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.
40078 Commits
32 Branches
283MB
Tree: e74433a8e6
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e74433a8e6'
${ noResults }
FFmpeg/libavcodec/dsputil.c

1536 lines
50KB
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. #define avg2(a, b) ((a + b + 1) >> 1)

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

  378. 

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

  380.                    int x16, int y16, int rounder)

  381. {

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

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

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

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

  386.     int i;

  387. 

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

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

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

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

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

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

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

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

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

  397.         dst   += stride;

  398.         src   += stride;

  399.     }

  400. }

  401. 

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

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

  404.               int width, int height)

  405. {

  406.     int y, vx, vy;

  407.     const int s = 1 << shift;

  408. 

  409.     width--;

  410.     height--;

  411. 

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

  413.         int x;

  414. 

  415.         vx = ox;

  416.         vy = oy;

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

  418.             int index;

  419.             int src_x  = vx >> 16;

  420.             int src_y  = vy >> 16;

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

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

  423. 

  424.             src_x >>= shift;

  425.             src_y >>= shift;

  426. 

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

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

  429.                     index = src_x + src_y * stride;

  430.                     dst[y * stride + x] =

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

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

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

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

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

  436.                 } else {

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

  438.                     dst[y * stride + x] =

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

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

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

  442.                 }

  443.             } else {

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

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

  446.                     dst[y * stride + x] =

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

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

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

  450.                 } else {

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

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

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

  454.                 }

  455.             }

  456. 

  457.             vx += dxx;

  458.             vy += dyx;

  459.         }

  460.         ox += dxy;

  461.         oy += dyy;

  462.     }

  463. }

  464. 

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

  466.                               int line_size, int h)

  467. {

  468.     int s = 0, i;

  469. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  487.         pix1 += line_size;

  488.         pix2 += line_size;

  489.     }

  490.     return s;

  491. }

  492. 

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

  494.                           int line_size, int h)

  495. {

  496.     int s = 0, i;

  497. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  515.         pix1 += line_size;

  516.         pix2 += line_size;

  517.     }

  518.     return s;

  519. }

  520. 

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

  522.                           int line_size, int h)

  523. {

  524.     int s = 0, i;

  525.     uint8_t *pix3 = pix2 + line_size;

  526. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  544.         pix1 += line_size;

  545.         pix2 += line_size;

  546.         pix3 += line_size;

  547.     }

  548.     return s;

  549. }

  550. 

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

  552.                            int line_size, int h)

  553. {

  554.     int s = 0, i;

  555.     uint8_t *pix3 = pix2 + line_size;

  556. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  574.         pix1 += line_size;

  575.         pix2 += line_size;

  576.         pix3 += line_size;

  577.     }

  578.     return s;

  579. }

  580. 

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

  582.                              int line_size, int h)

  583. {

  584.     int s = 0, i;

  585. 

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

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

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

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

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

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

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

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

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

  595.         pix1 += line_size;

  596.         pix2 += line_size;

  597.     }

  598.     return s;

  599. }

  600. 

  601. static int pix_abs8_x2_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] - avg2(pix2[0], pix2[1]));

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

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

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

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

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

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

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

  615.         pix1 += line_size;

  616.         pix2 += line_size;

  617.     }

  618.     return s;

  619. }

  620. 

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

  622.                          int line_size, int h)

  623. {

  624.     int s = 0, i;

  625.     uint8_t *pix3 = pix2 + line_size;

  626. 

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

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

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

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

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

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

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

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

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

  636.         pix1 += line_size;

  637.         pix2 += line_size;

  638.         pix3 += line_size;

  639.     }

  640.     return s;

  641. }

  642. 

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

  644.                           int line_size, int h)

  645. {

  646.     int s = 0, i;

  647.     uint8_t *pix3 = pix2 + line_size;

  648. 

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

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

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

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

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

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

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

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

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

  658.         pix1 += line_size;

  659.         pix2 += line_size;

  660.         pix3 += line_size;

  661.     }

  662.     return s;

  663. }

  664. 

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

  666. {

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

  668. 

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

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

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

  672.         if (y + 1 < h) {

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

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

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

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

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

  678.         }

  679.         s1 += stride;

  680.         s2 += stride;

  681.     }

  682. 

  683.     if (c)

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

  685.     else

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

  687. }

  688. 

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

  690. {

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

  692. 

