swscale: remove alp/chr/lumSrcOffset.
[ffmpeg.git] / libswscale / swscale.c
index bb274e6..d97c4e1 100644 (file)
@@ -1,24 +1,21 @@
 /*
  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
  *
- * FFmpeg is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
  *
- * You should have received a copy of the GNU General Public License
- * along with FFmpeg; if not, write to the Free Software
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- *
- * the C code (not assembly, mmx, ...) of this file can be used
- * under the LGPL license too
  */
 
 /*
@@ -27,7 +24,7 @@
   {BGR,RGB}{1,4,8,15,16} support dithering
 
   unscaled special converters (YV12=I420=IYUV, Y800=Y8)
-  YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
+  YV12 -> {BGR,RGB}{1,4,8,12,15,16,24,32}
   x -> x
   YUV9 -> YV12
   YUV9/YV12 -> Y800
@@ -39,7 +36,7 @@
 
 /*
 tested special converters (most are tested actually, but I did not write it down ...)
- YV12 -> BGR16
+ YV12 -> BGR12/BGR16
  YV12 -> YV12
  BGR15 -> BGR16
  BGR16 -> BGR16
@@ -54,154 +51,24 @@ untested special converters
   BGR24 -> YV12
 */
 
-#define _SVID_SOURCE //needed for MAP_ANONYMOUS
 #include <inttypes.h>
 #include <string.h>
 #include <math.h>
 #include <stdio.h>
 #include "config.h"
 #include <assert.h>
-#if HAVE_SYS_MMAN_H
-#include <sys/mman.h>
-#if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
-#define MAP_ANONYMOUS MAP_ANON
-#endif
-#endif
-#if HAVE_VIRTUALALLOC
-#define WIN32_LEAN_AND_MEAN
-#include <windows.h>
-#endif
 #include "swscale.h"
 #include "swscale_internal.h"
 #include "rgb2rgb.h"
 #include "libavutil/intreadwrite.h"
-#include "libavutil/x86_cpu.h"
+#include "libavutil/cpu.h"
 #include "libavutil/avutil.h"
+#include "libavutil/mathematics.h"
 #include "libavutil/bswap.h"
 #include "libavutil/pixdesc.h"
 
-unsigned swscale_version(void)
-{
-    return LIBSWSCALE_VERSION_INT;
-}
-
-const char *swscale_configuration(void)
-{
-    return FFMPEG_CONFIGURATION;
-}
-
-const char *swscale_license(void)
-{
-#define LICENSE_PREFIX "libswscale license: "
-    return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
-}
-
-#undef MOVNTQ
-#undef PAVGB
-
-//#undef HAVE_MMX2
-//#define HAVE_AMD3DNOW
-//#undef HAVE_MMX
-//#undef ARCH_X86
 #define DITHER1XBPP
 
-#define FAST_BGR2YV12 // use 7 bit coefficients instead of 15 bit
-
-#define RET 0xC3 //near return opcode for x86
-
-#ifdef M_PI
-#define PI M_PI
-#else
-#define PI 3.14159265358979323846
-#endif
-
-#define isSupportedIn(x)    (       \
-           (x)==PIX_FMT_YUV420P     \
-        || (x)==PIX_FMT_YUVA420P    \
-        || (x)==PIX_FMT_YUYV422     \
-        || (x)==PIX_FMT_UYVY422     \
-        || (x)==PIX_FMT_RGB48BE     \
-        || (x)==PIX_FMT_RGB48LE     \
-        || (x)==PIX_FMT_RGB32       \
-        || (x)==PIX_FMT_RGB32_1     \
-        || (x)==PIX_FMT_BGR24       \
-        || (x)==PIX_FMT_BGR565      \
-        || (x)==PIX_FMT_BGR555      \
-        || (x)==PIX_FMT_BGR32       \
-        || (x)==PIX_FMT_BGR32_1     \
-        || (x)==PIX_FMT_RGB24       \
-        || (x)==PIX_FMT_RGB565      \
-        || (x)==PIX_FMT_RGB555      \
-        || (x)==PIX_FMT_GRAY8       \
-        || (x)==PIX_FMT_YUV410P     \
-        || (x)==PIX_FMT_YUV440P     \
-        || (x)==PIX_FMT_NV12        \
-        || (x)==PIX_FMT_NV21        \
-        || (x)==PIX_FMT_GRAY16BE    \
-        || (x)==PIX_FMT_GRAY16LE    \
-        || (x)==PIX_FMT_YUV444P     \
-        || (x)==PIX_FMT_YUV422P     \
-        || (x)==PIX_FMT_YUV411P     \
-        || (x)==PIX_FMT_PAL8        \
-        || (x)==PIX_FMT_BGR8        \
-        || (x)==PIX_FMT_RGB8        \
-        || (x)==PIX_FMT_BGR4_BYTE   \
-        || (x)==PIX_FMT_RGB4_BYTE   \
-        || (x)==PIX_FMT_YUV440P     \
-        || (x)==PIX_FMT_MONOWHITE   \
-        || (x)==PIX_FMT_MONOBLACK   \
-        || (x)==PIX_FMT_YUV420P16LE   \
-        || (x)==PIX_FMT_YUV422P16LE   \
-        || (x)==PIX_FMT_YUV444P16LE   \
-        || (x)==PIX_FMT_YUV420P16BE   \
-        || (x)==PIX_FMT_YUV422P16BE   \
-        || (x)==PIX_FMT_YUV444P16BE   \
-    )
-
-int sws_isSupportedInput(enum PixelFormat pix_fmt)
-{
-    return isSupportedIn(pix_fmt);
-}
-
-#define isSupportedOut(x)   (       \
-           (x)==PIX_FMT_YUV420P     \
-        || (x)==PIX_FMT_YUVA420P    \
-        || (x)==PIX_FMT_YUYV422     \
-        || (x)==PIX_FMT_UYVY422     \
-        || (x)==PIX_FMT_YUV444P     \
-        || (x)==PIX_FMT_YUV422P     \
-        || (x)==PIX_FMT_YUV411P     \
-        || isRGB(x)                 \
-        || isBGR(x)                 \
-        || (x)==PIX_FMT_NV12        \
-        || (x)==PIX_FMT_NV21        \
-        || (x)==PIX_FMT_GRAY16BE    \
-        || (x)==PIX_FMT_GRAY16LE    \
-        || (x)==PIX_FMT_GRAY8       \
-        || (x)==PIX_FMT_YUV410P     \
-        || (x)==PIX_FMT_YUV440P     \
-        || (x)==PIX_FMT_YUV420P16LE   \
-        || (x)==PIX_FMT_YUV422P16LE   \
-        || (x)==PIX_FMT_YUV444P16LE   \
-        || (x)==PIX_FMT_YUV420P16BE   \
-        || (x)==PIX_FMT_YUV422P16BE   \
-        || (x)==PIX_FMT_YUV444P16BE   \
-    )
-
-int sws_isSupportedOutput(enum PixelFormat pix_fmt)
-{
-    return isSupportedOut(pix_fmt);
-}
-
-#define isPacked(x)         (       \
-           (x)==PIX_FMT_PAL8        \
-        || (x)==PIX_FMT_YUYV422     \
-        || (x)==PIX_FMT_UYVY422     \
-        || isRGB(x)                 \
-        || isBGR(x)                 \
-    )
-#define usePal(x) (av_pix_fmt_descriptors[x].flags & PIX_FMT_PAL)
-
 #define RGB2YUV_SHIFT 15
 #define BY ( (int)(0.114*219/255*(1<<RGB2YUV_SHIFT)+0.5))
 #define BV (-(int)(0.081*224/255*(1<<RGB2YUV_SHIFT)+0.5))
@@ -213,17 +80,15 @@ int sws_isSupportedOutput(enum PixelFormat pix_fmt)
 #define RV ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
 #define RU (-(int)(0.169*224/255*(1<<RGB2YUV_SHIFT)+0.5))
 
-extern const int32_t ff_yuv2rgb_coeffs[8][4];
-
 static const double rgb2yuv_table[8][9]={
-    {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
-    {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
-    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
-    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
+    {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5}, //ITU709
+    {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5}, //ITU709
+    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //DEFAULT / ITU601 / ITU624 / SMPTE 170M
+    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //DEFAULT / ITU601 / ITU624 / SMPTE 170M
     {0.59  , 0.11  , 0.30  , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
-    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
-    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
-    {0.701 , 0.087 , 0.212 , -0.384, 0.5  -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
+    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //DEFAULT / ITU601 / ITU624 / SMPTE 170M
+    {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //DEFAULT / ITU601 / ITU624 / SMPTE 170M
+    {0.701 , 0.087 , 0.212 , -0.384, 0.5, -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
 };
 
 /*
@@ -241,76 +106,24 @@ add BGR4 output support
 write special BGR->BGR scaler
 */
 
-#if ARCH_X86 && CONFIG_GPL
-DECLARE_ASM_CONST(8, uint64_t, bF8)=       0xF8F8F8F8F8F8F8F8LL;
-DECLARE_ASM_CONST(8, uint64_t, bFC)=       0xFCFCFCFCFCFCFCFCLL;
-DECLARE_ASM_CONST(8, uint64_t, w10)=       0x0010001000100010LL;
-DECLARE_ASM_CONST(8, uint64_t, w02)=       0x0002000200020002LL;
-DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
-DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
-DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
-DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
-
-const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
-        0x0103010301030103LL,
-        0x0200020002000200LL,};
-
-const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
-        0x0602060206020602LL,
-        0x0004000400040004LL,};
-
-DECLARE_ASM_CONST(8, uint64_t, b16Mask)=   0x001F001F001F001FLL;
-DECLARE_ASM_CONST(8, uint64_t, g16Mask)=   0x07E007E007E007E0LL;
-DECLARE_ASM_CONST(8, uint64_t, r16Mask)=   0xF800F800F800F800LL;
-DECLARE_ASM_CONST(8, uint64_t, b15Mask)=   0x001F001F001F001FLL;
-DECLARE_ASM_CONST(8, uint64_t, g15Mask)=   0x03E003E003E003E0LL;
-DECLARE_ASM_CONST(8, uint64_t, r15Mask)=   0x7C007C007C007C00LL;
-
-DECLARE_ALIGNED(8, const uint64_t, ff_M24A)         = 0x00FF0000FF0000FFLL;
-DECLARE_ALIGNED(8, const uint64_t, ff_M24B)         = 0xFF0000FF0000FF00LL;
-DECLARE_ALIGNED(8, const uint64_t, ff_M24C)         = 0x0000FF0000FF0000LL;
-
-#ifdef FAST_BGR2YV12
-DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff)   = 0x000000210041000DULL;
-DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff)   = 0x0000FFEEFFDC0038ULL;
-DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff)   = 0x00000038FFD2FFF8ULL;
-#else
-DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff)   = 0x000020E540830C8BULL;
-DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff)   = 0x0000ED0FDAC23831ULL;
-DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff)   = 0x00003831D0E6F6EAULL;
-#endif /* FAST_BGR2YV12 */
-DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset)  = 0x1010101010101010ULL;
-DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
-DECLARE_ALIGNED(8, const uint64_t, ff_w1111)        = 0x0001000100010001ULL;
-
-DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
-DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
-DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
-DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
-DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;
-
-DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUV[2][4]) = {
-    {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
-    {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
-};
-
-DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;
-
-#endif /* ARCH_X86 && CONFIG_GPL */
-
-static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
-
-DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_4[2][8])={
+DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_4)[2][8]={
 {  1,   3,   1,   3,   1,   3,   1,   3, },
 {  2,   0,   2,   0,   2,   0,   2,   0, },
 };
 
-DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_8[2][8])={
+DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_8)[2][8]={
 {  6,   2,   6,   2,   6,   2,   6,   2, },
 {  0,   4,   0,   4,   0,   4,   0,   4, },
 };
 
-DECLARE_ALIGNED(8, const uint8_t, dither_8x8_32[8][8])={
+DECLARE_ALIGNED(8, const uint8_t, dither_4x4_16)[4][8]={
+{  8,   4,  11,   7,   8,   4,  11,   7, },
+{  2,  14,   1,  13,   2,  14,   1,  13, },
+{ 10,   6,   9,   5,  10,   6,   9,   5, },
+{  0,  12,   3,  15,   0,  12,   3,  15, },
+};
+
+DECLARE_ALIGNED(8, const uint8_t, dither_8x8_32)[8][8]={
 { 17,   9,  23,  15,  16,   8,  22,  14, },
 {  5,  29,   3,  27,   4,  28,   2,  26, },
 { 21,  13,  19,  11,  20,  12,  18,  10, },
@@ -321,7 +134,7 @@ DECLARE_ALIGNED(8, const uint8_t, dither_8x8_32[8][8])={
 {  1,  25,   7,  31,   0,  24,   6,  30, },
 };
 
-DECLARE_ALIGNED(8, const uint8_t, dither_8x8_73[8][8])={
+DECLARE_ALIGNED(8, const uint8_t, dither_8x8_73)[8][8]={
 {  0,  55,  14,  68,   3,  58,  17,  72, },
 { 37,  18,  50,  32,  40,  22,  54,  35, },
 {  9,  64,   5,  59,  13,  67,   8,  63, },
@@ -333,7 +146,7 @@ DECLARE_ALIGNED(8, const uint8_t, dither_8x8_73[8][8])={
 };
 
 #if 1
-DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
+DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
 {117,  62, 158, 103, 113,  58, 155, 100, },
 { 34, 199,  21, 186,  31, 196,  17, 182, },
 {144,  89, 131,  76, 141,  86, 127,  72, },
@@ -345,7 +158,7 @@ DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
 };
 #elif 1
 // tries to correct a gamma of 1.5
-DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
+DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
 {  0, 143,  18, 200,   2, 156,  25, 215, },
 { 78,  28, 125,  64,  89,  36, 138,  74, },
 { 10, 180,   3, 161,  16, 195,   8, 175, },
@@ -357,7 +170,7 @@ DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
 };
 #elif 1
 // tries to correct a gamma of 2.0
-DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
+DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
 {  0, 124,   8, 193,   0, 140,  12, 213, },
 { 55,  14, 104,  42,  66,  19, 119,  52, },
 {  3, 168,   1, 145,   6, 187,   3, 162, },
@@ -369,7 +182,7 @@ DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
 };
 #else
 // tries to correct a gamma of 2.5
-DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
+DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
 {  0, 107,   3, 187,   0, 125,   6, 212, },
 { 39,   7,  86,  28,  49,  11, 102,  36, },
 {  1, 158,   0, 131,   3, 180,   1, 151, },
@@ -381,97 +194,133 @@ DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
 };
 #endif
 
-const char *sws_format_name(enum PixelFormat format)
-{
-    if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
-        return av_pix_fmt_descriptors[format].name;
-    else
-        return "Unknown format";
-}
-
-static av_always_inline void yuv2yuvX16inC_template(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
-                                                    const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
-                                                    const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest,
-                                                    int dstW, int chrDstW, int big_endian)
+static av_always_inline void
+yuv2yuvX16_c_template(const int16_t *lumFilter, const int16_t **lumSrc,
+                      int lumFilterSize, const int16_t *chrFilter,
+                      const int16_t **chrUSrc, const int16_t **chrVSrc,
+                      int chrFilterSize, const int16_t **alpSrc,
+                      uint16_t *dest, uint16_t *uDest, uint16_t *vDest,
+                      uint16_t *aDest, int dstW, int chrDstW,
+                      int big_endian, int output_bits)
 {
     //FIXME Optimize (just quickly written not optimized..)
     int i;
-
+    int shift = 11 + 16 - output_bits;
+
+#define output_pixel(pos, val) \
+    if (big_endian) { \
+        if (output_bits == 16) { \
+            AV_WB16(pos, av_clip_uint16(val >> shift)); \
+        } else { \
+            AV_WB16(pos, av_clip_uintp2(val >> shift, output_bits)); \
+        } \
+    } else { \
+        if (output_bits == 16) { \
+            AV_WL16(pos, av_clip_uint16(val >> shift)); \
+        } else { \
+            AV_WL16(pos, av_clip_uintp2(val >> shift, output_bits)); \
+        } \
+    }
     for (i = 0; i < dstW; i++) {
-        int val = 1 << 10;
+        int val = 1 << (26-output_bits);
         int j;
 
         for (j = 0; j < lumFilterSize; j++)
             val += lumSrc[j][i] * lumFilter[j];
 
-        if (big_endian) {
-            AV_WB16(&dest[i], av_clip_uint16(val >> 11));
-        } else {
-            AV_WL16(&dest[i], av_clip_uint16(val >> 11));
-        }
+        output_pixel(&dest[i], val);
     }
 
     if (uDest) {
         for (i = 0; i < chrDstW; i++) {
-            int u = 1 << 10;
-            int v = 1 << 10;
+            int u = 1 << (26-output_bits);
+            int v = 1 << (26-output_bits);
             int j;
 
             for (j = 0; j < chrFilterSize; j++) {
-                u += chrSrc[j][i       ] * chrFilter[j];
-                v += chrSrc[j][i + VOFW] * chrFilter[j];
+                u += chrUSrc[j][i] * chrFilter[j];
+                v += chrVSrc[j][i] * chrFilter[j];
             }
 
-            if (big_endian) {
-                AV_WB16(&uDest[i], av_clip_uint16(u >> 11));
-                AV_WB16(&vDest[i], av_clip_uint16(v >> 11));
-            } else {
-                AV_WL16(&uDest[i], av_clip_uint16(u >> 11));
-                AV_WL16(&vDest[i], av_clip_uint16(v >> 11));
-            }
+            output_pixel(&uDest[i], u);
+            output_pixel(&vDest[i], v);
         }
     }
 
     if (CONFIG_SWSCALE_ALPHA && aDest) {
         for (i = 0; i < dstW; i++) {
-            int val = 1 << 10;
+            int val = 1 << (26-output_bits);
             int j;
 
             for (j = 0; j < lumFilterSize; j++)
                 val += alpSrc[j][i] * lumFilter[j];
 
-            if (big_endian) {
-                AV_WB16(&aDest[i], av_clip_uint16(val >> 11));
-            } else {
-                AV_WL16(&aDest[i], av_clip_uint16(val >> 11));
-            }
+            output_pixel(&aDest[i], val);
         }
     }
-}
-
-static inline void yuv2yuvX16inC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
-                                 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
-                                 const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest, int dstW, int chrDstW,
-                                 enum PixelFormat dstFormat)
-{
-    if (isBE(dstFormat)) {
-        yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
-                               chrFilter, chrSrc, chrFilterSize,
-                               alpSrc,
-                               dest, uDest, vDest, aDest,
-                               dstW, chrDstW, 1);
+#undef output_pixel
+}
+
+#define yuv2NBPS(bits, BE_LE, is_be) \
+static void yuv2yuvX ## bits ## BE_LE ## _c(SwsContext *c, const int16_t *lumFilter, \
+                              const int16_t **lumSrc, int lumFilterSize, \
+                              const int16_t *chrFilter, const int16_t **chrUSrc, \
+                              const int16_t **chrVSrc, \
+                              int chrFilterSize, const int16_t **alpSrc, \
+                              uint8_t *_dest, uint8_t *_uDest, uint8_t *_vDest, \
+                              uint8_t *_aDest, int dstW, int chrDstW) \
+{ \
+    uint16_t *dest  = (uint16_t *) _dest,  *uDest = (uint16_t *) _uDest, \
+             *vDest = (uint16_t *) _vDest, *aDest = (uint16_t *) _aDest; \
+    yuv2yuvX16_c_template(lumFilter, lumSrc, lumFilterSize, \
+                          chrFilter, chrUSrc, chrVSrc, chrFilterSize, \
+                          alpSrc, \
+                          dest, uDest, vDest, aDest, \
+                          dstW, chrDstW, is_be, bits); \
+}
+yuv2NBPS( 9, BE, 1);
+yuv2NBPS( 9, LE, 0);
+yuv2NBPS(10, BE, 1);
+yuv2NBPS(10, LE, 0);
+yuv2NBPS(16, BE, 1);
+yuv2NBPS(16, LE, 0);
+
+static inline void yuv2yuvX16_c(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
+                                const int16_t *chrFilter, const int16_t **chrUSrc, const int16_t **chrVSrc, int chrFilterSize,
+                                const int16_t **alpSrc, uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, int dstW, int chrDstW,
+                                enum PixelFormat dstFormat)
+{
+#define conv16(bits) \
+    if (isBE(dstFormat)) { \
+        yuv2yuvX ## bits ## BE_c(c, lumFilter, lumSrc, lumFilterSize, \
+                                 chrFilter, chrUSrc, chrVSrc, chrFilterSize, \
+                                 alpSrc, \
+                                 dest, uDest, vDest, aDest, \
+                                 dstW, chrDstW); \
+    } else { \
+        yuv2yuvX ## bits ## LE_c(c, lumFilter, lumSrc, lumFilterSize, \
+                                 chrFilter, chrUSrc, chrVSrc, chrFilterSize, \
+                                 alpSrc, \
+                                 dest, uDest, vDest, aDest, \
+                                 dstW, chrDstW); \
+    }
+    if (is16BPS(dstFormat)) {
+        conv16(16);
+    } else if (av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1 == 8) {
+        conv16(9);
     } else {
-        yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
-                               chrFilter, chrSrc, chrFilterSize,
-                               alpSrc,
-                               dest, uDest, vDest, aDest,
-                               dstW, chrDstW, 0);
+        conv16(10);
     }
+#undef conv16
 }
 
