Move postproc ---> libswscale
authorLuca Abeni <lucabe72@email.it>
Fri, 30 Jun 2006 12:00:31 +0000 (12:00 +0000)
committerLuca Abeni <lucabe72@email.it>
Fri, 30 Jun 2006 12:00:31 +0000 (12:00 +0000)
Originally committed as revision 18866 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale

30 files changed:
libswscale/Makefile [new file with mode: 0644]
libswscale/cs_test.c [new file with mode: 0644]
libswscale/rgb2rgb.c [new file with mode: 0644]
libswscale/rgb2rgb.h [new file with mode: 0644]
libswscale/rgb2rgb_template.c [new file with mode: 0644]
libswscale/swscale-example.c [new file with mode: 0644]
libswscale/swscale.c [new file with mode: 0644]
libswscale/swscale.h [new file with mode: 0644]
libswscale/swscale_altivec_template.c [new file with mode: 0644]
libswscale/swscale_internal.h [new file with mode: 0644]
libswscale/swscale_template.c [new file with mode: 0644]
libswscale/yuv2rgb.c [new file with mode: 0644]
libswscale/yuv2rgb_altivec.c [new file with mode: 0644]
libswscale/yuv2rgb_mlib.c [new file with mode: 0644]
libswscale/yuv2rgb_template.c [new file with mode: 0644]
postproc/Makefile [deleted file]
postproc/cs_test.c [deleted file]
postproc/rgb2rgb.c [deleted file]
postproc/rgb2rgb.h [deleted file]
postproc/rgb2rgb_template.c [deleted file]
postproc/swscale-example.c [deleted file]
postproc/swscale.c [deleted file]
postproc/swscale.h [deleted file]
postproc/swscale_altivec_template.c [deleted file]
postproc/swscale_internal.h [deleted file]
postproc/swscale_template.c [deleted file]
postproc/yuv2rgb.c [deleted file]
postproc/yuv2rgb_altivec.c [deleted file]
postproc/yuv2rgb_mlib.c [deleted file]
postproc/yuv2rgb_template.c [deleted file]

diff --git a/libswscale/Makefile b/libswscale/Makefile
new file mode 100644 (file)
index 0000000..475bd31
--- /dev/null
@@ -0,0 +1,52 @@
+
+include ../config.mak
+
+SWSLIB = libswscale.a
+
+SWSSRCS=swscale.c rgb2rgb.c yuv2rgb.c
+
+SWSOBJS=$(SWSSRCS:.c=.o)
+
+ifeq ($(TARGET_ALTIVEC),yes)
+SWSOBJS +=  yuv2rgb_altivec.o
+endif
+
+CFLAGS  = $(OPTFLAGS) $(MLIB_INC) -I. -I.. $(EXTRA_INC)
+# -I/usr/X11R6/include/
+
+.SUFFIXES: .c .o
+
+# .PHONY: all clean
+
+.c.o:
+       $(CC) -c $(CFLAGS) -I.. -o $@ $<
+
+all:    $(SWSLIB)
+
+$(SWSLIB):     $(SWSOBJS)
+       $(AR) r $(SWSLIB) $(SWSOBJS)
+       $(RANLIB) $(SWSLIB)
+
+clean:
+       rm -f *.o *.a *~ *.so cs_test swscale-example
+
+distclean: clean
+       rm -f .depend
+
+dep:    depend
+
+depend:
+       $(CC) -MM $(CFLAGS) $(SWSSRCS) 1>.depend
+
+cs_test: cs_test.o $(SWSLIB)
+       $(CC) cs_test.o $(SWSLIB) ../cpudetect.o -DFOR_MENCODER ../mp_msg.c -o cs_test -W -Wall
+
+swscale-example: swscale-example.o $(SWSLIB)
+       $(CC) swscale-example.o $(SWSLIB) ../libmpcodecs/img_format.o -lm -o swscale-example -W -Wall
+#
+# include dependency files if they exist
+#
+ifneq ($(wildcard .depend),)
+include .depend
+endif
+
diff --git a/libswscale/cs_test.c b/libswscale/cs_test.c
new file mode 100644 (file)
index 0000000..b1963f9
--- /dev/null
@@ -0,0 +1,288 @@
+/*
+    Copyright (C) 2002 Michael Niedermayer <michaelni@gmx.at>
+
+    This program 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.
+
+    This program 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.
+
+    You should have received a copy of the GNU General Public License
+    along with this program; if not, write to the Free Software
+    Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <inttypes.h>
+
+#include "swscale.h"
+#include "rgb2rgb.h"
+#include "cpudetect.h"
+
+#define SIZE 1000
+#define srcByte 0x55
+#define dstByte 0xBB
+
+#ifdef __APPLE_CC__
+#define memalign(x,y) malloc(y)
+#endif
+
+static int get_sws_cpuflags()
+{
+    return (gCpuCaps.hasMMX ? SWS_CPU_CAPS_MMX : 0) |
+       (gCpuCaps.hasMMX2 ? SWS_CPU_CAPS_MMX2 : 0) |
+       (gCpuCaps.has3DNow ? SWS_CPU_CAPS_3DNOW : 0);
+}
+
+main(int argc, char **argv)
+{
+       int i, funcNum;
+       uint8_t *srcBuffer= (uint8_t*)memalign(128, SIZE);
+       uint8_t *dstBuffer= (uint8_t*)memalign(128, SIZE);
+       int failedNum=0;
+       int passedNum=0;
+       
+       printf("memory corruption test ...\n");
+       
+       if(argc==2){
+               GetCpuCaps(&gCpuCaps);
+               printf("testing mmx\n");
+       }
+       
+       sws_rgb2rgb_init(get_sws_cpuflags());
+       
+       for(funcNum=0; funcNum<100; funcNum++){
+               int width;
+               int failed=0;
+               int srcBpp=0;
+               int dstBpp=0;
+
+               printf("."); fflush(stdout);
+               memset(srcBuffer, srcByte, SIZE);
+
+               for(width=32; width<64; width++){
+                       int dstOffset;
+                       for(dstOffset=128; dstOffset<196; dstOffset++){
+                               int srcOffset;
+                               memset(dstBuffer, dstByte, SIZE);
+
+                               for(srcOffset=128; srcOffset<196; srcOffset++){
+                                       uint8_t *src= srcBuffer+srcOffset;
+                                       uint8_t *dst= dstBuffer+dstOffset;
+                                       char *name=NULL;
+                                       
+                                       if(failed) break; //don't fill the screen with shit ...
+
+                                       switch(funcNum){
+                                       case 0:
+                                               srcBpp=2;
+                                               dstBpp=2;
+                                               name="rgb15to16";
+                                               rgb15to16(src, dst, width*srcBpp);
+                                               break;
+                                       case 1:
+                                               srcBpp=2;
+                                               dstBpp=3;
+                                               name="rgb15to24";
+                                               rgb15to24(src, dst, width*srcBpp);
+                                               break;
+                                       case 2:
+                                               srcBpp=2;
+                                               dstBpp=4;
+                                               name="rgb15to32";
+                                               rgb15to32(src, dst, width*srcBpp);
+                                               break;
+                                       case 3:
+                                               srcBpp=2;
+                                               dstBpp=3;
+                                               name="rgb16to24";
+                                               rgb16to24(src, dst, width*srcBpp);
+                                               break;
+                                       case 4:
+                                               srcBpp=2;
+                                               dstBpp=4;
+                                               name="rgb16to32";
+                                               rgb16to32(src, dst, width*srcBpp);
+                                               break;
+                                       case 5:
+                                               srcBpp=3;
+                                               dstBpp=2;
+                                               name="rgb24to15";
+                                               rgb24to15(src, dst, width*srcBpp);
+                                               break;
+                                       case 6:
+                                               srcBpp=3;
+                                               dstBpp=2;
+                                               name="rgb24to16";
+                                               rgb24to16(src, dst, width*srcBpp);
+                                               break;
+                                       case 7:
+                                               srcBpp=3;
+                                               dstBpp=4;
+                                               name="rgb24to32";
+                                               rgb24to32(src, dst, width*srcBpp);
+                                               break;
+                                       case 8:
+                                               srcBpp=4;
+                                               dstBpp=2;
+                                               name="rgb32to15";
+                                               rgb32to15(src, dst, width*srcBpp);
+                                               break;
+                                       case 9:
+                                               srcBpp=4;
+                                               dstBpp=2;
+                                               name="rgb32to16";
+                                               rgb32to16(src, dst, width*srcBpp);
+                                               break;
+                                       case 10:
+                                               srcBpp=4;
+                                               dstBpp=3;
+                                               name="rgb32to24";
+                                               rgb32to24(src, dst, width*srcBpp);
+                                               break;
+                                       case 11:
+                                               srcBpp=2;
+                                               dstBpp=2;
+                                               name="rgb16to15";
+                                               rgb16to15(src, dst, width*srcBpp);
+                                               break;
+                                       
+                                       case 14:
+                                               srcBpp=2;
+                                               dstBpp=2;
+                                               name="rgb15tobgr15";
+                                               rgb15tobgr15(src, dst, width*srcBpp);
+                                               break;
+                                       case 15:
+                                               srcBpp=2;
+                                               dstBpp=2;
+                                               name="rgb15tobgr16";
+                                               rgb15tobgr16(src, dst, width*srcBpp);
+                                               break;
+                                       case 16:
+                                               srcBpp=2;
+                                               dstBpp=3;
+                                               name="rgb15tobgr24";
+                                               rgb15tobgr24(src, dst, width*srcBpp);
+                                               break;
+                                       case 17:
+                                               srcBpp=2;
+                                               dstBpp=4;
+                                               name="rgb15tobgr32";
+                                               rgb15tobgr32(src, dst, width*srcBpp);
+                                               break;
+                                       case 18:
+                                               srcBpp=2;
+                                               dstBpp=2;
+                                               name="rgb16tobgr15";
+                                               rgb16tobgr15(src, dst, width*srcBpp);
+                                               break;
+                                       case 19:
+                                               srcBpp=2;
+                                               dstBpp=2;
+                                               name="rgb16tobgr16";
+                                               rgb16tobgr16(src, dst, width*srcBpp);
+                                               break;
+                                       case 20:
+                                               srcBpp=2;
+                                               dstBpp=3;
+                                               name="rgb16tobgr24";
+                                               rgb16tobgr24(src, dst, width*srcBpp);
+                                               break;
+                                       case 21:
+                                               srcBpp=2;
+                                               dstBpp=4;
+                                               name="rgb16tobgr32";
+                                               rgb16tobgr32(src, dst, width*srcBpp);
+                                               break;
+                                       case 22:
+                                               srcBpp=3;
+                                               dstBpp=2;
+                                               name="rgb24tobgr15";
+                                               rgb24tobgr15(src, dst, width*srcBpp);
+                                               break;
+                                       case 23:
+                                               srcBpp=3;
+                                               dstBpp=2;
+                                               name="rgb24tobgr16";
+                                               rgb24tobgr16(src, dst, width*srcBpp);
+                                               break;
+                                       case 24:
+                                               srcBpp=3;
+                                               dstBpp=3;
+                                               name="rgb24tobgr24";
+                                               rgb24tobgr24(src, dst, width*srcBpp);
+                                               break;
+                                       case 25:
+                                               srcBpp=3;
+                                               dstBpp=4;
+                                               name="rgb24tobgr32";
+                                               rgb24tobgr32(src, dst, width*srcBpp);
+                                               break;
+                                       case 26:
+                                               srcBpp=4;
+                                               dstBpp=2;
+                                               name="rgb32tobgr15";
+                                               rgb32tobgr15(src, dst, width*srcBpp);
+                                               break;
+                                       case 27:
+                                               srcBpp=4;
+                                               dstBpp=2;
+                                               name="rgb32tobgr16";
+                                               rgb32tobgr16(src, dst, width*srcBpp);
+                                               break;
+                                       case 28:
+                                               srcBpp=4;
+                                               dstBpp=3;
+                                               name="rgb32tobgr24";
+                                               rgb32tobgr24(src, dst, width*srcBpp);
+                                               break;
+                                       case 29:
+                                               srcBpp=4;
+                                               dstBpp=4;
+                                               name="rgb32tobgr32";
+                                               rgb32tobgr32(src, dst, width*srcBpp);
+                                               break;
+
+                                       }
+                                       if(!srcBpp) break;
+
+                                       for(i=0; i<SIZE; i++){
+                                               if(srcBuffer[i]!=srcByte){
+                                                       printf("src damaged at %d w:%d src:%d dst:%d %s\n", 
+                                                               i, width, srcOffset, dstOffset, name);
+                                                       failed=1;
+                                                       break;
+                                               }
+                                       }
+                                       for(i=0; i<dstOffset; i++){
+                                               if(dstBuffer[i]!=dstByte){
+                                                       printf("dst damaged at %d w:%d src:%d dst:%d %s\n", 
+                                                               i, width, srcOffset, dstOffset, name);
+                                                       failed=1;
+                                                       break;
+                                               }
+                                       }
+                                       for(i=dstOffset + width*dstBpp; i<SIZE; i++){
+                                               if(dstBuffer[i]!=dstByte){
+                                                       printf("dst damaged at %d w:%d src:%d dst:%d %s\n", 
+                                                               i, width, srcOffset, dstOffset, name);
+                                                       failed=1;
+                                                       break;
+                                               }
+                                       }
+                               }
+                       }
+               }
+               if(failed) failedNum++;
+               else if(srcBpp) passedNum++;
+       }
+       
+       printf("%d converters passed, %d converters randomly overwrote memory\n", passedNum, failedNum);
+       return failedNum;
+}
diff --git a/libswscale/rgb2rgb.c b/libswscale/rgb2rgb.c
new file mode 100644 (file)
index 0000000..049e105
--- /dev/null
@@ -0,0 +1,644 @@
+/*
+ *
+ *  rgb2rgb.c, Software RGB to RGB convertor
+ *  pluralize by Software PAL8 to RGB convertor
+ *               Software YUV to YUV convertor
+ *               Software YUV to RGB convertor
+ *  Written by Nick Kurshev.
+ *  palette & yuv & runtime cpu stuff by Michael (michaelni@gmx.at) (under GPL)
+ */
+#include <inttypes.h>
+#include "config.h"
+#include "rgb2rgb.h"
+#include "swscale.h"
+#include "cpudetect.h"
+#include "mangle.h"
+#include "bswap.h"
+#include "libvo/fastmemcpy.h"
+
+#define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
+
+void (*rgb24to32)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb24to16)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb24to15)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb32to24)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb32to16)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb32to15)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb15to16)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb15to24)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb15to32)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb16to15)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb16to24)(const uint8_t *src,uint8_t *dst,long src_size);
+void (*rgb16to32)(const uint8_t *src,uint8_t *dst,long src_size);
+//void (*rgb24tobgr32)(const uint8_t *src, uint8_t *dst, long src_size);
+void (*rgb24tobgr24)(const uint8_t *src, uint8_t *dst, long src_size);
+void (*rgb24tobgr16)(const uint8_t *src, uint8_t *dst, long src_size);
+void (*rgb24tobgr15)(const uint8_t *src, uint8_t *dst, long src_size);
+void (*rgb32tobgr32)(const uint8_t *src, uint8_t *dst, long src_size);
+//void (*rgb32tobgr24)(const uint8_t *src, uint8_t *dst, long src_size);
+void (*rgb32tobgr16)(const uint8_t *src, uint8_t *dst, long src_size);
+void (*rgb32tobgr15)(const uint8_t *src, uint8_t *dst, long src_size);
+
+void (*yv12toyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride);
+void (*yv12touyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride);
+void (*yuv422ptoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride);
+void (*yuy2toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+       long width, long height,
+       long lumStride, long chromStride, long srcStride);
+void (*rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+       long width, long height,
+       long lumStride, long chromStride, long srcStride);
+void (*planar2x)(const uint8_t *src, uint8_t *dst, long width, long height,
+       long srcStride, long dstStride);
+void (*interleaveBytes)(uint8_t *src1, uint8_t *src2, uint8_t *dst,
+                           long width, long height, long src1Stride,
+                           long src2Stride, long dstStride);
+void (*vu9_to_vu12)(const uint8_t *src1, const uint8_t *src2,
+                       uint8_t *dst1, uint8_t *dst2,
+                       long width, long height,
+                       long srcStride1, long srcStride2,
+                       long dstStride1, long dstStride2);
+void (*yvu9_to_yuy2)(const uint8_t *src1, const uint8_t *src2, const uint8_t *src3,
+                       uint8_t *dst,
+                       long width, long height,
+                       long srcStride1, long srcStride2,
+                       long srcStride3, long dstStride);
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+static const uint64_t mmx_null  __attribute__((aligned(8))) = 0x0000000000000000ULL;
+static const uint64_t mmx_one   __attribute__((aligned(8))) = 0xFFFFFFFFFFFFFFFFULL;
+static const uint64_t mask32b  attribute_used __attribute__((aligned(8))) = 0x000000FF000000FFULL;
+static const uint64_t mask32g  attribute_used __attribute__((aligned(8))) = 0x0000FF000000FF00ULL;
+static const uint64_t mask32r  attribute_used __attribute__((aligned(8))) = 0x00FF000000FF0000ULL;
+static const uint64_t mask32   __attribute__((aligned(8))) = 0x00FFFFFF00FFFFFFULL;
+static const uint64_t mask3216br __attribute__((aligned(8)))=0x00F800F800F800F8ULL;
+static const uint64_t mask3216g  __attribute__((aligned(8)))=0x0000FC000000FC00ULL;
+static const uint64_t mask3215g  __attribute__((aligned(8)))=0x0000F8000000F800ULL;
+static const uint64_t mul3216  __attribute__((aligned(8))) = 0x2000000420000004ULL;
+static const uint64_t mul3215  __attribute__((aligned(8))) = 0x2000000820000008ULL;
+static const uint64_t mask24b  attribute_used __attribute__((aligned(8))) = 0x00FF0000FF0000FFULL;
+static const uint64_t mask24g  attribute_used __attribute__((aligned(8))) = 0xFF0000FF0000FF00ULL;
+static const uint64_t mask24r  attribute_used __attribute__((aligned(8))) = 0x0000FF0000FF0000ULL;
+static const uint64_t mask24l  __attribute__((aligned(8))) = 0x0000000000FFFFFFULL;
+static const uint64_t mask24h  __attribute__((aligned(8))) = 0x0000FFFFFF000000ULL;
+static const uint64_t mask24hh  __attribute__((aligned(8))) = 0xffff000000000000ULL;
+static const uint64_t mask24hhh  __attribute__((aligned(8))) = 0xffffffff00000000ULL;
+static const uint64_t mask24hhhh  __attribute__((aligned(8))) = 0xffffffffffff0000ULL;
+static const uint64_t mask15b  __attribute__((aligned(8))) = 0x001F001F001F001FULL; /* 00000000 00011111  xxB */
+static const uint64_t mask15rg __attribute__((aligned(8))) = 0x7FE07FE07FE07FE0ULL; /* 01111111 11100000  RGx */
+static const uint64_t mask15s  __attribute__((aligned(8))) = 0xFFE0FFE0FFE0FFE0ULL;
+static const uint64_t mask15g  __attribute__((aligned(8))) = 0x03E003E003E003E0ULL;
+static const uint64_t mask15r  __attribute__((aligned(8))) = 0x7C007C007C007C00ULL;
+#define mask16b mask15b
+static const uint64_t mask16g  __attribute__((aligned(8))) = 0x07E007E007E007E0ULL;
+static const uint64_t mask16r  __attribute__((aligned(8))) = 0xF800F800F800F800ULL;
+static const uint64_t red_16mask  __attribute__((aligned(8))) = 0x0000f8000000f800ULL;
+static const uint64_t green_16mask __attribute__((aligned(8)))= 0x000007e0000007e0ULL;
+static const uint64_t blue_16mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
+static const uint64_t red_15mask  __attribute__((aligned(8))) = 0x00007c000000f800ULL;
+static const uint64_t green_15mask __attribute__((aligned(8)))= 0x000003e0000007e0ULL;
+static const uint64_t blue_15mask __attribute__((aligned(8))) = 0x0000001f0000001fULL;
+
+#ifdef FAST_BGR2YV12
+static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
+static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
+static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
+#else
+static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
+static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
+static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
+#endif
+static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
+static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
+static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
+
+#if 0
+static volatile uint64_t __attribute__((aligned(8))) b5Dither;
+static volatile uint64_t __attribute__((aligned(8))) g5Dither;
+static volatile uint64_t __attribute__((aligned(8))) g6Dither;
+static volatile uint64_t __attribute__((aligned(8))) r5Dither;
+
+static uint64_t __attribute__((aligned(8))) dither4[2]={
+       0x0103010301030103LL,
+       0x0200020002000200LL,};
+
+static uint64_t __attribute__((aligned(8))) dither8[2]={
+       0x0602060206020602LL,
+       0x0004000400040004LL,};
+#endif
+#endif
+
+#define RGB2YUV_SHIFT 8
+#define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
+#define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
+#define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
+#define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
+#define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
+#define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
+#define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
+#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
+#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
+
+//Note: we have C, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
+//Plain C versions
+#undef HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#undef HAVE_SSE2
+#define RENAME(a) a ## _C
+#include "rgb2rgb_template.c"
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+
+//MMX versions
+#undef RENAME
+#define HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#undef HAVE_SSE2
+#define RENAME(a) a ## _MMX
+#include "rgb2rgb_template.c"
+
+//MMX2 versions
+#undef RENAME
+#define HAVE_MMX
+#define HAVE_MMX2
+#undef HAVE_3DNOW
+#undef HAVE_SSE2
+#define RENAME(a) a ## _MMX2
+#include "rgb2rgb_template.c"
+
+//3DNOW versions
+#undef RENAME
+#define HAVE_MMX
+#undef HAVE_MMX2
+#define HAVE_3DNOW
+#undef HAVE_SSE2
+#define RENAME(a) a ## _3DNOW
+#include "rgb2rgb_template.c"
+
+#endif //ARCH_X86 || ARCH_X86_64
+
+/*
+ rgb15->rgb16 Original by Strepto/Astral
+ ported to gcc & bugfixed : A'rpi
+ MMX2, 3DNOW optimization by Nick Kurshev
+ 32bit c version, and and&add trick by Michael Niedermayer
+*/
+
+void sws_rgb2rgb_init(int flags){
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+       if(flags & SWS_CPU_CAPS_MMX2){
+               rgb15to16= rgb15to16_MMX2;
+               rgb15to24= rgb15to24_MMX2;
+               rgb15to32= rgb15to32_MMX2;
+               rgb16to24= rgb16to24_MMX2;
+               rgb16to32= rgb16to32_MMX2;
+               rgb16to15= rgb16to15_MMX2;
+               rgb24to16= rgb24to16_MMX2;
+               rgb24to15= rgb24to15_MMX2;
+               rgb24to32= rgb24to32_MMX2;
+               rgb32to16= rgb32to16_MMX2;
+               rgb32to15= rgb32to15_MMX2;
+               rgb32to24= rgb32to24_MMX2;
+               rgb24tobgr15= rgb24tobgr15_MMX2;
+               rgb24tobgr16= rgb24tobgr16_MMX2;
+               rgb24tobgr24= rgb24tobgr24_MMX2;
+               rgb32tobgr32= rgb32tobgr32_MMX2;
+               rgb32tobgr16= rgb32tobgr16_MMX2;
+               rgb32tobgr15= rgb32tobgr15_MMX2;
+               yv12toyuy2= yv12toyuy2_MMX2;
+               yv12touyvy= yv12touyvy_MMX2;
+               yuv422ptoyuy2= yuv422ptoyuy2_MMX2;
+               yuy2toyv12= yuy2toyv12_MMX2;
+//             uyvytoyv12= uyvytoyv12_MMX2;
+//             yvu9toyv12= yvu9toyv12_MMX2;
+               planar2x= planar2x_MMX2;
+               rgb24toyv12= rgb24toyv12_MMX2;
+               interleaveBytes= interleaveBytes_MMX2;
+               vu9_to_vu12= vu9_to_vu12_MMX2;
+               yvu9_to_yuy2= yvu9_to_yuy2_MMX2;
+       }else if(flags & SWS_CPU_CAPS_3DNOW){
+               rgb15to16= rgb15to16_3DNOW;
+               rgb15to24= rgb15to24_3DNOW;
+               rgb15to32= rgb15to32_3DNOW;
+               rgb16to24= rgb16to24_3DNOW;
+               rgb16to32= rgb16to32_3DNOW;
+               rgb16to15= rgb16to15_3DNOW;
+               rgb24to16= rgb24to16_3DNOW;
+               rgb24to15= rgb24to15_3DNOW;
+               rgb24to32= rgb24to32_3DNOW;
+               rgb32to16= rgb32to16_3DNOW;
+               rgb32to15= rgb32to15_3DNOW;
+               rgb32to24= rgb32to24_3DNOW;
+               rgb24tobgr15= rgb24tobgr15_3DNOW;
+               rgb24tobgr16= rgb24tobgr16_3DNOW;
+               rgb24tobgr24= rgb24tobgr24_3DNOW;
+               rgb32tobgr32= rgb32tobgr32_3DNOW;
+               rgb32tobgr16= rgb32tobgr16_3DNOW;
+               rgb32tobgr15= rgb32tobgr15_3DNOW;
+               yv12toyuy2= yv12toyuy2_3DNOW;
+               yv12touyvy= yv12touyvy_3DNOW;
+               yuv422ptoyuy2= yuv422ptoyuy2_3DNOW;
+               yuy2toyv12= yuy2toyv12_3DNOW;
+//             uyvytoyv12= uyvytoyv12_3DNOW;
+//             yvu9toyv12= yvu9toyv12_3DNOW;
+               planar2x= planar2x_3DNOW;
+               rgb24toyv12= rgb24toyv12_3DNOW;
+               interleaveBytes= interleaveBytes_3DNOW;
+               vu9_to_vu12= vu9_to_vu12_3DNOW;
+               yvu9_to_yuy2= yvu9_to_yuy2_3DNOW;
+       }else if(flags & SWS_CPU_CAPS_MMX){
+               rgb15to16= rgb15to16_MMX;
+               rgb15to24= rgb15to24_MMX;
+               rgb15to32= rgb15to32_MMX;
+               rgb16to24= rgb16to24_MMX;
+               rgb16to32= rgb16to32_MMX;
+               rgb16to15= rgb16to15_MMX;
+               rgb24to16= rgb24to16_MMX;
+               rgb24to15= rgb24to15_MMX;
+               rgb24to32= rgb24to32_MMX;
+               rgb32to16= rgb32to16_MMX;
+               rgb32to15= rgb32to15_MMX;
+               rgb32to24= rgb32to24_MMX;
+               rgb24tobgr15= rgb24tobgr15_MMX;
+               rgb24tobgr16= rgb24tobgr16_MMX;
+               rgb24tobgr24= rgb24tobgr24_MMX;
+               rgb32tobgr32= rgb32tobgr32_MMX;
+               rgb32tobgr16= rgb32tobgr16_MMX;
+               rgb32tobgr15= rgb32tobgr15_MMX;
+               yv12toyuy2= yv12toyuy2_MMX;
+               yv12touyvy= yv12touyvy_MMX;
+               yuv422ptoyuy2= yuv422ptoyuy2_MMX;
+               yuy2toyv12= yuy2toyv12_MMX;
+//             uyvytoyv12= uyvytoyv12_MMX;
+//             yvu9toyv12= yvu9toyv12_MMX;
+               planar2x= planar2x_MMX;
+               rgb24toyv12= rgb24toyv12_MMX;
+               interleaveBytes= interleaveBytes_MMX;
+               vu9_to_vu12= vu9_to_vu12_MMX;
+               yvu9_to_yuy2= yvu9_to_yuy2_MMX;
+       }else
+#endif
+       {
+               rgb15to16= rgb15to16_C;
+               rgb15to24= rgb15to24_C;
+               rgb15to32= rgb15to32_C;
+               rgb16to24= rgb16to24_C;
+               rgb16to32= rgb16to32_C;
+               rgb16to15= rgb16to15_C;
+               rgb24to16= rgb24to16_C;
+               rgb24to15= rgb24to15_C;
+               rgb24to32= rgb24to32_C;
+               rgb32to16= rgb32to16_C;
+               rgb32to15= rgb32to15_C;
+               rgb32to24= rgb32to24_C;
+               rgb24tobgr15= rgb24tobgr15_C;
+               rgb24tobgr16= rgb24tobgr16_C;
+               rgb24tobgr24= rgb24tobgr24_C;
+               rgb32tobgr32= rgb32tobgr32_C;
+               rgb32tobgr16= rgb32tobgr16_C;
+               rgb32tobgr15= rgb32tobgr15_C;
+               yv12toyuy2= yv12toyuy2_C;
+               yv12touyvy= yv12touyvy_C;
+               yuv422ptoyuy2= yuv422ptoyuy2_C;
+               yuy2toyv12= yuy2toyv12_C;
+//             uyvytoyv12= uyvytoyv12_C;
+//             yvu9toyv12= yvu9toyv12_C;
+               planar2x= planar2x_C;
+               rgb24toyv12= rgb24toyv12_C;
+               interleaveBytes= interleaveBytes_C;
+               vu9_to_vu12= vu9_to_vu12_C;
+               yvu9_to_yuy2= yvu9_to_yuy2_C;
+       }
+}
+
+/**
+ * Pallete is assumed to contain bgr32
+ */
+void palette8torgb32(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette)
+{
+       long i;
+
+/*
+       for(i=0; i<num_pixels; i++)
+               ((unsigned *)dst)[i] = ((unsigned *)palette)[ src[i] ];
+*/
+
+       for(i=0; i<num_pixels; i++)
+       {
+               #ifdef WORDS_BIGENDIAN
+                       dst[3]= palette[ src[i]*4+2 ];
+                       dst[2]= palette[ src[i]*4+1 ];
+                       dst[1]= palette[ src[i]*4+0 ];
+               #else
+               //FIXME slow?