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

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

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

  696.         if (y + 1 < h) {

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

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

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

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

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

  702.         }

  703.         s1 += stride;

  704.         s2 += stride;

  705.     }

  706. 

  707.     if (c)

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

  709.     else

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

  711. }

  712. 

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

  714.                           int16_t basis[64], int scale)

  715. {

  716.     int i;

  717.     unsigned int sum = 0;

  718. 

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

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

  721.                            (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  722.                           (BASIS_SHIFT - RECON_SHIFT));

  723.         int w = weight[i];

  724.         b >>= RECON_SHIFT;

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

  726. 

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

  728.     }

  729.     return sum >> 2;

  730. }

  731. 

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

  733. {

  734.     int i;

  735. 

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

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

  738.                    (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  739.                   (BASIS_SHIFT - RECON_SHIFT);

  740. }

  741. 

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

  743.                     int stride, int h)

  744. {

  745.     return 0;

  746. }

  747. 

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

  749. {

  750.     int i;

  751. 

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

  753. 

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

  755.         switch (type & 0xFF) {

  756.         case FF_CMP_SAD:

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

  758.             break;

  759.         case FF_CMP_SATD:

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

  761.             break;

  762.         case FF_CMP_SSE:

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

  764.             break;

  765.         case FF_CMP_DCT:

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

  767.             break;

  768.         case FF_CMP_DCT264:

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

  770.             break;

  771.         case FF_CMP_DCTMAX:

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

  773.             break;

  774.         case FF_CMP_PSNR:

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

  776.             break;

  777.         case FF_CMP_BIT:

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

  779.             break;

  780.         case FF_CMP_RD:

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

  782.             break;

  783.         case FF_CMP_VSAD:

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

  785.             break;

  786.         case FF_CMP_VSSE:

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

  788.             break;

  789.         case FF_CMP_ZERO:

  790.             cmp[i] = zero_cmp;

  791.             break;

  792.         case FF_CMP_NSSE:

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

  794.             break;

  795.         default:

  796.             av_log(NULL, AV_LOG_ERROR,

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

  798.         }

  799.     }

  800. }

  801. 

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

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

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

  805. 

  806. #define BUTTERFLY1(x, y)                        \

  807.     {                                           \

  808.         int a, b;                               \

  809.         a = x;                                  \

  810.         b = y;                                  \

  811.         x = a + b;                              \

  812.         y = a - b;                              \

  813.     }

  814. 

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

  816. 

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

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

  819. {

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

  821. 

  822.     assert(h == 8);

  823. 

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

  825.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

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

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

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

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

  838. 

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

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

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

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

  843. 

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

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

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

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

  848.     }

  849. 

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

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

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

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

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

  855. 

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

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

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

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

  860. 

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

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

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

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

  865.     }

  866.     return sum;

  867. }

  868. 

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

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

  871. {

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

  873. 

  874.     assert(h == 8);

  875. 

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

  877.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

  886. 

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

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

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

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

  891. 

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

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

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

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

  896.     }

  897. 

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

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

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

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

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

  903. 

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

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

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

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

  908. 

  909.         sum +=

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

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

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

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

  914.     }

  915. 

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

  917. 

  918.     return sum;

  919. }

  920. 

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

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

  923. {

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

  925. 

  926.     assert(h == 8);

  927. 

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

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

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

  931. }

  932. 

  933. #if CONFIG_GPL

  934. #define DCT8_1D                                         \

  935.     {                                                   \

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

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

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

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

  940.         const int a0  = s07 + s34;                      \

  941.         const int a1  = s16 + s25;                      \

  942.         const int a2  = s07 - s34;                      \

  943.         const int a3  = s16 - s25;                      \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  960.     }

  961. 

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

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

  964. {

  965.     int16_t dct[8][8];

  966.     int i, sum = 0;

  967. 

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

  969. 

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

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

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

  973.         DCT8_1D

  974. #undef SRC

  975. #undef DST

  976. 

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

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

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

  980.             DCT8_1D

  981. #undef SRC

  982. #undef DST

  983.             return sum;

  984. }

  985. #endif

  986. 

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

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

  989. {

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

  991.     int sum = 0, i;

  992. 

  993.     assert(h == 8);

  994. 

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

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

  997. 

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

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

  1000. 

  1001.     return sum;

  1002. }

  1003. 

  1004. static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,

  1005.                            uint8_t *src2, int stride, int h)

  1006. {

  1007.     LOCAL_ALIGNED_16(int16_t, temp, [64 * 2]);

  1008.     int16_t *const bak = temp + 64;

  1009.     int sum = 0, i;

  1010. 

