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.
40016 Commits
32 Branches
541MB
Tree: b88cc5cca1
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'b88cc5cca1'
${ noResults }
FFmpeg/libavcodec/dsputil.c

1574 lines
51KB
Raw Blame History 

  1. /*

  2.  * DSP utils

  3.  * Copyright (c) 2000, 2001 Fabrice Bellard

  4.  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>

  5.  *

  6.  * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at>

  7.  *

  8.  * This file is part of Libav.

  9.  *

  10.  * Libav is free software; you can redistribute it and/or

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

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

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

  14.  *

  15.  * Libav is distributed in the hope that it will be useful,

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

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

  18.  * Lesser General Public License for more details.

  19.  *

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

  21.  * License along with Libav; if not, write to the Free Software

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

  23.  */

  24. 

  25. /**

  26.  * @file

  27.  * DSP utils

  28.  */

  29. 

  30. #include "libavutil/attributes.h"

  31. #include "libavutil/imgutils.h"

  32. #include "avcodec.h"

  33. #include "copy_block.h"

  34. #include "dct.h"

  35. #include "dsputil.h"

  36. #include "simple_idct.h"

  37. #include "faandct.h"

  38. #include "faanidct.h"

  39. #include "imgconvert.h"

  40. #include "mathops.h"

  41. #include "mpegvideo.h"

  42. #include "config.h"

  43. 

  44. uint32_t ff_square_tab[512] = { 0, };

  45. 

  46. #define BIT_DEPTH 16

  47. #include "dsputilenc_template.c"

  48. #undef BIT_DEPTH

  49. 

  50. #define BIT_DEPTH 8

  51. #include "dsputilenc_template.c"

  52. 

  53. /* Input permutation for the simple_idct_mmx */

  54. static const uint8_t simple_mmx_permutation[64] = {

  55.     0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D,

  56.     0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D,

  57.     0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D,

  58.     0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F,

  59.     0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F,

  60.     0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D,

  61.     0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F,

  62.     0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  63. };

  64. 

  65. static const uint8_t idct_sse2_row_perm[8] = { 0, 4, 1, 5, 2, 6, 3, 7 };

  66. 

  67. av_cold void ff_init_scantable(uint8_t *permutation, ScanTable *st,

  68.                                const uint8_t *src_scantable)

  69. {

  70.     int i, end;

  71. 

  72.     st->scantable = src_scantable;

  73. 

  74.     for (i = 0; i < 64; i++) {

  75.         int j = src_scantable[i];

  76.         st->permutated[i] = permutation[j];

  77.     }

  78. 

  79.     end = -1;

  80.     for (i = 0; i < 64; i++) {

  81.         int j = st->permutated[i];

  82.         if (j > end)

  83.             end = j;

  84.         st->raster_end[i] = end;

  85.     }

  86. }

  87. 

  88. av_cold void ff_init_scantable_permutation(uint8_t *idct_permutation,

  89.                                            int idct_permutation_type)

  90. {

  91.     int i;

  92. 

  93.     switch (idct_permutation_type) {

  94.     case FF_NO_IDCT_PERM:

  95.         for (i = 0; i < 64; i++)

  96.             idct_permutation[i] = i;

  97.         break;

  98.     case FF_LIBMPEG2_IDCT_PERM:

  99.         for (i = 0; i < 64; i++)

  100.             idct_permutation[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);

  101.         break;

  102.     case FF_SIMPLE_IDCT_PERM:

  103.         for (i = 0; i < 64; i++)

  104.             idct_permutation[i] = simple_mmx_permutation[i];

  105.         break;

  106.     case FF_TRANSPOSE_IDCT_PERM:

  107.         for (i = 0; i < 64; i++)

  108.             idct_permutation[i] = ((i & 7) << 3) | (i >> 3);

  109.         break;

  110.     case FF_PARTTRANS_IDCT_PERM:

  111.         for (i = 0; i < 64; i++)

  112.             idct_permutation[i] = (i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3);

  113.         break;

  114.     case FF_SSE2_IDCT_PERM:

  115.         for (i = 0; i < 64; i++)

  116.             idct_permutation[i] = (i & 0x38) | idct_sse2_row_perm[i & 7];

  117.         break;

  118.     default:

  119.         av_log(NULL, AV_LOG_ERROR,

  120.                "Internal error, IDCT permutation not set\n");

  121.     }

  122. }

  123. 

  124. static int pix_sum_c(uint8_t *pix, int line_size)

  125. {

  126.     int s = 0, i, j;

  127. 

  128.     for (i = 0; i < 16; i++) {

  129.         for (j = 0; j < 16; j += 8) {

  130.             s   += pix[0];

  131.             s   += pix[1];

  132.             s   += pix[2];

  133.             s   += pix[3];

  134.             s   += pix[4];

  135.             s   += pix[5];

  136.             s   += pix[6];

  137.             s   += pix[7];

  138.             pix += 8;

  139.         }

  140.         pix += line_size - 16;

  141.     }

  142.     return s;

  143. }

  144. 

  145. static int pix_norm1_c(uint8_t *pix, int line_size)

  146. {

  147.     int s = 0, i, j;

  148.     uint32_t *sq = ff_square_tab + 256;

  149. 

  150.     for (i = 0; i < 16; i++) {

  151.         for (j = 0; j < 16; j += 8) {

  152. #if 0

  153.             s += sq[pix[0]];

  154.             s += sq[pix[1]];

  155.             s += sq[pix[2]];

  156.             s += sq[pix[3]];

  157.             s += sq[pix[4]];

  158.             s += sq[pix[5]];

  159.             s += sq[pix[6]];

  160.             s += sq[pix[7]];

  161. #else

  162. #if HAVE_FAST_64BIT

  163.             register uint64_t x = *(uint64_t *) pix;

  164.             s += sq[x         & 0xff];

  165.             s += sq[(x >>  8) & 0xff];

  166.             s += sq[(x >> 16) & 0xff];

  167.             s += sq[(x >> 24) & 0xff];

  168.             s += sq[(x >> 32) & 0xff];

  169.             s += sq[(x >> 40) & 0xff];

  170.             s += sq[(x >> 48) & 0xff];

  171.             s += sq[(x >> 56) & 0xff];

  172. #else

  173.             register uint32_t x = *(uint32_t *) pix;

  174.             s += sq[x         & 0xff];

  175.             s += sq[(x >>  8) & 0xff];

  176.             s += sq[(x >> 16) & 0xff];

  177.             s += sq[(x >> 24) & 0xff];

  178.             x  = *(uint32_t *) (pix + 4);

  179.             s += sq[x         & 0xff];

  180.             s += sq[(x >>  8) & 0xff];

  181.             s += sq[(x >> 16) & 0xff];

  182.             s += sq[(x >> 24) & 0xff];

  183. #endif

  184. #endif

  185.             pix += 8;

  186.         }

  187.         pix += line_size - 16;

  188.     }

  189.     return s;

  190. }

  191. 

  192. static void bswap_buf(uint32_t *dst, const uint32_t *src, int w)

  193. {

  194.     int i;

  195. 

  196.     for (i = 0; i + 8 <= w; i += 8) {

  197.         dst[i + 0] = av_bswap32(src[i + 0]);

  198.         dst[i + 1] = av_bswap32(src[i + 1]);

  199.         dst[i + 2] = av_bswap32(src[i + 2]);

  200.         dst[i + 3] = av_bswap32(src[i + 3]);

  201.         dst[i + 4] = av_bswap32(src[i + 4]);

  202.         dst[i + 5] = av_bswap32(src[i + 5]);

  203.         dst[i + 6] = av_bswap32(src[i + 6]);

  204.         dst[i + 7] = av_bswap32(src[i + 7]);

  205.     }

  206.     for (; i < w; i++)

  207.         dst[i + 0] = av_bswap32(src[i + 0]);

  208. }

  209. 

  210. static void bswap16_buf(uint16_t *dst, const uint16_t *src, int len)

  211. {

  212.     while (len--)

  213.         *dst++ = av_bswap16(*src++);

  214. }

  215. 

