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.
71232 Commits
32 Branches
325MB
Tree: af5fec9538
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'af5fec9538'
${ noResults }
FFmpeg/libavcodec/ppc/fdctdsp.c

488 lines
19KB
Raw Blame History 

  1. /*

  2.  * Copyright (C) 2003  James Klicman <james@klicman.org>

  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 "config.h"

  22. #if HAVE_ALTIVEC_H

  23. #include <altivec.h>

  24. #endif

  25. 

  26. #include "libavutil/attributes.h"

  27. #include "libavutil/cpu.h"

  28. #include "libavutil/ppc/cpu.h"

  29. #include "libavcodec/fdctdsp.h"

  30. #include "fdct.h"

  31. 

  32. #if HAVE_ALTIVEC

  33. 

  34. #define vs16(v)   ((vector signed short) (v))

  35. #define vs32(v)     ((vector signed int) (v))

  36. #define  vu8(v)  ((vector unsigned char) (v))

  37. #define vu16(v) ((vector unsigned short) (v))

  38. #define vu32(v)   ((vector unsigned int) (v))

  39. 

  40. #define C1     0.98078525066375732421875000 /* cos(1 * PI / 16) */

  41. #define C2     0.92387950420379638671875000 /* cos(2 * PI / 16) */

  42. #define C3     0.83146959543228149414062500 /* cos(3 * PI / 16) */

  43. #define C4     0.70710676908493041992187500 /* cos(4 * PI / 16) */

  44. #define C5     0.55557024478912353515625000 /* cos(5 * PI / 16) */

  45. #define C6     0.38268342614173889160156250 /* cos(6 * PI / 16) */

  46. #define C7     0.19509032368659973144531250 /* cos(7 * PI / 16) */

  47. #define SQRT_2 1.41421353816986083984375000 /* sqrt(2)          */

  48. 

  49. #define W0 -(2 * C2)

  50. #define W1  (2 * C6)

  51. #define W2 (SQRT_2 * C6)

  52. #define W3 (SQRT_2 * C3)

  53. #define W4 (SQRT_2 * (-C1 + C3 + C5 - C7))

  54. #define W5 (SQRT_2 *  (C1 + C3 - C5 + C7))

  55. #define W6 (SQRT_2 *  (C1 + C3 + C5 - C7))

  56. #define W7 (SQRT_2 *  (C1 + C3 - C5 - C7))

  57. #define W8 (SQRT_2 *  (C7 - C3))

  58. #define W9 (SQRT_2 * (-C1 - C3))

  59. #define WA (SQRT_2 * (-C3 - C5))

  60. #define WB (SQRT_2 *  (C5 - C3))

  61. 

  62. static const vector float fdctconsts[3] = {

  63.     { W0, W1, W2, W3 },

  64.     { W4, W5, W6, W7 },

  65.     { W8, W9, WA, WB }

  66. };

  67. 

  68. #define LD_W0 vec_splat(cnsts0, 0)

  69. #define LD_W1 vec_splat(cnsts0, 1)

  70. #define LD_W2 vec_splat(cnsts0, 2)

  71. #define LD_W3 vec_splat(cnsts0, 3)

  72. #define LD_W4 vec_splat(cnsts1, 0)

  73. #define LD_W5 vec_splat(cnsts1, 1)

  74. #define LD_W6 vec_splat(cnsts1, 2)

  75. #define LD_W7 vec_splat(cnsts1, 3)

  76. #define LD_W8 vec_splat(cnsts2, 0)

  77. #define LD_W9 vec_splat(cnsts2, 1)

  78. #define LD_WA vec_splat(cnsts2, 2)

  79. #define LD_WB vec_splat(cnsts2, 3)

  80. 

  81. #if HAVE_BIGENDIAN

  82. #define VEC_FMERGEH(a, b) vec_mergeh(a, b)

  83. #define VEC_FMERGEL(a, b) vec_mergel(a, b)

  84. #else

  85. #define VEC_FMERGEH(a, b) vec_mergel(b, a)

  86. #define VEC_FMERGEL(a, b) vec_mergeh(b, a)

  87. #endif

  88. 