-static inline void yuv2yuvXinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
-                               const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
-                               const int16_t **alpSrc, uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, int dstW, int chrDstW)
+static inline void yuv2yuvX_c(SwsContext *c, const int16_t *lumFilter,
+                              const int16_t **lumSrc, int lumFilterSize,
+                              const int16_t *chrFilter, const int16_t **chrUSrc,
+                              const int16_t **chrVSrc,
+                              int chrFilterSize, const int16_t **alpSrc,
+                              uint8_t *dest, uint8_t *uDest, uint8_t *vDest,
+                              uint8_t *aDest, int dstW, int chrDstW)
 {
     //FIXME Optimize (just quickly written not optimized..)
     int i;
@@ -490,8 +339,8 @@ static inline void yuv2yuvXinC(const int16_t *lumFilter, const int16_t **lumSrc,
             int v=1<<18;
             int j;
             for (j=0; j<chrFilterSize; j++) {
-                u += chrSrc[j][i] * chrFilter[j];
-                v += chrSrc[j][i + VOFW] * chrFilter[j];
+                u += chrUSrc[j][i] * chrFilter[j];
+                v += chrVSrc[j][i] * chrFilter[j];
             }
 
             uDest[i]= av_clip_uint8(u>>19);
@@ -510,9 +359,12 @@ static inline void yuv2yuvXinC(const int16_t *lumFilter, const int16_t **lumSrc,
 
 }
 
-static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
-                                const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
-                                uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
+static inline void yuv2nv12X_c(SwsContext *c, const int16_t *lumFilter,
+                               const int16_t **lumSrc, int lumFilterSize,
+                               const int16_t *chrFilter, const int16_t **chrUSrc,
+                               const int16_t **chrVSrc,
+                               int chrFilterSize, uint8_t *dest, uint8_t *uDest,
+                               int dstW, int chrDstW, enum PixelFormat dstFormat)
 {
     //FIXME Optimize (just quickly written not optimized..)
     int i;
@@ -534,8 +386,8 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             int v=1<<18;
             int j;
             for (j=0; j<chrFilterSize; j++) {
-                u += chrSrc[j][i] * chrFilter[j];
-                v += chrSrc[j][i + VOFW] * chrFilter[j];
+                u += chrUSrc[j][i] * chrFilter[j];
+                v += chrVSrc[j][i] * chrFilter[j];
             }
 
             uDest[2*i]= av_clip_uint8(u>>19);
@@ -547,8 +399,8 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             int v=1<<18;
             int j;
             for (j=0; j<chrFilterSize; j++) {
-                u += chrSrc[j][i] * chrFilter[j];
-                v += chrSrc[j][i + VOFW] * chrFilter[j];
+                u += chrUSrc[j][i] * chrFilter[j];
+                v += chrVSrc[j][i] * chrFilter[j];
             }
 
             uDest[2*i]= av_clip_uint8(v>>19);
@@ -572,8 +424,8 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             Y2 += lumSrc[j][i2+1] * lumFilter[j];\
         }\
         for (j=0; j<chrFilterSize; j++) {\
-            U += chrSrc[j][i] * chrFilter[j];\
-            V += chrSrc[j][i+VOFW] * chrFilter[j];\
+            U += chrUSrc[j][i] * chrFilter[j];\
+            V += chrVSrc[j][i] * chrFilter[j];\
         }\
         Y1>>=19;\
         Y2>>=19;\
@@ -588,7 +440,7 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             }\
             A1>>=19;\
             A2>>=19;\
-        }\
+        }
 
 #define YSCALE_YUV_2_PACKEDX_C(type,alpha) \
         YSCALE_YUV_2_PACKEDX_NOCLIP_C(type,alpha)\
@@ -620,8 +472,8 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             Y += lumSrc[j][i     ] * lumFilter[j];\
         }\
         for (j=0; j<chrFilterSize; j++) {\
-            U += chrSrc[j][i     ] * chrFilter[j];\
-            V += chrSrc[j][i+VOFW] * chrFilter[j];\
+            U += chrUSrc[j][i] * chrFilter[j];\
+            V += chrVSrc[j][i] * chrFilter[j];\
         }\
         Y >>=10;\
         U >>=10;\
@@ -633,7 +485,7 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             A >>=19;\
             if (A&256)\
                 A = av_clip_uint8(A);\
-        }\
+        }
 
 #define YSCALE_YUV_2_RGBX_FULL_C(rnd,alpha) \
     YSCALE_YUV_2_PACKEDX_FULL_C(rnd>>3,alpha)\
@@ -650,8 +502,7 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             else if (G<0)G=0;   \
             if (B>=(256<<22))   B=(256<<22)-1; \
             else if (B<0)B=0;   \
-        }\
-
+        }
 
 #define YSCALE_YUV_2_GRAY16_C \
     for (i=0; i<(dstW>>1); i++) {\
@@ -680,79 +531,79 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
     YSCALE_YUV_2_PACKEDX_C(type,alpha)  /* FIXME fix tables so that clipping is not needed and then use _NOCLIP*/\
     r = (type *)c->table_rV[V];   \
     g = (type *)(c->table_gU[U] + c->table_gV[V]); \
-    b = (type *)c->table_bU[U];   \
+    b = (type *)c->table_bU[U];
 
 #define YSCALE_YUV_2_PACKED2_C(type,alpha)   \
     for (i=0; i<(dstW>>1); i++) { \
         const int i2= 2*i;       \
         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;           \
         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;           \
-        int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;  \
-        int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19;  \
+        int U= (ubuf0[i]*uvalpha1+ubuf1[i]*uvalpha)>>19;              \
+        int V= (vbuf0[i]*uvalpha1+vbuf1[i]*uvalpha)>>19;              \
         type av_unused *r, *b, *g;                                    \
         int av_unused A1, A2;                                         \
         if (alpha) {\
             A1= (abuf0[i2  ]*yalpha1+abuf1[i2  ]*yalpha)>>19;         \
             A2= (abuf0[i2+1]*yalpha1+abuf1[i2+1]*yalpha)>>19;         \
-        }\
+        }
 
 #define YSCALE_YUV_2_GRAY16_2_C   \
     for (i=0; i<(dstW>>1); i++) { \
         const int i2= 2*i;       \
         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>11;           \
-        int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11;           \
+        int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11;
 
 #define YSCALE_YUV_2_RGB2_C(type,alpha) \
     YSCALE_YUV_2_PACKED2_C(type,alpha)\
     r = (type *)c->table_rV[V];\
     g = (type *)(c->table_gU[U] + c->table_gV[V]);\
-    b = (type *)c->table_bU[U];\
+    b = (type *)c->table_bU[U];
 
 #define YSCALE_YUV_2_PACKED1_C(type,alpha) \
     for (i=0; i<(dstW>>1); i++) {\
         const int i2= 2*i;\
         int Y1= buf0[i2  ]>>7;\
         int Y2= buf0[i2+1]>>7;\
-        int U= (uvbuf1[i     ])>>7;\
-        int V= (uvbuf1[i+VOFW])>>7;\
+        int U= (ubuf1[i])>>7;\
+        int V= (vbuf1[i])>>7;\
         type av_unused *r, *b, *g;\
         int av_unused A1, A2;\
         if (alpha) {\
             A1= abuf0[i2  ]>>7;\
             A2= abuf0[i2+1]>>7;\
-        }\
+        }
 
 #define YSCALE_YUV_2_GRAY16_1_C \
     for (i=0; i<(dstW>>1); i++) {\
         const int i2= 2*i;\
         int Y1= buf0[i2  ]<<1;\
-        int Y2= buf0[i2+1]<<1;\
+        int Y2= buf0[i2+1]<<1;
 
 #define YSCALE_YUV_2_RGB1_C(type,alpha) \
     YSCALE_YUV_2_PACKED1_C(type,alpha)\
     r = (type *)c->table_rV[V];\
     g = (type *)(c->table_gU[U] + c->table_gV[V]);\
-    b = (type *)c->table_bU[U];\
+    b = (type *)c->table_bU[U];
 
 #define YSCALE_YUV_2_PACKED1B_C(type,alpha) \
     for (i=0; i<(dstW>>1); i++) {\
         const int i2= 2*i;\
         int Y1= buf0[i2  ]>>7;\
         int Y2= buf0[i2+1]>>7;\
-        int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
-        int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
+        int U= (ubuf0[i] + ubuf1[i])>>8;\
+        int V= (vbuf0[i] + vbuf1[i])>>8;\
         type av_unused *r, *b, *g;\
         int av_unused A1, A2;\
         if (alpha) {\
             A1= abuf0[i2  ]>>7;\
             A2= abuf0[i2+1]>>7;\
-        }\
+        }
 
 #define YSCALE_YUV_2_RGB1B_C(type,alpha) \
     YSCALE_YUV_2_PACKED1B_C(type,alpha)\
     r = (type *)c->table_rV[V];\
     g = (type *)(c->table_gU[U] + c->table_gV[V]);\
-    b = (type *)c->table_bU[U];\
+    b = (type *)c->table_bU[U];
 
 #define YSCALE_YUV_2_MONO2_C \
     const uint8_t * const d128=dither_8x8_220[y&7];\
@@ -769,8 +620,7 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
         acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
         ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
         dest++;\
-    }\
-
+    }
 
 #define YSCALE_YUV_2_MONOX_C \
     const uint8_t * const d128=dither_8x8_220[y&7];\
@@ -801,7 +651,6 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
         }\
     }
 
-
 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
     switch(c->dstFormat) {\
     case PIX_FMT_RGB48BE:\
@@ -822,6 +671,18 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             dest+=12;\
         }\
         break;\
+    case PIX_FMT_BGR48BE:\
+    case PIX_FMT_BGR48LE:\
+        func(uint8_t,0)\
+            ((uint8_t*)dest)[ 0] = ((uint8_t*)dest)[ 1] = b[Y1];\
+            ((uint8_t*)dest)[ 2] = ((uint8_t*)dest)[ 3] = g[Y1];\
+            ((uint8_t*)dest)[ 4] = ((uint8_t*)dest)[ 5] = r[Y1];\
+            ((uint8_t*)dest)[ 6] = ((uint8_t*)dest)[ 7] = b[Y2];\
+            ((uint8_t*)dest)[ 8] = ((uint8_t*)dest)[ 9] = g[Y2];\
+            ((uint8_t*)dest)[10] = ((uint8_t*)dest)[11] = r[Y2];\
+            dest+=12;\
+        }\
+        break;\
     case PIX_FMT_RGBA:\
     case PIX_FMT_BGRA:\
         if (CONFIG_SMALL) {\
@@ -888,8 +749,10 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             dest+=6;\
         }\
         break;\
-    case PIX_FMT_RGB565:\
-    case PIX_FMT_BGR565:\
+    case PIX_FMT_RGB565BE:\
+    case PIX_FMT_RGB565LE:\
+    case PIX_FMT_BGR565BE:\
+    case PIX_FMT_BGR565LE:\
         {\
             const int dr1= dither_2x2_8[y&1    ][0];\
             const int dg1= dither_2x2_4[y&1    ][0];\
@@ -903,8 +766,10 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             }\
         }\
         break;\
-    case PIX_FMT_RGB555:\
-    case PIX_FMT_BGR555:\
+    case PIX_FMT_RGB555BE:\
+    case PIX_FMT_RGB555LE:\
+    case PIX_FMT_BGR555BE:\
+    case PIX_FMT_BGR555LE:\
         {\
             const int dr1= dither_2x2_8[y&1    ][0];\
             const int dg1= dither_2x2_8[y&1    ][1];\
@@ -918,6 +783,23 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             }\
         }\
         break;\
+    case PIX_FMT_RGB444BE:\
+    case PIX_FMT_RGB444LE:\
+    case PIX_FMT_BGR444BE:\
+    case PIX_FMT_BGR444LE:\
+        {\
+            const int dr1= dither_4x4_16[y&3    ][0];\
+            const int dg1= dither_4x4_16[y&3    ][1];\
+            const int db1= dither_4x4_16[(y&3)^3][0];\
+            const int dr2= dither_4x4_16[y&3    ][1];\
+            const int dg2= dither_4x4_16[y&3    ][0];\
+            const int db2= dither_4x4_16[(y&3)^3][1];\
+            func(uint16_t,0)\
+                ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
+                ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
+            }\
+        }\
+        break;\
     case PIX_FMT_RGB8:\
     case PIX_FMT_BGR8:\
         {\
@@ -989,23 +871,26 @@ static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc
             ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
         }                \
         break;\
-    }\
-
+    }
 
-static inline void yuv2packedXinC(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
-                                  const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
-                                  const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
+static void yuv2packedX_c(SwsContext *c, const int16_t *lumFilter,
+                          const int16_t **lumSrc, int lumFilterSize,
+                          const int16_t *chrFilter, const int16_t **chrUSrc,
+                          const int16_t **chrVSrc, int chrFilterSize,
+                          const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
 {
     int i;
     YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGBX_C, YSCALE_YUV_2_PACKEDX_C(void,0), YSCALE_YUV_2_GRAY16_C, YSCALE_YUV_2_MONOX_C)
 }
 
-static inline void yuv2rgbXinC_full(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
-                                    const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
-                                    const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
+static inline void yuv2rgbX_c_full(SwsContext *c, const int16_t *lumFilter,
+                                   const int16_t **lumSrc, int lumFilterSize,
+                                   const int16_t *chrFilter, const int16_t **chrUSrc,
+                                   const int16_t **chrVSrc, int chrFilterSize,
+                                   const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
 {
     int i;
-    int step= fmt_depth(c->dstFormat)/8;
+    int step= c->dstFormatBpp/8;
     int aidx= 3;
 
     switch(c->dstFormat) {
@@ -1094,53 +979,62 @@ static void fillPlane(uint8_t* plane, int stride, int width, int height, int y,
     }
 }
 
-static inline void rgb48ToY(uint8_t *dst, const uint8_t *src, int width,
-                            uint32_t *unused)
-{
-    int i;
-    for (i = 0; i < width; i++) {
-        int r = src[i*6+0];
-        int g = src[i*6+2];
-        int b = src[i*6+4];
-
-        dst[i] = (RY*r + GY*g + BY*b + (33<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
-    }
-}
-
-static inline void rgb48ToUV(uint8_t *dstU, uint8_t *dstV,
-                             const uint8_t *src1, const uint8_t *src2,
-                             int width, uint32_t *unused)
-{
-    int i;
-    assert(src1==src2);
-    for (i = 0; i < width; i++) {
-        int r = src1[6*i + 0];
-        int g = src1[6*i + 2];
-        int b = src1[6*i + 4];
-
-        dstU[i] = (RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
-        dstV[i] = (RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
-    }
-}
-
-static inline void rgb48ToUV_half(uint8_t *dstU, uint8_t *dstV,
-                                  const uint8_t *src1, const uint8_t *src2,
-                                  int width, uint32_t *unused)
-{
-    int i;
-    assert(src1==src2);
-    for (i = 0; i < width; i++) {
-        int r= src1[12*i + 0] + src1[12*i + 6];
-        int g= src1[12*i + 2] + src1[12*i + 8];
-        int b= src1[12*i + 4] + src1[12*i + 10];
-
-        dstU[i]= (RU*r + GU*g + BU*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1);
-        dstV[i]= (RV*r + GV*g + BV*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1);
-    }
-}
+#define rgb48funcs(LE_BE, rfunc, compA, compB, compC) \
+static void compA ## compB ## compC ## 48 ## LE_BE ## ToY_c( \
+                       uint8_t *dst, const uint8_t *src, int width, \
+                       uint32_t *unused) \
+{ \
+    int i; \
+    for (i = 0; i < width; i++) { \
+        int compA = rfunc(&src[i*6+0]) >> 8; \
+        int compB = rfunc(&src[i*6+2]) >> 8; \
+        int compC = rfunc(&src[i*6+4]) >> 8; \
+ \
+        dst[i] = (RY*r + GY*g + BY*b + (33<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT; \
+    } \
+} \
+ \
+static void compA ## compB ## compC ## 48 ## LE_BE ## ToUV_c( \
+                        uint8_t *dstU, uint8_t *dstV, \
+                        const uint8_t *src1, const uint8_t *src2, \
+                        int width, uint32_t *unused) \
+{ \
+    int i; \
+    assert(src1==src2); \
+    for (i = 0; i < width; i++) { \
+        int compA = rfunc(&src1[6*i + 0]) >> 8; \
+        int compB = rfunc(&src1[6*i + 2]) >> 8; \
+        int compC = rfunc(&src1[6*i + 4]) >> 8; \
+ \
+        dstU[i] = (RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT; \
+        dstV[i] = (RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT; \
+    } \
+} \
+ \
+static void compA ## compB ## compC ## 48 ## LE_BE ## ToUV_half_c( \
+                            uint8_t *dstU, uint8_t *dstV, \
+                            const uint8_t *src1, const uint8_t *src2, \
+                            int width, uint32_t *unused) \
+{ \
+    int i; \
+    assert(src1==src2); \
+    for (i = 0; i < width; i++) { \
+        int compA = (rfunc(&src1[12*i + 0]) >> 8) + (rfunc(&src1[12*i + 6]) >> 8); \
+        int compB = (rfunc(&src1[12*i + 2]) >> 8) + (rfunc(&src1[12*i + 8]) >> 8); \
+        int compC = (rfunc(&src1[12*i + 4]) >> 8) + (rfunc(&src1[12*i + 10]) >> 8); \
+ \
+        dstU[i]= (RU*r + GU*g + BU*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1); \
+        dstV[i]= (RV*r + GV*g + BV*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1); \
+    } \
+}
+rgb48funcs(LE, AV_RL16, r, g, b);
+rgb48funcs(BE, AV_RB16, r, g, b);
+rgb48funcs(LE, AV_RL16, b, g, r);
+rgb48funcs(BE, AV_RB16, b, g, r);
 
 #define BGR2Y(type, name, shr, shg, shb, maskr, maskg, maskb, RY, GY, BY, S)\
-static inline void name(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)\
+static void name ## _c(uint8_t *dst, const uint8_t *src, \
+                       int width, uint32_t *unused)\
 {\
     int i;\
     for (i=0; i<width; i++) {\
@@ -1153,13 +1047,15 @@ static inline void name(uint8_t *dst, const uint8_t *src, long width, uint32_t *
 }
 
