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.
34805 Commits
32 Branches
959MB
Tree: 6980453569
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '6980453569'
${ noResults }
FFmpeg/libswscale/utils.c

1702 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 "rgb2rgb.h"

  50. #include "swscale.h"

  51. #include "swscale_internal.h"

  52. 

  53. unsigned swscale_version(void)

  54. {

  55.     return LIBSWSCALE_VERSION_INT;

  56. }

  57. 

  58. const char *swscale_configuration(void)

  59. {

  60.     return LIBAV_CONFIGURATION;

  61. }

  62. 

  63. const char *swscale_license(void)

  64. {

  65. #define LICENSE_PREFIX "libswscale license: "

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

  67. }

  68. 

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

  70. 

  71. typedef struct FormatEntry {

  72.     int is_supported_in, is_supported_out;

  73. } FormatEntry;

  74. 

  75. static const FormatEntry format_entries[PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  152. };

  153. 

  154. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  155. {

  156.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  157.            format_entries[pix_fmt].is_supported_in : 0;

  158. }

  159. 

  160. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  161. {

  162.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  163.            format_entries[pix_fmt].is_supported_out : 0;

  164. }

  165. 

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

  167. 

  168. const char *sws_format_name(enum PixelFormat format)

  169. {

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

  171.         return av_pix_fmt_descriptors[format].name;

  172.     else

  173.         return "Unknown format";

  174. }

  175. 

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

  177.                              double dist)

  178. {

  179.     if (dist <= 1.0)

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

  181.     else

  182.         return getSplineCoeff(0.0,

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

  184.                                c + 3.0 * d,

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

  186.                               dist - 1.0);

  187. }

  188. 

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

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

  191.                       int filterAlign, int one, int flags, int cpu_flags,

  192.                       SwsVector *srcFilter, SwsVector *dstFilter,

  193.                       double param[2], int is_horizontal)

  194. {

  195.     int i;

  196.     int filterSize;

  197.     int filter2Size;

  198.     int minFilterSize;

  199.     int64_t *filter    = NULL;

  200.     int64_t *filter2   = NULL;

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

  202.     int ret            = -1;

  203. 

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

  205. 

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

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

  208. 

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

  210.         int i;

  211.         filterSize = 1;

  212.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  214. 

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

  216.             filter[i * filterSize] = fone;

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

  218.         }

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

  220.         int i;

  221.         int xDstInSrc;

  222.         filterSize = 1;

  223.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  225. 

  226.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  229. 

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

  231.             filter[i]       = fone;

  232.             xDstInSrc      += xInc;

  233.         }

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

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

  236.         int i;

  237.         int xDstInSrc;

  238.         filterSize = 2;

  239.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  241. 

  242.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  245.             int j;

  246. 

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

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

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

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

  251.                                 (fone >> 16);

  252.                 if (coeff < 0)

  253.                     coeff = 0;

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

  255.                 xx++;

  256.             }

  257.             xDstInSrc += xInc;

  258.         }

  259.     } else {

  260.         int64_t xDstInSrc;

  261.         int sizeFactor;

  262. 

  263.         if (flags & SWS_BICUBIC)

  264.             sizeFactor = 4;

  265.         else if (flags & SWS_X)

  266.             sizeFactor = 8;

  267.         else if (flags & SWS_AREA)

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

  269.         else if (flags & SWS_GAUSS)

  270.             sizeFactor = 8;     // infinite ;)

  271.         else if (flags & SWS_LANCZOS)

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

  273.         else if (flags & SWS_SINC)

  274.             sizeFactor = 20;    // infinite ;)

  275.         else if (flags & SWS_SPLINE)

  276.             sizeFactor = 20;    // infinite ;)

  277.         else if (flags & SWS_BILINEAR)

  278.             sizeFactor = 2;

  279.         else {

  280.             sizeFactor = 0;     // GCC warning killer

  281.             assert(0);

  282.         }

  283. 

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

  285.             filterSize = 1 + sizeFactor;    // upscale

  286.         else

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

  288. 

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

  290.         filterSize = FFMAX(filterSize, 1);

  291. 

  292.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  294. 

  295.         xDstInSrc = xInc - 0x10000;

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

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

  298.             int j;

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

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

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

  302.                 double floatd;

  303.                 int64_t coeff;

  304. 

  305.                 if (xInc > 1 << 16)

  306.                     d = d * dstW / srcW;

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

  308. 

  309.                 if (flags & SWS_BICUBIC) {

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

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

  312. 

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

  314.                         coeff = 0.0;

  315.                     } else {

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

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

  318. 

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

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

  321.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  322.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  323.                         else

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

  325.                                       (6 * B + 30 * C) * dd  +

  326.                                     (-12 * B - 48 * C) * d   +

  327.                                       (8 * B + 24 * C) * (1 << 30);

  328.                     }

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

  330.                 }

  331. #if 0

  332.                 else if (flags & SWS_X) {

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

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

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

  336.                 }

  337. #endif

  338.                 else if (flags & SWS_X) {

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

  340.                     double c;

  341. 

