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.
34836 Commits
32 Branches
935MB
Tree: e0c6cce447
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e0c6cce447'
${ noResults }
FFmpeg/libswscale/utils.c

1700 lines
59KB
Raw Blame History 

  1. /*

  2.  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>

  3.  *

  4.  * This file is part of Libav.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. #include "config.h"

  22. 

  23. #define _SVID_SOURCE // needed for MAP_ANONYMOUS

  24. #include <assert.h>

  25. #include <inttypes.h>

  26. #include <math.h>

  27. #include <stdio.h>

  28. #include <string.h>

  29. #if HAVE_SYS_MMAN_H

  30. #include <sys/mman.h>

  31. #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)

  32. #define MAP_ANONYMOUS MAP_ANON

  33. #endif

  34. #endif

  35. #if HAVE_VIRTUALALLOC

  36. #define WIN32_LEAN_AND_MEAN

  37. #include <windows.h>

  38. #endif

  39. 

  40. #include "libavutil/attributes.h"

  41. #include "libavutil/avutil.h"

  42. #include "libavutil/bswap.h"

  43. #include "libavutil/cpu.h"

  44. #include "libavutil/intreadwrite.h"

  45. #include "libavutil/mathematics.h"

  46. #include "libavutil/opt.h"

  47. #include "libavutil/pixdesc.h"

  48. #include "libavutil/x86/asm.h"

  49. #include "libavutil/x86/cpu.h"

  50. #include "rgb2rgb.h"

  51. #include "swscale.h"

  52. #include "swscale_internal.h"

  53. 

  54. unsigned swscale_version(void)

  55. {

  56.     return LIBSWSCALE_VERSION_INT;

  57. }

  58. 

  59. const char *swscale_configuration(void)

  60. {

  61.     return LIBAV_CONFIGURATION;

  62. }

  63. 

  64. const char *swscale_license(void)

  65. {

  66. #define LICENSE_PREFIX "libswscale license: "

  67.     return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;

  68. }

  69. 

  70. #define RET 0xC3 // near return opcode for x86

  71. 

  72. typedef struct FormatEntry {

  73.     int is_supported_in, is_supported_out;

  74. } FormatEntry;

  75. 

  76. static const FormatEntry format_entries[PIX_FMT_NB] = {

  77.     [PIX_FMT_YUV420P]     = { 1, 1 },

  78.     [PIX_FMT_YUYV422]     = { 1, 1 },

  79.     [PIX_FMT_RGB24]       = { 1, 1 },

  80.     [PIX_FMT_BGR24]       = { 1, 1 },

  81.     [PIX_FMT_YUV422P]     = { 1, 1 },

  82.     [PIX_FMT_YUV444P]     = { 1, 1 },

  83.     [PIX_FMT_YUV410P]     = { 1, 1 },

  84.     [PIX_FMT_YUV411P]     = { 1, 1 },

  85.     [PIX_FMT_GRAY8]       = { 1, 1 },

  86.     [PIX_FMT_MONOWHITE]   = { 1, 1 },

  87.     [PIX_FMT_MONOBLACK]   = { 1, 1 },

  88.     [PIX_FMT_PAL8]        = { 1, 0 },

  89.     [PIX_FMT_YUVJ420P]    = { 1, 1 },

  90.     [PIX_FMT_YUVJ422P]    = { 1, 1 },

  91.     [PIX_FMT_YUVJ444P]    = { 1, 1 },

  92.     [PIX_FMT_UYVY422]     = { 1, 1 },

  93.     [PIX_FMT_UYYVYY411]   = { 0, 0 },

  94.     [PIX_FMT_BGR8]        = { 1, 1 },

  95.     [PIX_FMT_BGR4]        = { 0, 1 },

  96.     [PIX_FMT_BGR4_BYTE]   = { 1, 1 },

  97.     [PIX_FMT_RGB8]        = { 1, 1 },

  98.     [PIX_FMT_RGB4]        = { 0, 1 },

  99.     [PIX_FMT_RGB4_BYTE]   = { 1, 1 },

  100.     [PIX_FMT_NV12]        = { 1, 1 },

  101.     [PIX_FMT_NV21]        = { 1, 1 },

  102.     [PIX_FMT_ARGB]        = { 1, 1 },

  103.     [PIX_FMT_RGBA]        = { 1, 1 },

  104.     [PIX_FMT_ABGR]        = { 1, 1 },

  105.     [PIX_FMT_BGRA]        = { 1, 1 },

  106.     [PIX_FMT_GRAY16BE]    = { 1, 1 },

  107.     [PIX_FMT_GRAY16LE]    = { 1, 1 },

  108.     [PIX_FMT_YUV440P]     = { 1, 1 },

  109.     [PIX_FMT_YUVJ440P]    = { 1, 1 },

  110.     [PIX_FMT_YUVA420P]    = { 1, 1 },

  111.     [PIX_FMT_RGB48BE]     = { 1, 1 },

  112.     [PIX_FMT_RGB48LE]     = { 1, 1 },

  113.     [PIX_FMT_RGB565BE]    = { 1, 1 },

  114.     [PIX_FMT_RGB565LE]    = { 1, 1 },

  115.     [PIX_FMT_RGB555BE]    = { 1, 1 },

  116.     [PIX_FMT_RGB555LE]    = { 1, 1 },

  117.     [PIX_FMT_BGR565BE]    = { 1, 1 },

  118.     [PIX_FMT_BGR565LE]    = { 1, 1 },

  119.     [PIX_FMT_BGR555BE]    = { 1, 1 },

  120.     [PIX_FMT_BGR555LE]    = { 1, 1 },

  121.     [PIX_FMT_YUV420P16LE] = { 1, 1 },

  122.     [PIX_FMT_YUV420P16BE] = { 1, 1 },

  123.     [PIX_FMT_YUV422P16LE] = { 1, 1 },

  124.     [PIX_FMT_YUV422P16BE] = { 1, 1 },

  125.     [PIX_FMT_YUV444P16LE] = { 1, 1 },

  126.     [PIX_FMT_YUV444P16BE] = { 1, 1 },

  127.     [PIX_FMT_RGB444LE]    = { 1, 1 },

  128.     [PIX_FMT_RGB444BE]    = { 1, 1 },

  129.     [PIX_FMT_BGR444LE]    = { 1, 1 },

  130.     [PIX_FMT_BGR444BE]    = { 1, 1 },

  131.     [PIX_FMT_Y400A]       = { 1, 0 },

  132.     [PIX_FMT_BGR48BE]     = { 1, 1 },

  133.     [PIX_FMT_BGR48LE]     = { 1, 1 },

  134.     [PIX_FMT_YUV420P9BE]  = { 1, 1 },

  135.     [PIX_FMT_YUV420P9LE]  = { 1, 1 },

  136.     [PIX_FMT_YUV420P10BE] = { 1, 1 },

  137.     [PIX_FMT_YUV420P10LE] = { 1, 1 },

  138.     [PIX_FMT_YUV422P9BE]  = { 1, 1 },

  139.     [PIX_FMT_YUV422P9LE]  = { 1, 1 },

  140.     [PIX_FMT_YUV422P10BE] = { 1, 1 },

  141.     [PIX_FMT_YUV422P10LE] = { 1, 1 },

  142.     [PIX_FMT_YUV444P9BE]  = { 1, 1 },

  143.     [PIX_FMT_YUV444P9LE]  = { 1, 1 },

  144.     [PIX_FMT_YUV444P10BE] = { 1, 1 },

  145.     [PIX_FMT_YUV444P10LE] = { 1, 1 },

  146.     [PIX_FMT_GBRP]        = { 1, 0 },

  147.     [PIX_FMT_GBRP9LE]     = { 1, 0 },

  148.     [PIX_FMT_GBRP9BE]     = { 1, 0 },

  149.     [PIX_FMT_GBRP10LE]    = { 1, 0 },

  150.     [PIX_FMT_GBRP10BE]    = { 1, 0 },

  151.     [PIX_FMT_GBRP16LE]    = { 1, 0 },

  152.     [PIX_FMT_GBRP16BE]    = { 1, 0 },

  153. };

  154. 