 BGR2Y(uint32_t, bgr32ToY,16, 0, 0, 0x00FF, 0xFF00, 0x00FF, RY<< 8, GY   , BY<< 8, RGB2YUV_SHIFT+8)
+BGR2Y(uint32_t,bgr321ToY,16,16, 0, 0xFF00, 0x00FF, 0xFF00, RY    , GY<<8, BY    , RGB2YUV_SHIFT+8)
 BGR2Y(uint32_t, rgb32ToY, 0, 0,16, 0x00FF, 0xFF00, 0x00FF, RY<< 8, GY   , BY<< 8, RGB2YUV_SHIFT+8)
+BGR2Y(uint32_t,rgb321ToY, 0,16,16, 0xFF00, 0x00FF, 0xFF00, RY    , GY<<8, BY    , RGB2YUV_SHIFT+8)
 BGR2Y(uint16_t, bgr16ToY, 0, 0, 0, 0x001F, 0x07E0, 0xF800, RY<<11, GY<<5, BY    , RGB2YUV_SHIFT+8)
 BGR2Y(uint16_t, bgr15ToY, 0, 0, 0, 0x001F, 0x03E0, 0x7C00, RY<<10, GY<<5, BY    , RGB2YUV_SHIFT+7)
 BGR2Y(uint16_t, rgb16ToY, 0, 0, 0, 0xF800, 0x07E0, 0x001F, RY    , GY<<5, BY<<11, RGB2YUV_SHIFT+8)
 BGR2Y(uint16_t, rgb15ToY, 0, 0, 0, 0x7C00, 0x03E0, 0x001F, RY    , GY<<5, BY<<10, RGB2YUV_SHIFT+7)
 
-static inline void abgrToA(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
+static void abgrToA_c(uint8_t *dst, const uint8_t *src, int width, uint32_t *unused)
 {
     int i;
     for (i=0; i<width; i++) {
@@ -1167,25 +1063,37 @@ static inline void abgrToA(uint8_t *dst, const uint8_t *src, long width, uint32_
     }
 }
 
-#define BGR2UV(type, name, shr, shg, shb, maska, maskr, maskg, maskb, RU, GU, BU, RV, GV, BV, S)\
-static inline void name(uint8_t *dstU, uint8_t *dstV, const uint8_t *src, const uint8_t *dummy, long width, uint32_t *unused)\
+static void rgbaToA_c(uint8_t *dst, const uint8_t *src, int width, uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++) {
+        dst[i]= src[4*i+3];
+    }
+}
+
+#define BGR2UV(type, name, shr, shg, shb, shp, maskr, maskg, maskb, RU, GU, BU, RV, GV, BV, S) \
+static void name ## _c(uint8_t *dstU, uint8_t *dstV, \
+                       const uint8_t *src, const uint8_t *dummy, \
+                       int width, uint32_t *unused)\
 {\
     int i;\
     for (i=0; i<width; i++) {\
-        int b= (((const type*)src)[i]&maskb)>>shb;\
-        int g= (((const type*)src)[i]&maskg)>>shg;\
-        int r= (((const type*)src)[i]&maskr)>>shr;\
+        int b= ((((const type*)src)[i]>>shp)&maskb)>>shb;\
+        int g= ((((const type*)src)[i]>>shp)&maskg)>>shg;\
+        int r= ((((const type*)src)[i]>>shp)&maskr)>>shr;\
 \
         dstU[i]= ((RU)*r + (GU)*g + (BU)*b + (257<<((S)-1)))>>(S);\
         dstV[i]= ((RV)*r + (GV)*g + (BV)*b + (257<<((S)-1)))>>(S);\
     }\
 }\
-static inline void name ## _half(uint8_t *dstU, uint8_t *dstV, const uint8_t *src, const uint8_t *dummy, long width, uint32_t *unused)\
+static void name ## _half_c(uint8_t *dstU, uint8_t *dstV, \
+                            const uint8_t *src, const uint8_t *dummy, \
+                            int width, uint32_t *unused)\
 {\
     int i;\
     for (i=0; i<width; i++) {\
-        int pix0= ((const type*)src)[2*i+0];\
-        int pix1= ((const type*)src)[2*i+1];\
+        int pix0= ((const type*)src)[2*i+0]>>shp;\
+        int pix1= ((const type*)src)[2*i+1]>>shp;\
         int g= (pix0&~(maskr|maskb))+(pix1&~(maskr|maskb));\
         int b= ((pix0+pix1-g)&(maskb|(2*maskb)))>>shb;\
         int r= ((pix0+pix1-g)&(maskr|(2*maskr)))>>shr;\
@@ -1198,14 +1106,16 @@ static inline void name ## _half(uint8_t *dstU, uint8_t *dstV, const uint8_t *sr
     }\
 }
 
-BGR2UV(uint32_t, bgr32ToUV,16, 0, 0, 0xFF000000, 0xFF0000, 0xFF00,   0x00FF, RU<< 8, GU   , BU<< 8, RV<< 8, GV   , BV<< 8, RGB2YUV_SHIFT+8)
-BGR2UV(uint32_t, rgb32ToUV, 0, 0,16, 0xFF000000,   0x00FF, 0xFF00, 0xFF0000, RU<< 8, GU   , BU<< 8, RV<< 8, GV   , BV<< 8, RGB2YUV_SHIFT+8)
-BGR2UV(uint16_t, bgr16ToUV, 0, 0, 0,          0,   0x001F, 0x07E0,   0xF800, RU<<11, GU<<5, BU    , RV<<11, GV<<5, BV    , RGB2YUV_SHIFT+8)
-BGR2UV(uint16_t, bgr15ToUV, 0, 0, 0,          0,   0x001F, 0x03E0,   0x7C00, RU<<10, GU<<5, BU    , RV<<10, GV<<5, BV    , RGB2YUV_SHIFT+7)
-BGR2UV(uint16_t, rgb16ToUV, 0, 0, 0,          0,   0xF800, 0x07E0,   0x001F, RU    , GU<<5, BU<<11, RV    , GV<<5, BV<<11, RGB2YUV_SHIFT+8)
-BGR2UV(uint16_t, rgb15ToUV, 0, 0, 0,          0,   0x7C00, 0x03E0,   0x001F, RU    , GU<<5, BU<<10, RV    , GV<<5, BV<<10, RGB2YUV_SHIFT+7)
+BGR2UV(uint32_t, bgr32ToUV,16, 0, 0, 0, 0xFF0000, 0xFF00,   0x00FF, RU<< 8, GU   , BU<< 8, RV<< 8, GV   , BV<< 8, RGB2YUV_SHIFT+8)
+BGR2UV(uint32_t,bgr321ToUV,16, 0, 0, 8, 0xFF0000, 0xFF00,   0x00FF, RU<< 8, GU   , BU<< 8, RV<< 8, GV   , BV<< 8, RGB2YUV_SHIFT+8)
+BGR2UV(uint32_t, rgb32ToUV, 0, 0,16, 0,   0x00FF, 0xFF00, 0xFF0000, RU<< 8, GU   , BU<< 8, RV<< 8, GV   , BV<< 8, RGB2YUV_SHIFT+8)
+BGR2UV(uint32_t,rgb321ToUV, 0, 0,16, 8,   0x00FF, 0xFF00, 0xFF0000, RU<< 8, GU   , BU<< 8, RV<< 8, GV   , BV<< 8, RGB2YUV_SHIFT+8)
+BGR2UV(uint16_t, bgr16ToUV, 0, 0, 0, 0,   0x001F, 0x07E0,   0xF800, RU<<11, GU<<5, BU    , RV<<11, GV<<5, BV    , RGB2YUV_SHIFT+8)
+BGR2UV(uint16_t, bgr15ToUV, 0, 0, 0, 0,   0x001F, 0x03E0,   0x7C00, RU<<10, GU<<5, BU    , RV<<10, GV<<5, BV    , RGB2YUV_SHIFT+7)
+BGR2UV(uint16_t, rgb16ToUV, 0, 0, 0, 0,   0xF800, 0x07E0,   0x001F, RU    , GU<<5, BU<<11, RV    , GV<<5, BV<<11, RGB2YUV_SHIFT+8)
+BGR2UV(uint16_t, rgb15ToUV, 0, 0, 0, 0,   0x7C00, 0x03E0,   0x001F, RU    , GU<<5, BU<<10, RV    , GV<<5, BV<<10, RGB2YUV_SHIFT+7)
 
-static inline void palToY(uint8_t *dst, const uint8_t *src, long width, uint32_t *pal)
+static void palToY_c(uint8_t *dst, const uint8_t *src, int width, uint32_t *pal)
 {
     int i;
     for (i=0; i<width; i++) {
@@ -1215,9 +1125,9 @@ static inline void palToY(uint8_t *dst, const uint8_t *src, long width, uint32_t
     }
 }
 
-static inline void palToUV(uint8_t *dstU, uint8_t *dstV,
-                           const uint8_t *src1, const uint8_t *src2,
-                           long width, uint32_t *pal)
+static void palToUV_c(uint8_t *dstU, uint8_t *dstV,
+                      const uint8_t *src1, const uint8_t *src2,
+                      int width, uint32_t *pal)
 {
     int i;
     assert(src1 == src2);
@@ -1229,7 +1139,8 @@ static inline void palToUV(uint8_t *dstU, uint8_t *dstV,
     }
 }
 
-static inline void monowhite2Y(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
+static void monowhite2Y_c(uint8_t *dst, const uint8_t *src,
+                          int width, uint32_t *unused)
 {
     int i, j;
     for (i=0; i<width/8; i++) {
@@ -1239,7 +1150,8 @@ static inline void monowhite2Y(uint8_t *dst, const uint8_t *src, long width, uin
     }
 }
 
-static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
+static void monoblack2Y_c(uint8_t *dst, const uint8_t *src,
+                          int width, uint32_t *unused)
 {
     int i, j;
     for (i=0; i<width/8; i++) {
@@ -1249,2210 +1161,888 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
     }
 }
 
-
-//Note: we have C, MMX, MMX2, 3DNOW versions, there is no 3DNOW+MMX2 one
-//Plain C versions
-#if ((!HAVE_MMX || !CONFIG_GPL) && !HAVE_ALTIVEC) || CONFIG_RUNTIME_CPUDETECT
-#define COMPILE_C
-#endif
-
-#if ARCH_PPC
-#if HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT
-#define COMPILE_ALTIVEC
-#endif
-#endif //ARCH_PPC
-
-#if ARCH_X86
-
-#if ((HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
-#define COMPILE_MMX
-#endif
-
-#if (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
-#define COMPILE_MMX2
-#endif
-
-#if ((HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
-#define COMPILE_3DNOW
-#endif
-#endif //ARCH_X86
-
-#define COMPILE_TEMPLATE_MMX 0
-#define COMPILE_TEMPLATE_MMX2 0
-#define COMPILE_TEMPLATE_AMD3DNOW 0
-#define COMPILE_TEMPLATE_ALTIVEC 0
-
-#ifdef COMPILE_C
-#define RENAME(a) a ## _C
-#include "swscale_template.c"
-#endif
-
-#ifdef COMPILE_ALTIVEC
-#undef RENAME
-#undef COMPILE_TEMPLATE_ALTIVEC
-#define COMPILE_TEMPLATE_ALTIVEC 1
-#define RENAME(a) a ## _altivec
-#include "swscale_template.c"
-#endif
-
-#if ARCH_X86
-
-//MMX versions
-#ifdef COMPILE_MMX
-#undef RENAME
-#undef COMPILE_TEMPLATE_MMX
-#undef COMPILE_TEMPLATE_MMX2
-#undef COMPILE_TEMPLATE_AMD3DNOW
-#define COMPILE_TEMPLATE_MMX 1
-#define COMPILE_TEMPLATE_MMX2 0
-#define COMPILE_TEMPLATE_AMD3DNOW 0
-#define RENAME(a) a ## _MMX
-#include "swscale_template.c"
-#endif
-
-//MMX2 versions
-#ifdef COMPILE_MMX2
-#undef RENAME
-#undef COMPILE_TEMPLATE_MMX
-#undef COMPILE_TEMPLATE_MMX2
-#undef COMPILE_TEMPLATE_AMD3DNOW
-#define COMPILE_TEMPLATE_MMX 1
-#define COMPILE_TEMPLATE_MMX2 1
-#define COMPILE_TEMPLATE_AMD3DNOW 0
-#define RENAME(a) a ## _MMX2
-#include "swscale_template.c"
-#endif
-
-//3DNOW versions
-#ifdef COMPILE_3DNOW
-#undef RENAME
-#undef COMPILE_TEMPLATE_MMX
-#undef COMPILE_TEMPLATE_MMX2
-#undef COMPILE_TEMPLATE_AMD3DNOW
-#define COMPILE_TEMPLATE_MMX 1
-#define COMPILE_TEMPLATE_MMX2 0
-#define COMPILE_TEMPLATE_AMD3DNOW 1
-#define RENAME(a) a ## _3DNow
-#include "swscale_template.c"
-#endif
-
-#endif //ARCH_X86
-
-static double getSplineCoeff(double a, double b, double c, double d, double dist)
-{
-//    printf("%f %f %f %f %f\n", a,b,c,d,dist);
-    if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
-    else           return getSplineCoeff(        0.0,
-                                          b+ 2.0*c + 3.0*d,
-                                                 c + 3.0*d,
-                                         -b- 3.0*c - 6.0*d,
-                                         dist-1.0);
-}
-
-static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
-                             int srcW, int dstW, int filterAlign, int one, int flags,
-                             SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
+static void yuv2yuv1_c(SwsContext *c, const int16_t *lumSrc,
+                       const int16_t *chrUSrc, const int16_t *chrVSrc,
+                       const int16_t *alpSrc,
+                       uint8_t *dest, uint8_t *uDest, uint8_t *vDest,
+                       uint8_t *aDest, int dstW, int chrDstW)
 {
     int i;
-    int filterSize;
-    int filter2Size;
-    int minFilterSize;
-    int64_t *filter=NULL;
-    int64_t *filter2=NULL;
-    const int64_t fone= 1LL<<54;
-    int ret= -1;
-#if ARCH_X86
-    if (flags & SWS_CPU_CAPS_MMX)
-        __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
-#endif
-
-    // NOTE: the +1 is for the MMX scaler which reads over the end
-    FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
+    for (i=0; i<dstW; i++) {
+        int val= (lumSrc[i]+64)>>7;
+        dest[i]= av_clip_uint8(val);
+    }
 
-    if (FFABS(xInc - 0x10000) <10) { // unscaled
-        int i;
-        filterSize= 1;
-        FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
+    if (uDest)
+        for (i=0; i<chrDstW; i++) {
+            int u=(chrUSrc[i]+64)>>7;
+            int v=(chrVSrc[i]+64)>>7;
+            uDest[i]= av_clip_uint8(u);
+            vDest[i]= av_clip_uint8(v);
+        }
 
+    if (CONFIG_SWSCALE_ALPHA && aDest)
         for (i=0; i<dstW; i++) {
-            filter[i*filterSize]= fone;
-            (*filterPos)[i]=i;
+            int val= (alpSrc[i]+64)>>7;
+            aDest[i]= av_clip_uint8(val);
         }
+}
 
-    } else if (flags&SWS_POINT) { // lame looking point sampling mode
-        int i;
-        int xDstInSrc;
-        filterSize= 1;
-        FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
+/**
+ * vertical bilinear scale YV12 to RGB
+ */
+static void yuv2packed2_c(SwsContext *c, const uint16_t *buf0,
+                          const uint16_t *buf1, const uint16_t *ubuf0,
+                          const uint16_t *ubuf1, const uint16_t *vbuf0,
+                          const uint16_t *vbuf1, const uint16_t *abuf0,
+                          const uint16_t *abuf1, uint8_t *dest, int dstW,
+                          int yalpha, int uvalpha, int y)
+{
+    int  yalpha1=4095- yalpha;
+    int uvalpha1=4095-uvalpha;
+    int i;
 
-        xDstInSrc= xInc/2 - 0x8000;
-        for (i=0; i<dstW; i++) {
-            int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
+    YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGB2_C, YSCALE_YUV_2_PACKED2_C(void,0), YSCALE_YUV_2_GRAY16_2_C, YSCALE_YUV_2_MONO2_C)
+}
 
-            (*filterPos)[i]= xx;
-            filter[i]= fone;
-            xDstInSrc+= xInc;
-        }
-    } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
-        int i;
-        int xDstInSrc;
-        filterSize= 2;
-        FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
+/**
+ * YV12 to RGB without scaling or interpolating
+ */
+static void yuv2packed1_c(SwsContext *c, const uint16_t *buf0,
+                          const uint16_t *ubuf0, const uint16_t *ubuf1,
+                          const uint16_t *vbuf0, const uint16_t *vbuf1,
+                          const uint16_t *abuf0, uint8_t *dest, int dstW,
+                          int uvalpha, enum PixelFormat dstFormat,
+                          int flags, int y)
+{
+    const int yalpha1=0;
+    int i;
 
-        xDstInSrc= xInc/2 - 0x8000;
-        for (i=0; i<dstW; i++) {
-            int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
-            int j;
+    const uint16_t *buf1= buf0; //FIXME needed for RGB1/BGR1
+    const int yalpha= 4096; //FIXME ...
 
-            (*filterPos)[i]= xx;
-            //bilinear upscale / linear interpolate / area averaging
-            for (j=0; j<filterSize; j++) {
-                int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
-                if (coeff<0) coeff=0;
-                filter[i*filterSize + j]= coeff;
-                xx++;
-            }
-            xDstInSrc+= xInc;
-        }
+    if (uvalpha < 2048) {
+        YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGB1_C, YSCALE_YUV_2_PACKED1_C(void,0), YSCALE_YUV_2_GRAY16_1_C, YSCALE_YUV_2_MONO2_C)
     } else {
-        int xDstInSrc;
-        int sizeFactor;
-
-        if      (flags&SWS_BICUBIC)      sizeFactor=  4;
-        else if (flags&SWS_X)            sizeFactor=  8;
-        else if (flags&SWS_AREA)         sizeFactor=  1; //downscale only, for upscale it is bilinear
-        else if (flags&SWS_GAUSS)        sizeFactor=  8;   // infinite ;)
-        else if (flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
-        else if (flags&SWS_SINC)         sizeFactor= 20; // infinite ;)
-        else if (flags&SWS_SPLINE)       sizeFactor= 20;  // infinite ;)
-        else if (flags&SWS_BILINEAR)     sizeFactor=  2;
-        else {
-            sizeFactor= 0; //GCC warning killer
-            assert(0);
-        }
-
-        if (xInc <= 1<<16)      filterSize= 1 + sizeFactor; // upscale
-        else                    filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
+        YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGB1B_C, YSCALE_YUV_2_PACKED1B_C(void,0), YSCALE_YUV_2_GRAY16_1_C, YSCALE_YUV_2_MONO2_C)
+    }
+}
 
-        if (filterSize > srcW-2) filterSize=srcW-2;
+//FIXME yuy2* can read up to 7 samples too much
 
-        FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
+static void yuy2ToY_c(uint8_t *dst, const uint8_t *src, int width,
+                      uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++)
+        dst[i]= src[2*i];
+}
 
-        xDstInSrc= xInc - 0x10000;
-        for (i=0; i<dstW; i++) {
-            int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
-            int j;
-            (*filterPos)[i]= xx;
-            for (j=0; j<filterSize; j++) {
-                int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
-                double floatd;
-                int64_t coeff;
-
-                if (xInc > 1<<16)
-                    d= d*dstW/srcW;
-                floatd= d * (1.0/(1<<30));
-
-                if (flags & SWS_BICUBIC) {
-                    int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1<<24);
-                    int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
-                    int64_t dd = ( d*d)>>30;
-                    int64_t ddd= (dd*d)>>30;
-
-                    if      (d < 1LL<<30)
-                        coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
-                    else if (d < 1LL<<31)
-                        coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
-                    else
-                        coeff=0.0;
-                    coeff *= fone>>(30+24);
-                }
-/*                else if (flags & SWS_X) {
-                    double p= param ? param*0.01 : 0.3;
-                    coeff = d ? sin(d*PI)/(d*PI) : 1.0;
-                    coeff*= pow(2.0, - p*d*d);
-                }*/
-                else if (flags & SWS_X) {
-                    double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
-                    double c;
-
-                    if (floatd<1.0)
-                        c = cos(floatd*PI);
-                    else
-                        c=-1.0;
-                    if (c<0.0)      c= -pow(-c, A);
-                    else            c=  pow( c, A);
-                    coeff= (c*0.5 + 0.5)*fone;
-                } else if (flags & SWS_AREA) {
-                    int64_t d2= d - (1<<29);
-                    if      (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
-                    else if (d2*xInc <  (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
-                    else coeff=0.0;
-                    coeff *= fone>>(30+16);
-                } else if (flags & SWS_GAUSS) {
-                    double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
-                    coeff = (pow(2.0, - p*floatd*floatd))*fone;
-                } else if (flags & SWS_SINC) {
-                    coeff = (d ? sin(floatd*PI)/(floatd*PI) : 1.0)*fone;
-                } else if (flags & SWS_LANCZOS) {
-                    double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
-                    coeff = (d ? sin(floatd*PI)*sin(floatd*PI/p)/(floatd*floatd*PI*PI/p) : 1.0)*fone;
-                    if (floatd>p) coeff=0;
-                } else if (flags & SWS_BILINEAR) {
-                    coeff= (1<<30) - d;
-                    if (coeff<0) coeff=0;
-                    coeff *= fone >> 30;
-                } else if (flags & SWS_SPLINE) {
-                    double p=-2.196152422706632;
-                    coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
-                } else {
-                    coeff= 0.0; //GCC warning killer
-                    assert(0);
-                }
+static void yuy2ToUV_c(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1,
+                       const uint8_t *src2, int width, uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++) {
+        dstU[i]= src1[4*i + 1];
+        dstV[i]= src1[4*i + 3];
+    }
+    assert(src1 == src2);
+}
 