  342.                     if (floatd < 1.0)

  343.                         c = cos(floatd * M_PI);

  344.                     else

  345.                         c = -1.0;

  346.                     if (c < 0.0)

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

  348.                     else

  349.                         c = pow(c, A);

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

  351.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  357.                     else

  358.                         coeff = 0.0;

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

  360.                 } else if (flags & SWS_GAUSS) {

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

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

  363.                 } else if (flags & SWS_SINC) {

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

  365.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  369.                     if (floatd > p)

  370.                         coeff = 0;

  371.                 } else if (flags & SWS_BILINEAR) {

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

  373.                     if (coeff < 0)

  374.                         coeff = 0;

  375.                     coeff *= fone >> 30;

  376.                 } else if (flags & SWS_SPLINE) {

  377.                     double p = -2.196152422706632;

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

  379.                 } else {

  380.                     coeff = 0.0; // GCC warning killer

  381.                     assert(0);

  382.                 }

  383. 

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

  385.                 xx++;

  386.             }

  387.             xDstInSrc += 2 * xInc;

  388.         }

  389.     }

  390. 

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

  392.      * av_free(filter);

  393.      */

  394.     assert(filterSize > 0);

  395.     filter2Size = filterSize;

  396.     if (srcFilter)

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

  398.     if (dstFilter)

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

  400.     assert(filter2Size > 0);

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

  402. 

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

  404.         int j, k;

  405. 

  406.         if (srcFilter) {

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

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

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

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

  411.             }

  412.         } else {

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

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

  415.         }

  416.         // FIXME dstFilter

  417. 

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

  419.     }

  420.     av_freep(&filter);

  421. 

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

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

  424.     minFilterSize = 0;

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

  426.         int min = filter2Size;

  427.         int j;

  428.         int64_t cutOff = 0.0;

  429. 

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

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

  432.             int k;

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

  434. 

  435.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  436.                 break;

  437. 

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

  439.              * filter otherwise */

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

  441.                 break;

  442. 

  443.             // move filter coefficients left

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

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

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

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

  448.         }

  449. 

  450.         cutOff = 0;

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

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

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

  454. 

  455.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  456.                 break;

  457.             min--;

  458.         }

  459. 

  460.         if (min > minFilterSize)

  461.             minFilterSize = min;

  462.     }

  463. 

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

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

  466.         if (minFilterSize < 5)

  467.             filterAlign = 4;

  468. 

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

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

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

  472.         if (minFilterSize < 3)

  473.             filterAlign = 1;

  474.     }

  475. 

  476.     if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {

  477.         // special case for unscaled vertical filtering

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

  479.             filterAlign = 1;

  480.     }

  481. 

  482.     assert(minFilterSize > 0);

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

  484.     assert(filterSize > 0);

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

  486.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  488.         goto fail;

  489.     *outFilterSize = filterSize;

  490. 

  491.     if (flags & SWS_PRINT_INFO)

  492.         av_log(NULL, AV_LOG_VERBOSE,

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

  494.                filter2Size, filterSize);

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

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

  497.         int j;

  498. 

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

  500.             if (j >= filter2Size)

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

  502.             else

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

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

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

  506.         }

  507.     }

  508. 

  509.     // FIXME try to align filterPos if possible

  510. 

  511.     // fix borders

  512.     if (is_horizontal) {

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

  514.             int j;

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

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

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

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

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

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

  521.                 }

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

  523.             }

  524. 

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

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

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

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

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

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

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

  532.                 }

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

  534.             }

  535.         }

  536.     }

  537. 

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

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

  540.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  542. 

  543.     /* normalize & store in outFilter */

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

  545.         int j;

  546.         int64_t error = 0;

  547.         int64_t sum   = 0;

  548. 

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

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

  551.         }

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

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

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

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

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

  557.             error                                  = v - intV * sum;

  558.         }

  559.     }

  560. 

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

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

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

  564.                                                       * read over the end */

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

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

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

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

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

  570.     }

  571. 

  572.     ret = 0;

  573. 

  574. fail:

  575.     av_free(filter);

  576.     av_free(filter2);

  577.     return ret;

  578. }

  579. 

  580. #if HAVE_MMXEXT_INLINE

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

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

  583. {

  584.     uint8_t *fragmentA;

  585.     x86_reg imm8OfPShufW1A;

  586.     x86_reg imm8OfPShufW2A;

  587.     x86_reg fragmentLengthA;

  588.     uint8_t *fragmentB;

  589.     x86_reg imm8OfPShufW1B;

  590.     x86_reg imm8OfPShufW2B;

  591.     x86_reg fragmentLengthB;

  592.     int fragmentPos;

  593. 

  594.     int xpos, i;

  595. 

  596.     // create an optimized horizontal scaling routine

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

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

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

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

  601.      */

  602. 

  603.     // code fragment

  604. 

