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.
75489 Commits
32 Branches
321MB
Tree: b01f3ddad3
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'b01f3ddad3'
${ noResults }
FFmpeg/libavcodec/mips/aaccoder_mips.c

2357 lines
97KB
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. #include "libavcodec/aacenctab.h"

  66. 

  67. #if HAVE_INLINE_ASM

  68. typedef struct BandCodingPath {

  69.     int prev_idx;

  70.     float cost;

  71.     int run;

  72. } BandCodingPath;

  73. 

  74. static const uint8_t uquad_sign_bits[81] = {

  75.     0, 1, 1, 1, 2, 2, 1, 2, 2,

  76.     1, 2, 2, 2, 3, 3, 2, 3, 3,

  77.     1, 2, 2, 2, 3, 3, 2, 3, 3,

  78.     1, 2, 2, 2, 3, 3, 2, 3, 3,

  79.     2, 3, 3, 3, 4, 4, 3, 4, 4,

  80.     2, 3, 3, 3, 4, 4, 3, 4, 4,

  81.     1, 2, 2, 2, 3, 3, 2, 3, 3,

  82.     2, 3, 3, 3, 4, 4, 3, 4, 4,

  83.     2, 3, 3, 3, 4, 4, 3, 4, 4

  84. };

  85. 

  86. static const uint8_t upair7_sign_bits[64] = {

  87.     0, 1, 1, 1, 1, 1, 1, 1,

  88.     1, 2, 2, 2, 2, 2, 2, 2,

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

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

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

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

  93.     1, 2, 2, 2, 2, 2, 2, 2,

  94.     1, 2, 2, 2, 2, 2, 2, 2,

  95. };

  96. 

  97. static const uint8_t upair12_sign_bits[169] = {

  98.     0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,

  99.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  100.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  101.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

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

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

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

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

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

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

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

  109.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  110.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2

  111. };

  112. 

  113. static const uint8_t esc_sign_bits[289] = {

  114.     0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,

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

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

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

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

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

  125.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  126.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  127.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  128.     1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,

  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. };

  132. 

  133. #define ROUND_STANDARD 0.4054f

  134. #define ROUND_TO_ZERO 0.1054f

  135. 

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

  137. #ifndef USE_REALLY_FULL_SEARCH

  138.     int i;

  139.     float a, b, c, d;

  140.     float ax, bx, cx, dx;

  141. 

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

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

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

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

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

  147. 

  148.         ax = sqrtf(a);

  149.         bx = sqrtf(b);

  150.         cx = sqrtf(c);

  151.         dx = sqrtf(d);

  152. 

  153.         a = a * ax;

  154.         b = b * bx;

  155.         c = c * cx;

  156.         d = d * dx;

  157. 

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

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

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

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

  162.     }

  163. #endif /* USE_REALLY_FULL_SEARCH */

  164. }

  165. 

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

  167.     float maxval = 0.0f;

  168.     int w2, i;

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

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

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

  172.         }

  173.     }

  174.     return maxval;

  175. }

  176. 

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

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

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

  180.     int qmaxval, cb;

  181.     if (qmaxval >= (FF_ARRAY_ELEMS(aac_maxval_cb)))

  182.         cb = 11;

  183.     else

  184.         cb = aac_maxval_cb[qmaxval];

  185.     return cb;

  186. }

  187. 

  188. /**

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

  190.  */

  191. static void quantize_and_encode_band_cost_SQUAD_mips(struct AACEncContext *s,

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

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

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

  195.                                                      int *bits, const float ROUNDING)

  196. {

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

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

  199.     int i;

  200.     int qc1, qc2, qc3, qc4;

  201. 

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

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

  204.     float    *p_vec   = (float    *)ff_aac_codebook_vectors[cb-1];

  205. 

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

  207.     scaled = s->scoefs;

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

  209.         int curidx;

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

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

  212.         const float *vec;

  213. 

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

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

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

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

  218. 

  219.         __asm__ volatile (

  220.             ".set push                      \n\t"

  221.             ".set noreorder                 \n\t"

  222. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  243. 

  244.             ".set pop                       \n\t"

  245. 

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

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

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

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

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

  251.             : "memory"

  252.         );

  253. 

  254.         curidx = qc1;

  255.         curidx *= 3;

  256.         curidx += qc2;

  257.         curidx *= 3;

  258.         curidx += qc3;

  259.         curidx *= 3;

  260.         curidx += qc4;

  261.         curidx += 40;

  262. 

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

  264. 

  265.         if (out) {

  266.            vec = &p_vec[curidx*4];

  267.            out[i+0] = vec[0] * IQ;

  268.            out[i+1] = vec[1] * IQ;

  269.            out[i+2] = vec[2] * IQ;

  270.            out[i+3] = vec[3] * IQ;

  271.         }

  272.     }

  273. }

  274. 

  275. static void quantize_and_encode_band_cost_UQUAD_mips(struct AACEncContext *s,

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

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

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

  279.                                                      int *bits, const float ROUNDING)

  280. {

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

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

  283.     int i;

  284.     int qc1, qc2, qc3, qc4;

  285. 

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

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

  288.     float    *p_vec   = (float    *)ff_aac_codebook_vectors[cb-1];

  289. 

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

  291.     scaled = s->scoefs;

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

  293.         int curidx, sign, count;

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

  295.         uint8_t v_bits;

  296.         unsigned int v_codes;

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

  298.         const float *vec;

  299. 

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

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

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

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

  304. 

  305.         __asm__ volatile (

  306.             ".set push                              \n\t"

  307.             ".set noreorder                         \n\t"

  308. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  344. 

  345.             ".set pop                               \n\t"

  346. 

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

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

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

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

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

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

  353.             : "memory"

  354.         );

  355. 

  356.         curidx = qc1;

  357.         curidx *= 3;

  358.         curidx += qc2;

  359.         curidx *= 3;

  360.         curidx += qc3;

  361.         curidx *= 3;

  362.         curidx += qc4;

  363. 

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

  365.         v_bits  = p_bits[curidx] + count;

  366.         put_bits(pb, v_bits, v_codes);

  367. 

  368.         if (out) {

  369.            vec = &p_vec[curidx*4];

  370.            out[i+0] = copysignf(vec[0] * IQ, in[i+0]);

  371.            out[i+1] = copysignf(vec[1] * IQ, in[i+1]);

  372.            out[i+2] = copysignf(vec[2] * IQ, in[i+2]);

  373.            out[i+3] = copysignf(vec[3] * IQ, in[i+3]);

  374.         }

  375.     }

  376. }

  377. 

  378. static void quantize_and_encode_band_cost_SPAIR_mips(struct AACEncContext *s,

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

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

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

  382.                                                      int *bits, const float ROUNDING)

  383. {

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

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

  386.     int i;

  387.     int qc1, qc2, qc3, qc4;

  388. 

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

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

  391.     float    *p_vec   = (float    *)ff_aac_codebook_vectors[cb-1];

  392. 

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

  394.     scaled = s->scoefs;

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

  396.         int curidx, curidx2;

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

  398.         uint8_t v_bits;

  399.         unsigned int v_codes;

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

  401.         const float *vec1, *vec2;

  402. 

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

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

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

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

  407. 

  408.         __asm__ volatile (

  409.             ".set push                      \n\t"

  410.             ".set noreorder                 \n\t"

  411. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  437. 

  438.             ".set pop                       \n\t"

  439. 

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

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

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

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

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

  445.             : "memory"

  446.         );

  447. 

  448.         curidx = 9 * qc1;

  449.         curidx += qc2 + 40;

  450. 

  451.         curidx2 = 9 * qc3;

  452.         curidx2 += qc4 + 40;

  453. 

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

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

  456.         put_bits(pb, v_bits, v_codes);

  457. 

  458.         if (out) {

  459.            vec1 = &p_vec[curidx*2 ];

  460.            vec2 = &p_vec[curidx2*2];

  461.            out[i+0] = vec1[0] * IQ;

  462.            out[i+1] = vec1[1] * IQ;

  463.            out[i+2] = vec2[0] * IQ;

  464.            out[i+3] = vec2[1] * IQ;

  465.         }

  466.     }

  467. }

  468. 

  469. static void quantize_and_encode_band_cost_UPAIR7_mips(struct AACEncContext *s,

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

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

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

  473.                                                       int *bits, const float ROUNDING)

  474. {

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

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

  477.     int i;

  478.     int qc1, qc2, qc3, qc4;

  479. 

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

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

  482.     float    *p_vec   = (float    *)ff_aac_codebook_vectors[cb-1];

  483. 

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

  485.     scaled = s->scoefs;

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

  487.         int curidx1, curidx2, sign1, count1, sign2, count2;

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

  489.         uint8_t v_bits;

  490.         unsigned int v_codes;

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

  492.         const float *vec1, *vec2;

  493. 

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

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

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

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

  498. 

  499.         __asm__ volatile (

  500.             ".set push                              \n\t"

  501.             ".set noreorder                         \n\t"

  502. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  536. 

  537.             ".set pop                               \n\t"

  538. 

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

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

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

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

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

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

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

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

  547.               "memory"

  548.         );

  549. 

  550.         curidx1  = 8 * qc1;

  551.         curidx1 += qc2;

  552. 

  553.         v_codes = (p_codes[curidx1] << count1) | sign1;

  554.         v_bits  = p_bits[curidx1] + count1;

  555.         put_bits(pb, v_bits, v_codes);

  556. 

  557.         curidx2  = 8 * qc3;

  558.         curidx2 += qc4;

  559. 

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

  561.         v_bits  = p_bits[curidx2] + count2;

  562.         put_bits(pb, v_bits, v_codes);

  563. 

  564.         if (out) {

  565.            vec1 = &p_vec[curidx1*2];

  566.            vec2 = &p_vec[curidx2*2];

  567.            out[i+0] = copysignf(vec1[0] * IQ, in[i+0]);

  568.            out[i+1] = copysignf(vec1[1] * IQ, in[i+1]);

  569.            out[i+2] = copysignf(vec2[0] * IQ, in[i+2]);

  570.            out[i+3] = copysignf(vec2[1] * IQ, in[i+3]);

  571.         }

  572.     }

  573. }

  574. 

  575. static void quantize_and_encode_band_cost_UPAIR12_mips(struct AACEncContext *s,

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

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

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

  579.                                                        int *bits, const float ROUNDING)

  580. {

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

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

  583.     int i;

  584.     int qc1, qc2, qc3, qc4;

  585. 

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

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

  588.     float    *p_vec   = (float   *)ff_aac_codebook_vectors[cb-1];

  589. 

