + sce1->coeffs[start+w2*128+i  ]) * 0.5;

  2336.                         M[i+1] = (sce0->coeffs[start+w2*128+i+1]

  2337.                                 + sce1->coeffs[start+w2*128+i+1]) * 0.5;

  2338.                         M[i+2] = (sce0->coeffs[start+w2*128+i+2]

  2339.                                 + sce1->coeffs[start+w2*128+i+2]) * 0.5;

  2340.                         M[i+3] = (sce0->coeffs[start+w2*128+i+3]

  2341.                                 + sce1->coeffs[start+w2*128+i+3]) * 0.5;

  2342. 

  2343.                         S[i  ] =  M[i  ]

  2344.                                 - sce1->coeffs[start+w2*128+i  ];

  2345.                         S[i+1] =  M[i+1]

  2346.                                 - sce1->coeffs[start+w2*128+i+1];

  2347.                         S[i+2] =  M[i+2]

  2348.                                 - sce1->coeffs[start+w2*128+i+2];

  2349.                         S[i+3] =  M[i+3]

  2350.                                 - sce1->coeffs[start+w2*128+i+3];

  2351.                    }

  2352.                     abs_pow34_v(L34, sce0->coeffs+start+w2*128, sce0->ics.swb_sizes[g]);

  2353.                     abs_pow34_v(R34, sce1->coeffs+start+w2*128, sce0->ics.swb_sizes[g]);

  2354.                     abs_pow34_v(M34, M,                         sce0->ics.swb_sizes[g]);

  2355.                     abs_pow34_v(S34, S,                         sce0->ics.swb_sizes[g]);

  2356.                     dist1 += quantize_band_cost(s, sce0->coeffs + start + w2*128,

  2357.                                                 L34,

  2358.                                                 sce0->ics.swb_sizes[g],

  2359.                                                 sce0->sf_idx[(w+w2)*16+g],

  2360.                                                 sce0->band_type[(w+w2)*16+g],

  2361.                                                 s->lambda / band0->threshold, INFINITY, NULL);

  2362.                     dist1 += quantize_band_cost(s, sce1->coeffs + start + w2*128,

  2363.                                                 R34,

  2364.                                                 sce1->ics.swb_sizes[g],

  2365.                                                 sce1->sf_idx[(w+w2)*16+g],

  2366.                                                 sce1->band_type[(w+w2)*16+g],

  2367.                                                 s->lambda / band1->threshold, INFINITY, NULL);

  2368.                     dist2 += quantize_band_cost(s, M,

  2369.                                                 M34,

  2370.                                                 sce0->ics.swb_sizes[g],

  2371.                                                 sce0->sf_idx[(w+w2)*16+g],

  2372.                                                 sce0->band_type[(w+w2)*16+g],

  2373.                                                 s->lambda / maxthr, INFINITY, NULL);

  2374.                     dist2 += quantize_band_cost(s, S,

  2375.                                                 S34,

  2376.                                                 sce1->ics.swb_sizes[g],

  2377.                                                 sce1->sf_idx[(w+w2)*16+g],

  2378.                                                 sce1->band_type[(w+w2)*16+g],

  2379.                                                 s->lambda / minthr, INFINITY, NULL);

  2380.                 }

  2381.                 cpe->ms_mask[w*16+g] = dist2 < dist1;

  2382.             }

  2383.             start += sce0->ics.swb_sizes[g];

  2384.         }

  2385.     }

  2386. }

  2387. #endif /*HAVE_MIPSFPU */

  2388. 

  2389. static void codebook_trellis_rate_mips(AACEncContext *s, SingleChannelElement *sce,

  2390.                                        int win, int group_len, const float lambda)

  2391. {

  2392.     BandCodingPath path[120][12];

  2393.     int w, swb, cb, start, size;

  2394.     int i, j;

  2395.     const int max_sfb  = sce->ics.max_sfb;

  2396.     const int run_bits = sce->ics.num_windows == 1 ? 5 : 3;

  2397.     const int run_esc  = (1 << run_bits) - 1;

  2398.     int idx, ppos, count;

  2399.     int stackrun[120], stackcb[120], stack_len;

  2400.     float next_minbits = INFINITY;

  2401.     int next_mincb = 0;

  2402. 

  2403.     abs_pow34_v(s->scoefs, sce->coeffs, 1024);

  2404.     start = win*128;

  2405.     for (cb = 0; cb < 12; cb++) {

  2406.         path[0][cb].cost     = run_bits+4;

  2407.         path[0][cb].prev_idx = -1;

  2408.         path[0][cb].run      = 0;

  2409.     }

  2410.     for (swb = 0; swb < max_sfb; swb++) {

  2411.         size = sce->ics.swb_sizes[swb];

  2412.         if (sce->zeroes[win*16 + swb]) {

  2413.             float cost_stay_here = path[swb][0].cost;

  2414.             float cost_get_here  = next_minbits + run_bits + 4;

  2415.             if (   run_value_bits[sce->ics.num_windows == 8][path[swb][0].run]

  2416.                 != run_value_bits[sce->ics.num_windows == 8][path[swb][0].run+1])

  2417.                 cost_stay_here += run_bits;

  2418.             if (cost_get_here < cost_stay_here) {

  2419.                 path[swb+1][0].prev_idx = next_mincb;

  2420.                 path[swb+1][0].cost     = cost_get_here;