  605.     __asm__ volatile (

  606.         "jmp                         9f                 \n\t"

  607.         // Begin

  608.         "0:                                             \n\t"

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

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

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

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

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

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

  615.         "1:                                             \n\t"

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

  617.         "2:                                             \n\t"

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

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

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

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

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

  623. 

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

  625. 

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

  627.         // End

  628.         "9:                                             \n\t"

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

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

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

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

  633.         "dec                         %1                 \n\t"

  634.         "dec                         %2                 \n\t"

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

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

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

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

  639. 

  640. 

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

  642.           "=r" (fragmentLengthA)

  643.         );

  644. 

  645.     __asm__ volatile (

  646.         "jmp                         9f                 \n\t"

  647.         // Begin

  648.         "0:                                             \n\t"

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

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

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

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

  653.         "1:                                             \n\t"

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

  655.         "2:                                             \n\t"

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

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

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

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

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

  661. 

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

  663. 

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

  665.         // End

  666.         "9:                                             \n\t"

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

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

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

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

  671.         "dec                         %1                 \n\t"

  672.         "dec                         %2                 \n\t"

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

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

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

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

  677. 

  678. 

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

  680.           "=r" (fragmentLengthB)

  681.         );

  682. 

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

  684.     fragmentPos = 0;

  685. 

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

  687.         int xx = xpos >> 16;

  688. 

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

  690.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  700.             int shift              = 0;

  701. 

  702.             if (filterCode) {

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

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

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

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

  707.                 filterPos[i / 2] = xx;

  708. 

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

  710. 

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

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

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

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

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

  716.                                                                (c << 4) |

  717.                                                                (d << 6);

  718. 

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

  720.                     shift = maxShift;               // avoid overread

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

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

  723. 

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

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

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

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

  728.                 }

  729.             }

  730. 

  731.             fragmentPos += fragmentLength;

  732. 

  733.             if (filterCode)

  734.                 filterCode[fragmentPos] = RET;

  735.         }

  736.         xpos += xInc;

  737.     }

  738.     if (filterCode)

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

  740. 

  741.     return fragmentPos + 1;

  742. }

  743. #endif /* HAVE_MMXEXT_INLINE */

  744. 

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

  746. {

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

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

  749. }

  750. 

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

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

  753.                              int brightness, int contrast, int saturation)

  754. {

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

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

  757. 

  758.     c->brightness = brightness;

  759.     c->contrast   = contrast;

  760.     c->saturation = saturation;

  761.     c->srcRange   = srcRange;

  762.     c->dstRange   = dstRange;

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

  764.         return -1;

  765. 

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

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

  768. 

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

  770.                              contrast, saturation);

  771.     // FIXME factorize

  772. 

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

  774.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  775.                                        contrast, saturation);

  776.     return 0;

  777. }

  778. 

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

  780.                              int *srcRange, int **table, int *dstRange,

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

  782. {

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

  784.         return -1;

  785. 

  786.     *inv_table  = c->srcColorspaceTable;

  787.     *table      = c->dstColorspaceTable;

  788.     *srcRange   = c->srcRange;

  789.     *dstRange   = c->dstRange;

  790.     *brightness = c->brightness;

  791.     *contrast   = c->contrast;

  792.     *saturation = c->saturation;

  793. 

  794.     return 0;

  795. }

  796. 

  797. static int handle_jpeg(enum PixelFormat *format)

  798. {

  799.     switch (*format) {

  800.     case PIX_FMT_YUVJ420P:

  801.         *format = PIX_FMT_YUV420P;

  802.         return 1;

  803.     case PIX_FMT_YUVJ422P:

  804.         *format = PIX_FMT_YUV422P;

  805.         return 1;

  806.     case PIX_FMT_YUVJ444P:

  807.         *format = PIX_FMT_YUV444P;

  808.         return 1;

  809.     case PIX_FMT_YUVJ440P:

  810.         *format = PIX_FMT_YUV440P;

  811.         return 1;

  812.     default:

  813.         return 0;

  814.     }

  815. }

  816. 

  817. SwsContext *sws_alloc_context(void)

  818. {

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

  820. 

  821.     c->av_class = &sws_context_class;

  822.     av_opt_set_defaults(c);

  823. 

  824.     return c;

  825. }

  826. 

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

  828.                              SwsFilter *dstFilter)

  829. {

  830.     int i;

  831.     int usesVFilter, usesHFilter;

  832.     int unscaled;

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

  834.     int srcW              = c->srcW;

  835.     int srcH              = c->srcH;

  836.     int dstW              = c->dstW;

  837.     int dstH              = c->dstH;

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

  839.     int dst_stride_px     = dst_stride >> 1;

  840.     int flags, cpu_flags;

  841.     enum PixelFormat srcFormat = c->srcFormat;

  842.     enum PixelFormat dstFormat = c->dstFormat;

  843. 

  844.     cpu_flags = av_get_cpu_flags();

  845.     flags     = c->flags;

  846.     emms_c();

  847.     if (!rgb15to16)

  848.         sws_rgb2rgb_init();

  849. 