  155. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  156. {

  157.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  158.            format_entries[pix_fmt].is_supported_in : 0;

  159. }

  160. 

  161. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  162. {

  163.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  164.            format_entries[pix_fmt].is_supported_out : 0;

  165. }

  166. 

  167. extern const int32_t ff_yuv2rgb_coeffs[8][4];

  168. 

  169. const char *sws_format_name(enum PixelFormat format)

  170. {

  171.     if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)

  172.         return av_pix_fmt_descriptors[format].name;

  173.     else

  174.         return "Unknown format";

  175. }

  176. 

  177. static double getSplineCoeff(double a, double b, double c, double d,

  178.                              double dist)

  179. {

  180.     if (dist <= 1.0)

  181.         return ((d * dist + c) * dist + b) * dist + a;

  182.     else

  183.         return getSplineCoeff(0.0,

  184.                                b + 2.0 * c + 3.0 * d,

  185.                                c + 3.0 * d,

  186.                               -b - 3.0 * c - 6.0 * d,

  187.                               dist - 1.0);

  188. }

  189. 

  190. static int initFilter(int16_t **outFilter, int32_t **filterPos,

  191.                       int *outFilterSize, int xInc, int srcW, int dstW,

  192.                       int filterAlign, int one, int flags, int cpu_flags,

  193.                       SwsVector *srcFilter, SwsVector *dstFilter,

  194.                       double param[2], int is_horizontal)

  195. {

  196.     int i;

  197.     int filterSize;

  198.     int filter2Size;

  199.     int minFilterSize;

  200.     int64_t *filter    = NULL;

  201.     int64_t *filter2   = NULL;

  202.     const int64_t fone = 1LL << 54;

  203.     int ret            = -1;

  204. 

  205.     emms_c(); // FIXME should not be required but IS (even for non-MMX versions)

  206. 

  207.     // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end

  208.     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);

  209. 

  210.     if (FFABS(xInc - 0x10000) < 10) { // unscaled

  211.         int i;

  212.         filterSize = 1;

  213.         FF_ALLOCZ_OR_GOTO(NULL, filter,

  214.                           dstW * sizeof(*filter) * filterSize, fail);

  215. 

  216.         for (i = 0; i < dstW; i++) {

  217.             filter[i * filterSize] = fone;

  218.             (*filterPos)[i]        = i;

  219.         }

  220.     } else if (flags & SWS_POINT) { // lame looking point sampling mode

  221.         int i;

  222.         int xDstInSrc;

  223.         filterSize = 1;

  224.         FF_ALLOC_OR_GOTO(NULL, filter,

  225.                          dstW * sizeof(*filter) * filterSize, fail);

  226. 

  227.         xDstInSrc = xInc / 2 - 0x8000;

  228.         for (i = 0; i < dstW; i++) {

  229.             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;

  230. 

  231.             (*filterPos)[i] = xx;

  232.             filter[i]       = fone;

  233.             xDstInSrc      += xInc;

  234.         }

  235.     } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||

  236.                (flags & SWS_FAST_BILINEAR)) { // bilinear upscale

  237.         int i;

  238.         int xDstInSrc;

  239.         filterSize = 2;

  240.         FF_ALLOC_OR_GOTO(NULL, filter,

  241.                          dstW * sizeof(*filter) * filterSize, fail);

  242. 

  243.         xDstInSrc = xInc / 2 - 0x8000;

  244.         for (i = 0; i < dstW; i++) {

  245.             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;

  246.             int j;

  247. 

  248.             (*filterPos)[i] = xx;

  249.             // bilinear upscale / linear interpolate / area averaging

  250.             for (j = 0; j < filterSize; j++) {

  251.                 int64_t coeff = fone - FFABS((xx << 16) - xDstInSrc) *

  252.                                 (fone >> 16);

  253.                 if (coeff < 0)

  254.                     coeff = 0;

  255.                 filter[i * filterSize + j] = coeff;

  256.                 xx++;

  257.             }

  258.             xDstInSrc += xInc;

  259.         }

  260.     } else {

  261.         int64_t xDstInSrc;

  262.         int sizeFactor;

  263. 

  264.         if (flags & SWS_BICUBIC)

  265.             sizeFactor = 4;

  266.         else if (flags & SWS_X)

  267.             sizeFactor = 8;

  268.         else if (flags & SWS_AREA)

  269.             sizeFactor = 1;     // downscale only, for upscale it is bilinear

  270.         else if (flags & SWS_GAUSS)

  271.             sizeFactor = 8;     // infinite ;)

  272.         else if (flags & SWS_LANCZOS)

  273.             sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;

  274.         else if (flags & SWS_SINC)

  275.             sizeFactor = 20;    // infinite ;)

  276.         else if (flags & SWS_SPLINE)

  277.             sizeFactor = 20;    // infinite ;)

  278.         else if (flags & SWS_BILINEAR)

  279.             sizeFactor = 2;

  280.         else {

  281.             sizeFactor = 0;     // GCC warning killer

  282.             assert(0);

  283.         }

  284. 

  285.         if (xInc <= 1 << 16)

  286.             filterSize = 1 + sizeFactor;    // upscale

  287.         else

  288.             filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;

  289. 

  290.         filterSize = FFMIN(filterSize, srcW - 2);

  291.         filterSize = FFMAX(filterSize, 1);

  292. 

  293.         FF_ALLOC_OR_GOTO(NULL, filter,

  294.                          dstW * sizeof(*filter) * filterSize, fail);

  295. 

  296.         xDstInSrc = xInc - 0x10000;

  297.         for (i = 0; i < dstW; i++) {

  298.             int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);

  299.             int j;

  300.             (*filterPos)[i] = xx;

  301.             for (j = 0; j < filterSize; j++) {

  302.                 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;

  303.                 double floatd;

  304.                 int64_t coeff;

  305. 

  306.                 if (xInc > 1 << 16)

  307.                     d = d * dstW / srcW;

  308.                 floatd = d * (1.0 / (1 << 30));

  309. 

  310.                 if (flags & SWS_BICUBIC) {

  311.                     int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1 << 24);

  312.                     int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);

  313. 

  314.                     if (d >= 1LL << 31) {

  315.                         coeff = 0.0;

  316.                     } else {

  317.                         int64_t dd  = (d  * d) >> 30;

  318.                         int64_t ddd = (dd * d) >> 30;

  319. 

  320.                         if (d < 1LL << 30)

  321.                             coeff =  (12 * (1 << 24) -  9 * B - 6 * C) * ddd +

  322.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  323.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  324.                         else

  325.                             coeff =      (-B -  6 * C) * ddd +

  326.                                       (6 * B + 30 * C) * dd  +

  327.                                     (-12 * B - 48 * C) * d   +

  328.                                       (8 * B + 24 * C) * (1 << 30);

  329.                     }

  330.                     coeff *= fone >> (30 + 24);

  331.                 }

  332. #if 0

  333.                 else if (flags & SWS_X) {

  334.                     double p  = param ? param * 0.01 : 0.3;

  335.                     coeff     = d ? sin(d * M_PI) / (d * M_PI) : 1.0;

  336.                     coeff    *= pow(2.0, -p * d * d);

  337.                 }

  338. #endif

  339.                 else if (flags & SWS_X) {

  340.                     double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;

  341.                     double c;

  342. 