+                       dst[0]= palette[ src[i]*4+2 ];
+                       dst[1]= palette[ src[i]*4+1 ];
+                       dst[2]= palette[ src[i]*4+0 ];
+                       //dst[3]= 0; /* do we need this cleansing? */
+               #endif
+               dst+= 4;
+       }
+}
+
+void palette8tobgr32(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette)
+{
+       long i;
+       for(i=0; i<num_pixels; i++)
+       {
+               #ifdef WORDS_BIGENDIAN
+                       dst[3]= palette[ src[i]*4+0 ];
+                       dst[2]= palette[ src[i]*4+1 ];
+                       dst[1]= palette[ src[i]*4+2 ];
+               #else
+                       //FIXME slow?
+                       dst[0]= palette[ src[i]*4+0 ];
+                       dst[1]= palette[ src[i]*4+1 ];
+                       dst[2]= palette[ src[i]*4+2 ];
+                       //dst[3]= 0; /* do we need this cleansing? */
+               #endif
+               
+               dst+= 4;
+       }
+}
+
+/**
+ * Pallete is assumed to contain bgr32
+ */
+void palette8torgb24(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette)
+{
+       long i;
+/*
+       writes 1 byte o much and might cause alignment issues on some architectures?
+       for(i=0; i<num_pixels; i++)
+               ((unsigned *)(&dst[i*3])) = ((unsigned *)palette)[ src[i] ];
+*/
+       for(i=0; i<num_pixels; i++)
+       {
+               //FIXME slow?
+               dst[0]= palette[ src[i]*4+2 ];
+               dst[1]= palette[ src[i]*4+1 ];
+               dst[2]= palette[ src[i]*4+0 ];
+               dst+= 3;
+       }
+}
+
+void palette8tobgr24(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette)
+{
+       long i;
+/*
+       writes 1 byte o much and might cause alignment issues on some architectures?
+       for(i=0; i<num_pixels; i++)
+               ((unsigned *)(&dst[i*3])) = ((unsigned *)palette)[ src[i] ];
+*/
+       for(i=0; i<num_pixels; i++)
+       {
+               //FIXME slow?
+               dst[0]= palette[ src[i]*4+0 ];
+               dst[1]= palette[ src[i]*4+1 ];
+               dst[2]= palette[ src[i]*4+2 ];
+               dst+= 3;
+       }
+}
+
+/**
+ * Palette is assumed to contain bgr16, see rgb32to16 to convert the palette
+ */
+void palette8torgb16(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette)
+{
+       long i;
+       for(i=0; i<num_pixels; i++)
+               ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
+}
+void palette8tobgr16(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette)
+{
+       long i;
+       for(i=0; i<num_pixels; i++)
+               ((uint16_t *)dst)[i] = bswap_16(((uint16_t *)palette)[ src[i] ]);
+}
+
+/**
+ * Pallete is assumed to contain bgr15, see rgb32to15 to convert the palette
+ */
+void palette8torgb15(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette)
+{
+       long i;
+       for(i=0; i<num_pixels; i++)
+               ((uint16_t *)dst)[i] = ((uint16_t *)palette)[ src[i] ];
+}
+void palette8tobgr15(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette)
+{
+       long i;
+       for(i=0; i<num_pixels; i++)
+               ((uint16_t *)dst)[i] = bswap_16(((uint16_t *)palette)[ src[i] ]);
+}
+
+void rgb32tobgr24(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       long i;
+       long num_pixels = src_size >> 2;
+       for(i=0; i<num_pixels; i++)
+       {
+               #ifdef WORDS_BIGENDIAN
+                       /* RGB32 (= A,B,G,R) -> BGR24 (= B,G,R) */
+                       dst[3*i + 0] = src[4*i + 1];
+                       dst[3*i + 1] = src[4*i + 2];
+                       dst[3*i + 2] = src[4*i + 3];
+               #else
+                       dst[3*i + 0] = src[4*i + 2];
+                       dst[3*i + 1] = src[4*i + 1];
+                       dst[3*i + 2] = src[4*i + 0];
+               #endif
+       }
+}
+
+void rgb24tobgr32(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       long i;
+       for(i=0; 3*i<src_size; i++)
+       {
+               #ifdef WORDS_BIGENDIAN
+                       /* RGB24 (= R,G,B) -> BGR32 (= A,R,G,B) */
+                       dst[4*i + 0] = 0;
+                       dst[4*i + 1] = src[3*i + 0];
+                       dst[4*i + 2] = src[3*i + 1];
+                       dst[4*i + 3] = src[3*i + 2];
+               #else
+                       dst[4*i + 0] = src[3*i + 2];
+                       dst[4*i + 1] = src[3*i + 1];
+                       dst[4*i + 2] = src[3*i + 0];
+                       dst[4*i + 3] = 0;
+               #endif
+       }
+}
+
+void rgb16tobgr32(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint16_t *end;
+       uint8_t *d = (uint8_t *)dst;
+       const uint16_t *s = (uint16_t *)src;
+       end = s + src_size/2;
+       while(s < end)
+       {
+               register uint16_t bgr;
+               bgr = *s++;
+               #ifdef WORDS_BIGENDIAN
+                       *d++ = 0;
+                       *d++ = (bgr&0x1F)<<3;
+                       *d++ = (bgr&0x7E0)>>3;
+                       *d++ = (bgr&0xF800)>>8;
+               #else
+                       *d++ = (bgr&0xF800)>>8;
+                       *d++ = (bgr&0x7E0)>>3;
+                       *d++ = (bgr&0x1F)<<3;
+                       *d++ = 0;
+               #endif
+       }
+}
+
+void rgb16tobgr24(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint16_t *end;
+       uint8_t *d = (uint8_t *)dst;
+       const uint16_t *s = (const uint16_t *)src;
+       end = s + src_size/2;
+       while(s < end)
+       {
+               register uint16_t bgr;
+               bgr = *s++;
+               *d++ = (bgr&0xF800)>>8;
+               *d++ = (bgr&0x7E0)>>3;
+               *d++ = (bgr&0x1F)<<3;
+       }
+}
+
+void rgb16tobgr16(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       long i;
+       long num_pixels = src_size >> 1;
+       
+       for(i=0; i<num_pixels; i++)
+       {
+           unsigned b,g,r;
+           register uint16_t rgb;
+           rgb = src[2*i];
+           r = rgb&0x1F;
+           g = (rgb&0x7E0)>>5;
+           b = (rgb&0xF800)>>11;
+           dst[2*i] = (b&0x1F) | ((g&0x3F)<<5) | ((r&0x1F)<<11);
+       }
+}
+
+void rgb16tobgr15(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       long i;
+       long num_pixels = src_size >> 1;
+       
+       for(i=0; i<num_pixels; i++)
+       {
+           unsigned b,g,r;
+           register uint16_t rgb;
+           rgb = src[2*i];
+           r = rgb&0x1F;
+           g = (rgb&0x7E0)>>5;
+           b = (rgb&0xF800)>>11;
+           dst[2*i] = (b&0x1F) | ((g&0x1F)<<5) | ((r&0x1F)<<10);
+       }
+}
+
+void rgb15tobgr32(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint16_t *end;
+       uint8_t *d = (uint8_t *)dst;
+       const uint16_t *s = (const uint16_t *)src;
+       end = s + src_size/2;
+       while(s < end)
+       {
+               register uint16_t bgr;
+               bgr = *s++;
+               #ifdef WORDS_BIGENDIAN
+                       *d++ = 0;
+                       *d++ = (bgr&0x1F)<<3;
+                       *d++ = (bgr&0x3E0)>>2;
+                       *d++ = (bgr&0x7C00)>>7;
+               #else
+                       *d++ = (bgr&0x7C00)>>7;
+                       *d++ = (bgr&0x3E0)>>2;
+                       *d++ = (bgr&0x1F)<<3;
+                       *d++ = 0;
+               #endif
+       }
+}
+
+void rgb15tobgr24(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint16_t *end;
+       uint8_t *d = (uint8_t *)dst;
+       const uint16_t *s = (uint16_t *)src;
+       end = s + src_size/2;
+       while(s < end)
+       {
+               register uint16_t bgr;
+               bgr = *s++;
+               *d++ = (bgr&0x7C00)>>7;
+               *d++ = (bgr&0x3E0)>>2;
+               *d++ = (bgr&0x1F)<<3;
+       }
+}
+
+void rgb15tobgr16(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       long i;
+       long num_pixels = src_size >> 1;
+       
+       for(i=0; i<num_pixels; i++)
+       {
+           unsigned b,g,r;
+           register uint16_t rgb;
+           rgb = src[2*i];
+           r = rgb&0x1F;
+           g = (rgb&0x3E0)>>5;
+           b = (rgb&0x7C00)>>10;
+           dst[2*i] = (b&0x1F) | ((g&0x3F)<<5) | ((r&0x1F)<<11);
+       }
+}
+
+void rgb15tobgr15(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       long i;
+       long num_pixels = src_size >> 1;
+       
+       for(i=0; i<num_pixels; i++)
+       {
+           unsigned b,g,r;
+           register uint16_t rgb;
+           rgb = src[2*i];
+           r = rgb&0x1F;
+           g = (rgb&0x3E0)>>5;
+           b = (rgb&0x7C00)>>10;
+           dst[2*i] = (b&0x1F) | ((g&0x1F)<<5) | ((r&0x1F)<<10);
+       }
+}
+
+void rgb8tobgr8(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       long i;
+       long num_pixels = src_size;
+       for(i=0; i<num_pixels; i++)
+       {
+           unsigned b,g,r;
+           register uint8_t rgb;
+           rgb = src[i];
+           r = (rgb&0x07);
+           g = (rgb&0x38)>>3;
+           b = (rgb&0xC0)>>6;
+           dst[i] = ((b<<1)&0x07) | ((g&0x07)<<3) | ((r&0x03)<<6);
+       }
+}
diff --git a/libswscale/rgb2rgb.h b/libswscale/rgb2rgb.h
new file mode 100644 (file)
index 0000000..e38893d
--- /dev/null
@@ -0,0 +1,144 @@
+/*
+ *
+ *  rgb2rgb.h, Software RGB to RGB convertor
+ *  pluralize by Software PAL8 to RGB convertor
+ *               Software YUV to YUV convertor
+ *               Software YUV to RGB convertor
+ */
+
+#ifndef RGB2RGB_INCLUDED
+#define RGB2RGB_INCLUDED
+
+// Note: do not fix the dependence on stdio.h
+
+/* A full collection of rgb to rgb(bgr) convertors */
+extern void (*rgb24to32)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb24to16)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb24to15)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb32to24)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb32to16)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb32to15)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb15to16)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb15to24)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb15to32)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb16to15)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb16to24)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb16to32)(const uint8_t *src,uint8_t *dst,long src_size);
+extern void (*rgb24tobgr24)(const uint8_t *src, uint8_t *dst, long src_size);
+extern void (*rgb24tobgr16)(const uint8_t *src, uint8_t *dst, long src_size);
+extern void (*rgb24tobgr15)(const uint8_t *src, uint8_t *dst, long src_size);
+extern void (*rgb32tobgr32)(const uint8_t *src, uint8_t *dst, long src_size);
+extern void (*rgb32tobgr16)(const uint8_t *src, uint8_t *dst, long src_size);
+extern void (*rgb32tobgr15)(const uint8_t *src, uint8_t *dst, long src_size);
+
+extern void rgb24tobgr32(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb32tobgr24(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb16tobgr32(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb16tobgr24(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb16tobgr16(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb16tobgr15(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb15tobgr32(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb15tobgr24(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb15tobgr16(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb15tobgr15(const uint8_t *src, uint8_t *dst, long src_size);
+extern void rgb8tobgr8(const uint8_t *src, uint8_t *dst, long src_size);
+
+
+extern void palette8torgb32(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette);
+extern void palette8tobgr32(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette);
+extern void palette8torgb24(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette);
+extern void palette8tobgr24(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette);
+extern void palette8torgb16(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette);
+extern void palette8tobgr16(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette);
+extern void palette8torgb15(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette);
+extern void palette8tobgr15(const uint8_t *src, uint8_t *dst, long num_pixels, const uint8_t *palette);
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
+ */
+//void uyvytoyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ */
+extern void (*yv12toyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride);
+
+/**
+ *
+ * width should be a multiple of 16
+ */
+extern void (*yuv422ptoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride);
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ */
+extern void (*yuy2toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+       long width, long height,
+       long lumStride, long chromStride, long srcStride);
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ */
+extern void (*yv12touyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride);
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 2 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
+ */
+extern void (*rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+       long width, long height,
+       long lumStride, long chromStride, long srcStride);
+extern void (*planar2x)(const uint8_t *src, uint8_t *dst, long width, long height,
+       long srcStride, long dstStride);
+
+extern void (*interleaveBytes)(uint8_t *src1, uint8_t *src2, uint8_t *dst,
+                           long width, long height, long src1Stride,
+                           long src2Stride, long dstStride);
+
+extern void (*vu9_to_vu12)(const uint8_t *src1, const uint8_t *src2,
+                       uint8_t *dst1, uint8_t *dst2,
+                       long width, long height,
+                       long srcStride1, long srcStride2,
+                       long dstStride1, long dstStride2);
+
+extern void (*yvu9_to_yuy2)(const uint8_t *src1, const uint8_t *src2, const uint8_t *src3,
+                       uint8_t *dst,
+                       long width, long height,
+                       long srcStride1, long srcStride2,
+                       long srcStride3, long dstStride);
+       
+
+#define MODE_RGB  0x1
+#define MODE_BGR  0x2
+
+static void yuv2rgb(uint8_t * image, uint8_t * py,
+                             uint8_t * pu, uint8_t * pv,
+                             unsigned h_size, unsigned v_size,
+                             int rgb_stride, int y_stride, int uv_stride){
+printf("broken, this should use the swscaler\n");
+}
+
+static void yuv2rgb_init (unsigned bpp, int mode){
+printf("broken, this should use the swscaler\n");
+}
+
+void sws_rgb2rgb_init(int flags);
+
+#endif
diff --git a/libswscale/rgb2rgb_template.c b/libswscale/rgb2rgb_template.c
new file mode 100644 (file)
index 0000000..807da61
--- /dev/null
@@ -0,0 +1,2675 @@
+/*
+ *
+ *  rgb2rgb.c, Software RGB to RGB convertor
+ *  pluralize by Software PAL8 to RGB convertor
+ *               Software YUV to YUV convertor
+ *               Software YUV to RGB convertor
+ *  Written by Nick Kurshev.
+ *  palette & yuv & runtime cpu stuff by Michael (michaelni@gmx.at) (under GPL)
+ *  lot of big-endian byteorder fixes by Alex Beregszaszi
+ */
+
+#include <stddef.h>
+#include <inttypes.h> /* for __WORDSIZE */
+
+#include "asmalign.h"
+
+#ifndef __WORDSIZE
+// #warning You have misconfigured system and probably will lose performance!