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

  851. 

  852.     if (!sws_isSupportedInput(srcFormat)) {

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

  854.                sws_format_name(srcFormat));

  855.         return AVERROR(EINVAL);

  856.     }

  857.     if (!sws_isSupportedOutput(dstFormat)) {

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

  859.                sws_format_name(dstFormat));

  860.         return AVERROR(EINVAL);

  861.     }

  862. 

  863.     i = flags & (SWS_POINT         |

  864.                  SWS_AREA          |

  865.                  SWS_BILINEAR      |

  866.                  SWS_FAST_BILINEAR |

  867.                  SWS_BICUBIC       |

  868.                  SWS_X             |

  869.                  SWS_GAUSS         |

  870.                  SWS_LANCZOS       |

  871.                  SWS_SINC          |

  872.                  SWS_SPLINE        |

  873.                  SWS_BICUBLIN);

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

  875.         av_log(c, AV_LOG_ERROR,

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

  877.         return AVERROR(EINVAL);

  878.     }

  879.     /* sanity check */

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

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

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

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

  884.                srcW, srcH, dstW, dstH);

  885.         return AVERROR(EINVAL);

  886.     }

  887. 

  888.     if (!dstFilter)

  889.         dstFilter = &dummyFilter;

  890.     if (!srcFilter)

  891.         srcFilter = &dummyFilter;

  892. 

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

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

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

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

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

  898. 

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

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

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

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

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

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

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

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

  907. 

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

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

  910. 

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

  912.      * chroma interpolation */

  913.     if (flags & SWS_FULL_CHR_H_INT &&

  914.         isAnyRGB(dstFormat)        &&

  915.         dstFormat != PIX_FMT_RGBA  &&

  916.         dstFormat != PIX_FMT_ARGB  &&

  917.         dstFormat != PIX_FMT_BGRA  &&

  918.         dstFormat != PIX_FMT_ABGR  &&

  919.         dstFormat != PIX_FMT_RGB24 &&

  920.         dstFormat != PIX_FMT_BGR24) {

  921.         av_log(c, AV_LOG_ERROR,

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

  923.                sws_format_name(dstFormat));

  924.         flags   &= ~SWS_FULL_CHR_H_INT;

  925.         c->flags = flags;

  926.     }

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

  928.         c->chrDstHSubSample = 1;

  929. 

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

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

  932.                            SWS_SRC_V_CHR_DROP_SHIFT;

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

  934. 

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

  936.      * wants full chroma */

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

  938.         srcFormat != PIX_FMT_RGB8 && srcFormat != PIX_FMT_BGR8 &&

  939.         srcFormat != PIX_FMT_RGB4 && srcFormat != PIX_FMT_BGR4 &&

  940.         srcFormat != PIX_FMT_RGB4_BYTE && srcFormat != PIX_FMT_BGR4_BYTE &&

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

  942.          (flags & SWS_FAST_BILINEAR)))

  943.         c->chrSrcHSubSample = 1;

  944. 

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

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

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

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

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

  950. 

  951.     /* unscaled special cases */

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

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

  954.         ff_get_unscaled_swscale(c);

  955. 

  956.         if (c->swScale) {

  957.             if (flags & SWS_PRINT_INFO)

  958.                 av_log(c, AV_LOG_INFO,

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

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

  961.             return 0;

  962.         }

  963.     }

  964. 

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

  966.     if (c->srcBpc < 8)

  967.         c->srcBpc = 8;

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

  969.     if (c->dstBpc < 8)

  970.         c->dstBpc = 8;

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

  972.         dst_stride <<= 1;

  973.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

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

  975.                      fail);

  976.     if (HAVE_MMXEXT && HAVE_INLINE_ASM && cpu_flags & AV_CPU_FLAG_MMXEXT &&

  977.         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 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {

  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 =

  1054.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1055.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1056.                 1;

  1057. 

  1058.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1059.                            &c->hLumFilterSize, c->lumXInc,

  1060.                            srcW, dstW, filterAlign, 1 << 14,

  1061.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1062.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1063.                            c->param, 1) < 0)

  1064.                 goto fail;

  1065.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1066.                            &c->hChrFilterSize, c->chrXInc,

  1067.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1068.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1069.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1070.                            c->param, 1) < 0)

  1071.                 goto fail;

  1072.         }

  1073.     } // initialize horizontal stuff

  1074. 

  1075.     /* precalculate vertical scaler filter coefficients */

  1076.     {

  1077.         const int filterAlign =

  1078.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1079.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1080.             1;

  1081. 

  1082.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1083.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1084.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1085.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1086.                        c->param, 0) < 0)

  1087.             goto fail;

  1088.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1089.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1090.                        filterAlign, (1 << 12),

  1091.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1092.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1093.                        c->param, 0) < 0)

  1094.             goto fail;

  1095. 

  1096. #if HAVE_ALTIVEC

  1097.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1098.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1099. 