  216. static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  217.                   int line_size, int h)

  218. {

  219.     int s = 0, i;

  220.     uint32_t *sq = ff_square_tab + 256;

  221. 

  222.     for (i = 0; i < h; i++) {

  223.         s    += sq[pix1[0] - pix2[0]];

  224.         s    += sq[pix1[1] - pix2[1]];

  225.         s    += sq[pix1[2] - pix2[2]];

  226.         s    += sq[pix1[3] - pix2[3]];

  227.         pix1 += line_size;

  228.         pix2 += line_size;

  229.     }

  230.     return s;

  231. }

  232. 

  233. static int sse8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  234.                   int line_size, int h)

  235. {

  236.     int s = 0, i;

  237.     uint32_t *sq = ff_square_tab + 256;

  238. 

  239.     for (i = 0; i < h; i++) {

  240.         s    += sq[pix1[0] - pix2[0]];

  241.         s    += sq[pix1[1] - pix2[1]];

  242.         s    += sq[pix1[2] - pix2[2]];

  243.         s    += sq[pix1[3] - pix2[3]];

  244.         s    += sq[pix1[4] - pix2[4]];

  245.         s    += sq[pix1[5] - pix2[5]];

  246.         s    += sq[pix1[6] - pix2[6]];

  247.         s    += sq[pix1[7] - pix2[7]];

  248.         pix1 += line_size;

  249.         pix2 += line_size;

  250.     }

  251.     return s;

  252. }

  253. 

  254. static int sse16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  255.                    int line_size, int h)

  256. {

  257.     int s = 0, i;

  258.     uint32_t *sq = ff_square_tab + 256;

  259. 

  260.     for (i = 0; i < h; i++) {

  261.         s += sq[pix1[0]  - pix2[0]];

  262.         s += sq[pix1[1]  - pix2[1]];

  263.         s += sq[pix1[2]  - pix2[2]];

  264.         s += sq[pix1[3]  - pix2[3]];

  265.         s += sq[pix1[4]  - pix2[4]];

  266.         s += sq[pix1[5]  - pix2[5]];

  267.         s += sq[pix1[6]  - pix2[6]];

  268.         s += sq[pix1[7]  - pix2[7]];

  269.         s += sq[pix1[8]  - pix2[8]];

  270.         s += sq[pix1[9]  - pix2[9]];

  271.         s += sq[pix1[10] - pix2[10]];

  272.         s += sq[pix1[11] - pix2[11]];

  273.         s += sq[pix1[12] - pix2[12]];

  274.         s += sq[pix1[13] - pix2[13]];

  275.         s += sq[pix1[14] - pix2[14]];

  276.         s += sq[pix1[15] - pix2[15]];

  277. 

  278.         pix1 += line_size;

  279.         pix2 += line_size;

  280.     }

  281.     return s;

  282. }

  283. 

  284. static void diff_pixels_c(int16_t *restrict block, const uint8_t *s1,

  285.                           const uint8_t *s2, int stride)

  286. {

  287.     int i;

  288. 

  289.     /* read the pixels */

  290.     for (i = 0; i < 8; i++) {

  291.         block[0] = s1[0] - s2[0];

  292.         block[1] = s1[1] - s2[1];

  293.         block[2] = s1[2] - s2[2];

  294.         block[3] = s1[3] - s2[3];

  295.         block[4] = s1[4] - s2[4];

  296.         block[5] = s1[5] - s2[5];

  297.         block[6] = s1[6] - s2[6];

  298.         block[7] = s1[7] - s2[7];

  299.         s1      += stride;

  300.         s2      += stride;

  301.         block   += 8;

  302.     }

  303. }

  304. 

  305. static void put_pixels_clamped_c(const int16_t *block, uint8_t *restrict pixels,

  306.                                  int line_size)

  307. {

  308.     int i;

  309. 

  310.     /* read the pixels */

  311.     for (i = 0; i < 8; i++) {

  312.         pixels[0] = av_clip_uint8(block[0]);

  313.         pixels[1] = av_clip_uint8(block[1]);

  314.         pixels[2] = av_clip_uint8(block[2]);

  315.         pixels[3] = av_clip_uint8(block[3]);

  316.         pixels[4] = av_clip_uint8(block[4]);

  317.         pixels[5] = av_clip_uint8(block[5]);

  318.         pixels[6] = av_clip_uint8(block[6]);

  319.         pixels[7] = av_clip_uint8(block[7]);

  320. 

  321.         pixels += line_size;

  322.         block  += 8;

  323.     }

  324. }

  325. 

  326. static void put_signed_pixels_clamped_c(const int16_t *block,

  327.                                         uint8_t *restrict pixels,

  328.                                         int line_size)

  329. {

  330.     int i, j;

  331. 

  332.     for (i = 0; i < 8; i++) {

  333.         for (j = 0; j < 8; j++) {

  334.             if (*block < -128)

  335.                 *pixels = 0;

  336.             else if (*block > 127)

  337.                 *pixels = 255;

  338.             else

  339.                 *pixels = (uint8_t) (*block + 128);

  340.             block++;

  341.             pixels++;

  342.         }

  343.         pixels += (line_size - 8);

  344.     }

  345. }

  346. 

  347. static void add_pixels_clamped_c(const int16_t *block, uint8_t *restrict pixels,

  348.                                  int line_size)

  349. {

  350.     int i;

  351. 

  352.     /* read the pixels */

  353.     for (i = 0; i < 8; i++) {

  354.         pixels[0] = av_clip_uint8(pixels[0] + block[0]);

  355.         pixels[1] = av_clip_uint8(pixels[1] + block[1]);

  356.         pixels[2] = av_clip_uint8(pixels[2] + block[2]);

  357.         pixels[3] = av_clip_uint8(pixels[3] + block[3]);

  358.         pixels[4] = av_clip_uint8(pixels[4] + block[4]);

  359.         pixels[5] = av_clip_uint8(pixels[5] + block[5]);

  360.         pixels[6] = av_clip_uint8(pixels[6] + block[6]);

  361.         pixels[7] = av_clip_uint8(pixels[7] + block[7]);

  362.         pixels   += line_size;

  363.         block    += 8;

  364.     }

  365. }

  366. 

  367. static int sum_abs_dctelem_c(int16_t *block)

  368. {

  369.     int sum = 0, i;

  370. 

  371.     for (i = 0; i < 64; i++)

  372.         sum += FFABS(block[i]);

  373.     return sum;

  374. }

  375. 

  376. static void fill_block16_c(uint8_t *block, uint8_t value, int line_size, int h)

  377. {

  378.     int i;

  379. 

  380.     for (i = 0; i < h; i++) {

  381.         memset(block, value, 16);

  382.         block += line_size;

  383.     }

  384. }

  385. 

  386. static void fill_block8_c(uint8_t *block, uint8_t value, int line_size, int h)

  387. {

  388.     int i;

  389. 

  390.     for (i = 0; i < h; i++) {

  391.         memset(block, value, 8);

  392.         block += line_size;

  393.     }

  394. }

  395. 

  396. #define avg2(a, b) ((a + b + 1) >> 1)

  397. #define avg4(a, b, c, d) ((a + b + c + d + 2) >> 2)

  398. 

  399. static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h,

  400.                    int x16, int y16, int rounder)

  401. {

  402.     const int A = (16 - x16) * (16 - y16);

  403.     const int B = (x16)      * (16 - y16);

  404.     const int C = (16 - x16) * (y16);

  405.     const int D = (x16)      * (y16);

  406.     int i;

  407. 