-                filter[i*filterSize + j]= coeff;
-                xx++;
-            }
-            xDstInSrc+= 2*xInc;
-        }
+static void LEToUV_c(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1,
+                     const uint8_t *src2, int width, uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++) {
+        dstU[i]= src1[2*i + 1];
+        dstV[i]= src2[2*i + 1];
     }
+}
 
-    /* apply src & dst Filter to filter -> filter2
-       av_free(filter);
-    */
-    assert(filterSize>0);
-    filter2Size= filterSize;
-    if (srcFilter) filter2Size+= srcFilter->length - 1;
-    if (dstFilter) filter2Size+= dstFilter->length - 1;
-    assert(filter2Size>0);
-    FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
+/* This is almost identical to the previous, end exists only because
+ * yuy2ToY/UV)(dst, src+1, ...) would have 100% unaligned accesses. */
+static void uyvyToY_c(uint8_t *dst, const uint8_t *src, int width,
+                      uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++)
+        dst[i]= src[2*i+1];
+}
 
-    for (i=0; i<dstW; i++) {
-        int j, k;
+static void uyvyToUV_c(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1,
+                       const uint8_t *src2, int width, uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++) {
+        dstU[i]= src1[4*i + 0];
+        dstV[i]= src1[4*i + 2];
+    }
+    assert(src1 == src2);
+}
 
-        if(srcFilter) {
-            for (k=0; k<srcFilter->length; k++) {
-                for (j=0; j<filterSize; j++)
-                    filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
-            }
-        } else {
-            for (j=0; j<filterSize; j++)
-                filter2[i*filter2Size + j]= filter[i*filterSize + j];
-        }
-        //FIXME dstFilter
+static void BEToUV_c(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1,
+                     const uint8_t *src2, int width, uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++) {
+        dstU[i]= src1[2*i];
+        dstV[i]= src2[2*i];
+    }
+}
 
-        (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
+static av_always_inline void nvXXtoUV_c(uint8_t *dst1, uint8_t *dst2,
+                                        const uint8_t *src, int width)
+{
+    int i;
+    for (i = 0; i < width; i++) {
+        dst1[i] = src[2*i+0];
+        dst2[i] = src[2*i+1];
     }
-    av_freep(&filter);
+}
 
-    /* try to reduce the filter-size (step1 find size and shift left) */
-    // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
-    minFilterSize= 0;
-    for (i=dstW-1; i>=0; i--) {
-        int min= filter2Size;
-        int j;
-        int64_t cutOff=0.0;
+static void nv12ToUV_c(uint8_t *dstU, uint8_t *dstV,
+                       const uint8_t *src1, const uint8_t *src2,
+                       int width, uint32_t *unused)
+{
+    nvXXtoUV_c(dstU, dstV, src1, width);
+}
 
-        /* get rid off near zero elements on the left by shifting left */
-        for (j=0; j<filter2Size; j++) {
-            int k;
-            cutOff += FFABS(filter2[i*filter2Size]);
+static void nv21ToUV_c(uint8_t *dstU, uint8_t *dstV,
+                       const uint8_t *src1, const uint8_t *src2,
+                       int width, uint32_t *unused)
+{
+    nvXXtoUV_c(dstV, dstU, src1, width);
+}
 
-            if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
+// FIXME Maybe dither instead.
+#define YUV_NBPS(depth, endianness, rfunc) \
+static void endianness ## depth ## ToUV_c(uint8_t *dstU, uint8_t *dstV, \
+                                          const uint8_t *_srcU, const uint8_t *_srcV, \
+                                          int width, uint32_t *unused) \
+{ \
+    int i; \
+    const uint16_t *srcU = (const uint16_t*)_srcU; \
+    const uint16_t *srcV = (const uint16_t*)_srcV; \
+    for (i = 0; i < width; i++) { \
+        dstU[i] = rfunc(&srcU[i])>>(depth-8); \
+        dstV[i] = rfunc(&srcV[i])>>(depth-8); \
+    } \
+} \
+\
+static void endianness ## depth ## ToY_c(uint8_t *dstY, const uint8_t *_srcY, \
+                                         int width, uint32_t *unused) \
+{ \
+    int i; \
+    const uint16_t *srcY = (const uint16_t*)_srcY; \
+    for (i = 0; i < width; i++) \
+        dstY[i] = rfunc(&srcY[i])>>(depth-8); \
+} \
+
+YUV_NBPS( 9, LE, AV_RL16)
+YUV_NBPS( 9, BE, AV_RB16)
+YUV_NBPS(10, LE, AV_RL16)
+YUV_NBPS(10, BE, AV_RB16)
+
+static void bgr24ToY_c(uint8_t *dst, const uint8_t *src,
+                       int width, uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++) {
+        int b= src[i*3+0];
+        int g= src[i*3+1];
+        int r= src[i*3+2];
 
-            /* preserve monotonicity because the core can't handle the filter otherwise */
-            if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
+        dst[i]= ((RY*r + GY*g + BY*b + (33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
+    }
+}
 
-            // move filter coefficients left
-            for (k=1; k<filter2Size; k++)
-                filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
-            filter2[i*filter2Size + k - 1]= 0;
-            (*filterPos)[i]++;
-        }
+static void bgr24ToUV_c(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1,
+                        const uint8_t *src2, int width, uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++) {
+        int b= src1[3*i + 0];
+        int g= src1[3*i + 1];
+        int r= src1[3*i + 2];
 
-        cutOff=0;
-        /* count near zeros on the right */
-        for (j=filter2Size-1; j>0; j--) {
-            cutOff += FFABS(filter2[i*filter2Size + j]);
+        dstU[i]= (RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT;
+        dstV[i]= (RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT;
+    }
+    assert(src1 == src2);
+}
 
-            if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
-            min--;
-        }
+static void bgr24ToUV_half_c(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1,
+                             const uint8_t *src2, int width, uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++) {
+        int b= src1[6*i + 0] + src1[6*i + 3];
+        int g= src1[6*i + 1] + src1[6*i + 4];
+        int r= src1[6*i + 2] + src1[6*i + 5];
 
-        if (min>minFilterSize) minFilterSize= min;
+        dstU[i]= (RU*r + GU*g + BU*b + (257<<RGB2YUV_SHIFT))>>(RGB2YUV_SHIFT+1);
+        dstV[i]= (RV*r + GV*g + BV*b + (257<<RGB2YUV_SHIFT))>>(RGB2YUV_SHIFT+1);
     }
+    assert(src1 == src2);
+}
 
-    if (flags & SWS_CPU_CAPS_ALTIVEC) {
-        // we can handle the special case 4,
-        // so we don't want to go to the full 8
-        if (minFilterSize < 5)
-            filterAlign = 4;
-
-        // We really don't want to waste our time
-        // doing useless computation, so fall back on
-        // the scalar C code for very small filters.
-        // Vectorizing is worth it only if you have a
-        // decent-sized vector.
-        if (minFilterSize < 3)
-            filterAlign = 1;
-    }
+static void rgb24ToY_c(uint8_t *dst, const uint8_t *src, int width,
+                       uint32_t *unused)
+{
+    int i;
+    for (i=0; i<width; i++) {
+        int r= src[i*3+0];
+        int g= src[i*3+1];
+        int b= src[i*3+2];
 
-    if (flags & SWS_CPU_CAPS_MMX) {
-        // special case for unscaled vertical filtering
-        if (minFilterSize == 1 && filterAlign == 2)
-            filterAlign= 1;
+        dst[i]= ((RY*r + GY*g + BY*b + (33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
     }
+}
 
-    assert(minFilterSize > 0);
-    filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
-    assert(filterSize > 0);
-    filter= av_malloc(filterSize*dstW*sizeof(*filter));
-    if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
-        goto fail;
-    *outFilterSize= filterSize;
-
-    if (flags&SWS_PRINT_INFO)
-        av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
-    /* try to reduce the filter-size (step2 reduce it) */
-    for (i=0; i<dstW; i++) {
-        int j;
+static void rgb24ToUV_c(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1,
+                        const uint8_t *src2, int width, uint32_t *unused)
+{
+    int i;
+    assert(src1==src2);
+    for (i=0; i<width; i++) {
+        int r= src1[3*i + 0];
+        int g= src1[3*i + 1];
+        int b= src1[3*i + 2];
 
-        for (j=0; j<filterSize; j++) {
-            if (j>=filter2Size) filter[i*filterSize + j]= 0;
-            else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
-            if((flags & SWS_BITEXACT) && j>=minFilterSize)
-                filter[i*filterSize + j]= 0;
-        }
+        dstU[i]= (RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT;
+        dstV[i]= (RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT;
     }
+}
+
+static void rgb24ToUV_half_c(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1,
+                             const uint8_t *src2, int width, uint32_t *unused)
+{
+    int i;
+    assert(src1==src2);
+    for (i=0; i<width; i++) {
+        int r= src1[6*i + 0] + src1[6*i + 3];
+        int g= src1[6*i + 1] + src1[6*i + 4];
+        int b= src1[6*i + 2] + src1[6*i + 5];
 
+        dstU[i]= (RU*r + GU*g + BU*b + (257<<RGB2YUV_SHIFT))>>(RGB2YUV_SHIFT+1);
+        dstV[i]= (RV*r + GV*g + BV*b + (257<<RGB2YUV_SHIFT))>>(RGB2YUV_SHIFT+1);
+    }
+}
 
-    //FIXME try to align filterPos if possible
 
-    //fix borders
+// bilinear / bicubic scaling
+static void hScale_c(int16_t *dst, int dstW, const uint8_t *src,
+                     int srcW, int xInc,
+                     const int16_t *filter, const int16_t *filterPos,
+                     int filterSize)
+{
+    int i;
     for (i=0; i<dstW; i++) {
         int j;
-        if ((*filterPos)[i] < 0) {
-            // move filter coefficients left to compensate for filterPos
-            for (j=1; j<filterSize; j++) {
-                int left= FFMAX(j + (*filterPos)[i], 0);
-                filter[i*filterSize + left] += filter[i*filterSize + j];
-                filter[i*filterSize + j]=0;
-            }
-            (*filterPos)[i]= 0;
+        int srcPos= filterPos[i];
+        int val=0;
+        for (j=0; j<filterSize; j++) {
+            val += ((int)src[srcPos + j])*filter[filterSize*i + j];
         }
+        //filter += hFilterSize;
+        dst[i] = FFMIN(val>>7, (1<<15)-1); // the cubic equation does overflow ...
+        //dst[i] = val>>7;
+    }
+}
 
-        if ((*filterPos)[i] + filterSize > srcW) {
-            int shift= (*filterPos)[i] + filterSize - srcW;
-            // move filter coefficients right to compensate for filterPos
-            for (j=filterSize-2; j>=0; j--) {
-                int right= FFMIN(j + shift, filterSize-1);
-                filter[i*filterSize +right] += filter[i*filterSize +j];
-                filter[i*filterSize +j]=0;
-            }
-            (*filterPos)[i]= srcW - filterSize;
-        }
-    }
-
-    // Note the +1 is for the MMX scaler which reads over the end
-    /* align at 16 for AltiVec (needed by hScale_altivec_real) */
-    FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
-
-    /* normalize & store in outFilter */
-    for (i=0; i<dstW; i++) {
-        int j;
-        int64_t error=0;
-        int64_t sum=0;
-
-        for (j=0; j<filterSize; j++) {
-            sum+= filter[i*filterSize + j];
-        }
-        sum= (sum + one/2)/ one;
-        for (j=0; j<*outFilterSize; j++) {
-            int64_t v= filter[i*filterSize + j] + error;
-            int intV= ROUNDED_DIV(v, sum);
-            (*outFilter)[i*(*outFilterSize) + j]= intV;
-            error= v - intV*sum;
-        }
-    }
-
-    (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
-    for (i=0; i<*outFilterSize; i++) {
-        int j= dstW*(*outFilterSize);
-        (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
-    }
-
-    ret=0;
-fail:
-    av_free(filter);
-    av_free(filter2);
-    return ret;
-}
-
-#ifdef COMPILE_MMX2
-static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
+//FIXME all pal and rgb srcFormats could do this convertion as well
+//FIXME all scalers more complex than bilinear could do half of this transform
+static void chrRangeToJpeg_c(uint16_t *dstU, uint16_t *dstV, int width)
 {
-    uint8_t *fragmentA;
-    x86_reg imm8OfPShufW1A;
-    x86_reg imm8OfPShufW2A;
-    x86_reg fragmentLengthA;
-    uint8_t *fragmentB;
-    x86_reg imm8OfPShufW1B;
-    x86_reg imm8OfPShufW2B;
-    x86_reg fragmentLengthB;
-    int fragmentPos;
-
-    int xpos, i;
-
-    // create an optimized horizontal scaling routine
-    /* This scaler is made of runtime-generated MMX2 code using specially
-     * tuned pshufw instructions. For every four output pixels, if four
-     * input pixels are enough for the fast bilinear scaling, then a chunk
-     * of fragmentB is used. If five input pixels are needed, then a chunk
-     * of fragmentA is used.
-     */
-
-    //code fragment
-
-    __asm__ volatile(
-        "jmp                         9f                 \n\t"
-    // Begin
-        "0:                                             \n\t"
-        "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
-        "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
-        "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"
-        "punpcklbw                %%mm7, %%mm1          \n\t"
-        "punpcklbw                %%mm7, %%mm0          \n\t"
-        "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"
-        "1:                                             \n\t"
-        "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
-        "2:                                             \n\t"
-        "psubw                    %%mm1, %%mm0          \n\t"
-        "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
-        "pmullw                   %%mm3, %%mm0          \n\t"
-        "psllw                       $7, %%mm1          \n\t"
-        "paddw                    %%mm1, %%mm0          \n\t"
-
-        "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
-
-        "add                         $8, %%"REG_a"      \n\t"
-    // End
-        "9:                                             \n\t"
-//        "int $3                                         \n\t"
-        "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
-        "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
-        "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
-        "dec                         %1                 \n\t"
-        "dec                         %2                 \n\t"
-        "sub                         %0, %1             \n\t"
-        "sub                         %0, %2             \n\t"
-        "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
-        "sub                         %0, %3             \n\t"
-
-
-        :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
-        "=r" (fragmentLengthA)
-    );
-
-    __asm__ volatile(
-        "jmp                         9f                 \n\t"
-    // Begin
-        "0:                                             \n\t"
-        "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
-        "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
-        "punpcklbw                %%mm7, %%mm0          \n\t"
-        "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"
-        "1:                                             \n\t"
-        "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
-        "2:                                             \n\t"
-        "psubw                    %%mm1, %%mm0          \n\t"
-        "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
-        "pmullw                   %%mm3, %%mm0          \n\t"
-        "psllw                       $7, %%mm1          \n\t"
-        "paddw                    %%mm1, %%mm0          \n\t"
-
-        "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
-
-        "add                         $8, %%"REG_a"      \n\t"
-    // End
-        "9:                                             \n\t"
-//        "int                       $3                   \n\t"
-        "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
-        "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
-        "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
-        "dec                         %1                 \n\t"
-        "dec                         %2                 \n\t"
-        "sub                         %0, %1             \n\t"
-        "sub                         %0, %2             \n\t"
-        "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
-        "sub                         %0, %3             \n\t"
-
-
-        :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
-        "=r" (fragmentLengthB)
-    );
-
-    xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
-    fragmentPos=0;
-
-    for (i=0; i<dstW/numSplits; i++) {
-        int xx=xpos>>16;
-
-        if ((i&3) == 0) {
-            int a=0;
-            int b=((xpos+xInc)>>16) - xx;
-            int c=((xpos+xInc*2)>>16) - xx;
-            int d=((xpos+xInc*3)>>16) - xx;
-            int inc                = (d+1<4);
-            uint8_t *fragment      = (d+1<4) ? fragmentB       : fragmentA;
-            x86_reg imm8OfPShufW1  = (d+1<4) ? imm8OfPShufW1B  : imm8OfPShufW1A;
-            x86_reg imm8OfPShufW2  = (d+1<4) ? imm8OfPShufW2B  : imm8OfPShufW2A;
-            x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
-            int maxShift= 3-(d+inc);
-            int shift=0;
-
-            if (filterCode) {
-                filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
-                filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
-                filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
-                filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
-                filterPos[i/2]= xx;
-
-                memcpy(filterCode + fragmentPos, fragment, fragmentLength);
-
-                filterCode[fragmentPos + imm8OfPShufW1]=
-                    (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
-                filterCode[fragmentPos + imm8OfPShufW2]=
-                    a | (b<<2) | (c<<4) | (d<<6);
-
-                if (i+4-inc>=dstW) shift=maxShift; //avoid overread
-                else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
-
-                if (shift && i>=shift) {
-                    filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
-                    filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
-                    filterPos[i/2]-=shift;
-                }
-            }
-
-            fragmentPos+= fragmentLength;
-
-            if (filterCode)
-                filterCode[fragmentPos]= RET;
-        }
-        xpos+=xInc;
-    }
-    if (filterCode)
-        filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
-
-    return fragmentPos + 1;
-}
-#endif /* COMPILE_MMX2 */
-
-static SwsFunc getSwsFunc(SwsContext *c)
-{
-#if CONFIG_RUNTIME_CPUDETECT
-    int flags = c->flags;
-
-#if ARCH_X86 && CONFIG_GPL
-    // ordered per speed fastest first
-    if (flags & SWS_CPU_CAPS_MMX2) {
-        sws_init_swScale_MMX2(c);
-        return swScale_MMX2;
-    } else if (flags & SWS_CPU_CAPS_3DNOW) {
-        sws_init_swScale_3DNow(c);
-        return swScale_3DNow;
-    } else if (flags & SWS_CPU_CAPS_MMX) {
-        sws_init_swScale_MMX(c);
-        return swScale_MMX;
-    } else {
-        sws_init_swScale_C(c);
-        return swScale_C;
-    }
-
-#else
-#if ARCH_PPC
-    if (flags & SWS_CPU_CAPS_ALTIVEC) {
-        sws_init_swScale_altivec(c);
-        return swScale_altivec;
-    } else {
-        sws_init_swScale_C(c);
-        return swScale_C;
+    int i;
+    for (i = 0; i < width; i++) {
+        dstU[i] = (FFMIN(dstU[i],30775)*4663 - 9289992)>>12; //-264
+        dstV[i] = (FFMIN(dstV[i],30775)*4663 - 9289992)>>12; //-264
     }
-#endif
-    sws_init_swScale_C(c);
-    return swScale_C;
-#endif /* ARCH_X86 && CONFIG_GPL */
-#else //CONFIG_RUNTIME_CPUDETECT
-#if   COMPILE_TEMPLATE_MMX2
-    sws_init_swScale_MMX2(c);
-    return swScale_MMX2;
-#elif COMPILE_TEMPLATE_AMD3DNOW
-    sws_init_swScale_3DNow(c);
-    return swScale_3DNow;
-#elif COMPILE_TEMPLATE_MMX
-    sws_init_swScale_MMX(c);
-    return swScale_MMX;
-#elif COMPILE_TEMPLATE_ALTIVEC
-    sws_init_swScale_altivec(c);
-    return swScale_altivec;
-#else
-    sws_init_swScale_C(c);
-    return swScale_C;
-#endif
-#endif //!CONFIG_RUNTIME_CPUDETECT
 }
-
-static int PlanarToNV12Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                               int srcSliceH, uint8_t* dstParam[], int dstStride[])
+static void chrRangeFromJpeg_c(uint16_t *dstU, uint16_t *dstV, int width)
 {
-    uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
-    /* Copy Y plane */
-    if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
-        memcpy(dst, src[0], srcSliceH*dstStride[0]);
-    else {
-        int i;
-        const uint8_t *srcPtr= src[0];
-        uint8_t *dstPtr= dst;
-        for (i=0; i<srcSliceH; i++) {
-            memcpy(dstPtr, srcPtr, c->srcW);
-            srcPtr+= srcStride[0];
-            dstPtr+= dstStride[0];
-        }
+    int i;
+    for (i = 0; i < width; i++) {
+        dstU[i] = (dstU[i]*1799 + 4081085)>>11; //1469
+        dstV[i] = (dstV[i]*1799 + 4081085)>>11; //1469
     }
-    dst = dstParam[1] + dstStride[1]*srcSliceY/2;
-    if (c->dstFormat == PIX_FMT_NV12)
-        interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
-    else
-        interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
-
-    return srcSliceH;
-}
-
-static int PlanarToYuy2Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                               int srcSliceH, uint8_t* dstParam[], int dstStride[])
-{
-    uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
-
-    yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
-
-    return srcSliceH;
 }
-
-static int PlanarToUyvyWrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                               int srcSliceH, uint8_t* dstParam[], int dstStride[])
+static void lumRangeToJpeg_c(uint16_t *dst, int width)
 {
-    uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
-
-    yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
-
-    return srcSliceH;
-}
-
-static int YUV422PToYuy2Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                                int srcSliceH, uint8_t* dstParam[], int dstStride[])
-{
-    uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
-
-    yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
-
-    return srcSliceH;
-}
-
-static int YUV422PToUyvyWrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                                int srcSliceH, uint8_t* dstParam[], int dstStride[])
-{
-    uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
-
-    yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
-
-    return srcSliceH;
+    int i;
+    for (i = 0; i < width; i++)
+        dst[i] = (FFMIN(dst[i],30189)*19077 - 39057361)>>14;
 }
-
-static int YUYV2YUV420Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                               int srcSliceH, uint8_t* dstParam[], int dstStride[])
+static void lumRangeFromJpeg_c(uint16_t *dst, int width)
 {
-    uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
-    uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
-    uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
-
-    yuyvtoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
-
-    if (dstParam[3])
-        fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
-
-    return srcSliceH;
+    int i;
+    for (i = 0; i < width; i++)
+        dst[i] = (dst[i]*14071 + 33561947)>>14;
 }
 