  89. #define FDCTROW(b0, b1, b2, b3, b4, b5, b6, b7) /* {{{ */           \

  90.     x0 = vec_add(b0, b7);               /* x0 = b0 + b7; */         \

  91.     x7 = vec_sub(b0, b7);               /* x7 = b0 - b7; */         \

  92.     x1 = vec_add(b1, b6);               /* x1 = b1 + b6; */         \

  93.     x6 = vec_sub(b1, b6);               /* x6 = b1 - b6; */         \

  94.     x2 = vec_add(b2, b5);               /* x2 = b2 + b5; */         \

  95.     x5 = vec_sub(b2, b5);               /* x5 = b2 - b5; */         \

  96.     x3 = vec_add(b3, b4);               /* x3 = b3 + b4; */         \

  97.     x4 = vec_sub(b3, b4);               /* x4 = b3 - b4; */         \

  98.                                                                     \

  99.     b7 = vec_add(x0, x3);               /* b7 = x0 + x3; */         \

  100.     b1 = vec_add(x1, x2);               /* b1 = x1 + x2; */         \

  101.     b0 = vec_add(b7, b1);               /* b0 = b7 + b1; */         \

  102.     b4 = vec_sub(b7, b1);               /* b4 = b7 - b1; */         \

  103.                                                                     \

  104.     b2   = vec_sub(x0, x3);             /* b2 = x0 - x3; */         \

  105.     b6   = vec_sub(x1, x2);             /* b6 = x1 - x2; */         \

  106.     b5   = vec_add(b6, b2);             /* b5 = b6 + b2; */         \

  107.     cnst = LD_W2;                                                   \

  108.     b5   = vec_madd(cnst, b5, mzero);   /* b5 = b5 * W2; */         \

  109.     cnst = LD_W1;                                                   \

  110.     b2   = vec_madd(cnst, b2, b5);      /* b2 = b5 + b2 * W1; */    \

  111.     cnst = LD_W0;                                                   \

  112.     b6   = vec_madd(cnst, b6, b5);      /* b6 = b5 + b6 * W0; */    \

  113.                                                                     \

  114.     x0   = vec_add(x4, x7);             /* x0 = x4 + x7; */         \

  115.     x1   = vec_add(x5, x6);             /* x1 = x5 + x6; */         \

  116.     x2   = vec_add(x4, x6);             /* x2 = x4 + x6; */         \

  117.     x3   = vec_add(x5, x7);             /* x3 = x5 + x7; */         \

  118.     x8   = vec_add(x2, x3);             /* x8 = x2 + x3; */         \

  119.     cnst = LD_W3;                                                   \

  120.     x8   = vec_madd(cnst, x8, mzero);   /* x8 = x8 * W3; */         \

  121.                                                                     \

  122.     cnst = LD_W8;                                                   \

  123.     x0   = vec_madd(cnst, x0, mzero);   /* x0 *= W8; */             \

  124.     cnst = LD_W9;                                                   \

  125.     x1   = vec_madd(cnst, x1, mzero);   /* x1 *= W9; */             \

  126.     cnst = LD_WA;                                                   \

  127.     x2   = vec_madd(cnst, x2, x8);      /* x2 = x2 * WA + x8; */    \

  128.     cnst = LD_WB;                                                   \

  129.     x3   = vec_madd(cnst, x3, x8);      /* x3 = x3 * WB + x8; */    \

  130.                                                                     \

  131.     cnst = LD_W4;                                                   \

  132.     b7   = vec_madd(cnst, x4, x0);      /* b7 = x4 * W4 + x0; */    \

  133.     cnst = LD_W5;                                                   \

  134.     b5   = vec_madd(cnst, x5, x1);      /* b5 = x5 * W5 + x1; */    \

  135.     cnst = LD_W6;                                                   \

  136.     b3   = vec_madd(cnst, x6, x1);      /* b3 = x6 * W6 + x1; */    \

  137.     cnst = LD_W7;                                                   \

  138.     b1   = vec_madd(cnst, x7, x0);      /* b1 = x7 * W7 + x0; */    \

  139.                                                                     \

  140.     b7 = vec_add(b7, x2);               /* b7 = b7 + x2; */         \

  141.     b5 = vec_add(b5, x3);               /* b5 = b5 + x3; */         \

  142.     b3 = vec_add(b3, x2);               /* b3 = b3 + x2; */         \

  143.     b1 = vec_add(b1, x3)                /* b1 = b1 + x3; */         \

  144.     /* }}} */

  145. 

