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.
100005 Commits
32 Branches
303MB
Tree: 16bb8247b4
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '16bb8247b4'
${ noResults }
FFmpeg/libavcodec/mpegaudiodsp_template.c

372 lines
11KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2001, 2002 Fabrice Bellard

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. #include <stdint.h>

  22. 

  23. #include "libavutil/attributes.h"

  24. #include "libavutil/mem.h"

  25. #include "libavutil/thread.h"

  26. #include "dct32.h"

  27. #include "mathops.h"

  28. #include "mpegaudiodsp.h"

  29. #include "mpegaudio.h"

  30. 

  31. #if USE_FLOATS

  32. #define RENAME(n) n##_float

  33. 

  34. static inline float round_sample(float *sum)

  35. {

  36.     float sum1=*sum;

  37.     *sum = 0;

  38.     return sum1;

  39. }

  40. 

  41. #define MACS(rt, ra, rb) rt+=(ra)*(rb)

  42. #define MULS(ra, rb) ((ra)*(rb))

  43. #define MULH3(x, y, s) ((s)*(y)*(x))

  44. #define MLSS(rt, ra, rb) rt-=(ra)*(rb)

  45. #define MULLx(x, y, s) ((y)*(x))

  46. #define FIXHR(x)        ((float)(x))

  47. #define FIXR(x)        ((float)(x))

  48. #define SHR(a,b)       ((a)*(1.0f/(1<<(b))))

  49. 

  50. #else

  51. 

  52. #define RENAME(n) n##_fixed

  53. #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)

  54. 

  55. static inline int round_sample(int64_t *sum)

  56. {

  57.     int sum1;

  58.     sum1 = (int)((*sum) >> OUT_SHIFT);

  59.     *sum &= (1<<OUT_SHIFT)-1;

  60.     return av_clip_int16(sum1);

  61. }

  62. 

  63. #   define MULS(ra, rb) MUL64(ra, rb)

  64. #   define MACS(rt, ra, rb) MAC64(rt, ra, rb)

  65. #   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)

  66. #   define MULH3(x, y, s) MULH((s)*(x), y)

  67. #   define MULLx(x, y, s) MULL((int)(x),(y),s)

  68. #   define SHR(a,b)       (((int)(a))>>(b))

  69. #   define FIXR(a)        ((int)((a) * FRAC_ONE + 0.5))

  70. #   define FIXHR(a)       ((int)((a) * (1LL<<32) + 0.5))

  71. #endif

  72. 

  73. /** Window for MDCT. Actually only the elements in [0,17] and

  74.     [MDCT_BUF_SIZE/2, MDCT_BUF_SIZE/2 + 17] are actually used. The rest

  75.     is just to preserve alignment for SIMD implementations.

  76. */

  77. DECLARE_ALIGNED(16, INTFLOAT, RENAME(ff_mdct_win))[8][MDCT_BUF_SIZE];

  78. 

  79. DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256];

  80. 

  81. #define SUM8(op, sum, w, p)               \

  82. {                                         \

  83.     op(sum, (w)[0 * 64], (p)[0 * 64]);    \

  84.     op(sum, (w)[1 * 64], (p)[1 * 64]);    \

  85.     op(sum, (w)[2 * 64], (p)[2 * 64]);    \

  86.     op(sum, (w)[3 * 64], (p)[3 * 64]);    \

  87.     op(sum, (w)[4 * 64], (p)[4 * 64]);    \

  88.     op(sum, (w)[5 * 64], (p)[5 * 64]);    \

  89.     op(sum, (w)[6 * 64], (p)[6 * 64]);    \

  90.     op(sum, (w)[7 * 64], (p)[7 * 64]);    \

  91. }

  92. 