  343.                     if (floatd < 1.0)

  344.                         c = cos(floatd * M_PI);

  345.                     else

  346.                         c = -1.0;

  347.                     if (c < 0.0)

  348.                         c = -pow(-c, A);

  349.                     else

  350.                         c = pow(c, A);

  351.                     coeff = (c * 0.5 + 0.5) * fone;

  352.                 } else if (flags & SWS_AREA) {

  353.                     int64_t d2 = d - (1 << 29);

  354.                     if (d2 * xInc < -(1LL << (29 + 16)))

  355.                         coeff = 1.0 * (1LL << (30 + 16));

  356.                     else if (d2 * xInc < (1LL << (29 + 16)))

  357.                         coeff = -d2 * xInc + (1LL << (29 + 16));

  358.                     else

  359.                         coeff = 0.0;

  360.                     coeff *= fone >> (30 + 16);

  361.                 } else if (flags & SWS_GAUSS) {

  362.                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;

  363.                     coeff = (pow(2.0, -p * floatd * floatd)) * fone;

  364.                 } else if (flags & SWS_SINC) {

  365.                     coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;

  366.                 } else if (flags & SWS_LANCZOS) {

  367.                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;

  368.                     coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /

  369.                              (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;

  370.                     if (floatd > p)

  371.                         coeff = 0;

  372.                 } else if (flags & SWS_BILINEAR) {

  373.                     coeff = (1 << 30) - d;

  374.                     if (coeff < 0)

  375.                         coeff = 0;

  376.                     coeff *= fone >> 30;

  377.                 } else if (flags & SWS_SPLINE) {

  378.                     double p = -2.196152422706632;

  379.                     coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;

  380.                 } else {

  381.                     coeff = 0.0; // GCC warning killer

  382.                     assert(0);

  383.                 }

  384. 

  385.                 filter[i * filterSize + j] = coeff;

  386.                 xx++;

  387.             }

  388.             xDstInSrc += 2 * xInc;

  389.         }

  390.     }

  391. 

  392.     /* apply src & dst Filter to filter -> filter2

  393.      * av_free(filter);

  394.      */

  395.     assert(filterSize > 0);

  396.     filter2Size = filterSize;

  397.     if (srcFilter)

  398.         filter2Size += srcFilter->length - 1;

  399.     if (dstFilter)

  400.         filter2Size += dstFilter->length - 1;

  401.     assert(filter2Size > 0);

  402.     FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);

  403. 

  404.     for (i = 0; i < dstW; i++) {

  405.         int j, k;

  406. 

  407.         if (srcFilter) {

  408.             for (k = 0; k < srcFilter->length; k++) {

  409.                 for (j = 0; j < filterSize; j++)

  410.                     filter2[i * filter2Size + k + j] +=

  411.                         srcFilter->coeff[k] * filter[i * filterSize + j];

  412.             }

  413.         } else {

  414.             for (j = 0; j < filterSize; j++)

  415.                 filter2[i * filter2Size + j] = filter[i * filterSize + j];

  416.         }

  417.         // FIXME dstFilter

  418. 

  419.         (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;

  420.     }

  421.     av_freep(&filter);

  422. 

  423.     /* try to reduce the filter-size (step1 find size and shift left) */

  424.     // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).

  425.     minFilterSize = 0;

  426.     for (i = dstW - 1; i >= 0; i--) {

  427.         int min = filter2Size;

  428.         int j;

  429.         int64_t cutOff = 0.0;

  430. 

  431.         /* get rid of near zero elements on the left by shifting left */

  432.         for (j = 0; j < filter2Size; j++) {

  433.             int k;

  434.             cutOff += FFABS(filter2[i * filter2Size]);

  435. 

  436.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  437.                 break;

  438. 

  439.             /* preserve monotonicity because the core can't handle the

  440.              * filter otherwise */

  441.             if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])

  442.                 break;

  443. 

  444.             // move filter coefficients left

  445.             for (k = 1; k < filter2Size; k++)

  446.                 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];

  447.             filter2[i * filter2Size + k - 1] = 0;

  448.             (*filterPos)[i]++;

  449.         }

  450. 

  451.         cutOff = 0;

  452.         /* count near zeros on the right */

  453.         for (j = filter2Size - 1; j > 0; j--) {

  454.             cutOff += FFABS(filter2[i * filter2Size + j]);

  455. 

  456.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  457.                 break;

  458.             min--;

  459.         }

  460. 

  461.         if (min > minFilterSize)

  462.             minFilterSize = min;

  463.     }

  464. 

  465.     if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {

  466.         // we can handle the special case 4, so we don't want to go the full 8

  467.         if (minFilterSize < 5)

  468.             filterAlign = 4;

  469. 

  470.         /* We really don't want to waste our time doing useless computation, so

  471.          * fall back on the scalar C code for very small filters.

  472.          * Vectorizing is worth it only if you have a decent-sized vector. */

  473.         if (minFilterSize < 3)

  474.             filterAlign = 1;

  475.     }

  476. 

  477.     if (INLINE_MMX(cpu_flags)) {

  478.         // special case for unscaled vertical filtering

  479.         if (minFilterSize == 1 && filterAlign == 2)

  480.             filterAlign = 1;

  481.     }

  482. 

  483.     assert(minFilterSize > 0);

  484.     filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));

  485.     assert(filterSize > 0);

  486.     filter = av_malloc(filterSize * dstW * sizeof(*filter));

  487.     if (filterSize >= MAX_FILTER_SIZE * 16 /

  488.                       ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)

  489.         goto fail;

  490.     *outFilterSize = filterSize;

  491. 

  492.     if (flags & SWS_PRINT_INFO)

  493.         av_log(NULL, AV_LOG_VERBOSE,

  494.                "SwScaler: reducing / aligning filtersize %d -> %d\n",

  495.                filter2Size, filterSize);

  496.     /* try to reduce the filter-size (step2 reduce it) */

  497.     for (i = 0; i < dstW; i++) {

  498.         int j;

  499. 

  500.         for (j = 0; j < filterSize; j++) {

  501.             if (j >= filter2Size)

  502.                 filter[i * filterSize + j] = 0;

  503.             else

  504.                 filter[i * filterSize + j] = filter2[i * filter2Size + j];

  505.             if ((flags & SWS_BITEXACT) && j >= minFilterSize)

  506.                 filter[i * filterSize + j] = 0;

  507.         }

  508.     }

  509. 

  510.     // FIXME try to align filterPos if possible

  511. 

  512.     // fix borders

  513.     if (is_horizontal) {

  514.         for (i = 0; i < dstW; i++) {

  515.             int j;

  516.             if ((*filterPos)[i] < 0) {

  517.                 // move filter coefficients left to compensate for filterPos

  518.                 for (j = 1; j < filterSize; j++) {

  519.                     int left = FFMAX(j + (*filterPos)[i], 0);

  520.                     filter[i * filterSize + left] += filter[i * filterSize + j];

  521.                     filter[i * filterSize + j]     = 0;

  522.                 }

  523.                 (*filterPos)[i] = 0;

  524.             }

  525. 

  526.             if ((*filterPos)[i] + filterSize > srcW) {

  527.                 int shift = (*filterPos)[i] + filterSize - srcW;

  528.                 // move filter coefficients right to compensate for filterPos

  529.                 for (j = filterSize - 2; j >= 0; j--) {

  530.                     int right = FFMIN(j + shift, filterSize - 1);

  531.                     filter[i * filterSize + right] += filter[i * filterSize + j];

  532.                     filter[i * filterSize + j]      = 0;

  533.                 }

  534.                 (*filterPos)[i] = srcW - filterSize;

  535.             }

  536.         }

  537.     }

  538. 

  539.     // Note the +1 is for the MMX scaler which reads over the end

  540.     /* align at 16 for AltiVec (needed by hScale_altivec_real) */

  541.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

  542.                       *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);

  543. 

  544.     /* normalize & store in outFilter */

  545.     for (i = 0; i < dstW; i++) {

  546.         int j;

  547.         int64_t error = 0;

  548.         int64_t sum   = 0;

  549. 

  550.         for (j = 0; j < filterSize; j++) {

  551.             sum += filter[i * filterSize + j];

  552.         }

  553.         sum = (sum + one / 2) / one;

  554.         for (j = 0; j < *outFilterSize; j++) {

  555.             int64_t v = filter[i * filterSize + j] + error;

  556.             int intV  = ROUNDED_DIV(v, sum);

  557.             (*outFilter)[i * (*outFilterSize) + j] = intV;

  558.             error                                  = v - intV * sum;

  559.         }

  560.     }

  561. 