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

  591.     scaled = s->scoefs;

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

  593.         int curidx1, curidx2, sign1, count1, sign2, count2;

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

  595.         uint8_t v_bits;

  596.         unsigned int v_codes;

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

  598.         const float *vec1, *vec2;

  599. 

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

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

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

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

  604. 

  605.         __asm__ volatile (

  606.             ".set push                              \n\t"

  607.             ".set noreorder                         \n\t"

  608. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  642. 

  643.             ".set pop                               \n\t"

  644. 

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

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

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

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

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

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

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

  652.             : "memory"

  653.         );

  654. 

  655.         curidx1  = 13 * qc1;

  656.         curidx1 += qc2;

  657. 

  658.         v_codes = (p_codes[curidx1] << count1) | sign1;

  659.         v_bits  = p_bits[curidx1] + count1;

  660.         put_bits(pb, v_bits, v_codes);

  661. 

  662.         curidx2  = 13 * qc3;

  663.         curidx2 += qc4;

  664. 

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

  666.         v_bits  = p_bits[curidx2] + count2;

  667.         put_bits(pb, v_bits, v_codes);

  668. 

  669.         if (out) {

  670.            vec1 = &p_vec[curidx1*2];

  671.            vec2 = &p_vec[curidx2*2];

  672.            out[i+0] = copysignf(vec1[0] * IQ, in[i+0]);

  673.            out[i+1] = copysignf(vec1[1] * IQ, in[i+1]);

  674.            out[i+2] = copysignf(vec2[0] * IQ, in[i+2]);

  675.            out[i+3] = copysignf(vec2[1] * IQ, in[i+3]);

  676.         }

  677.     }

  678. }

  679. 

  680. static void quantize_and_encode_band_cost_ESC_mips(struct AACEncContext *s,

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

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

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

  684.                                                    int *bits, const float ROUNDING)

  685. {

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

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

  688.     int i;

  689.     int qc1, qc2, qc3, qc4;

  690. 

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

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

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

  694. 

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

  696.     scaled = s->scoefs;

  697. 

  698.     if (cb < 11) {

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

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

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

  702.             uint8_t v_bits;

  703.             unsigned int v_codes;

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

  705.             const float *vec1, *vec2;

  706. 

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

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

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

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

  711. 

  712.             __asm__ volatile (

  713.                 ".set push                                  \n\t"

  714.                 ".set noreorder                             \n\t"

  715. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  749. 

  750.                 ".set pop                                   \n\t"

  751. 

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

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

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

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

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

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

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

  759.                 : "memory"

  760.             );

  761. 

  762.             curidx = 17 * qc1;

  763.             curidx += qc2;

  764.             curidx2 = 17 * qc3;

  765.             curidx2 += qc4;

  766. 

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

  768.             v_bits  = p_bits[curidx] + count1;

  769.             put_bits(pb, v_bits, v_codes);

  770. 

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

  772.             v_bits  = p_bits[curidx2] + count2;

  773.             put_bits(pb, v_bits, v_codes);

  774. 

  775.             if (out) {

  776.                vec1 = &p_vectors[curidx*2 ];

  777.                vec2 = &p_vectors[curidx2*2];

  778.                out[i+0] = copysignf(vec1[0] * IQ, in[i+0]);

  779.                out[i+1] = copysignf(vec1[1] * IQ, in[i+1]);

  780.                out[i+2] = copysignf(vec2[0] * IQ, in[i+2]);

  781.                out[i+3] = copysignf(vec2[1] * IQ, in[i+3]);

  782.             }

  783.         }

  784.     } else {

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

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

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

  788.             uint8_t v_bits;

  789.             unsigned int v_codes;

  790.             int c1, c2, c3, c4;

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

  792.             const float *vec1, *vec2;

  793. 

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

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

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

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

  798. 

  799.             __asm__ volatile (

  800.                 ".set push                                  \n\t"

  801.                 ".set noreorder                             \n\t"

  802. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  844. 

  845.                 ".set pop                                   \n\t"

  846. 

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

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

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

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

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

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

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

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

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

  856.                 : "memory"

  857.             );

  858. 

  859.             curidx = 17 * qc1;

  860.             curidx += qc2;

  861. 

  862.             curidx2 = 17 * qc3;

  863.             curidx2 += qc4;

  864. 

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

  866.             v_bits  = p_bits[curidx] + count1;

  867.             put_bits(pb, v_bits, v_codes);

  868. 

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

  870.                 int len = av_log2(c1);

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

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

  873.             }

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

  875.                 int len = av_log2(c2);

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

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

  878.             }

  879. 

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

  881.             v_bits  = p_bits[curidx2] + count2;

  882.             put_bits(pb, v_bits, v_codes);

  883. 

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

  885.                 int len = av_log2(c3);

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

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

  888.             }

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

  890.                 int len = av_log2(c4);

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

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

  893.             }

  894. 

  895.             if (out) {

  896.                vec1 = &p_vectors[curidx*2];

  897.                vec2 = &p_vectors[curidx2*2];

  898.                out[i+0] = copysignf(c1 * cbrtf(c1) * IQ, in[i+0]);

  899.                out[i+1] = copysignf(c2 * cbrtf(c2) * IQ, in[i+1]);

  900.                out[i+2] = copysignf(c3 * cbrtf(c3) * IQ, in[i+2]);

  901.                out[i+3] = copysignf(c4 * cbrtf(c4) * IQ, in[i+3]);

  902.             }

  903.         }

  904.     }

  905. }

  906. 

  907. static void quantize_and_encode_band_cost_NONE_mips(struct AACEncContext *s,

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

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

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

  911.                                                          int *bits, const float ROUNDING) {

  912.     av_assert0(0);

  913. }

  914. 

  915. static void quantize_and_encode_band_cost_ZERO_mips(struct AACEncContext *s,

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

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

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

  919.                                                          int *bits, const float ROUNDING) {

  920.     int i;

  921.     if (bits)

  922.         *bits = 0;

  923.     if (out) {

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

  925.            out[i  ] = 0.0f;

  926.            out[i+1] = 0.0f;

  927.            out[i+2] = 0.0f;

  928.            out[i+3] = 0.0f;

  929.         }

  930.     }

  931. }

  932. 

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

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

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

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

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

  938.     quantize_and_encode_band_cost_ZERO_mips,

  939.     quantize_and_encode_band_cost_SQUAD_mips,

  940.     quantize_and_encode_band_cost_SQUAD_mips,

  941.     quantize_and_encode_band_cost_UQUAD_mips,

  942.     quantize_and_encode_band_cost_UQUAD_mips,

  943.     quantize_and_encode_band_cost_SPAIR_mips,

  944.     quantize_and_encode_band_cost_SPAIR_mips,

  945.     quantize_and_encode_band_cost_UPAIR7_mips,

  946.     quantize_and_encode_band_cost_UPAIR7_mips,

  947.     quantize_and_encode_band_cost_UPAIR12_mips,

  948.     quantize_and_encode_band_cost_UPAIR12_mips,

  949.     quantize_and_encode_band_cost_ESC_mips,

  950.     quantize_and_encode_band_cost_NONE_mips, /* cb 12 doesn't exist */

  951.     quantize_and_encode_band_cost_ZERO_mips,

  952.     quantize_and_encode_band_cost_ZERO_mips,

  953.     quantize_and_encode_band_cost_ZERO_mips,

  954. };

  955. 

  956. #define quantize_and_encode_band_cost(                                       \

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

  958.                                 lambda, uplim, bits, ROUNDING)               \

  959.     quantize_and_encode_band_cost_arr[cb](                                   \

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

  961.                                 lambda, uplim, bits, ROUNDING)

  962. 

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

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

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

  966. {

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

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

  969. }

  970. 

  971. /**

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

  973.  */

  974. static float get_band_numbits_ZERO_mips(struct AACEncContext *s,

  975.                                         PutBitContext *pb, const float *in,

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

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

  978.                                         int *bits)

  979. {

  980.     return 0;

  981. }

  982. 

  983. static float get_band_numbits_NONE_mips(struct AACEncContext *s,

  984.                                         PutBitContext *pb, const float *in,

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

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

  987.                                         int *bits)

  988. {

  989.     av_assert0(0);

  990.     return 0;

  991. }

  992. 

  993. static float get_band_numbits_SQUAD_mips(struct AACEncContext *s,

  994.                                          PutBitContext *pb, const float *in,

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

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

  997.                                          int *bits)

  998. {

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

  1000.     int i;

  1001.     int qc1, qc2, qc3, qc4;

  1002.     int curbits = 0;

  1003. 

  1004.     uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1005. 

  1006.     for (i = 0; i < size; i += 4) {

  1007.         int curidx;

  1008.         int *in_int = (int *)&in[i];

  1009.         int t0, t1, t2, t3, t4, t5, t6, t7;

  1010. 

  1011.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1012.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1013.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1014.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1015. 

  1016.         __asm__ volatile (

  1017.             ".set push                      \n\t"

  1018.             ".set noreorder                 \n\t"

  1019. 

  1020.             "slt    %[qc1], $zero,  %[qc1]  \n\t"

  1021.             "slt    %[qc2], $zero,  %[qc2]  \n\t"

  1022.             "slt    %[qc3], $zero,  %[qc3]  \n\t"

  1023.             "slt    %[qc4], $zero,  %[qc4]  \n\t"

  1024.             "lw     %[t0],  0(%[in_int])    \n\t"

  1025.             "lw     %[t1],  4(%[in_int])    \n\t"

  1026.             "lw     %[t2],  8(%[in_int])    \n\t"

  1027.             "lw     %[t3],  12(%[in_int])   \n\t"

  1028.             "srl    %[t0],  %[t0],  31      \n\t"

  1029.             "srl    %[t1],  %[t1],  31      \n\t"

  1030.             "srl    %[t2],  %[t2],  31      \n\t"

  1031.             "srl    %[t3],  %[t3],  31      \n\t"

  1032.             "subu   %[t4],  $zero,  %[qc1]  \n\t"

  1033.             "subu   %[t5],  $zero,  %[qc2]  \n\t"

  1034.             "subu   %[t6],  $zero,  %[qc3]  \n\t"

  1035.             "subu   %[t7],  $zero,  %[qc4]  \n\t"

  1036.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1037.             "movn   %[qc2], %[t5],  %[t1]   \n\t"

  1038.             "movn   %[qc3], %[t6],  %[t2]   \n\t"

  1039.             "movn   %[qc4], %[t7],  %[t3]   \n\t"

  1040. 

  1041.             ".set pop                       \n\t"

  1042. 

  1043.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1044.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1045.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1046.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  1047.             : [in_int]"r"(in_int)

  1048.             : "memory"

  1049.         );

  1050. 