  146. #define FDCTCOL(b0, b1, b2, b3, b4, b5, b6, b7) /* {{{ */           \

  147.     x0 = vec_add(b0, b7);               /* x0 = b0 + b7; */         \

  148.     x7 = vec_sub(b0, b7);               /* x7 = b0 - b7; */         \

  149.     x1 = vec_add(b1, b6);               /* x1 = b1 + b6; */         \

  150.     x6 = vec_sub(b1, b6);               /* x6 = b1 - b6; */         \

  151.     x2 = vec_add(b2, b5);               /* x2 = b2 + b5; */         \

  152.     x5 = vec_sub(b2, b5);               /* x5 = b2 - b5; */         \

  153.     x3 = vec_add(b3, b4);               /* x3 = b3 + b4; */         \

  154.     x4 = vec_sub(b3, b4);               /* x4 = b3 - b4; */         \

  155.                                                                     \

  156.     b7 = vec_add(x0, x3);               /* b7 = x0 + x3; */         \

  157.     b1 = vec_add(x1, x2);               /* b1 = x1 + x2; */         \

  158.     b0 = vec_add(b7, b1);               /* b0 = b7 + b1; */         \

  159.     b4 = vec_sub(b7, b1);               /* b4 = b7 - b1; */         \

  160.                                                                     \

  161.     b2   = vec_sub(x0, x3);             /* b2 = x0 - x3; */         \

  162.     b6   = vec_sub(x1, x2);             /* b6 = x1 - x2; */         \

  163.     b5   = vec_add(b6, b2);             /* b5 = b6 + b2; */         \

  164.     cnst = LD_W2;                                                   \

  165.     b5   = vec_madd(cnst, b5, mzero);   /* b5 = b5 * W2; */         \

  166.     cnst = LD_W1;                                                   \

  167.     b2   = vec_madd(cnst, b2, b5);      /* b2 = b5 + b2 * W1; */    \

  168.     cnst = LD_W0;                                                   \

  169.     b6   = vec_madd(cnst, b6, b5);      /* b6 = b5 + b6 * W0; */    \

  170.                                                                     \

  171.     x0   = vec_add(x4, x7);             /* x0 = x4 + x7; */         \

  172.     x1   = vec_add(x5, x6);             /* x1 = x5 + x6; */         \

  173.     x2   = vec_add(x4, x6);             /* x2 = x4 + x6; */         \

  174.     x3   = vec_add(x5, x7);             /* x3 = x5 + x7; */         \

  175.     x8   = vec_add(x2, x3);             /* x8 = x2 + x3; */         \

  176.     cnst = LD_W3;                                                   \

  177.     x8   = vec_madd(cnst, x8, mzero);   /* x8 = x8 * W3; */         \

  178.                                                                     \

  179.     cnst = LD_W8;                                                   \

  180.     x0   = vec_madd(cnst, x0, mzero);   /* x0 *= W8; */             \

  181.     cnst = LD_W9;                                                   \

  182.     x1   = vec_madd(cnst, x1, mzero);   /* x1 *= W9; */             \

  183.     cnst = LD_WA;                                                   \

  184.     x2   = vec_madd(cnst, x2, x8);      /* x2 = x2 * WA + x8; */    \

  185.     cnst = LD_WB;                                                   \