  562.     (*filterPos)[dstW + 0] =

  563.     (*filterPos)[dstW + 1] =

  564.     (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will

  565.                                                       * read over the end */

  566.     for (i = 0; i < *outFilterSize; i++) {

  567.         int k = (dstW - 1) * (*outFilterSize) + i;

  568.         (*outFilter)[k + 1 * (*outFilterSize)] =

  569.         (*outFilter)[k + 2 * (*outFilterSize)] =

  570.         (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];

  571.     }

  572. 

  573.     ret = 0;

  574. 

  575. fail:

  576.     av_free(filter);

  577.     av_free(filter2);

  578.     return ret;

  579. }

  580. 

  581. #if HAVE_MMXEXT_INLINE

  582. static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode,

  583.                            int16_t *filter, int32_t *filterPos, int numSplits)

  584. {

  585.     uint8_t *fragmentA;

  586.     x86_reg imm8OfPShufW1A;

  587.     x86_reg imm8OfPShufW2A;

  588.     x86_reg fragmentLengthA;

  589.     uint8_t *fragmentB;

  590.     x86_reg imm8OfPShufW1B;

  591.     x86_reg imm8OfPShufW2B;

  592.     x86_reg fragmentLengthB;

  593.     int fragmentPos;

  594. 

  595.     int xpos, i;

  596. 

  597.     // create an optimized horizontal scaling routine

  598.     /* This scaler is made of runtime-generated MMX2 code using specially tuned

  599.      * pshufw instructions. For every four output pixels, if four input pixels

  600.      * are enough for the fast bilinear scaling, then a chunk of fragmentB is

  601.      * used. If five input pixels are needed, then a chunk of fragmentA is used.

  602.      */

  603. 

  604.     // code fragment

  605. 

  606.     __asm__ volatile (

  607.         "jmp                         9f                 \n\t"

  608.         // Begin

  609.         "0:                                             \n\t"

  610.         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"

  611.         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"

  612.         "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"

  613.         "punpcklbw                %%mm7, %%mm1          \n\t"

  614.         "punpcklbw                %%mm7, %%mm0          \n\t"

  615.         "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"

  616.         "1:                                             \n\t"

  617.         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"

  618.         "2:                                             \n\t"

  619.         "psubw                    %%mm1, %%mm0          \n\t"

  620.         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"

  621.         "pmullw                   %%mm3, %%mm0          \n\t"

  622.         "psllw                       $7, %%mm1          \n\t"

  623.         "paddw                    %%mm1, %%mm0          \n\t"

  624. 

  625.         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"

  626. 

  627.         "add                         $8, %%"REG_a"      \n\t"

  628.         // End

  629.         "9:                                             \n\t"

  630.         // "int $3                                         \n\t"

  631.         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"

  632.         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"

  633.         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"

  634.         "dec                         %1                 \n\t"

  635.         "dec                         %2                 \n\t"

  636.         "sub                         %0, %1             \n\t"

  637.         "sub                         %0, %2             \n\t"

  638.         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"

  639.         "sub                         %0, %3             \n\t"

  640. 

  641. 

  642.         : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),

  643.           "=r" (fragmentLengthA)

  644.         );

  645. 

  646.     __asm__ volatile (

  647.         "jmp                         9f                 \n\t"

  648.         // Begin

  649.         "0:                                             \n\t"

  650.         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"

  651.         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"

  652.         "punpcklbw                %%mm7, %%mm0          \n\t"

  653.         "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"

  654.         "1:                                             \n\t"

  655.         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"

  656.         "2:                                             \n\t"

  657.         "psubw                    %%mm1, %%mm0          \n\t"

  658.         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"

  659.         "pmullw                   %%mm3, %%mm0          \n\t"

  660.         "psllw                       $7, %%mm1          \n\t"

  661.         "paddw                    %%mm1, %%mm0          \n\t"

  662. 

  663.         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"

  664. 

  665.         "add                         $8, %%"REG_a"      \n\t"

  666.         // End

  667.         "9:                                             \n\t"

  668.         // "int                       $3                   \n\t"

  669.         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"

  670.         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"

  671.         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"

  672.         "dec                         %1                 \n\t"

  673.         "dec                         %2                 \n\t"

  674.         "sub                         %0, %1             \n\t"

  675.         "sub                         %0, %2             \n\t"

  676.         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"

  677.         "sub                         %0, %3             \n\t"

  678. 

  679. 

  680.         : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),

  681.           "=r" (fragmentLengthB)

  682.         );

  683. 

  684.     xpos        = 0; // lumXInc/2 - 0x8000; // difference between pixel centers

  685.     fragmentPos = 0;

  686. 

  687.     for (i = 0; i < dstW / numSplits; i++) {

  688.         int xx = xpos >> 16;

  689. 

  690.         if ((i & 3) == 0) {

  691.             int a                  = 0;

  692.             int b                  = ((xpos + xInc) >> 16) - xx;

  693.             int c                  = ((xpos + xInc * 2) >> 16) - xx;

  694.             int d                  = ((xpos + xInc * 3) >> 16) - xx;

  695.             int inc                = (d + 1 < 4);

  696.             uint8_t *fragment      = (d + 1 < 4) ? fragmentB : fragmentA;

  697.             x86_reg imm8OfPShufW1  = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;

  698.             x86_reg imm8OfPShufW2  = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;

  699.             x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;

  700.             int maxShift           = 3 - (d + inc);

  701.             int shift              = 0;

  702. 

  703.             if (filterCode) {

  704.                 filter[i]        = ((xpos              & 0xFFFF) ^ 0xFFFF) >> 9;

  705.                 filter[i + 1]    = (((xpos + xInc)     & 0xFFFF) ^ 0xFFFF) >> 9;

  706.                 filter[i + 2]    = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;

  707.                 filter[i + 3]    = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;

  708.                 filterPos[i / 2] = xx;

  709. 

  710.                 memcpy(filterCode + fragmentPos, fragment, fragmentLength);

  711. 

  712.                 filterCode[fragmentPos + imm8OfPShufW1] =  (a + inc)       |

  713.                                                           ((b + inc) << 2) |

  714.                                                           ((c + inc) << 4) |

  715.                                                           ((d + inc) << 6);

  716.                 filterCode[fragmentPos + imm8OfPShufW2] =  a | (b << 2) |

  717.                                                                (c << 4) |

  718.                                                                (d << 6);

  719. 

  720.                 if (i + 4 - inc >= dstW)

  721.                     shift = maxShift;               // avoid overread

  722.                 else if ((filterPos[i / 2] & 3) <= maxShift)

  723.                     shift = filterPos[i / 2] & 3;   // align

  724. 

  725.                 if (shift && i >= shift) {

  726.                     filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;

  727.                     filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;

  728.                     filterPos[i / 2]                        -= shift;

  729.                 }

  730.             }

  731. 

  732.             fragmentPos += fragmentLength;

  733. 

  734.             if (filterCode)

  735.                 filterCode[fragmentPos] = RET;

  736.         }

  737.         xpos += xInc;

  738.     }

  739.     if (filterCode)

  740.         filterPos[((i / 2) + 1) & (~1)] = xpos >> 16;  // needed to jump to the next part

  741. 

  742.     return fragmentPos + 1;

  743. }

  744. #endif /* HAVE_MMXEXT_INLINE */

  745. 

  746. static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)

  747. {

  748.     *h = av_pix_fmt_descriptors[format].log2_chroma_w;

  749.     *v = av_pix_fmt_descriptors[format].log2_chroma_h;

  750. }

  751. 

  752. int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],

  753.                              int srcRange, const int table[4], int dstRange,

  754.                              int brightness, int contrast, int saturation)

  755. {

  756.     memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);

  757.     memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);

  758. 

  759.     c->brightness = brightness;

  760.     c->contrast   = contrast;

  761.     c->saturation = saturation;

  762.     c->srcRange   = srcRange;

  763.     c->dstRange   = dstRange;

  764.     if (isYUV(c->dstFormat) || isGray(c->dstFormat))

  765.         return -1;

  766. 

  767.     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);

  768.     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);

  769. 

  770.     ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,

  771.                              contrast, saturation);

  772.     // FIXME factorize

  773. 