+#define __WORDSIZE MP_WORDSIZE
+#endif
+
+#undef PREFETCH
+#undef MOVNTQ
+#undef EMMS
+#undef SFENCE
+#undef MMREG_SIZE
+#undef PREFETCHW
+#undef PAVGB
+
+#ifdef HAVE_SSE2
+#define MMREG_SIZE 16
+#else
+#define MMREG_SIZE 8
+#endif
+
+#ifdef HAVE_3DNOW
+#define PREFETCH  "prefetch"
+#define PREFETCHW "prefetchw"
+#define PAVGB    "pavgusb"
+#elif defined ( HAVE_MMX2 )
+#define PREFETCH "prefetchnta"
+#define PREFETCHW "prefetcht0"
+#define PAVGB    "pavgb"
+#else
+#ifdef __APPLE__
+#define PREFETCH "#"
+#define PREFETCHW "#"
+#else
+#define PREFETCH "/nop"
+#define PREFETCHW "/nop"
+#endif
+#endif
+
+#ifdef HAVE_3DNOW
+/* On K6 femms is faster of emms. On K7 femms is directly mapped on emms. */
+#define EMMS     "femms"
+#else
+#define EMMS     "emms"
+#endif
+
+#ifdef HAVE_MMX2
+#define MOVNTQ "movntq"
+#define SFENCE "sfence"
+#else
+#define MOVNTQ "movq"
+#ifdef __APPLE__
+#define SFENCE "#"
+#else
+#define SFENCE "/nop"
+#endif
+#endif
+
+static inline void RENAME(rgb24to32)(const uint8_t *src,uint8_t *dst,long src_size)
+{
+  uint8_t *dest = dst;
+  const uint8_t *s = src;
+  const uint8_t *end;
+#ifdef HAVE_MMX
+  const uint8_t *mm_end;
+#endif
+  end = s + src_size;
+#ifdef HAVE_MMX
+  __asm __volatile(PREFETCH"   %0"::"m"(*s):"memory");
+  mm_end = end - 23;
+  __asm __volatile("movq       %0, %%mm7"::"m"(mask32):"memory");
+  while(s < mm_end)
+  {
+    __asm __volatile(
+       PREFETCH"       32%1\n\t"
+       "movd   %1, %%mm0\n\t"
+       "punpckldq 3%1, %%mm0\n\t"
+       "movd   6%1, %%mm1\n\t"
+       "punpckldq 9%1, %%mm1\n\t"
+       "movd   12%1, %%mm2\n\t"
+       "punpckldq 15%1, %%mm2\n\t"
+       "movd   18%1, %%mm3\n\t"
+       "punpckldq 21%1, %%mm3\n\t"
+       "pand   %%mm7, %%mm0\n\t"
+       "pand   %%mm7, %%mm1\n\t"
+       "pand   %%mm7, %%mm2\n\t"
+       "pand   %%mm7, %%mm3\n\t"
+       MOVNTQ" %%mm0, %0\n\t"
+       MOVNTQ" %%mm1, 8%0\n\t"
+       MOVNTQ" %%mm2, 16%0\n\t"
+       MOVNTQ" %%mm3, 24%0"
+       :"=m"(*dest)
+       :"m"(*s)
+       :"memory");
+    dest += 32;
+    s += 24;
+  }
+  __asm __volatile(SFENCE:::"memory");
+  __asm __volatile(EMMS:::"memory");
+#endif
+  while(s < end)
+  {
+#ifdef WORDS_BIGENDIAN
+    /* RGB24 (= R,G,B) -> RGB32 (= A,B,G,R) */
+    *dest++ = 0;
+    *dest++ = s[2];
+    *dest++ = s[1];
+    *dest++ = s[0];
+    s+=3;
+#else
+    *dest++ = *s++;
+    *dest++ = *s++;
+    *dest++ = *s++;
+    *dest++ = 0;
+#endif
+  }
+}
+
+static inline void RENAME(rgb32to24)(const uint8_t *src,uint8_t *dst,long src_size)
+{
+  uint8_t *dest = dst;
+  const uint8_t *s = src;
+  const uint8_t *end;
+#ifdef HAVE_MMX
+  const uint8_t *mm_end;
+#endif
+  end = s + src_size;
+#ifdef HAVE_MMX
+  __asm __volatile(PREFETCH"   %0"::"m"(*s):"memory");
+  mm_end = end - 31;
+  while(s < mm_end)
+  {
+    __asm __volatile(
+       PREFETCH"       32%1\n\t"
+       "movq   %1, %%mm0\n\t"
+       "movq   8%1, %%mm1\n\t"
+       "movq   16%1, %%mm4\n\t"
+       "movq   24%1, %%mm5\n\t"
+       "movq   %%mm0, %%mm2\n\t"
+       "movq   %%mm1, %%mm3\n\t"
+       "movq   %%mm4, %%mm6\n\t"
+       "movq   %%mm5, %%mm7\n\t"
+       "psrlq  $8, %%mm2\n\t"
+       "psrlq  $8, %%mm3\n\t"
+       "psrlq  $8, %%mm6\n\t"
+       "psrlq  $8, %%mm7\n\t"
+       "pand   %2, %%mm0\n\t"
+       "pand   %2, %%mm1\n\t"
+       "pand   %2, %%mm4\n\t"
+       "pand   %2, %%mm5\n\t"
+       "pand   %3, %%mm2\n\t"
+       "pand   %3, %%mm3\n\t"
+       "pand   %3, %%mm6\n\t"
+       "pand   %3, %%mm7\n\t"
+       "por    %%mm2, %%mm0\n\t"
+       "por    %%mm3, %%mm1\n\t"
+       "por    %%mm6, %%mm4\n\t"
+       "por    %%mm7, %%mm5\n\t"
+
+       "movq   %%mm1, %%mm2\n\t"
+       "movq   %%mm4, %%mm3\n\t"
+       "psllq  $48, %%mm2\n\t"
+       "psllq  $32, %%mm3\n\t"
+       "pand   %4, %%mm2\n\t"
+       "pand   %5, %%mm3\n\t"
+       "por    %%mm2, %%mm0\n\t"
+       "psrlq  $16, %%mm1\n\t"
+       "psrlq  $32, %%mm4\n\t"
+       "psllq  $16, %%mm5\n\t"
+       "por    %%mm3, %%mm1\n\t"
+       "pand   %6, %%mm5\n\t"
+       "por    %%mm5, %%mm4\n\t"
+
+       MOVNTQ" %%mm0, %0\n\t"
+       MOVNTQ" %%mm1, 8%0\n\t"
+       MOVNTQ" %%mm4, 16%0"
+       :"=m"(*dest)
+       :"m"(*s),"m"(mask24l),
+        "m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
+       :"memory");
+    dest += 24;
+    s += 32;
+  }
+  __asm __volatile(SFENCE:::"memory");
+  __asm __volatile(EMMS:::"memory");
+#endif
+  while(s < end)
+  {
+#ifdef WORDS_BIGENDIAN
+    /* RGB32 (= A,B,G,R) -> RGB24 (= R,G,B) */
+    s++;
+    dest[2] = *s++;
+    dest[1] = *s++;
+    dest[0] = *s++;
+    dest += 3;
+#else
+    *dest++ = *s++;
+    *dest++ = *s++;
+    *dest++ = *s++;
+    s++;
+#endif
+  }
+}
+
+/*
+ Original by Strepto/Astral
+ ported to gcc & bugfixed : A'rpi
+ MMX2, 3DNOW optimization by Nick Kurshev
+ 32bit c version, and and&add trick by Michael Niedermayer
+*/
+static inline void RENAME(rgb15to16)(const uint8_t *src,uint8_t *dst,long src_size)
+{
+  register const uint8_t* s=src;
+  register uint8_t* d=dst;
+  register const uint8_t *end;
+  const uint8_t *mm_end;
+  end = s + src_size;
+#ifdef HAVE_MMX
+  __asm __volatile(PREFETCH"   %0"::"m"(*s));
+  __asm __volatile("movq       %0, %%mm4"::"m"(mask15s));
+  mm_end = end - 15;
+  while(s<mm_end)
+  {
+       __asm __volatile(
+               PREFETCH"       32%1\n\t"
+               "movq   %1, %%mm0\n\t"
+               "movq   8%1, %%mm2\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm2, %%mm3\n\t"
+               "pand   %%mm4, %%mm0\n\t"
+               "pand   %%mm4, %%mm2\n\t"
+               "paddw  %%mm1, %%mm0\n\t"
+               "paddw  %%mm3, %%mm2\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               MOVNTQ" %%mm2, 8%0"
+               :"=m"(*d)
+               :"m"(*s)
+               );
+       d+=16;
+       s+=16;
+  }
+  __asm __volatile(SFENCE:::"memory");
+  __asm __volatile(EMMS:::"memory");
+#endif
+    mm_end = end - 3;
+    while(s < mm_end)
+    {
+       register unsigned x= *((uint32_t *)s);
+       *((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0);
+       d+=4;
+       s+=4;
+    }
+    if(s < end)
+    {
+       register unsigned short x= *((uint16_t *)s);
+       *((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0);
+    }
+}
+
+static inline void RENAME(rgb16to15)(const uint8_t *src,uint8_t *dst,long src_size)
+{
+  register const uint8_t* s=src;
+  register uint8_t* d=dst;
+  register const uint8_t *end;
+  const uint8_t *mm_end;
+  end = s + src_size;
+#ifdef HAVE_MMX
+  __asm __volatile(PREFETCH"   %0"::"m"(*s));
+  __asm __volatile("movq       %0, %%mm7"::"m"(mask15rg));
+  __asm __volatile("movq       %0, %%mm6"::"m"(mask15b));
+  mm_end = end - 15;
+  while(s<mm_end)
+  {
+       __asm __volatile(
+               PREFETCH"       32%1\n\t"
+               "movq   %1, %%mm0\n\t"
+               "movq   8%1, %%mm2\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm2, %%mm3\n\t"
+               "psrlq  $1, %%mm0\n\t"
+               "psrlq  $1, %%mm2\n\t"
+               "pand   %%mm7, %%mm0\n\t"
+               "pand   %%mm7, %%mm2\n\t"
+               "pand   %%mm6, %%mm1\n\t"
+               "pand   %%mm6, %%mm3\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm3, %%mm2\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               MOVNTQ" %%mm2, 8%0"
+               :"=m"(*d)
+               :"m"(*s)
+               );
+       d+=16;
+       s+=16;
+  }
+  __asm __volatile(SFENCE:::"memory");
+  __asm __volatile(EMMS:::"memory");
+#endif
+    mm_end = end - 3;
+    while(s < mm_end)
+    {
+       register uint32_t x= *((uint32_t *)s);
+       *((uint32_t *)d) = ((x>>1)&0x7FE07FE0) | (x&0x001F001F);
+       s+=4;
+       d+=4;
+    }
+    if(s < end)
+    {
+       register uint16_t x= *((uint16_t *)s);
+       *((uint16_t *)d) = ((x>>1)&0x7FE0) | (x&0x001F);
+       s+=2;
+       d+=2;
+    }
+}
+
+static inline void RENAME(rgb32to16)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint8_t *s = src;
+       const uint8_t *end;
+#ifdef HAVE_MMX
+       const uint8_t *mm_end;
+#endif
+       uint16_t *d = (uint16_t *)dst;
+       end = s + src_size;
+#ifdef HAVE_MMX
+       mm_end = end - 15;
+#if 1 //is faster only if multiplies are reasonable fast (FIXME figure out on which cpus this is faster, on Athlon its slightly faster)
+       asm volatile(
+               "movq %3, %%mm5                 \n\t"
+               "movq %4, %%mm6                 \n\t"
+               "movq %5, %%mm7                 \n\t"
+               ASMALIGN16
+               "1:                             \n\t"
+               PREFETCH" 32(%1)                \n\t"
+               "movd   (%1), %%mm0             \n\t"
+               "movd   4(%1), %%mm3            \n\t"
+               "punpckldq 8(%1), %%mm0         \n\t"
+               "punpckldq 12(%1), %%mm3        \n\t"
+               "movq %%mm0, %%mm1              \n\t"
+               "movq %%mm3, %%mm4              \n\t"
+               "pand %%mm6, %%mm0              \n\t"
+               "pand %%mm6, %%mm3              \n\t"
+               "pmaddwd %%mm7, %%mm0           \n\t"
+               "pmaddwd %%mm7, %%mm3           \n\t"
+               "pand %%mm5, %%mm1              \n\t"
+               "pand %%mm5, %%mm4              \n\t"
+               "por %%mm1, %%mm0               \n\t"   
+               "por %%mm4, %%mm3               \n\t"
+               "psrld $5, %%mm0                \n\t"
+               "pslld $11, %%mm3               \n\t"
+               "por %%mm3, %%mm0               \n\t"
+               MOVNTQ" %%mm0, (%0)             \n\t"
+               "add $16, %1                    \n\t"
+               "add $8, %0                     \n\t"
+               "cmp %2, %1                     \n\t"
+               " jb 1b                         \n\t"
+               : "+r" (d), "+r"(s)
+               : "r" (mm_end), "m" (mask3216g), "m" (mask3216br), "m" (mul3216)
+       );
+#else
+       __asm __volatile(PREFETCH"      %0"::"m"(*src):"memory");
+       __asm __volatile(
+           "movq       %0, %%mm7\n\t"
+           "movq       %1, %%mm6\n\t"
+           ::"m"(red_16mask),"m"(green_16mask));
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movd   %1, %%mm0\n\t"
+               "movd   4%1, %%mm3\n\t"
+               "punpckldq 8%1, %%mm0\n\t"
+               "punpckldq 12%1, %%mm3\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm3, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "psrlq  $3, %%mm0\n\t"
+               "psrlq  $3, %%mm3\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %2, %%mm3\n\t"
+               "psrlq  $5, %%mm1\n\t"
+               "psrlq  $5, %%mm4\n\t"
+               "pand   %%mm6, %%mm1\n\t"
+               "pand   %%mm6, %%mm4\n\t"
+               "psrlq  $8, %%mm2\n\t"
+               "psrlq  $8, %%mm5\n\t"
+               "pand   %%mm7, %%mm2\n\t"
+               "pand   %%mm7, %%mm5\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               "psllq  $16, %%mm3\n\t"
+               "por    %%mm3, %%mm0\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
+               d += 4;
+               s += 16;
+       }
+#endif
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               register int rgb = *(uint32_t*)s; s += 4;
+               *d++ = ((rgb&0xFF)>>3) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>8);
+       }
+}
+
+static inline void RENAME(rgb32tobgr16)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint8_t *s = src;
+       const uint8_t *end;
+#ifdef HAVE_MMX
+       const uint8_t *mm_end;
+#endif
+       uint16_t *d = (uint16_t *)dst;
+       end = s + src_size;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*src):"memory");
+       __asm __volatile(
+           "movq       %0, %%mm7\n\t"
+           "movq       %1, %%mm6\n\t"
+           ::"m"(red_16mask),"m"(green_16mask));
+       mm_end = end - 15;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movd   %1, %%mm0\n\t"
+               "movd   4%1, %%mm3\n\t"
+               "punpckldq 8%1, %%mm0\n\t"
+               "punpckldq 12%1, %%mm3\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm3, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "psllq  $8, %%mm0\n\t"
+               "psllq  $8, %%mm3\n\t"
+               "pand   %%mm7, %%mm0\n\t"
+               "pand   %%mm7, %%mm3\n\t"
+               "psrlq  $5, %%mm1\n\t"
+               "psrlq  $5, %%mm4\n\t"
+               "pand   %%mm6, %%mm1\n\t"
+               "pand   %%mm6, %%mm4\n\t"
+               "psrlq  $19, %%mm2\n\t"
+               "psrlq  $19, %%mm5\n\t"
+               "pand   %2, %%mm2\n\t"
+               "pand   %2, %%mm5\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               "psllq  $16, %%mm3\n\t"
+               "por    %%mm3, %%mm0\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
+               d += 4;
+               s += 16;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               register int rgb = *(uint32_t*)s; s += 4;
+               *d++ = ((rgb&0xF8)<<8) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>19);
+       }
+}
+
+static inline void RENAME(rgb32to15)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint8_t *s = src;
+       const uint8_t *end;
+#ifdef HAVE_MMX
+       const uint8_t *mm_end;
+#endif
+       uint16_t *d = (uint16_t *)dst;
+       end = s + src_size;
+#ifdef HAVE_MMX
+       mm_end = end - 15;
+#if 1 //is faster only if multiplies are reasonable fast (FIXME figure out on which cpus this is faster, on Athlon its slightly faster)
+       asm volatile(
+               "movq %3, %%mm5                 \n\t"
+               "movq %4, %%mm6                 \n\t"
+               "movq %5, %%mm7                 \n\t"
+               ASMALIGN16
+               "1:                             \n\t"
+               PREFETCH" 32(%1)                \n\t"
+               "movd   (%1), %%mm0             \n\t"
+               "movd   4(%1), %%mm3            \n\t"
+               "punpckldq 8(%1), %%mm0         \n\t"
+               "punpckldq 12(%1), %%mm3        \n\t"
+               "movq %%mm0, %%mm1              \n\t"
+               "movq %%mm3, %%mm4              \n\t"
+               "pand %%mm6, %%mm0              \n\t"
+               "pand %%mm6, %%mm3              \n\t"
+               "pmaddwd %%mm7, %%mm0           \n\t"
+               "pmaddwd %%mm7, %%mm3           \n\t"
+               "pand %%mm5, %%mm1              \n\t"
+               "pand %%mm5, %%mm4              \n\t"
+               "por %%mm1, %%mm0               \n\t"   
+               "por %%mm4, %%mm3               \n\t"
+               "psrld $6, %%mm0                \n\t"
+               "pslld $10, %%mm3               \n\t"
+               "por %%mm3, %%mm0               \n\t"
+               MOVNTQ" %%mm0, (%0)             \n\t"
+               "add $16, %1                    \n\t"
+               "add $8, %0                     \n\t"
+               "cmp %2, %1                     \n\t"
+               " jb 1b                         \n\t"
+               : "+r" (d), "+r"(s)
+               : "r" (mm_end), "m" (mask3215g), "m" (mask3216br), "m" (mul3215)
+       );
+#else
+       __asm __volatile(PREFETCH"      %0"::"m"(*src):"memory");
+       __asm __volatile(
+           "movq       %0, %%mm7\n\t"
+           "movq       %1, %%mm6\n\t"
+           ::"m"(red_15mask),"m"(green_15mask));
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movd   %1, %%mm0\n\t"
+               "movd   4%1, %%mm3\n\t"
+               "punpckldq 8%1, %%mm0\n\t"
+               "punpckldq 12%1, %%mm3\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm3, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "psrlq  $3, %%mm0\n\t"
+               "psrlq  $3, %%mm3\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %2, %%mm3\n\t"
+               "psrlq  $6, %%mm1\n\t"
+               "psrlq  $6, %%mm4\n\t"
+               "pand   %%mm6, %%mm1\n\t"
+               "pand   %%mm6, %%mm4\n\t"
+               "psrlq  $9, %%mm2\n\t"
+               "psrlq  $9, %%mm5\n\t"
+               "pand   %%mm7, %%mm2\n\t"
+               "pand   %%mm7, %%mm5\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               "psllq  $16, %%mm3\n\t"
+               "por    %%mm3, %%mm0\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
+               d += 4;
+               s += 16;
+       }
+#endif
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               register int rgb = *(uint32_t*)s; s += 4;
+               *d++ = ((rgb&0xFF)>>3) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>9);
+       }
+}
+
+static inline void RENAME(rgb32tobgr15)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint8_t *s = src;
+       const uint8_t *end;
+#ifdef HAVE_MMX
+       const uint8_t *mm_end;
+#endif
+       uint16_t *d = (uint16_t *)dst;
+       end = s + src_size;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*src):"memory");
+       __asm __volatile(
+           "movq       %0, %%mm7\n\t"
+           "movq       %1, %%mm6\n\t"
+           ::"m"(red_15mask),"m"(green_15mask));
+       mm_end = end - 15;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movd   %1, %%mm0\n\t"
+               "movd   4%1, %%mm3\n\t"
+               "punpckldq 8%1, %%mm0\n\t"
+               "punpckldq 12%1, %%mm3\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm3, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "psllq  $7, %%mm0\n\t"
+               "psllq  $7, %%mm3\n\t"
+               "pand   %%mm7, %%mm0\n\t"
+               "pand   %%mm7, %%mm3\n\t"
+               "psrlq  $6, %%mm1\n\t"
+               "psrlq  $6, %%mm4\n\t"
+               "pand   %%mm6, %%mm1\n\t"
+               "pand   %%mm6, %%mm4\n\t"
+               "psrlq  $19, %%mm2\n\t"
+               "psrlq  $19, %%mm5\n\t"
+               "pand   %2, %%mm2\n\t"
+               "pand   %2, %%mm5\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               "psllq  $16, %%mm3\n\t"
+               "por    %%mm3, %%mm0\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
+               d += 4;
+               s += 16;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               register int rgb = *(uint32_t*)s; s += 4;
+               *d++ = ((rgb&0xF8)<<7) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>19);
+       }
+}
+
+static inline void RENAME(rgb24to16)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint8_t *s = src;
+       const uint8_t *end;
+#ifdef HAVE_MMX
+       const uint8_t *mm_end;
+#endif
+       uint16_t *d = (uint16_t *)dst;
+       end = s + src_size;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*src):"memory");
+       __asm __volatile(
+           "movq       %0, %%mm7\n\t"
+           "movq       %1, %%mm6\n\t"
+           ::"m"(red_16mask),"m"(green_16mask));
+       mm_end = end - 11;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movd   %1, %%mm0\n\t"
+               "movd   3%1, %%mm3\n\t"
+               "punpckldq 6%1, %%mm0\n\t"
+               "punpckldq 9%1, %%mm3\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm3, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "psrlq  $3, %%mm0\n\t"
+               "psrlq  $3, %%mm3\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %2, %%mm3\n\t"
+               "psrlq  $5, %%mm1\n\t"
+               "psrlq  $5, %%mm4\n\t"
+               "pand   %%mm6, %%mm1\n\t"
+               "pand   %%mm6, %%mm4\n\t"
+               "psrlq  $8, %%mm2\n\t"
+               "psrlq  $8, %%mm5\n\t"
+               "pand   %%mm7, %%mm2\n\t"
+               "pand   %%mm7, %%mm5\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               "psllq  $16, %%mm3\n\t"
+               "por    %%mm3, %%mm0\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
+               d += 4;
+               s += 12;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               const int b= *s++;
+               const int g= *s++;
+               const int r= *s++;
+               *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
+       }
+}
+
+static inline void RENAME(rgb24tobgr16)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint8_t *s = src;
+       const uint8_t *end;
+#ifdef HAVE_MMX
+       const uint8_t *mm_end;
+#endif
+       uint16_t *d = (uint16_t *)dst;
+       end = s + src_size;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*src):"memory");
+       __asm __volatile(
+           "movq       %0, %%mm7\n\t"
+           "movq       %1, %%mm6\n\t"
+           ::"m"(red_16mask),"m"(green_16mask));
+       mm_end = end - 15;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movd   %1, %%mm0\n\t"
+               "movd   3%1, %%mm3\n\t"
+               "punpckldq 6%1, %%mm0\n\t"
+               "punpckldq 9%1, %%mm3\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm3, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "psllq  $8, %%mm0\n\t"
+               "psllq  $8, %%mm3\n\t"
+               "pand   %%mm7, %%mm0\n\t"
+               "pand   %%mm7, %%mm3\n\t"
+               "psrlq  $5, %%mm1\n\t"
+               "psrlq  $5, %%mm4\n\t"
+               "pand   %%mm6, %%mm1\n\t"
+               "pand   %%mm6, %%mm4\n\t"
+               "psrlq  $19, %%mm2\n\t"
+               "psrlq  $19, %%mm5\n\t"
+               "pand   %2, %%mm2\n\t"
+               "pand   %2, %%mm5\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               "psllq  $16, %%mm3\n\t"
+               "por    %%mm3, %%mm0\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               :"=m"(*d):"m"(*s),"m"(blue_16mask):"memory");
+               d += 4;
+               s += 12;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               const int r= *s++;
+               const int g= *s++;
+               const int b= *s++;
+               *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);
+       }
+}
+
+static inline void RENAME(rgb24to15)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint8_t *s = src;
+       const uint8_t *end;
+#ifdef HAVE_MMX
+       const uint8_t *mm_end;
+#endif
+       uint16_t *d = (uint16_t *)dst;
+       end = s + src_size;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*src):"memory");
+       __asm __volatile(
+           "movq       %0, %%mm7\n\t"
+           "movq       %1, %%mm6\n\t"
+           ::"m"(red_15mask),"m"(green_15mask));
+       mm_end = end - 11;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movd   %1, %%mm0\n\t"
+               "movd   3%1, %%mm3\n\t"
+               "punpckldq 6%1, %%mm0\n\t"
+               "punpckldq 9%1, %%mm3\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm3, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "psrlq  $3, %%mm0\n\t"
+               "psrlq  $3, %%mm3\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %2, %%mm3\n\t"
+               "psrlq  $6, %%mm1\n\t"
+               "psrlq  $6, %%mm4\n\t"
+               "pand   %%mm6, %%mm1\n\t"
+               "pand   %%mm6, %%mm4\n\t"
+               "psrlq  $9, %%mm2\n\t"
+               "psrlq  $9, %%mm5\n\t"
+               "pand   %%mm7, %%mm2\n\t"
+               "pand   %%mm7, %%mm5\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               "psllq  $16, %%mm3\n\t"
+               "por    %%mm3, %%mm0\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
+               d += 4;
+               s += 12;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               const int b= *s++;
+               const int g= *s++;
+               const int r= *s++;
+               *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
+       }
+}
+
+static inline void RENAME(rgb24tobgr15)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint8_t *s = src;
+       const uint8_t *end;
+#ifdef HAVE_MMX
+       const uint8_t *mm_end;
+#endif
+       uint16_t *d = (uint16_t *)dst;
+       end = s + src_size;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*src):"memory");
+       __asm __volatile(
+           "movq       %0, %%mm7\n\t"
+           "movq       %1, %%mm6\n\t"
+           ::"m"(red_15mask),"m"(green_15mask));
+       mm_end = end - 15;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movd   %1, %%mm0\n\t"
+               "movd   3%1, %%mm3\n\t"
+               "punpckldq 6%1, %%mm0\n\t"
+               "punpckldq 9%1, %%mm3\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm3, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "psllq  $7, %%mm0\n\t"
+               "psllq  $7, %%mm3\n\t"
+               "pand   %%mm7, %%mm0\n\t"
+               "pand   %%mm7, %%mm3\n\t"
+               "psrlq  $6, %%mm1\n\t"
+               "psrlq  $6, %%mm4\n\t"
+               "pand   %%mm6, %%mm1\n\t"
+               "pand   %%mm6, %%mm4\n\t"
+               "psrlq  $19, %%mm2\n\t"
+               "psrlq  $19, %%mm5\n\t"
+               "pand   %2, %%mm2\n\t"
+               "pand   %2, %%mm5\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               "psllq  $16, %%mm3\n\t"
+               "por    %%mm3, %%mm0\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               :"=m"(*d):"m"(*s),"m"(blue_15mask):"memory");
+               d += 4;
+               s += 12;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               const int r= *s++;
+               const int g= *s++;
+               const int b= *s++;
+               *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);
+       }
+}
+
+/*
+  I use here less accurate approximation by simply
+ left-shifting the input
+  value and filling the low order bits with
+ zeroes. This method improves png's
+  compression but this scheme cannot reproduce white exactly, since it does not
+  generate an all-ones maximum value; the net effect is to darken the
+  image slightly.