-static int YUYV2YUV422Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                               int srcSliceH, uint8_t* dstParam[], int dstStride[])
+static void hyscale_fast_c(SwsContext *c, int16_t *dst, int dstWidth,
+                           const uint8_t *src, int srcW, int xInc)
 {
-    uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
-    uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
-    uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
-
-    yuyvtoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
-
-    return srcSliceH;
+    int i;
+    unsigned int xpos=0;
+    for (i=0;i<dstWidth;i++) {
+        register unsigned int xx=xpos>>16;
+        register unsigned int xalpha=(xpos&0xFFFF)>>9;
+        dst[i]= (src[xx]<<7) + (src[xx+1] - src[xx])*xalpha;
+        xpos+=xInc;
+    }
 }
 
-static int UYVY2YUV420Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                               int srcSliceH, uint8_t* dstParam[], int dstStride[])
+// *** horizontal scale Y line to temp buffer
+static inline void hyscale(SwsContext *c, uint16_t *dst, int dstWidth,
+                           const uint8_t *src, int srcW, int xInc,
+                           const int16_t *hLumFilter,
+                           const int16_t *hLumFilterPos, int hLumFilterSize,
+                           uint8_t *formatConvBuffer,
+                           uint32_t *pal, int isAlpha)
 {
-    uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
-    uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
-    uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
+    void (*toYV12)(uint8_t *, const uint8_t *, int, uint32_t *) = isAlpha ? c->alpToYV12 : c->lumToYV12;
+    void (*convertRange)(uint16_t *, int) = isAlpha ? NULL : c->lumConvertRange;
 
-    uyvytoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
+    if (toYV12) {
+        toYV12(formatConvBuffer, src, srcW, pal);
+        src= formatConvBuffer;
+    }
 
-    if (dstParam[3])
-        fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
+    if (!c->hyscale_fast) {
+        c->hScale(dst, dstWidth, src, srcW, xInc, hLumFilter, hLumFilterPos, hLumFilterSize);
+    } else { // fast bilinear upscale / crap downscale
+        c->hyscale_fast(c, dst, dstWidth, src, srcW, xInc);
+    }
 
-    return srcSliceH;
+    if (convertRange)
+        convertRange(dst, dstWidth);
 }
 
-static int UYVY2YUV422Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                               int srcSliceH, uint8_t* dstParam[], int dstStride[])
+static void hcscale_fast_c(SwsContext *c, int16_t *dst1, int16_t *dst2,
+                           int dstWidth, const uint8_t *src1,
+                           const uint8_t *src2, int srcW, int xInc)
 {
-    uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
-    uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
-    uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
-
-    uyvytoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
-
-    return srcSliceH;
+    int i;
+    unsigned int xpos=0;
+    for (i=0;i<dstWidth;i++) {
+        register unsigned int xx=xpos>>16;
+        register unsigned int xalpha=(xpos&0xFFFF)>>9;
+        dst1[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
+        dst2[i]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
+        xpos+=xInc;
+    }
 }
 
-static int pal2rgbWrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                          int srcSliceH, uint8_t* dst[], int dstStride[])
+static inline void hcscale(SwsContext *c, uint16_t *dst1, uint16_t *dst2, int dstWidth,
+                           const uint8_t *src1, const uint8_t *src2,
+                           int srcW, int xInc, const int16_t *hChrFilter,
+                           const int16_t *hChrFilterPos, int hChrFilterSize,
+                           uint8_t *formatConvBuffer, uint32_t *pal)
 {
-    const enum PixelFormat srcFormat= c->srcFormat;
-    const enum PixelFormat dstFormat= c->dstFormat;
-    void (*conv)(const uint8_t *src, uint8_t *dst, long num_pixels,
-                 const uint8_t *palette)=NULL;
-    int i;
-    uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
-    const uint8_t *srcPtr= src[0];
-
-    if (!usePal(srcFormat))
-        av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
-               sws_format_name(srcFormat), sws_format_name(dstFormat));
-
-    switch(dstFormat) {
-    case PIX_FMT_RGB32  : conv = palette8topacked32; break;
-    case PIX_FMT_BGR32  : conv = palette8topacked32; break;
-    case PIX_FMT_BGR32_1: conv = palette8topacked32; break;
-    case PIX_FMT_RGB32_1: conv = palette8topacked32; break;
-    case PIX_FMT_RGB24  : conv = palette8topacked24; break;
-    case PIX_FMT_BGR24  : conv = palette8topacked24; break;
-    default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
-                    sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
+    if (c->chrToYV12) {
+        uint8_t *buf2 = formatConvBuffer + FFALIGN(srcW, 16);
+        c->chrToYV12(formatConvBuffer, buf2, src1, src2, srcW, pal);
+        src1= formatConvBuffer;
+        src2= buf2;
     }
 
-
-    for (i=0; i<srcSliceH; i++) {
-        conv(srcPtr, dstPtr, c->srcW, (uint8_t *) c->pal_rgb);
-        srcPtr+= srcStride[0];
-        dstPtr+= dstStride[0];
+    if (!c->hcscale_fast) {
+        c->hScale(dst1, dstWidth, src1, srcW, xInc, hChrFilter, hChrFilterPos, hChrFilterSize);
+        c->hScale(dst2, dstWidth, src2, srcW, xInc, hChrFilter, hChrFilterPos, hChrFilterSize);
+    } else { // fast bilinear upscale / crap downscale
+        c->hcscale_fast(c, dst1, dst2, dstWidth, src1, src2, srcW, xInc);
     }
 
-    return srcSliceH;
+    if (c->chrConvertRange)
+        c->chrConvertRange(dst1, dst2, dstWidth);
 }
 
-/* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
-static int rgb2rgbWrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                          int srcSliceH, uint8_t* dst[], int dstStride[])
+#define DEBUG_SWSCALE_BUFFERS 0
+#define DEBUG_BUFFERS(...) if (DEBUG_SWSCALE_BUFFERS) av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
+
+static int swScale(SwsContext *c, const uint8_t* src[],
+                   int srcStride[], int srcSliceY,
+                   int srcSliceH, uint8_t* dst[], int dstStride[])
 {
-    const enum PixelFormat srcFormat= c->srcFormat;
+    /* load a few things into local vars to make the code more readable? and faster */
+    const int srcW= c->srcW;
+    const int dstW= c->dstW;
+    const int dstH= c->dstH;
+    const int chrDstW= c->chrDstW;
+    const int chrSrcW= c->chrSrcW;
+    const int lumXInc= c->lumXInc;
+    const int chrXInc= c->chrXInc;
     const enum PixelFormat dstFormat= c->dstFormat;
-    const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
-    const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
-    const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
-    const int dstId= fmt_depth(dstFormat) >> 2;
-    void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
-
-    /* BGR -> BGR */
-    if (  (isBGR(srcFormat) && isBGR(dstFormat))
-       || (isRGB(srcFormat) && isRGB(dstFormat))) {
-        switch(srcId | (dstId<<4)) {
-        case 0x34: conv= rgb16to15; break;
-        case 0x36: conv= rgb24to15; break;
-        case 0x38: conv= rgb32to15; break;
-        case 0x43: conv= rgb15to16; break;
-        case 0x46: conv= rgb24to16; break;
-        case 0x48: conv= rgb32to16; break;
-        case 0x63: conv= rgb15to24; break;
-        case 0x64: conv= rgb16to24; break;
-        case 0x68: conv= rgb32to24; break;
-        case 0x83: conv= rgb15to32; break;
-        case 0x84: conv= rgb16to32; break;
-        case 0x86: conv= rgb24to32; break;
-        default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
-                        sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
-        }
-    } else if (  (isBGR(srcFormat) && isRGB(dstFormat))
-             || (isRGB(srcFormat) && isBGR(dstFormat))) {
-        switch(srcId | (dstId<<4)) {
-        case 0x33: conv= rgb15tobgr15; break;
-        case 0x34: conv= rgb16tobgr15; break;
-        case 0x36: conv= rgb24tobgr15; break;
-        case 0x38: conv= rgb32tobgr15; break;
-        case 0x43: conv= rgb15tobgr16; break;
-        case 0x44: conv= rgb16tobgr16; break;
-        case 0x46: conv= rgb24tobgr16; break;
-        case 0x48: conv= rgb32tobgr16; break;
-        case 0x63: conv= rgb15tobgr24; break;
-        case 0x64: conv= rgb16tobgr24; break;
-        case 0x66: conv= rgb24tobgr24; break;
-        case 0x68: conv= rgb32tobgr24; break;
-        case 0x83: conv= rgb15tobgr32; break;
-        case 0x84: conv= rgb16tobgr32; break;
-        case 0x86: conv= rgb24tobgr32; break;
-        case 0x88: conv= rgb32tobgr32; break;
-        default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
-                        sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
+    const int flags= c->flags;
+    int16_t *vLumFilterPos= c->vLumFilterPos;
+    int16_t *vChrFilterPos= c->vChrFilterPos;
+    int16_t *hLumFilterPos= c->hLumFilterPos;
+    int16_t *hChrFilterPos= c->hChrFilterPos;
+    int16_t *vLumFilter= c->vLumFilter;
+    int16_t *vChrFilter= c->vChrFilter;
+    int16_t *hLumFilter= c->hLumFilter;
+    int16_t *hChrFilter= c->hChrFilter;
+    int32_t *lumMmxFilter= c->lumMmxFilter;
+    int32_t *chrMmxFilter= c->chrMmxFilter;
+    int32_t av_unused *alpMmxFilter= c->alpMmxFilter;
+    const int vLumFilterSize= c->vLumFilterSize;
+    const int vChrFilterSize= c->vChrFilterSize;
+    const int hLumFilterSize= c->hLumFilterSize;
+    const int hChrFilterSize= c->hChrFilterSize;
+    int16_t **lumPixBuf= c->lumPixBuf;
+    int16_t **chrUPixBuf= c->chrUPixBuf;
+    int16_t **chrVPixBuf= c->chrVPixBuf;
+    int16_t **alpPixBuf= c->alpPixBuf;
+    const int vLumBufSize= c->vLumBufSize;
+    const int vChrBufSize= c->vChrBufSize;
+    uint8_t *formatConvBuffer= c->formatConvBuffer;
+    const int chrSrcSliceY= srcSliceY >> c->chrSrcVSubSample;
+    const int chrSrcSliceH= -((-srcSliceH) >> c->chrSrcVSubSample);
+    int lastDstY;
+    uint32_t *pal=c->pal_yuv;
+
+    /* vars which will change and which we need to store back in the context */
+    int dstY= c->dstY;
+    int lumBufIndex= c->lumBufIndex;
+    int chrBufIndex= c->chrBufIndex;
+    int lastInLumBuf= c->lastInLumBuf;
+    int lastInChrBuf= c->lastInChrBuf;
+
+    if (isPacked(c->srcFormat)) {
+        src[0]=
+        src[1]=
+        src[2]=
+        src[3]= src[0];
+        srcStride[0]=
+        srcStride[1]=
+        srcStride[2]=
+        srcStride[3]= srcStride[0];
+    }
+    srcStride[1]<<= c->vChrDrop;
+    srcStride[2]<<= c->vChrDrop;
+
+    DEBUG_BUFFERS("swScale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
+                  src[0], srcStride[0], src[1], srcStride[1], src[2], srcStride[2], src[3], srcStride[3],
+                  dst[0], dstStride[0], dst[1], dstStride[1], dst[2], dstStride[2], dst[3], dstStride[3]);
+    DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
+                   srcSliceY,    srcSliceH,    dstY,    dstH);
+    DEBUG_BUFFERS("vLumFilterSize: %d vLumBufSize: %d vChrFilterSize: %d vChrBufSize: %d\n",
+                   vLumFilterSize,    vLumBufSize,    vChrFilterSize,    vChrBufSize);
+
+    if (dstStride[0]%8 !=0 || dstStride[1]%8 !=0 || dstStride[2]%8 !=0 || dstStride[3]%8 != 0) {
+        static int warnedAlready=0; //FIXME move this into the context perhaps
+        if (flags & SWS_PRINT_INFO && !warnedAlready) {
+            av_log(c, AV_LOG_WARNING, "Warning: dstStride is not aligned!\n"
+                   "         ->cannot do aligned memory accesses anymore\n");
+            warnedAlready=1;
         }
-    } else {
-        av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
-               sws_format_name(srcFormat), sws_format_name(dstFormat));
     }
 
-    if(conv) {
-        const uint8_t *srcPtr= src[0];
-        if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
-            srcPtr += ALT32_CORR;
-
-        if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
-            conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
-        else {
-            int i;
-            uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
-
-            for (i=0; i<srcSliceH; i++) {
-                conv(srcPtr, dstPtr, c->srcW*srcBpp);
-                srcPtr+= srcStride[0];
-                dstPtr+= dstStride[0];
-            }
+    /* Note the user might start scaling the picture in the middle so this
+       will not get executed. This is not really intended but works
+       currently, so people might do it. */
+    if (srcSliceY ==0) {
+        lumBufIndex=-1;
+        chrBufIndex=-1;
+        dstY=0;
+        lastInLumBuf= -1;
+        lastInChrBuf= -1;
+    }
+
+    lastDstY= dstY;
+
+    for (;dstY < dstH; dstY++) {
+        unsigned char *dest =dst[0]+dstStride[0]*dstY;
+        const int chrDstY= dstY>>c->chrDstVSubSample;
+        unsigned char *uDest=dst[1]+dstStride[1]*chrDstY;
+        unsigned char *vDest=dst[2]+dstStride[2]*chrDstY;
+        unsigned char *aDest=(CONFIG_SWSCALE_ALPHA && alpPixBuf) ? dst[3]+dstStride[3]*dstY : NULL;
+
+        const int firstLumSrcY= vLumFilterPos[dstY]; //First line needed as input
+        const int firstLumSrcY2= vLumFilterPos[FFMIN(dstY | ((1<<c->chrDstVSubSample) - 1), dstH-1)];
+        const int firstChrSrcY= vChrFilterPos[chrDstY]; //First line needed as input
+        int lastLumSrcY= firstLumSrcY + vLumFilterSize -1; // Last line needed as input
+        int lastLumSrcY2=firstLumSrcY2+ vLumFilterSize -1; // Last line needed as input
+        int lastChrSrcY= firstChrSrcY + vChrFilterSize -1; // Last line needed as input
+        int enough_lines;
+
+        //handle holes (FAST_BILINEAR & weird filters)
+        if (firstLumSrcY > lastInLumBuf) lastInLumBuf= firstLumSrcY-1;
+        if (firstChrSrcY > lastInChrBuf) lastInChrBuf= firstChrSrcY-1;
+        assert(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);
+        assert(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);
+
+        DEBUG_BUFFERS("dstY: %d\n", dstY);
+        DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
+                         firstLumSrcY,    lastLumSrcY,    lastInLumBuf);
+        DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
+                         firstChrSrcY,    lastChrSrcY,    lastInChrBuf);
+
+        // Do we have enough lines in this slice to output the dstY line
+        enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH && lastChrSrcY < -((-srcSliceY - srcSliceH)>>c->chrSrcVSubSample);
+
+        if (!enough_lines) {
+            lastLumSrcY = srcSliceY + srcSliceH - 1;
+            lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
+            DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
+                                            lastLumSrcY, lastChrSrcY);
         }
-    }
-    return srcSliceH;
-}
-
-static int bgr24toyv12Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                              int srcSliceH, uint8_t* dst[], int dstStride[])
-{
 