  1051.         curidx = qc1;

  1052.         curidx *= 3;

  1053.         curidx += qc2;

  1054.         curidx *= 3;

  1055.         curidx += qc3;

  1056.         curidx *= 3;

  1057.         curidx += qc4;

  1058.         curidx += 40;

  1059. 

  1060.         curbits += p_bits[curidx];

  1061.     }

  1062.     return curbits;

  1063. }

  1064. 

  1065. static float get_band_numbits_UQUAD_mips(struct AACEncContext *s,

  1066.                                          PutBitContext *pb, const float *in,

  1067.                                          const float *scaled, int size, int scale_idx,

  1068.                                          int cb, const float lambda, const float uplim,

  1069.                                          int *bits)

  1070. {

  1071.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1072.     int i;

  1073.     int curbits = 0;

  1074.     int qc1, qc2, qc3, qc4;

  1075. 

  1076.     uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1077. 

  1078.     for (i = 0; i < size; i += 4) {

  1079.         int curidx;

  1080.         int t0, t1, t2, t3, t4;

  1081. 

  1082.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1083.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1084.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1085.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1086. 

  1087.         __asm__ volatile (

  1088.             ".set push                      \n\t"

  1089.             ".set noreorder                 \n\t"

  1090. 

  1091.             "ori    %[t4],  $zero,  2       \n\t"

  1092.             "slt    %[t0],  %[t4],  %[qc1]  \n\t"

  1093.             "slt    %[t1],  %[t4],  %[qc2]  \n\t"

  1094.             "slt    %[t2],  %[t4],  %[qc3]  \n\t"

  1095.             "slt    %[t3],  %[t4],  %[qc4]  \n\t"

  1096.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1097.             "movn   %[qc2], %[t4],  %[t1]   \n\t"

  1098.             "movn   %[qc3], %[t4],  %[t2]   \n\t"

  1099.             "movn   %[qc4], %[t4],  %[t3]   \n\t"

  1100. 

  1101.             ".set pop                       \n\t"

  1102. 

  1103.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1104.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1105.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1106.               [t4]"=&r"(t4)

  1107.         );

  1108. 

  1109.         curidx = qc1;

  1110.         curidx *= 3;

  1111.         curidx += qc2;

  1112.         curidx *= 3;

  1113.         curidx += qc3;

  1114.         curidx *= 3;

  1115.         curidx += qc4;

  1116. 

  1117.         curbits += p_bits[curidx];

  1118.         curbits += uquad_sign_bits[curidx];

  1119.     }

  1120.     return curbits;

  1121. }

  1122. 

  1123. static float get_band_numbits_SPAIR_mips(struct AACEncContext *s,

  1124.                                          PutBitContext *pb, const float *in,

  1125.                                          const float *scaled, int size, int scale_idx,

  1126.                                          int cb, const float lambda, const float uplim,

  1127.                                          int *bits)

  1128. {

  1129.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1130.     int i;

  1131.     int qc1, qc2, qc3, qc4;

  1132.     int curbits = 0;

  1133. 

  1134.     uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1];

  1135. 

  1136.     for (i = 0; i < size; i += 4) {

  1137.         int curidx, curidx2;

  1138.         int *in_int = (int *)&in[i];

  1139.         int t0, t1, t2, t3, t4, t5, t6, t7;

  1140. 

  1141.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1142.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1143.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1144.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1145. 

  1146.         __asm__ volatile (

  1147.             ".set push                      \n\t"

  1148.             ".set noreorder                 \n\t"

  1149. 

  1150.             "ori    %[t4],  $zero,  4       \n\t"

  1151.             "slt    %[t0],  %[t4],  %[qc1]  \n\t"

  1152.             "slt    %[t1],  %[t4],  %[qc2]  \n\t"

  1153.             "slt    %[t2],  %[t4],  %[qc3]  \n\t"

  1154.             "slt    %[t3],  %[t4],  %[qc4]  \n\t"

  1155.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1156.             "movn   %[qc2], %[t4],  %[t1]   \n\t"

  1157.             "movn   %[qc3], %[t4],  %[t2]   \n\t"

  1158.             "movn   %[qc4], %[t4],  %[t3]   \n\t"

  1159.             "lw     %[t0],  0(%[in_int])    \n\t"

  1160.             "lw     %[t1],  4(%[in_int])    \n\t"

  1161.             "lw     %[t2],  8(%[in_int])    \n\t"

  1162.             "lw     %[t3],  12(%[in_int])   \n\t"

  1163.             "srl    %[t0],  %[t0],  31      \n\t"

  1164.             "srl    %[t1],  %[t1],  31      \n\t"

  1165.             "srl    %[t2],  %[t2],  31      \n\t"

  1166.             "srl    %[t3],  %[t3],  31      \n\t"

  1167.             "subu   %[t4],  $zero,  %[qc1]  \n\t"

  1168.             "subu   %[t5],  $zero,  %[qc2]  \n\t"

  1169.             "subu   %[t6],  $zero,  %[qc3]  \n\t"

  1170.             "subu   %[t7],  $zero,  %[qc4]  \n\t"

  1171.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1172.             "movn   %[qc2], %[t5],  %[t1]   \n\t"

  1173.             "movn   %[qc3], %[t6],  %[t2]   \n\t"

  1174.             "movn   %[qc4], %[t7],  %[t3]   \n\t"

  1175. 

  1176.             ".set pop                       \n\t"

  1177. 

  1178.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1179.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1180.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1181.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  1182.             : [in_int]"r"(in_int)

  1183.             : "memory"

  1184.         );

  1185. 

  1186.         curidx  = 9 * qc1;

  1187.         curidx += qc2 + 40;

  1188. 

  1189.         curidx2  = 9 * qc3;

  1190.         curidx2 += qc4 + 40;

  1191. 

  1192.         curbits += p_bits[curidx] + p_bits[curidx2];

  1193.     }

  1194.     return curbits;

  1195. }

  1196. 

  1197. static float get_band_numbits_UPAIR7_mips(struct AACEncContext *s,

  1198.                                           PutBitContext *pb, const float *in,

  1199.                                           const float *scaled, int size, int scale_idx,

  1200.                                           int cb, const float lambda, const float uplim,

  1201.                                           int *bits)

  1202. {

  1203.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1204.     int i;

  1205.     int qc1, qc2, qc3, qc4;

  1206.     int curbits = 0;

  1207. 

  1208.     uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1209. 

  1210.     for (i = 0; i < size; i += 4) {

  1211.         int curidx, curidx2;

  1212.         int t0, t1, t2, t3, t4;

  1213. 

  1214.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1215.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1216.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1217.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1218. 

  1219.         __asm__ volatile (

  1220.             ".set push                      \n\t"

  1221.             ".set noreorder                 \n\t"

  1222. 

  1223.             "ori    %[t4],  $zero,  7       \n\t"

  1224.             "slt    %[t0],  %[t4],  %[qc1]  \n\t"

  1225.             "slt    %[t1],  %[t4],  %[qc2]  \n\t"

  1226.             "slt    %[t2],  %[t4],  %[qc3]  \n\t"

  1227.             "slt    %[t3],  %[t4],  %[qc4]  \n\t"

  1228.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1229.             "movn   %[qc2], %[t4],  %[t1]   \n\t"

  1230.             "movn   %[qc3], %[t4],  %[t2]   \n\t"

  1231.             "movn   %[qc4], %[t4],  %[t3]   \n\t"

  1232. 

  1233.             ".set pop                       \n\t"

  1234. 

  1235.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1236.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1237.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1238.               [t4]"=&r"(t4)

  1239.         );

  1240. 

  1241.         curidx  = 8 * qc1;

  1242.         curidx += qc2;

  1243. 

  1244.         curidx2  = 8 * qc3;

  1245.         curidx2 += qc4;

  1246. 

  1247.         curbits += p_bits[curidx] +

  1248.                    upair7_sign_bits[curidx] +

  1249.                    p_bits[curidx2] +

  1250.                    upair7_sign_bits[curidx2];

  1251.     }

  1252.     return curbits;

  1253. }

  1254. 

  1255. static float get_band_numbits_UPAIR12_mips(struct AACEncContext *s,

  1256.                                            PutBitContext *pb, const float *in,

  1257.                                            const float *scaled, int size, int scale_idx,

  1258.                                            int cb, const float lambda, const float uplim,

  1259.                                            int *bits)

  1260. {

  1261.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1262.     int i;

  1263.     int qc1, qc2, qc3, qc4;

  1264.     int curbits = 0;

  1265. 

  1266.     uint8_t *p_bits = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1267. 

  1268.     for (i = 0; i < size; i += 4) {

  1269.         int curidx, curidx2;

  1270.         int t0, t1, t2, t3, t4;

  1271. 

  1272.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1273.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1274.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1275.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1276. 

  1277.         __asm__ volatile (

  1278.             ".set push                      \n\t"

  1279.             ".set noreorder                 \n\t"

  1280. 

  1281.             "ori    %[t4],  $zero,  12      \n\t"

  1282.             "slt    %[t0],  %[t4],  %[qc1]  \n\t"

  1283.             "slt    %[t1],  %[t4],  %[qc2]  \n\t"

  1284.             "slt    %[t2],  %[t4],  %[qc3]  \n\t"

  1285.             "slt    %[t3],  %[t4],  %[qc4]  \n\t"

  1286.             "movn   %[qc1], %[t4],  %[t0]   \n\t"

  1287.             "movn   %[qc2], %[t4],  %[t1]   \n\t"

  1288.             "movn   %[qc3], %[t4],  %[t2]   \n\t"

  1289.             "movn   %[qc4], %[t4],  %[t3]   \n\t"

  1290. 

  1291.             ".set pop                       \n\t"

  1292. 

  1293.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1294.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1295.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1296.               [t4]"=&r"(t4)

  1297.         );

  1298. 

  1299.         curidx  = 13 * qc1;

  1300.         curidx += qc2;

  1301. 

  1302.         curidx2  = 13 * qc3;

  1303.         curidx2 += qc4;

  1304. 

  1305.         curbits += p_bits[curidx] +

  1306.                    p_bits[curidx2] +

  1307.                    upair12_sign_bits[curidx] +

  1308.                    upair12_sign_bits[curidx2];

  1309.     }

  1310.     return curbits;

  1311. }

  1312. 

  1313. static float get_band_numbits_ESC_mips(struct AACEncContext *s,

  1314.                                        PutBitContext *pb, const float *in,

  1315.                                        const float *scaled, int size, int scale_idx,

  1316.                                        int cb, const float lambda, const float uplim,

  1317.                                        int *bits)

  1318. {

  1319.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1320.     int i;

  1321.     int qc1, qc2, qc3, qc4;

  1322.     int curbits = 0;

  1323. 