+
+  The better method should be "left bit replication":
+
+   4 3 2 1 0
+   ---------
+   1 1 0 1 1
+
+   7 6 5 4 3  2 1 0
+   ----------------
+   1 1 0 1 1  1 1 0
+   |=======|  |===|
+       |      Leftmost Bits Repeated to Fill Open Bits
+       |
+   Original Bits
+*/
+static inline void RENAME(rgb15to24)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint16_t *end;
+#ifdef HAVE_MMX
+       const uint16_t *mm_end;
+#endif
+       uint8_t *d = (uint8_t *)dst;
+       const uint16_t *s = (uint16_t *)src;
+       end = s + src_size/2;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*s):"memory");
+       mm_end = end - 7;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movq   %1, %%mm0\n\t"
+               "movq   %1, %%mm1\n\t"
+               "movq   %1, %%mm2\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %3, %%mm1\n\t"
+               "pand   %4, %%mm2\n\t"
+               "psllq  $3, %%mm0\n\t"
+               "psrlq  $2, %%mm1\n\t"
+               "psrlq  $7, %%mm2\n\t"
+               "movq   %%mm0, %%mm3\n\t"
+               "movq   %%mm1, %%mm4\n\t"
+               "movq   %%mm2, %%mm5\n\t"
+               "punpcklwd %5, %%mm0\n\t"
+               "punpcklwd %5, %%mm1\n\t"
+               "punpcklwd %5, %%mm2\n\t"
+               "punpckhwd %5, %%mm3\n\t"
+               "punpckhwd %5, %%mm4\n\t"
+               "punpckhwd %5, %%mm5\n\t"
+               "psllq  $8, %%mm1\n\t"
+               "psllq  $16, %%mm2\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "psllq  $8, %%mm4\n\t"
+               "psllq  $16, %%mm5\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm5, %%mm3\n\t"
+
+               "movq   %%mm0, %%mm6\n\t"
+               "movq   %%mm3, %%mm7\n\t"
+               
+               "movq   8%1, %%mm0\n\t"
+               "movq   8%1, %%mm1\n\t"
+               "movq   8%1, %%mm2\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %3, %%mm1\n\t"
+               "pand   %4, %%mm2\n\t"
+               "psllq  $3, %%mm0\n\t"
+               "psrlq  $2, %%mm1\n\t"
+               "psrlq  $7, %%mm2\n\t"
+               "movq   %%mm0, %%mm3\n\t"
+               "movq   %%mm1, %%mm4\n\t"
+               "movq   %%mm2, %%mm5\n\t"
+               "punpcklwd %5, %%mm0\n\t"
+               "punpcklwd %5, %%mm1\n\t"
+               "punpcklwd %5, %%mm2\n\t"
+               "punpckhwd %5, %%mm3\n\t"
+               "punpckhwd %5, %%mm4\n\t"
+               "punpckhwd %5, %%mm5\n\t"
+               "psllq  $8, %%mm1\n\t"
+               "psllq  $16, %%mm2\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "psllq  $8, %%mm4\n\t"
+               "psllq  $16, %%mm5\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm5, %%mm3\n\t"
+
+               :"=m"(*d)
+               :"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r), "m"(mmx_null)
+               :"memory");
+           /* Borrowed 32 to 24 */
+           __asm __volatile(
+               "movq   %%mm0, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "movq   %%mm6, %%mm0\n\t"
+               "movq   %%mm7, %%mm1\n\t"
+               
+               "movq   %%mm4, %%mm6\n\t"
+               "movq   %%mm5, %%mm7\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm1, %%mm3\n\t"
+
+               "psrlq  $8, %%mm2\n\t"
+               "psrlq  $8, %%mm3\n\t"
+               "psrlq  $8, %%mm6\n\t"
+               "psrlq  $8, %%mm7\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %2, %%mm1\n\t"
+               "pand   %2, %%mm4\n\t"
+               "pand   %2, %%mm5\n\t"
+               "pand   %3, %%mm2\n\t"
+               "pand   %3, %%mm3\n\t"
+               "pand   %3, %%mm6\n\t"
+               "pand   %3, %%mm7\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm3, %%mm1\n\t"
+               "por    %%mm6, %%mm4\n\t"
+               "por    %%mm7, %%mm5\n\t"
+
+               "movq   %%mm1, %%mm2\n\t"
+               "movq   %%mm4, %%mm3\n\t"
+               "psllq  $48, %%mm2\n\t"
+               "psllq  $32, %%mm3\n\t"
+               "pand   %4, %%mm2\n\t"
+               "pand   %5, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "psrlq  $16, %%mm1\n\t"
+               "psrlq  $32, %%mm4\n\t"
+               "psllq  $16, %%mm5\n\t"
+               "por    %%mm3, %%mm1\n\t"
+               "pand   %6, %%mm5\n\t"
+               "por    %%mm5, %%mm4\n\t"
+
+               MOVNTQ" %%mm0, %0\n\t"
+               MOVNTQ" %%mm1, 8%0\n\t"
+               MOVNTQ" %%mm4, 16%0"
+
+               :"=m"(*d)
+               :"m"(*s),"m"(mask24l),"m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
+               :"memory");
+               d += 24;
+               s += 8;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               register uint16_t bgr;
+               bgr = *s++;
+               *d++ = (bgr&0x1F)<<3;
+               *d++ = (bgr&0x3E0)>>2;
+               *d++ = (bgr&0x7C00)>>7;
+       }
+}
+
+static inline void RENAME(rgb16to24)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint16_t *end;
+#ifdef HAVE_MMX
+       const uint16_t *mm_end;
+#endif
+       uint8_t *d = (uint8_t *)dst;
+       const uint16_t *s = (const uint16_t *)src;
+       end = s + src_size/2;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*s):"memory");
+       mm_end = end - 7;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movq   %1, %%mm0\n\t"
+               "movq   %1, %%mm1\n\t"
+               "movq   %1, %%mm2\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %3, %%mm1\n\t"
+               "pand   %4, %%mm2\n\t"
+               "psllq  $3, %%mm0\n\t"
+               "psrlq  $3, %%mm1\n\t"
+               "psrlq  $8, %%mm2\n\t"
+               "movq   %%mm0, %%mm3\n\t"
+               "movq   %%mm1, %%mm4\n\t"
+               "movq   %%mm2, %%mm5\n\t"
+               "punpcklwd %5, %%mm0\n\t"
+               "punpcklwd %5, %%mm1\n\t"
+               "punpcklwd %5, %%mm2\n\t"
+               "punpckhwd %5, %%mm3\n\t"
+               "punpckhwd %5, %%mm4\n\t"
+               "punpckhwd %5, %%mm5\n\t"
+               "psllq  $8, %%mm1\n\t"
+               "psllq  $16, %%mm2\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "psllq  $8, %%mm4\n\t"
+               "psllq  $16, %%mm5\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               
+               "movq   %%mm0, %%mm6\n\t"
+               "movq   %%mm3, %%mm7\n\t"
+
+               "movq   8%1, %%mm0\n\t"
+               "movq   8%1, %%mm1\n\t"
+               "movq   8%1, %%mm2\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %3, %%mm1\n\t"
+               "pand   %4, %%mm2\n\t"
+               "psllq  $3, %%mm0\n\t"
+               "psrlq  $3, %%mm1\n\t"
+               "psrlq  $8, %%mm2\n\t"
+               "movq   %%mm0, %%mm3\n\t"
+               "movq   %%mm1, %%mm4\n\t"
+               "movq   %%mm2, %%mm5\n\t"
+               "punpcklwd %5, %%mm0\n\t"
+               "punpcklwd %5, %%mm1\n\t"
+               "punpcklwd %5, %%mm2\n\t"
+               "punpckhwd %5, %%mm3\n\t"
+               "punpckhwd %5, %%mm4\n\t"
+               "punpckhwd %5, %%mm5\n\t"
+               "psllq  $8, %%mm1\n\t"
+               "psllq  $16, %%mm2\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "psllq  $8, %%mm4\n\t"
+               "psllq  $16, %%mm5\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               :"=m"(*d)
+               :"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null)           
+               :"memory");
+           /* Borrowed 32 to 24 */
+           __asm __volatile(
+               "movq   %%mm0, %%mm4\n\t"
+               "movq   %%mm3, %%mm5\n\t"
+               "movq   %%mm6, %%mm0\n\t"
+               "movq   %%mm7, %%mm1\n\t"
+               
+               "movq   %%mm4, %%mm6\n\t"
+               "movq   %%mm5, %%mm7\n\t"
+               "movq   %%mm0, %%mm2\n\t"
+               "movq   %%mm1, %%mm3\n\t"
+
+               "psrlq  $8, %%mm2\n\t"
+               "psrlq  $8, %%mm3\n\t"
+               "psrlq  $8, %%mm6\n\t"
+               "psrlq  $8, %%mm7\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %2, %%mm1\n\t"
+               "pand   %2, %%mm4\n\t"
+               "pand   %2, %%mm5\n\t"
+               "pand   %3, %%mm2\n\t"
+               "pand   %3, %%mm3\n\t"
+               "pand   %3, %%mm6\n\t"
+               "pand   %3, %%mm7\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "por    %%mm3, %%mm1\n\t"
+               "por    %%mm6, %%mm4\n\t"
+               "por    %%mm7, %%mm5\n\t"
+
+               "movq   %%mm1, %%mm2\n\t"
+               "movq   %%mm4, %%mm3\n\t"
+               "psllq  $48, %%mm2\n\t"
+               "psllq  $32, %%mm3\n\t"
+               "pand   %4, %%mm2\n\t"
+               "pand   %5, %%mm3\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "psrlq  $16, %%mm1\n\t"
+               "psrlq  $32, %%mm4\n\t"
+               "psllq  $16, %%mm5\n\t"
+               "por    %%mm3, %%mm1\n\t"
+               "pand   %6, %%mm5\n\t"
+               "por    %%mm5, %%mm4\n\t"
+
+               MOVNTQ" %%mm0, %0\n\t"
+               MOVNTQ" %%mm1, 8%0\n\t"
+               MOVNTQ" %%mm4, 16%0"
+
+               :"=m"(*d)
+               :"m"(*s),"m"(mask24l),"m"(mask24h),"m"(mask24hh),"m"(mask24hhh),"m"(mask24hhhh)
+               :"memory");
+               d += 24;
+               s += 8;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               register uint16_t bgr;
+               bgr = *s++;
+               *d++ = (bgr&0x1F)<<3;
+               *d++ = (bgr&0x7E0)>>3;
+               *d++ = (bgr&0xF800)>>8;
+       }
+}
+
+static inline void RENAME(rgb15to32)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint16_t *end;
+#ifdef HAVE_MMX
+       const uint16_t *mm_end;
+#endif
+       uint8_t *d = (uint8_t *)dst;
+       const uint16_t *s = (const uint16_t *)src;
+       end = s + src_size/2;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*s):"memory");
+       __asm __volatile("pxor  %%mm7,%%mm7\n\t":::"memory");
+       mm_end = end - 3;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movq   %1, %%mm0\n\t"
+               "movq   %1, %%mm1\n\t"
+               "movq   %1, %%mm2\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %3, %%mm1\n\t"
+               "pand   %4, %%mm2\n\t"
+               "psllq  $3, %%mm0\n\t"
+               "psrlq  $2, %%mm1\n\t"
+               "psrlq  $7, %%mm2\n\t"
+               "movq   %%mm0, %%mm3\n\t"
+               "movq   %%mm1, %%mm4\n\t"
+               "movq   %%mm2, %%mm5\n\t"
+               "punpcklwd %%mm7, %%mm0\n\t"
+               "punpcklwd %%mm7, %%mm1\n\t"
+               "punpcklwd %%mm7, %%mm2\n\t"
+               "punpckhwd %%mm7, %%mm3\n\t"
+               "punpckhwd %%mm7, %%mm4\n\t"
+               "punpckhwd %%mm7, %%mm5\n\t"
+               "psllq  $8, %%mm1\n\t"
+               "psllq  $16, %%mm2\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "psllq  $8, %%mm4\n\t"
+               "psllq  $16, %%mm5\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               MOVNTQ" %%mm3, 8%0\n\t"
+               :"=m"(*d)
+               :"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r)
+               :"memory");
+               d += 16;
+               s += 4;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+#if 0 //slightly slower on athlon
+               int bgr= *s++;
+               *((uint32_t*)d)++ = ((bgr&0x1F)<<3) + ((bgr&0x3E0)<<6) + ((bgr&0x7C00)<<9);
+#else
+               register uint16_t bgr;
+               bgr = *s++;
+#ifdef WORDS_BIGENDIAN
+               *d++ = 0;
+               *d++ = (bgr&0x7C00)>>7;
+               *d++ = (bgr&0x3E0)>>2;
+               *d++ = (bgr&0x1F)<<3;
+#else
+               *d++ = (bgr&0x1F)<<3;
+               *d++ = (bgr&0x3E0)>>2;
+               *d++ = (bgr&0x7C00)>>7;
+               *d++ = 0;
+#endif
+
+#endif
+       }
+}
+
+static inline void RENAME(rgb16to32)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       const uint16_t *end;
+#ifdef HAVE_MMX
+       const uint16_t *mm_end;
+#endif
+       uint8_t *d = (uint8_t *)dst;
+       const uint16_t *s = (uint16_t *)src;
+       end = s + src_size/2;
+#ifdef HAVE_MMX
+       __asm __volatile(PREFETCH"      %0"::"m"(*s):"memory");
+       __asm __volatile("pxor  %%mm7,%%mm7\n\t":::"memory");
+       mm_end = end - 3;
+       while(s < mm_end)
+       {
+           __asm __volatile(
+               PREFETCH" 32%1\n\t"
+               "movq   %1, %%mm0\n\t"
+               "movq   %1, %%mm1\n\t"
+               "movq   %1, %%mm2\n\t"
+               "pand   %2, %%mm0\n\t"
+               "pand   %3, %%mm1\n\t"
+               "pand   %4, %%mm2\n\t"
+               "psllq  $3, %%mm0\n\t"
+               "psrlq  $3, %%mm1\n\t"
+               "psrlq  $8, %%mm2\n\t"
+               "movq   %%mm0, %%mm3\n\t"
+               "movq   %%mm1, %%mm4\n\t"
+               "movq   %%mm2, %%mm5\n\t"
+               "punpcklwd %%mm7, %%mm0\n\t"
+               "punpcklwd %%mm7, %%mm1\n\t"
+               "punpcklwd %%mm7, %%mm2\n\t"
+               "punpckhwd %%mm7, %%mm3\n\t"
+               "punpckhwd %%mm7, %%mm4\n\t"
+               "punpckhwd %%mm7, %%mm5\n\t"
+               "psllq  $8, %%mm1\n\t"
+               "psllq  $16, %%mm2\n\t"
+               "por    %%mm1, %%mm0\n\t"
+               "por    %%mm2, %%mm0\n\t"
+               "psllq  $8, %%mm4\n\t"
+               "psllq  $16, %%mm5\n\t"
+               "por    %%mm4, %%mm3\n\t"
+               "por    %%mm5, %%mm3\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               MOVNTQ" %%mm3, 8%0\n\t"
+               :"=m"(*d)
+               :"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r)
+               :"memory");
+               d += 16;
+               s += 4;
+       }
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#endif
+       while(s < end)
+       {
+               register uint16_t bgr;
+               bgr = *s++;
+#ifdef WORDS_BIGENDIAN
+               *d++ = 0;
+               *d++ = (bgr&0xF800)>>8;
+               *d++ = (bgr&0x7E0)>>3;
+               *d++ = (bgr&0x1F)<<3;
+#else
+               *d++ = (bgr&0x1F)<<3;
+               *d++ = (bgr&0x7E0)>>3;
+               *d++ = (bgr&0xF800)>>8;
+               *d++ = 0;
+#endif
+       }
+}
+
+static inline void RENAME(rgb32tobgr32)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+#ifdef HAVE_MMX
+/* TODO: unroll this loop */
+       asm volatile (
+               "xor %%"REG_a", %%"REG_a"       \n\t"
+               ASMALIGN16
+               "1:                             \n\t"
+               PREFETCH" 32(%0, %%"REG_a")     \n\t"
+               "movq (%0, %%"REG_a"), %%mm0    \n\t"
+               "movq %%mm0, %%mm1              \n\t"
+               "movq %%mm0, %%mm2              \n\t"
+               "pslld $16, %%mm0               \n\t"
+               "psrld $16, %%mm1               \n\t"
+               "pand "MANGLE(mask32r)", %%mm0  \n\t"
+               "pand "MANGLE(mask32g)", %%mm2  \n\t"
+               "pand "MANGLE(mask32b)", %%mm1  \n\t"
+               "por %%mm0, %%mm2               \n\t"
+               "por %%mm1, %%mm2               \n\t"
+               MOVNTQ" %%mm2, (%1, %%"REG_a")  \n\t"
+               "add $8, %%"REG_a"              \n\t"
+               "cmp %2, %%"REG_a"              \n\t"
+               " jb 1b                         \n\t"
+               :: "r" (src), "r"(dst), "r" (src_size-7)
+               : "%"REG_a
+       );
+
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+#else
+       unsigned i;
+       unsigned num_pixels = src_size >> 2;
+       for(i=0; i<num_pixels; i++)
+       {
+#ifdef WORDS_BIGENDIAN  
+         dst[4*i + 1] = src[4*i + 3];
+         dst[4*i + 2] = src[4*i + 2];
+         dst[4*i + 3] = src[4*i + 1];
+#else
+         dst[4*i + 0] = src[4*i + 2];
+         dst[4*i + 1] = src[4*i + 1];
+         dst[4*i + 2] = src[4*i + 0];
+#endif
+       }
+#endif
+}
+
+static inline void RENAME(rgb24tobgr24)(const uint8_t *src, uint8_t *dst, long src_size)
+{
+       unsigned i;
+#ifdef HAVE_MMX
+       long mmx_size= 23 - src_size;
+       asm volatile (
+               "movq "MANGLE(mask24r)", %%mm5  \n\t"
+               "movq "MANGLE(mask24g)", %%mm6  \n\t"
+               "movq "MANGLE(mask24b)", %%mm7  \n\t"
+               ASMALIGN16
+               "1:                             \n\t"
+               PREFETCH" 32(%1, %%"REG_a")     \n\t"
+               "movq   (%1, %%"REG_a"), %%mm0  \n\t" // BGR BGR BG
+               "movq   (%1, %%"REG_a"), %%mm1  \n\t" // BGR BGR BG
+               "movq  2(%1, %%"REG_a"), %%mm2  \n\t" // R BGR BGR B
+               "psllq $16, %%mm0               \n\t" // 00 BGR BGR
+               "pand %%mm5, %%mm0              \n\t"
+               "pand %%mm6, %%mm1              \n\t"
+               "pand %%mm7, %%mm2              \n\t"
+               "por %%mm0, %%mm1               \n\t"
+               "por %%mm2, %%mm1               \n\t"                
+               "movq  6(%1, %%"REG_a"), %%mm0  \n\t" // BGR BGR BG
+               MOVNTQ" %%mm1,   (%2, %%"REG_a")\n\t" // RGB RGB RG
+               "movq  8(%1, %%"REG_a"), %%mm1  \n\t" // R BGR BGR B
+               "movq 10(%1, %%"REG_a"), %%mm2  \n\t" // GR BGR BGR
+               "pand %%mm7, %%mm0              \n\t"
+               "pand %%mm5, %%mm1              \n\t"
+               "pand %%mm6, %%mm2              \n\t"
+               "por %%mm0, %%mm1               \n\t"
+               "por %%mm2, %%mm1               \n\t"                
+               "movq 14(%1, %%"REG_a"), %%mm0  \n\t" // R BGR BGR B
+               MOVNTQ" %%mm1,  8(%2, %%"REG_a")\n\t" // B RGB RGB R
+               "movq 16(%1, %%"REG_a"), %%mm1  \n\t" // GR BGR BGR
+               "movq 18(%1, %%"REG_a"), %%mm2  \n\t" // BGR BGR BG
+               "pand %%mm6, %%mm0              \n\t"
+               "pand %%mm7, %%mm1              \n\t"
+               "pand %%mm5, %%mm2              \n\t"
+               "por %%mm0, %%mm1               \n\t"
+               "por %%mm2, %%mm1               \n\t"                
+               MOVNTQ" %%mm1, 16(%2, %%"REG_a")\n\t"
+               "add $24, %%"REG_a"             \n\t"
+               " js 1b                         \n\t"
+               : "+a" (mmx_size)
+               : "r" (src-mmx_size), "r"(dst-mmx_size)
+       );
+
+       __asm __volatile(SFENCE:::"memory");
+       __asm __volatile(EMMS:::"memory");
+
+       if(mmx_size==23) return; //finihsed, was multiple of 8
+
+       src+= src_size;
+       dst+= src_size;
+       src_size= 23-mmx_size;
+       src-= src_size;
+       dst-= src_size;
+#endif
+       for(i=0; i<src_size; i+=3)
+       {
+               register uint8_t x;
+               x          = src[i + 2];
+               dst[i + 1] = src[i + 1];
+               dst[i + 2] = src[i + 0];
+               dst[i + 0] = x;
+       }
+}
+
+static inline void RENAME(yuvPlanartoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride, long vertLumPerChroma)
+{
+       long y;
+       const long chromWidth= width>>1;
+       for(y=0; y<height; y++)
+       {
+#ifdef HAVE_MMX
+//FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway)
+               asm volatile(
+                       "xor %%"REG_a", %%"REG_a"       \n\t"
+                       ASMALIGN16
+                       "1:                             \n\t"
+                       PREFETCH" 32(%1, %%"REG_a", 2)  \n\t"
+                       PREFETCH" 32(%2, %%"REG_a")     \n\t"
+                       PREFETCH" 32(%3, %%"REG_a")     \n\t"
+                       "movq (%2, %%"REG_a"), %%mm0    \n\t" // U(0)
+                       "movq %%mm0, %%mm2              \n\t" // U(0)
+                       "movq (%3, %%"REG_a"), %%mm1    \n\t" // V(0)
+                       "punpcklbw %%mm1, %%mm0         \n\t" // UVUV UVUV(0)
+                       "punpckhbw %%mm1, %%mm2         \n\t" // UVUV UVUV(8)
+
+                       "movq (%1, %%"REG_a",2), %%mm3  \n\t" // Y(0)
+                       "movq 8(%1, %%"REG_a",2), %%mm5 \n\t" // Y(8)
+                       "movq %%mm3, %%mm4              \n\t" // Y(0)
+                       "movq %%mm5, %%mm6              \n\t" // Y(8)
+                       "punpcklbw %%mm0, %%mm3         \n\t" // YUYV YUYV(0)
+                       "punpckhbw %%mm0, %%mm4         \n\t" // YUYV YUYV(4)
+                       "punpcklbw %%mm2, %%mm5         \n\t" // YUYV YUYV(8)
+                       "punpckhbw %%mm2, %%mm6         \n\t" // YUYV YUYV(12)
+
+                       MOVNTQ" %%mm3, (%0, %%"REG_a", 4)\n\t"
+                       MOVNTQ" %%mm4, 8(%0, %%"REG_a", 4)\n\t"
+                       MOVNTQ" %%mm5, 16(%0, %%"REG_a", 4)\n\t"
+                       MOVNTQ" %%mm6, 24(%0, %%"REG_a", 4)\n\t"
+
+                       "add $8, %%"REG_a"              \n\t"
+                       "cmp %4, %%"REG_a"              \n\t"
+                       " jb 1b                         \n\t"
+                       ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth)
+                       : "%"REG_a
+               );
+#else
+
+#if defined ARCH_ALPHA && defined HAVE_MVI
+#define pl2yuy2(n)                                     \
+       y1 = yc[n];                                     \
+       y2 = yc2[n];                                    \
+       u = uc[n];                                      \
+       v = vc[n];                                      \
+       asm("unpkbw %1, %0" : "=r"(y1) : "r"(y1));      \
+       asm("unpkbw %1, %0" : "=r"(y2) : "r"(y2));      \
+       asm("unpkbl %1, %0" : "=r"(u) : "r"(u));        \
+       asm("unpkbl %1, %0" : "=r"(v) : "r"(v));        \
+       yuv1 = (u << 8) + (v << 24);                    \
+       yuv2 = yuv1 + y2;                               \
+       yuv1 += y1;                                     \
+       qdst[n] = yuv1;                                 \
+       qdst2[n] = yuv2;
+
+               int i;
+               uint64_t *qdst = (uint64_t *) dst;
+               uint64_t *qdst2 = (uint64_t *) (dst + dstStride);
+               const uint32_t *yc = (uint32_t *) ysrc;
+               const uint32_t *yc2 = (uint32_t *) (ysrc + lumStride);
+               const uint16_t *uc = (uint16_t*) usrc, *vc = (uint16_t*) vsrc;
+               for(i = 0; i < chromWidth; i += 8){
+                       uint64_t y1, y2, yuv1, yuv2;
+                       uint64_t u, v;
+                       /* Prefetch */
+                       asm("ldq $31,64(%0)" :: "r"(yc));
+                       asm("ldq $31,64(%0)" :: "r"(yc2));
+                       asm("ldq $31,64(%0)" :: "r"(uc));
+                       asm("ldq $31,64(%0)" :: "r"(vc));
+
+                       pl2yuy2(0);
+                       pl2yuy2(1);
+                       pl2yuy2(2);
+                       pl2yuy2(3);
+
+                       yc += 4;
+                       yc2 += 4;
+                       uc += 4;
+                       vc += 4;
+                       qdst += 4;
+                       qdst2 += 4;
+               }
+               y++;
+               ysrc += lumStride;
+               dst += dstStride;
+
+#elif __WORDSIZE >= 64
+               int i;
+               uint64_t *ldst = (uint64_t *) dst;
+               const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
+               for(i = 0; i < chromWidth; i += 2){
+                       uint64_t k, l;
+                       k = yc[0] + (uc[0] << 8) +
+                           (yc[1] << 16) + (vc[0] << 24);
+                       l = yc[2] + (uc[1] << 8) +
+                           (yc[3] << 16) + (vc[1] << 24);
+                       *ldst++ = k + (l << 32);
+                       yc += 4;
+                       uc += 2;
+                       vc += 2;
+               }
+
+#else
+               int i, *idst = (int32_t *) dst;
+               const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
+               for(i = 0; i < chromWidth; i++){
+#ifdef WORDS_BIGENDIAN
+                       *idst++ = (yc[0] << 24)+ (uc[0] << 16) +
+                           (yc[1] << 8) + (vc[0] << 0);
+#else
+                       *idst++ = yc[0] + (uc[0] << 8) +
+                           (yc[1] << 16) + (vc[0] << 24);
+#endif
+                       yc += 2;
+                       uc++;
+                       vc++;
+               }
+#endif
+#endif
+               if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
+               {
+                       usrc += chromStride;
+                       vsrc += chromStride;
+               }
+               ysrc += lumStride;
+               dst += dstStride;
+       }
+#ifdef HAVE_MMX
+asm(    EMMS" \n\t"
+        SFENCE" \n\t"
+        :::"memory");
+#endif
+}
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ */
+static inline void RENAME(yv12toyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride)
+{
+       //FIXME interpolate chroma
+       RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2);
+}
+
+static inline void RENAME(yuvPlanartouyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride, long vertLumPerChroma)
+{
+       long y;
+       const long chromWidth= width>>1;
+       for(y=0; y<height; y++)
+       {
+#ifdef HAVE_MMX
+//FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway)
+               asm volatile(
+                       "xor %%"REG_a", %%"REG_a"       \n\t"
+                       ASMALIGN16
+                       "1:                             \n\t"
+                       PREFETCH" 32(%1, %%"REG_a", 2)  \n\t"
+                       PREFETCH" 32(%2, %%"REG_a")     \n\t"
+                       PREFETCH" 32(%3, %%"REG_a")     \n\t"
+                       "movq (%2, %%"REG_a"), %%mm0    \n\t" // U(0)
+                       "movq %%mm0, %%mm2              \n\t" // U(0)
+                       "movq (%3, %%"REG_a"), %%mm1    \n\t" // V(0)
+                       "punpcklbw %%mm1, %%mm0         \n\t" // UVUV UVUV(0)
+                       "punpckhbw %%mm1, %%mm2         \n\t" // UVUV UVUV(8)
+
+                       "movq (%1, %%"REG_a",2), %%mm3  \n\t" // Y(0)
+                       "movq 8(%1, %%"REG_a",2), %%mm5 \n\t" // Y(8)
+                       "movq %%mm0, %%mm4              \n\t" // Y(0)
+                       "movq %%mm2, %%mm6              \n\t" // Y(8)
+                       "punpcklbw %%mm3, %%mm0         \n\t" // YUYV YUYV(0)
+                       "punpckhbw %%mm3, %%mm4         \n\t" // YUYV YUYV(4)
+                       "punpcklbw %%mm5, %%mm2         \n\t" // YUYV YUYV(8)
+                       "punpckhbw %%mm5, %%mm6         \n\t" // YUYV YUYV(12)
+
+                       MOVNTQ" %%mm0, (%0, %%"REG_a", 4)\n\t"
+                       MOVNTQ" %%mm4, 8(%0, %%"REG_a", 4)\n\t"
+                       MOVNTQ" %%mm2, 16(%0, %%"REG_a", 4)\n\t"
+                       MOVNTQ" %%mm6, 24(%0, %%"REG_a", 4)\n\t"
+
+                       "add $8, %%"REG_a"              \n\t"
+                       "cmp %4, %%"REG_a"              \n\t"
+                       " jb 1b                         \n\t"
+                       ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth)
+                       : "%"REG_a
+               );
+#else
+//FIXME adapt the alpha asm code from yv12->yuy2
+
+#if __WORDSIZE >= 64
+               int i;
+               uint64_t *ldst = (uint64_t *) dst;
+               const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
+               for(i = 0; i < chromWidth; i += 2){
+                       uint64_t k, l;
+                       k = uc[0] + (yc[0] << 8) +
+                           (vc[0] << 16) + (yc[1] << 24);
+                       l = uc[1] + (yc[2] << 8) +
+                           (vc[1] << 16) + (yc[3] << 24);
+                       *ldst++ = k + (l << 32);
+                       yc += 4;
+                       uc += 2;
+                       vc += 2;
+               }
+
+#else
+               int i, *idst = (int32_t *) dst;
+               const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
+               for(i = 0; i < chromWidth; i++){
+#ifdef WORDS_BIGENDIAN
+                       *idst++ = (uc[0] << 24)+ (yc[0] << 16) +
+                           (vc[0] << 8) + (yc[1] << 0);
+#else
+                       *idst++ = uc[0] + (yc[0] << 8) +
+                           (vc[0] << 16) + (yc[1] << 24);
+#endif
+                       yc += 2;
+                       uc++;
+                       vc++;
+               }
+#endif
+#endif
+               if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
+               {
+                       usrc += chromStride;
+                       vsrc += chromStride;
+               }
+               ysrc += lumStride;
+               dst += dstStride;
+       }
+#ifdef HAVE_MMX
+asm(    EMMS" \n\t"
+        SFENCE" \n\t"
+        :::"memory");
+#endif
+}
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ */
+static inline void RENAME(yv12touyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride)
+{
+       //FIXME interpolate chroma
+       RENAME(yuvPlanartouyvy)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2);
+}
+
+/**
+ *
+ * width should be a multiple of 16
+ */
+static inline void RENAME(yuv422ptoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+       long width, long height,
+       long lumStride, long chromStride, long dstStride)
+{
+       RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 1);
+}
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ */
+static inline void RENAME(yuy2toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+       long width, long height,
+       long lumStride, long chromStride, long srcStride)
+{
+       long y;
+       const long chromWidth= width>>1;
+       for(y=0; y<height; y+=2)
+       {
+#ifdef HAVE_MMX
+               asm volatile(
+                       "xor %%"REG_a", %%"REG_a"       \n\t"
+                       "pcmpeqw %%mm7, %%mm7           \n\t"
+                       "psrlw $8, %%mm7                \n\t" // FF,00,FF,00...