  1011.     assert(h == 8);

  1012.     s->mb_intra = 0;

  1013. 

  1014.     s->dsp.diff_pixels(temp, src1, src2, stride);

  1015. 

  1016.     memcpy(bak, temp, 64 * sizeof(int16_t));

  1017. 

  1018.     s->block_last_index[0 /* FIXME */] =

  1019.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  1020.     s->dct_unquantize_inter(s, temp, 0, s->qscale);

  1021.     ff_simple_idct_8(temp); // FIXME

  1022. 

  1023.     for (i = 0; i < 64; i++)

  1024.         sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);

  1025. 

  1026.     return sum;

  1027. }

  1028. 

  1029. static int rd8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,

  1030.                    int stride, int h)

  1031. {

  1032.     const uint8_t *scantable = s->intra_scantable.permutated;

  1033.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  1034.     LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]);

  1035.     LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]);

  1036.     int i, last, run, bits, level, distortion, start_i;

  1037.     const int esc_length = s->ac_esc_length;

  1038.     uint8_t *length, *last_length;

  1039. 

  1040.     assert(h == 8);

  1041. 

  1042.     copy_block8(lsrc1, src1, 8, stride, 8);

  1043.     copy_block8(lsrc2, src2, 8, stride, 8);

  1044. 

  1045.     s->dsp.diff_pixels(temp, lsrc1, lsrc2, 8);

  1046. 

  1047.     s->block_last_index[0 /* FIXME */] =

  1048.     last                               =

  1049.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  1050. 

  1051.     bits = 0;

  1052. 

  1053.     if (s->mb_intra) {

  1054.         start_i     = 1;

  1055.         length      = s->intra_ac_vlc_length;

  1056.         last_length = s->intra_ac_vlc_last_length;

  1057.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  1058.     } else {

  1059.         start_i     = 0;

  1060.         length      = s->inter_ac_vlc_length;

  1061.         last_length = s->inter_ac_vlc_last_length;

  1062.     }

  1063. 

  1064.     if (last >= start_i) {

  1065.         run = 0;

  1066.         for (i = start_i; i < last; i++) {

  1067.             int j = scantable[i];

  1068.             level = temp[j];

  1069. 

  1070.             if (level) {

  1071.                 level += 64;

  1072.                 if ((level & (~127)) == 0)

  1073.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  1074.                 else

  1075.                     bits += esc_length;

  1076.                 run = 0;

  1077.             } else

  1078.                 run++;

  1079.         }

  1080.         i = scantable[last];

  1081. 

  1082.         level = temp[i] + 64;

  1083. 

  1084.         assert(level - 64);

  1085. 

  1086.         if ((level & (~127)) == 0) {

  1087.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  1088.         } else

  1089.             bits += esc_length;

  1090.     }

  1091. 

  1092.     if (last >= 0) {

  1093.         if (s->mb_intra)

  1094.             s->dct_unquantize_intra(s, temp, 0, s->qscale);

  1095.         else

  1096.             s->dct_unquantize_inter(s, temp, 0, s->qscale);

  1097.     }

  1098. 

  1099.     s->dsp.idct_add(lsrc2, 8, temp);

  1100. 

  1101.     distortion = s->dsp.sse[1](NULL, lsrc2, lsrc1, 8, 8);

  1102. 

  1103.     return distortion + ((bits * s->qscale * s->qscale * 109 + 64) >> 7);

  1104. }

  1105. 

  1106. static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,

  1107.                     int stride, int h)

  1108. {

  1109.     const uint8_t *scantable = s->intra_scantable.permutated;

  1110.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  1111.     int i, last, run, bits, level, start_i;

  1112.     const int esc_length = s->ac_esc_length;

  1113.     uint8_t *length, *last_length;

  1114. 

  1115.     assert(h == 8);

  1116. 

  1117.     s->dsp.diff_pixels(temp, src1, src2, stride);

  1118. 

  1119.     s->block_last_index[0 /* FIXME */] =

  1120.     last                               =

  1121.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  1122. 

  1123.     bits = 0;

  1124. 

  1125.     if (s->mb_intra) {

  1126.         start_i     = 1;

  1127.         length      = s->intra_ac_vlc_length;

  1128.         last_length = s->intra_ac_vlc_last_length;

  1129.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  1130.     } else {

  1131.         start_i     = 0;

  1132.         length      = s->inter_ac_vlc_length;

  1133.         last_length = s->inter_ac_vlc_last_length;

  1134.     }

  1135. 