-    rgb24toyv12(
-        src[0],
-        dst[0]+ srcSliceY    *dstStride[0],
-        dst[1]+(srcSliceY>>1)*dstStride[1],
-        dst[2]+(srcSliceY>>1)*dstStride[2],
-        c->srcW, srcSliceH,
-        dstStride[0], dstStride[1], srcStride[0]);
-    if (dst[3])
-        fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
-    return srcSliceH;
-}
-
-static int yvu9toyv12Wrapper(SwsContext *c, const const uint8_t* src[], int srcStride[], int srcSliceY,
-                             int srcSliceH, uint8_t* dst[], int dstStride[])
-{
-    int i;
-
-    /* copy Y */
-    if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
-        memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
-    else {
-        const uint8_t *srcPtr= src[0];
-        uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
-
-        for (i=0; i<srcSliceH; i++) {
-            memcpy(dstPtr, srcPtr, c->srcW);
-            srcPtr+= srcStride[0];
-            dstPtr+= dstStride[0];
+        //Do horizontal scaling
+        while(lastInLumBuf < lastLumSrcY) {
+            const uint8_t *src1= src[0]+(lastInLumBuf + 1 - srcSliceY)*srcStride[0];
+            const uint8_t *src2= src[3]+(lastInLumBuf + 1 - srcSliceY)*srcStride[3];
+            lumBufIndex++;
+            assert(lumBufIndex < 2*vLumBufSize);
+            assert(lastInLumBuf + 1 - srcSliceY < srcSliceH);
+            assert(lastInLumBuf + 1 - srcSliceY >= 0);
+            hyscale(c, lumPixBuf[ lumBufIndex ], dstW, src1, srcW, lumXInc,
+                    hLumFilter, hLumFilterPos, hLumFilterSize,
+                    formatConvBuffer,
+                    pal, 0);
+            if (CONFIG_SWSCALE_ALPHA && alpPixBuf)
+                hyscale(c, alpPixBuf[ lumBufIndex ], dstW, src2, srcW,
+                        lumXInc, hLumFilter, hLumFilterPos, hLumFilterSize,
+                        formatConvBuffer,
+                        pal, 1);
+            lastInLumBuf++;
+            DEBUG_BUFFERS("\t\tlumBufIndex %d: lastInLumBuf: %d\n",
+                               lumBufIndex,    lastInLumBuf);
         }
-    }
-
-    if (c->dstFormat==PIX_FMT_YUV420P || c->dstFormat==PIX_FMT_YUVA420P) {
-        planar2x(src[1], dst[1] + dstStride[1]*(srcSliceY >> 1), c->chrSrcW,
-                 srcSliceH >> 2, srcStride[1], dstStride[1]);
-        planar2x(src[2], dst[2] + dstStride[2]*(srcSliceY >> 1), c->chrSrcW,
-                 srcSliceH >> 2, srcStride[2], dstStride[2]);
-    } else {
-        planar2x(src[1], dst[2] + dstStride[2]*(srcSliceY >> 1), c->chrSrcW,
-                 srcSliceH >> 2, srcStride[1], dstStride[2]);
-        planar2x(src[2], dst[1] + dstStride[1]*(srcSliceY >> 1), c->chrSrcW,
-                 srcSliceH >> 2, srcStride[2], dstStride[1]);
-    }
-    if (dst[3])
-        fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
-    return srcSliceH;
-}
-
-/* unscaled copy like stuff (assumes nearly identical formats) */
-static int packedCopy(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                      int srcSliceH, uint8_t* dst[], int dstStride[])
-{
-    if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
-        memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
-    else {
-        int i;
-        const uint8_t *srcPtr= src[0];
-        uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
-        int length=0;
-
-        /* universal length finder */
-        while(length+c->srcW <= FFABS(dstStride[0])
-           && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
-        assert(length!=0);
-
-        for (i=0; i<srcSliceH; i++) {
-            memcpy(dstPtr, srcPtr, length);
-            srcPtr+= srcStride[0];
-            dstPtr+= dstStride[0];
+        while(lastInChrBuf < lastChrSrcY) {
+            const uint8_t *src1= src[1]+(lastInChrBuf + 1 - chrSrcSliceY)*srcStride[1];
+            const uint8_t *src2= src[2]+(lastInChrBuf + 1 - chrSrcSliceY)*srcStride[2];
+            chrBufIndex++;
+            assert(chrBufIndex < 2*vChrBufSize);
+            assert(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));
+            assert(lastInChrBuf + 1 - chrSrcSliceY >= 0);
+            //FIXME replace parameters through context struct (some at least)
+
+            if (c->needs_hcscale)
+                hcscale(c, chrUPixBuf[chrBufIndex], chrVPixBuf[chrBufIndex],
+                          chrDstW, src1, src2, chrSrcW, chrXInc,
+                          hChrFilter, hChrFilterPos, hChrFilterSize,
+                          formatConvBuffer, pal);
+            lastInChrBuf++;
+            DEBUG_BUFFERS("\t\tchrBufIndex %d: lastInChrBuf: %d\n",
+                               chrBufIndex,    lastInChrBuf);
         }
-    }
-    return srcSliceH;
-}
-
-static int planarCopy(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                      int srcSliceH, uint8_t* dst[], int dstStride[])
-{
-    int plane, i, j;
-    for (plane=0; plane<4; plane++) {
-        int length= (plane==0 || plane==3) ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
-        int y=      (plane==0 || plane==3) ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
-        int height= (plane==0 || plane==3) ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
-        const uint8_t *srcPtr= src[plane];
-        uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
-
-        if (!dst[plane]) continue;
-        // ignore palette for GRAY8
-        if (plane == 1 && !dst[2]) continue;
-        if (!src[plane] || (plane == 1 && !src[2])) {
-            if(is16BPS(c->dstFormat))
-                length*=2;
-            fillPlane(dst[plane], dstStride[plane], length, height, y, (plane==3) ? 255 : 128);
-        } else {
-            if(is16BPS(c->srcFormat) && !is16BPS(c->dstFormat)) {
-                if (!isBE(c->srcFormat)) srcPtr++;
-                for (i=0; i<height; i++) {
-                    for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
-                    srcPtr+= srcStride[plane];
-                    dstPtr+= dstStride[plane];
+        //wrap buf index around to stay inside the ring buffer
+        if (lumBufIndex >= vLumBufSize) lumBufIndex-= vLumBufSize;
+        if (chrBufIndex >= vChrBufSize) chrBufIndex-= vChrBufSize;
+        if (!enough_lines)
+            break; //we can't output a dstY line so let's try with the next slice
+
+#if HAVE_MMX
+        updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex, lastInLumBuf, lastInChrBuf);
+#endif
+        if (dstY < dstH-2) {
+            const int16_t **lumSrcPtr= (const int16_t **) lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
+            const int16_t **chrUSrcPtr= (const int16_t **) chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
+            const int16_t **chrVSrcPtr= (const int16_t **) chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
+            const int16_t **alpSrcPtr= (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? (const int16_t **) alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
+            if (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21) {
+                const int chrSkipMask= (1<<c->chrDstVSubSample)-1;
+                if (dstY&chrSkipMask) uDest= NULL; //FIXME split functions in lumi / chromi
+                c->yuv2nv12X(c,
+                             vLumFilter+dstY*vLumFilterSize   , lumSrcPtr, vLumFilterSize,
+                             vChrFilter+chrDstY*vChrFilterSize, chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
+                             dest, uDest, dstW, chrDstW, dstFormat);
+            } else if (isPlanarYUV(dstFormat) || dstFormat==PIX_FMT_GRAY8) { //YV12 like
+                const int chrSkipMask= (1<<c->chrDstVSubSample)-1;
+                if ((dstY&chrSkipMask) || isGray(dstFormat)) uDest=vDest= NULL; //FIXME split functions in lumi / chromi
+                if (c->yuv2yuv1 && vLumFilterSize == 1 && vChrFilterSize == 1) { // unscaled YV12
+                    const int16_t *lumBuf = lumSrcPtr[0];
+                    const int16_t *chrUBuf= chrUSrcPtr[0];
+                    const int16_t *chrVBuf= chrVSrcPtr[0];
+                    const int16_t *alpBuf= (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? alpSrcPtr[0] : NULL;
+                    c->yuv2yuv1(c, lumBuf, chrUBuf, chrVBuf, alpBuf, dest,
+                                uDest, vDest, aDest, dstW, chrDstW);
+                } else { //General YV12
+                    c->yuv2yuvX(c,
+                                vLumFilter+dstY*vLumFilterSize   , lumSrcPtr, vLumFilterSize,
+                                vChrFilter+chrDstY*vChrFilterSize, chrUSrcPtr,
+                                chrVSrcPtr, vChrFilterSize,
+                                alpSrcPtr, dest, uDest, vDest, aDest, dstW, chrDstW);
                 }
-            } else if(!is16BPS(c->srcFormat) && is16BPS(c->dstFormat)) {
-                for (i=0; i<height; i++) {
-                    for (j=0; j<length; j++) {
-                        dstPtr[ j<<1   ] = srcPtr[j];
-                        dstPtr[(j<<1)+1] = srcPtr[j];
+            } else {
+                assert(lumSrcPtr  + vLumFilterSize - 1 < lumPixBuf  + vLumBufSize*2);
+                assert(chrUSrcPtr + vChrFilterSize - 1 < chrUPixBuf + vChrBufSize*2);
+                if (vLumFilterSize == 1 && vChrFilterSize == 2) { //unscaled RGB
+                    int chrAlpha= vChrFilter[2*dstY+1];
+                    if(flags & SWS_FULL_CHR_H_INT) {
+                        yuv2rgbX_c_full(c, //FIXME write a packed1_full function
+                                        vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize,
+                                        vChrFilter+dstY*vChrFilterSize, chrUSrcPtr,
+                                        chrVSrcPtr, vChrFilterSize,
+                                        alpSrcPtr, dest, dstW, dstY);
+                    } else {
+                        c->yuv2packed1(c, *lumSrcPtr, *chrUSrcPtr, *(chrUSrcPtr+1),
+                                       *chrVSrcPtr, *(chrVSrcPtr+1),
+                                       alpPixBuf ? *alpSrcPtr : NULL,
+                                       dest, dstW, chrAlpha, dstFormat, flags, dstY);
+                    }
+                } else if (vLumFilterSize == 2 && vChrFilterSize == 2) { //bilinear upscale RGB
+                    int lumAlpha= vLumFilter[2*dstY+1];
+                    int chrAlpha= vChrFilter[2*dstY+1];
+                    lumMmxFilter[2]=
+                    lumMmxFilter[3]= vLumFilter[2*dstY   ]*0x10001;
+                    chrMmxFilter[2]=
+                    chrMmxFilter[3]= vChrFilter[2*chrDstY]*0x10001;
+                    if(flags & SWS_FULL_CHR_H_INT) {
+                        yuv2rgbX_c_full(c, //FIXME write a packed2_full function
+                                        vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize,
+                                        vChrFilter+dstY*vChrFilterSize, chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
+                                        alpSrcPtr, dest, dstW, dstY);
+                    } else {
+                        c->yuv2packed2(c, *lumSrcPtr, *(lumSrcPtr+1), *chrUSrcPtr, *(chrUSrcPtr+1),
+                                       *chrVSrcPtr, *(chrVSrcPtr+1),
+                                       alpPixBuf ? *alpSrcPtr : NULL, alpPixBuf ? *(alpSrcPtr+1) : NULL,
+                                       dest, dstW, lumAlpha, chrAlpha, dstY);
+                    }
+                } else { //general RGB
+                    if(flags & SWS_FULL_CHR_H_INT) {
+                        yuv2rgbX_c_full(c,
+                                        vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize,
+                                        vChrFilter+dstY*vChrFilterSize, chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
+                                        alpSrcPtr, dest, dstW, dstY);
+                    } else {
+                        c->yuv2packedX(c,
+                                       vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize,
+                                       vChrFilter+dstY*vChrFilterSize, chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
+                                       alpSrcPtr, dest, dstW, dstY);
                     }
-                    srcPtr+= srcStride[plane];
-                    dstPtr+= dstStride[plane];
                 }
-            } else if(is16BPS(c->srcFormat) && is16BPS(c->dstFormat)
-                  && isBE(c->srcFormat) != isBE(c->dstFormat)) {
-
-                for (i=0; i<height; i++) {
-                    for (j=0; j<length; j++)
-                        ((uint16_t*)dstPtr)[j] = bswap_16(((uint16_t*)srcPtr)[j]);
-                    srcPtr+= srcStride[plane];
-                    dstPtr+= dstStride[plane];
+            }
+        } else { // hmm looks like we can't use MMX here without overwriting this array's tail
+            const int16_t **lumSrcPtr= (const int16_t **)lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
+            const int16_t **chrUSrcPtr= (const int16_t **)chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
+            const int16_t **chrVSrcPtr= (const int16_t **)chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
+            const int16_t **alpSrcPtr= (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? (const int16_t **)alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
+            if (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21) {
+                const int chrSkipMask= (1<<c->chrDstVSubSample)-1;
+                if (dstY&chrSkipMask) uDest= NULL; //FIXME split functions in lumi / chromi
+                yuv2nv12X_c(c, vLumFilter+dstY*vLumFilterSize,
+                            lumSrcPtr, vLumFilterSize,
+                            vChrFilter+chrDstY*vChrFilterSize,
+                            chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
+                            dest, uDest, dstW, chrDstW, dstFormat);
+            } else if (isPlanarYUV(dstFormat) || dstFormat==PIX_FMT_GRAY8) { //YV12
+                const int chrSkipMask= (1<<c->chrDstVSubSample)-1;
+                if ((dstY&chrSkipMask) || isGray(dstFormat)) uDest=vDest= NULL; //FIXME split functions in lumi / chromi
+                if (is16BPS(dstFormat) || is9_OR_10BPS(dstFormat)) {
+                    yuv2yuvX16_c(c, vLumFilter+dstY*vLumFilterSize   , lumSrcPtr, vLumFilterSize,
+                                 vChrFilter+chrDstY*vChrFilterSize, chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
+                                 alpSrcPtr, dest, uDest, vDest, aDest, dstW, chrDstW,
+                                 dstFormat);
+                } else {
+                    yuv2yuvX_c(c, vLumFilter+dstY*vLumFilterSize,
+                               lumSrcPtr, vLumFilterSize,
+                               vChrFilter+chrDstY*vChrFilterSize,
+                               chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
+                               alpSrcPtr, dest, uDest, vDest, aDest,
+                               dstW, chrDstW);
                 }
-            } else if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
-                memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
-            else {
-                if(is16BPS(c->srcFormat) && is16BPS(c->dstFormat))
-                    length*=2;
-                for (i=0; i<height; i++) {
-                    memcpy(dstPtr, srcPtr, length);
-                    srcPtr+= srcStride[plane];
-                    dstPtr+= dstStride[plane];
+            } else {
+                assert(lumSrcPtr + vLumFilterSize - 1 < lumPixBuf + vLumBufSize*2);
+                assert(chrUSrcPtr + vChrFilterSize - 1 < chrUPixBuf + vChrBufSize*2);
+                if(flags & SWS_FULL_CHR_H_INT) {
+                    yuv2rgbX_c_full(c,
+                                    vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize,
+                                    vChrFilter+dstY*vChrFilterSize, chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
+                                    alpSrcPtr, dest, dstW, dstY);
+                } else {
+                    yuv2packedX_c(c,
+                                  vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize,
+                                  vChrFilter+dstY*vChrFilterSize, chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
+                                  alpSrcPtr, dest, dstW, dstY);
                 }
             }
         }
     }
-    return srcSliceH;
-}
-
 
-static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
-{
-    *h = av_pix_fmt_descriptors[format].log2_chroma_w;
-    *v = av_pix_fmt_descriptors[format].log2_chroma_h;
-}
+    if ((dstFormat == PIX_FMT_YUVA420P) && !alpPixBuf)
+        fillPlane(dst[3], dstStride[3], dstW, dstY-lastDstY, lastDstY, 255);
 
-static uint16_t roundToInt16(int64_t f)
-{
-    int r= (f + (1<<15))>>16;
-         if (r<-0x7FFF) return 0x8000;
-    else if (r> 0x7FFF) return 0x7FFF;
-    else                return r;
-}
-
-int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
-{
-    int64_t crv =  inv_table[0];
-    int64_t cbu =  inv_table[1];
-    int64_t cgu = -inv_table[2];
-    int64_t cgv = -inv_table[3];
-    int64_t cy  = 1<<16;
-    int64_t oy  = 0;
-
-    memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
-    memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
-
-    c->brightness= brightness;
-    c->contrast  = contrast;
-    c->saturation= saturation;
-    c->srcRange  = srcRange;
-    c->dstRange  = dstRange;
-    if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
-
-    c->uOffset=   0x0400040004000400LL;
-    c->vOffset=   0x0400040004000400LL;
-
-    if (!srcRange) {
-        cy= (cy*255) / 219;
-        oy= 16<<16;
-    } else {
-        crv= (crv*224) / 255;
-        cbu= (cbu*224) / 255;
-        cgu= (cgu*224) / 255;
-        cgv= (cgv*224) / 255;
-    }
-
-    cy = (cy *contrast             )>>16;
-    crv= (crv*contrast * saturation)>>32;
-    cbu= (cbu*contrast * saturation)>>32;
-    cgu= (cgu*contrast * saturation)>>32;
-    cgv= (cgv*contrast * saturation)>>32;
-
-    oy -= 256*brightness;
-
-    c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
-    c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
-    c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
-    c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
-    c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
-    c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
-
-    c->yuv2rgb_y_coeff  = (int16_t)roundToInt16(cy <<13);
-    c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
-    c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
-    c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
-    c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
-    c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
-
-    ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
-    //FIXME factorize
-
-#ifdef COMPILE_ALTIVEC
-    if (c->flags & SWS_CPU_CAPS_ALTIVEC)
-        ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
+#if HAVE_MMX2
+    if (av_get_cpu_flags() & AV_CPU_FLAG_MMX2)
+        __asm__ volatile("sfence":::"memory");
 #endif
-    return 0;
-}
+    emms_c();
 
-int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
-{
-    if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
-
-    *inv_table = c->srcColorspaceTable;
-    *table     = c->dstColorspaceTable;
-    *srcRange  = c->srcRange;
-    *dstRange  = c->dstRange;
-    *brightness= c->brightness;
-    *contrast  = c->contrast;
-    *saturation= c->saturation;
-
-    return 0;
-}
+    /* store changed local vars back in the context */
+    c->dstY= dstY;
+    c->lumBufIndex= lumBufIndex;
+    c->chrBufIndex= chrBufIndex;
+    c->lastInLumBuf= lastInLumBuf;
+    c->lastInChrBuf= lastInChrBuf;
 
-static int handle_jpeg(enum PixelFormat *format)
-{
-    switch (*format) {
-    case PIX_FMT_YUVJ420P:
-        *format = PIX_FMT_YUV420P;
-        return 1;
-    case PIX_FMT_YUVJ422P:
-        *format = PIX_FMT_YUV422P;
-        return 1;
-    case PIX_FMT_YUVJ444P:
-        *format = PIX_FMT_YUV444P;
-        return 1;
-    case PIX_FMT_YUVJ440P:
-        *format = PIX_FMT_YUV440P;
-        return 1;
-    default:
-        return 0;
-    }
+    return dstY - lastDstY;
 }
 
-SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat, int dstW, int dstH, enum PixelFormat dstFormat, int flags,
-                           SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
+static void sws_init_swScale_c(SwsContext *c)
 {
+    enum PixelFormat srcFormat = c->srcFormat,
+                     dstFormat = c->dstFormat;
 