  186.     x3   = vec_madd(cnst, x3, x8);      /* x3 = x3 * WB + x8; */    \

  187.                                                                     \

  188.     cnst = LD_W4;                                                   \

  189.     b7   = vec_madd(cnst, x4, x0);      /* b7 = x4 * W4 + x0; */    \

  190.     cnst = LD_W5;                                                   \

  191.     b5   = vec_madd(cnst, x5, x1);      /* b5 = x5 * W5 + x1; */    \

  192.     cnst = LD_W6;                                                   \

  193.     b3   = vec_madd(cnst, x6, x1);      /* b3 = x6 * W6 + x1; */    \

  194.     cnst = LD_W7;                                                   \

  195.     b1   = vec_madd(cnst, x7, x0);      /* b1 = x7 * W7 + x0; */    \

  196.                                                                     \

  197.     b7 = vec_add(b7, x2);               /* b7 += x2; */             \

  198.     b5 = vec_add(b5, x3);               /* b5 += x3; */             \

  199.     b3 = vec_add(b3, x2);               /* b3 += x2; */             \

  200.     b1 = vec_add(b1, x3)                /* b1 += x3; */             \

  201.     /* }}} */

  202. 

  203. /* two dimensional discrete cosine transform */

  204. void ff_fdct_altivec(int16_t *block)

  205. {

  206.     vector signed short *bp;

  207.     const vector float *cp = fdctconsts;

  208.     vector float b00, b10, b20, b30, b40, b50, b60, b70;

  209.     vector float b01, b11, b21, b31, b41, b51, b61, b71;

  210.     vector float mzero, cnst, cnsts0, cnsts1, cnsts2;

  211.     vector float x0, x1, x2, x3, x4, x5, x6, x7, x8;

  212. 

  213.     /* setup constants {{{ */

  214.     /* mzero = -0.0 */

  215.     mzero  = ((vector float) vec_splat_u32(-1));

  216.     mzero  = ((vector float) vec_sl(vu32(mzero), vu32(mzero)));

  217.     cnsts0 = vec_ld(0, cp);

  218.     cp++;

  219.     cnsts1 = vec_ld(0, cp);

  220.     cp++;

  221.     cnsts2 = vec_ld(0, cp);

  222.     /* }}} */

  223. 

  224.     /* 8x8 matrix transpose (vector short[8]) {{{ */

  225. #define MERGE_S16(hl, a, b) vec_merge ## hl(vs16(a), vs16(b))

  226. 

  227.     bp  = (vector signed short *) block;

  228.     b00 = ((vector float) vec_ld(0,      bp));

  229.     b40 = ((vector float) vec_ld(16 * 4, bp));

  230.     b01 = ((vector float) MERGE_S16(h, b00, b40));

  231.     b11 = ((vector float) MERGE_S16(l, b00, b40));

  232.     bp++;

  233.     b10 = ((vector float) vec_ld(0,      bp));

  234.     b50 = ((vector float) vec_ld(16 * 4, bp));

  235.     b21 = ((vector float) MERGE_S16(h, b10, b50));

  236.     b31 = ((vector float) MERGE_S16(l, b10, b50));

  237.     bp++;

  238.     b20 = ((vector float) vec_ld(0,      bp));

  239.     b60 = ((vector float) vec_ld(16 * 4, bp));

  240.     b41 = ((vector float) MERGE_S16(h, b20, b60));

  241.     b51 = ((vector float) MERGE_S16(l, b20, b60));

  242.     bp++;

  243.     b30 = ((vector float) vec_ld(0,      bp));

  244.     b70 = ((vector float) vec_ld(16 * 4, bp));

  245.     b61 = ((vector float) MERGE_S16(h, b30, b70));

  246.     b71 = ((vector float) MERGE_S16(l, b30, b70));

  247. 