  1136.     if (last >= start_i) {

  1137.         run = 0;

  1138.         for (i = start_i; i < last; i++) {

  1139.             int j = scantable[i];

  1140.             level = temp[j];

  1141. 

  1142.             if (level) {

  1143.                 level += 64;

  1144.                 if ((level & (~127)) == 0)

  1145.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  1146.                 else

  1147.                     bits += esc_length;

  1148.                 run = 0;

  1149.             } else

  1150.                 run++;

  1151.         }

  1152.         i = scantable[last];

  1153. 

  1154.         level = temp[i] + 64;

  1155. 

  1156.         assert(level - 64);

  1157. 

  1158.         if ((level & (~127)) == 0)

  1159.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  1160.         else

  1161.             bits += esc_length;

  1162.     }

  1163. 

  1164.     return bits;

  1165. }

  1166. 

  1167. #define VSAD_INTRA(size)                                                \

  1168. static int vsad_intra ## size ## _c(MpegEncContext *c,                  \

  1169.                                     uint8_t *s, uint8_t *dummy,         \

  1170.                                     int stride, int h)                  \

  1171. {                                                                       \

  1172.     int score = 0, x, y;                                                \

  1173.                                                                         \

  1174.     for (y = 1; y < h; y++) {                                           \

  1175.         for (x = 0; x < size; x += 4) {                                 \

  1176.             score += FFABS(s[x]     - s[x + stride])     +              \

  1177.                      FFABS(s[x + 1] - s[x + stride + 1]) +              \

  1178.                      FFABS(s[x + 2] - s[x + 2 + stride]) +              \

  1179.                      FFABS(s[x + 3] - s[x + 3 + stride]);               \

  1180.         }                                                               \

  1181.         s += stride;                                                    \

  1182.     }                                                                   \

  1183.                                                                         \

  1184.     return score;                                                       \

  1185. }

  1186. VSAD_INTRA(8)

  1187. VSAD_INTRA(16)

  1188. 

  1189. static int vsad16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  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 < 16; x++)

  1196.             score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  1197.         s1 += stride;

  1198.         s2 += stride;

  1199.     }

  1200. 

  1201.     return score;

  1202. }

  1203. 

  1204. #define SQ(a) ((a) * (a))

  1205. #define VSSE_INTRA(size)                                                \

  1206. static int vsse_intra ## size ## _c(MpegEncContext *c,                  \

  1207.                                     uint8_t *s, uint8_t *dummy,         \

  1208.                                     int stride, int h)                  \

  1209. {                                                                       \

  1210.     int score = 0, x, y;                                                \

  1211.                                                                         \

  1212.     for (y = 1; y < h; y++) {                                           \

  1213.         for (x = 0; x < size; x += 4) {                                 \

  1214.             score += SQ(s[x]     - s[x + stride]) +                     \

  1215.                      SQ(s[x + 1] - s[x + stride + 1]) +                 \

  1216.                      SQ(s[x + 2] - s[x + stride + 2]) +                 \

  1217.                      SQ(s[x + 3] - s[x + stride + 3]);                  \

  1218.         }                                                               \

  1219.         s += stride;                                                    \

  1220.     }                                                                   \

  1221.                                                                         \

  1222.     return score;                                                       \

  1223. }

  1224. VSSE_INTRA(8)

  1225. VSSE_INTRA(16)

  1226. 

  1227. static int vsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  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 < 16; x++)

  1234.             score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  1235.         s1 += stride;

  1236.         s2 += stride;

  1237.     }

  1238. 

  1239.     return score;

  1240. }

  1241. 

  1242. #define WRAPPER8_16_SQ(name8, name16)                                   \

  1243. static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src,        \

  1244.                   int stride, int h)                                    \

  1245. {                                                                       \

  1246.     int score = 0;                                                      \

  1247.                                                                         \

  1248.     score += name8(s, dst, src, stride, 8);                             \

  1249.     score += name8(s, dst + 8, src + 8, stride, 8);                     \

  1250.     if (h == 16) {                                                      \

  1251.         dst   += 8 * stride;                                            \

  1252.         src   += 8 * stride;                                            \

  1253.         score += name8(s, dst, src, stride, 8);                         \

  1254.         score += name8(s, dst + 8, src + 8, stride, 8);                 \

  1255.     }                                                                   \

  1256.     return score;                                                       \

  1257. }

  1258. 