  1100.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1101.             int j;

  1102.             short *p = (short *)&c->vYCoeffsBank[i];

  1103.             for (j = 0; j < 8; j++)

  1104.                 p[j] = c->vLumFilter[i];

  1105.         }

  1106. 

  1107.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1108.             int j;

  1109.             short *p = (short *)&c->vCCoeffsBank[i];

  1110.             for (j = 0; j < 8; j++)

  1111.                 p[j] = c->vChrFilter[i];

  1112.         }

  1113. #endif

  1114.     }

  1115. 

  1116.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1117.     c->vLumBufSize = c->vLumFilterSize;

  1118.     c->vChrBufSize = c->vChrFilterSize;

  1119.     for (i = 0; i < dstH; i++) {

  1120.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1121.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1122.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1123.                                << c->chrSrcVSubSample));

  1124. 

  1125.         nextSlice >>= c->chrSrcVSubSample;

  1126.         nextSlice <<= c->chrSrcVSubSample;

  1127.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1128.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1129.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1130.             (nextSlice >> c->chrSrcVSubSample))

  1131.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1132.                              c->vChrFilterPos[chrI];

  1133.     }

  1134. 

  1135.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1136.      * need to allocate several megabytes to handle all possible cases) */

  1137.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1138.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1139.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1140.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1141.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1142.     /* Note we need at least one pixel more at the end because of the MMX code

  1143.      * (just in case someone wants to replace the 4000/8000). */

  1144.     /* align at 16 bytes for AltiVec */

  1145.     for (i = 0; i < c->vLumBufSize; i++) {

  1146.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1147.                           dst_stride + 16, fail);

  1148.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1149.     }

  1150.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1151.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc & ~7);

  1152.     c->uv_off_byte = dst_stride + 16;

  1153.     for (i = 0; i < c->vChrBufSize; i++) {

  1154.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1155.                          dst_stride * 2 + 32, fail);

  1156.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1157.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1158.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1159.     }

  1160.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1161.         for (i = 0; i < c->vLumBufSize; i++) {

  1162.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1163.                               dst_stride + 16, fail);

  1164.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1165.         }

  1166. 

  1167.     // try to avoid drawing green stuff between the right end and the stride end

  1168.     for (i = 0; i < c->vChrBufSize; i++)

  1169.         memset(c->chrUPixBuf[i], 64, dst_stride * 2 + 1);

  1170. 

  1171.     assert(c->chrDstH <= dstH);

  1172. 

  1173.     if (flags & SWS_PRINT_INFO) {

  1174.         if (flags & SWS_FAST_BILINEAR)

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

  1176.         else if (flags & SWS_BILINEAR)

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

  1178.         else if (flags & SWS_BICUBIC)

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

  1180.         else if (flags & SWS_X)

  1181.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1182.         else if (flags & SWS_POINT)

  1183.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1184.         else if (flags & SWS_AREA)

  1185.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1186.         else if (flags & SWS_BICUBLIN)

  1187.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1188.         else if (flags & SWS_GAUSS)

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

  1190.         else if (flags & SWS_SINC)

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

  1192.         else if (flags & SWS_LANCZOS)

  1193.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1194.         else if (flags & SWS_SPLINE)

  1195.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1196.         else

  1197.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1198. 

  1199.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1200.                sws_format_name(srcFormat),

  1201. #ifdef DITHER1XBPP

  1202.                dstFormat == PIX_FMT_BGR555   || dstFormat == PIX_FMT_BGR565   ||

  1203.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1204.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ?

  1205.                                                              "dithered " : "",

  1206. #else

  1207.                "",

  1208. #endif

  1209.                sws_format_name(dstFormat));

  1210. 

  1211.         if (HAVE_MMXEXT && cpu_flags & AV_CPU_FLAG_MMXEXT)

  1212.             av_log(c, AV_LOG_INFO, "using MMX2\n");

  1213.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)

  1214.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1215.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)

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

  1217.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

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

  1219.         else

  1220.             av_log(c, AV_LOG_INFO, "using C\n");

  1221. 

  1222.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1223.         av_log(c, AV_LOG_DEBUG,

  1224.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1225.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1226.         av_log(c, AV_LOG_DEBUG,

  1227.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1228.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1229.                c->chrXInc, c->chrYInc);

  1230.     }

  1231. 

  1232.     c->swScale = ff_getSwsFunc(c);

  1233.     return 0;

  1234. fail: // FIXME replace things by appropriate error codes

  1235.     return -1;

  1236. }

  1237. 

  1238. #if FF_API_SWS_GETCONTEXT

  1239. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1240.                            int dstW, int dstH, enum PixelFormat dstFormat,

  1241.                            int flags, SwsFilter *srcFilter,

  1242.                            SwsFilter *dstFilter, const double *param)

  1243. {

  1244.     SwsContext *c;

  1245. 

  1246.     if (!(c = sws_alloc_context()))

  1247.         return NULL;

  1248. 