  93. #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \

  94. {                                               \

  95.     INTFLOAT tmp;\

  96.     tmp = p[0 * 64];\

  97.     op1(sum1, (w1)[0 * 64], tmp);\

  98.     op2(sum2, (w2)[0 * 64], tmp);\

  99.     tmp = p[1 * 64];\

  100.     op1(sum1, (w1)[1 * 64], tmp);\

  101.     op2(sum2, (w2)[1 * 64], tmp);\

  102.     tmp = p[2 * 64];\

  103.     op1(sum1, (w1)[2 * 64], tmp);\

  104.     op2(sum2, (w2)[2 * 64], tmp);\

  105.     tmp = p[3 * 64];\

  106.     op1(sum1, (w1)[3 * 64], tmp);\

  107.     op2(sum2, (w2)[3 * 64], tmp);\

  108.     tmp = p[4 * 64];\

  109.     op1(sum1, (w1)[4 * 64], tmp);\

  110.     op2(sum2, (w2)[4 * 64], tmp);\

  111.     tmp = p[5 * 64];\

  112.     op1(sum1, (w1)[5 * 64], tmp);\

  113.     op2(sum2, (w2)[5 * 64], tmp);\

  114.     tmp = p[6 * 64];\

  115.     op1(sum1, (w1)[6 * 64], tmp);\

  116.     op2(sum2, (w2)[6 * 64], tmp);\

  117.     tmp = p[7 * 64];\

  118.     op1(sum1, (w1)[7 * 64], tmp);\

  119.     op2(sum2, (w2)[7 * 64], tmp);\

  120. }

  121. 

  122. void RENAME(ff_mpadsp_apply_window)(MPA_INT *synth_buf, MPA_INT *window,

  123.                                   int *dither_state, OUT_INT *samples,

  124.                                   ptrdiff_t incr)

  125. {

  126.     register const MPA_INT *w, *w2, *p;

  127.     int j;

  128.     OUT_INT *samples2;

  129. #if USE_FLOATS

  130.     float sum, sum2;

  131. #else

  132.     int64_t sum, sum2;

  133. #endif

  134. 

  135.     /* copy to avoid wrap */

  136.     memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));

  137. 

  138.     samples2 = samples + 31 * incr;

  139.     w = window;

  140.     w2 = window + 31;

  141. 

  142.     sum = *dither_state;

  143.     p = synth_buf + 16;

  144.     SUM8(MACS, sum, w, p);

  145.     p = synth_buf + 48;

  146.     SUM8(MLSS, sum, w + 32, p);

  147.     *samples = round_sample(&sum);

  148.     samples += incr;

  149.     w++;

  150. 

  151.     /* we calculate two samples at the same time to avoid one memory

  152.        access per two sample */

  153.     for(j=1;j<16;j++) {

  154.         sum2 = 0;

  155.         p = synth_buf + 16 + j;

  156.         SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);

  157.         p = synth_buf + 48 - j;

  158.         SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);

  159. 

  160.         *samples = round_sample(&sum);

  161.         samples += incr;

  162.         sum += sum2;

  163.         *samples2 = round_sample(&sum);

  164.         samples2 -= incr;

  165.         w++;

  166.         w2--;

  167.     }

  168. 

  169.     p = synth_buf + 32;

  170.     SUM8(MLSS, sum, w + 32, p);

  171.     *samples = round_sample(&sum);

  172.     *dither_state= sum;

  173. }

  174. 

  175. /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:

  176.    32 samples. */

  177. void RENAME(ff_mpa_synth_filter)(MPADSPContext *s, MPA_INT *synth_buf_ptr,

  178.                                  int *synth_buf_offset,

  179.                                  MPA_INT *window, int *dither_state,

  180.                                  OUT_INT *samples, ptrdiff_t incr,

  181.                                  MPA_INT *sb_samples)

  182. {

  183.     MPA_INT *synth_buf;

  184.     int offset;

  185. 

  186.     offset = *synth_buf_offset;

  187.     synth_buf = synth_buf_ptr + offset;

  188. 

  189.     s->RENAME(dct32)(synth_buf, sb_samples);

  190.     s->RENAME(apply_window)(synth_buf, window, dither_state, samples, incr);

  191. 

  192.     offset = (offset - 32) & 511;

  193.     *synth_buf_offset = offset;

  194. }

  195. 

  196. static av_cold void mpa_synth_init(MPA_INT *window)

  197. {

  198.     int i, j;

  199. 