  1259. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  1260. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  1261. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  1262. #if CONFIG_GPL

  1263. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  1264. #endif

  1265. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  1266. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  1267. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  1268. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  1269. 

  1270. static inline uint32_t clipf_c_one(uint32_t a, uint32_t mini,

  1271.                                    uint32_t maxi, uint32_t maxisign)

  1272. {

  1273.     if (a > mini)

  1274.         return mini;

  1275.     else if ((a ^ (1U << 31)) > maxisign)

  1276.         return maxi;

  1277.     else

  1278.         return a;

  1279. }

  1280. 

  1281. static void vector_clipf_c_opposite_sign(float *dst, const float *src,

  1282.                                          float *min, float *max, int len)

  1283. {

  1284.     int i;

  1285.     uint32_t mini        = *(uint32_t *) min;

  1286.     uint32_t maxi        = *(uint32_t *) max;

  1287.     uint32_t maxisign    = maxi ^ (1U << 31);

  1288.     uint32_t *dsti       = (uint32_t *) dst;

  1289.     const uint32_t *srci = (const uint32_t *) src;

  1290. 

  1291.     for (i = 0; i < len; i += 8) {

  1292.         dsti[i + 0] = clipf_c_one(srci[i + 0], mini, maxi, maxisign);

  1293.         dsti[i + 1] = clipf_c_one(srci[i + 1], mini, maxi, maxisign);

  1294.         dsti[i + 2] = clipf_c_one(srci[i + 2], mini, maxi, maxisign);

  1295.         dsti[i + 3] = clipf_c_one(srci[i + 3], mini, maxi, maxisign);

  1296.         dsti[i + 4] = clipf_c_one(srci[i + 4], mini, maxi, maxisign);

  1297.         dsti[i + 5] = clipf_c_one(srci[i + 5], mini, maxi, maxisign);

  1298.         dsti[i + 6] = clipf_c_one(srci[i + 6], mini, maxi, maxisign);

  1299.         dsti[i + 7] = clipf_c_one(srci[i + 7], mini, maxi, maxisign);

  1300.     }

  1301. }

  1302. 

  1303. static void vector_clipf_c(float *dst, const float *src,

  1304.                            float min, float max, int len)

  1305. {

  1306.     int i;

  1307. 

  1308.     if (min < 0 && max > 0) {

  1309.         vector_clipf_c_opposite_sign(dst, src, &min, &max, len);

  1310.     } else {

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

  1312.             dst[i]     = av_clipf(src[i], min, max);

  1313.             dst[i + 1] = av_clipf(src[i + 1], min, max);

  1314.             dst[i + 2] = av_clipf(src[i + 2], min, max);

  1315.             dst[i + 3] = av_clipf(src[i + 3], min, max);

  1316.             dst[i + 4] = av_clipf(src[i + 4], min, max);

  1317.             dst[i + 5] = av_clipf(src[i + 5], min, max);

  1318.             dst[i + 6] = av_clipf(src[i + 6], min, max);

  1319.             dst[i + 7] = av_clipf(src[i + 7], min, max);

  1320.         }

  1321.     }

  1322. }

  1323. 

  1324. static int32_t scalarproduct_int16_c(const int16_t *v1, const int16_t *v2,

  1325.                                      int order)

  1326. {

  1327.     int res = 0;

  1328. 

  1329.     while (order--)

  1330.         res += *v1++ **v2++;

  1331. 

  1332.     return res;

  1333. }

  1334. 

  1335. static void vector_clip_int32_c(int32_t *dst, const int32_t *src, int32_t min,

  1336.                                 int32_t max, unsigned int len)

  1337. {

  1338.     do {

  1339.         *dst++ = av_clip(*src++, min, max);

  1340.         *dst++ = av_clip(*src++, min, max);

  1341.         *dst++ = av_clip(*src++, min, max);

  1342.         *dst++ = av_clip(*src++, min, max);

  1343.         *dst++ = av_clip(*src++, min, max);

  1344.         *dst++ = av_clip(*src++, min, max);

  1345.         *dst++ = av_clip(*src++, min, max);

  1346.         *dst++ = av_clip(*src++, min, max);

  1347.         len   -= 8;

  1348.     } while (len > 0);

  1349. }

  1350. 