  774.     if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)

  775.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  776.                                        contrast, saturation);

  777.     return 0;

  778. }

  779. 

  780. int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,

  781.                              int *srcRange, int **table, int *dstRange,

  782.                              int *brightness, int *contrast, int *saturation)

  783. {

  784.     if (isYUV(c->dstFormat) || isGray(c->dstFormat))

  785.         return -1;

  786. 

  787.     *inv_table  = c->srcColorspaceTable;

  788.     *table      = c->dstColorspaceTable;

  789.     *srcRange   = c->srcRange;

  790.     *dstRange   = c->dstRange;

  791.     *brightness = c->brightness;

  792.     *contrast   = c->contrast;

  793.     *saturation = c->saturation;

  794. 

  795.     return 0;

  796. }

  797. 

  798. static int handle_jpeg(enum PixelFormat *format)

  799. {

  800.     switch (*format) {

  801.     case PIX_FMT_YUVJ420P:

  802.         *format = PIX_FMT_YUV420P;

  803.         return 1;

  804.     case PIX_FMT_YUVJ422P:

  805.         *format = PIX_FMT_YUV422P;

  806.         return 1;

  807.     case PIX_FMT_YUVJ444P:

  808.         *format = PIX_FMT_YUV444P;

  809.         return 1;

  810.     case PIX_FMT_YUVJ440P:

  811.         *format = PIX_FMT_YUV440P;

  812.         return 1;

  813.     default:

  814.         return 0;

  815.     }

  816. }

  817. 

  818. SwsContext *sws_alloc_context(void)

  819. {

  820.     SwsContext *c = av_mallocz(sizeof(SwsContext));

  821. 

  822.     c->av_class = &sws_context_class;

  823.     av_opt_set_defaults(c);

  824. 

  825.     return c;

  826. }

  827. 

  828. av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,

  829.                              SwsFilter *dstFilter)

  830. {

  831.     int i;

  832.     int usesVFilter, usesHFilter;

  833.     int unscaled;

  834.     SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };

  835.     int srcW              = c->srcW;

  836.     int srcH              = c->srcH;

  837.     int dstW              = c->dstW;

  838.     int dstH              = c->dstH;

  839.     int dst_stride        = FFALIGN(dstW * sizeof(int16_t) + 16, 16);

  840.     int dst_stride_px     = dst_stride >> 1;

  841.     int flags, cpu_flags;

  842.     enum PixelFormat srcFormat = c->srcFormat;

  843.     enum PixelFormat dstFormat = c->dstFormat;

  844. 

  845.     cpu_flags = av_get_cpu_flags();

  846.     flags     = c->flags;

  847.     emms_c();

  848.     if (!rgb15to16)

  849.         sws_rgb2rgb_init();

  850. 

  851.     unscaled = (srcW == dstW && srcH == dstH);

  852. 

  853.     if (!sws_isSupportedInput(srcFormat)) {

  854.         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",

  855.                sws_format_name(srcFormat));

  856.         return AVERROR(EINVAL);

  857.     }

  858.     if (!sws_isSupportedOutput(dstFormat)) {

  859.         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",

  860.                sws_format_name(dstFormat));

  861.         return AVERROR(EINVAL);

  862.     }

  863. 

  864.     i = flags & (SWS_POINT         |

  865.                  SWS_AREA          |

  866.                  SWS_BILINEAR      |

  867.                  SWS_FAST_BILINEAR |

  868.                  SWS_BICUBIC       |

  869.                  SWS_X             |

  870.                  SWS_GAUSS         |

  871.                  SWS_LANCZOS       |

  872.                  SWS_SINC          |

  873.                  SWS_SPLINE        |

  874.                  SWS_BICUBLIN);

  875.     if (!i || (i & (i - 1))) {

  876.         av_log(c, AV_LOG_ERROR,

  877.                "Exactly one scaler algorithm must be chosen\n");

  878.         return AVERROR(EINVAL);

  879.     }

  880.     /* sanity check */

  881.     if (srcW < 4 || srcH < 1 || dstW < 8 || dstH < 1) {

  882.         /* FIXME check if these are enough and try to lower them after

  883.          * fixing the relevant parts of the code */

  884.         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",

  885.                srcW, srcH, dstW, dstH);

  886.         return AVERROR(EINVAL);

  887.     }

  888. 

  889.     if (!dstFilter)

  890.         dstFilter = &dummyFilter;

  891.     if (!srcFilter)

  892.         srcFilter = &dummyFilter;

  893. 

  894.     c->lumXInc      = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;

  895.     c->lumYInc      = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;

  896.     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);

  897.     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);

  898.     c->vRounder     = 4 * 0x0001000100010001ULL;

  899. 

  900.     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||

  901.                   (srcFilter->chrV && srcFilter->chrV->length > 1) ||

  902.                   (dstFilter->lumV && dstFilter->lumV->length > 1) ||

  903.                   (dstFilter->chrV && dstFilter->chrV->length > 1);

  904.     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||

  905.                   (srcFilter->chrH && srcFilter->chrH->length > 1) ||

  906.                   (dstFilter->lumH && dstFilter->lumH->length > 1) ||

  907.                   (dstFilter->chrH && dstFilter->chrH->length > 1);

  908. 

  909.     getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);

  910.     getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);

  911. 

  912.     /* reuse chroma for 2 pixels RGB/BGR unless user wants full

  913.      * chroma interpolation */

  914.     if (flags & SWS_FULL_CHR_H_INT &&

  915.         isAnyRGB(dstFormat)        &&

  916.         dstFormat != PIX_FMT_RGBA  &&

  917.         dstFormat != PIX_FMT_ARGB  &&

  918.         dstFormat != PIX_FMT_BGRA  &&

  919.         dstFormat != PIX_FMT_ABGR  &&

  920.         dstFormat != PIX_FMT_RGB24 &&

  921.         dstFormat != PIX_FMT_BGR24) {

  922.         av_log(c, AV_LOG_ERROR,

  923.                "full chroma interpolation for destination format '%s' not yet implemented\n",

  924.                sws_format_name(dstFormat));

  925.         flags   &= ~SWS_FULL_CHR_H_INT;

  926.         c->flags = flags;

  927.     }

  928.     if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))

  929.         c->chrDstHSubSample = 1;

  930. 

  931.     // drop some chroma lines if the user wants it

  932.     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>

  933.                            SWS_SRC_V_CHR_DROP_SHIFT;

  934.     c->chrSrcVSubSample += c->vChrDrop;

  935. 

  936.     /* drop every other pixel for chroma calculation unless user

  937.      * wants full chroma */

  938.     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&

  939.         srcFormat != PIX_FMT_RGB8 && srcFormat != PIX_FMT_BGR8 &&

  940.         srcFormat != PIX_FMT_RGB4 && srcFormat != PIX_FMT_BGR4 &&

  941.         srcFormat != PIX_FMT_RGB4_BYTE && srcFormat != PIX_FMT_BGR4_BYTE &&

  942.         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||

  943.          (flags & SWS_FAST_BILINEAR)))

  944.         c->chrSrcHSubSample = 1;

  945. 

  946.     // Note the -((-x)>>y) is so that we always round toward +inf.

  947.     c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);

  948.     c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);

  949.     c->chrDstW = -((-dstW) >> c->chrDstHSubSample);

  950.     c->chrDstH = -((-dstH) >> c->chrDstVSubSample);

  951. 

  952.     /* unscaled special cases */

  953.     if (unscaled && !usesHFilter && !usesVFilter &&

  954.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  955.         ff_get_unscaled_swscale(c);

  956. 

  957.         if (c->swScale) {

  958.             if (flags & SWS_PRINT_INFO)

  959.                 av_log(c, AV_LOG_INFO,

  960.                        "using unscaled %s -> %s special converter\n",

  961.                        sws_format_name(srcFormat), sws_format_name(dstFormat));

  962.             return 0;

  963.         }

  964.     }

  965. 