  1324.     uint8_t *p_bits = (uint8_t*)ff_aac_spectral_bits[cb-1];

  1325. 

  1326.     for (i = 0; i < size; i += 4) {

  1327.         int curidx, curidx2;

  1328.         int cond0, cond1, cond2, cond3;

  1329.         int c1, c2, c3, c4;

  1330.         int t4, t5;

  1331. 

  1332.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1333.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1334.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1335.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1336. 

  1337.         __asm__ volatile (

  1338.             ".set push                                  \n\t"

  1339.             ".set noreorder                             \n\t"

  1340. 

  1341.             "ori        %[t4],      $zero,  15          \n\t"

  1342.             "ori        %[t5],      $zero,  16          \n\t"

  1343.             "shll_s.w   %[c1],      %[qc1], 18          \n\t"

  1344.             "shll_s.w   %[c2],      %[qc2], 18          \n\t"

  1345.             "shll_s.w   %[c3],      %[qc3], 18          \n\t"

  1346.             "shll_s.w   %[c4],      %[qc4], 18          \n\t"

  1347.             "srl        %[c1],      %[c1],  18          \n\t"

  1348.             "srl        %[c2],      %[c2],  18          \n\t"

  1349.             "srl        %[c3],      %[c3],  18          \n\t"

  1350.             "srl        %[c4],      %[c4],  18          \n\t"

  1351.             "slt        %[cond0],   %[t4],  %[qc1]      \n\t"

  1352.             "slt        %[cond1],   %[t4],  %[qc2]      \n\t"

  1353.             "slt        %[cond2],   %[t4],  %[qc3]      \n\t"

  1354.             "slt        %[cond3],   %[t4],  %[qc4]      \n\t"

  1355.             "movn       %[qc1],     %[t5],  %[cond0]    \n\t"

  1356.             "movn       %[qc2],     %[t5],  %[cond1]    \n\t"

  1357.             "movn       %[qc3],     %[t5],  %[cond2]    \n\t"

  1358.             "movn       %[qc4],     %[t5],  %[cond3]    \n\t"

  1359.             "ori        %[t5],      $zero,  31          \n\t"

  1360.             "clz        %[c1],      %[c1]               \n\t"

  1361.             "clz        %[c2],      %[c2]               \n\t"

  1362.             "clz        %[c3],      %[c3]               \n\t"

  1363.             "clz        %[c4],      %[c4]               \n\t"

  1364.             "subu       %[c1],      %[t5],  %[c1]       \n\t"

  1365.             "subu       %[c2],      %[t5],  %[c2]       \n\t"

  1366.             "subu       %[c3],      %[t5],  %[c3]       \n\t"

  1367.             "subu       %[c4],      %[t5],  %[c4]       \n\t"

  1368.             "sll        %[c1],      %[c1],  1           \n\t"

  1369.             "sll        %[c2],      %[c2],  1           \n\t"

  1370.             "sll        %[c3],      %[c3],  1           \n\t"

  1371.             "sll        %[c4],      %[c4],  1           \n\t"

  1372.             "addiu      %[c1],      %[c1],  -3          \n\t"

  1373.             "addiu      %[c2],      %[c2],  -3          \n\t"

  1374.             "addiu      %[c3],      %[c3],  -3          \n\t"

  1375.             "addiu      %[c4],      %[c4],  -3          \n\t"

  1376.             "subu       %[cond0],   $zero,  %[cond0]    \n\t"

  1377.             "subu       %[cond1],   $zero,  %[cond1]    \n\t"

  1378.             "subu       %[cond2],   $zero,  %[cond2]    \n\t"

  1379.             "subu       %[cond3],   $zero,  %[cond3]    \n\t"

  1380.             "and        %[c1],      %[c1],  %[cond0]    \n\t"

  1381.             "and        %[c2],      %[c2],  %[cond1]    \n\t"

  1382.             "and        %[c3],      %[c3],  %[cond2]    \n\t"

  1383.             "and        %[c4],      %[c4],  %[cond3]    \n\t"

  1384. 

  1385.             ".set pop                                   \n\t"

  1386. 

  1387.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1388.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1389.               [cond0]"=&r"(cond0), [cond1]"=&r"(cond1),

  1390.               [cond2]"=&r"(cond2), [cond3]"=&r"(cond3),

  1391.               [c1]"=&r"(c1), [c2]"=&r"(c2),

  1392.               [c3]"=&r"(c3), [c4]"=&r"(c4),

  1393.               [t4]"=&r"(t4), [t5]"=&r"(t5)

  1394.         );

  1395. 

  1396.         curidx = 17 * qc1;

  1397.         curidx += qc2;

  1398. 

  1399.         curidx2 = 17 * qc3;

  1400.         curidx2 += qc4;

  1401. 

  1402.         curbits += p_bits[curidx];

  1403.         curbits += esc_sign_bits[curidx];

  1404.         curbits += p_bits[curidx2];

  1405.         curbits += esc_sign_bits[curidx2];

  1406. 

  1407.         curbits += c1;

  1408.         curbits += c2;

  1409.         curbits += c3;

  1410.         curbits += c4;

  1411.     }

  1412.     return curbits;

  1413. }

  1414. 

  1415. static float (*const get_band_numbits_arr[])(struct AACEncContext *s,

  1416.                                              PutBitContext *pb, const float *in,

  1417.                                              const float *scaled, int size, int scale_idx,

  1418.                                              int cb, const float lambda, const float uplim,

  1419.                                              int *bits) = {

  1420.     get_band_numbits_ZERO_mips,

  1421.     get_band_numbits_SQUAD_mips,

  1422.     get_band_numbits_SQUAD_mips,

  1423.     get_band_numbits_UQUAD_mips,

  1424.     get_band_numbits_UQUAD_mips,

  1425.     get_band_numbits_SPAIR_mips,

  1426.     get_band_numbits_SPAIR_mips,

  1427.     get_band_numbits_UPAIR7_mips,

  1428.     get_band_numbits_UPAIR7_mips,

  1429.     get_band_numbits_UPAIR12_mips,

  1430.     get_band_numbits_UPAIR12_mips,

  1431.     get_band_numbits_ESC_mips,

  1432.     get_band_numbits_NONE_mips, /* cb 12 doesn't exist */

  1433.     get_band_numbits_ZERO_mips,

  1434.     get_band_numbits_ZERO_mips,

  1435.     get_band_numbits_ZERO_mips,

  1436. };

  1437. 

  1438. #define get_band_numbits(                                  \

  1439.                                 s, pb, in, scaled, size, scale_idx, cb, \

  1440.                                 lambda, uplim, bits)                    \

  1441.     get_band_numbits_arr[cb](                              \

  1442.                                 s, pb, in, scaled, size, scale_idx, cb, \

  1443.                                 lambda, uplim, bits)

  1444. 

  1445. static float quantize_band_cost_bits(struct AACEncContext *s, const float *in,

  1446.                                      const float *scaled, int size, int scale_idx,

  1447.                                      int cb, const float lambda, const float uplim,

  1448.                                      int *bits, int rtz)

  1449. {

  1450.     return get_band_numbits(s, NULL, in, scaled, size, scale_idx, cb, lambda, uplim, bits);

  1451. }

  1452. 

  1453. /**

  1454.  * Functions developed from template function and optimized for getting the band cost

  1455.  */

  1456. #if HAVE_MIPSFPU

  1457. static float get_band_cost_ZERO_mips(struct AACEncContext *s,

  1458.                                      PutBitContext *pb, const float *in,

  1459.                                      const float *scaled, int size, int scale_idx,

  1460.                                      int cb, const float lambda, const float uplim,

  1461.                                      int *bits)

  1462. {

  1463.     int i;

  1464.     float cost = 0;

  1465. 

  1466.     for (i = 0; i < size; i += 4) {

  1467.         cost += in[i  ] * in[i  ];

  1468.         cost += in[i+1] * in[i+1];

  1469.         cost += in[i+2] * in[i+2];

  1470.         cost += in[i+3] * in[i+3];

  1471.     }

  1472.     if (bits)

  1473.         *bits = 0;

  1474.     return cost * lambda;

  1475. }

  1476. 

  1477. static float get_band_cost_NONE_mips(struct AACEncContext *s,

  1478.                                      PutBitContext *pb, const float *in,

  1479.                                      const float *scaled, int size, int scale_idx,

  1480.                                      int cb, const float lambda, const float uplim,

  1481.                                      int *bits)

  1482. {

  1483.     av_assert0(0);

  1484.     return 0;

  1485. }

  1486. 

  1487. static float get_band_cost_SQUAD_mips(struct AACEncContext *s,

  1488.                                       PutBitContext *pb, const float *in,

  1489.                                       const float *scaled, int size, int scale_idx,

  1490.                                       int cb, const float lambda, const float uplim,

  1491.                                       int *bits)

  1492. {

  1493.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1494.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1495.     int i;

  1496.     float cost = 0;

  1497.     int qc1, qc2, qc3, qc4;

  1498.     int curbits = 0;

  1499. 

  1500.     uint8_t *p_bits  = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1501.     float   *p_codes = (float   *)ff_aac_codebook_vectors[cb-1];

  1502. 

  1503.     for (i = 0; i < size; i += 4) {

  1504.         const float *vec;

  1505.         int curidx;

  1506.         int   *in_int = (int   *)&in[i];

  1507.         float *in_pos = (float *)&in[i];

  1508.         float di0, di1, di2, di3;

  1509.         int t0, t1, t2, t3, t4, t5, t6, t7;

  1510. 

  1511.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1512.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1513.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1514.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1515. 

  1516.         __asm__ volatile (

  1517.             ".set push                                  \n\t"

  1518.             ".set noreorder                             \n\t"

  1519. 

  1520.             "slt        %[qc1], $zero,  %[qc1]          \n\t"

  1521.             "slt        %[qc2], $zero,  %[qc2]          \n\t"

  1522.             "slt        %[qc3], $zero,  %[qc3]          \n\t"

  1523.             "slt        %[qc4], $zero,  %[qc4]          \n\t"

  1524.             "lw         %[t0],  0(%[in_int])            \n\t"

  1525.             "lw         %[t1],  4(%[in_int])            \n\t"

  1526.             "lw         %[t2],  8(%[in_int])            \n\t"

  1527.             "lw         %[t3],  12(%[in_int])           \n\t"

  1528.             "srl        %[t0],  %[t0],  31              \n\t"

  1529.             "srl        %[t1],  %[t1],  31              \n\t"

  1530.             "srl        %[t2],  %[t2],  31              \n\t"

  1531.             "srl        %[t3],  %[t3],  31              \n\t"

  1532.             "subu       %[t4],  $zero,  %[qc1]          \n\t"

  1533.             "subu       %[t5],  $zero,  %[qc2]          \n\t"

  1534.             "subu       %[t6],  $zero,  %[qc3]          \n\t"

  1535.             "subu       %[t7],  $zero,  %[qc4]          \n\t"

  1536.             "movn       %[qc1], %[t4],  %[t0]           \n\t"

  1537.             "movn       %[qc2], %[t5],  %[t1]           \n\t"

  1538.             "movn       %[qc3], %[t6],  %[t2]           \n\t"

  1539.             "movn       %[qc4], %[t7],  %[t3]           \n\t"

  1540. 