  1351. static void jref_idct_put(uint8_t *dest, int line_size, int16_t *block)

  1352. {

  1353.     ff_j_rev_dct(block);

  1354.     put_pixels_clamped_c(block, dest, line_size);

  1355. }

  1356. 

  1357. static void jref_idct_add(uint8_t *dest, int line_size, int16_t *block)

  1358. {

  1359.     ff_j_rev_dct(block);

  1360.     add_pixels_clamped_c(block, dest, line_size);

  1361. }

  1362. 

  1363. /* draw the edges of width 'w' of an image of size width, height */

  1364. // FIXME: Check that this is OK for MPEG-4 interlaced.

  1365. static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,

  1366.                            int w, int h, int sides)

  1367. {

  1368.     uint8_t *ptr = buf, *last_line;

  1369.     int i;

  1370. 

  1371.     /* left and right */

  1372.     for (i = 0; i < height; i++) {

  1373.         memset(ptr - w, ptr[0], w);

  1374.         memset(ptr + width, ptr[width - 1], w);

  1375.         ptr += wrap;

  1376.     }

  1377. 

  1378.     /* top and bottom + corners */

  1379.     buf -= w;

  1380.     last_line = buf + (height - 1) * wrap;

  1381.     if (sides & EDGE_TOP)

  1382.         for (i = 0; i < h; i++)

  1383.             // top

  1384.             memcpy(buf - (i + 1) * wrap, buf, width + w + w);

  1385.     if (sides & EDGE_BOTTOM)

  1386.         for (i = 0; i < h; i++)

  1387.             // bottom

  1388.             memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);

  1389. }

  1390. 

  1391. /* init static data */

  1392. av_cold void ff_dsputil_static_init(void)

  1393. {

  1394.     int i;

  1395. 

  1396.     for (i = 0; i < 512; i++)

  1397.         ff_square_tab[i] = (i - 256) * (i - 256);

  1398. }

  1399. 

  1400. av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)

  1401. {

  1402.     const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;

  1403. 

  1404. #if CONFIG_ENCODERS

  1405.     if (avctx->bits_per_raw_sample == 10) {

  1406.         c->fdct    = ff_jpeg_fdct_islow_10;

  1407.         c->fdct248 = ff_fdct248_islow_10;

  1408.     } else {

  1409.         if (avctx->dct_algo == FF_DCT_FASTINT) {

  1410.             c->fdct    = ff_fdct_ifast;

  1411.             c->fdct248 = ff_fdct_ifast248;

  1412.         } else if (avctx->dct_algo == FF_DCT_FAAN) {

  1413.             c->fdct    = ff_faandct;

  1414.             c->fdct248 = ff_faandct248;

  1415.         } else {

  1416.             c->fdct    = ff_jpeg_fdct_islow_8; // slow/accurate/default

  1417.             c->fdct248 = ff_fdct248_islow_8;

  1418.         }

  1419.     }

  1420. #endif /* CONFIG_ENCODERS */

  1421. 

  1422.     if (avctx->bits_per_raw_sample == 10) {

  1423.         c->idct_put              = ff_simple_idct_put_10;

  1424.         c->idct_add              = ff_simple_idct_add_10;

  1425.         c->idct                  = ff_simple_idct_10;

  1426.         c->idct_permutation_type = FF_NO_IDCT_PERM;

  1427.     } else {

  1428.         if (avctx->idct_algo == FF_IDCT_INT) {

  1429.             c->idct_put              = jref_idct_put;

  1430.             c->idct_add              = jref_idct_add;

  1431.             c->idct                  = ff_j_rev_dct;

  1432.             c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM;

  1433.         } else if (avctx->idct_algo == FF_IDCT_FAAN) {

  1434.             c->idct_put              = ff_faanidct_put;

  1435.             c->idct_add              = ff_faanidct_add;

  1436.             c->idct                  = ff_faanidct;

  1437.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1438.         } else { // accurate/default

  1439.             c->idct_put              = ff_simple_idct_put_8;

  1440.             c->idct_add              = ff_simple_idct_add_8;

  1441.             c->idct                  = ff_simple_idct_8;

  1442.             c->idct_permutation_type = FF_NO_IDCT_PERM;

  1443.         }

  1444.     }

  1445. 

  1446.     c->diff_pixels = diff_pixels_c;

  1447. 