  408.     for (i = 0; i < h; i++) {

  409.         dst[0] = (A * src[0] + B * src[1] + C * src[stride + 0] + D * src[stride + 1] + rounder) >> 8;

  410.         dst[1] = (A * src[1] + B * src[2] + C * src[stride + 1] + D * src[stride + 2] + rounder) >> 8;

  411.         dst[2] = (A * src[2] + B * src[3] + C * src[stride + 2] + D * src[stride + 3] + rounder) >> 8;

  412.         dst[3] = (A * src[3] + B * src[4] + C * src[stride + 3] + D * src[stride + 4] + rounder) >> 8;

  413.         dst[4] = (A * src[4] + B * src[5] + C * src[stride + 4] + D * src[stride + 5] + rounder) >> 8;

  414.         dst[5] = (A * src[5] + B * src[6] + C * src[stride + 5] + D * src[stride + 6] + rounder) >> 8;

  415.         dst[6] = (A * src[6] + B * src[7] + C * src[stride + 6] + D * src[stride + 7] + rounder) >> 8;

  416.         dst[7] = (A * src[7] + B * src[8] + C * src[stride + 7] + D * src[stride + 8] + rounder) >> 8;

  417.         dst   += stride;

  418.         src   += stride;

  419.     }

  420. }

  421. 

  422. void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,

  423.               int dxx, int dxy, int dyx, int dyy, int shift, int r,

  424.               int width, int height)

  425. {

  426.     int y, vx, vy;

  427.     const int s = 1 << shift;

  428. 

  429.     width--;

  430.     height--;

  431. 

  432.     for (y = 0; y < h; y++) {

  433.         int x;

  434. 

  435.         vx = ox;

  436.         vy = oy;

  437.         for (x = 0; x < 8; x++) { // FIXME: optimize

  438.             int index;

  439.             int src_x  = vx >> 16;

  440.             int src_y  = vy >> 16;

  441.             int frac_x = src_x & (s - 1);

  442.             int frac_y = src_y & (s - 1);

  443. 

  444.             src_x >>= shift;

  445.             src_y >>= shift;

  446. 

  447.             if ((unsigned) src_x < width) {

  448.                 if ((unsigned) src_y < height) {

  449.                     index = src_x + src_y * stride;

  450.                     dst[y * stride + x] =

  451.                         ((src[index]                        * (s - frac_x) +

  452.                           src[index + 1]          * frac_x) * (s - frac_y) +

  453.                          (src[index + stride]               * (s - frac_x) +

  454.                           src[index + stride + 1] * frac_x) *      frac_y  +

  455.                          r) >> (shift * 2);

  456.                 } else {

  457.                     index = src_x + av_clip(src_y, 0, height) * stride;

  458.                     dst[y * stride + x] =

  459.                         ((src[index]               * (s - frac_x) +

  460.                           src[index + 1] * frac_x) *  s           +

  461.                          r) >> (shift * 2);

  462.                 }

  463.             } else {

  464.                 if ((unsigned) src_y < height) {

  465.                     index = av_clip(src_x, 0, width) + src_y * stride;

  466.                     dst[y * stride + x] =

  467.                         ((src[index]                    * (s - frac_y) +

  468.                           src[index + stride] * frac_y) *  s           +

  469.                          r) >> (shift * 2);

  470.                 } else {

  471.                     index = av_clip(src_x, 0, width) +

  472.                             av_clip(src_y, 0, height) * stride;

  473.                     dst[y * stride + x] = src[index];

  474.                 }

  475.             }

  476. 

  477.             vx += dxx;

  478.             vy += dyx;

  479.         }

  480.         ox += dxy;

  481.         oy += dyy;

  482.     }

  483. }

  484. 

  485. static inline int pix_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  486.                               int line_size, int h)

  487. {

  488.     int s = 0, i;

  489. 

  490.     for (i = 0; i < h; i++) {

  491.         s    += abs(pix1[0]  - pix2[0]);

  492.         s    += abs(pix1[1]  - pix2[1]);

  493.         s    += abs(pix1[2]  - pix2[2]);

  494.         s    += abs(pix1[3]  - pix2[3]);

  495.         s    += abs(pix1[4]  - pix2[4]);

  496.         s    += abs(pix1[5]  - pix2[5]);

  497.         s    += abs(pix1[6]  - pix2[6]);

  498.         s    += abs(pix1[7]  - pix2[7]);

  499.         s    += abs(pix1[8]  - pix2[8]);

  500.         s    += abs(pix1[9]  - pix2[9]);

  501.         s    += abs(pix1[10] - pix2[10]);

  502.         s    += abs(pix1[11] - pix2[11]);

  503.         s    += abs(pix1[12] - pix2[12]);

  504.         s    += abs(pix1[13] - pix2[13]);

  505.         s    += abs(pix1[14] - pix2[14]);

  506.         s    += abs(pix1[15] - pix2[15]);

  507.         pix1 += line_size;

  508.         pix2 += line_size;

  509.     }

  510.     return s;

  511. }

  512. 

  513. static int pix_abs16_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  514.                           int line_size, int h)

  515. {

  516.     int s = 0, i;

  517. 

  518.     for (i = 0; i < h; i++) {

  519.         s    += abs(pix1[0]  - avg2(pix2[0],  pix2[1]));

  520.         s    += abs(pix1[1]  - avg2(pix2[1],  pix2[2]));

  521.         s    += abs(pix1[2]  - avg2(pix2[2],  pix2[3]));

  522.         s    += abs(pix1[3]  - avg2(pix2[3],  pix2[4]));

  523.         s    += abs(pix1[4]  - avg2(pix2[4],  pix2[5]));

  524.         s    += abs(pix1[5]  - avg2(pix2[5],  pix2[6]));

  525.         s    += abs(pix1[6]  - avg2(pix2[6],  pix2[7]));

  526.         s    += abs(pix1[7]  - avg2(pix2[7],  pix2[8]));

  527.         s    += abs(pix1[8]  - avg2(pix2[8],  pix2[9]));

  528.         s    += abs(pix1[9]  - avg2(pix2[9],  pix2[10]));

  529.         s    += abs(pix1[10] - avg2(pix2[10], pix2[11]));

  530.         s    += abs(pix1[11] - avg2(pix2[11], pix2[12]));

  531.         s    += abs(pix1[12] - avg2(pix2[12], pix2[13]));

  532.         s    += abs(pix1[13] - avg2(pix2[13], pix2[14]));

  533.         s    += abs(pix1[14] - avg2(pix2[14], pix2[15]));

  534.         s    += abs(pix1[15] - avg2(pix2[15], pix2[16]));

  535.         pix1 += line_size;

  536.         pix2 += line_size;

  537.     }

  538.     return s;

  539. }

  540. 

  541. static int pix_abs16_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  542.                           int line_size, int h)

  543. {

  544.     int s = 0, i;

  545.     uint8_t *pix3 = pix2 + line_size;

  546. 

  547.     for (i = 0; i < h; i++) {

  548.         s    += abs(pix1[0]  - avg2(pix2[0],  pix3[0]));

  549.         s    += abs(pix1[1]  - avg2(pix2[1],  pix3[1]));

  550.         s    += abs(pix1[2]  - avg2(pix2[2],  pix3[2]));

  551.         s    += abs(pix1[3]  - avg2(pix2[3],  pix3[3]));

  552.         s    += abs(pix1[4]  - avg2(pix2[4],  pix3[4]));

  553.         s    += abs(pix1[5]  - avg2(pix2[5],  pix3[5]));

  554.         s    += abs(pix1[6]  - avg2(pix2[6],  pix3[6]));

  555.         s    += abs(pix1[7]  - avg2(pix2[7],  pix3[7]));

  556.         s    += abs(pix1[8]  - avg2(pix2[8],  pix3[8]));

  557.         s    += abs(pix1[9]  - avg2(pix2[9],  pix3[9]));

  558.         s    += abs(pix1[10] - avg2(pix2[10], pix3[10]));

  559.         s    += abs(pix1[11] - avg2(pix2[11], pix3[11]));

  560.         s    += abs(pix1[12] - avg2(pix2[12], pix3[12]));

  561.         s    += abs(pix1[13] - avg2(pix2[13], pix3[13]));

  562.         s    += abs(pix1[14] - avg2(pix2[14], pix3[14]));

  563.         s    += abs(pix1[15] - avg2(pix2[15], pix3[15]));

  564.         pix1 += line_size;

  565.         pix2 += line_size;

  566.         pix3 += line_size;

  567.     }

  568.     return s;

  569. }

  570. 