+                       ASMALIGN16
+                       "1:                             \n\t"
+                       PREFETCH" 64(%0, %%"REG_a", 4)  \n\t"
+                       "movq (%0, %%"REG_a", 4), %%mm0 \n\t" // YUYV YUYV(0)
+                       "movq 8(%0, %%"REG_a", 4), %%mm1\n\t" // YUYV YUYV(4)
+                       "movq %%mm0, %%mm2              \n\t" // YUYV YUYV(0)
+                       "movq %%mm1, %%mm3              \n\t" // YUYV YUYV(4)
+                       "psrlw $8, %%mm0                \n\t" // U0V0 U0V0(0)
+                       "psrlw $8, %%mm1                \n\t" // U0V0 U0V0(4)
+                       "pand %%mm7, %%mm2              \n\t" // Y0Y0 Y0Y0(0)
+                       "pand %%mm7, %%mm3              \n\t" // Y0Y0 Y0Y0(4)
+                       "packuswb %%mm1, %%mm0          \n\t" // UVUV UVUV(0)
+                       "packuswb %%mm3, %%mm2          \n\t" // YYYY YYYY(0)
+
+                       MOVNTQ" %%mm2, (%1, %%"REG_a", 2)\n\t"
+
+                       "movq 16(%0, %%"REG_a", 4), %%mm1\n\t" // YUYV YUYV(8)
+                       "movq 24(%0, %%"REG_a", 4), %%mm2\n\t" // YUYV YUYV(12)
+                       "movq %%mm1, %%mm3              \n\t" // YUYV YUYV(8)
+                       "movq %%mm2, %%mm4              \n\t" // YUYV YUYV(12)
+                       "psrlw $8, %%mm1                \n\t" // U0V0 U0V0(8)
+                       "psrlw $8, %%mm2                \n\t" // U0V0 U0V0(12)
+                       "pand %%mm7, %%mm3              \n\t" // Y0Y0 Y0Y0(8)
+                       "pand %%mm7, %%mm4              \n\t" // Y0Y0 Y0Y0(12)
+                       "packuswb %%mm2, %%mm1          \n\t" // UVUV UVUV(8)
+                       "packuswb %%mm4, %%mm3          \n\t" // YYYY YYYY(8)
+
+                       MOVNTQ" %%mm3, 8(%1, %%"REG_a", 2)\n\t"
+
+                       "movq %%mm0, %%mm2              \n\t" // UVUV UVUV(0)
+                       "movq %%mm1, %%mm3              \n\t" // UVUV UVUV(8)
+                       "psrlw $8, %%mm0                \n\t" // V0V0 V0V0(0)
+                       "psrlw $8, %%mm1                \n\t" // V0V0 V0V0(8)
+                       "pand %%mm7, %%mm2              \n\t" // U0U0 U0U0(0)
+                       "pand %%mm7, %%mm3              \n\t" // U0U0 U0U0(8)
+                       "packuswb %%mm1, %%mm0          \n\t" // VVVV VVVV(0)
+                       "packuswb %%mm3, %%mm2          \n\t" // UUUU UUUU(0)
+
+                       MOVNTQ" %%mm0, (%3, %%"REG_a")  \n\t"
+                       MOVNTQ" %%mm2, (%2, %%"REG_a")  \n\t"
+
+                       "add $8, %%"REG_a"              \n\t"
+                       "cmp %4, %%"REG_a"              \n\t"
+                       " jb 1b                         \n\t"
+                       ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
+                       : "memory", "%"REG_a
+               );
+
+               ydst += lumStride;
+               src  += srcStride;
+
+               asm volatile(
+                       "xor %%"REG_a", %%"REG_a"       \n\t"
+                       ASMALIGN16
+                       "1:                             \n\t"
+                       PREFETCH" 64(%0, %%"REG_a", 4)  \n\t"
+                       "movq (%0, %%"REG_a", 4), %%mm0 \n\t" // YUYV YUYV(0)
+                       "movq 8(%0, %%"REG_a", 4), %%mm1\n\t" // YUYV YUYV(4)
+                       "movq 16(%0, %%"REG_a", 4), %%mm2\n\t" // YUYV YUYV(8)
+                       "movq 24(%0, %%"REG_a", 4), %%mm3\n\t" // YUYV YUYV(12)
+                       "pand %%mm7, %%mm0              \n\t" // Y0Y0 Y0Y0(0)
+                       "pand %%mm7, %%mm1              \n\t" // Y0Y0 Y0Y0(4)
+                       "pand %%mm7, %%mm2              \n\t" // Y0Y0 Y0Y0(8)
+                       "pand %%mm7, %%mm3              \n\t" // Y0Y0 Y0Y0(12)
+                       "packuswb %%mm1, %%mm0          \n\t" // YYYY YYYY(0)
+                       "packuswb %%mm3, %%mm2          \n\t" // YYYY YYYY(8)
+
+                       MOVNTQ" %%mm0, (%1, %%"REG_a", 2)\n\t"
+                       MOVNTQ" %%mm2, 8(%1, %%"REG_a", 2)\n\t"
+
+                       "add $8, %%"REG_a"              \n\t"
+                       "cmp %4, %%"REG_a"              \n\t"
+                       " jb 1b                         \n\t"
+
+                       ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
+                       : "memory", "%"REG_a
+               );
+#else
+               long i;
+               for(i=0; i<chromWidth; i++)
+               {
+                       ydst[2*i+0]     = src[4*i+0];
+                       udst[i]         = src[4*i+1];
+                       ydst[2*i+1]     = src[4*i+2];
+                       vdst[i]         = src[4*i+3];
+               }
+               ydst += lumStride;
+               src  += srcStride;
+
+               for(i=0; i<chromWidth; i++)
+               {
+                       ydst[2*i+0]     = src[4*i+0];
+                       ydst[2*i+1]     = src[4*i+2];
+               }
+#endif
+               udst += chromStride;
+               vdst += chromStride;
+               ydst += lumStride;
+               src  += srcStride;
+       }
+#ifdef HAVE_MMX
+asm volatile(   EMMS" \n\t"
+               SFENCE" \n\t"
+               :::"memory");
+#endif
+}
+
+static inline void RENAME(yvu9toyv12)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc,
+       uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+       long width, long height, long lumStride, long chromStride)
+{
+       /* Y Plane */
+       memcpy(ydst, ysrc, width*height);
+
+       /* XXX: implement upscaling for U,V */
+}
+
+static inline void RENAME(planar2x)(const uint8_t *src, uint8_t *dst, long srcWidth, long srcHeight, long srcStride, long dstStride)
+{
+       long x,y;
+       
+       dst[0]= src[0];
+        
+       // first line
+       for(x=0; x<srcWidth-1; x++){
+               dst[2*x+1]= (3*src[x] +   src[x+1])>>2;
+               dst[2*x+2]= (  src[x] + 3*src[x+1])>>2;
+       }
+       dst[2*srcWidth-1]= src[srcWidth-1];
+       
+        dst+= dstStride;
+
+       for(y=1; y<srcHeight; y++){
+#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+               const long mmxSize= srcWidth&~15;
+               asm volatile(
+                       "mov %4, %%"REG_a"              \n\t"
+                       "1:                             \n\t"
+                       "movq (%0, %%"REG_a"), %%mm0    \n\t"
+                       "movq (%1, %%"REG_a"), %%mm1    \n\t"
+                       "movq 1(%0, %%"REG_a"), %%mm2   \n\t"
+                       "movq 1(%1, %%"REG_a"), %%mm3   \n\t"
+                       "movq -1(%0, %%"REG_a"), %%mm4  \n\t"
+                       "movq -1(%1, %%"REG_a"), %%mm5  \n\t"
+                       PAVGB" %%mm0, %%mm5             \n\t"
+                       PAVGB" %%mm0, %%mm3             \n\t"
+                       PAVGB" %%mm0, %%mm5             \n\t"
+                       PAVGB" %%mm0, %%mm3             \n\t"
+                       PAVGB" %%mm1, %%mm4             \n\t"
+                       PAVGB" %%mm1, %%mm2             \n\t"
+                       PAVGB" %%mm1, %%mm4             \n\t"
+                       PAVGB" %%mm1, %%mm2             \n\t"
+                       "movq %%mm5, %%mm7              \n\t"
+                       "movq %%mm4, %%mm6              \n\t"
+                       "punpcklbw %%mm3, %%mm5         \n\t"
+                       "punpckhbw %%mm3, %%mm7         \n\t"
+                       "punpcklbw %%mm2, %%mm4         \n\t"
+                       "punpckhbw %%mm2, %%mm6         \n\t"
+#if 1
+                       MOVNTQ" %%mm5, (%2, %%"REG_a", 2)\n\t"
+                       MOVNTQ" %%mm7, 8(%2, %%"REG_a", 2)\n\t"
+                       MOVNTQ" %%mm4, (%3, %%"REG_a", 2)\n\t"
+                       MOVNTQ" %%mm6, 8(%3, %%"REG_a", 2)\n\t"
+#else
+                       "movq %%mm5, (%2, %%"REG_a", 2) \n\t"
+                       "movq %%mm7, 8(%2, %%"REG_a", 2)\n\t"
+                       "movq %%mm4, (%3, %%"REG_a", 2) \n\t"
+                       "movq %%mm6, 8(%3, %%"REG_a", 2)\n\t"
+#endif
+                       "add $8, %%"REG_a"              \n\t"
+                       " js 1b                         \n\t"
+                       :: "r" (src + mmxSize  ), "r" (src + srcStride + mmxSize  ),
+                          "r" (dst + mmxSize*2), "r" (dst + dstStride + mmxSize*2),
+                          "g" (-mmxSize)
+                       : "%"REG_a
+
+               );
+#else
+               const long mmxSize=1;
+#endif
+               dst[0        ]= (3*src[0] +   src[srcStride])>>2;
+               dst[dstStride]= (  src[0] + 3*src[srcStride])>>2;
+
+               for(x=mmxSize-1; x<srcWidth-1; x++){
+                       dst[2*x          +1]= (3*src[x+0] +   src[x+srcStride+1])>>2;
+                       dst[2*x+dstStride+2]= (  src[x+0] + 3*src[x+srcStride+1])>>2;
+                       dst[2*x+dstStride+1]= (  src[x+1] + 3*src[x+srcStride  ])>>2;
+                       dst[2*x          +2]= (3*src[x+1] +   src[x+srcStride  ])>>2;
+               }
+               dst[srcWidth*2 -1            ]= (3*src[srcWidth-1] +   src[srcWidth-1 + srcStride])>>2;
+               dst[srcWidth*2 -1 + dstStride]= (  src[srcWidth-1] + 3*src[srcWidth-1 + srcStride])>>2;
+
+               dst+=dstStride*2;
+               src+=srcStride;
+       }
+       
+       // last line
+#if 1
+       dst[0]= src[0];
+        
+       for(x=0; x<srcWidth-1; x++){
+               dst[2*x+1]= (3*src[x] +   src[x+1])>>2;
+               dst[2*x+2]= (  src[x] + 3*src[x+1])>>2;
+       }
+       dst[2*srcWidth-1]= src[srcWidth-1];
+#else
+       for(x=0; x<srcWidth; x++){
+               dst[2*x+0]=
+               dst[2*x+1]= src[x];
+       }
+#endif
+
+#ifdef HAVE_MMX
+asm volatile(   EMMS" \n\t"
+               SFENCE" \n\t"
+               :::"memory");
+#endif
+}
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 16 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ * chrominance data is only taken from every secound line others are ignored FIXME write HQ version
+ */
+static inline void RENAME(uyvytoyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+       long width, long height,
+       long lumStride, long chromStride, long srcStride)
+{
+       long y;
+       const long chromWidth= width>>1;
+       for(y=0; y<height; y+=2)
+       {
+#ifdef HAVE_MMX
+               asm volatile(
+                       "xorl %%eax, %%eax              \n\t"
+                       "pcmpeqw %%mm7, %%mm7           \n\t"
+                       "psrlw $8, %%mm7                \n\t" // FF,00,FF,00...
+                       ASMALIGN16
+                       "1:                             \n\t"
+                       PREFETCH" 64(%0, %%eax, 4)      \n\t"
+                       "movq (%0, %%eax, 4), %%mm0     \n\t" // UYVY UYVY(0)
+                       "movq 8(%0, %%eax, 4), %%mm1    \n\t" // UYVY UYVY(4)
+                       "movq %%mm0, %%mm2              \n\t" // UYVY UYVY(0)
+                       "movq %%mm1, %%mm3              \n\t" // UYVY UYVY(4)
+                       "pand %%mm7, %%mm0              \n\t" // U0V0 U0V0(0)
+                       "pand %%mm7, %%mm1              \n\t" // U0V0 U0V0(4)
+                       "psrlw $8, %%mm2                \n\t" // Y0Y0 Y0Y0(0)
+                       "psrlw $8, %%mm3                \n\t" // Y0Y0 Y0Y0(4)
+                       "packuswb %%mm1, %%mm0          \n\t" // UVUV UVUV(0)
+                       "packuswb %%mm3, %%mm2          \n\t" // YYYY YYYY(0)
+
+                       MOVNTQ" %%mm2, (%1, %%eax, 2)   \n\t"
+
+                       "movq 16(%0, %%eax, 4), %%mm1   \n\t" // UYVY UYVY(8)
+                       "movq 24(%0, %%eax, 4), %%mm2   \n\t" // UYVY UYVY(12)
+                       "movq %%mm1, %%mm3              \n\t" // UYVY UYVY(8)
+                       "movq %%mm2, %%mm4              \n\t" // UYVY UYVY(12)
+                       "pand %%mm7, %%mm1              \n\t" // U0V0 U0V0(8)
+                       "pand %%mm7, %%mm2              \n\t" // U0V0 U0V0(12)
+                       "psrlw $8, %%mm3                \n\t" // Y0Y0 Y0Y0(8)
+                       "psrlw $8, %%mm4                \n\t" // Y0Y0 Y0Y0(12)
+                       "packuswb %%mm2, %%mm1          \n\t" // UVUV UVUV(8)
+                       "packuswb %%mm4, %%mm3          \n\t" // YYYY YYYY(8)
+
+                       MOVNTQ" %%mm3, 8(%1, %%eax, 2)  \n\t"
+
+                       "movq %%mm0, %%mm2              \n\t" // UVUV UVUV(0)
+                       "movq %%mm1, %%mm3              \n\t" // UVUV UVUV(8)
+                       "psrlw $8, %%mm0                \n\t" // V0V0 V0V0(0)
+                       "psrlw $8, %%mm1                \n\t" // V0V0 V0V0(8)
+                       "pand %%mm7, %%mm2              \n\t" // U0U0 U0U0(0)
+                       "pand %%mm7, %%mm3              \n\t" // U0U0 U0U0(8)
+                       "packuswb %%mm1, %%mm0          \n\t" // VVVV VVVV(0)
+                       "packuswb %%mm3, %%mm2          \n\t" // UUUU UUUU(0)
+
+                       MOVNTQ" %%mm0, (%3, %%eax)      \n\t"
+                       MOVNTQ" %%mm2, (%2, %%eax)      \n\t"
+
+                       "addl $8, %%eax                 \n\t"
+                       "cmpl %4, %%eax                 \n\t"
+                       " jb 1b                         \n\t"
+                       ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
+                       : "memory", "%eax"
+               );
+
+               ydst += lumStride;
+               src  += srcStride;
+
+               asm volatile(
+                       "xorl %%eax, %%eax              \n\t"
+                       ASMALIGN16
+                       "1:                             \n\t"
+                       PREFETCH" 64(%0, %%eax, 4)      \n\t"
+                       "movq (%0, %%eax, 4), %%mm0     \n\t" // YUYV YUYV(0)
+                       "movq 8(%0, %%eax, 4), %%mm1    \n\t" // YUYV YUYV(4)
+                       "movq 16(%0, %%eax, 4), %%mm2   \n\t" // YUYV YUYV(8)
+                       "movq 24(%0, %%eax, 4), %%mm3   \n\t" // YUYV YUYV(12)
+                       "psrlw $8, %%mm0                \n\t" // Y0Y0 Y0Y0(0)
+                       "psrlw $8, %%mm1                \n\t" // Y0Y0 Y0Y0(4)
+                       "psrlw $8, %%mm2                \n\t" // Y0Y0 Y0Y0(8)
+                       "psrlw $8, %%mm3                \n\t" // Y0Y0 Y0Y0(12)
+                       "packuswb %%mm1, %%mm0          \n\t" // YYYY YYYY(0)
+                       "packuswb %%mm3, %%mm2          \n\t" // YYYY YYYY(8)
+
+                       MOVNTQ" %%mm0, (%1, %%eax, 2)   \n\t"
+                       MOVNTQ" %%mm2, 8(%1, %%eax, 2)  \n\t"
+
+                       "addl $8, %%eax                 \n\t"
+                       "cmpl %4, %%eax                 \n\t"
+                       " jb 1b                         \n\t"
+
+                       ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)
+                       : "memory", "%eax"
+               );
+#else
+               long i;
+               for(i=0; i<chromWidth; i++)
+               {
+                       udst[i]         = src[4*i+0];
+                       ydst[2*i+0]     = src[4*i+1];
+                       vdst[i]         = src[4*i+2];
+                       ydst[2*i+1]     = src[4*i+3];
+               }
+               ydst += lumStride;
+               src  += srcStride;
+
+               for(i=0; i<chromWidth; i++)
+               {
+                       ydst[2*i+0]     = src[4*i+1];
+                       ydst[2*i+1]     = src[4*i+3];
+               }
+#endif
+               udst += chromStride;
+               vdst += chromStride;
+               ydst += lumStride;
+               src  += srcStride;
+       }
+#ifdef HAVE_MMX
+asm volatile(   EMMS" \n\t"
+               SFENCE" \n\t"
+               :::"memory");
+#endif
+}
+
+/**
+ *
+ * height should be a multiple of 2 and width should be a multiple of 2 (if this is a
+ * problem for anyone then tell me, and ill fix it)
+ * chrominance data is only taken from every secound line others are ignored in the C version FIXME write HQ version
+ */
+static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
+       long width, long height,
+       long lumStride, long chromStride, long srcStride)
+{
+       long y;
+       const long chromWidth= width>>1;
+#ifdef HAVE_MMX
+       for(y=0; y<height-2; y+=2)
+       {
+               long i;
+               for(i=0; i<2; i++)
+               {
+                       asm volatile(
+                               "mov %2, %%"REG_a"              \n\t"
+                               "movq "MANGLE(bgr2YCoeff)", %%mm6               \n\t"
+                               "movq "MANGLE(w1111)", %%mm5            \n\t"
+                               "pxor %%mm7, %%mm7              \n\t"
+                               "lea (%%"REG_a", %%"REG_a", 2), %%"REG_b"\n\t"
+                               ASMALIGN16
+                               "1:                             \n\t"
+                               PREFETCH" 64(%0, %%"REG_b")     \n\t"
+                               "movd (%0, %%"REG_b"), %%mm0    \n\t"
+                               "movd 3(%0, %%"REG_b"), %%mm1   \n\t"
+                               "punpcklbw %%mm7, %%mm0         \n\t"
+                               "punpcklbw %%mm7, %%mm1         \n\t"
+                               "movd 6(%0, %%"REG_b"), %%mm2   \n\t"
+                               "movd 9(%0, %%"REG_b"), %%mm3   \n\t"
+                               "punpcklbw %%mm7, %%mm2         \n\t"
+                               "punpcklbw %%mm7, %%mm3         \n\t"
+                               "pmaddwd %%mm6, %%mm0           \n\t"
+                               "pmaddwd %%mm6, %%mm1           \n\t"
+                               "pmaddwd %%mm6, %%mm2           \n\t"
+                               "pmaddwd %%mm6, %%mm3           \n\t"
+#ifndef FAST_BGR2YV12
+                               "psrad $8, %%mm0                \n\t"
+                               "psrad $8, %%mm1                \n\t"
+                               "psrad $8, %%mm2                \n\t"
+                               "psrad $8, %%mm3                \n\t"
+#endif
+                               "packssdw %%mm1, %%mm0          \n\t"
+                               "packssdw %%mm3, %%mm2          \n\t"
+                               "pmaddwd %%mm5, %%mm0           \n\t"
+                               "pmaddwd %%mm5, %%mm2           \n\t"
+                               "packssdw %%mm2, %%mm0          \n\t"
+                               "psraw $7, %%mm0                \n\t"
+
+                               "movd 12(%0, %%"REG_b"), %%mm4  \n\t"
+                               "movd 15(%0, %%"REG_b"), %%mm1  \n\t"
+                               "punpcklbw %%mm7, %%mm4         \n\t"
+                               "punpcklbw %%mm7, %%mm1         \n\t"
+                               "movd 18(%0, %%"REG_b"), %%mm2  \n\t"
+                               "movd 21(%0, %%"REG_b"), %%mm3  \n\t"
+                               "punpcklbw %%mm7, %%mm2         \n\t"
+                               "punpcklbw %%mm7, %%mm3         \n\t"
+                               "pmaddwd %%mm6, %%mm4           \n\t"
+                               "pmaddwd %%mm6, %%mm1           \n\t"
+                               "pmaddwd %%mm6, %%mm2           \n\t"
+                               "pmaddwd %%mm6, %%mm3           \n\t"
+#ifndef FAST_BGR2YV12
+                               "psrad $8, %%mm4                \n\t"
+                               "psrad $8, %%mm1                \n\t"
+                               "psrad $8, %%mm2                \n\t"
+                               "psrad $8, %%mm3                \n\t"
+#endif
+                               "packssdw %%mm1, %%mm4          \n\t"
+                               "packssdw %%mm3, %%mm2          \n\t"
+                               "pmaddwd %%mm5, %%mm4           \n\t"
+                               "pmaddwd %%mm5, %%mm2           \n\t"
+                               "add $24, %%"REG_b"             \n\t"
+                               "packssdw %%mm2, %%mm4          \n\t"
+                               "psraw $7, %%mm4                \n\t"
+
+                               "packuswb %%mm4, %%mm0          \n\t"
+                               "paddusb "MANGLE(bgr2YOffset)", %%mm0   \n\t"
+
+                               MOVNTQ" %%mm0, (%1, %%"REG_a")  \n\t"
+                               "add $8, %%"REG_a"              \n\t"
+                               " js 1b                         \n\t"
+                               : : "r" (src+width*3), "r" (ydst+width), "g" (-width)
+                               : "%"REG_a, "%"REG_b
+                       );
+                       ydst += lumStride;
+                       src  += srcStride;
+               }
+               src -= srcStride*2;
+               asm volatile(
+                       "mov %4, %%"REG_a"              \n\t"
+                       "movq "MANGLE(w1111)", %%mm5            \n\t"
+                       "movq "MANGLE(bgr2UCoeff)", %%mm6               \n\t"
+                       "pxor %%mm7, %%mm7              \n\t"
+                       "lea (%%"REG_a", %%"REG_a", 2), %%"REG_b"\n\t"
+                       "add %%"REG_b", %%"REG_b"       \n\t"
+                       ASMALIGN16
+                       "1:                             \n\t"
+                       PREFETCH" 64(%0, %%"REG_b")     \n\t"
+                       PREFETCH" 64(%1, %%"REG_b")     \n\t"
+#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+                       "movq (%0, %%"REG_b"), %%mm0    \n\t"
+                       "movq (%1, %%"REG_b"), %%mm1    \n\t"
+                       "movq 6(%0, %%"REG_b"), %%mm2   \n\t"
+                       "movq 6(%1, %%"REG_b"), %%mm3   \n\t"
+                       PAVGB" %%mm1, %%mm0             \n\t"
+                       PAVGB" %%mm3, %%mm2             \n\t"
+                       "movq %%mm0, %%mm1              \n\t"
+                       "movq %%mm2, %%mm3              \n\t"
+                       "psrlq $24, %%mm0               \n\t"
+                       "psrlq $24, %%mm2               \n\t"
+                       PAVGB" %%mm1, %%mm0             \n\t"
+                       PAVGB" %%mm3, %%mm2             \n\t"
+                       "punpcklbw %%mm7, %%mm0         \n\t"
+                       "punpcklbw %%mm7, %%mm2         \n\t"
+#else
+                       "movd (%0, %%"REG_b"), %%mm0    \n\t"
+                       "movd (%1, %%"REG_b"), %%mm1    \n\t"
+                       "movd 3(%0, %%"REG_b"), %%mm2   \n\t"
+                       "movd 3(%1, %%"REG_b"), %%mm3   \n\t"
+                       "punpcklbw %%mm7, %%mm0         \n\t"
+                       "punpcklbw %%mm7, %%mm1         \n\t"
+                       "punpcklbw %%mm7, %%mm2         \n\t"
+                       "punpcklbw %%mm7, %%mm3         \n\t"
+                       "paddw %%mm1, %%mm0             \n\t"
+                       "paddw %%mm3, %%mm2             \n\t"
+                       "paddw %%mm2, %%mm0             \n\t"
+                       "movd 6(%0, %%"REG_b"), %%mm4   \n\t"
+                       "movd 6(%1, %%"REG_b"), %%mm1   \n\t"
+                       "movd 9(%0, %%"REG_b"), %%mm2   \n\t"
+                       "movd 9(%1, %%"REG_b"), %%mm3   \n\t"
+                       "punpcklbw %%mm7, %%mm4         \n\t"
+                       "punpcklbw %%mm7, %%mm1         \n\t"
+                       "punpcklbw %%mm7, %%mm2         \n\t"
+                       "punpcklbw %%mm7, %%mm3         \n\t"
+                       "paddw %%mm1, %%mm4             \n\t"
+                       "paddw %%mm3, %%mm2             \n\t"
+                       "paddw %%mm4, %%mm2             \n\t"
+                       "psrlw $2, %%mm0                \n\t"
+                       "psrlw $2, %%mm2                \n\t"
+#endif
+                       "movq "MANGLE(bgr2VCoeff)", %%mm1               \n\t"
+                       "movq "MANGLE(bgr2VCoeff)", %%mm3               \n\t"
+
+                       "pmaddwd %%mm0, %%mm1           \n\t"
+                       "pmaddwd %%mm2, %%mm3           \n\t"
+                       "pmaddwd %%mm6, %%mm0           \n\t"
+                       "pmaddwd %%mm6, %%mm2           \n\t"
+#ifndef FAST_BGR2YV12
+                       "psrad $8, %%mm0                \n\t"
+                       "psrad $8, %%mm1                \n\t"
+                       "psrad $8, %%mm2                \n\t"