  1448.     c->put_pixels_clamped        = put_pixels_clamped_c;

  1449.     c->put_signed_pixels_clamped = put_signed_pixels_clamped_c;

  1450.     c->add_pixels_clamped        = add_pixels_clamped_c;

  1451. 

  1452.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  1453. 

  1454.     c->gmc1 = gmc1_c;

  1455.     c->gmc  = ff_gmc_c;

  1456. 

  1457.     c->pix_sum   = pix_sum_c;

  1458.     c->pix_norm1 = pix_norm1_c;

  1459. 

  1460.     /* TODO [0] 16  [1] 8 */

  1461.     c->pix_abs[0][0] = pix_abs16_c;

  1462.     c->pix_abs[0][1] = pix_abs16_x2_c;

  1463.     c->pix_abs[0][2] = pix_abs16_y2_c;

  1464.     c->pix_abs[0][3] = pix_abs16_xy2_c;

  1465.     c->pix_abs[1][0] = pix_abs8_c;

  1466.     c->pix_abs[1][1] = pix_abs8_x2_c;

  1467.     c->pix_abs[1][2] = pix_abs8_y2_c;

  1468.     c->pix_abs[1][3] = pix_abs8_xy2_c;

  1469. 

  1470. #define SET_CMP_FUNC(name)                      \

  1471.     c->name[0] = name ## 16_c;                  \

  1472.     c->name[1] = name ## 8x8_c;

  1473. 

  1474.     SET_CMP_FUNC(hadamard8_diff)

  1475.     c->hadamard8_diff[4] = hadamard8_intra16_c;

  1476.     c->hadamard8_diff[5] = hadamard8_intra8x8_c;

  1477.     SET_CMP_FUNC(dct_sad)

  1478.     SET_CMP_FUNC(dct_max)

  1479. #if CONFIG_GPL

  1480.     SET_CMP_FUNC(dct264_sad)

  1481. #endif

  1482.     c->sad[0] = pix_abs16_c;

  1483.     c->sad[1] = pix_abs8_c;

  1484.     c->sse[0] = sse16_c;

  1485.     c->sse[1] = sse8_c;

  1486.     c->sse[2] = sse4_c;

  1487.     SET_CMP_FUNC(quant_psnr)

  1488.     SET_CMP_FUNC(rd)

  1489.     SET_CMP_FUNC(bit)

  1490.     c->vsad[0] = vsad16_c;

  1491.     c->vsad[4] = vsad_intra16_c;

  1492.     c->vsad[5] = vsad_intra8_c;

  1493.     c->vsse[0] = vsse16_c;

  1494.     c->vsse[4] = vsse_intra16_c;

  1495.     c->vsse[5] = vsse_intra8_c;

  1496.     c->nsse[0] = nsse16_c;

  1497.     c->nsse[1] = nsse8_c;

  1498. 

  1499.     c->bswap_buf   = bswap_buf;

  1500.     c->bswap16_buf = bswap16_buf;

  1501. 

  1502.     c->try_8x8basis = try_8x8basis_c;

  1503.     c->add_8x8basis = add_8x8basis_c;

  1504. 

  1505.     c->scalarproduct_int16 = scalarproduct_int16_c;

  1506.     c->vector_clip_int32   = vector_clip_int32_c;

  1507.     c->vector_clipf        = vector_clipf_c;

  1508. 

  1509.     c->shrink[0] = av_image_copy_plane;

  1510.     c->shrink[1] = ff_shrink22;

  1511.     c->shrink[2] = ff_shrink44;

  1512.     c->shrink[3] = ff_shrink88;

  1513. 

  1514.     c->draw_edges = draw_edges_8_c;

  1515. 

  1516.     switch (avctx->bits_per_raw_sample) {

  1517.     case 9:

  1518.     case 10:

  1519.         c->get_pixels = get_pixels_16_c;

  1520.         break;

  1521.     default:

  1522.         c->get_pixels = get_pixels_8_c;

  1523.         break;

  1524.     }

  1525. 

  1526.     if (ARCH_ARM)

  1527.         ff_dsputil_init_arm(c, avctx, high_bit_depth);

  1528.     if (ARCH_PPC)

  1529.         ff_dsputil_init_ppc(c, avctx, high_bit_depth);

  1530.     if (ARCH_X86)

  1531.         ff_dsputil_init_x86(c, avctx, high_bit_depth);

  1532. 

  1533.     ff_init_scantable_permutation(c->idct_permutation,

  1534.                                   c->idct_permutation_type);

  1535. }