-    SwsContext *c;
-    int i;
-    int usesVFilter, usesHFilter;
-    int unscaled, needsDither;
-    int srcRange, dstRange;
-    SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
-#if ARCH_X86
-    if (flags & SWS_CPU_CAPS_MMX)
-        __asm__ volatile("emms\n\t"::: "memory");
-#endif
-
-#if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
-    flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
-#if   COMPILE_TEMPLATE_MMX2
-    flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
-#elif COMPILE_TEMPLATE_AMD3DNOW
-    flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
-#elif COMPILE_TEMPLATE_MMX
-    flags |= SWS_CPU_CAPS_MMX;
-#elif COMPILE_TEMPLATE_ALTIVEC
-    flags |= SWS_CPU_CAPS_ALTIVEC;
-#elif ARCH_BFIN
-    flags |= SWS_CPU_CAPS_BFIN;
-#endif
-#endif /* CONFIG_RUNTIME_CPUDETECT */
-    if (!rgb15to16) sws_rgb2rgb_init(flags);
-
-    unscaled = (srcW == dstW && srcH == dstH);
-    needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
-        && (fmt_depth(dstFormat))<24
-        && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
-
-    srcRange = handle_jpeg(&srcFormat);
-    dstRange = handle_jpeg(&dstFormat);
-
-    if (!isSupportedIn(srcFormat)) {
-        av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
-        return NULL;
-    }
-    if (!isSupportedOut(dstFormat)) {
-        av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
-        return NULL;
-    }
-
-    i= flags & ( SWS_POINT
-                |SWS_AREA
-                |SWS_BILINEAR
-                |SWS_FAST_BILINEAR
-                |SWS_BICUBIC
-                |SWS_X
-                |SWS_GAUSS
-                |SWS_LANCZOS
-                |SWS_SINC
-                |SWS_SPLINE
-                |SWS_BICUBLIN);
-    if(!i || (i & (i-1))) {
-        av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
-        return NULL;
-    }
-
-    /* sanity check */
-    if (srcW<4 || srcH<1 || dstW<8 || dstH<1) { //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code
-        av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
-               srcW, srcH, dstW, dstH);
-        return NULL;
-    }
-    if(srcW > VOFW || dstW > VOFW) {
-        av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
-        return NULL;
-    }
-
-    if (!dstFilter) dstFilter= &dummyFilter;
-    if (!srcFilter) srcFilter= &dummyFilter;
-
-    FF_ALLOCZ_OR_GOTO(NULL, c, sizeof(SwsContext), fail);
-
-    c->av_class = &sws_context_class;
-    c->srcW= srcW;
-    c->srcH= srcH;
-    c->dstW= dstW;
-    c->dstH= dstH;
-    c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
-    c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
-    c->flags= flags;
-    c->dstFormat= dstFormat;
-    c->srcFormat= srcFormat;
-    c->vRounder= 4* 0x0001000100010001ULL;
-
-    usesHFilter= usesVFilter= 0;
-    if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
-    if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
-    if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
-    if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
-    if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
-    if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
-    if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
-    if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
-
-    getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
-    getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
-
-    // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
-    if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
-
-    // drop some chroma lines if the user wants it
-    c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
-    c->chrSrcVSubSample+= c->vChrDrop;
-
-    // drop every other pixel for chroma calculation unless user wants full chroma
-    if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
-      && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8
-      && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4
-      && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
-      && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
-        c->chrSrcHSubSample=1;
-
-    if (param) {
-        c->param[0] = param[0];
-        c->param[1] = param[1];
-    } else {
-        c->param[0] =
-        c->param[1] = SWS_PARAM_DEFAULT;
-    }
-
-    // Note the -((-x)>>y) is so that we always round toward +inf.
-    c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
-    c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
-    c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
-    c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
-
-    sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
-
-    /* unscaled special cases */
-    if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat))) {
-        /* yv12_to_nv12 */
-        if ((srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P) && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21)) {
-            c->swScale= PlanarToNV12Wrapper;
-        }
-        /* yuv2bgr */
-        if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P || srcFormat==PIX_FMT_YUVA420P) && (isBGR(dstFormat) || isRGB(dstFormat))
-            && !(flags & SWS_ACCURATE_RND) && !(dstH&1)) {
-            c->swScale= ff_yuv2rgb_get_func_ptr(c);
-        }
-
-        if (srcFormat==PIX_FMT_YUV410P && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_BITEXACT)) {
-            c->swScale= yvu9toyv12Wrapper;
-        }
-
-        /* bgr24toYV12 */
-        if (srcFormat==PIX_FMT_BGR24 && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_ACCURATE_RND))
-            c->swScale= bgr24toyv12Wrapper;
-
-        /* RGB/BGR -> RGB/BGR (no dither needed forms) */
-        if (  (isBGR(srcFormat) || isRGB(srcFormat))
-           && (isBGR(dstFormat) || isRGB(dstFormat))
-           && srcFormat != PIX_FMT_BGR8      && dstFormat != PIX_FMT_BGR8
-           && srcFormat != PIX_FMT_RGB8      && dstFormat != PIX_FMT_RGB8
-           && srcFormat != PIX_FMT_BGR4      && dstFormat != PIX_FMT_BGR4
-           && srcFormat != PIX_FMT_RGB4      && dstFormat != PIX_FMT_RGB4
-           && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
-           && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
-           && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
-           && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
-                                             && dstFormat != PIX_FMT_RGB32_1
-                                             && dstFormat != PIX_FMT_BGR32_1
-           && srcFormat != PIX_FMT_RGB48LE   && dstFormat != PIX_FMT_RGB48LE
-           && srcFormat != PIX_FMT_RGB48BE   && dstFormat != PIX_FMT_RGB48BE
-           && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
-             c->swScale= rgb2rgbWrapper;
-
-        if ((usePal(srcFormat) && (
-                 dstFormat == PIX_FMT_RGB32   ||
-                 dstFormat == PIX_FMT_RGB32_1 ||
-                 dstFormat == PIX_FMT_RGB24   ||
-                 dstFormat == PIX_FMT_BGR32   ||
-                 dstFormat == PIX_FMT_BGR32_1 ||
-                 dstFormat == PIX_FMT_BGR24)))
-             c->swScale= pal2rgbWrapper;
-
-        if (srcFormat == PIX_FMT_YUV422P) {
-            if (dstFormat == PIX_FMT_YUYV422)
-                c->swScale= YUV422PToYuy2Wrapper;
-            else if (dstFormat == PIX_FMT_UYVY422)
-                c->swScale= YUV422PToUyvyWrapper;
-        }
-
-        /* LQ converters if -sws 0 or -sws 4*/
-        if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)) {
-            /* yv12_to_yuy2 */
-            if (srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P) {
-                if (dstFormat == PIX_FMT_YUYV422)
-                    c->swScale= PlanarToYuy2Wrapper;
-                else if (dstFormat == PIX_FMT_UYVY422)
-                    c->swScale= PlanarToUyvyWrapper;
-            }
-        }
-        if(srcFormat == PIX_FMT_YUYV422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
-            c->swScale= YUYV2YUV420Wrapper;
-        if(srcFormat == PIX_FMT_UYVY422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
-            c->swScale= UYVY2YUV420Wrapper;
-        if(srcFormat == PIX_FMT_YUYV422 && dstFormat == PIX_FMT_YUV422P)
-            c->swScale= YUYV2YUV422Wrapper;
-        if(srcFormat == PIX_FMT_UYVY422 && dstFormat == PIX_FMT_YUV422P)
-            c->swScale= UYVY2YUV422Wrapper;
-
-#ifdef COMPILE_ALTIVEC
-        if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
-            !(c->flags & SWS_BITEXACT) &&
-            srcFormat == PIX_FMT_YUV420P) {
-          // unscaled YV12 -> packed YUV, we want speed
-          if (dstFormat == PIX_FMT_YUYV422)
-              c->swScale= yv12toyuy2_unscaled_altivec;
-          else if (dstFormat == PIX_FMT_UYVY422)
-              c->swScale= yv12touyvy_unscaled_altivec;
-        }
-#endif
-
-        /* simple copy */
-        if (  srcFormat == dstFormat
-            || (srcFormat == PIX_FMT_YUVA420P && dstFormat == PIX_FMT_YUV420P)
-            || (srcFormat == PIX_FMT_YUV420P && dstFormat == PIX_FMT_YUVA420P)
-            || (isPlanarYUV(srcFormat) && isGray(dstFormat))
-            || (isPlanarYUV(dstFormat) && isGray(srcFormat))
-            || (isGray(dstFormat) && isGray(srcFormat))
-            || (isPlanarYUV(srcFormat) && isPlanarYUV(dstFormat)
-                && c->chrDstHSubSample == c->chrSrcHSubSample
-                && c->chrDstVSubSample == c->chrSrcVSubSample
-                && dstFormat != PIX_FMT_NV12 && dstFormat != PIX_FMT_NV21
-                && srcFormat != PIX_FMT_NV12 && srcFormat != PIX_FMT_NV21))
-        {
-            if (isPacked(c->srcFormat))
-                c->swScale= packedCopy;
-            else /* Planar YUV or gray */
-                c->swScale= planarCopy;
-        }
-#if ARCH_BFIN
-        if (flags & SWS_CPU_CAPS_BFIN)
-            ff_bfin_get_unscaled_swscale (c);
-#endif
-
-        if (c->swScale) {
-            if (flags&SWS_PRINT_INFO)
-                av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
-                       sws_format_name(srcFormat), sws_format_name(dstFormat));
-            return c;
-        }
-    }
-
-    if (flags & SWS_CPU_CAPS_MMX2) {
-        c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
-        if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
-            if (flags&SWS_PRINT_INFO)
-                av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
-        }
-        if (usesHFilter) c->canMMX2BeUsed=0;
-    }
-    else
-        c->canMMX2BeUsed=0;
-
-    c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
-    c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
-
-    // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
-    // but only for the FAST_BILINEAR mode otherwise do correct scaling
-    // n-2 is the last chrominance sample available
-    // this is not perfect, but no one should notice the difference, the more correct variant
-    // would be like the vertical one, but that would require some special code for the
-    // first and last pixel
-    if (flags&SWS_FAST_BILINEAR) {
-        if (c->canMMX2BeUsed) {
-            c->lumXInc+= 20;
-            c->chrXInc+= 20;
-        }
-        //we don't use the x86 asm scaler if MMX is available
-        else if (flags & SWS_CPU_CAPS_MMX) {
-            c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
-            c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
+    c->yuv2nv12X    = yuv2nv12X_c;
+    if (is16BPS(dstFormat)) {
+        c->yuv2yuvX     = isBE(dstFormat) ? yuv2yuvX16BE_c  : yuv2yuvX16LE_c;
+    } else if (is9_OR_10BPS(dstFormat)) {
+        if (dstFormat == PIX_FMT_YUV420P9BE || dstFormat == PIX_FMT_YUV420P9LE) {
+            c->yuv2yuvX = isBE(dstFormat) ? yuv2yuvX9BE_c :  yuv2yuvX9LE_c;
+        } else {
+            c->yuv2yuvX = isBE(dstFormat) ? yuv2yuvX10BE_c : yuv2yuvX10LE_c;
         }
+    } else {
+        c->yuv2yuv1     = yuv2yuv1_c;
+        c->yuv2yuvX     = yuv2yuvX_c;
     }
+    c->yuv2packed1  = yuv2packed1_c;
+    c->yuv2packed2  = yuv2packed2_c;
+    c->yuv2packedX  = yuv2packedX_c;
 
-    /* precalculate horizontal scaler filter coefficients */
-    {
-#if defined(COMPILE_MMX2)
-// can't downscale !!!
-        if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
-            c->lumMmx2FilterCodeSize = initMMX2HScaler(      dstW, c->lumXInc, NULL, NULL, NULL, 8);
-            c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
-
-#ifdef MAP_ANONYMOUS
-            c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
-            c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
-#elif HAVE_VIRTUALALLOC
-            c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
-            c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
-#else
-            c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
-            c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
-#endif
-
-            FF_ALLOCZ_OR_GOTO(c, c->hLumFilter   , (dstW        /8+8)*sizeof(int16_t), fail);
-            FF_ALLOCZ_OR_GOTO(c, c->hChrFilter   , (c->chrDstW  /4+8)*sizeof(int16_t), fail);
-            FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW      /2/8+8)*sizeof(int32_t), fail);
-            FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
+    c->hScale       = hScale_c;
 
-            initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
-            initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
-
-#ifdef MAP_ANONYMOUS
-            mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
-            mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
-#endif
-        } else
-#endif /* defined(COMPILE_MMX2) */
-        {
-            const int filterAlign=
-                (flags & SWS_CPU_CAPS_MMX) ? 4 :
-                (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
-                1;
-
-            if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
-                           srcW      ,       dstW, filterAlign, 1<<14,
-                           (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
-                           srcFilter->lumH, dstFilter->lumH, c->param) < 0)
-                goto fail;
-            if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
-                           c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
-                           (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
-                           srcFilter->chrH, dstFilter->chrH, c->param) < 0)
-                goto fail;
-        }
-    } // initialize horizontal stuff
-
-
-
-    /* precalculate vertical scaler filter coefficients */
+    if (c->flags & SWS_FAST_BILINEAR)
     {
-        const int filterAlign=
-            (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
-            (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
-            1;
-
-        if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
-                       srcH      ,        dstH, filterAlign, (1<<12),
-                       (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
-                       srcFilter->lumV, dstFilter->lumV, c->param) < 0)
-            goto fail;
-        if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
-                       c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
-                       (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
-                       srcFilter->chrV, dstFilter->chrV, c->param) < 0)
-            goto fail;
-
-#ifdef COMPILE_ALTIVEC
-        FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
-        FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
-
-        for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
-            int j;
-            short *p = (short *)&c->vYCoeffsBank[i];
-            for (j=0;j<8;j++)
-                p[j] = c->vLumFilter[i];
+        c->hyscale_fast = hyscale_fast_c;
+        c->hcscale_fast = hcscale_fast_c;
+    }
+
+    c->chrToYV12 = NULL;
+    switch(srcFormat) {
+        case PIX_FMT_YUYV422  : c->chrToYV12 = yuy2ToUV_c; break;
+        case PIX_FMT_UYVY422  : c->chrToYV12 = uyvyToUV_c; break;
+        case PIX_FMT_NV12     : c->chrToYV12 = nv12ToUV_c; break;
+        case PIX_FMT_NV21     : c->chrToYV12 = nv21ToUV_c; break;
+        case PIX_FMT_RGB8     :
+        case PIX_FMT_BGR8     :
+        case PIX_FMT_PAL8     :
+        case PIX_FMT_BGR4_BYTE:
+        case PIX_FMT_RGB4_BYTE: c->chrToYV12 = palToUV_c; break;
+        case PIX_FMT_YUV420P9BE: c->chrToYV12 = BE9ToUV_c; break;
+        case PIX_FMT_YUV420P9LE: c->chrToYV12 = LE9ToUV_c; break;
+        case PIX_FMT_YUV420P10BE: c->chrToYV12 = BE10ToUV_c; break;
+        case PIX_FMT_YUV420P10LE: c->chrToYV12 = LE10ToUV_c; break;
+        case PIX_FMT_YUV420P16BE:
+        case PIX_FMT_YUV422P16BE:
+        case PIX_FMT_YUV444P16BE: c->chrToYV12 = BEToUV_c; break;
+        case PIX_FMT_YUV420P16LE:
+        case PIX_FMT_YUV422P16LE:
+        case PIX_FMT_YUV444P16LE: c->chrToYV12 = LEToUV_c; break;
+    }
+    if (c->chrSrcHSubSample) {
+        switch(srcFormat) {
+        case PIX_FMT_RGB48BE: c->chrToYV12 = rgb48BEToUV_half_c; break;
+        case PIX_FMT_RGB48LE: c->chrToYV12 = rgb48LEToUV_half_c; break;
+        case PIX_FMT_BGR48BE: c->chrToYV12 = bgr48BEToUV_half_c; break;
+        case PIX_FMT_BGR48LE: c->chrToYV12 = bgr48LEToUV_half_c; break;
+        case PIX_FMT_RGB32  : c->chrToYV12 = bgr32ToUV_half_c;  break;
+        case PIX_FMT_RGB32_1: c->chrToYV12 = bgr321ToUV_half_c; break;
+        case PIX_FMT_BGR24  : c->chrToYV12 = bgr24ToUV_half_c; break;
+        case PIX_FMT_BGR565 : c->chrToYV12 = bgr16ToUV_half_c; break;
+        case PIX_FMT_BGR555 : c->chrToYV12 = bgr15ToUV_half_c; break;
+        case PIX_FMT_BGR32  : c->chrToYV12 = rgb32ToUV_half_c;  break;
+        case PIX_FMT_BGR32_1: c->chrToYV12 = rgb321ToUV_half_c; break;
+        case PIX_FMT_RGB24  : c->chrToYV12 = rgb24ToUV_half_c; break;
+        case PIX_FMT_RGB565 : c->chrToYV12 = rgb16ToUV_half_c; break;
+        case PIX_FMT_RGB555 : c->chrToYV12 = rgb15ToUV_half_c; break;
         }
-
-        for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
-            int j;
-            short *p = (short *)&c->vCCoeffsBank[i];
-            for (j=0;j<8;j++)
-                p[j] = c->vChrFilter[i];
+    } else {
+        switch(srcFormat) {
+        case PIX_FMT_RGB48BE: c->chrToYV12 = rgb48BEToUV_c; break;
+        case PIX_FMT_RGB48LE: c->chrToYV12 = rgb48LEToUV_c; break;
+        case PIX_FMT_BGR48BE: c->chrToYV12 = bgr48BEToUV_c; break;
+        case PIX_FMT_BGR48LE: c->chrToYV12 = bgr48LEToUV_c; break;
+        case PIX_FMT_RGB32  : c->chrToYV12 = bgr32ToUV_c;  break;
+        case PIX_FMT_RGB32_1: c->chrToYV12 = bgr321ToUV_c; break;
+        case PIX_FMT_BGR24  : c->chrToYV12 = bgr24ToUV_c; break;
+        case PIX_FMT_BGR565 : c->chrToYV12 = bgr16ToUV_c; break;
+        case PIX_FMT_BGR555 : c->chrToYV12 = bgr15ToUV_c; break;
+        case PIX_FMT_BGR32  : c->chrToYV12 = rgb32ToUV_c;  break;
+        case PIX_FMT_BGR32_1: c->chrToYV12 = rgb321ToUV_c; break;
+        case PIX_FMT_RGB24  : c->chrToYV12 = rgb24ToUV_c; break;
+        case PIX_FMT_RGB565 : c->chrToYV12 = rgb16ToUV_c; break;
+        case PIX_FMT_RGB555 : c->chrToYV12 = rgb15ToUV_c; break;
         }
-#endif
-    }
-
-    // calculate buffer sizes so that they won't run out while handling these damn slices
-    c->vLumBufSize= c->vLumFilterSize;
-    c->vChrBufSize= c->vChrFilterSize;
-    for (i=0; i<dstH; i++) {
-        int chrI= i*c->chrDstH / dstH;
-        int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
-                           ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
-
-        nextSlice>>= c->chrSrcVSubSample;
-        nextSlice<<= c->chrSrcVSubSample;
-        if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
-            c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
-        if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
-            c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
     }
 
-    // allocate pixbufs (we use dynamic allocation because otherwise we would need to
-    // allocate several megabytes to handle all possible cases)
-    FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
-    FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
-    if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
-        FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
-    //Note we need at least one pixel more at the end because of the MMX code (just in case someone wanna replace the 4000/8000)
-    /* align at 16 bytes for AltiVec */
-    for (i=0; i<c->vLumBufSize; i++) {
-        FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);
-        c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
-    }
-    for (i=0; i<c->vChrBufSize; i++) {
-        FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);
-        c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];
-    }
-    if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
-        for (i=0; i<c->vLumBufSize; i++) {
-            FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);
-            c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
+    c->lumToYV12 = NULL;
+    c->alpToYV12 = NULL;
+    switch (srcFormat) {
+    case PIX_FMT_YUV420P9BE: c->lumToYV12 = BE9ToY_c; break;
+    case PIX_FMT_YUV420P9LE: c->lumToYV12 = LE9ToY_c; break;
+    case PIX_FMT_YUV420P10BE: c->lumToYV12 = BE10ToY_c; break;
+    case PIX_FMT_YUV420P10LE: c->lumToYV12 = LE10ToY_c; break;
+    case PIX_FMT_YUYV422  :
+    case PIX_FMT_YUV420P16BE:
+    case PIX_FMT_YUV422P16BE:
+    case PIX_FMT_YUV444P16BE:
+    case PIX_FMT_Y400A    :
+    case PIX_FMT_GRAY16BE : c->lumToYV12 = yuy2ToY_c; break;
+    case PIX_FMT_UYVY422  :
+    case PIX_FMT_YUV420P16LE:
+    case PIX_FMT_YUV422P16LE:
+    case PIX_FMT_YUV444P16LE:
+    case PIX_FMT_GRAY16LE : c->lumToYV12 = uyvyToY_c; break;
+    case PIX_FMT_BGR24    : c->lumToYV12 = bgr24ToY_c; break;
+    case PIX_FMT_BGR565   : c->lumToYV12 = bgr16ToY_c; break;
+    case PIX_FMT_BGR555   : c->lumToYV12 = bgr15ToY_c; break;
+    case PIX_FMT_RGB24    : c->lumToYV12 = rgb24ToY_c; break;
+    case PIX_FMT_RGB565   : c->lumToYV12 = rgb16ToY_c; break;
+    case PIX_FMT_RGB555   : c->lumToYV12 = rgb15ToY_c; break;
+    case PIX_FMT_RGB8     :
+    case PIX_FMT_BGR8     :
+    case PIX_FMT_PAL8     :
+    case PIX_FMT_BGR4_BYTE:
+    case PIX_FMT_RGB4_BYTE: c->lumToYV12 = palToY_c; break;
+    case PIX_FMT_MONOBLACK: c->lumToYV12 = monoblack2Y_c; break;
+    case PIX_FMT_MONOWHITE: c->lumToYV12 = monowhite2Y_c; break;
+    case PIX_FMT_RGB32  : c->lumToYV12 = bgr32ToY_c;  break;
+    case PIX_FMT_RGB32_1: c->lumToYV12 = bgr321ToY_c; break;
+    case PIX_FMT_BGR32  : c->lumToYV12 = rgb32ToY_c;  break;
+    case PIX_FMT_BGR32_1: c->lumToYV12 = rgb321ToY_c; break;
+    case PIX_FMT_RGB48BE: c->lumToYV12 = rgb48BEToY_c; break;
+    case PIX_FMT_RGB48LE: c->lumToYV12 = rgb48LEToY_c; break;
+    case PIX_FMT_BGR48BE: c->lumToYV12 = bgr48BEToY_c; break;
+    case PIX_FMT_BGR48LE: c->lumToYV12 = bgr48LEToY_c; break;
+    }
+    if (c->alpPixBuf) {
+        switch (srcFormat) {
+        case PIX_FMT_BGRA:
+        case PIX_FMT_RGBA:  c->alpToYV12 = rgbaToA_c; break;
+        case PIX_FMT_ABGR:
+        case PIX_FMT_ARGB:  c->alpToYV12 = abgrToA_c; break;
+        case PIX_FMT_Y400A: c->alpToYV12 = uyvyToY_c; break;
         }
-
-    //try to avoid drawing green stuff between the right end and the stride end
-    for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
-
-    assert(2*VOFW == VOF);
-
-    assert(c->chrDstH <= dstH);
-
-    if (flags&SWS_PRINT_INFO) {
-        if (flags&SWS_FAST_BILINEAR)
-            av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
-        else if (flags&SWS_BILINEAR)
-            av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
-        else if (flags&SWS_BICUBIC)
-            av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
-        else if (flags&SWS_X)
-            av_log(c, AV_LOG_INFO, "Experimental scaler, ");
-        else if (flags&SWS_POINT)
-            av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
-        else if (flags&SWS_AREA)
-            av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
-        else if (flags&SWS_BICUBLIN)
-            av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
-        else if (flags&SWS_GAUSS)
-            av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
-        else if (flags&SWS_SINC)
-            av_log(c, AV_LOG_INFO, "Sinc scaler, ");
-        else if (flags&SWS_LANCZOS)
-            av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
-        else if (flags&SWS_SPLINE)
-            av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
-        else
-            av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
-
-        av_log(c, AV_LOG_INFO, "from %s to %s%s ",
-               sws_format_name(srcFormat),
-#ifdef DITHER1XBPP
-               dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ? "dithered " : "",
-#else
-               "",
-#endif
-               sws_format_name(dstFormat));
-
-        if (flags & SWS_CPU_CAPS_MMX2)
-            av_log(c, AV_LOG_INFO, "using MMX2\n");
-        else if (flags & SWS_CPU_CAPS_3DNOW)
-            av_log(c, AV_LOG_INFO, "using 3DNOW\n");
-        else if (flags & SWS_CPU_CAPS_MMX)
-            av_log(c, AV_LOG_INFO, "using MMX\n");
-        else if (flags & SWS_CPU_CAPS_ALTIVEC)
-            av_log(c, AV_LOG_INFO, "using AltiVec\n");
-        else
-            av_log(c, AV_LOG_INFO, "using C\n");
     }
 