  200.     /* max = 18760, max sum over all 16 coefs : 44736 */

  201.     for(i=0;i<257;i++) {

  202.         INTFLOAT v;

  203.         v = ff_mpa_enwindow[i];

  204. #if USE_FLOATS

  205.         v *= 1.0 / (1LL<<(16 + FRAC_BITS));

  206. #endif

  207.         window[i] = v;

  208.         if ((i & 63) != 0)

  209.             v = -v;

  210.         if (i != 0)

  211.             window[512 - i] = v;

  212.     }

  213. 

  214. 

  215.     // Needed for avoiding shuffles in ASM implementations

  216.     for(i=0; i < 8; i++)

  217.         for(j=0; j < 16; j++)

  218.             window[512+16*i+j] = window[64*i+32-j];

  219. 

  220.     for(i=0; i < 8; i++)

  221.         for(j=0; j < 16; j++)

  222.             window[512+128+16*i+j] = window[64*i+48-j];

  223. }

  224. 

  225. static av_cold void mpa_synth_window_init(void)

  226. {

  227.     mpa_synth_init(RENAME(ff_mpa_synth_window));

  228. }

  229. 

  230. av_cold void RENAME(ff_mpa_synth_init)(void)

  231. {

  232.     static AVOnce init_static_once = AV_ONCE_INIT;

  233.     ff_thread_once(&init_static_once, mpa_synth_window_init);

  234. }

  235. 

  236. /* cos(pi*i/18) */

  237. #define C1 FIXHR(0.98480775301220805936/2)

  238. #define C2 FIXHR(0.93969262078590838405/2)

  239. #define C3 FIXHR(0.86602540378443864676/2)

  240. #define C4 FIXHR(0.76604444311897803520/2)

  241. #define C5 FIXHR(0.64278760968653932632/2)

  242. #define C6 FIXHR(0.5/2)

  243. #define C7 FIXHR(0.34202014332566873304/2)

  244. #define C8 FIXHR(0.17364817766693034885/2)

  245. 

  246. /* 0.5 / cos(pi*(2*i+1)/36) */

  247. static const INTFLOAT icos36[9] = {

  248.     FIXR(0.50190991877167369479),

  249.     FIXR(0.51763809020504152469), //0

  250.     FIXR(0.55168895948124587824),

  251.     FIXR(0.61038729438072803416),

  252.     FIXR(0.70710678118654752439), //1

  253.     FIXR(0.87172339781054900991),

  254.     FIXR(1.18310079157624925896),

  255.     FIXR(1.93185165257813657349), //2

  256.     FIXR(5.73685662283492756461),

  257. };

  258. 

  259. /* 0.5 / cos(pi*(2*i+1)/36) */

  260. static const INTFLOAT icos36h[9] = {

  261.     FIXHR(0.50190991877167369479/2),

  262.     FIXHR(0.51763809020504152469/2), //0

  263.     FIXHR(0.55168895948124587824/2),

  264.     FIXHR(0.61038729438072803416/2),

  265.     FIXHR(0.70710678118654752439/2), //1

  266.     FIXHR(0.87172339781054900991/2),

  267.     FIXHR(1.18310079157624925896/4),

  268.     FIXHR(1.93185165257813657349/4), //2

  269. //    FIXHR(5.73685662283492756461),

  270. };

  271. 

  272. /* using Lee like decomposition followed by hand coded 9 points DCT */

  273. static void imdct36(INTFLOAT *out, INTFLOAT *buf, SUINTFLOAT *in, INTFLOAT *win)

  274. {

  275.     int i, j;

  276.     SUINTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;

  277.     SUINTFLOAT tmp[18], *tmp1, *in1;

  278. 

  279.     for (i = 17; i >= 1; i--)

  280.         in[i] += in[i-1];

  281.     for (i = 17; i >= 3; i -= 2)

  282.         in[i] += in[i-2];

  283. 

  284.     for (j = 0; j < 2; j++) {

  285.         tmp1 = tmp + j;

  286.         in1 = in + j;

  287. 

  288.         t2 = in1[2*4] + in1[2*8] - in1[2*2];

  289. 