  966.     c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;

  967.     if (c->srcBpc < 8)

  968.         c->srcBpc = 8;

  969.     c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;

  970.     if (c->dstBpc < 8)

  971.         c->dstBpc = 8;

  972.     if (c->dstBpc == 16)

  973.         dst_stride <<= 1;

  974.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

  975.                      (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16,

  976.                      fail);

  977.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 10) {

  978.         c->canMMX2BeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  979.                             (srcW & 15) == 0) ? 1 : 0;

  980.         if (!c->canMMX2BeUsed && dstW >= srcW && (srcW & 15) == 0

  981.             && (flags & SWS_FAST_BILINEAR)) {

  982.             if (flags & SWS_PRINT_INFO)

  983.                 av_log(c, AV_LOG_INFO,

  984.                        "output width is not a multiple of 32 -> no MMX2 scaler\n");

  985.         }

  986.         if (usesHFilter)

  987.             c->canMMX2BeUsed = 0;

  988.     } else

  989.         c->canMMX2BeUsed = 0;

  990. 

  991.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  992.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  993. 

  994.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  995.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  996.      * correct scaling.

  997.      * n-2 is the last chrominance sample available.

  998.      * This is not perfect, but no one should notice the difference, the more

  999.      * correct variant would be like the vertical one, but that would require

  1000.      * some special code for the first and last pixel */

  1001.     if (flags & SWS_FAST_BILINEAR) {

  1002.         if (c->canMMX2BeUsed) {

  1003.             c->lumXInc += 20;

  1004.             c->chrXInc += 20;

  1005.         }

  1006.         // we don't use the x86 asm scaler if MMX is available

  1007.         else if (INLINE_MMX(cpu_flags)) {

  1008.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1009.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1010.         }

  1011.     }

  1012. 

  1013.     /* precalculate horizontal scaler filter coefficients */

  1014.     {

  1015. #if HAVE_MMXEXT_INLINE

  1016. // can't downscale !!!

  1017.         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {

  1018.             c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,

  1019.                                                        NULL, NULL, 8);

  1020.             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,

  1021.                                                        NULL, NULL, NULL, 4);

  1022. 

  1023. #ifdef MAP_ANONYMOUS

  1024.             c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1025.             c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1026. #elif HAVE_VIRTUALALLOC

  1027.             c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1028.             c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1029. #else

  1030.             c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);

  1031.             c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);

  1032. #endif

  1033. 

  1034.             if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)

  1035.                 return AVERROR(ENOMEM);

  1036.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1037.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1038.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1039.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1040. 

  1041.             initMMX2HScaler(dstW, c->lumXInc, c->lumMmx2FilterCode,

  1042.                             c->hLumFilter, c->hLumFilterPos, 8);

  1043.             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,

  1044.                             c->hChrFilter, c->hChrFilterPos, 4);

  1045. 

  1046. #ifdef MAP_ANONYMOUS

  1047.             mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1048.             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1049. #endif

  1050.         } else

  1051. #endif /* HAVE_MMXEXT_INLINE */

  1052.         {

  1053.             const int filterAlign = INLINE_MMX(cpu_flags)  ? 4 :

  1054.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1055.                 1;

  1056. 

  1057.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1058.                            &c->hLumFilterSize, c->lumXInc,

  1059.                            srcW, dstW, filterAlign, 1 << 14,

  1060.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1061.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1062.                            c->param, 1) < 0)

  1063.                 goto fail;

  1064.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1065.                            &c->hChrFilterSize, c->chrXInc,

  1066.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1067.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1068.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1069.                            c->param, 1) < 0)

  1070.                 goto fail;

  1071.         }

  1072.     } // initialize horizontal stuff

  1073. 

  1074.     /* precalculate vertical scaler filter coefficients */

  1075.     {

  1076.         const int filterAlign = INLINE_MMX(cpu_flags)  ? 2 :

  1077.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1078.             1;

  1079. 

  1080.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1081.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1082.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1083.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1084.                        c->param, 0) < 0)

  1085.             goto fail;

  1086.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1087.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1088.                        filterAlign, (1 << 12),

  1089.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1090.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1091.                        c->param, 0) < 0)

  1092.             goto fail;

  1093. 

  1094. #if HAVE_ALTIVEC

  1095.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1096.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1097. 

  1098.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1099.             int j;

  1100.             short *p = (short *)&c->vYCoeffsBank[i];

  1101.             for (j = 0; j < 8; j++)

  1102.                 p[j] = c->vLumFilter[i];

  1103.         }

  1104. 

  1105.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1106.             int j;

  1107.             short *p = (short *)&c->vCCoeffsBank[i];

  1108.             for (j = 0; j < 8; j++)

  1109.                 p[j] = c->vChrFilter[i];

  1110.         }

  1111. #endif

  1112.     }

  1113. 

  1114.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1115.     c->vLumBufSize = c->vLumFilterSize;

  1116.     c->vChrBufSize = c->vChrFilterSize;

  1117.     for (i = 0; i < dstH; i++) {

  1118.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1119.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1120.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1121.                                << c->chrSrcVSubSample));

  1122. 

  1123.         nextSlice >>= c->chrSrcVSubSample;

  1124.         nextSlice <<= c->chrSrcVSubSample;

  1125.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1126.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1127.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1128.             (nextSlice >> c->chrSrcVSubSample))

  1129.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1130.                              c->vChrFilterPos[chrI];

  1131.     }

  1132. 

  1133.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1134.      * need to allocate several megabytes to handle all possible cases) */

  1135.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1136.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1137.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1138.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1139.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1140.     /* Note we need at least one pixel more at the end because of the MMX code

  1141.      * (just in case someone wants to replace the 4000/8000). */

  1142.     /* align at 16 bytes for AltiVec */

  1143.     for (i = 0; i < c->vLumBufSize; i++) {

  1144.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1145.                           dst_stride + 16, fail);

  1146.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1147.     }

  1148.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1149.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc & ~7);

  1150.     c->uv_off_byte = dst_stride + 16;

  1151.     for (i = 0; i < c->vChrBufSize; i++) {

  1152.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1153.                          dst_stride * 2 + 32, fail);

  1154.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1155.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1156.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1157.     }

  1158.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1159.         for (i = 0; i < c->vLumBufSize; i++) {

  1160.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1161.                               dst_stride + 16, fail);

  1162.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1163.         }

  1164. 

  1165.     // try to avoid drawing green stuff between the right end and the stride end

  1166.     for (i = 0; i < c->vChrBufSize; i++)

  1167.         memset(c->chrUPixBuf[i], 64, dst_stride * 2 + 1);

  1168. 

  1169.     assert(c->chrDstH <= dstH);

  1170. 

  1171.     if (flags & SWS_PRINT_INFO) {

  1172.         if (flags & SWS_FAST_BILINEAR)

  1173.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1174.         else if (flags & SWS_BILINEAR)

  1175.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1176.         else if (flags & SWS_BICUBIC)

  1177.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1178.         else if (flags & SWS_X)

  1179.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1180.         else if (flags & SWS_POINT)

  1181.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1182.         else if (flags & SWS_AREA)

  1183.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1184.         else if (flags & SWS_BICUBLIN)

  1185.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1186.         else if (flags & SWS_GAUSS)

  1187.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1188.         else if (flags & SWS_SINC)

  1189.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1190.         else if (flags & SWS_LANCZOS)

  1191.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1192.         else if (flags & SWS_SPLINE)

  1193.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1194.         else

  1195.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1196. 

  1197.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1198.                sws_format_name(srcFormat),

  1199. #ifdef DITHER1XBPP

  1200.                dstFormat == PIX_FMT_BGR555   || dstFormat == PIX_FMT_BGR565   ||

  1201.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1202.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ?

  1203.                                                              "dithered " : "",

  1204. #else

  1205.                "",

  1206. #endif

  1207.                sws_format_name(dstFormat));

  1208. 

  1209.         if (INLINE_MMXEXT(cpu_flags))

  1210.             av_log(c, AV_LOG_INFO, "using MMX2\n");

  1211.         else if (INLINE_AMD3DNOW(cpu_flags))

  1212.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1213.         else if (INLINE_MMX(cpu_flags))

  1214.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1215.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1216.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1217.         else

  1218.             av_log(c, AV_LOG_INFO, "using C\n");

  1219. 