+                       "psrad $8, %%mm3                \n\t"
+#endif
+                       "packssdw %%mm2, %%mm0          \n\t"
+                       "packssdw %%mm3, %%mm1          \n\t"
+                       "pmaddwd %%mm5, %%mm0           \n\t"
+                       "pmaddwd %%mm5, %%mm1           \n\t"
+                       "packssdw %%mm1, %%mm0          \n\t" // V1 V0 U1 U0
+                       "psraw $7, %%mm0                \n\t"
+
+#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+                       "movq 12(%0, %%"REG_b"), %%mm4  \n\t"
+                       "movq 12(%1, %%"REG_b"), %%mm1  \n\t"
+                       "movq 18(%0, %%"REG_b"), %%mm2  \n\t"
+                       "movq 18(%1, %%"REG_b"), %%mm3  \n\t"
+                       PAVGB" %%mm1, %%mm4             \n\t"
+                       PAVGB" %%mm3, %%mm2             \n\t"
+                       "movq %%mm4, %%mm1              \n\t"
+                       "movq %%mm2, %%mm3              \n\t"
+                       "psrlq $24, %%mm4               \n\t"
+                       "psrlq $24, %%mm2               \n\t"
+                       PAVGB" %%mm1, %%mm4             \n\t"
+                       PAVGB" %%mm3, %%mm2             \n\t"
+                       "punpcklbw %%mm7, %%mm4         \n\t"
+                       "punpcklbw %%mm7, %%mm2         \n\t"
+#else
+                       "movd 12(%0, %%"REG_b"), %%mm4  \n\t"
+                       "movd 12(%1, %%"REG_b"), %%mm1  \n\t"
+                       "movd 15(%0, %%"REG_b"), %%mm2  \n\t"
+                       "movd 15(%1, %%"REG_b"), %%mm3  \n\t"
+                       "punpcklbw %%mm7, %%mm4         \n\t"
+                       "punpcklbw %%mm7, %%mm1         \n\t"
+                       "punpcklbw %%mm7, %%mm2         \n\t"
+                       "punpcklbw %%mm7, %%mm3         \n\t"
+                       "paddw %%mm1, %%mm4             \n\t"
+                       "paddw %%mm3, %%mm2             \n\t"
+                       "paddw %%mm2, %%mm4             \n\t"
+                       "movd 18(%0, %%"REG_b"), %%mm5  \n\t"
+                       "movd 18(%1, %%"REG_b"), %%mm1  \n\t"
+                       "movd 21(%0, %%"REG_b"), %%mm2  \n\t"
+                       "movd 21(%1, %%"REG_b"), %%mm3  \n\t"
+                       "punpcklbw %%mm7, %%mm5         \n\t"
+                       "punpcklbw %%mm7, %%mm1         \n\t"
+                       "punpcklbw %%mm7, %%mm2         \n\t"
+                       "punpcklbw %%mm7, %%mm3         \n\t"
+                       "paddw %%mm1, %%mm5             \n\t"
+                       "paddw %%mm3, %%mm2             \n\t"
+                       "paddw %%mm5, %%mm2             \n\t"
+                       "movq "MANGLE(w1111)", %%mm5            \n\t"
+                       "psrlw $2, %%mm4                \n\t"
+                       "psrlw $2, %%mm2                \n\t"
+#endif
+                       "movq "MANGLE(bgr2VCoeff)", %%mm1               \n\t"
+                       "movq "MANGLE(bgr2VCoeff)", %%mm3               \n\t"
+
+                       "pmaddwd %%mm4, %%mm1           \n\t"
+                       "pmaddwd %%mm2, %%mm3           \n\t"
+                       "pmaddwd %%mm6, %%mm4           \n\t"
+                       "pmaddwd %%mm6, %%mm2           \n\t"
+#ifndef FAST_BGR2YV12
+                       "psrad $8, %%mm4                \n\t"
+                       "psrad $8, %%mm1                \n\t"
+                       "psrad $8, %%mm2                \n\t"
+                       "psrad $8, %%mm3                \n\t"
+#endif
+                       "packssdw %%mm2, %%mm4          \n\t"
+                       "packssdw %%mm3, %%mm1          \n\t"
+                       "pmaddwd %%mm5, %%mm4           \n\t"
+                       "pmaddwd %%mm5, %%mm1           \n\t"
+                       "add $24, %%"REG_b"             \n\t"
+                       "packssdw %%mm1, %%mm4          \n\t" // V3 V2 U3 U2
+                       "psraw $7, %%mm4                \n\t"
+
+                       "movq %%mm0, %%mm1              \n\t"
+                       "punpckldq %%mm4, %%mm0         \n\t"
+                       "punpckhdq %%mm4, %%mm1         \n\t"
+                       "packsswb %%mm1, %%mm0          \n\t"
+                       "paddb "MANGLE(bgr2UVOffset)", %%mm0    \n\t"
+                       "movd %%mm0, (%2, %%"REG_a")    \n\t"
+                       "punpckhdq %%mm0, %%mm0         \n\t"
+                       "movd %%mm0, (%3, %%"REG_a")    \n\t"
+                       "add $4, %%"REG_a"              \n\t"
+                       " js 1b                         \n\t"
+                       : : "r" (src+chromWidth*6), "r" (src+srcStride+chromWidth*6), "r" (udst+chromWidth), "r" (vdst+chromWidth), "g" (-chromWidth)
+                       : "%"REG_a, "%"REG_b
+               );
+
+               udst += chromStride;
+               vdst += chromStride;
+               src  += srcStride*2;
+       }
+
+       asm volatile(   EMMS" \n\t"
+                       SFENCE" \n\t"
+                       :::"memory");
+#else
+       y=0;
+#endif
+       for(; y<height; y+=2)
+       {
+               long i;
+               for(i=0; i<chromWidth; i++)
+               {
+                       unsigned int b= src[6*i+0];
+                       unsigned int g= src[6*i+1];
+                       unsigned int r= src[6*i+2];
+
+                       unsigned int Y  =  ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;
+                       unsigned int V  =  ((RV*r + GV*g + BV*b)>>RGB2YUV_SHIFT) + 128;
+                       unsigned int U  =  ((RU*r + GU*g + BU*b)>>RGB2YUV_SHIFT) + 128;
+
+                       udst[i]         = U;
+                       vdst[i]         = V;
+                       ydst[2*i]       = Y;
+
+                       b= src[6*i+3];
+                       g= src[6*i+4];
+                       r= src[6*i+5];
+
+                       Y  =  ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;
+                       ydst[2*i+1]     = Y;
+               }
+               ydst += lumStride;
+               src  += srcStride;
+
+               for(i=0; i<chromWidth; i++)
+               {
+                       unsigned int b= src[6*i+0];
+                       unsigned int g= src[6*i+1];
+                       unsigned int r= src[6*i+2];
+
+                       unsigned int Y  =  ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;
+
+                       ydst[2*i]       = Y;
+
+                       b= src[6*i+3];
+                       g= src[6*i+4];
+                       r= src[6*i+5];
+
+                       Y  =  ((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16;
+                       ydst[2*i+1]     = Y;
+               }
+               udst += chromStride;
+               vdst += chromStride;
+               ydst += lumStride;
+               src  += srcStride;
+       }
+}
+
+void RENAME(interleaveBytes)(uint8_t *src1, uint8_t *src2, uint8_t *dest,
+                           long width, long height, long src1Stride,
+                           long src2Stride, long dstStride){
+       long h;
+
+       for(h=0; h < height; h++)
+       {
+               long w;
+
+#ifdef HAVE_MMX
+#ifdef HAVE_SSE2
+               asm(
+                       "xor %%"REG_a", %%"REG_a"       \n\t"
+                       "1:                             \n\t"
+                       PREFETCH" 64(%1, %%"REG_a")     \n\t"
+                       PREFETCH" 64(%2, %%"REG_a")     \n\t"
+                       "movdqa (%1, %%"REG_a"), %%xmm0 \n\t"
+                       "movdqa (%1, %%"REG_a"), %%xmm1 \n\t"
+                       "movdqa (%2, %%"REG_a"), %%xmm2 \n\t"
+                       "punpcklbw %%xmm2, %%xmm0       \n\t"
+                       "punpckhbw %%xmm2, %%xmm1       \n\t"
+                       "movntdq %%xmm0, (%0, %%"REG_a", 2)\n\t"
+                       "movntdq %%xmm1, 16(%0, %%"REG_a", 2)\n\t"
+                       "add $16, %%"REG_a"             \n\t"
+                       "cmp %3, %%"REG_a"              \n\t"
+                       " jb 1b                         \n\t"
+                       ::"r"(dest), "r"(src1), "r"(src2), "r" (width-15)
+                       : "memory", "%"REG_a""
+               );
+#else
+               asm(
+                       "xor %%"REG_a", %%"REG_a"       \n\t"
+                       "1:                             \n\t"
+                       PREFETCH" 64(%1, %%"REG_a")     \n\t"
+                       PREFETCH" 64(%2, %%"REG_a")     \n\t"
+                       "movq (%1, %%"REG_a"), %%mm0    \n\t"
+                       "movq 8(%1, %%"REG_a"), %%mm2   \n\t"
+                       "movq %%mm0, %%mm1              \n\t"
+                       "movq %%mm2, %%mm3              \n\t"
+                       "movq (%2, %%"REG_a"), %%mm4    \n\t"
+                       "movq 8(%2, %%"REG_a"), %%mm5   \n\t"
+                       "punpcklbw %%mm4, %%mm0         \n\t"
+                       "punpckhbw %%mm4, %%mm1         \n\t"
+                       "punpcklbw %%mm5, %%mm2         \n\t"
+                       "punpckhbw %%mm5, %%mm3         \n\t"
+                       MOVNTQ" %%mm0, (%0, %%"REG_a", 2)\n\t"
+                       MOVNTQ" %%mm1, 8(%0, %%"REG_a", 2)\n\t"
+                       MOVNTQ" %%mm2, 16(%0, %%"REG_a", 2)\n\t"
+                       MOVNTQ" %%mm3, 24(%0, %%"REG_a", 2)\n\t"
+                       "add $16, %%"REG_a"             \n\t"
+                       "cmp %3, %%"REG_a"              \n\t"
+                       " jb 1b                         \n\t"
+                       ::"r"(dest), "r"(src1), "r"(src2), "r" (width-15)
+                       : "memory", "%"REG_a
+               );
+#endif
+               for(w= (width&(~15)); w < width; w++)
+               {
+                       dest[2*w+0] = src1[w];
+                       dest[2*w+1] = src2[w];
+               }
+#else
+               for(w=0; w < width; w++)
+               {
+                       dest[2*w+0] = src1[w];
+                       dest[2*w+1] = src2[w];
+               }
+#endif
+               dest += dstStride;
+                src1 += src1Stride;
+                src2 += src2Stride;
+       }
+#ifdef HAVE_MMX
+       asm(
+               EMMS" \n\t"
+               SFENCE" \n\t"
+               ::: "memory"
+               );
+#endif
+}
+
+static inline void RENAME(vu9_to_vu12)(const uint8_t *src1, const uint8_t *src2,
+                       uint8_t *dst1, uint8_t *dst2,
+                       long width, long height,
+                       long srcStride1, long srcStride2,
+                       long dstStride1, long dstStride2)
+{
+    long y,x,w,h;
+    w=width/2; h=height/2;
+#ifdef HAVE_MMX
+    asm volatile(
+       PREFETCH" %0\n\t"
+       PREFETCH" %1\n\t"
+       ::"m"(*(src1+srcStride1)),"m"(*(src2+srcStride2)):"memory");
+#endif
+    for(y=0;y<h;y++){
+       const uint8_t* s1=src1+srcStride1*(y>>1);
+       uint8_t* d=dst1+dstStride1*y;
+       x=0;
+#ifdef HAVE_MMX
+       for(;x<w-31;x+=32)
+       {
+           asm volatile(
+               PREFETCH" 32%1\n\t"
+               "movq   %1, %%mm0\n\t"
+               "movq   8%1, %%mm2\n\t"
+               "movq   16%1, %%mm4\n\t"
+               "movq   24%1, %%mm6\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm2, %%mm3\n\t"
+               "movq   %%mm4, %%mm5\n\t"
+               "movq   %%mm6, %%mm7\n\t"
+               "punpcklbw %%mm0, %%mm0\n\t"
+               "punpckhbw %%mm1, %%mm1\n\t"
+               "punpcklbw %%mm2, %%mm2\n\t"
+               "punpckhbw %%mm3, %%mm3\n\t"
+               "punpcklbw %%mm4, %%mm4\n\t"
+               "punpckhbw %%mm5, %%mm5\n\t"
+               "punpcklbw %%mm6, %%mm6\n\t"
+               "punpckhbw %%mm7, %%mm7\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               MOVNTQ" %%mm1, 8%0\n\t"
+               MOVNTQ" %%mm2, 16%0\n\t"
+               MOVNTQ" %%mm3, 24%0\n\t"
+               MOVNTQ" %%mm4, 32%0\n\t"
+               MOVNTQ" %%mm5, 40%0\n\t"
+               MOVNTQ" %%mm6, 48%0\n\t"
+               MOVNTQ" %%mm7, 56%0"
+               :"=m"(d[2*x])
+               :"m"(s1[x])
+               :"memory");
+       }
+#endif
+       for(;x<w;x++) d[2*x]=d[2*x+1]=s1[x];
+    }
+    for(y=0;y<h;y++){
+       const uint8_t* s2=src2+srcStride2*(y>>1);
+       uint8_t* d=dst2+dstStride2*y;
+       x=0;
+#ifdef HAVE_MMX
+       for(;x<w-31;x+=32)
+       {
+           asm volatile(
+               PREFETCH" 32%1\n\t"
+               "movq   %1, %%mm0\n\t"
+               "movq   8%1, %%mm2\n\t"
+               "movq   16%1, %%mm4\n\t"
+               "movq   24%1, %%mm6\n\t"
+               "movq   %%mm0, %%mm1\n\t"
+               "movq   %%mm2, %%mm3\n\t"
+               "movq   %%mm4, %%mm5\n\t"
+               "movq   %%mm6, %%mm7\n\t"
+               "punpcklbw %%mm0, %%mm0\n\t"
+               "punpckhbw %%mm1, %%mm1\n\t"
+               "punpcklbw %%mm2, %%mm2\n\t"
+               "punpckhbw %%mm3, %%mm3\n\t"
+               "punpcklbw %%mm4, %%mm4\n\t"
+               "punpckhbw %%mm5, %%mm5\n\t"
+               "punpcklbw %%mm6, %%mm6\n\t"
+               "punpckhbw %%mm7, %%mm7\n\t"
+               MOVNTQ" %%mm0, %0\n\t"
+               MOVNTQ" %%mm1, 8%0\n\t"
+               MOVNTQ" %%mm2, 16%0\n\t"
+               MOVNTQ" %%mm3, 24%0\n\t"
+               MOVNTQ" %%mm4, 32%0\n\t"
+               MOVNTQ" %%mm5, 40%0\n\t"
+               MOVNTQ" %%mm6, 48%0\n\t"
+               MOVNTQ" %%mm7, 56%0"
+               :"=m"(d[2*x])
+               :"m"(s2[x])
+               :"memory");
+       }
+#endif
+       for(;x<w;x++) d[2*x]=d[2*x+1]=s2[x];
+    }
+#ifdef HAVE_MMX
+       asm(
+               EMMS" \n\t"
+               SFENCE" \n\t"
+               ::: "memory"
+               );
+#endif
+}
+
+static inline void RENAME(yvu9_to_yuy2)(const uint8_t *src1, const uint8_t *src2, const uint8_t *src3,
+                       uint8_t *dst,
+                       long width, long height,
+                       long srcStride1, long srcStride2,
+                       long srcStride3, long dstStride)
+{
+    long y,x,w,h;
+    w=width/2; h=height;
+    for(y=0;y<h;y++){
+       const uint8_t* yp=src1+srcStride1*y;
+       const uint8_t* up=src2+srcStride2*(y>>2);
+       const uint8_t* vp=src3+srcStride3*(y>>2);
+       uint8_t* d=dst+dstStride*y;
+       x=0;
+#ifdef HAVE_MMX
+       for(;x<w-7;x+=8)
+       {
+           asm volatile(
+               PREFETCH" 32(%1, %0)\n\t"
+               PREFETCH" 32(%2, %0)\n\t"
+               PREFETCH" 32(%3, %0)\n\t"
+               "movq   (%1, %0, 4), %%mm0\n\t"       /* Y0Y1Y2Y3Y4Y5Y6Y7 */
+               "movq   (%2, %0), %%mm1\n\t"       /* U0U1U2U3U4U5U6U7 */
+               "movq   (%3, %0), %%mm2\n\t"         /* V0V1V2V3V4V5V6V7 */
+               "movq   %%mm0, %%mm3\n\t"    /* Y0Y1Y2Y3Y4Y5Y6Y7 */
+               "movq   %%mm1, %%mm4\n\t"    /* U0U1U2U3U4U5U6U7 */
+               "movq   %%mm2, %%mm5\n\t"    /* V0V1V2V3V4V5V6V7 */
+               "punpcklbw %%mm1, %%mm1\n\t" /* U0U0 U1U1 U2U2 U3U3 */
+               "punpcklbw %%mm2, %%mm2\n\t" /* V0V0 V1V1 V2V2 V3V3 */
+               "punpckhbw %%mm4, %%mm4\n\t" /* U4U4 U5U5 U6U6 U7U7 */
+               "punpckhbw %%mm5, %%mm5\n\t" /* V4V4 V5V5 V6V6 V7V7 */
+
+               "movq   %%mm1, %%mm6\n\t"
+               "punpcklbw %%mm2, %%mm1\n\t" /* U0V0 U0V0 U1V1 U1V1*/
+               "punpcklbw %%mm1, %%mm0\n\t" /* Y0U0 Y1V0 Y2U0 Y3V0*/
+               "punpckhbw %%mm1, %%mm3\n\t" /* Y4U1 Y5V1 Y6U1 Y7V1*/
+               MOVNTQ" %%mm0, (%4, %0, 8)\n\t"
+               MOVNTQ" %%mm3, 8(%4, %0, 8)\n\t"
+               
+               "punpckhbw %%mm2, %%mm6\n\t" /* U2V2 U2V2 U3V3 U3V3*/
+               "movq   8(%1, %0, 4), %%mm0\n\t"
+               "movq   %%mm0, %%mm3\n\t"
+               "punpcklbw %%mm6, %%mm0\n\t" /* Y U2 Y V2 Y U2 Y V2*/
+               "punpckhbw %%mm6, %%mm3\n\t" /* Y U3 Y V3 Y U3 Y V3*/
+               MOVNTQ" %%mm0, 16(%4, %0, 8)\n\t"
+               MOVNTQ" %%mm3, 24(%4, %0, 8)\n\t"
+
+               "movq   %%mm4, %%mm6\n\t"
+               "movq   16(%1, %0, 4), %%mm0\n\t"
+               "movq   %%mm0, %%mm3\n\t"
+               "punpcklbw %%mm5, %%mm4\n\t"
+               "punpcklbw %%mm4, %%mm0\n\t" /* Y U4 Y V4 Y U4 Y V4*/
+               "punpckhbw %%mm4, %%mm3\n\t" /* Y U5 Y V5 Y U5 Y V5*/
+               MOVNTQ" %%mm0, 32(%4, %0, 8)\n\t"
+               MOVNTQ" %%mm3, 40(%4, %0, 8)\n\t"
+               
+               "punpckhbw %%mm5, %%mm6\n\t"
+               "movq   24(%1, %0, 4), %%mm0\n\t"
+               "movq   %%mm0, %%mm3\n\t"
+               "punpcklbw %%mm6, %%mm0\n\t" /* Y U6 Y V6 Y U6 Y V6*/
+               "punpckhbw %%mm6, %%mm3\n\t" /* Y U7 Y V7 Y U7 Y V7*/
+               MOVNTQ" %%mm0, 48(%4, %0, 8)\n\t"
+               MOVNTQ" %%mm3, 56(%4, %0, 8)\n\t"
+
+               : "+r" (x)
+                : "r"(yp), "r" (up), "r"(vp), "r"(d)
+               :"memory");
+       }
+#endif
+       for(; x<w; x++)
+       {
+           const long x2= x<<2;
+           d[8*x+0]=yp[x2];
+           d[8*x+1]=up[x];
+           d[8*x+2]=yp[x2+1];
+           d[8*x+3]=vp[x];
+           d[8*x+4]=yp[x2+2];
+           d[8*x+5]=up[x];
+           d[8*x+6]=yp[x2+3];
+           d[8*x+7]=vp[x];
+       }
+    }
+#ifdef HAVE_MMX
+       asm(
+               EMMS" \n\t"
+               SFENCE" \n\t"
+               ::: "memory"
+               );
+#endif
+}
diff --git a/libswscale/swscale-example.c b/libswscale/swscale-example.c
new file mode 100644 (file)
index 0000000..f8e0ef4
--- /dev/null
@@ -0,0 +1,216 @@
+/*
+    Copyright (C) 2003 Michael Niedermayer <michaelni@gmx.at>
+
+    This program 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.
+
+    This program 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.
+
+    You should have received a copy of the GNU General Public License
+    along with this program; if not, write to the Free Software
+    Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+*/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <inttypes.h>
+#include <stdarg.h>
+
+#include "config.h"
+
+#include "swscale.h"
+#include "libvo/img_format.h"
+
+static int testFormat[]={
+IMGFMT_YVU9,
+IMGFMT_YV12,
+//IMGFMT_IYUV,
+IMGFMT_I420,
+IMGFMT_BGR15,
+IMGFMT_BGR16,
+IMGFMT_BGR24,
+IMGFMT_BGR32,
+IMGFMT_RGB24,
+IMGFMT_RGB32,
+//IMGFMT_Y8,
+IMGFMT_Y800,
+//IMGFMT_YUY2,
+0
+};
+
+static uint64_t getSSD(uint8_t *src1, uint8_t *src2, int stride1, int stride2, int w, int h){
+       int x,y;
+       uint64_t ssd=0;
+
+//printf("%d %d\n", w, h);
+        
+       for(y=0; y<h; y++){
+               for(x=0; x<w; x++){
+                       int d= src1[x + y*stride1] - src2[x + y*stride2];
+                       ssd+= d*d;
+//printf("%d", abs(src1[x + y*stride1] - src2[x + y*stride2])/26 );
+               }
+//printf("\n");
+       }
+       return ssd;
+}
+
+// test by ref -> src -> dst -> out & compare out against ref
+// ref & out are YV12
+static void doTest(uint8_t *ref[3], int refStride[3], int w, int h, int srcFormat, int dstFormat, 
+                   int srcW, int srcH, int dstW, int dstH, int flags){
+       uint8_t *src[3];
+       uint8_t *dst[3];
+       uint8_t *out[3];
+       int srcStride[3], dstStride[3];
+       int i;
+       uint64_t ssdY, ssdU, ssdV;
+       struct SwsContext *srcContext, *dstContext, *outContext;
+       
+       for(i=0; i<3; i++){
+               // avoid stride % bpp != 0
+               if(srcFormat==IMGFMT_RGB24 || srcFormat==IMGFMT_BGR24)
+                       srcStride[i]= srcW*3;
+               else
+                       srcStride[i]= srcW*4;
+               
+               if(dstFormat==IMGFMT_RGB24 || dstFormat==IMGFMT_BGR24)
+                       dstStride[i]= dstW*3;
+               else
+                       dstStride[i]= dstW*4;
+       
+               src[i]= (uint8_t*) malloc(srcStride[i]*srcH);
+               dst[i]= (uint8_t*) malloc(dstStride[i]*dstH);
+               out[i]= (uint8_t*) malloc(refStride[i]*h);
+       }
+
+       srcContext= sws_getContext(w, h, IMGFMT_YV12, srcW, srcH, srcFormat, flags, NULL, NULL, NULL);
+       dstContext= sws_getContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, NULL, NULL, NULL);
+       outContext= sws_getContext(dstW, dstH, dstFormat, w, h, IMGFMT_YV12, flags, NULL, NULL, NULL);
+       if(srcContext==NULL ||dstContext==NULL ||outContext==NULL){
+               printf("Failed allocating swsContext\n");
+               goto end;
+       }
+//     printf("test %X %X %X -> %X %X %X\n", (int)ref[0], (int)ref[1], (int)ref[2],
+//             (int)src[0], (int)src[1], (int)src[2]);
+
+       sws_scale(srcContext, ref, refStride, 0, h   , src, srcStride);
+       sws_scale(dstContext, src, srcStride, 0, srcH, dst, dstStride);
+       sws_scale(outContext, dst, dstStride, 0, dstH, out, refStride);
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+       asm volatile ("emms\n\t");
+#endif
+            
+       ssdY= getSSD(ref[0], out[0], refStride[0], refStride[0], w, h);
+       ssdU= getSSD(ref[1], out[1], refStride[1], refStride[1], (w+1)>>1, (h+1)>>1);
+       ssdV= getSSD(ref[2], out[2], refStride[2], refStride[2], (w+1)>>1, (h+1)>>1);
+       
+       if(srcFormat == IMGFMT_Y800 || dstFormat==IMGFMT_Y800) ssdU=ssdV=0; //FIXME check that output is really gray
+       
+       ssdY/= w*h;
+       ssdU/= w*h/4;
+       ssdV/= w*h/4;
+       
+       if(ssdY>100 || ssdU>100 || ssdV>100){
+               printf(" %s %dx%d -> %s %4dx%4d flags=%2d SSD=%5lld,%5lld,%5lld\n", 
+                       vo_format_name(srcFormat), srcW, srcH, 
+                       vo_format_name(dstFormat), dstW, dstH,
+                       flags,
+                       ssdY, ssdU, ssdV);
+       }
+
+       end:
+       
+       sws_freeContext(srcContext);
+       sws_freeContext(dstContext);
+       sws_freeContext(outContext);
+
+       for(i=0; i<3; i++){
+               free(src[i]);
+               free(dst[i]);
+               free(out[i]);
+       }
+}
+
+void mp_msg( int x, int y, const char *format, ... ){
+    va_list va;
+    va_start(va, format);
+    vfprintf(stderr, format, va);
+    va_end(va);
+}
+
+void fast_memcpy(void *a, void *b, int s){ //FIXME
+    memcpy(a, b, s);
+}
+
+static void selfTest(uint8_t *src[3], int stride[3], int w, int h){
+       int srcFormat, dstFormat, srcFormatIndex, dstFormatIndex;
+       int srcW, srcH, dstW, dstH;
+       int flags;
+
+       for(srcFormatIndex=0; ;srcFormatIndex++){
+               srcFormat= testFormat[srcFormatIndex];
+               if(!srcFormat) break;
+               for(dstFormatIndex=0; ;dstFormatIndex++){
+                       dstFormat= testFormat[dstFormatIndex];
+                       if(!dstFormat) break;
+//                     if(!isSupportedOut(dstFormat)) continue;
+printf("%s -> %s\n", 
+       vo_format_name(srcFormat),
+       vo_format_name(dstFormat));
+                       srcW= w;
+                       srcH= h;
+                       for(dstW=w - w/3; dstW<= 4*w/3; dstW+= w/3){
+                               for(dstH=h - h/3; dstH<= 4*h/3; dstH+= h/3){
+                                       for(flags=1; flags<33; flags*=2)
+                                               doTest(src, stride, w, h, srcFormat, dstFormat,
+                                                       srcW, srcH, dstW, dstH, flags);
+                               }
+                       }
+               }
+       }
+}
+
+#define W 96
+#define H 96
+
+int main(int argc, char **argv){
+       uint8_t rgb_data[W*H*4];
+       uint8_t *rgb_src[3]= {rgb_data, NULL, NULL};
+       int rgb_stride[3]={4*W, 0, 0};
+       uint8_t data[3][W*H];
+       uint8_t *src[3]= {data[0], data[1], data[2]};
+       int stride[3]={W, W, W};
+       int x, y;
+       struct SwsContext *sws;
+
+       sws= sws_getContext(W/12, H/12, IMGFMT_BGR32, W, H, IMGFMT_YV12, 2, NULL, NULL, NULL);
+        
+       for(y=0; y<H; y++){
+               for(x=0; x<W*4; x++){
+                       rgb_data[ x + y*4*W]= random();
+               }
+       }
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+       sws_rgb2rgb_init(SWS_CPU_CAPS_MMX*0);
+#else
+       sws_rgb2rgb_init(0);
+#endif
+       sws_scale(sws, rgb_src, rgb_stride, 0, H   , src, stride);
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+       asm volatile ("emms\n\t");
+#endif
+
+       selfTest(src,  stride, W, H);
+
+        return 123;
+}
diff --git a/libswscale/swscale.c b/libswscale/swscale.c
new file mode 100644 (file)
index 0000000..11f774d
--- /dev/null
@@ -0,0 +1,2707 @@
+/*
+    Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
+
+    This program 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.