  1249.     c->flags     = flags;

  1250.     c->srcW      = srcW;

  1251.     c->srcH      = srcH;

  1252.     c->dstW      = dstW;

  1253.     c->dstH      = dstH;

  1254.     c->srcRange  = handle_jpeg(&srcFormat);

  1255.     c->dstRange  = handle_jpeg(&dstFormat);

  1256.     c->srcFormat = srcFormat;

  1257.     c->dstFormat = dstFormat;

  1258. 

  1259.     if (param) {

  1260.         c->param[0] = param[0];

  1261.         c->param[1] = param[1];

  1262.     }

  1263.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1264.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1265.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1266. 

  1267.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1268.         sws_freeContext(c);

  1269.         return NULL;

  1270.     }

  1271. 

  1272.     return c;

  1273. }

  1274. #endif

  1275. 

  1276. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1277.                                 float lumaSharpen, float chromaSharpen,

  1278.                                 float chromaHShift, float chromaVShift,

  1279.                                 int verbose)

  1280. {

  1281.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1282.     if (!filter)

  1283.         return NULL;

  1284. 

  1285.     if (lumaGBlur != 0.0) {

  1286.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1287.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1288.     } else {

  1289.         filter->lumH = sws_getIdentityVec();

  1290.         filter->lumV = sws_getIdentityVec();

  1291.     }

  1292. 

  1293.     if (chromaGBlur != 0.0) {

  1294.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1295.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1296.     } else {

  1297.         filter->chrH = sws_getIdentityVec();

  1298.         filter->chrV = sws_getIdentityVec();

  1299.     }

  1300. 

  1301.     if (chromaSharpen != 0.0) {

  1302.         SwsVector *id = sws_getIdentityVec();

  1303.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1304.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1305.         sws_addVec(filter->chrH, id);

  1306.         sws_addVec(filter->chrV, id);

  1307.         sws_freeVec(id);

  1308.     }

  1309. 

  1310.     if (lumaSharpen != 0.0) {

  1311.         SwsVector *id = sws_getIdentityVec();

  1312.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1313.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1314.         sws_addVec(filter->lumH, id);

  1315.         sws_addVec(filter->lumV, id);

  1316.         sws_freeVec(id);

  1317.     }

  1318. 

  1319.     if (chromaHShift != 0.0)

  1320.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1321. 

  1322.     if (chromaVShift != 0.0)

  1323.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1324. 

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

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

  1327.     sws_normalizeVec(filter->lumH, 1.0);

  1328.     sws_normalizeVec(filter->lumV, 1.0);

  1329. 

  1330.     if (verbose)

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

  1332.     if (verbose)

  1333.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1334. 

  1335.     return filter;

  1336. }

  1337. 

  1338. SwsVector *sws_allocVec(int length)

  1339. {

  1340.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1341.     if (!vec)

  1342.         return NULL;

  1343.     vec->length = length;

  1344.     vec->coeff  = av_malloc(sizeof(double) * length);

  1345.     if (!vec->coeff)

  1346.         av_freep(&vec);

  1347.     return vec;

  1348. }

  1349. 

  1350. SwsVector *sws_getGaussianVec(double variance, double quality)

  1351. {

  1352.     const int length = (int)(variance * quality + 0.5) | 1;

  1353.     int i;

  1354.     double middle  = (length - 1) * 0.5;

  1355.     SwsVector *vec = sws_allocVec(length);

  1356. 

  1357.     if (!vec)

  1358.         return NULL;

  1359. 

  1360.     for (i = 0; i < length; i++) {

  1361.         double dist = i - middle;

  1362.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1363.                         sqrt(2 * variance * M_PI);

  1364.     }

  1365. 

  1366.     sws_normalizeVec(vec, 1.0);

  1367. 

  1368.     return vec;

  1369. }

  1370. 

  1371. SwsVector *sws_getConstVec(double c, int length)

  1372. {

  1373.     int i;

  1374.     SwsVector *vec = sws_allocVec(length);

  1375. 

  1376.     if (!vec)

  1377.         return NULL;

  1378. 

  1379.     for (i = 0; i < length; i++)

  1380.         vec->coeff[i] = c;

  1381. 

  1382.     return vec;

  1383. }

  1384. 

  1385. SwsVector *sws_getIdentityVec(void)

  1386. {

  1387.     return sws_getConstVec(1.0, 1);

  1388. }

  1389. 

  1390. static double sws_dcVec(SwsVector *a)

  1391. {

  1392.     int i;

  1393.     double sum = 0;

  1394. 

  1395.     for (i = 0; i < a->length; i++)

  1396.         sum += a->coeff[i];

  1397. 

  1398.     return sum;

  1399. }

  1400. 

  1401. void sws_scaleVec(SwsVector *a, double scalar)

  1402. {

  1403.     int i;

  1404. 

  1405.     for (i = 0; i < a->length; i++)

  1406.         a->coeff[i] *= scalar;

  1407. }

  1408. 

  1409. void sws_normalizeVec(SwsVector *a, double height)

  1410. {

  1411.     sws_scaleVec(a, height / sws_dcVec(a));

  1412. }

  1413. 