  290.         t3 = in1[2*0] + SHR(in1[2*6],1);

  291.         t1 = in1[2*0] - in1[2*6];

  292.         tmp1[ 6] = t1 - SHR(t2,1);

  293.         tmp1[16] = t1 + t2;

  294. 

  295.         t0 = MULH3(in1[2*2] + in1[2*4] ,    C2, 2);

  296.         t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);

  297.         t2 = MULH3(in1[2*2] + in1[2*8] ,   -C4, 2);

  298. 

  299.         tmp1[10] = t3 - t0 - t2;

  300.         tmp1[ 2] = t3 + t0 + t1;

  301.         tmp1[14] = t3 + t2 - t1;

  302. 

  303.         tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);

  304.         t2 = MULH3(in1[2*1] + in1[2*5],    C1, 2);

  305.         t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);

  306.         t0 = MULH3(in1[2*3], C3, 2);

  307. 

  308.         t1 = MULH3(in1[2*1] + in1[2*7],   -C5, 2);

  309. 

  310.         tmp1[ 0] = t2 + t3 + t0;

  311.         tmp1[12] = t2 + t1 - t0;

  312.         tmp1[ 8] = t3 - t1 - t0;

  313.     }

  314. 

  315.     i = 0;

  316.     for (j = 0; j < 4; j++) {

  317.         t0 = tmp[i];

  318.         t1 = tmp[i + 2];

  319.         s0 = t1 + t0;

  320.         s2 = t1 - t0;

  321. 

  322.         t2 = tmp[i + 1];

  323.         t3 = tmp[i + 3];

  324.         s1 = MULH3(t3 + t2, icos36h[    j], 2);

  325.         s3 = MULLx(t3 - t2, icos36 [8 - j], FRAC_BITS);

  326. 

  327.         t0 = s0 + s1;

  328.         t1 = s0 - s1;

  329.         out[(9 + j) * SBLIMIT] = MULH3(t1, win[     9 + j], 1) + buf[4*(9 + j)];

  330.         out[(8 - j) * SBLIMIT] = MULH3(t1, win[     8 - j], 1) + buf[4*(8 - j)];

  331.         buf[4 * ( 9 + j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + j], 1);

  332.         buf[4 * ( 8 - j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - j], 1);

  333. 

  334.         t0 = s2 + s3;

  335.         t1 = s2 - s3;

  336.         out[(9 + 8 - j) * SBLIMIT] = MULH3(t1, win[     9 + 8 - j], 1) + buf[4*(9 + 8 - j)];

  337.         out[         j  * SBLIMIT] = MULH3(t1, win[             j], 1) + buf[4*(        j)];

  338.         buf[4 * ( 9 + 8 - j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 8 - j], 1);

  339.         buf[4 * (         j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2         + j], 1);

  340.         i += 4;

  341.     }

  342. 

  343.     s0 = tmp[16];

  344.     s1 = MULH3(tmp[17], icos36h[4], 2);

  345.     t0 = s0 + s1;

  346.     t1 = s0 - s1;

  347.     out[(9 + 4) * SBLIMIT] = MULH3(t1, win[     9 + 4], 1) + buf[4*(9 + 4)];

  348.     out[(8 - 4) * SBLIMIT] = MULH3(t1, win[     8 - 4], 1) + buf[4*(8 - 4)];

  349.     buf[4 * ( 9 + 4     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 4], 1);

  350.     buf[4 * ( 8 - 4     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - 4], 1);

  351. }

  352. 

  353. void RENAME(ff_imdct36_blocks)(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in,

  354.                                int count, int switch_point, int block_type)

  355. {

  356.     int j;

  357.     for (j=0 ; j < count; j++) {

  358.         /* apply window & overlap with previous buffer */

  359. 

  360.         /* select window */

  361.         int win_idx = (switch_point && j < 2) ? 0 : block_type;

  362.         INTFLOAT *win = RENAME(ff_mdct_win)[win_idx + (4 & -(j & 1))];

  363. 

  364.         imdct36(out, buf, in, win);

  365. 

  366.         in  += 18;

  367.         buf += ((j&3) != 3 ? 1 : (72-3));

  368.         out++;

  369.     }

  370. }