  1541.             ".set pop                                   \n\t"

  1542. 

  1543.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1544.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1545.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1546.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  1547.             : [in_int]"r"(in_int)

  1548.             : "memory"

  1549.         );

  1550. 

  1551.         curidx = qc1;

  1552.         curidx *= 3;

  1553.         curidx += qc2;

  1554.         curidx *= 3;

  1555.         curidx += qc3;

  1556.         curidx *= 3;

  1557.         curidx += qc4;

  1558.         curidx += 40;

  1559. 

  1560.         curbits += p_bits[curidx];

  1561.         vec     = &p_codes[curidx*4];

  1562. 

  1563.         __asm__ volatile (

  1564.             ".set push                                  \n\t"

  1565.             ".set noreorder                             \n\t"

  1566. 

  1567.             "lwc1       $f0,    0(%[in_pos])            \n\t"

  1568.             "lwc1       $f1,    0(%[vec])               \n\t"

  1569.             "lwc1       $f2,    4(%[in_pos])            \n\t"

  1570.             "lwc1       $f3,    4(%[vec])               \n\t"

  1571.             "lwc1       $f4,    8(%[in_pos])            \n\t"

  1572.             "lwc1       $f5,    8(%[vec])               \n\t"

  1573.             "lwc1       $f6,    12(%[in_pos])           \n\t"

  1574.             "lwc1       $f7,    12(%[vec])              \n\t"

  1575.             "nmsub.s    %[di0], $f0,    $f1,    %[IQ]   \n\t"

  1576.             "nmsub.s    %[di1], $f2,    $f3,    %[IQ]   \n\t"

  1577.             "nmsub.s    %[di2], $f4,    $f5,    %[IQ]   \n\t"

  1578.             "nmsub.s    %[di3], $f6,    $f7,    %[IQ]   \n\t"

  1579. 

  1580.             ".set pop                                   \n\t"

  1581. 

  1582.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  1583.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  1584.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  1585.               [IQ]"f"(IQ)

  1586.             : "$f0", "$f1", "$f2", "$f3",

  1587.               "$f4", "$f5", "$f6", "$f7",

  1588.               "memory"

  1589.         );

  1590. 

  1591.         cost += di0 * di0 + di1 * di1

  1592.                 + di2 * di2 + di3 * di3;

  1593.     }

  1594. 

  1595.     if (bits)

  1596.         *bits = curbits;

  1597.     return cost * lambda + curbits;

  1598. }

  1599. 

  1600. static float get_band_cost_UQUAD_mips(struct AACEncContext *s,

  1601.                                       PutBitContext *pb, const float *in,

  1602.                                       const float *scaled, int size, int scale_idx,

  1603.                                       int cb, const float lambda, const float uplim,

  1604.                                       int *bits)

  1605. {

  1606.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1607.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1608.     int i;

  1609.     float cost = 0;

  1610.     int curbits = 0;

  1611.     int qc1, qc2, qc3, qc4;

  1612. 

  1613.     uint8_t *p_bits  = (uint8_t*)ff_aac_spectral_bits[cb-1];

  1614.     float   *p_codes = (float  *)ff_aac_codebook_vectors[cb-1];

  1615. 

  1616.     for (i = 0; i < size; i += 4) {

  1617.         const float *vec;

  1618.         int curidx;

  1619.         float *in_pos = (float *)&in[i];

  1620.         float di0, di1, di2, di3;

  1621.         int t0, t1, t2, t3, t4;

  1622. 

  1623.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1624.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1625.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1626.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1627. 

  1628.         __asm__ volatile (

  1629.             ".set push                                  \n\t"

  1630.             ".set noreorder                             \n\t"

  1631. 

  1632.             "ori        %[t4],  $zero,  2               \n\t"

  1633.             "slt        %[t0],  %[t4],  %[qc1]          \n\t"

  1634.             "slt        %[t1],  %[t4],  %[qc2]          \n\t"

  1635.             "slt        %[t2],  %[t4],  %[qc3]          \n\t"

  1636.             "slt        %[t3],  %[t4],  %[qc4]          \n\t"

  1637.             "movn       %[qc1], %[t4],  %[t0]           \n\t"

  1638.             "movn       %[qc2], %[t4],  %[t1]           \n\t"

  1639.             "movn       %[qc3], %[t4],  %[t2]           \n\t"

  1640.             "movn       %[qc4], %[t4],  %[t3]           \n\t"

  1641. 

  1642.             ".set pop                                   \n\t"

  1643. 

  1644.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1645.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1646.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1647.               [t4]"=&r"(t4)

  1648.         );

  1649. 

  1650.         curidx = qc1;

  1651.         curidx *= 3;

  1652.         curidx += qc2;

  1653.         curidx *= 3;

  1654.         curidx += qc3;

  1655.         curidx *= 3;

  1656.         curidx += qc4;

  1657. 

  1658.         curbits += p_bits[curidx];

  1659.         curbits += uquad_sign_bits[curidx];

  1660.         vec     = &p_codes[curidx*4];

  1661. 

  1662.         __asm__ volatile (

  1663.             ".set push                                  \n\t"

  1664.             ".set noreorder                             \n\t"

  1665. 

  1666.             "lwc1       %[di0], 0(%[in_pos])            \n\t"

  1667.             "lwc1       %[di1], 4(%[in_pos])            \n\t"

  1668.             "lwc1       %[di2], 8(%[in_pos])            \n\t"

  1669.             "lwc1       %[di3], 12(%[in_pos])           \n\t"

  1670.             "abs.s      %[di0], %[di0]                  \n\t"

  1671.             "abs.s      %[di1], %[di1]                  \n\t"

  1672.             "abs.s      %[di2], %[di2]                  \n\t"

  1673.             "abs.s      %[di3], %[di3]                  \n\t"

  1674.             "lwc1       $f0,    0(%[vec])               \n\t"

  1675.             "lwc1       $f1,    4(%[vec])               \n\t"

  1676.             "lwc1       $f2,    8(%[vec])               \n\t"

  1677.             "lwc1       $f3,    12(%[vec])              \n\t"

  1678.             "nmsub.s    %[di0], %[di0], $f0,    %[IQ]   \n\t"

  1679.             "nmsub.s    %[di1], %[di1], $f1,    %[IQ]   \n\t"

  1680.             "nmsub.s    %[di2], %[di2], $f2,    %[IQ]   \n\t"

  1681.             "nmsub.s    %[di3], %[di3], $f3,    %[IQ]   \n\t"

  1682. 

  1683.             ".set pop                                   \n\t"

  1684. 

  1685.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  1686.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  1687.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  1688.               [IQ]"f"(IQ)

  1689.             : "$f0", "$f1", "$f2", "$f3",

  1690.               "memory"

  1691.         );

  1692. 

  1693.         cost += di0 * di0 + di1 * di1

  1694.                 + di2 * di2 + di3 * di3;

  1695.     }

  1696. 

  1697.     if (bits)

  1698.         *bits = curbits;

  1699.     return cost * lambda + curbits;

  1700. }

  1701. 

  1702. static float get_band_cost_SPAIR_mips(struct AACEncContext *s,

  1703.                                       PutBitContext *pb, const float *in,

  1704.                                       const float *scaled, int size, int scale_idx,

  1705.                                       int cb, const float lambda, const float uplim,

  1706.                                       int *bits)

  1707. {

  1708.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1709.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1710.     int i;

  1711.     float cost = 0;

  1712.     int qc1, qc2, qc3, qc4;

  1713.     int curbits = 0;

  1714. 

  1715.     uint8_t *p_bits  = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1716.     float   *p_codes = (float   *)ff_aac_codebook_vectors[cb-1];

  1717. 

  1718.     for (i = 0; i < size; i += 4) {

  1719.         const float *vec, *vec2;

  1720.         int curidx, curidx2;

  1721.         int   *in_int = (int   *)&in[i];

  1722.         float *in_pos = (float *)&in[i];

  1723.         float di0, di1, di2, di3;

  1724.         int t0, t1, t2, t3, t4, t5, t6, t7;

  1725. 

  1726.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1727.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1728.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1729.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1730. 

  1731.         __asm__ volatile (

  1732.             ".set push                                  \n\t"

  1733.             ".set noreorder                             \n\t"

  1734. 

  1735.             "ori        %[t4],  $zero,  4               \n\t"

  1736.             "slt        %[t0],  %[t4],  %[qc1]          \n\t"

  1737.             "slt        %[t1],  %[t4],  %[qc2]          \n\t"

  1738.             "slt        %[t2],  %[t4],  %[qc3]          \n\t"

  1739.             "slt        %[t3],  %[t4],  %[qc4]          \n\t"

  1740.             "movn       %[qc1], %[t4],  %[t0]           \n\t"

  1741.             "movn       %[qc2], %[t4],  %[t1]           \n\t"

  1742.             "movn       %[qc3], %[t4],  %[t2]           \n\t"

  1743.             "movn       %[qc4], %[t4],  %[t3]           \n\t"

  1744.             "lw         %[t0],  0(%[in_int])            \n\t"

  1745.             "lw         %[t1],  4(%[in_int])            \n\t"

  1746.             "lw         %[t2],  8(%[in_int])            \n\t"

  1747.             "lw         %[t3],  12(%[in_int])           \n\t"

  1748.             "srl        %[t0],  %[t0],  31              \n\t"

  1749.             "srl        %[t1],  %[t1],  31              \n\t"

  1750.             "srl        %[t2],  %[t2],  31              \n\t"

  1751.             "srl        %[t3],  %[t3],  31              \n\t"

  1752.             "subu       %[t4],  $zero,  %[qc1]          \n\t"

  1753.             "subu       %[t5],  $zero,  %[qc2]          \n\t"

  1754.             "subu       %[t6],  $zero,  %[qc3]          \n\t"

  1755.             "subu       %[t7],  $zero,  %[qc4]          \n\t"

  1756.             "movn       %[qc1], %[t4],  %[t0]           \n\t"

  1757.             "movn       %[qc2], %[t5],  %[t1]           \n\t"

  1758.             "movn       %[qc3], %[t6],  %[t2]           \n\t"

  1759.             "movn       %[qc4], %[t7],  %[t3]           \n\t"

  1760. 