  1414. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1415. {

  1416.     int length = a->length + b->length - 1;

  1417.     int i, j;

  1418.     SwsVector *vec = sws_getConstVec(0.0, length);

  1419. 

  1420.     if (!vec)

  1421.         return NULL;

  1422. 

  1423.     for (i = 0; i < a->length; i++) {

  1424.         for (j = 0; j < b->length; j++) {

  1425.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1426.         }

  1427.     }

  1428. 

  1429.     return vec;

  1430. }

  1431. 

  1432. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1433. {

  1434.     int length = FFMAX(a->length, b->length);

  1435.     int i;

  1436.     SwsVector *vec = sws_getConstVec(0.0, length);

  1437. 

  1438.     if (!vec)

  1439.         return NULL;

  1440. 

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

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

  1443.     for (i = 0; i < b->length; i++)

  1444.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1445. 

  1446.     return vec;

  1447. }

  1448. 

  1449. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1450. {

  1451.     int length = FFMAX(a->length, b->length);

  1452.     int i;

  1453.     SwsVector *vec = sws_getConstVec(0.0, length);

  1454. 

  1455.     if (!vec)

  1456.         return NULL;

  1457. 

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

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

  1460.     for (i = 0; i < b->length; i++)

  1461.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1462. 

  1463.     return vec;

  1464. }

  1465. 

  1466. /* shift left / or right if "shift" is negative */

  1467. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1468. {

  1469.     int length = a->length + FFABS(shift) * 2;

  1470.     int i;

  1471.     SwsVector *vec = sws_getConstVec(0.0, length);

  1472. 

  1473.     if (!vec)

  1474.         return NULL;

  1475. 

  1476.     for (i = 0; i < a->length; i++) {

  1477.         vec->coeff[i + (length    - 1) / 2 -

  1478.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1479.     }

  1480. 

  1481.     return vec;

  1482. }

  1483. 

  1484. void sws_shiftVec(SwsVector *a, int shift)

  1485. {

  1486.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1487.     av_free(a->coeff);

  1488.     a->coeff  = shifted->coeff;

  1489.     a->length = shifted->length;

  1490.     av_free(shifted);

  1491. }

  1492. 

  1493. void sws_addVec(SwsVector *a, SwsVector *b)

  1494. {

  1495.     SwsVector *sum = sws_sumVec(a, b);

  1496.     av_free(a->coeff);

  1497.     a->coeff  = sum->coeff;

  1498.     a->length = sum->length;

  1499.     av_free(sum);

  1500. }

  1501. 

  1502. void sws_subVec(SwsVector *a, SwsVector *b)

  1503. {

  1504.     SwsVector *diff = sws_diffVec(a, b);

  1505.     av_free(a->coeff);

  1506.     a->coeff  = diff->coeff;

  1507.     a->length = diff->length;

  1508.     av_free(diff);

  1509. }

  1510. 

  1511. void sws_convVec(SwsVector *a, SwsVector *b)

  1512. {

  1513.     SwsVector *conv = sws_getConvVec(a, b);

  1514.     av_free(a->coeff);

  1515.     a->coeff  = conv->coeff;

  1516.     a->length = conv->length;

  1517.     av_free(conv);

  1518. }

  1519. 

  1520. SwsVector *sws_cloneVec(SwsVector *a)

  1521. {

  1522.     int i;

  1523.     SwsVector *vec = sws_allocVec(a->length);

  1524. 

  1525.     if (!vec)

  1526.         return NULL;

  1527. 

  1528.     for (i = 0; i < a->length; i++)

  1529.         vec->coeff[i] = a->coeff[i];

  1530. 

  1531.     return vec;

  1532. }

  1533. 

  1534. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1535. {

  1536.     int i;

  1537.     double max = 0;

  1538.     double min = 0;

  1539.     double range;

  1540. 

  1541.     for (i = 0; i < a->length; i++)

  1542.         if (a->coeff[i] > max)

  1543.             max = a->coeff[i];

  1544. 

  1545.     for (i = 0; i < a->length; i++)

  1546.         if (a->coeff[i] < min)

  1547.             min = a->coeff[i];

  1548. 

  1549.     range = max - min;

  1550. 

  1551.     for (i = 0; i < a->length; i++) {

  1552.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1553.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1554.         for (; x > 0; x--)

  1555.             av_log(log_ctx, log_level, " ");

  1556.         av_log(log_ctx, log_level, "|\n");

  1557.     }

  1558. }

  1559. 

  1560. void sws_freeVec(SwsVector *a)

  1561. {

  1562.     if (!a)

  1563.         return;

  1564.     av_freep(&a->coeff);

  1565.     a->length = 0;

  1566.     av_free(a);

  1567. }

  1568. 

  1569. void sws_freeFilter(SwsFilter *filter)

  1570. {

  1571.     if (!filter)

  1572.         return;

  1573. 