  2421.                 path[swb+1][0].run      = 1;

  2422.             } else {

  2423.                 path[swb+1][0].prev_idx = 0;

  2424.                 path[swb+1][0].cost     = cost_stay_here;

  2425.                 path[swb+1][0].run      = path[swb][0].run + 1;

  2426.             }

  2427.             next_minbits = path[swb+1][0].cost;

  2428.             next_mincb = 0;

  2429.             for (cb = 1; cb < 12; cb++) {

  2430.                 path[swb+1][cb].cost = 61450;

  2431.                 path[swb+1][cb].prev_idx = -1;

  2432.                 path[swb+1][cb].run = 0;

  2433.             }

  2434.         } else {

  2435.             float minbits = next_minbits;

  2436.             int mincb = next_mincb;

  2437.             int startcb = sce->band_type[win*16+swb];

  2438.             next_minbits = INFINITY;

  2439.             next_mincb = 0;

  2440.             for (cb = 0; cb < startcb; cb++) {

  2441.                 path[swb+1][cb].cost = 61450;

  2442.                 path[swb+1][cb].prev_idx = -1;

  2443.                 path[swb+1][cb].run = 0;

  2444.             }

  2445.             for (cb = startcb; cb < 12; cb++) {

  2446.                 float cost_stay_here, cost_get_here;

  2447.                 float bits = 0.0f;

  2448.                 for (w = 0; w < group_len; w++) {

  2449.                     bits += quantize_band_cost_bits(s, sce->coeffs + start + w*128,

  2450.                                                     s->scoefs + start + w*128, size,

  2451.                                                     sce->sf_idx[(win+w)*16+swb], cb,

  2452.                                                     0, INFINITY, NULL);

  2453.                 }

  2454.                 cost_stay_here = path[swb][cb].cost + bits;

  2455.                 cost_get_here  = minbits            + bits + run_bits + 4;

  2456.                 if (   run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run]

  2457.                     != run_value_bits[sce->ics.num_windows == 8][path[swb][cb].run+1])

  2458.                     cost_stay_here += run_bits;

  2459.                 if (cost_get_here < cost_stay_here) {

  2460.                     path[swb+1][cb].prev_idx = mincb;

  2461.                     path[swb+1][cb].cost     = cost_get_here;

  2462.                     path[swb+1][cb].run      = 1;

  2463.                 } else {

  2464.                     path[swb+1][cb].prev_idx = cb;

  2465.                     path[swb+1][cb].cost     = cost_stay_here;

  2466.                     path[swb+1][cb].run      = path[swb][cb].run + 1;

  2467.                 }

  2468.                 if (path[swb+1][cb].cost < next_minbits) {

  2469.                     next_minbits = path[swb+1][cb].cost;

  2470.                     next_mincb = cb;

  2471.                 }

  2472.             }

  2473.         }

  2474.         start += sce->ics.swb_sizes[swb];

  2475.     }

  2476. 

  2477.     stack_len = 0;

  2478.     idx       = 0;

  2479.     for (cb = 1; cb < 12; cb++)

  2480.         if (path[max_sfb][cb].cost < path[max_sfb][idx].cost)

  2481.             idx = cb;

  2482.     ppos = max_sfb;

  2483.     while (ppos > 0) {

  2484.         av_assert1(idx >= 0);

  2485.         cb = idx;

  2486.         stackrun[stack_len] = path[ppos][cb].run;

  2487.         stackcb [stack_len] = cb;

  2488.         idx = path[ppos-path[ppos][cb].run+1][cb].prev_idx;

  2489.         ppos -= path[ppos][cb].run;

  2490.         stack_len++;

  2491.     }

  2492. 

  2493.     start = 0;

  2494.     for (i = stack_len - 1; i >= 0; i--) {

  2495.         put_bits(&s->pb, 4, stackcb[i]);

  2496.         count = stackrun[i];

  2497.         memset(sce->zeroes + win*16 + start, !stackcb[i], count);

  2498.         for (j = 0; j < count; j++) {

  2499.             sce->band_type[win*16 + start] =  stackcb[i];

  2500.             start++;

  2501.         }

  2502.         while (count >= run_esc) {

  2503.             put_bits(&s->pb, run_bits, run_esc);

  2504.             count -= run_esc;

  2505.         }

  2506.         put_bits(&s->pb, run_bits, count);

  2507.     }

  2508. }

  2509. #endif /* HAVE_INLINE_ASM */

  2510. 

  2511. void ff_aac_coder_init_mips(AACEncContext *c) {

  2512. #if HAVE_INLINE_ASM

  2513.     AACCoefficientsEncoder *e = c->coder;

  2514.     int option = c->options.aac_coder;

  2515. 

  2516.     if (option == 2) {

  2517. // Disabled due to failure with fate-aac-pns-encode

  2518. //         e->quantize_and_encode_band = quantize_and_encode_band_mips;

  2519. //         e->encode_window_bands_info = codebook_trellis_rate_mips;

  2520. #if HAVE_MIPSFPU

  2521.         e->search_for_quantizers    = search_for_quantizers_twoloop_mips;

  2522.         e->search_for_ms            = search_for_ms_mips;

  2523. #endif /* HAVE_MIPSFPU */

  2524.     }

  2525. #endif /* HAVE_INLINE_ASM */

  2526. }