  1761.             ".set pop                                   \n\t"

  1762. 

  1763.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1764.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1765.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1766.               [t4]"=&r"(t4), [t5]"=&r"(t5), [t6]"=&r"(t6), [t7]"=&r"(t7)

  1767.             : [in_int]"r"(in_int)

  1768.             : "memory"

  1769.         );

  1770. 

  1771.         curidx = 9 * qc1;

  1772.         curidx += qc2 + 40;

  1773. 

  1774.         curidx2 = 9 * qc3;

  1775.         curidx2 += qc4 + 40;

  1776. 

  1777.         curbits += p_bits[curidx];

  1778.         curbits += p_bits[curidx2];

  1779. 

  1780.         vec     = &p_codes[curidx*2];

  1781.         vec2    = &p_codes[curidx2*2];

  1782. 

  1783.         __asm__ volatile (

  1784.             ".set push                                  \n\t"

  1785.             ".set noreorder                             \n\t"

  1786. 

  1787.             "lwc1       $f0,    0(%[in_pos])            \n\t"

  1788.             "lwc1       $f1,    0(%[vec])               \n\t"

  1789.             "lwc1       $f2,    4(%[in_pos])            \n\t"

  1790.             "lwc1       $f3,    4(%[vec])               \n\t"

  1791.             "lwc1       $f4,    8(%[in_pos])            \n\t"

  1792.             "lwc1       $f5,    0(%[vec2])              \n\t"

  1793.             "lwc1       $f6,    12(%[in_pos])           \n\t"

  1794.             "lwc1       $f7,    4(%[vec2])              \n\t"

  1795.             "nmsub.s    %[di0], $f0,    $f1,    %[IQ]   \n\t"

  1796.             "nmsub.s    %[di1], $f2,    $f3,    %[IQ]   \n\t"

  1797.             "nmsub.s    %[di2], $f4,    $f5,    %[IQ]   \n\t"

  1798.             "nmsub.s    %[di3], $f6,    $f7,    %[IQ]   \n\t"

  1799. 

  1800.             ".set pop                                   \n\t"

  1801. 

  1802.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  1803.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  1804.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  1805.               [vec2]"r"(vec2), [IQ]"f"(IQ)

  1806.             : "$f0", "$f1", "$f2", "$f3",

  1807.               "$f4", "$f5", "$f6", "$f7",

  1808.               "memory"

  1809.         );

  1810. 

  1811.         cost += di0 * di0 + di1 * di1

  1812.                 + di2 * di2 + di3 * di3;

  1813.     }

  1814. 

  1815.     if (bits)

  1816.         *bits = curbits;

  1817.     return cost * lambda + curbits;

  1818. }

  1819. 

  1820. static float get_band_cost_UPAIR7_mips(struct AACEncContext *s,

  1821.                                        PutBitContext *pb, const float *in,

  1822.                                        const float *scaled, int size, int scale_idx,

  1823.                                        int cb, const float lambda, const float uplim,

  1824.                                        int *bits)

  1825. {

  1826.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1827.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1828.     int i;

  1829.     float cost = 0;

  1830.     int qc1, qc2, qc3, qc4;

  1831.     int curbits = 0;

  1832. 

  1833.     uint8_t *p_bits  = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1834.     float   *p_codes = (float   *)ff_aac_codebook_vectors[cb-1];

  1835. 

  1836.     for (i = 0; i < size; i += 4) {

  1837.         const float *vec, *vec2;

  1838.         int curidx, curidx2, sign1, count1, sign2, count2;

  1839.         int   *in_int = (int   *)&in[i];

  1840.         float *in_pos = (float *)&in[i];

  1841.         float di0, di1, di2, di3;

  1842.         int t0, t1, t2, t3, t4;

  1843. 

  1844.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1845.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1846.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1847.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1848. 

  1849.         __asm__ volatile (

  1850.             ".set push                                          \n\t"

  1851.             ".set noreorder                                     \n\t"

  1852. 

  1853.             "ori        %[t4],      $zero,      7               \n\t"

  1854.             "ori        %[sign1],   $zero,      0               \n\t"

  1855.             "ori        %[sign2],   $zero,      0               \n\t"

  1856.             "slt        %[t0],      %[t4],      %[qc1]          \n\t"

  1857.             "slt        %[t1],      %[t4],      %[qc2]          \n\t"

  1858.             "slt        %[t2],      %[t4],      %[qc3]          \n\t"

  1859.             "slt        %[t3],      %[t4],      %[qc4]          \n\t"

  1860.             "movn       %[qc1],     %[t4],      %[t0]           \n\t"

  1861.             "movn       %[qc2],     %[t4],      %[t1]           \n\t"

  1862.             "movn       %[qc3],     %[t4],      %[t2]           \n\t"

  1863.             "movn       %[qc4],     %[t4],      %[t3]           \n\t"

  1864.             "lw         %[t0],      0(%[in_int])                \n\t"

  1865.             "lw         %[t1],      4(%[in_int])                \n\t"

  1866.             "lw         %[t2],      8(%[in_int])                \n\t"

  1867.             "lw         %[t3],      12(%[in_int])               \n\t"

  1868.             "slt        %[t0],      %[t0],      $zero           \n\t"

  1869.             "movn       %[sign1],   %[t0],      %[qc1]          \n\t"

  1870.             "slt        %[t2],      %[t2],      $zero           \n\t"

  1871.             "movn       %[sign2],   %[t2],      %[qc3]          \n\t"

  1872.             "slt        %[t1],      %[t1],      $zero           \n\t"

  1873.             "sll        %[t0],      %[sign1],   1               \n\t"

  1874.             "or         %[t0],      %[t0],      %[t1]           \n\t"

  1875.             "movn       %[sign1],   %[t0],      %[qc2]          \n\t"

  1876.             "slt        %[t3],      %[t3],      $zero           \n\t"

  1877.             "sll        %[t0],      %[sign2],   1               \n\t"

  1878.             "or         %[t0],      %[t0],      %[t3]           \n\t"

  1879.             "movn       %[sign2],   %[t0],      %[qc4]          \n\t"

  1880.             "slt        %[count1],  $zero,      %[qc1]          \n\t"

  1881.             "slt        %[t1],      $zero,      %[qc2]          \n\t"

  1882.             "slt        %[count2],  $zero,      %[qc3]          \n\t"

  1883.             "slt        %[t2],      $zero,      %[qc4]          \n\t"

  1884.             "addu       %[count1],  %[count1],  %[t1]           \n\t"

  1885.             "addu       %[count2],  %[count2],  %[t2]           \n\t"

  1886. 

  1887.             ".set pop                                           \n\t"

  1888. 

  1889.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  1890.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  1891.               [sign1]"=&r"(sign1), [count1]"=&r"(count1),

  1892.               [sign2]"=&r"(sign2), [count2]"=&r"(count2),

  1893.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  1894.               [t4]"=&r"(t4)

  1895.             : [in_int]"r"(in_int)

  1896.             : "memory"

  1897.         );

  1898. 

  1899.         curidx = 8 * qc1;

  1900.         curidx += qc2;

  1901. 

  1902.         curidx2 = 8 * qc3;

  1903.         curidx2 += qc4;

  1904. 

  1905.         curbits += p_bits[curidx];

  1906.         curbits += upair7_sign_bits[curidx];

  1907.         vec     = &p_codes[curidx*2];

  1908. 

  1909.         curbits += p_bits[curidx2];

  1910.         curbits += upair7_sign_bits[curidx2];

  1911.         vec2    = &p_codes[curidx2*2];

  1912. 

  1913.         __asm__ volatile (

  1914.             ".set push                                          \n\t"

  1915.             ".set noreorder                                     \n\t"

  1916. 

  1917.             "lwc1       %[di0],     0(%[in_pos])                \n\t"

  1918.             "lwc1       %[di1],     4(%[in_pos])                \n\t"

  1919.             "lwc1       %[di2],     8(%[in_pos])                \n\t"

  1920.             "lwc1       %[di3],     12(%[in_pos])               \n\t"

  1921.             "abs.s      %[di0],     %[di0]                      \n\t"

  1922.             "abs.s      %[di1],     %[di1]                      \n\t"

  1923.             "abs.s      %[di2],     %[di2]                      \n\t"

  1924.             "abs.s      %[di3],     %[di3]                      \n\t"

  1925.             "lwc1       $f0,        0(%[vec])                   \n\t"

  1926.             "lwc1       $f1,        4(%[vec])                   \n\t"

  1927.             "lwc1       $f2,        0(%[vec2])                  \n\t"

  1928.             "lwc1       $f3,        4(%[vec2])                  \n\t"

  1929.             "nmsub.s    %[di0],     %[di0],     $f0,    %[IQ]   \n\t"

  1930.             "nmsub.s    %[di1],     %[di1],     $f1,    %[IQ]   \n\t"

  1931.             "nmsub.s    %[di2],     %[di2],     $f2,    %[IQ]   \n\t"

  1932.             "nmsub.s    %[di3],     %[di3],     $f3,    %[IQ]   \n\t"

  1933. 

  1934.             ".set pop                                           \n\t"

  1935. 

  1936.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  1937.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  1938.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  1939.               [vec2]"r"(vec2), [IQ]"f"(IQ)

  1940.             : "$f0", "$f1", "$f2", "$f3",

  1941.               "memory"

  1942.         );

  1943. 

  1944.         cost += di0 * di0 + di1 * di1

  1945.                 + di2 * di2 + di3 * di3;

  1946.     }

  1947. 

  1948.     if (bits)

  1949.         *bits = curbits;

  1950.     return cost * lambda + curbits;

  1951. }

  1952. 

  1953. static float get_band_cost_UPAIR12_mips(struct AACEncContext *s,

  1954.                                         PutBitContext *pb, const float *in,

  1955.                                         const float *scaled, int size, int scale_idx,

  1956.                                         int cb, const float lambda, const float uplim,

  1957.                                         int *bits)

  1958. {

  1959.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  1960.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  1961.     int i;

  1962.     float cost = 0;

  1963.     int qc1, qc2, qc3, qc4;

  1964.     int curbits = 0;

  1965. 

  1966.     uint8_t *p_bits  = (uint8_t *)ff_aac_spectral_bits[cb-1];

  1967.     float   *p_codes = (float   *)ff_aac_codebook_vectors[cb-1];

  1968. 