  571. static int pix_abs16_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  572.                            int line_size, int h)

  573. {

  574.     int s = 0, i;

  575.     uint8_t *pix3 = pix2 + line_size;

  576. 

  577.     for (i = 0; i < h; i++) {

  578.         s    += abs(pix1[0]  - avg4(pix2[0],  pix2[1],  pix3[0],  pix3[1]));

  579.         s    += abs(pix1[1]  - avg4(pix2[1],  pix2[2],  pix3[1],  pix3[2]));

  580.         s    += abs(pix1[2]  - avg4(pix2[2],  pix2[3],  pix3[2],  pix3[3]));

  581.         s    += abs(pix1[3]  - avg4(pix2[3],  pix2[4],  pix3[3],  pix3[4]));

  582.         s    += abs(pix1[4]  - avg4(pix2[4],  pix2[5],  pix3[4],  pix3[5]));

  583.         s    += abs(pix1[5]  - avg4(pix2[5],  pix2[6],  pix3[5],  pix3[6]));

  584.         s    += abs(pix1[6]  - avg4(pix2[6],  pix2[7],  pix3[6],  pix3[7]));

  585.         s    += abs(pix1[7]  - avg4(pix2[7],  pix2[8],  pix3[7],  pix3[8]));

  586.         s    += abs(pix1[8]  - avg4(pix2[8],  pix2[9],  pix3[8],  pix3[9]));

  587.         s    += abs(pix1[9]  - avg4(pix2[9],  pix2[10], pix3[9],  pix3[10]));

  588.         s    += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));

  589.         s    += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));

  590.         s    += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));

  591.         s    += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));

  592.         s    += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));

  593.         s    += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));

  594.         pix1 += line_size;

  595.         pix2 += line_size;

  596.         pix3 += line_size;

  597.     }

  598.     return s;

  599. }

  600. 

  601. static inline int pix_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  602.                              int line_size, int h)

  603. {

  604.     int s = 0, i;

  605. 

  606.     for (i = 0; i < h; i++) {

  607.         s    += abs(pix1[0] - pix2[0]);

  608.         s    += abs(pix1[1] - pix2[1]);

  609.         s    += abs(pix1[2] - pix2[2]);

  610.         s    += abs(pix1[3] - pix2[3]);

  611.         s    += abs(pix1[4] - pix2[4]);

  612.         s    += abs(pix1[5] - pix2[5]);

  613.         s    += abs(pix1[6] - pix2[6]);

  614.         s    += abs(pix1[7] - pix2[7]);

  615.         pix1 += line_size;

  616.         pix2 += line_size;

  617.     }

  618.     return s;

  619. }

  620. 

  621. static int pix_abs8_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  622.                          int line_size, int h)

  623. {

  624.     int s = 0, i;

  625. 

  626.     for (i = 0; i < h; i++) {

  627.         s    += abs(pix1[0] - avg2(pix2[0], pix2[1]));

  628.         s    += abs(pix1[1] - avg2(pix2[1], pix2[2]));

  629.         s    += abs(pix1[2] - avg2(pix2[2], pix2[3]));

  630.         s    += abs(pix1[3] - avg2(pix2[3], pix2[4]));

  631.         s    += abs(pix1[4] - avg2(pix2[4], pix2[5]));

  632.         s    += abs(pix1[5] - avg2(pix2[5], pix2[6]));

  633.         s    += abs(pix1[6] - avg2(pix2[6], pix2[7]));

  634.         s    += abs(pix1[7] - avg2(pix2[7], pix2[8]));

  635.         pix1 += line_size;

  636.         pix2 += line_size;

  637.     }

  638.     return s;

  639. }

  640. 

  641. static int pix_abs8_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  642.                          int line_size, int h)

  643. {

  644.     int s = 0, i;

  645.     uint8_t *pix3 = pix2 + line_size;

  646. 

  647.     for (i = 0; i < h; i++) {

  648.         s    += abs(pix1[0] - avg2(pix2[0], pix3[0]));

  649.         s    += abs(pix1[1] - avg2(pix2[1], pix3[1]));

  650.         s    += abs(pix1[2] - avg2(pix2[2], pix3[2]));

  651.         s    += abs(pix1[3] - avg2(pix2[3], pix3[3]));

  652.         s    += abs(pix1[4] - avg2(pix2[4], pix3[4]));

  653.         s    += abs(pix1[5] - avg2(pix2[5], pix3[5]));

  654.         s    += abs(pix1[6] - avg2(pix2[6], pix3[6]));

  655.         s    += abs(pix1[7] - avg2(pix2[7], pix3[7]));

  656.         pix1 += line_size;

  657.         pix2 += line_size;

  658.         pix3 += line_size;

  659.     }

  660.     return s;

  661. }

  662. 

  663. static int pix_abs8_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  664.                           int line_size, int h)

  665. {

  666.     int s = 0, i;

  667.     uint8_t *pix3 = pix2 + line_size;

  668. 

  669.     for (i = 0; i < h; i++) {

  670.         s    += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));

  671.         s    += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));

  672.         s    += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));

  673.         s    += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));

  674.         s    += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));

  675.         s    += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));

  676.         s    += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));

  677.         s    += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));

  678.         pix1 += line_size;

  679.         pix2 += line_size;

  680.         pix3 += line_size;

  681.     }

  682.     return s;

  683. }

  684. 

  685. static int nsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)

  686. {

  687.     int score1 = 0, score2 = 0, x, y;

  688. 

  689.     for (y = 0; y < h; y++) {

  690.         for (x = 0; x < 16; x++)

  691.             score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);

  692.         if (y + 1 < h) {

  693.             for (x = 0; x < 15; x++)

  694.                 score2 += FFABS(s1[x]     - s1[x + stride] -

  695.                                 s1[x + 1] + s1[x + stride + 1]) -

  696.                           FFABS(s2[x]     - s2[x + stride] -

  697.                                 s2[x + 1] + s2[x + stride + 1]);

  698.         }

  699.         s1 += stride;

  700.         s2 += stride;

  701.     }

  702. 

  703.     if (c)

  704.         return score1 + FFABS(score2) * c->avctx->nsse_weight;

  705.     else

  706.         return score1 + FFABS(score2) * 8;

  707. }

  708. 

  709. static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)

  710. {

  711.     int score1 = 0, score2 = 0, x, y;

  712. 

  713.     for (y = 0; y < h; y++) {

  714.         for (x = 0; x < 8; x++)

  715.             score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);

  716.         if (y + 1 < h) {

  717.             for (x = 0; x < 7; x++)

  718.                 score2 += FFABS(s1[x]     - s1[x + stride] -

  719.                                 s1[x + 1] + s1[x + stride + 1]) -

  720.                           FFABS(s2[x]     - s2[x + stride] -

  721.                                 s2[x + 1] + s2[x + stride + 1]);

  722.         }

  723.         s1 += stride;

  724.         s2 += stride;

  725.     }

  726. 

  727.     if (c)

  728.         return score1 + FFABS(score2) * c->avctx->nsse_weight;

  729.     else

  730.         return score1 + FFABS(score2) * 8;

  731. }

  732. 

  733. static int try_8x8basis_c(int16_t rem[64], int16_t weight[64],

  734.                           int16_t basis[64], int scale)

  735. {

  736.     int i;

  737.     unsigned int sum = 0;

  738. 

  739.     for (i = 0; i < 8 * 8; i++) {

  740.         int b = rem[i] + ((basis[i] * scale +

  741.                            (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  742.                           (BASIS_SHIFT - RECON_SHIFT));

  743.         int w = weight[i];

  744.         b >>= RECON_SHIFT;

  745.         assert(-512 < b && b < 512);

  746. 

  747.         sum += (w * b) * (w * b) >> 4;

  748.     }

  749.     return sum >> 2;

  750. }

  751. 

  752. static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale)

  753. {

  754.     int i;

  755. 

  756.     for (i = 0; i < 8 * 8; i++)

  757.         rem[i] += (basis[i] * scale +

  758.                    (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  759.                   (BASIS_SHIFT - RECON_SHIFT);

  760. }

  761. 