  248.     x0 = ((vector float) MERGE_S16(h, b01, b41));

  249.     x1 = ((vector float) MERGE_S16(l, b01, b41));

  250.     x2 = ((vector float) MERGE_S16(h, b11, b51));

  251.     x3 = ((vector float) MERGE_S16(l, b11, b51));

  252.     x4 = ((vector float) MERGE_S16(h, b21, b61));

  253.     x5 = ((vector float) MERGE_S16(l, b21, b61));

  254.     x6 = ((vector float) MERGE_S16(h, b31, b71));

  255.     x7 = ((vector float) MERGE_S16(l, b31, b71));

  256. 

  257.     b00 = ((vector float) MERGE_S16(h, x0, x4));

  258.     b10 = ((vector float) MERGE_S16(l, x0, x4));

  259.     b20 = ((vector float) MERGE_S16(h, x1, x5));

  260.     b30 = ((vector float) MERGE_S16(l, x1, x5));

  261.     b40 = ((vector float) MERGE_S16(h, x2, x6));

  262.     b50 = ((vector float) MERGE_S16(l, x2, x6));

  263.     b60 = ((vector float) MERGE_S16(h, x3, x7));

  264.     b70 = ((vector float) MERGE_S16(l, x3, x7));

  265. 

  266. #undef MERGE_S16

  267.     /* }}} */

  268. 

  269.     /* Some of the initial calculations can be done as vector short

  270.      * before conversion to vector float.  The following code section

  271.      * takes advantage of this. */

  272. 

  273.     /* fdct rows {{{ */

  274.     x0 = ((vector float) vec_add(vs16(b00), vs16(b70)));

  275.     x7 = ((vector float) vec_sub(vs16(b00), vs16(b70)));

  276.     x1 = ((vector float) vec_add(vs16(b10), vs16(b60)));

  277.     x6 = ((vector float) vec_sub(vs16(b10), vs16(b60)));

  278.     x2 = ((vector float) vec_add(vs16(b20), vs16(b50)));

  279.     x5 = ((vector float) vec_sub(vs16(b20), vs16(b50)));

  280.     x3 = ((vector float) vec_add(vs16(b30), vs16(b40)));

  281.     x4 = ((vector float) vec_sub(vs16(b30), vs16(b40)));

  282. 

  283.     b70 = ((vector float) vec_add(vs16(x0), vs16(x3)));

  284.     b10 = ((vector float) vec_add(vs16(x1), vs16(x2)));

  285. 

  286.     b00 = ((vector float) vec_add(vs16(b70), vs16(b10)));

  287.     b40 = ((vector float) vec_sub(vs16(b70), vs16(b10)));

  288. 

  289. #define CTF0(n)                                                    \

  290.     b ## n ## 1 = ((vector float) vec_unpackl(vs16(b ## n ## 0))); \

  291.     b ## n ## 0 = ((vector float) vec_unpackh(vs16(b ## n ## 0))); \

  292.     b ## n ## 1 = vec_ctf(vs32(b ## n ## 1), 0);                   \

  293.     b ## n ## 0 = vec_ctf(vs32(b ## n ## 0), 0)

  294. 

  295.     CTF0(0);

  296.     CTF0(4);

  297. 

  298.     b20 = ((vector float) vec_sub(vs16(x0), vs16(x3)));

  299.     b60 = ((vector float) vec_sub(vs16(x1), vs16(x2)));

  300. 

  301.     CTF0(2);

  302.     CTF0(6);

  303. 

  304. #undef CTF0

  305. 

  306.     x0 = vec_add(b60, b20);

  307.     x1 = vec_add(b61, b21);

  308. 

  309.     cnst = LD_W2;

  310.     x0   = vec_madd(cnst, x0, mzero);

  311.     x1   = vec_madd(cnst, x1, mzero);

  312.     cnst = LD_W1;

  313.     b20  = vec_madd(cnst, b20, x0);

  314.     b21  = vec_madd(cnst, b21, x1);

  315.     cnst = LD_W0;

  316.     b60  = vec_madd(cnst, b60, x0);

  317.     b61  = vec_madd(cnst, b61, x1);

  318. 