  1969.     for (i = 0; i < size; i += 4) {

  1970.         const float *vec, *vec2;

  1971.         int curidx, curidx2;

  1972.         int sign1, count1, sign2, count2;

  1973.         int   *in_int = (int   *)&in[i];

  1974.         float *in_pos = (float *)&in[i];

  1975.         float di0, di1, di2, di3;

  1976.         int t0, t1, t2, t3, t4;

  1977. 

  1978.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  1979.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  1980.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  1981.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  1982. 

  1983.         __asm__ volatile (

  1984.             ".set push                                          \n\t"

  1985.             ".set noreorder                                     \n\t"

  1986. 

  1987.             "ori        %[t4],      $zero,      12              \n\t"

  1988.             "ori        %[sign1],   $zero,      0               \n\t"

  1989.             "ori        %[sign2],   $zero,      0               \n\t"

  1990.             "slt        %[t0],      %[t4],      %[qc1]          \n\t"

  1991.             "slt        %[t1],      %[t4],      %[qc2]          \n\t"

  1992.             "slt        %[t2],      %[t4],      %[qc3]          \n\t"

  1993.             "slt        %[t3],      %[t4],      %[qc4]          \n\t"

  1994.             "movn       %[qc1],     %[t4],      %[t0]           \n\t"

  1995.             "movn       %[qc2],     %[t4],      %[t1]           \n\t"

  1996.             "movn       %[qc3],     %[t4],      %[t2]           \n\t"

  1997.             "movn       %[qc4],     %[t4],      %[t3]           \n\t"

  1998.             "lw         %[t0],      0(%[in_int])                \n\t"

  1999.             "lw         %[t1],      4(%[in_int])                \n\t"

  2000.             "lw         %[t2],      8(%[in_int])                \n\t"

  2001.             "lw         %[t3],      12(%[in_int])               \n\t"

  2002.             "slt        %[t0],      %[t0],      $zero           \n\t"

  2003.             "movn       %[sign1],   %[t0],      %[qc1]          \n\t"

  2004.             "slt        %[t2],      %[t2],      $zero           \n\t"

  2005.             "movn       %[sign2],   %[t2],      %[qc3]          \n\t"

  2006.             "slt        %[t1],      %[t1],      $zero           \n\t"

  2007.             "sll        %[t0],      %[sign1],   1               \n\t"

  2008.             "or         %[t0],      %[t0],      %[t1]           \n\t"

  2009.             "movn       %[sign1],   %[t0],      %[qc2]          \n\t"

  2010.             "slt        %[t3],      %[t3],      $zero           \n\t"

  2011.             "sll        %[t0],      %[sign2],   1               \n\t"

  2012.             "or         %[t0],      %[t0],      %[t3]           \n\t"

  2013.             "movn       %[sign2],   %[t0],      %[qc4]          \n\t"

  2014.             "slt        %[count1],  $zero,      %[qc1]          \n\t"

  2015.             "slt        %[t1],      $zero,      %[qc2]          \n\t"

  2016.             "slt        %[count2],  $zero,      %[qc3]          \n\t"

  2017.             "slt        %[t2],      $zero,      %[qc4]          \n\t"

  2018.             "addu       %[count1],  %[count1],  %[t1]           \n\t"

  2019.             "addu       %[count2],  %[count2],  %[t2]           \n\t"

  2020. 

  2021.             ".set pop                                           \n\t"

  2022. 

  2023.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  2024.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  2025.               [sign1]"=&r"(sign1), [count1]"=&r"(count1),

  2026.               [sign2]"=&r"(sign2), [count2]"=&r"(count2),

  2027.               [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3),

  2028.               [t4]"=&r"(t4)

  2029.             : [in_int]"r"(in_int)

  2030.             : "memory"

  2031.         );

  2032. 

  2033.         curidx = 13 * qc1;

  2034.         curidx += qc2;

  2035. 

  2036.         curidx2 = 13 * qc3;

  2037.         curidx2 += qc4;

  2038. 

  2039.         curbits += p_bits[curidx];

  2040.         curbits += p_bits[curidx2];

  2041.         curbits += upair12_sign_bits[curidx];

  2042.         curbits += upair12_sign_bits[curidx2];

  2043.         vec     = &p_codes[curidx*2];

  2044.         vec2    = &p_codes[curidx2*2];

  2045. 

  2046.         __asm__ volatile (

  2047.             ".set push                                          \n\t"

  2048.             ".set noreorder                                     \n\t"

  2049. 

  2050.             "lwc1       %[di0],     0(%[in_pos])                \n\t"

  2051.             "lwc1       %[di1],     4(%[in_pos])                \n\t"

  2052.             "lwc1       %[di2],     8(%[in_pos])                \n\t"

  2053.             "lwc1       %[di3],     12(%[in_pos])               \n\t"

  2054.             "abs.s      %[di0],     %[di0]                      \n\t"

  2055.             "abs.s      %[di1],     %[di1]                      \n\t"

  2056.             "abs.s      %[di2],     %[di2]                      \n\t"

  2057.             "abs.s      %[di3],     %[di3]                      \n\t"

  2058.             "lwc1       $f0,        0(%[vec])                   \n\t"

  2059.             "lwc1       $f1,        4(%[vec])                   \n\t"

  2060.             "lwc1       $f2,        0(%[vec2])                  \n\t"

  2061.             "lwc1       $f3,        4(%[vec2])                  \n\t"

  2062.             "nmsub.s    %[di0],     %[di0],     $f0,    %[IQ]   \n\t"

  2063.             "nmsub.s    %[di1],     %[di1],     $f1,    %[IQ]   \n\t"

  2064.             "nmsub.s    %[di2],     %[di2],     $f2,    %[IQ]   \n\t"

  2065.             "nmsub.s    %[di3],     %[di3],     $f3,    %[IQ]   \n\t"

  2066. 

  2067.             ".set pop                                           \n\t"

  2068. 

  2069.             : [di0]"=&f"(di0), [di1]"=&f"(di1),

  2070.               [di2]"=&f"(di2), [di3]"=&f"(di3)

  2071.             : [in_pos]"r"(in_pos), [vec]"r"(vec),

  2072.               [vec2]"r"(vec2), [IQ]"f"(IQ)

  2073.             : "$f0", "$f1", "$f2", "$f3",

  2074.               "memory"

  2075.         );

  2076. 

  2077.         cost += di0 * di0 + di1 * di1

  2078.                 + di2 * di2 + di3 * di3;

  2079.     }

  2080. 

  2081.     if (bits)

  2082.         *bits = curbits;

  2083.     return cost * lambda + curbits;

  2084. }

  2085. 

  2086. static float get_band_cost_ESC_mips(struct AACEncContext *s,

  2087.                                     PutBitContext *pb, const float *in,

  2088.                                     const float *scaled, int size, int scale_idx,

  2089.                                     int cb, const float lambda, const float uplim,

  2090.                                     int *bits)

  2091. {

  2092.     const float Q34 = ff_aac_pow34sf_tab[POW_SF2_ZERO - scale_idx + SCALE_ONE_POS - SCALE_DIV_512];

  2093.     const float IQ  = ff_aac_pow2sf_tab [POW_SF2_ZERO + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];

  2094.     const float CLIPPED_ESCAPE = 165140.0f * IQ;

  2095.     int i;

  2096.     float cost = 0;

  2097.     int qc1, qc2, qc3, qc4;

  2098.     int curbits = 0;

  2099. 

  2100.     uint8_t *p_bits  = (uint8_t*)ff_aac_spectral_bits[cb-1];

  2101.     float   *p_codes = (float*  )ff_aac_codebook_vectors[cb-1];

  2102. 

  2103.     for (i = 0; i < size; i += 4) {

  2104.         const float *vec, *vec2;

  2105.         int curidx, curidx2;

  2106.         float t1, t2, t3, t4;

  2107.         float di1, di2, di3, di4;

  2108.         int cond0, cond1, cond2, cond3;

  2109.         int c1, c2, c3, c4;

  2110.         int t6, t7;

  2111. 

  2112.         qc1 = scaled[i  ] * Q34 + ROUND_STANDARD;

  2113.         qc2 = scaled[i+1] * Q34 + ROUND_STANDARD;

  2114.         qc3 = scaled[i+2] * Q34 + ROUND_STANDARD;

  2115.         qc4 = scaled[i+3] * Q34 + ROUND_STANDARD;

  2116. 

  2117.         __asm__ volatile (

  2118.             ".set push                                  \n\t"

  2119.             ".set noreorder                             \n\t"

  2120. 

  2121.             "ori        %[t6],      $zero,  15          \n\t"

  2122.             "ori        %[t7],      $zero,  16          \n\t"

  2123.             "shll_s.w   %[c1],      %[qc1], 18          \n\t"

  2124.             "shll_s.w   %[c2],      %[qc2], 18          \n\t"

  2125.             "shll_s.w   %[c3],      %[qc3], 18          \n\t"

  2126.             "shll_s.w   %[c4],      %[qc4], 18          \n\t"

  2127.             "srl        %[c1],      %[c1],  18          \n\t"

  2128.             "srl        %[c2],      %[c2],  18          \n\t"

  2129.             "srl        %[c3],      %[c3],  18          \n\t"

  2130.             "srl        %[c4],      %[c4],  18          \n\t"

  2131.             "slt        %[cond0],   %[t6],  %[qc1]      \n\t"

  2132.             "slt        %[cond1],   %[t6],  %[qc2]      \n\t"

  2133.             "slt        %[cond2],   %[t6],  %[qc3]      \n\t"

  2134.             "slt        %[cond3],   %[t6],  %[qc4]      \n\t"

  2135.             "movn       %[qc1],     %[t7],  %[cond0]    \n\t"

  2136.             "movn       %[qc2],     %[t7],  %[cond1]    \n\t"

  2137.             "movn       %[qc3],     %[t7],  %[cond2]    \n\t"

  2138.             "movn       %[qc4],     %[t7],  %[cond3]    \n\t"

  2139. 

  2140.             ".set pop                                   \n\t"

  2141. 

  2142.             : [qc1]"+r"(qc1), [qc2]"+r"(qc2),

  2143.               [qc3]"+r"(qc3), [qc4]"+r"(qc4),

  2144.               [cond0]"=&r"(cond0), [cond1]"=&r"(cond1),

  2145.               [cond2]"=&r"(cond2), [cond3]"=&r"(cond3),

  2146.               [c1]"=&r"(c1), [c2]"=&r"(c2),

  2147.               [c3]"=&r"(c3), [c4]"=&r"(c4),

  2148.               [t6]"=&r"(t6), [t7]"=&r"(t7)

  2149.         );

  2150. 

  2151.         curidx = 17 * qc1;

  2152.         curidx += qc2;

  2153. 

  2154.         curidx2 = 17 * qc3;

  2155.         curidx2 += qc4;

  2156. 