  1220.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1221.         av_log(c, AV_LOG_DEBUG,

  1222.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1223.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1224.         av_log(c, AV_LOG_DEBUG,

  1225.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1226.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1227.                c->chrXInc, c->chrYInc);

  1228.     }

  1229. 

  1230.     c->swScale = ff_getSwsFunc(c);

  1231.     return 0;

  1232. fail: // FIXME replace things by appropriate error codes

  1233.     return -1;

  1234. }

  1235. 

  1236. #if FF_API_SWS_GETCONTEXT

  1237. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1238.                            int dstW, int dstH, enum PixelFormat dstFormat,

  1239.                            int flags, SwsFilter *srcFilter,

  1240.                            SwsFilter *dstFilter, const double *param)

  1241. {

  1242.     SwsContext *c;

  1243. 

  1244.     if (!(c = sws_alloc_context()))

  1245.         return NULL;

  1246. 

  1247.     c->flags     = flags;

  1248.     c->srcW      = srcW;

  1249.     c->srcH      = srcH;

  1250.     c->dstW      = dstW;

  1251.     c->dstH      = dstH;

  1252.     c->srcRange  = handle_jpeg(&srcFormat);

  1253.     c->dstRange  = handle_jpeg(&dstFormat);

  1254.     c->srcFormat = srcFormat;

  1255.     c->dstFormat = dstFormat;

  1256. 

  1257.     if (param) {

  1258.         c->param[0] = param[0];

  1259.         c->param[1] = param[1];

  1260.     }

  1261.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1262.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1263.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1264. 

  1265.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1266.         sws_freeContext(c);

  1267.         return NULL;

  1268.     }

  1269. 

  1270.     return c;

  1271. }

  1272. #endif

  1273. 

  1274. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1275.                                 float lumaSharpen, float chromaSharpen,

  1276.                                 float chromaHShift, float chromaVShift,

  1277.                                 int verbose)

  1278. {

  1279.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1280.     if (!filter)

  1281.         return NULL;

  1282. 

  1283.     if (lumaGBlur != 0.0) {

  1284.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1285.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1286.     } else {

  1287.         filter->lumH = sws_getIdentityVec();

  1288.         filter->lumV = sws_getIdentityVec();

  1289.     }

  1290. 

  1291.     if (chromaGBlur != 0.0) {

  1292.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1293.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1294.     } else {

  1295.         filter->chrH = sws_getIdentityVec();

  1296.         filter->chrV = sws_getIdentityVec();

  1297.     }

  1298. 

  1299.     if (chromaSharpen != 0.0) {

  1300.         SwsVector *id = sws_getIdentityVec();

  1301.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1302.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1303.         sws_addVec(filter->chrH, id);

  1304.         sws_addVec(filter->chrV, id);

  1305.         sws_freeVec(id);

  1306.     }

  1307. 

  1308.     if (lumaSharpen != 0.0) {

  1309.         SwsVector *id = sws_getIdentityVec();

  1310.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1311.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1312.         sws_addVec(filter->lumH, id);

  1313.         sws_addVec(filter->lumV, id);

  1314.         sws_freeVec(id);

  1315.     }

  1316. 

  1317.     if (chromaHShift != 0.0)

  1318.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1319. 

  1320.     if (chromaVShift != 0.0)

  1321.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1322. 

  1323.     sws_normalizeVec(filter->chrH, 1.0);

  1324.     sws_normalizeVec(filter->chrV, 1.0);

  1325.     sws_normalizeVec(filter->lumH, 1.0);

  1326.     sws_normalizeVec(filter->lumV, 1.0);

  1327. 

  1328.     if (verbose)

  1329.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1330.     if (verbose)

  1331.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1332. 

  1333.     return filter;

  1334. }

  1335. 

  1336. SwsVector *sws_allocVec(int length)

  1337. {

  1338.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1339.     if (!vec)

  1340.         return NULL;

  1341.     vec->length = length;

  1342.     vec->coeff  = av_malloc(sizeof(double) * length);

  1343.     if (!vec->coeff)

  1344.         av_freep(&vec);

  1345.     return vec;

  1346. }

  1347. 

  1348. SwsVector *sws_getGaussianVec(double variance, double quality)

  1349. {

  1350.     const int length = (int)(variance * quality + 0.5) | 1;

  1351.     int i;

  1352.     double middle  = (length - 1) * 0.5;

  1353.     SwsVector *vec = sws_allocVec(length);

  1354. 

  1355.     if (!vec)

  1356.         return NULL;

  1357. 

  1358.     for (i = 0; i < length; i++) {

  1359.         double dist = i - middle;

  1360.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1361.                         sqrt(2 * variance * M_PI);

  1362.     }

  1363. 

  1364.     sws_normalizeVec(vec, 1.0);

  1365. 

  1366.     return vec;

  1367. }

  1368. 

  1369. SwsVector *sws_getConstVec(double c, int length)

  1370. {

  1371.     int i;

  1372.     SwsVector *vec = sws_allocVec(length);

  1373. 

  1374.     if (!vec)

  1375.         return NULL;

  1376. 

  1377.     for (i = 0; i < length; i++)

  1378.         vec->coeff[i] = c;

  1379. 

  1380.     return vec;

  1381. }

  1382. 

  1383. SwsVector *sws_getIdentityVec(void)

  1384. {

  1385.     return sws_getConstVec(1.0, 1);

  1386. }

  1387. 

  1388. static double sws_dcVec(SwsVector *a)

  1389. {

  1390.     int i;

  1391.     double sum = 0;

  1392. 

  1393.     for (i = 0; i < a->length; i++)

  1394.         sum += a->coeff[i];

  1395. 

  1396.     return sum;

  1397. }

  1398. 

  1399. void sws_scaleVec(SwsVector *a, double scalar)

  1400. {

  1401.     int i;

  1402. 

  1403.     for (i = 0; i < a->length; i++)

  1404.         a->coeff[i] *= scalar;

  1405. }

  1406. 

  1407. void sws_normalizeVec(SwsVector *a, double height)

  1408. {

  1409.     sws_scaleVec(a, height / sws_dcVec(a));

  1410. }

  1411. 

  1412. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1413. {

  1414.     int length = a->length + b->length - 1;

  1415.     int i, j;

  1416.     SwsVector *vec = sws_getConstVec(0.0, length);

  1417. 

  1418.     if (!vec)

  1419.         return NULL;

  1420. 

  1421.     for (i = 0; i < a->length; i++) {

  1422.         for (j = 0; j < b->length; j++) {

  1423.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1424.         }

  1425.     }

  1426. 

  1427.     return vec;

  1428. }

  1429. 

  1430. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1431. {

  1432.     int length = FFMAX(a->length, b->length);

  1433.     int i;

  1434.     SwsVector *vec = sws_getConstVec(0.0, length);

  1435. 

  1436.     if (!vec)

  1437.         return NULL;

  1438. 

  1439.     for (i = 0; i < a->length; i++)

  1440.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1441.     for (i = 0; i < b->length; i++)

  1442.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1443. 

  1444.     return vec;

  1445. }

  1446. 

  1447. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1448. {

  1449.     int length = FFMAX(a->length, b->length);

  1450.     int i;

  1451.     SwsVector *vec = sws_getConstVec(0.0, length);

  1452. 

  1453.     if (!vec)

  1454.         return NULL;

  1455. 

  1456.     for (i = 0; i < a->length; i++)

  1457.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1458.     for (i = 0; i < b->length; i++)

  1459.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1460. 

  1461.     return vec;

  1462. }

  1463. 

  1464. /* shift left / or right if "shift" is negative */

  1465. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1466. {

  1467.     int length = a->length + FFABS(shift) * 2;

  1468.     int i;

  1469.     SwsVector *vec = sws_getConstVec(0.0, length);

  1470. 

  1471.     if (!vec)

  1472.         return NULL;

  1473. 