  319. #define CTFX(x, b)                                  \

  320.     b ## 0 = ((vector float) vec_unpackh(vs16(x))); \

  321.     b ## 1 = ((vector float) vec_unpackl(vs16(x))); \

  322.     b ## 0 = vec_ctf(vs32(b ## 0), 0);              \

  323.     b ## 1 = vec_ctf(vs32(b ## 1), 0)

  324. 

  325.     CTFX(x4, b7);

  326.     CTFX(x5, b5);

  327.     CTFX(x6, b3);

  328.     CTFX(x7, b1);

  329. 

  330. #undef CTFX

  331. 

  332.     x0   = vec_add(b70, b10);

  333.     x1   = vec_add(b50, b30);

  334.     x2   = vec_add(b70, b30);

  335.     x3   = vec_add(b50, b10);

  336.     x8   = vec_add(x2, x3);

  337.     cnst = LD_W3;

  338.     x8   = vec_madd(cnst, x8, mzero);

  339. 

  340.     cnst = LD_W8;

  341.     x0   = vec_madd(cnst, x0, mzero);

  342.     cnst = LD_W9;

  343.     x1   = vec_madd(cnst, x1, mzero);

  344.     cnst = LD_WA;

  345.     x2   = vec_madd(cnst, x2, x8);

  346.     cnst = LD_WB;

  347.     x3   = vec_madd(cnst, x3, x8);

  348. 

  349.     cnst = LD_W4;

  350.     b70  = vec_madd(cnst, b70, x0);

  351.     cnst = LD_W5;

  352.     b50  = vec_madd(cnst, b50, x1);

  353.     cnst = LD_W6;

  354.     b30  = vec_madd(cnst, b30, x1);

  355.     cnst = LD_W7;

  356.     b10  = vec_madd(cnst, b10, x0);

  357. 

  358.     b70 = vec_add(b70, x2);

  359.     b50 = vec_add(b50, x3);

  360.     b30 = vec_add(b30, x2);

  361.     b10 = vec_add(b10, x3);

  362. 

  363.     x0   = vec_add(b71, b11);

  364.     x1   = vec_add(b51, b31);

  365.     x2   = vec_add(b71, b31);

  366.     x3   = vec_add(b51, b11);

  367.     x8   = vec_add(x2, x3);

  368.     cnst = LD_W3;

  369.     x8   = vec_madd(cnst, x8, mzero);

  370. 

  371.     cnst = LD_W8;

  372.     x0   = vec_madd(cnst, x0, mzero);

  373.     cnst = LD_W9;

  374.     x1   = vec_madd(cnst, x1, mzero);

  375.     cnst = LD_WA;

  376.     x2   = vec_madd(cnst, x2, x8);

  377.     cnst = LD_WB;

  378.     x3   = vec_madd(cnst, x3, x8);

  379. 

  380.     cnst = LD_W4;

  381.     b71  = vec_madd(cnst, b71, x0);

  382.     cnst = LD_W5;

  383.     b51  = vec_madd(cnst, b51, x1);

  384.     cnst = LD_W6;

  385.     b31  = vec_madd(cnst, b31, x1);

  386.     cnst = LD_W7;

  387.     b11  = vec_madd(cnst, b11, x0);

  388. 

  389.     b71 = vec_add(b71, x2);

  390.     b51 = vec_add(b51, x3);

  391.     b31 = vec_add(b31, x2);

  392.     b11 = vec_add(b11, x3);

  393.     /* }}} */

  394. 

  395.     /* 8x8 matrix transpose (vector float[8][2]) {{{ */

  396.     x0 = VEC_FMERGEL(b00, b20);

  397.     x1 = VEC_FMERGEH(b00, b20);

  398.     x2 = VEC_FMERGEL(b10, b30);

  399.     x3 = VEC_FMERGEH(b10, b30);

  400. 