+
+    This program 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.
+
+    You should have received a copy of the GNU General Public License
+    along with this program; if not, write to the Free Software
+    Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+*/
+
+/*
+  supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
+  supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
+  {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}
+  x -> x
+  YUV9 -> YV12
+  YUV9/YV12 -> Y800
+  Y800 -> YUV9/YV12
+  BGR24 -> BGR32 & RGB24 -> RGB32
+  BGR32 -> BGR24 & RGB32 -> RGB24
+  BGR15 -> BGR16
+*/
+
+/* 
+tested special converters (most are tested actually but i didnt write it down ...)
+ YV12 -> BGR16
+ YV12 -> YV12
+ BGR15 -> BGR16
+ BGR16 -> BGR16
+ YVU9 -> YV12
+
+untested special converters
+  YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
+  YV12/I420 -> YV12/I420
+  YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
+  BGR24 -> BGR32 & RGB24 -> RGB32
+  BGR32 -> BGR24 & RGB32 -> RGB24
+  BGR24 -> YV12
+*/
+
+#include <inttypes.h>
+#include <string.h>
+#include <math.h>
+#include <stdio.h>
+#include <unistd.h>
+#include "config.h"
+#include "mangle.h"
+#include <assert.h>
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#else
+#include <stdlib.h>
+#endif
+#ifdef HAVE_SYS_MMAN_H
+#include <sys/mman.h>
+#if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
+#define MAP_ANONYMOUS MAP_ANON
+#endif
+#endif
+#include "swscale.h"
+#include "swscale_internal.h"
+#include "cpudetect.h"
+#include "bswap.h"
+#include "libvo/img_format.h"
+#include "rgb2rgb.h"
+#include "libvo/fastmemcpy.h"
+
+#undef MOVNTQ
+#undef PAVGB
+
+//#undef HAVE_MMX2
+//#define HAVE_3DNOW
+//#undef HAVE_MMX
+//#undef ARCH_X86
+//#define WORDS_BIGENDIAN
+#define DITHER1XBPP
+
+#define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
+
+#define RET 0xC3 //near return opcode for X86
+
+#ifdef MP_DEBUG
+#define ASSERT(x) assert(x);
+#else
+#define ASSERT(x) ;
+#endif
+
+#ifdef M_PI
+#define PI M_PI
+#else
+#define PI 3.14159265358979323846
+#endif
+
+//FIXME replace this with something faster
+#define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
+                       || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21 \
+                       || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
+#define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
+#define isGray(x)      ((x)==IMGFMT_Y800)
+#define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
+#define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
+#define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
+                       || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
+                       || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
+                       || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
+                       || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
+#define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
+                       || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
+                       || isRGB(x) || isBGR(x)\
+                       || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21\
+                       || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
+#define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
+
+#define RGB2YUV_SHIFT 16
+#define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
+#define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
+#define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
+#define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
+#define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
+#define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
+#define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
+#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
+#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
+
+extern const int32_t Inverse_Table_6_9[8][4];
+
+/*
+NOTES
+Special versions: fast Y 1:1 scaling (no interpolation in y direction)
+
+TODO
+more intelligent missalignment avoidance for the horizontal scaler
+write special vertical cubic upscale version
+Optimize C code (yv12 / minmax)
+add support for packed pixel yuv input & output
+add support for Y8 output
+optimize bgr24 & bgr32
+add BGR4 output support
+write special BGR->BGR scaler
+*/
+
+#define ABS(a) ((a) > 0 ? (a) : (-(a)))
+#define MIN(a,b) ((a) > (b) ? (b) : (a))
+#define MAX(a,b) ((a) < (b) ? (b) : (a))
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
+static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
+static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
+static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
+static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
+static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
+static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
+static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
+
+static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
+static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
+static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
+static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
+
+static uint64_t __attribute__((aligned(8))) dither4[2]={
+       0x0103010301030103LL,
+       0x0200020002000200LL,};
+
+static uint64_t __attribute__((aligned(8))) dither8[2]={
+       0x0602060206020602LL,
+       0x0004000400040004LL,};
+
+static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
+static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
+static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
+static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
+static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
+static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
+
+static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
+static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
+static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
+
+#ifdef FAST_BGR2YV12
+static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
+static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
+static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
+#else
+static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
+static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
+static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
+#endif
+static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
+static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
+static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
+#endif
+
+// clipping helper table for C implementations:
+static unsigned char clip_table[768];
+
+static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
+                 
+extern const uint8_t dither_2x2_4[2][8];
+extern const uint8_t dither_2x2_8[2][8];
+extern const uint8_t dither_8x8_32[8][8];
+extern const uint8_t dither_8x8_73[8][8];
+extern const uint8_t dither_8x8_220[8][8];
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+void in_asm_used_var_warning_killer()
+{
+ volatile int i= bF8+bFC+w10+
+ bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
+ M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
+ if(i) i=0;
+}
+#endif
+
+static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
+                                   int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
+                                   uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
+{
+       //FIXME Optimize (just quickly writen not opti..)
+       int i;
+       for(i=0; i<dstW; i++)
+       {
+               int val=1<<18;
+               int j;
+               for(j=0; j<lumFilterSize; j++)
+                       val += lumSrc[j][i] * lumFilter[j];
+
+               dest[i]= MIN(MAX(val>>19, 0), 255);
+       }
+
+       if(uDest != NULL)
+               for(i=0; i<chrDstW; i++)
+               {
+                       int u=1<<18;
+                       int v=1<<18;
+                       int j;
+                       for(j=0; j<chrFilterSize; j++)
+                       {
+                               u += chrSrc[j][i] * chrFilter[j];
+                               v += chrSrc[j][i + 2048] * chrFilter[j];
+                       }
+
+                       uDest[i]= MIN(MAX(u>>19, 0), 255);
+                       vDest[i]= MIN(MAX(v>>19, 0), 255);
+               }
+}
+
+static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
+                               int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
+                               uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
+{
+       //FIXME Optimize (just quickly writen not opti..)
+       int i;
+       for(i=0; i<dstW; i++)
+       {
+               int val=1<<18;
+               int j;
+               for(j=0; j<lumFilterSize; j++)
+                       val += lumSrc[j][i] * lumFilter[j];
+
+               dest[i]= MIN(MAX(val>>19, 0), 255);
+       }
+
+       if(uDest == NULL)
+               return;
+
+       if(dstFormat == IMGFMT_NV12)
+               for(i=0; i<chrDstW; i++)
+               {
+                       int u=1<<18;
+                       int v=1<<18;
+                       int j;
+                       for(j=0; j<chrFilterSize; j++)
+                       {
+                               u += chrSrc[j][i] * chrFilter[j];
+                               v += chrSrc[j][i + 2048] * chrFilter[j];
+                       }
+
+                       uDest[2*i]= MIN(MAX(u>>19, 0), 255);
+                       uDest[2*i+1]= MIN(MAX(v>>19, 0), 255);
+               }
+       else
+               for(i=0; i<chrDstW; i++)
+               {
+                       int u=1<<18;
+                       int v=1<<18;
+                       int j;
+                       for(j=0; j<chrFilterSize; j++)
+                       {
+                               u += chrSrc[j][i] * chrFilter[j];
+                               v += chrSrc[j][i + 2048] * chrFilter[j];
+                       }
+
+                       uDest[2*i]= MIN(MAX(v>>19, 0), 255);
+                       uDest[2*i+1]= MIN(MAX(u>>19, 0), 255);
+               }
+}
+
+#define YSCALE_YUV_2_PACKEDX_C(type) \
+               for(i=0; i<(dstW>>1); i++){\
+                       int j;\
+                       int Y1=1<<18;\
+                       int Y2=1<<18;\
+                       int U=1<<18;\
+                       int V=1<<18;\
+                       type *r, *b, *g;\
+                       const int i2= 2*i;\
+                       \
+                       for(j=0; j<lumFilterSize; j++)\
+                       {\
+                               Y1 += lumSrc[j][i2] * lumFilter[j];\
+                               Y2 += lumSrc[j][i2+1] * lumFilter[j];\
+                       }\
+                       for(j=0; j<chrFilterSize; j++)\
+                       {\
+                               U += chrSrc[j][i] * chrFilter[j];\
+                               V += chrSrc[j][i+2048] * chrFilter[j];\
+                       }\
+                       Y1>>=19;\
+                       Y2>>=19;\
+                       U >>=19;\
+                       V >>=19;\
+                       if((Y1|Y2|U|V)&256)\
+                       {\
+                               if(Y1>255)   Y1=255;\
+                               else if(Y1<0)Y1=0;\
+                               if(Y2>255)   Y2=255;\
+                               else if(Y2<0)Y2=0;\
+                               if(U>255)    U=255;\
+                               else if(U<0) U=0;\
+                               if(V>255)    V=255;\
+                               else if(V<0) V=0;\
+                       }
+                        
+#define YSCALE_YUV_2_RGBX_C(type) \
+                       YSCALE_YUV_2_PACKEDX_C(type)\
+                       r = c->table_rV[V];\
+                       g = c->table_gU[U] + c->table_gV[V];\
+                       b = c->table_bU[U];\
+
+#define YSCALE_YUV_2_PACKED2_C \
+               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+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
+
+#define YSCALE_YUV_2_RGB2_C(type) \
+                       YSCALE_YUV_2_PACKED2_C\
+                       type *r, *b, *g;\
+                       r = c->table_rV[V];\
+                       g = c->table_gU[U] + c->table_gV[V];\
+                       b = c->table_bU[U];\
+
+#define YSCALE_YUV_2_PACKED1_C \
+               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+2048])>>7;\
+
+#define YSCALE_YUV_2_RGB1_C(type) \
+                       YSCALE_YUV_2_PACKED1_C\
+                       type *r, *b, *g;\
+                       r = c->table_rV[V];\
+                       g = c->table_gU[U] + c->table_gV[V];\
+                       b = c->table_bU[U];\
+
+#define YSCALE_YUV_2_PACKED1B_C \
+               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+2048] + uvbuf1[i+2048])>>8;\
+
+#define YSCALE_YUV_2_RGB1B_C(type) \
+                       YSCALE_YUV_2_PACKED1B_C\
+                       type *r, *b, *g;\
+                       r = c->table_rV[V];\
+                       g = c->table_gU[U] + c->table_gV[V];\
+                       b = c->table_bU[U];\
+
+#define YSCALE_YUV_2_ANYRGB_C(func, func2)\
+       switch(c->dstFormat)\
+       {\
+       case IMGFMT_BGR32:\
+       case IMGFMT_RGB32:\
+               func(uint32_t)\
+                       ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
+                       ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
+               }               \
+               break;\
+       case IMGFMT_RGB24:\
+               func(uint8_t)\
+                       ((uint8_t*)dest)[0]= r[Y1];\
+                       ((uint8_t*)dest)[1]= g[Y1];\
+                       ((uint8_t*)dest)[2]= b[Y1];\
+                       ((uint8_t*)dest)[3]= r[Y2];\
+                       ((uint8_t*)dest)[4]= g[Y2];\
+                       ((uint8_t*)dest)[5]= b[Y2];\
+                       dest+=6;\
+               }\
+               break;\
+       case IMGFMT_BGR24:\
+               func(uint8_t)\
+                       ((uint8_t*)dest)[0]= b[Y1];\
+                       ((uint8_t*)dest)[1]= g[Y1];\
+                       ((uint8_t*)dest)[2]= r[Y1];\
+                       ((uint8_t*)dest)[3]= b[Y2];\
+                       ((uint8_t*)dest)[4]= g[Y2];\
+                       ((uint8_t*)dest)[5]= r[Y2];\
+                       dest+=6;\
+               }\
+               break;\
+       case IMGFMT_RGB16:\
+       case IMGFMT_BGR16:\
+               {\
+                       const int dr1= dither_2x2_8[y&1    ][0];\
+                       const int dg1= dither_2x2_4[y&1    ][0];\
+                       const int db1= dither_2x2_8[(y&1)^1][0];\
+                       const int dr2= dither_2x2_8[y&1    ][1];\
+                       const int dg2= dither_2x2_4[y&1    ][1];\
+                       const int db2= dither_2x2_8[(y&1)^1][1];\
+                       func(uint16_t)\
+                               ((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 IMGFMT_RGB15:\
+       case IMGFMT_BGR15:\
+               {\
+                       const int dr1= dither_2x2_8[y&1    ][0];\
+                       const int dg1= dither_2x2_8[y&1    ][1];\
+                       const int db1= dither_2x2_8[(y&1)^1][0];\
+                       const int dr2= dither_2x2_8[y&1    ][1];\
+                       const int dg2= dither_2x2_8[y&1    ][0];\
+                       const int db2= dither_2x2_8[(y&1)^1][1];\
+                       func(uint16_t)\
+                               ((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 IMGFMT_RGB8:\
+       case IMGFMT_BGR8:\
+               {\
+                       const uint8_t * const d64= dither_8x8_73[y&7];\
+                       const uint8_t * const d32= dither_8x8_32[y&7];\
+                       func(uint8_t)\
+                               ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
+                               ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
+                       }\
+               }\
+               break;\
+       case IMGFMT_RGB4:\
+       case IMGFMT_BGR4:\
+               {\
+                       const uint8_t * const d64= dither_8x8_73 [y&7];\
+                       const uint8_t * const d128=dither_8x8_220[y&7];\
+                       func(uint8_t)\
+                               ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
+                                                + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
+                       }\
+               }\
+               break;\
+       case IMGFMT_RG4B:\
+       case IMGFMT_BG4B:\
+               {\
+                       const uint8_t * const d64= dither_8x8_73 [y&7];\
+                       const uint8_t * const d128=dither_8x8_220[y&7];\
+                       func(uint8_t)\
+                               ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
+                               ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
+                       }\
+               }\
+               break;\
+       case IMGFMT_RGB1:\
+       case IMGFMT_BGR1:\
+               {\
+                       const uint8_t * const d128=dither_8x8_220[y&7];\
+                       uint8_t *g= c->table_gU[128] + c->table_gV[128];\
+                       for(i=0; i<dstW-7; i+=8){\
+                               int acc;\
+                               acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
+                               acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
+                               acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
+                               acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
+                               acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
+                               acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
+                               acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
+                               acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
+                               ((uint8_t*)dest)[0]= acc;\
+                               dest++;\
+                       }\
+\
+/*\
+((uint8_t*)dest)-= dstW>>4;\
+{\
+                       int acc=0;\
+                       int left=0;\
+                       static int top[1024];\
+                       static int last_new[1024][1024];\
+                       static int last_in3[1024][1024];\
+                       static int drift[1024][1024];\
+                       int topLeft=0;\
+                       int shift=0;\
+                       int count=0;\
+                       const uint8_t * const d128=dither_8x8_220[y&7];\
+                       int error_new=0;\
+                       int error_in3=0;\
+                       int f=0;\
+                       \
+                       for(i=dstW>>1; i<dstW; i++){\
+                               int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
+                               int in2 = (76309 * (in - 16) + 32768) >> 16;\
+                               int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
+                               int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
+                                       + (last_new[y][i] - in3)*f/256;\
+                               int new= old> 128 ? 255 : 0;\
+\
+                               error_new+= ABS(last_new[y][i] - new);\
+                               error_in3+= ABS(last_in3[y][i] - in3);\
+                               f= error_new - error_in3*4;\
+                               if(f<0) f=0;\
+                               if(f>256) f=256;\
+\
+                               topLeft= top[i];\
+                               left= top[i]= old - new;\
+                               last_new[y][i]= new;\
+                               last_in3[y][i]= in3;\
+\
+                               acc+= acc + (new&1);\
+                               if((i&7)==6){\
+                                       ((uint8_t*)dest)[0]= acc;\
+                                       ((uint8_t*)dest)++;\
+                               }\
+                       }\
+}\
+*/\
+               }\
+               break;\
+       case IMGFMT_YUY2:\
+               func2\
+                       ((uint8_t*)dest)[2*i2+0]= Y1;\
+                       ((uint8_t*)dest)[2*i2+1]= U;\
+                       ((uint8_t*)dest)[2*i2+2]= Y2;\
+                       ((uint8_t*)dest)[2*i2+3]= V;\
+               }               \
+               break;\
+       case IMGFMT_UYVY:\
+               func2\
+                       ((uint8_t*)dest)[2*i2+0]= U;\
+                       ((uint8_t*)dest)[2*i2+1]= Y1;\
+                       ((uint8_t*)dest)[2*i2+2]= V;\
+                       ((uint8_t*)dest)[2*i2+3]= Y2;\
+               }               \
+               break;\
+       }\
+
+
+static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
+                                   int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
+                                   uint8_t *dest, int dstW, int y)
+{
+       int i;
+       switch(c->dstFormat)
+       {
+       case IMGFMT_RGB32:
+       case IMGFMT_BGR32:
+               YSCALE_YUV_2_RGBX_C(uint32_t)
+                       ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
+                       ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
+               }
+               break;
+       case IMGFMT_RGB24:
+               YSCALE_YUV_2_RGBX_C(uint8_t)
+                       ((uint8_t*)dest)[0]= r[Y1];
+                       ((uint8_t*)dest)[1]= g[Y1];
+                       ((uint8_t*)dest)[2]= b[Y1];
+                       ((uint8_t*)dest)[3]= r[Y2];
+                       ((uint8_t*)dest)[4]= g[Y2];
+                       ((uint8_t*)dest)[5]= b[Y2];
+                       dest+=6;
+               }
+               break;
+       case IMGFMT_BGR24:
+               YSCALE_YUV_2_RGBX_C(uint8_t)
+                       ((uint8_t*)dest)[0]= b[Y1];
+                       ((uint8_t*)dest)[1]= g[Y1];
+                       ((uint8_t*)dest)[2]= r[Y1];
+                       ((uint8_t*)dest)[3]= b[Y2];
+                       ((uint8_t*)dest)[4]= g[Y2];
+                       ((uint8_t*)dest)[5]= r[Y2];
+                       dest+=6;
+               }
+               break;
+       case IMGFMT_RGB16:
+       case IMGFMT_BGR16:
+               {
+                       const int dr1= dither_2x2_8[y&1    ][0];
+                       const int dg1= dither_2x2_4[y&1    ][0];
+                       const int db1= dither_2x2_8[(y&1)^1][0];
+                       const int dr2= dither_2x2_8[y&1    ][1];
+                       const int dg2= dither_2x2_4[y&1    ][1];
+                       const int db2= dither_2x2_8[(y&1)^1][1];
+                       YSCALE_YUV_2_RGBX_C(uint16_t)
+                               ((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 IMGFMT_RGB15:
+       case IMGFMT_BGR15:
+               {
+                       const int dr1= dither_2x2_8[y&1    ][0];
+                       const int dg1= dither_2x2_8[y&1    ][1];
+                       const int db1= dither_2x2_8[(y&1)^1][0];
+                       const int dr2= dither_2x2_8[y&1    ][1];
+                       const int dg2= dither_2x2_8[y&1    ][0];
+                       const int db2= dither_2x2_8[(y&1)^1][1];
+                       YSCALE_YUV_2_RGBX_C(uint16_t)
+                               ((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 IMGFMT_RGB8:
+       case IMGFMT_BGR8:
+               {
+                       const uint8_t * const d64= dither_8x8_73[y&7];
+                       const uint8_t * const d32= dither_8x8_32[y&7];
+                       YSCALE_YUV_2_RGBX_C(uint8_t)
+                               ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
+                               ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
+                       }
+               }
+               break;
+       case IMGFMT_RGB4:
+       case IMGFMT_BGR4:
+               {
+                       const uint8_t * const d64= dither_8x8_73 [y&7];
+                       const uint8_t * const d128=dither_8x8_220[y&7];
+                       YSCALE_YUV_2_RGBX_C(uint8_t)
+                               ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
+                                                 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
+                       }
+               }
+               break;
+       case IMGFMT_RG4B:
+       case IMGFMT_BG4B:
+               {
+                       const uint8_t * const d64= dither_8x8_73 [y&7];
+                       const uint8_t * const d128=dither_8x8_220[y&7];
+                       YSCALE_YUV_2_RGBX_C(uint8_t)
+                               ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
+                               ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
+                       }
+               }
+               break;
+       case IMGFMT_RGB1:
+       case IMGFMT_BGR1:
+               {
+                       const uint8_t * const d128=dither_8x8_220[y&7];
+                       uint8_t *g= c->table_gU[128] + c->table_gV[128];
+                       int acc=0;
+                       for(i=0; i<dstW-1; i+=2){
+                               int j;
+                               int Y1=1<<18;
+                               int Y2=1<<18;
+
+                               for(j=0; j<lumFilterSize; j++)
+                               {
+                                       Y1 += lumSrc[j][i] * lumFilter[j];
+                                       Y2 += lumSrc[j][i+1] * lumFilter[j];
+                               }
+                               Y1>>=19;
+                               Y2>>=19;
+                               if((Y1|Y2)&256)
+                               {
+                                       if(Y1>255)   Y1=255;
+                                       else if(Y1<0)Y1=0;
+                                       if(Y2>255)   Y2=255;
+                                       else if(Y2<0)Y2=0;
+                               }
+                               acc+= acc + g[Y1+d128[(i+0)&7]];
+                               acc+= acc + g[Y2+d128[(i+1)&7]];
+                               if((i&7)==6){
+                                       ((uint8_t*)dest)[0]= acc;
+                                       dest++;
+                               }
+                       }
+               }
+               break;
+       case IMGFMT_YUY2:
+               YSCALE_YUV_2_PACKEDX_C(void)
+                       ((uint8_t*)dest)[2*i2+0]= Y1;
+                       ((uint8_t*)dest)[2*i2+1]= U;
+                       ((uint8_t*)dest)[2*i2+2]= Y2;
+                       ((uint8_t*)dest)[2*i2+3]= V;
+               }
+                break;
+       case IMGFMT_UYVY:
+               YSCALE_YUV_2_PACKEDX_C(void)
+                       ((uint8_t*)dest)[2*i2+0]= U;
+                       ((uint8_t*)dest)[2*i2+1]= Y1;
+                       ((uint8_t*)dest)[2*i2+2]= V;
+                       ((uint8_t*)dest)[2*i2+3]= Y2;
+               }
+                break;
+       }
+}
+
+
+//Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
+//Plain C versions
+#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
+#define COMPILE_C
+#endif
+
+#ifdef ARCH_POWERPC
+#if defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)
+#define COMPILE_ALTIVEC
+#endif //HAVE_ALTIVEC
+#endif //ARCH_POWERPC
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+
+#if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
+#define COMPILE_MMX
+#endif
+
+#if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
+#define COMPILE_MMX2
+#endif
+
+#if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
+#define COMPILE_3DNOW
+#endif
+#endif //ARCH_X86 || ARCH_X86_64
+
+#undef HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+
+#ifdef COMPILE_C
+#undef HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#undef HAVE_ALTIVEC
+#define RENAME(a) a ## _C
+#include "swscale_template.c"
+#endif
+
+#ifdef ARCH_POWERPC
+#ifdef COMPILE_ALTIVEC
+#undef RENAME
+#define HAVE_ALTIVEC
+#define RENAME(a) a ## _altivec
+#include "swscale_template.c"
+#endif
+#endif //ARCH_POWERPC
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+
+//X86 versions
+/*
+#undef RENAME
+#undef HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#define ARCH_X86
+#define RENAME(a) a ## _X86
+#include "swscale_template.c"
+*/
+//MMX versions
+#ifdef COMPILE_MMX
+#undef RENAME
+#define HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#define RENAME(a) a ## _MMX
+#include "swscale_template.c"
+#endif
+
+//MMX2 versions
+#ifdef COMPILE_MMX2
+#undef RENAME
+#define HAVE_MMX
+#define HAVE_MMX2
+#undef HAVE_3DNOW
+#define RENAME(a) a ## _MMX2
+#include "swscale_template.c"
+#endif
+
+//3DNOW versions
+#ifdef COMPILE_3DNOW
+#undef RENAME
+#define HAVE_MMX
+#undef HAVE_MMX2
+#define HAVE_3DNOW
+#define RENAME(a) a ## _3DNow
+#include "swscale_template.c"
+#endif
+
+#endif //ARCH_X86 || ARCH_X86_64
+
+// minor note: the HAVE_xyz is messed up after that line so don't use it
+
+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 void 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])
+{
+       int i;
+       int filterSize;
+       int filter2Size;
+       int minFilterSize;
+       double *filter=NULL;
+       double *filter2=NULL;
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+       if(flags & SWS_CPU_CAPS_MMX)
+               asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
+#endif
+
+       // Note the +1 is for the MMXscaler which reads over the end
+       *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
+
+       if(ABS(xInc - 0x10000) <10) // unscaled
+       {
+               int i;
+               filterSize= 1;
+               filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
+               for(i=0; i<dstW*filterSize; i++) filter[i]=0;
+
+               for(i=0; i<dstW; i++)
+               {
+                       filter[i*filterSize]=1;
+                       (*filterPos)[i]=i;
+               }
+
+       }
+       else if(flags&SWS_POINT) // lame looking point sampling mode
+       {
+               int i;
+               int xDstInSrc;
+               filterSize= 1;
+               filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
+               
+               xDstInSrc= xInc/2 - 0x8000;
+               for(i=0; i<dstW; i++)
+               {
+                       int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
+
+                       (*filterPos)[i]= xx;
+                       filter[i]= 1.0;
+                       xDstInSrc+= xInc;
+               }
+       }
+       else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
+       {
+               int i;
+               int xDstInSrc;
+               if     (flags&SWS_BICUBIC) filterSize= 4;
+               else if(flags&SWS_X      ) filterSize= 4;
+               else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
+               filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
+
+               xDstInSrc= xInc/2 - 0x8000;
+               for(i=0; i<dstW; i++)
+               {
+                       int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
+                       int j;
+
+                       (*filterPos)[i]= xx;
+                               //Bilinear upscale / linear interpolate / Area averaging
+                               for(j=0; j<filterSize; j++)
+                               {
+                                       double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
+                                       double coeff= 1.0 - d;
+                                       if(coeff<0) coeff=0;
+                                       filter[i*filterSize + j]= coeff;
+                                       xx++;
+                               }
+                       xDstInSrc+= xInc;
+               }
+       }
+       else
+       {
+               double xDstInSrc;
+               double sizeFactor, filterSizeInSrc;
+               const double xInc1= (double)xInc / (double)(1<<16);
+
+               if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
+               else if(flags&SWS_X)            sizeFactor= 8.0;
+               else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
+               else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
+               else if(flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
+               else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
+               else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
+               else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
+               else {
+                       sizeFactor= 0.0; //GCC warning killer
+                       ASSERT(0)
+               }
+               
+               if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
+               else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
+
+               filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
+               if(filterSize > srcW-2) filterSize=srcW-2;
+
+               filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
+
+               xDstInSrc= xInc1 / 2.0 - 0.5;
+               for(i=0; i<dstW; i++)
+               {
+                       int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
+                       int j;
+                       (*filterPos)[i]= xx;
+                       for(j=0; j<filterSize; j++)
+                       {
+                               double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
+                               double coeff;
+                               if(flags & SWS_BICUBIC)
+                               {
+                                       double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
+                                       double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
+
+                                       if(d<1.0) 
+                                               coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
+                                       else if(d<2.0)
+                                               coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
+                                       else
+                                               coeff=0.0;
+                               }
+/*                             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;
+                                       
+                                       if(d<1.0)
+                                               coeff = cos(d*PI);
+                                       else
+                                               coeff=-1.0;
+                                       if(coeff<0.0)   coeff= -pow(-coeff, A);
+                                       else            coeff=  pow( coeff, A);
+                                       coeff= coeff*0.5 + 0.5;
+                               }
+                               else if(flags & SWS_AREA)
+                               {
+                                       double srcPixelSize= 1.0/xInc1;
+                                       if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
+                                       else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
+                                       else coeff=0.0;
+                               }
+                               else if(flags & SWS_GAUSS)
+                               {
+                                       double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
+                                       coeff = pow(2.0, - p*d*d);
+                               }
+                               else if(flags & SWS_SINC)
+                               {
+                                       coeff = d ? sin(d*PI)/(d*PI) : 1.0;
+                               }
+                               else if(flags & SWS_LANCZOS)
+                               {
+                                       double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; 
+                                       coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
+                                       if(d>p) coeff=0;
+                               }
+                               else if(flags & SWS_BILINEAR)
+                               {
+                                       coeff= 1.0 - d;
+                                       if(coeff<0) coeff=0;
+                               }
+                               else if(flags & SWS_SPLINE)
+                               {
+                                       double p=-2.196152422706632;
+                                       coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
+                               }
+                               else {
+                                       coeff= 0.0; //GCC warning killer
+                                       ASSERT(0)
+                               }
+
+                               filter[i*filterSize + j]= coeff;
+                               xx++;
+                       }
+                       xDstInSrc+= xInc1;
+               }
+       }
+
+       /* apply src & dst Filter to filter -> filter2
+          free(filter);
+       */
+       ASSERT(filterSize>0)
+       filter2Size= filterSize;
+       if(srcFilter) filter2Size+= srcFilter->length - 1;
+       if(dstFilter) filter2Size+= dstFilter->length - 1;
+       ASSERT(filter2Size>0)
+       filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
+
+       for(i=0; i<dstW; i++)
+       {
+               int j;
+               SwsVector scaleFilter;
+               SwsVector *outVec;
+
+               scaleFilter.coeff= filter + i*filterSize;
+               scaleFilter.length= filterSize;
+
+               if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
+               else          outVec= &scaleFilter;
+
+               ASSERT(outVec->length == filter2Size)
+               //FIXME dstFilter
+
+               for(j=0; j<outVec->length; j++)
+               {
+                       filter2[i*filter2Size + j]= outVec->coeff[j];
+               }
+
+               (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
+
+               if(outVec != &scaleFilter) sws_freeVec(outVec);
+       }
+       free(filter); filter=NULL;
+
+       /* try to reduce the filter-size (step1 find size and shift left) */
+       // Assume its 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;
+               double cutOff=0.0;
+
+               /* get rid off near zero elements on the left by shifting left */
+               for(j=0; j<filter2Size; j++)
+               {
+                       int k;
+                       cutOff += ABS(filter2[i*filter2Size]);
+
+                       if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
+
+                       /* preserve Monotonicity because the core can't handle the filter otherwise */
+                       if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
+
+                       // Move filter coeffs left
+                       for(k=1; k<filter2Size; k++)
+                               filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
+                       filter2[i*filter2Size + k - 1]= 0.0;
+                       (*filterPos)[i]++;
+               }
+
+               cutOff=0.0;
+               /* count near zeros on the right */
+               for(j=filter2Size-1; j>0; j--)
+               {
+                       cutOff += ABS(filter2[i*filter2Size + j]);
+
+                       if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
+                       min--;
+               }
+
+               if(min>minFilterSize) minFilterSize= min;
+       }
+
+        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 filter.