  762. static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,

  763.                     int stride, int h)

  764. {

  765.     return 0;

  766. }

  767. 

  768. void ff_set_cmp(DSPContext *c, me_cmp_func *cmp, int type)

  769. {

  770.     int i;

  771. 

  772.     memset(cmp, 0, sizeof(void *) * 6);

  773. 

  774.     for (i = 0; i < 6; i++) {

  775.         switch (type & 0xFF) {

  776.         case FF_CMP_SAD:

  777.             cmp[i] = c->sad[i];

  778.             break;

  779.         case FF_CMP_SATD:

  780.             cmp[i] = c->hadamard8_diff[i];

  781.             break;

  782.         case FF_CMP_SSE:

  783.             cmp[i] = c->sse[i];

  784.             break;

  785.         case FF_CMP_DCT:

  786.             cmp[i] = c->dct_sad[i];

  787.             break;

  788.         case FF_CMP_DCT264:

  789.             cmp[i] = c->dct264_sad[i];

  790.             break;

  791.         case FF_CMP_DCTMAX:

  792.             cmp[i] = c->dct_max[i];

  793.             break;

  794.         case FF_CMP_PSNR:

  795.             cmp[i] = c->quant_psnr[i];

  796.             break;

  797.         case FF_CMP_BIT:

  798.             cmp[i] = c->bit[i];

  799.             break;

  800.         case FF_CMP_RD:

  801.             cmp[i] = c->rd[i];

  802.             break;

  803.         case FF_CMP_VSAD:

  804.             cmp[i] = c->vsad[i];

  805.             break;

  806.         case FF_CMP_VSSE:

  807.             cmp[i] = c->vsse[i];

  808.             break;

  809.         case FF_CMP_ZERO:

  810.             cmp[i] = zero_cmp;

  811.             break;

  812.         case FF_CMP_NSSE:

  813.             cmp[i] = c->nsse[i];

  814.             break;

  815.         default:

  816.             av_log(NULL, AV_LOG_ERROR,

  817.                    "internal error in cmp function selection\n");

  818.         }

  819.     }

  820. }

  821. 

  822. #define BUTTERFLY2(o1, o2, i1, i2)              \

  823.     o1 = (i1) + (i2);                           \

  824.     o2 = (i1) - (i2);

  825. 

  826. #define BUTTERFLY1(x, y)                        \

  827.     {                                           \

  828.         int a, b;                               \

  829.         a = x;                                  \

  830.         b = y;                                  \

  831.         x = a + b;                              \

  832.         y = a - b;                              \

  833.     }

  834. 

  835. #define BUTTERFLYA(x, y) (FFABS((x) + (y)) + FFABS((x) - (y)))

  836. 

  837. static int hadamard8_diff8x8_c(MpegEncContext *s, uint8_t *dst,

  838.                                uint8_t *src, int stride, int h)

  839. {

  840.     int i, temp[64], sum = 0;

  841. 

  842.     assert(h == 8);

  843. 

  844.     for (i = 0; i < 8; i++) {

  845.         // FIXME: try pointer walks

  846.         BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],

  847.                    src[stride * i + 0] - dst[stride * i + 0],

  848.                    src[stride * i + 1] - dst[stride * i + 1]);

  849.         BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],

  850.                    src[stride * i + 2] - dst[stride * i + 2],

  851.                    src[stride * i + 3] - dst[stride * i + 3]);

  852.         BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],

  853.                    src[stride * i + 4] - dst[stride * i + 4],

  854.                    src[stride * i + 5] - dst[stride * i + 5]);

  855.         BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],

  856.                    src[stride * i + 6] - dst[stride * i + 6],

  857.                    src[stride * i + 7] - dst[stride * i + 7]);

  858. 

  859.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);

  860.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);

  861.         BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);

  862.         BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);

  863. 

  864.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);

  865.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);

  866.         BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);

  867.         BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);

  868.     }

  869. 

  870.     for (i = 0; i < 8; i++) {

  871.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);

  872.         BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);

  873.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);

  874.         BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);

  875. 

  876.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);

  877.         BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);

  878.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);

  879.         BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);

  880. 

  881.         sum += BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i]) +

  882.                BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i]) +

  883.                BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i]) +

  884.                BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);

  885.     }

  886.     return sum;

  887. }

  888. 

  889. static int hadamard8_intra8x8_c(MpegEncContext *s, uint8_t *src,

  890.                                 uint8_t *dummy, int stride, int h)

  891. {

  892.     int i, temp[64], sum = 0;

  893. 

  894.     assert(h == 8);

  895. 

  896.     for (i = 0; i < 8; i++) {

  897.         // FIXME: try pointer walks

  898.         BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],

  899.                    src[stride * i + 0], src[stride * i + 1]);

  900.         BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],

  901.                    src[stride * i + 2], src[stride * i + 3]);

  902.         BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],

  903.                    src[stride * i + 4], src[stride * i + 5]);

  904.         BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],

  905.                    src[stride * i + 6], src[stride * i + 7]);

  906. 

  907.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);

  908.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);

  909.         BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);

  910.         BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);

  911. 

  912.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);

  913.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);

  914.         BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);

  915.         BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);

  916.     }

  917. 

  918.     for (i = 0; i < 8; i++) {

  919.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);

  920.         BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);

  921.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);

  922.         BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);

  923. 

  924.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);

  925.         BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);

  926.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);

  927.         BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);

  928. 

  929.         sum +=

  930.             BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i])

  931.             + BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i])

  932.             + BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i])

  933.             + BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);

  934.     }

  935. 

  936.     sum -= FFABS(temp[8 * 0] + temp[8 * 4]); // -mean

  937. 

  938.     return sum;

  939. }

  940. 

  941. static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,

  942.                         uint8_t *src2, int stride, int h)

  943. {

  944.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  945. 

  946.     assert(h == 8);

  947. 

  948.     s->dsp.diff_pixels(temp, src1, src2, stride);

  949.     s->dsp.fdct(temp);

  950.     return s->dsp.sum_abs_dctelem(temp);

  951. }

  952. 

  953. #if CONFIG_GPL

  954. #define DCT8_1D                                         \

  955.     {                                                   \

  956.         const int s07 = SRC(0) + SRC(7);                \

  957.         const int s16 = SRC(1) + SRC(6);                \

  958.         const int s25 = SRC(2) + SRC(5);                \

  959.         const int s34 = SRC(3) + SRC(4);                \

  960.         const int a0  = s07 + s34;                      \

  961.         const int a1  = s16 + s25;                      \

  962.         const int a2  = s07 - s34;                      \

  963.         const int a3  = s16 - s25;                      \

  964.         const int d07 = SRC(0) - SRC(7);                \

  965.         const int d16 = SRC(1) - SRC(6);                \

  966.         const int d25 = SRC(2) - SRC(5);                \

  967.         const int d34 = SRC(3) - SRC(4);                \

  968.         const int a4  = d16 + d25 + (d07 + (d07 >> 1)); \

  969.         const int a5  = d07 - d34 - (d25 + (d25 >> 1)); \

  970.         const int a6  = d07 + d34 - (d16 + (d16 >> 1)); \

  971.         const int a7  = d16 - d25 + (d34 + (d34 >> 1)); \

  972.         DST(0, a0 + a1);                                \

  973.         DST(1, a4 + (a7 >> 2));                         \

  974.         DST(2, a2 + (a3 >> 1));                         \

  975.         DST(3, a5 + (a6 >> 2));                         \

  976.         DST(4, a0 - a1);                                \

  977.         DST(5, a6 - (a5 >> 2));                         \

  978.         DST(6, (a2 >> 1) - a3);                         \

  979.         DST(7, (a4 >> 2) - a7);                         \

  980.     }

  981. 

  982. static int dct264_sad8x8_c(MpegEncContext *s, uint8_t *src1,

  983.                            uint8_t *src2, int stride, int h)

  984. {

  985.     int16_t dct[8][8];

  986.     int i, sum = 0;

  987. 