-    if (flags & SWS_PRINT_INFO) {
-        if (flags & SWS_CPU_CAPS_MMX) {
-            if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
-                av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
-            else {
-                if (c->hLumFilterSize==4)
-                    av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
-                else if (c->hLumFilterSize==8)
-                    av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
-                else
-                    av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
-
-                if (c->hChrFilterSize==4)
-                    av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
-                else if (c->hChrFilterSize==8)
-                    av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
-                else
-                    av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
-            }
+    if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
+        if (c->srcRange) {
+            c->lumConvertRange = lumRangeFromJpeg_c;
+            c->chrConvertRange = chrRangeFromJpeg_c;
         } else {
-#if ARCH_X86
-            av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
-#else
-            if (flags & SWS_FAST_BILINEAR)
-                av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
-            else
-                av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
-#endif
-        }
-        if (isPlanarYUV(dstFormat)) {
-            if (c->vLumFilterSize==1)
-                av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
-            else
-                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
-        } else {
-            if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
-                av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
-                       "      2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
-            else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
-                av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
-            else
-                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
-        }
-
-        if (dstFormat==PIX_FMT_BGR24)
-            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
-                   (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
-        else if (dstFormat==PIX_FMT_RGB32)
-            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
-        else if (dstFormat==PIX_FMT_BGR565)
-            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
-        else if (dstFormat==PIX_FMT_BGR555)
-            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
-
-        av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
-    }
-    if (flags & SWS_PRINT_INFO) {
-        av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
-               c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
-        av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
-               c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
-    }
-
-    c->swScale= getSwsFunc(c);
-    return c;
-
-fail:
-    sws_freeContext(c);
-    return NULL;
-}
-
-static void reset_ptr(const uint8_t* src[], int format)
-{
-    if(!isALPHA(format))
-        src[3]=NULL;
-    if(!isPlanarYUV(format)) {
-        src[3]=src[2]=NULL;
-
-        if (!usePal(format))
-            src[1]= NULL;
-    }
-}
-
-/**
- * swscale wrapper, so we don't need to export the SwsContext.
- * Assumes planar YUV to be in YUV order instead of YVU.
- */
-int sws_scale(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-              int srcSliceH, uint8_t* dst[], int dstStride[])
-{
-    int i;
-    const uint8_t* src2[4]= {src[0], src[1], src[2], src[3]};
-    uint8_t* dst2[4]= {dst[0], dst[1], dst[2], dst[3]};
-
-    // do not mess up sliceDir if we have a "trailing" 0-size slice
-    if (srcSliceH == 0)
-        return 0;
-
-    if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
-        av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
-        return 0;
-    }
-    if (c->sliceDir == 0) {
-        if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
-    }
-
-    if (usePal(c->srcFormat)) {
-        for (i=0; i<256; i++) {
-            int p, r, g, b,y,u,v;
-            if(c->srcFormat == PIX_FMT_PAL8) {
-                p=((uint32_t*)(src[1]))[i];
-                r= (p>>16)&0xFF;
-                g= (p>> 8)&0xFF;
-                b=  p     &0xFF;
-            } else if(c->srcFormat == PIX_FMT_RGB8) {
-                r= (i>>5    )*36;
-                g= ((i>>2)&7)*36;
-                b= (i&3     )*85;
-            } else if(c->srcFormat == PIX_FMT_BGR8) {
-                b= (i>>6    )*85;
-                g= ((i>>3)&7)*36;
-                r= (i&7     )*36;
-            } else if(c->srcFormat == PIX_FMT_RGB4_BYTE) {
-                r= (i>>3    )*255;
-                g= ((i>>1)&3)*85;
-                b= (i&1     )*255;
-            } else {
-                assert(c->srcFormat == PIX_FMT_BGR4_BYTE);
-                b= (i>>3    )*255;
-                g= ((i>>1)&3)*85;
-                r= (i&1     )*255;
-            }
-            y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
-            u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
-            v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
-            c->pal_yuv[i]= y + (u<<8) + (v<<16);
-
-
-            switch(c->dstFormat) {
-            case PIX_FMT_BGR32:
-#if !HAVE_BIGENDIAN
-            case PIX_FMT_RGB24:
-#endif
-                c->pal_rgb[i]=  r + (g<<8) + (b<<16);
-                break;
-            case PIX_FMT_BGR32_1:
-#if HAVE_BIGENDIAN
-            case PIX_FMT_BGR24:
-#endif
-                c->pal_rgb[i]= (r + (g<<8) + (b<<16)) << 8;
-                break;
-            case PIX_FMT_RGB32_1:
-#if HAVE_BIGENDIAN
-            case PIX_FMT_RGB24:
-#endif
-                c->pal_rgb[i]= (b + (g<<8) + (r<<16)) << 8;
-                break;
-            case PIX_FMT_RGB32:
-#if !HAVE_BIGENDIAN
-            case PIX_FMT_BGR24:
-#endif
-            default:
-                c->pal_rgb[i]=  b + (g<<8) + (r<<16);
-            }
-        }
-    }
-
-    // copy strides, so they can safely be modified
-    if (c->sliceDir == 1) {
-        // slices go from top to bottom
-        int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2], srcStride[3]};
-        int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2], dstStride[3]};
-
-        reset_ptr(src2, c->srcFormat);
-        reset_ptr((const uint8_t**)dst2, c->dstFormat);
-
-        /* reset slice direction at end of frame */
-        if (srcSliceY + srcSliceH == c->srcH)
-            c->sliceDir = 0;
-
-        return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst2, dstStride2);
-    } else {
-        // slices go from bottom to top => we flip the image internally
-        int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2], -srcStride[3]};
-        int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2], -dstStride[3]};
-
-        src2[0] += (srcSliceH-1)*srcStride[0];
-        if (!usePal(c->srcFormat))
-            src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
-        src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
-        src2[3] += (srcSliceH-1)*srcStride[3];
-        dst2[0] += ( c->dstH                      -1)*dstStride[0];
-        dst2[1] += ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1];
-        dst2[2] += ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2];
-        dst2[3] += ( c->dstH                      -1)*dstStride[3];
-
-        reset_ptr(src2, c->srcFormat);
-        reset_ptr((const uint8_t**)dst2, c->dstFormat);
-
-        /* reset slice direction at end of frame */
-        if (!srcSliceY)
-            c->sliceDir = 0;
-
-        return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
-    }
-}
-
-#if LIBSWSCALE_VERSION_MAJOR < 1
-int sws_scale_ordered(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
-                      int srcSliceH, uint8_t* dst[], int dstStride[])
-{
-    return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
-}
-#endif
-
-SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
-                                float lumaSharpen, float chromaSharpen,
-                                float chromaHShift, float chromaVShift,
-                                int verbose)
-{
-    SwsFilter *filter= av_malloc(sizeof(SwsFilter));
-    if (!filter)
-        return NULL;
-
-    if (lumaGBlur!=0.0) {
-        filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
-        filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
-    } else {
-        filter->lumH= sws_getIdentityVec();
-        filter->lumV= sws_getIdentityVec();
-    }
-
-    if (chromaGBlur!=0.0) {
-        filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
-        filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
-    } else {
-        filter->chrH= sws_getIdentityVec();
-        filter->chrV= sws_getIdentityVec();
-    }
-
-    if (chromaSharpen!=0.0) {
-        SwsVector *id= sws_getIdentityVec();
-        sws_scaleVec(filter->chrH, -chromaSharpen);
-        sws_scaleVec(filter->chrV, -chromaSharpen);
-        sws_addVec(filter->chrH, id);
-        sws_addVec(filter->chrV, id);
-        sws_freeVec(id);
-    }
-
-    if (lumaSharpen!=0.0) {
-        SwsVector *id= sws_getIdentityVec();
-        sws_scaleVec(filter->lumH, -lumaSharpen);
-        sws_scaleVec(filter->lumV, -lumaSharpen);
-        sws_addVec(filter->lumH, id);
-        sws_addVec(filter->lumV, id);
-        sws_freeVec(id);
-    }
-
-    if (chromaHShift != 0.0)
-        sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
-
-    if (chromaVShift != 0.0)
-        sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
-
-    sws_normalizeVec(filter->chrH, 1.0);
-    sws_normalizeVec(filter->chrV, 1.0);
-    sws_normalizeVec(filter->lumH, 1.0);
-    sws_normalizeVec(filter->lumV, 1.0);
-
-    if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
-    if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
-
-    return filter;
-}
-
-SwsVector *sws_allocVec(int length)
-{
-    SwsVector *vec = av_malloc(sizeof(SwsVector));
-    if (!vec)
-        return NULL;
-    vec->length = length;
-    vec->coeff  = av_malloc(sizeof(double) * length);
-    if (!vec->coeff)
-        av_freep(&vec);
-    return vec;
-}
-
-SwsVector *sws_getGaussianVec(double variance, double quality)
-{
-    const int length= (int)(variance*quality + 0.5) | 1;
-    int i;
-    double middle= (length-1)*0.5;
-    SwsVector *vec= sws_allocVec(length);
-
-    if (!vec)
-        return NULL;
-
-    for (i=0; i<length; i++) {
-        double dist= i-middle;
-        vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
-    }
-
-    sws_normalizeVec(vec, 1.0);
-
-    return vec;
-}
-
-SwsVector *sws_getConstVec(double c, int length)
-{
-    int i;
-    SwsVector *vec= sws_allocVec(length);
-
-    if (!vec)
-        return NULL;
-
-    for (i=0; i<length; i++)
-        vec->coeff[i]= c;
-
-    return vec;
-}
-
-
-SwsVector *sws_getIdentityVec(void)
-{
-    return sws_getConstVec(1.0, 1);
-}
-
-double sws_dcVec(SwsVector *a)
-{
-    int i;
-    double sum=0;
-
-    for (i=0; i<a->length; i++)
-        sum+= a->coeff[i];
-
-    return sum;
-}
-
-void sws_scaleVec(SwsVector *a, double scalar)
-{
-    int i;
-
-    for (i=0; i<a->length; i++)
-        a->coeff[i]*= scalar;
-}
-
-void sws_normalizeVec(SwsVector *a, double height)
-{
-    sws_scaleVec(a, height/sws_dcVec(a));
-}
-
-static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
-{
-    int length= a->length + b->length - 1;
-    int i, j;
-    SwsVector *vec= sws_getConstVec(0.0, length);
-
-    if (!vec)
-        return NULL;
-
-    for (i=0; i<a->length; i++) {
-        for (j=0; j<b->length; j++) {
-            vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
+            c->lumConvertRange = lumRangeToJpeg_c;
+            c->chrConvertRange = chrRangeToJpeg_c;
         }
     }
 
-    return vec;
-}
-
-static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
-{
-    int length= FFMAX(a->length, b->length);
-    int i;
-    SwsVector *vec= sws_getConstVec(0.0, length);
-
-    if (!vec)
-        return NULL;
-
-    for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
-    for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
-
-    return vec;
-}
-
-static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
-{
-    int length= FFMAX(a->length, b->length);
-    int i;
-    SwsVector *vec= sws_getConstVec(0.0, length);
-
-    if (!vec)
-        return NULL;
-
-    for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
-    for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
-
-    return vec;
-}
-
-/* shift left / or right if "shift" is negative */
-static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
-{
-    int length= a->length + FFABS(shift)*2;
-    int i;
-    SwsVector *vec= sws_getConstVec(0.0, length);
-
-    if (!vec)
-        return NULL;
-
-    for (i=0; i<a->length; i++) {
-        vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
-    }
-
-    return vec;
-}
-
-void sws_shiftVec(SwsVector *a, int shift)
-{
-    SwsVector *shifted= sws_getShiftedVec(a, shift);
-    av_free(a->coeff);
-    a->coeff= shifted->coeff;
-    a->length= shifted->length;
-    av_free(shifted);
-}
-
-void sws_addVec(SwsVector *a, SwsVector *b)
-{
-    SwsVector *sum= sws_sumVec(a, b);
-    av_free(a->coeff);
-    a->coeff= sum->coeff;
-    a->length= sum->length;
-    av_free(sum);
-}
-
-void sws_subVec(SwsVector *a, SwsVector *b)
-{
-    SwsVector *diff= sws_diffVec(a, b);
-    av_free(a->coeff);
-    a->coeff= diff->coeff;
-    a->length= diff->length;
-    av_free(diff);
-}
-
-void sws_convVec(SwsVector *a, SwsVector *b)
-{
-    SwsVector *conv= sws_getConvVec(a, b);
-    av_free(a->coeff);
-    a->coeff= conv->coeff;
-    a->length= conv->length;
-    av_free(conv);
-}
-
-SwsVector *sws_cloneVec(SwsVector *a)
-{
-    int i;
-    SwsVector *vec= sws_allocVec(a->length);
-
-    if (!vec)
-        return NULL;
-
-    for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
-
-    return vec;
-}
-
-void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
-{
-    int i;
-    double max=0;
-    double min=0;
-    double range;
-
-    for (i=0; i<a->length; i++)
-        if (a->coeff[i]>max) max= a->coeff[i];
-
-    for (i=0; i<a->length; i++)
-        if (a->coeff[i]<min) min= a->coeff[i];
-
-    range= max - min;
-
-    for (i=0; i<a->length; i++) {
-        int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
-        av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
-        for (;x>0; x--) av_log(log_ctx, log_level, " ");
-        av_log(log_ctx, log_level, "|\n");
-    }
-}
-
-#if LIBSWSCALE_VERSION_MAJOR < 1
-void sws_printVec(SwsVector *a)
-{
-    sws_printVec2(a, NULL, AV_LOG_DEBUG);
+    if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
+          srcFormat == PIX_FMT_MONOBLACK || srcFormat == PIX_FMT_MONOWHITE))
+        c->needs_hcscale = 1;
 }
-#endif
 
-void sws_freeVec(SwsVector *a)
+SwsFunc ff_getSwsFunc(SwsContext *c)
 {
-    if (!a) return;
-    av_freep(&a->coeff);
-    a->length=0;
-    av_free(a);
-}
+    sws_init_swScale_c(c);
 
-void sws_freeFilter(SwsFilter *filter)
-{
-    if (!filter) return;
+    if (HAVE_MMX)
+        ff_sws_init_swScale_mmx(c);
+    if (HAVE_ALTIVEC)
+        ff_sws_init_swScale_altivec(c);
 
-    if (filter->lumH) sws_freeVec(filter->lumH);
-    if (filter->lumV) sws_freeVec(filter->lumV);
-    if (filter->chrH) sws_freeVec(filter->chrH);
-    if (filter->chrV) sws_freeVec(filter->chrV);
-    av_free(filter);
+    return swScale;
 }
-
-
-void sws_freeContext(SwsContext *c)
-{
-    int i;
-    if (!c) return;
-
-    if (c->lumPixBuf) {
-        for (i=0; i<c->vLumBufSize; i++)
-            av_freep(&c->lumPixBuf[i]);
-        av_freep(&c->lumPixBuf);
-    }
-
-    if (c->chrPixBuf) {
-        for (i=0; i<c->vChrBufSize; i++)
-            av_freep(&c->chrPixBuf[i]);
-        av_freep(&c->chrPixBuf);
-    }
-
-    if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
-        for (i=0; i<c->vLumBufSize; i++)
-            av_freep(&c->alpPixBuf[i]);
-        av_freep(&c->alpPixBuf);
-    }
-
-    av_freep(&c->vLumFilter);
-    av_freep(&c->vChrFilter);
-    av_freep(&c->hLumFilter);
-    av_freep(&c->hChrFilter);
-#ifdef COMPILE_ALTIVEC
-    av_freep(&c->vYCoeffsBank);
-    av_freep(&c->vCCoeffsBank);
-#endif
-
-    av_freep(&c->vLumFilterPos);
-    av_freep(&c->vChrFilterPos);
-    av_freep(&c->hLumFilterPos);
-    av_freep(&c->hChrFilterPos);
-
-#if ARCH_X86 && CONFIG_GPL
-#ifdef MAP_ANONYMOUS
-    if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
-    if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
-#elif HAVE_VIRTUALALLOC
-    if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, MEM_RELEASE);
-    if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, MEM_RELEASE);
-#else
-    av_free(c->lumMmx2FilterCode);
-    av_free(c->chrMmx2FilterCode);
-#endif
-    c->lumMmx2FilterCode=NULL;
-    c->chrMmx2FilterCode=NULL;
-#endif /* ARCH_X86 && CONFIG_GPL */
-
-    av_freep(&c->yuvTable);
-
-    av_free(c);
-}
-
-struct SwsContext *sws_getCachedContext(struct SwsContext *context,
-                                        int srcW, int srcH, enum PixelFormat srcFormat,
-                                        int dstW, int dstH, enum PixelFormat dstFormat, int flags,
-                                        SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
-{
-    static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
-
-    if (!param)
-        param = default_param;
-
-    if (context) {
-        if (context->srcW != srcW || context->srcH != srcH ||
-            context->srcFormat != srcFormat ||
-            context->dstW != dstW || context->dstH != dstH ||
-            context->dstFormat != dstFormat || context->flags != flags ||
-            context->param[0] != param[0] || context->param[1] != param[1])
-        {
-            sws_freeContext(context);
-            context = NULL;
-        }
-    }
-    if (!context) {
-        return sws_getContext(srcW, srcH, srcFormat,
-                              dstW, dstH, dstFormat, flags,
-                              srcFilter, dstFilter, param);
-    }
-    return context;
-}
-