  1474.     for (i = 0; i < a->length; i++) {

  1475.         vec->coeff[i + (length    - 1) / 2 -

  1476.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1477.     }

  1478. 

  1479.     return vec;

  1480. }

  1481. 

  1482. void sws_shiftVec(SwsVector *a, int shift)

  1483. {

  1484.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1485.     av_free(a->coeff);

  1486.     a->coeff  = shifted->coeff;

  1487.     a->length = shifted->length;

  1488.     av_free(shifted);

  1489. }

  1490. 

  1491. void sws_addVec(SwsVector *a, SwsVector *b)

  1492. {

  1493.     SwsVector *sum = sws_sumVec(a, b);

  1494.     av_free(a->coeff);

  1495.     a->coeff  = sum->coeff;

  1496.     a->length = sum->length;

  1497.     av_free(sum);

  1498. }

  1499. 

  1500. void sws_subVec(SwsVector *a, SwsVector *b)

  1501. {

  1502.     SwsVector *diff = sws_diffVec(a, b);

  1503.     av_free(a->coeff);

  1504.     a->coeff  = diff->coeff;

  1505.     a->length = diff->length;

  1506.     av_free(diff);

  1507. }

  1508. 

  1509. void sws_convVec(SwsVector *a, SwsVector *b)

  1510. {

  1511.     SwsVector *conv = sws_getConvVec(a, b);

  1512.     av_free(a->coeff);

  1513.     a->coeff  = conv->coeff;

  1514.     a->length = conv->length;

  1515.     av_free(conv);

  1516. }

  1517. 

  1518. SwsVector *sws_cloneVec(SwsVector *a)

  1519. {

  1520.     int i;

  1521.     SwsVector *vec = sws_allocVec(a->length);

  1522. 

  1523.     if (!vec)

  1524.         return NULL;

  1525. 

  1526.     for (i = 0; i < a->length; i++)

  1527.         vec->coeff[i] = a->coeff[i];

  1528. 

  1529.     return vec;

  1530. }

  1531. 

  1532. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1533. {

  1534.     int i;

  1535.     double max = 0;

  1536.     double min = 0;

  1537.     double range;

  1538. 

  1539.     for (i = 0; i < a->length; i++)

  1540.         if (a->coeff[i] > max)

  1541.             max = a->coeff[i];

  1542. 

  1543.     for (i = 0; i < a->length; i++)

  1544.         if (a->coeff[i] < min)

  1545.             min = a->coeff[i];

  1546. 

  1547.     range = max - min;

  1548. 

  1549.     for (i = 0; i < a->length; i++) {

  1550.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1551.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1552.         for (; x > 0; x--)

  1553.             av_log(log_ctx, log_level, " ");

  1554.         av_log(log_ctx, log_level, "|\n");

  1555.     }

  1556. }

  1557. 

  1558. void sws_freeVec(SwsVector *a)

  1559. {

  1560.     if (!a)

  1561.         return;

  1562.     av_freep(&a->coeff);

  1563.     a->length = 0;

  1564.     av_free(a);

  1565. }

  1566. 

  1567. void sws_freeFilter(SwsFilter *filter)

  1568. {

  1569.     if (!filter)

  1570.         return;

  1571. 

  1572.     if (filter->lumH)

  1573.         sws_freeVec(filter->lumH);

  1574.     if (filter->lumV)

  1575.         sws_freeVec(filter->lumV);

  1576.     if (filter->chrH)

  1577.         sws_freeVec(filter->chrH);

  1578.     if (filter->chrV)

  1579.         sws_freeVec(filter->chrV);

  1580.     av_free(filter);

  1581. }

  1582. 

  1583. void sws_freeContext(SwsContext *c)

  1584. {

  1585.     int i;

  1586.     if (!c)

  1587.         return;

  1588. 

  1589.     if (c->lumPixBuf) {

  1590.         for (i = 0; i < c->vLumBufSize; i++)

  1591.             av_freep(&c->lumPixBuf[i]);

  1592.         av_freep(&c->lumPixBuf);

  1593.     }

  1594. 

  1595.     if (c->chrUPixBuf) {

  1596.         for (i = 0; i < c->vChrBufSize; i++)

  1597.             av_freep(&c->chrUPixBuf[i]);

  1598.         av_freep(&c->chrUPixBuf);

  1599.         av_freep(&c->chrVPixBuf);

  1600.     }

  1601. 

  1602.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1603.         for (i = 0; i < c->vLumBufSize; i++)

  1604.             av_freep(&c->alpPixBuf[i]);

  1605.         av_freep(&c->alpPixBuf);

  1606.     }

  1607. 

  1608.     av_freep(&c->vLumFilter);

  1609.     av_freep(&c->vChrFilter);

  1610.     av_freep(&c->hLumFilter);

  1611.     av_freep(&c->hChrFilter);

  1612. #if HAVE_ALTIVEC

  1613.     av_freep(&c->vYCoeffsBank);

  1614.     av_freep(&c->vCCoeffsBank);

  1615. #endif

  1616. 

  1617.     av_freep(&c->vLumFilterPos);

  1618.     av_freep(&c->vChrFilterPos);

  1619.     av_freep(&c->hLumFilterPos);

  1620.     av_freep(&c->hChrFilterPos);

  1621. 

  1622. #if HAVE_MMX_INLINE

  1623. #ifdef MAP_ANONYMOUS

  1624.     if (c->lumMmx2FilterCode)

  1625.         munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1626.     if (c->chrMmx2FilterCode)

  1627.         munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1628. #elif HAVE_VIRTUALALLOC

  1629.     if (c->lumMmx2FilterCode)

  1630.         VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1631.     if (c->chrMmx2FilterCode)

  1632.         VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1633. #else

  1634.     av_free(c->lumMmx2FilterCode);

  1635.     av_free(c->chrMmx2FilterCode);

  1636. #endif

  1637.     c->lumMmx2FilterCode = NULL;

  1638.     c->chrMmx2FilterCode = NULL;

  1639. #endif /* HAVE_MMX_INLINE */

  1640. 

  1641.     av_freep(&c->yuvTable);

  1642.     av_free(c->formatConvBuffer);

  1643. 

  1644.     av_free(c);

  1645. }

  1646. 

  1647. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1648.                                         int srcH, enum PixelFormat srcFormat,

  1649.                                         int dstW, int dstH,

  1650.                                         enum PixelFormat dstFormat, int flags,

  1651.                                         SwsFilter *srcFilter,

  1652.                                         SwsFilter *dstFilter,

  1653.                                         const double *param)

  1654. {

  1655.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1656.                                              SWS_PARAM_DEFAULT };

  1657. 

  1658.     if (!param)

  1659.         param = default_param;

  1660. 

  1661.     if (context &&

  1662.         (context->srcW      != srcW      ||

  1663.          context->srcH      != srcH      ||

  1664.          context->srcFormat != srcFormat ||

  1665.          context->dstW      != dstW      ||

  1666.          context->dstH      != dstH      ||

  1667.          context->dstFormat != dstFormat ||

  1668.          context->flags     != flags     ||

  1669.          context->param[0]  != param[0]  ||

  1670.          context->param[1]  != param[1])) {

  1671.         sws_freeContext(context);

  1672.         context = NULL;

  1673.     }

  1674. 

  1675.     if (!context) {

  1676.         if (!(context = sws_alloc_context()))

  1677.             return NULL;

  1678.         context->srcW      = srcW;

  1679.         context->srcH      = srcH;

  1680.         context->srcRange  = handle_jpeg(&srcFormat);

  1681.         context->srcFormat = srcFormat;

  1682.         context->dstW      = dstW;

  1683.         context->dstH      = dstH;

  1684.         context->dstRange  = handle_jpeg(&dstFormat);

  1685.         context->dstFormat = dstFormat;

  1686.         context->flags     = flags;

  1687.         context->param[0]  = param[0];

  1688.         context->param[1]  = param[1];

  1689.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1690.                                  context->srcRange,

  1691.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1692.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1693.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1694.             sws_freeContext(context);

  1695.             return NULL;

  1696.         }

  1697.     }

  1698.     return context;

  1699. }