  401.     b00 = VEC_FMERGEH(x1, x3);

  402.     b10 = VEC_FMERGEL(x1, x3);

  403.     b20 = VEC_FMERGEH(x0, x2);

  404.     b30 = VEC_FMERGEL(x0, x2);

  405. 

  406.     x4 = VEC_FMERGEL(b41, b61);

  407.     x5 = VEC_FMERGEH(b41, b61);

  408.     x6 = VEC_FMERGEL(b51, b71);

  409.     x7 = VEC_FMERGEH(b51, b71);

  410. 

  411.     b41 = VEC_FMERGEH(x5, x7);

  412.     b51 = VEC_FMERGEL(x5, x7);

  413.     b61 = VEC_FMERGEH(x4, x6);

  414.     b71 = VEC_FMERGEL(x4, x6);

  415. 

  416.     x0 = VEC_FMERGEL(b01, b21);

  417.     x1 = VEC_FMERGEH(b01, b21);

  418.     x2 = VEC_FMERGEL(b11, b31);

  419.     x3 = VEC_FMERGEH(b11, b31);

  420. 

  421.     x4 = VEC_FMERGEL(b40, b60);

  422.     x5 = VEC_FMERGEH(b40, b60);

  423.     x6 = VEC_FMERGEL(b50, b70);

  424.     x7 = VEC_FMERGEH(b50, b70);

  425. 

  426.     b40 = VEC_FMERGEH(x1, x3);

  427.     b50 = VEC_FMERGEL(x1, x3);

  428.     b60 = VEC_FMERGEH(x0, x2);

  429.     b70 = VEC_FMERGEL(x0, x2);

  430. 

  431.     b01 = VEC_FMERGEH(x5, x7);

  432.     b11 = VEC_FMERGEL(x5, x7);

  433.     b21 = VEC_FMERGEH(x4, x6);

  434.     b31 = VEC_FMERGEL(x4, x6);

  435.     /* }}} */

  436. 

  437.     FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70);

  438.     FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71);

  439. 

  440.     /* round, convert back to short {{{ */

  441. #define CTS(n)                                                  \

  442.     b ## n ## 0 = vec_round(b ## n ## 0);                       \

  443.     b ## n ## 1 = vec_round(b ## n ## 1);                       \

  444.     b ## n ## 0 = ((vector float) vec_cts(b ## n ## 0, 0));     \

  445.     b ## n ## 1 = ((vector float) vec_cts(b ## n ## 1, 0));     \

  446.     b ## n ## 0 = ((vector float) vec_pack(vs32(b ## n ## 0),   \

  447.                                            vs32(b ## n ## 1))); \

  448.     vec_st(vs16(b ## n ## 0), 0, bp)

  449. 

  450.     bp = (vector signed short *) block;

  451.     CTS(0);

  452.     bp++;

  453.     CTS(1);

  454.     bp++;

  455.     CTS(2);

  456.     bp++;

  457.     CTS(3);

  458.     bp++;

  459.     CTS(4);

  460.     bp++;

  461.     CTS(5);

  462.     bp++;

  463.     CTS(6);

  464.     bp++;

  465.     CTS(7);

  466. 

  467. #undef CTS

  468.     /* }}} */

  469. }

  470. 

  471. #endif /* HAVE_ALTIVEC */

  472. 

  473. av_cold void ff_fdctdsp_init_ppc(FDCTDSPContext *c, AVCodecContext *avctx,

  474.                                  unsigned high_bit_depth)

  475. {

  476. #if HAVE_ALTIVEC

  477.     if (!PPC_ALTIVEC(av_get_cpu_flags()))

  478.         return;

  479. 

  480.     if (!high_bit_depth) {

  481.         if (avctx->dct_algo == FF_DCT_AUTO ||

  482.             avctx->dct_algo == FF_DCT_ALTIVEC) {

  483.             c->fdct = ff_fdct_altivec;

  484.         }

  485.     }

  486. #endif /* HAVE_ALTIVEC */

  487. }