  988.     s->dsp.diff_pixels(dct[0], src1, src2, stride);

  989. 

  990. #define SRC(x) dct[i][x]

  991. #define DST(x, v) dct[i][x] = v

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

  993.         DCT8_1D

  994. #undef SRC

  995. #undef DST

  996. 

  997. #define SRC(x) dct[x][i]

  998. #define DST(x, v) sum += FFABS(v)

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

  1000.             DCT8_1D

  1001. #undef SRC

  1002. #undef DST

  1003.             return sum;

  1004. }

  1005. #endif

  1006. 

  1007. static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,

  1008.                         uint8_t *src2, int stride, int h)

  1009. {

  1010.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  1011.     int sum = 0, i;

  1012. 

  1013.     assert(h == 8);

  1014. 

  1015.     s->dsp.diff_pixels(temp, src1, src2, stride);

  1016.     s->dsp.fdct(temp);

  1017. 

  1018.     for (i = 0; i < 64; i++)

  1019.         sum = FFMAX(sum, FFABS(temp[i]));

  1020. 

  1021.     return sum;

  1022. }

  1023. 

  1024. static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,

  1025.                            uint8_t *src2, int stride, int h)

  1026. {

  1027.     LOCAL_ALIGNED_16(int16_t, temp, [64 * 2]);

  1028.     int16_t *const bak = temp + 64;

  1029.     int sum = 0, i;

  1030. 

  1031.     assert(h == 8);

  1032.     s->mb_intra = 0;

  1033. 

  1034.     s->dsp.diff_pixels(temp, src1, src2, stride);

  1035. 

  1036.     memcpy(bak, temp, 64 * sizeof(int16_t));

  1037. 

  1038.     s->block_last_index[0 /* FIXME */] =

  1039.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  1040.     s->dct_unquantize_inter(s, temp, 0, s->qscale);

  1041.     ff_simple_idct_8(temp); // FIXME

  1042. 

  1043.     for (i = 0; i < 64; i++)

  1044.         sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);

  1045. 

  1046.     return sum;

  1047. }

  1048. 

  1049. static int rd8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,

  1050.                    int stride, int h)

  1051. {

  1052.     const uint8_t *scantable = s->intra_scantable.permutated;

  1053.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  1054.     LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]);

  1055.     LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]);

  1056.     int i, last, run, bits, level, distortion, start_i;

  1057.     const int esc_length = s->ac_esc_length;

  1058.     uint8_t *length, *last_length;

  1059. 

  1060.     assert(h == 8);

  1061. 

  1062.     copy_block8(lsrc1, src1, 8, stride, 8);

  1063.     copy_block8(lsrc2, src2, 8, stride, 8);

  1064. 

  1065.     s->dsp.diff_pixels(temp, lsrc1, lsrc2, 8);

  1066. 

  1067.     s->block_last_index[0 /* FIXME */] =

  1068.     last                               =

  1069.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  1070. 

  1071.     bits = 0;

  1072. 

  1073.     if (s->mb_intra) {

  1074.         start_i     = 1;

  1075.         length      = s->intra_ac_vlc_length;

  1076.         last_length = s->intra_ac_vlc_last_length;

  1077.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  1078.     } else {

  1079.         start_i     = 0;

  1080.         length      = s->inter_ac_vlc_length;

  1081.         last_length = s->inter_ac_vlc_last_length;

  1082.     }

  1083. 

  1084.     if (last >= start_i) {

  1085.         run = 0;

  1086.         for (i = start_i; i < last; i++) {

  1087.             int j = scantable[i];

  1088.             level = temp[j];

  1089. 

  1090.             if (level) {

  1091.                 level += 64;

  1092.                 if ((level & (~127)) == 0)

  1093.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  1094.                 else

  1095.                     bits += esc_length;

  1096.                 run = 0;

  1097.             } else

  1098.                 run++;

  1099.         }

  1100.         i = scantable[last];

  1101. 

  1102.         level = temp[i] + 64;

  1103. 

  1104.         assert(level - 64);

  1105. 

  1106.         if ((level & (~127)) == 0) {

  1107.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  1108.         } else

  1109.             bits += esc_length;

  1110.     }

  1111. 

  1112.     if (last >= 0) {

  1113.         if (s->mb_intra)

  1114.             s->dct_unquantize_intra(s, temp, 0, s->qscale);

  1115.         else

  1116.             s->dct_unquantize_inter(s, temp, 0, s->qscale);

  1117.     }

  1118. 

  1119.     s->dsp.idct_add(lsrc2, 8, temp);

  1120. 

  1121.     distortion = s->dsp.sse[1](NULL, lsrc2, lsrc1, 8, 8);

  1122. 

  1123.     return distortion + ((bits * s->qscale * s->qscale * 109 + 64) >> 7);

  1124. }

  1125. 

  1126. static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,

  1127.                     int stride, int h)

  1128. {

  1129.     const uint8_t *scantable = s->intra_scantable.permutated;

  1130.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  1131.     int i, last, run, bits, level, start_i;

  1132.     const int esc_length = s->ac_esc_length;

  1133.     uint8_t *length, *last_length;

  1134. 

  1135.     assert(h == 8);

  1136. 

  1137.     s->dsp.diff_pixels(temp, src1, src2, stride);

  1138. 

  1139.     s->block_last_index[0 /* FIXME */] =

  1140.     last                               =

  1141.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  1142. 

  1143.     bits = 0;

  1144. 

  1145.     if (s->mb_intra) {

  1146.         start_i     = 1;

  1147.         length      = s->intra_ac_vlc_length;

  1148.         last_length = s->intra_ac_vlc_last_length;

  1149.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  1150.     } else {

  1151.         start_i     = 0;

  1152.         length      = s->inter_ac_vlc_length;

  1153.         last_length = s->inter_ac_vlc_last_length;

  1154.     }

  1155. 

  1156.     if (last >= start_i) {

  1157.         run = 0;

  1158.         for (i = start_i; i < last; i++) {

  1159.             int j = scantable[i];

  1160.             level = temp[j];

  1161. 

  1162.             if (level) {

  1163.                 level += 64;

  1164.                 if ((level & (~127)) == 0)

  1165.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  1166.                 else

  1167.                     bits += esc_length;

  1168.                 run = 0;

  1169.             } else

  1170.                 run++;

  1171.         }

  1172.         i = scantable[last];

  1173. 

  1174.         level = temp[i] + 64;

  1175. 

  1176.         assert(level - 64);

  1177. 

  1178.         if ((level & (~127)) == 0)

  1179.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  1180.         else

  1181.             bits += esc_length;

  1182.     }

  1183. 

  1184.     return bits;

  1185. }

  1186. 

  1187. #define VSAD_INTRA(size)                                                \

  1188. static int vsad_intra ## size ## _c(MpegEncContext *c,                  \

  1189.                                     uint8_t *s, uint8_t *dummy,         \

  1190.                                     int stride, int h)                  \

  1191. {                                                                       \

  1192.     int score = 0, x, y;                                                \

  1193.                                                                         \

  1194.     for (y = 1; y < h; y++) {                                           \

  1195.         for (x = 0; x < size; x += 4) {                                 \

  1196.             score += FFABS(s[x]     - s[x + stride])     +              \

  1197.                      FFABS(s[x + 1] - s[x + stride + 1]) +              \

  1198.                      FFABS(s[x + 2] - s[x + 2 + stride]) +              \

  1199.                      FFABS(s[x + 3] - s[x + 3 + stride]);               \

  1200.         }                                                               \

  1201.         s += stride;                                                    \

  1202.     }                                                                   \

  1203.                                                                         \

  1204.     return score;                                                       \

  1205. }

  1206. VSAD_INTRA(8)

  1207. VSAD_INTRA(16)

  1208. 

  1209. static int vsad16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  1210.                     int stride, int h)

  1211. {

  1212.     int score = 0, x, y;

  1213. 

  1214.     for (y = 1; y < h; y++) {

  1215.         for (x = 0; x < 16; x++)

  1216.             score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  1217.         s1 += stride;

  1218.         s2 += stride;

  1219.     }

  1220. 