  1574.     if (filter->lumH)

  1575.         sws_freeVec(filter->lumH);

  1576.     if (filter->lumV)

  1577.         sws_freeVec(filter->lumV);

  1578.     if (filter->chrH)

  1579.         sws_freeVec(filter->chrH);

  1580.     if (filter->chrV)

  1581.         sws_freeVec(filter->chrV);

  1582.     av_free(filter);

  1583. }

  1584. 

  1585. void sws_freeContext(SwsContext *c)

  1586. {

  1587.     int i;

  1588.     if (!c)

  1589.         return;

  1590. 

  1591.     if (c->lumPixBuf) {

  1592.         for (i = 0; i < c->vLumBufSize; i++)

  1593.             av_freep(&c->lumPixBuf[i]);

  1594.         av_freep(&c->lumPixBuf);

  1595.     }

  1596. 

  1597.     if (c->chrUPixBuf) {

  1598.         for (i = 0; i < c->vChrBufSize; i++)

  1599.             av_freep(&c->chrUPixBuf[i]);

  1600.         av_freep(&c->chrUPixBuf);

  1601.         av_freep(&c->chrVPixBuf);

  1602.     }

  1603. 

  1604.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1605.         for (i = 0; i < c->vLumBufSize; i++)

  1606.             av_freep(&c->alpPixBuf[i]);

  1607.         av_freep(&c->alpPixBuf);

  1608.     }

  1609. 

  1610.     av_freep(&c->vLumFilter);

  1611.     av_freep(&c->vChrFilter);

  1612.     av_freep(&c->hLumFilter);

  1613.     av_freep(&c->hChrFilter);

  1614. #if HAVE_ALTIVEC

  1615.     av_freep(&c->vYCoeffsBank);

  1616.     av_freep(&c->vCCoeffsBank);

  1617. #endif

  1618. 

  1619.     av_freep(&c->vLumFilterPos);

  1620.     av_freep(&c->vChrFilterPos);

  1621.     av_freep(&c->hLumFilterPos);

  1622.     av_freep(&c->hChrFilterPos);

  1623. 

  1624. #if HAVE_MMX_INLINE

  1625. #ifdef MAP_ANONYMOUS

  1626.     if (c->lumMmx2FilterCode)

  1627.         munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1628.     if (c->chrMmx2FilterCode)

  1629.         munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1630. #elif HAVE_VIRTUALALLOC

  1631.     if (c->lumMmx2FilterCode)

  1632.         VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1633.     if (c->chrMmx2FilterCode)

  1634.         VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1635. #else

  1636.     av_free(c->lumMmx2FilterCode);

  1637.     av_free(c->chrMmx2FilterCode);

  1638. #endif

  1639.     c->lumMmx2FilterCode = NULL;

  1640.     c->chrMmx2FilterCode = NULL;

  1641. #endif /* HAVE_MMX_INLINE */

  1642. 

  1643.     av_freep(&c->yuvTable);

  1644.     av_free(c->formatConvBuffer);

  1645. 

  1646.     av_free(c);

  1647. }

  1648. 

  1649. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1650.                                         int srcH, enum PixelFormat srcFormat,

  1651.                                         int dstW, int dstH,

  1652.                                         enum PixelFormat dstFormat, int flags,

  1653.                                         SwsFilter *srcFilter,

  1654.                                         SwsFilter *dstFilter,

  1655.                                         const double *param)

  1656. {

  1657.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1658.                                              SWS_PARAM_DEFAULT };

  1659. 

  1660.     if (!param)

  1661.         param = default_param;

  1662. 

  1663.     if (context &&

  1664.         (context->srcW      != srcW      ||

  1665.          context->srcH      != srcH      ||

  1666.          context->srcFormat != srcFormat ||

  1667.          context->dstW      != dstW      ||

  1668.          context->dstH      != dstH      ||

  1669.          context->dstFormat != dstFormat ||

  1670.          context->flags     != flags     ||

  1671.          context->param[0]  != param[0]  ||

  1672.          context->param[1]  != param[1])) {

  1673.         sws_freeContext(context);

  1674.         context = NULL;

  1675.     }

  1676. 

  1677.     if (!context) {

  1678.         if (!(context = sws_alloc_context()))

  1679.             return NULL;

  1680.         context->srcW      = srcW;

  1681.         context->srcH      = srcH;

  1682.         context->srcRange  = handle_jpeg(&srcFormat);

  1683.         context->srcFormat = srcFormat;

  1684.         context->dstW      = dstW;

  1685.         context->dstH      = dstH;

  1686.         context->dstRange  = handle_jpeg(&dstFormat);

  1687.         context->dstFormat = dstFormat;

  1688.         context->flags     = flags;

  1689.         context->param[0]  = param[0];

  1690.         context->param[1]  = param[1];

  1691.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1692.                                  context->srcRange,

  1693.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1694.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1695.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1696.             sws_freeContext(context);

  1697.             return NULL;

  1698.         }

  1699.     }

  1700.     return context;

  1701. }