  2157.         curbits += p_bits[curidx];

  2158.         curbits += esc_sign_bits[curidx];

  2159.         vec     = &p_codes[curidx*2];

  2160. 

  2161.         curbits += p_bits[curidx2];

  2162.         curbits += esc_sign_bits[curidx2];

  2163.         vec2     = &p_codes[curidx2*2];

  2164. 

  2165.         curbits += (av_log2(c1) * 2 - 3) & (-cond0);

  2166.         curbits += (av_log2(c2) * 2 - 3) & (-cond1);

  2167.         curbits += (av_log2(c3) * 2 - 3) & (-cond2);

  2168.         curbits += (av_log2(c4) * 2 - 3) & (-cond3);

  2169. 

  2170.         t1 = fabsf(in[i  ]);

  2171.         t2 = fabsf(in[i+1]);

  2172.         t3 = fabsf(in[i+2]);

  2173.         t4 = fabsf(in[i+3]);

  2174. 

  2175.         if (cond0) {

  2176.             if (t1 >= CLIPPED_ESCAPE) {

  2177.                 di1 = t1 - CLIPPED_ESCAPE;

  2178.             } else {

  2179.                 di1 = t1 - c1 * cbrtf(c1) * IQ;

  2180.             }

  2181.         } else

  2182.             di1 = t1 - vec[0] * IQ;

  2183. 

  2184.         if (cond1) {

  2185.             if (t2 >= CLIPPED_ESCAPE) {

  2186.                 di2 = t2 - CLIPPED_ESCAPE;

  2187.             } else {

  2188.                 di2 = t2 - c2 * cbrtf(c2) * IQ;

  2189.             }

  2190.         } else

  2191.             di2 = t2 - vec[1] * IQ;

  2192. 

  2193.         if (cond2) {

  2194.             if (t3 >= CLIPPED_ESCAPE) {

  2195.                 di3 = t3 - CLIPPED_ESCAPE;

  2196.             } else {

  2197.                 di3 = t3 - c3 * cbrtf(c3) * IQ;

  2198.             }

  2199.         } else

  2200.             di3 = t3 - vec2[0] * IQ;

  2201. 

  2202.         if (cond3) {

  2203.             if (t4 >= CLIPPED_ESCAPE) {

  2204.                 di4 = t4 - CLIPPED_ESCAPE;

  2205.             } else {

  2206.                 di4 = t4 - c4 * cbrtf(c4) * IQ;

  2207.             }

  2208.         } else

  2209.             di4 = t4 - vec2[1]*IQ;

  2210. 

  2211.         cost += di1 * di1 + di2 * di2

  2212.                 + di3 * di3 + di4 * di4;

  2213.     }

  2214. 

  2215.     if (bits)

  2216.         *bits = curbits;

  2217.     return cost * lambda + curbits;

  2218. }

  2219. 

  2220. static float (*const get_band_cost_arr[])(struct AACEncContext *s,

  2221.                                           PutBitContext *pb, const float *in,

  2222.                                           const float *scaled, int size, int scale_idx,

  2223.                                           int cb, const float lambda, const float uplim,

  2224.                                           int *bits) = {

  2225.     get_band_cost_ZERO_mips,

  2226.     get_band_cost_SQUAD_mips,

  2227.     get_band_cost_SQUAD_mips,

  2228.     get_band_cost_UQUAD_mips,

  2229.     get_band_cost_UQUAD_mips,

  2230.     get_band_cost_SPAIR_mips,

  2231.     get_band_cost_SPAIR_mips,

  2232.     get_band_cost_UPAIR7_mips,

  2233.     get_band_cost_UPAIR7_mips,

  2234.     get_band_cost_UPAIR12_mips,

  2235.     get_band_cost_UPAIR12_mips,

  2236.     get_band_cost_ESC_mips,

  2237.     get_band_cost_NONE_mips, /* cb 12 doesn't exist */

  2238.     get_band_cost_ZERO_mips,

  2239.     get_band_cost_ZERO_mips,

  2240.     get_band_cost_ZERO_mips,

  2241. };

  2242. 

  2243. #define get_band_cost(                                  \

  2244.                                 s, pb, in, scaled, size, scale_idx, cb, \

  2245.                                 lambda, uplim, bits)                    \

  2246.     get_band_cost_arr[cb](                              \

  2247.                                 s, pb, in, scaled, size, scale_idx, cb, \

  2248.                                 lambda, uplim, bits)

  2249. 

  2250. static float quantize_band_cost(struct AACEncContext *s, const float *in,

  2251.                                 const float *scaled, int size, int scale_idx,

  2252.                                 int cb, const float lambda, const float uplim,

  2253.                                 int *bits, int rtz)

  2254. {

  2255.     return get_band_cost(s, NULL, in, scaled, size, scale_idx, cb, lambda, uplim, bits);

  2256. }

  2257. 

  2258. #include "libavcodec/aaccoder_twoloop.h"

  2259. 

  2260. static void search_for_ms_mips(AACEncContext *s, ChannelElement *cpe)

  2261. {

  2262.     int start = 0, i, w, w2, g;

  2263.     float M[128], S[128];

  2264.     float *L34 = s->scoefs, *R34 = s->scoefs + 128, *M34 = s->scoefs + 128*2, *S34 = s->scoefs + 128*3;

  2265.     const float lambda = s->lambda;

  2266.     SingleChannelElement *sce0 = &cpe->ch[0];

  2267.     SingleChannelElement *sce1 = &cpe->ch[1];

  2268.     if (!cpe->common_window)

  2269.         return;

  2270.     for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {

  2271.         start = 0;

  2272.         for (g = 0;  g < sce0->ics.num_swb; g++) {

  2273.             if (!cpe->ch[0].zeroes[w*16+g] && !cpe->ch[1].zeroes[w*16+g]) {

  2274.                 float dist1 = 0.0f, dist2 = 0.0f;

  2275.                 for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {

  2276.                     FFPsyBand *band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)*16+g];

  2277.                     FFPsyBand *band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)*16+g];

  2278.                     float minthr = FFMIN(band0->threshold, band1->threshold);

  2279.                     float maxthr = FFMAX(band0->threshold, band1->threshold);

  2280.                     for (i = 0; i < sce0->ics.swb_sizes[g]; i+=4) {

  2281.                         M[i  ] = (sce0->coeffs[start+w2*128+i  ]

  2282.                                 + sce1->coeffs[start+w2*128+i  ]) * 0.5;

  2283.                         M[i+1] = (sce0->coeffs[start+w2*128+i+1]

  2284.                                 + sce1->coeffs[start+w2*128+i+1]) * 0.5;

  2285.                         M[i+2] = (sce0->coeffs[start+w2*128+i+2]

  2286.                                 + sce1->coeffs[start+w2*128+i+2]) * 0.5;

  2287.                         M[i+3] = (sce0->coeffs[start+w2*128+i+3]

  2288.                                 + sce1->coeffs[start+w2*128+i+3]) * 0.5;

  2289. 

  2290.                         S[i  ] =  M[i  ]

  2291.                                 - sce1->coeffs[start+w2*128+i  ];

  2292.                         S[i+1] =  M[i+1]

  2293.                                 - sce1->coeffs[start+w2*128+i+1];

  2294.                         S[i+2] =  M[i+2]

  2295.                                 - sce1->coeffs[start+w2*128+i+2];

  2296.                         S[i+3] =  M[i+3]

  2297.                                 - sce1->coeffs[start+w2*128+i+3];

  2298.                    }

  2299.                     abs_pow34_v(L34, sce0->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);

  2300.                     abs_pow34_v(R34, sce1->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);

  2301.                     abs_pow34_v(M34, M,                         sce0->ics.swb_sizes[g]);

  2302.                     abs_pow34_v(S34, S,                         sce0->ics.swb_sizes[g]);

  2303.                     dist1 += quantize_band_cost(s, &sce0->coeffs[start + (w+w2)*128],

  2304.                                                 L34,

  2305.                                                 sce0->ics.swb_sizes[g],

  2306.                                                 sce0->sf_idx[(w+w2)*16+g],

  2307.                                                 sce0->band_type[(w+w2)*16+g],

  2308.                                                 lambda / band0->threshold, INFINITY, NULL, 0);

  2309.                     dist1 += quantize_band_cost(s, &sce1->coeffs[start + (w+w2)*128],

  2310.                                                 R34,

  2311.                                                 sce1->ics.swb_sizes[g],

  2312.                                                 sce1->sf_idx[(w+w2)*16+g],

  2313.                                                 sce1->band_type[(w+w2)*16+g],

  2314.                                                 lambda / band1->threshold, INFINITY, NULL, 0);

  2315.                     dist2 += quantize_band_cost(s, M,

  2316.                                                 M34,

  2317.                                                 sce0->ics.swb_sizes[g],

  2318.                                                 sce0->sf_idx[(w+w2)*16+g],

  2319.                                                 sce0->band_type[(w+w2)*16+g],

  2320.                                                 lambda / maxthr, INFINITY, NULL, 0);

  2321.                     dist2 += quantize_band_cost(s, S,

  2322.                                                 S34,

  2323.                                                 sce1->ics.swb_sizes[g],

  2324.                                                 sce1->sf_idx[(w+w2)*16+g],

  2325.                                                 sce1->band_type[(w+w2)*16+g],

  2326.                                                 lambda / minthr, INFINITY, NULL, 0);

  2327.                 }

  2328.                 cpe->ms_mask[w*16+g] = dist2 < dist1;

  2329.             }

  2330.             start += sce0->ics.swb_sizes[g];

  2331.         }

  2332.     }

  2333. }

  2334. #endif /*HAVE_MIPSFPU */

  2335. 

  2336. #include "libavcodec/aaccoder_trellis.h"

  2337. 

  2338. #endif /* HAVE_INLINE_ASM */

  2339. 

  2340. void ff_aac_coder_init_mips(AACEncContext *c) {

  2341. #if HAVE_INLINE_ASM

  2342.     AACCoefficientsEncoder *e = c->coder;

  2343.     int option = c->options.aac_coder;

  2344. 

  2345.     if (option == 2) {

  2346.         e->quantize_and_encode_band = quantize_and_encode_band_mips;

  2347.         e->encode_window_bands_info = codebook_trellis_rate;

  2348. #if HAVE_MIPSFPU

  2349.         e->search_for_quantizers    = search_for_quantizers_twoloop;

  2350. #endif /* HAVE_MIPSFPU */

  2351.     }

  2352. #if HAVE_MIPSFPU

  2353.     e->search_for_ms            = search_for_ms_mips;

  2354. #endif /* HAVE_MIPSFPU */

  2355. #endif /* HAVE_INLINE_ASM */

  2356. }