  1221.     return score;

  1222. }

  1223. 

  1224. #define SQ(a) ((a) * (a))

  1225. #define VSSE_INTRA(size)                                                \

  1226. static int vsse_intra ## size ## _c(MpegEncContext *c,                  \

  1227.                                     uint8_t *s, uint8_t *dummy,         \

  1228.                                     int stride, int h)                  \

  1229. {                                                                       \

  1230.     int score = 0, x, y;                                                \

  1231.                                                                         \

  1232.     for (y = 1; y < h; y++) {                                           \

  1233.         for (x = 0; x < size; x += 4) {                                 \

  1234.             score += SQ(s[x]     - s[x + stride]) +                     \

  1235.                      SQ(s[x + 1] - s[x + stride + 1]) +                 \

  1236.                      SQ(s[x + 2] - s[x + stride + 2]) +                 \

  1237.                      SQ(s[x + 3] - s[x + stride + 3]);                  \

  1238.         }                                                               \

  1239.         s += stride;                                                    \

  1240.     }                                                                   \

  1241.                                                                         \

  1242.     return score;                                                       \

  1243. }

  1244. VSSE_INTRA(8)

  1245. VSSE_INTRA(16)

  1246. 

  1247. static int vsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  1248.                     int stride, int h)

  1249. {

  1250.     int score = 0, x, y;

  1251. 

  1252.     for (y = 1; y < h; y++) {

  1253.         for (x = 0; x < 16; x++)

  1254.             score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  1255.         s1 += stride;

  1256.         s2 += stride;

  1257.     }

  1258. 

  1259.     return score;

  1260. }

  1261. 

  1262. #define WRAPPER8_16_SQ(name8, name16)                                   \

  1263. static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src,        \

  1264.                   int stride, int h)                                    \

  1265. {                                                                       \

  1266.     int score = 0;                                                      \

  1267.                                                                         \

  1268.     score += name8(s, dst, src, stride, 8);                             \

  1269.     score += name8(s, dst + 8, src + 8, stride, 8);                     \

  1270.     if (h == 16) {                                                      \

  1271.         dst   += 8 * stride;                                            \

  1272.         src   += 8 * stride;                                            \

  1273.         score += name8(s, dst, src, stride, 8);                         \

  1274.         score += name8(s, dst + 8, src + 8, stride, 8);                 \

  1275.     }                                                                   \

  1276.     return score;                                                       \

  1277. }

  1278. 

  1279. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  1280. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  1281. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  1282. #if CONFIG_GPL

  1283. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  1284. #endif

  1285. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  1286. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  1287. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  1288. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  1289. 

  1290. static inline uint32_t clipf_c_one(uint32_t a, uint32_t mini,

  1291.                                    uint32_t maxi, uint32_t maxisign)

  1292. {

  1293.     if (a > mini)

  1294.         return mini;

  1295.     else if ((a ^ (1U << 31)) > maxisign)

  1296.         return maxi;

  1297.     else

  1298.         return a;

  1299. }

  1300. 

  1301. static void vector_clipf_c_opposite_sign(float *dst, const float *src,

  1302.                                          float *min, float *max, int len)

  1303. {

  1304.     int i;

  1305.     uint32_t mini        = *(uint32_t *) min;

  1306.     uint32_t maxi        = *(uint32_t *) max;

  1307.     uint32_t maxisign    = maxi ^ (1U << 31);

  1308.     uint32_t *dsti       = (uint32_t *) dst;

  1309.     const uint32_t *srci = (const uint32_t *) src;

  1310. 

  1311.     for (i = 0; i < len; i += 8) {

  1312.         dsti[i + 0] = clipf_c_one(srci[i + 0], mini, maxi, maxisign);

  1313.         dsti[i + 1] = clipf_c_one(srci[i + 1], mini, maxi, maxisign);

  1314.         dsti[i + 2] = clipf_c_one(srci[i + 2], mini, maxi, maxisign);

  1315.         dsti[i + 3] = clipf_c_one(srci[i + 3], mini, maxi, maxisign);

  1316.         dsti[i + 4] = clipf_c_one(srci[i + 4], mini, maxi, maxisign);

  1317.         dsti[i + 5] = clipf_c_one(srci[i + 5], mini, maxi, maxisign);

  1318.         dsti[i + 6] = clipf_c_one(srci[i + 6], mini, maxi, maxisign);

  1319.         dsti[i + 7] = clipf_c_one(srci[i + 7], mini, maxi, maxisign);

  1320.     }

  1321. }

  1322. 

  1323. static void vector_clipf_c(float *dst, const float *src,

  1324.                            float min, float max, int len)

  1325. {

  1326.     int i;

  1327. 

  1328.     if (min < 0 && max > 0) {

  1329.         vector_clipf_c_opposite_sign(dst, src, &min, &max, len);

  1330.     } else {

  1331.         for (i = 0; i < len; i += 8) {

  1332.             dst[i]     = av_clipf(src[i], min, max);

  1333.             dst[i + 1] = av_clipf(src[i + 1], min, max);

  1334.             dst[i + 2] = av_clipf(src[i + 2], min, max);

  1335.             dst[i + 3] = av_clipf(src[i + 3], min, max);

  1336.             dst[i + 4] = av_clipf(src[i + 4], min, max);

  1337.             dst[i + 5] = av_clipf(src[i + 5], min, max);

  1338.             dst[i + 6] = av_clipf(src[i + 6], min, max);

  1339.             dst[i + 7] = av_clipf(src[i + 7], min, max);

  1340.         }

  1341.     }

  1342. }

  1343. 

  1344. static int32_t scalarproduct_int16_c(const int16_t *v1, const int16_t *v2,

  1345.                                      int order)

  1346. {

  1347.     int res = 0;

  1348. 

  1349.     while (order--)

  1350.         res += *v1++ **v2++;

  1351. 

  1352.     return res;

  1353. }

  1354. 

  1355. static void vector_clip_int32_c(int32_t *dst, const int32_t *src, int32_t min,

  1356.                                 int32_t max, unsigned int len)

  1357. {

  1358.     do {

  1359.         *dst++ = av_clip(*src++, min, max);

  1360.         *dst++ = av_clip(*src++, min, max);

  1361.         *dst++ = av_clip(*src++, min, max);

  1362.         *dst++ = av_clip(*src++, min, max);

  1363.         *dst++ = av_clip(*src++, min, max);

  1364.         *dst++ = av_clip(*src++, min, max);

  1365.         *dst++ = av_clip(*src++, min, max);

  1366.         *dst++ = av_clip(*src++, min, max);

  1367.         len   -= 8;

  1368.     } while (len > 0);

  1369. }

  1370. 

  1371. static void jref_idct_put(uint8_t *dest, int line_size, int16_t *block)

  1372. {

  1373.     ff_j_rev_dct(block);

  1374.     put_pixels_clamped_c(block, dest, line_size);

  1375. }

  1376. 

  1377. static void jref_idct_add(uint8_t *dest, int line_size, int16_t *block)

  1378. {

  1379.     ff_j_rev_dct(block);

  1380.     add_pixels_clamped_c(block, dest, line_size);

  1381. }

  1382. 

  1383. /* draw the edges of width 'w' of an image of size width, height */

  1384. // FIXME: Check that this is OK for MPEG-4 interlaced.

  1385. static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,

  1386.                            int w, int h, int sides)

  1387. {

  1388.     uint8_t *ptr = buf, *last_line;

  1389.     int i;

  1390. 

  1391.     /* left and right */

  1392.     for (i = 0; i < height; i++) {

  1393.         memset(ptr - w, ptr[0], w);

  1394.         memset(ptr + width, ptr[width - 1], w);

  1395.         ptr += wrap;

  1396.     }

  1397. 

  1398.     /* top and bottom + corners */

  1399.     buf -= w;

  1400.     last_line = buf + (height - 1) * wrap;

  1401.     if (sides & EDGE_TOP)

  1402.         for (i = 0; i < h; i++)

  1403.             // top

  1404.             memcpy(buf - (i + 1) * wrap, buf, width + w + w);

  1405.     if (sides & EDGE_BOTTOM)

  1406.         for (i = 0; i < h; i++)

  1407.             // bottom

  1408.             memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);

  1409. }

  1410. 

  1411. static void clear_block_8_c(int16_t *block)

  1412. {

  1413.     memset(block, 0, sizeof(int16_t) * 64);

  1414. }

  1415. 

  1416. static void clear_blocks_8_c(int16_t *blocks)

  1417. {

  1418.     memset(blocks, 0, sizeof(int16_t) * 6 * 64);

  1419. }

  1420. 

  1421. /* init static data */

  1422. av_cold void ff_dsputil_static_init(void)

  1423. {

  1424.     int i;

  1425. 

  1426.     for (i = 0; i < 512; i++)

  1427.         ff_square_tab[i] = (i - 256) * (i - 256);

  1428. }

  1429. 

  1430. av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)

  1431. {

  1432.     const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;

  1433. 

  1434. #if CONFIG_ENCODERS

  1435.     if (avctx->bits_per_raw_sample == 10) {

  1436.         c->fdct    = ff_jpeg_fdct_islow_10;

  1437.         c->fdct248 = ff_fdct248_islow_10;

  1438.     } else {

  1439.         if (avctx->dct_algo == FF_DCT_FASTINT) {

  1440.             c->fdct    = ff_fdct_ifast;

  1441.             c->fdct248 = ff_fdct_ifast248;

  1442.         } else if (avctx->dct_algo == FF_DCT_FAAN) {

  1443.             c->fdct    = ff_faandct;

  1444.             c->fdct248 = ff_faandct248;

  1445.         } else {

  1446.             c->fdct    = ff_jpeg_fdct_islow_8; // slow/accurate/default

  1447.             c->fdct248 = ff_fdct248_islow_8;

  1448.         }

  1449.     }

  1450. #endif /* CONFIG_ENCODERS */

  1451. 

  1452.     if (avctx->bits_per_raw_sample == 10) {

  1453.         c->idct_put              = ff_simple_idct_put_10;

  1454.         c->idct_add              = ff_simple_idct_add_10;

  1455.         c->idct                  = ff_simple_idct_10;

  1456.         c->idct_permutation_type = FF_NO_IDCT_PERM;

  1457.     } else {

  1458.         if (avctx->idct_algo == FF_IDCT_INT) {

  1459.             c->idct_put              = jref_idct_put;

  1460.             c->idct_add              = jref_idct_add;

  1461.             c->idct                  = ff_j_rev_dct;

  1462.             c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM;

  1463.         } else if (avctx->idct_algo == FF_IDCT_FAAN) {

  1464.             c->idct_put              = ff_faanidct_put;

  1465.             c->idct_add              = ff_faanidct_add;

  1466.             c->idct                  = ff_faanidct;

  1467.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1468.         } else { // accurate/default

  1469.             c->idct_put              = ff_simple_idct_put_8;

  1470.             c->idct_add              = ff_simple_idct_add_8;

  1471.             c->idct                  = ff_simple_idct_8;

  1472.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1473.         }

  1474.     }

  1475. 

  1476.     c->diff_pixels = diff_pixels_c;

  1477. 

  1478.     c->put_pixels_clamped        = put_pixels_clamped_c;

  1479.     c->put_signed_pixels_clamped = put_signed_pixels_clamped_c;

  1480.     c->add_pixels_clamped        = add_pixels_clamped_c;

  1481. 

  1482.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  1483. 

  1484.     c->gmc1 = gmc1_c;

  1485.     c->gmc  = ff_gmc_c;

  1486. 

  1487.     c->pix_sum   = pix_sum_c;

  1488.     c->pix_norm1 = pix_norm1_c;

  1489. 

  1490.     c->fill_block_tab[0] = fill_block16_c;

  1491.     c->fill_block_tab[1] = fill_block8_c;

  1492. 

  1493.     /* TODO [0] 16  [1] 8 */

  1494.     c->pix_abs[0][0] = pix_abs16_c;

  1495.     c->pix_abs[0][1] = pix_abs16_x2_c;

  1496.     c->pix_abs[0][2] = pix_abs16_y2_c;

  1497.     c->pix_abs[0][3] = pix_abs16_xy2_c;

  1498.     c->pix_abs[1][0] = pix_abs8_c;

  1499.     c->pix_abs[1][1] = pix_abs8_x2_c;

  1500.     c->pix_abs[1][2] = pix_abs8_y2_c;

  1501.     c->pix_abs[1][3] = pix_abs8_xy2_c;

  1502. 

  1503. #define SET_CMP_FUNC(name)                      \

  1504.     c->name[0] = name ## 16_c;                  \

  1505.     c->name[1] = name ## 8x8_c;

  1506. 

  1507.     SET_CMP_FUNC(hadamard8_diff)

  1508.     c->hadamard8_diff[4] = hadamard8_intra16_c;

  1509.     c->hadamard8_diff[5] = hadamard8_intra8x8_c;

  1510.     SET_CMP_FUNC(dct_sad)

  1511.     SET_CMP_FUNC(dct_max)

  1512. #if CONFIG_GPL

  1513.     SET_CMP_FUNC(dct264_sad)

  1514. #endif

  1515.     c->sad[0] = pix_abs16_c;

  1516.     c->sad[1] = pix_abs8_c;

  1517.     c->sse[0] = sse16_c;

  1518.     c->sse[1] = sse8_c;

  1519.     c->sse[2] = sse4_c;

  1520.     SET_CMP_FUNC(quant_psnr)

  1521.     SET_CMP_FUNC(rd)

  1522.     SET_CMP_FUNC(bit)

  1523.     c->vsad[0] = vsad16_c;

  1524.     c->vsad[4] = vsad_intra16_c;

  1525.     c->vsad[5] = vsad_intra8_c;

  1526.     c->vsse[0] = vsse16_c;

  1527.     c->vsse[4] = vsse_intra16_c;

  1528.     c->vsse[5] = vsse_intra8_c;

  1529.     c->nsse[0] = nsse16_c;

  1530.     c->nsse[1] = nsse8_c;

  1531. 

  1532.     c->bswap_buf   = bswap_buf;

  1533.     c->bswap16_buf = bswap16_buf;

  1534. 

  1535.     c->try_8x8basis = try_8x8basis_c;

  1536.     c->add_8x8basis = add_8x8basis_c;

  1537. 

  1538.     c->scalarproduct_int16 = scalarproduct_int16_c;

  1539.     c->vector_clip_int32   = vector_clip_int32_c;

  1540.     c->vector_clipf        = vector_clipf_c;

  1541. 

  1542.     c->shrink[0] = av_image_copy_plane;

  1543.     c->shrink[1] = ff_shrink22;

  1544.     c->shrink[2] = ff_shrink44;

  1545.     c->shrink[3] = ff_shrink88;

  1546. 

  1547.     c->draw_edges = draw_edges_8_c;

  1548. 

  1549.     c->clear_block  = clear_block_8_c;

  1550.     c->clear_blocks = clear_blocks_8_c;

  1551. 

  1552.     switch (avctx->bits_per_raw_sample) {

  1553.     case 9:

  1554.     case 10:

  1555.         c->get_pixels = get_pixels_16_c;

  1556.         break;

  1557.     default:

  1558.         c->get_pixels = get_pixels_8_c;

  1559.         break;

  1560.     }

  1561. 

  1562.     if (ARCH_ARM)

  1563.         ff_dsputil_init_arm(c, avctx, high_bit_depth);

  1564.     if (ARCH_BFIN)

  1565.         ff_dsputil_init_bfin(c, avctx, high_bit_depth);

  1566.     if (ARCH_PPC)

  1567.         ff_dsputil_init_ppc(c, avctx, high_bit_depth);

  1568.     if (ARCH_X86)

  1569.         ff_dsputil_init_x86(c, avctx, high_bit_depth);

  1570. 

  1571.     ff_init_scantable_permutation(c->idct_permutation,

  1572.                                   c->idct_permutation_type);

  1573. }