+          // vectorizing is worth it only if you have
+          // decent-sized vector.
+          if (minFilterSize < 3)
+            filterAlign = 1;
+        }
+
+       ASSERT(minFilterSize > 0)
+       filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
+       ASSERT(filterSize > 0)
+       filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
+       *outFilterSize= filterSize;
+
+       if(flags&SWS_PRINT_INFO)
+               MSG_V("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;
+
+               for(j=0; j<filterSize; j++)
+               {
+                       if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
+                       else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
+               }
+       }
+       free(filter2); filter2=NULL;
+       
+
+       //FIXME try to align filterpos if possible
+
+       //fix borders
+       for(i=0; i<dstW; i++)
+       {
+               int j;
+               if((*filterPos)[i] < 0)
+               {
+                       // Move filter coeffs left to compensate for filterPos
+                       for(j=1; j<filterSize; j++)
+                       {
+                               int left= MAX(j + (*filterPos)[i], 0);
+                               filter[i*filterSize + left] += filter[i*filterSize + j];
+                               filter[i*filterSize + j]=0;
+                       }
+                       (*filterPos)[i]= 0;
+               }
+
+               if((*filterPos)[i] + filterSize > srcW)
+               {
+                       int shift= (*filterPos)[i] + filterSize - srcW;
+                       // Move filter coeffs right to compensate for filterPos
+                       for(j=filterSize-2; j>=0; j--)
+                       {
+                               int right= MIN(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 MMXscaler which reads over the end
+       /* align at 16 for AltiVec (needed by hScale_altivec_real) */
+       *outFilter= (int16_t*)memalign(16, *outFilterSize*(dstW+1)*sizeof(int16_t));
+       memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
+
+       /* Normalize & Store in outFilter */
+       for(i=0; i<dstW; i++)
+       {
+               int j;
+               double error=0;
+               double sum=0;
+               double scale= one;
+
+               for(j=0; j<filterSize; j++)
+               {
+                       sum+= filter[i*filterSize + j];
+               }
+               scale/= sum;
+               for(j=0; j<*outFilterSize; j++)
+               {
+                       double v= filter[i*filterSize + j]*scale + error;
+                       int intV= floor(v + 0.5);
+                       (*outFilter)[i*(*outFilterSize) + j]= intV;
+                       error = v - intV;
+               }
+       }
+       
+       (*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)];
+       }
+
+       free(filter);
+}
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
+{
+       uint8_t *fragmentA;
+       long imm8OfPShufW1A;
+       long imm8OfPShufW2A;
+       long fragmentLengthA;
+       uint8_t *fragmentB;
+       long imm8OfPShufW1B;
+       long imm8OfPShufW2B;
+       long fragmentLengthB;
+       int fragmentPos;
+
+       int xpos, i;
+
+       // create an optimized horizontal scaling routine
+
+       //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 0b, %0                     \n\t"
+               "lea 1b, %1                     \n\t"
+               "lea 2b, %2                     \n\t"
+               "dec %1                         \n\t"
+               "dec %2                         \n\t"
+               "sub %0, %1                     \n\t"
+               "sub %0, %2                     \n\t"
+               "lea 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 0b, %0                     \n\t"
+               "lea 1b, %1                     \n\t"
+               "lea 2b, %2                     \n\t"
+               "dec %1                         \n\t"
+               "dec %2                         \n\t"
+               "sub %0, %1                     \n\t"
+               "sub %0, %2                     \n\t"
+               "lea 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;
+
+                       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;
+
+                       if(d+1<4)
+                       {
+                               int maxShift= 3-(d+1);
+                               int shift=0;
+
+                               memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
+
+                               funnyCode[fragmentPos + imm8OfPShufW1B]=
+                                       (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
+                               funnyCode[fragmentPos + imm8OfPShufW2B]=
+                                       a | (b<<2) | (c<<4) | (d<<6);
+
+                               if(i+3>=dstW) shift=maxShift; //avoid overread
+                               else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
+
+                               if(shift && i>=shift)
+                               {
+                                       funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
+                                       funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
+                                       filterPos[i/2]-=shift;
+                               }
+
+                               fragmentPos+= fragmentLengthB;
+                       }
+                       else
+                       {
+                               int maxShift= 3-d;
+                               int shift=0;
+
+                               memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
+
+                               funnyCode[fragmentPos + imm8OfPShufW1A]=
+                               funnyCode[fragmentPos + imm8OfPShufW2A]=
+                                       a | (b<<2) | (c<<4) | (d<<6);
+
+                               if(i+4>=dstW) shift=maxShift; //avoid overread
+                               else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
+
+                               if(shift && i>=shift)
+                               {
+                                       funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
+                                       funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
+                                       filterPos[i/2]-=shift;
+                               }
+
+                               fragmentPos+= fragmentLengthA;
+                       }
+
+                       funnyCode[fragmentPos]= RET;
+               }
+               xpos+=xInc;
+       }
+       filterPos[i/2]= xpos>>16; // needed to jump to the next part
+}
+#endif // ARCH_X86 || ARCH_X86_64
+
+static void globalInit(void){
+    // generating tables:
+    int i;
+    for(i=0; i<768; i++){
+       int c= MIN(MAX(i-256, 0), 255);
+       clip_table[i]=c;
+    }
+}
+
+static SwsFunc getSwsFunc(int flags){
+    
+#ifdef RUNTIME_CPUDETECT
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+       // ordered per speed fasterst first
+       if(flags & SWS_CPU_CAPS_MMX2)
+               return swScale_MMX2;
+       else if(flags & SWS_CPU_CAPS_3DNOW)
+               return swScale_3DNow;
+       else if(flags & SWS_CPU_CAPS_MMX)
+               return swScale_MMX;
+       else
+               return swScale_C;
+
+#else
+#ifdef ARCH_POWERPC
+       if(flags & SWS_CPU_CAPS_ALTIVEC)
+         return swScale_altivec;
+       else
+         return swScale_C;
+#endif
+       return swScale_C;
+#endif
+#else //RUNTIME_CPUDETECT
+#ifdef HAVE_MMX2
+       return swScale_MMX2;
+#elif defined (HAVE_3DNOW)
+       return swScale_3DNow;
+#elif defined (HAVE_MMX)
+       return swScale_MMX;
+#elif defined (HAVE_ALTIVEC)
+       return swScale_altivec;
+#else
+       return swScale_C;
+#endif
+#endif //!RUNTIME_CPUDETECT
+}
+
+static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
+             int srcSliceH, uint8_t* dstParam[], int dstStride[]){
+       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;
+               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];
+               }
+       }
+       dst = dstParam[1] + dstStride[1]*srcSliceY/2;
+       if (c->dstFormat == IMGFMT_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, 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, uint8_t* src[], int srcStride[], int srcSliceY,
+             int srcSliceH, uint8_t* dstParam[], int dstStride[]){
+       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;
+}
+
+/* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
+static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
+                          int srcSliceH, uint8_t* dst[], int dstStride[]){
+       const int srcFormat= c->srcFormat;
+       const int dstFormat= c->dstFormat;
+       const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
+       const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
+       const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 
+       const int dstId= (dstFormat&0xFF)>>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: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
+                                vo_format_name(srcFormat), vo_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: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
+                                vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
+               }
+       }else{
+               MSG_ERR("swScaler: internal error %s -> %s converter\n", 
+                        vo_format_name(srcFormat), vo_format_name(dstFormat));
+       }
+
+       if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
+               conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
+       else
+       {
+               int i;
+               uint8_t *srcPtr= src[0];
+               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];
+               }
+       }     
+       return srcSliceH;
+}
+
+static int bgr24toyv12Wrapper(SwsContext *c, 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]);
+       return srcSliceH;
+}
+
+static int yvu9toyv12Wrapper(SwsContext *c, 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{
+               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];
+               }
+       }
+
+       if(c->dstFormat==IMGFMT_YV12){
+               planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
+               planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
+       }else{
+               planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
+               planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
+       }
+       return srcSliceH;
+}
+
+/**
+ * bring pointers in YUV order instead of YVU
+ */
+static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
+       if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
+           || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
+               sortedP[0]= p[0];
+               sortedP[1]= p[2];
+               sortedP[2]= p[1];
+               sortedStride[0]= stride[0];
+               sortedStride[1]= stride[2];
+               sortedStride[2]= stride[1];
+       }
+       else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
+       {
+               sortedP[0]= p[0];
+               sortedP[1]= 
+               sortedP[2]= NULL;
+               sortedStride[0]= stride[0];
+               sortedStride[1]= 
+               sortedStride[2]= 0;
+       }
+       else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
+       {
+               sortedP[0]= p[0];
+               sortedP[1]= p[1];
+               sortedP[2]= p[2];
+               sortedStride[0]= stride[0];
+               sortedStride[1]= stride[1];
+               sortedStride[2]= stride[2];
+       }
+       else if(format == IMGFMT_NV12 || format == IMGFMT_NV21)
+       {
+               sortedP[0]= p[0];
+               sortedP[1]= p[1];
+               sortedP[2]= NULL;
+               sortedStride[0]= stride[0];
+               sortedStride[1]= stride[1];
+               sortedStride[2]= 0;
+       }else{
+               MSG_ERR("internal error in orderYUV\n");
+       }
+}
+
+/* unscaled copy like stuff (assumes nearly identical formats) */
+static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
+             int srcSliceH, uint8_t* dst[], int dstStride[]){
+
+       if(isPacked(c->srcFormat))
+       {
+               if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
+                       memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
+               else
+               {
+                       int i;
+                       uint8_t *srcPtr= src[0];
+                       uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
+                       int length=0;
+
+                       /* universal length finder */
+                       while(length+c->srcW <= ABS(dstStride[0]) 
+                          && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
+                       ASSERT(length!=0);
+
+                       for(i=0; i<srcSliceH; i++)
+                       {
+                               memcpy(dstPtr, srcPtr, length);
+                               srcPtr+= srcStride[0];
+                               dstPtr+= dstStride[0];
+                       }
+               }
+       }
+       else 
+       { /* Planar YUV or gray */
+               int plane;
+               for(plane=0; plane<3; plane++)
+               {
+                       int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
+                       int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
+                       int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
+
+                       if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
+                       {
+                               if(!isGray(c->dstFormat))
+                                       memset(dst[plane], 128, dstStride[plane]*height);
+                       }
+                       else
+                       {
+                               if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
+                                       memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
+                               else
+                               {
+                                       int i;
+                                       uint8_t *srcPtr= src[plane];
+                                       uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
+                                       for(i=0; i<height; i++)
+                                       {
+                                               memcpy(dstPtr, srcPtr, length);
+                                               srcPtr+= srcStride[plane];
+                                               dstPtr+= dstStride[plane];
+                                       }
+                               }
+                       }
+               }
+       }
+       return srcSliceH;
+}
+
+static int remove_dup_fourcc(int fourcc)
+{
+       switch(fourcc)
+       {
+           case IMGFMT_I420:
+           case IMGFMT_IYUV: return IMGFMT_YV12;
+           case IMGFMT_Y8  : return IMGFMT_Y800;
+           case IMGFMT_IF09: return IMGFMT_YVU9;
+           default: return fourcc;
+       }
+}
+
+static void getSubSampleFactors(int *h, int *v, int format){
+       switch(format){
+       case IMGFMT_UYVY:
+       case IMGFMT_YUY2:
+               *h=1;
+               *v=0;
+               break;
+       case IMGFMT_YV12:
+       case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
+       case IMGFMT_NV12:
+       case IMGFMT_NV21:
+               *h=1;
+               *v=1;
+               break;
+       case IMGFMT_YVU9:
+               *h=2;
+               *v=2;
+               break;
+       case IMGFMT_444P:
+               *h=0;
+               *v=0;
+               break;
+       case IMGFMT_422P:
+               *h=1;
+               *v=0;
+               break;
+       case IMGFMT_411P:
+               *h=2;
+               *v=0;
+               break;
+       default:
+               *h=0;
+               *v=0;
+               break;
+       }
+}
+
+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;
+}
+
+/**
+ * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
+ * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
+ * @return -1 if not supported
+ */
+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;
+
+       if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
+       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;
+
+       c->uOffset=   0x0400040004000400LL;
+       c->vOffset=   0x0400040004000400LL;
+
+       if(!srcRange){
+               cy= (cy*255) / 219;
+               oy= 16<<16;
+       }
+
+       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;
+
+       yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
+       //FIXME factorize
+
+#ifdef COMPILE_ALTIVEC
+       if (c->flags & SWS_CPU_CAPS_ALTIVEC)
+           yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
+#endif 
+       return 0;
+}
+
+/**
+ * @return -1 if not supported
+ */
+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;       
+}
+
+SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
+                         SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
+
+       SwsContext *c;
+       int i;
+       int usesVFilter, usesHFilter;
+       int unscaled, needsDither;
+       int srcFormat, dstFormat;
+       SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+       if(flags & SWS_CPU_CAPS_MMX)
+               asm volatile("emms\n\t"::: "memory");
+#endif
+
+#ifndef 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);
+#ifdef HAVE_MMX2
+       flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
+#elif defined (HAVE_3DNOW)
+       flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
+#elif defined (HAVE_MMX)
+       flags |= SWS_CPU_CAPS_MMX;
+#elif defined (HAVE_ALTIVEC)
+       flags |= SWS_CPU_CAPS_ALTIVEC;
+#endif
+#endif
+       if(clip_table[512] != 255) globalInit();
+       if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
+
+       /* avoid duplicate Formats, so we don't need to check to much */
+       srcFormat = remove_dup_fourcc(origSrcFormat);
+       dstFormat = remove_dup_fourcc(origDstFormat);
+
+       unscaled = (srcW == dstW && srcH == dstH);
+       needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
+                    && (dstFormat&0xFF)<24
+                    && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
+
+       if(!isSupportedIn(srcFormat)) 
+       {
+               MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
+               return NULL;
+       }
+       if(!isSupportedOut(dstFormat))
+       {
+               MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
+               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
+       {
+                MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
+                       srcW, srcH, dstW, dstH);
+               return NULL;
+       }
+
+       if(!dstFilter) dstFilter= &dummyFilter;
+       if(!srcFilter) srcFilter= &dummyFilter;
+
+       c= memalign(64, sizeof(SwsContext));
+       memset(c, 0, sizeof(SwsContext));
+
+       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->origDstFormat= origDstFormat;
+       c->origSrcFormat= origSrcFormat;
+        c->vRounder= 4* 0x0001000100010001ULL;
+
+       usesHFilter= usesVFilter= 0;
+       if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
+       if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
+       if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
+       if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
+       if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
+       if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
+       if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
+       if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
+
+       getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
+       getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
+
+       // reuse chroma for 2 pixles 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 2. pixel for chroma calculation unless user wants full chroma
+       if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 
+               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;
+       }
+
+       c->chrIntHSubSample= c->chrDstHSubSample;
+       c->chrIntVSubSample= c->chrSrcVSubSample;
+
+       // note the -((-x)>>y) is so that we allways 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, Inverse_Table_6_9[SWS_CS_DEFAULT], 0, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, 0, 0, 1<<16, 1<<16); 
+
+       /* unscaled special Cases */
+       if(unscaled && !usesHFilter && !usesVFilter)
+       {
+               /* yv12_to_nv12 */
+               if(srcFormat == IMGFMT_YV12 && (dstFormat == IMGFMT_NV12 || dstFormat == IMGFMT_NV21))
+               {
+                       c->swScale= PlanarToNV12Wrapper;
+               }
+               /* yuv2bgr */
+               if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
+               {
+                       c->swScale= yuv2rgb_get_func_ptr(c);
+               }
+               
+               if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
+               {
+                       c->swScale= yvu9toyv12Wrapper;
+               }
+
+               /* bgr24toYV12 */
+               if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
+                       c->swScale= bgr24toyv12Wrapper;
+               
+               /* rgb/bgr -> rgb/bgr (no dither needed forms) */
+               if(   (isBGR(srcFormat) || isRGB(srcFormat))
+                  && (isBGR(dstFormat) || isRGB(dstFormat)) 
+                  && !needsDither)
+                       c->swScale= rgb2rgbWrapper;
+
+               /* LQ converters if -sws 0 or -sws 4*/
+               if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
+                       /* rgb/bgr -> rgb/bgr (dither needed forms) */
+                       if(  (isBGR(srcFormat) || isRGB(srcFormat))
+                         && (isBGR(dstFormat) || isRGB(dstFormat)) 
+                         && needsDither)
+                               c->swScale= rgb2rgbWrapper;
+
+                       /* yv12_to_yuy2 */
+                       if(srcFormat == IMGFMT_YV12 && 
+                           (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
+                       {
+                               if (dstFormat == IMGFMT_YUY2)
+                                   c->swScale= PlanarToYuy2Wrapper;
+                               else
+                                   c->swScale= PlanarToUyvyWrapper;
+                       }
+               }
+
+#ifdef COMPILE_ALTIVEC
+               if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
+                   ((srcFormat == IMGFMT_YV12 && 
+                     (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
+                 // unscaled YV12 -> packed YUV, we want speed
+                 if (dstFormat == IMGFMT_YUY2)
+                   c->swScale= yv12toyuy2_unscaled_altivec;
+                 else
+                   c->swScale= yv12touyvy_unscaled_altivec;
+               }
+#endif
+
+               /* simple copy */
+               if(   srcFormat == dstFormat
+                  || (isPlanarYUV(srcFormat) && isGray(dstFormat))
+                  || (isPlanarYUV(dstFormat) && isGray(srcFormat))
+                 )
+               {
+                       c->swScale= simpleCopy;
+               }
+
+               if(c->swScale){
+                       if(flags&SWS_PRINT_INFO)
+                               MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
+                                       vo_format_name(srcFormat), vo_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)
+                               MSG_INFO("SwScaler: 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 noone shuld 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 x86asm 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;
+               }
+       }
+
+       /* precalculate horizontal scaler filter coefficients */
+       {
+               const int filterAlign=
+                 (flags & SWS_CPU_CAPS_MMX) ? 4 :
+                 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
+                 1;
+
+               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);
+               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);
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+// can't downscale !!!
+               if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
+               {
+#define MAX_FUNNY_CODE_SIZE 10000
+#ifdef MAP_ANONYMOUS
+                       c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
+                       c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
+#else
+                       c->funnyYCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
+                       c->funnyUVCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
+#endif
+
+                       c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
+                       c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
+                       c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
+                       c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
+
+                       initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
+                       initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
+               }
+#endif
+       } // Init Horizontal stuff
+
+
+
+       /* precalculate vertical scaler filter coefficients */
+       {
+               const int filterAlign=
+                 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
+                 1;
+
+               initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
+                               srcH      ,        dstH, filterAlign, (1<<12)-4,
+                               (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
+                               srcFilter->lumV, dstFilter->lumV, c->param);
+               initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
+                               c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
+                               (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
+                               srcFilter->chrV, dstFilter->chrV, c->param);
+
+#ifdef HAVE_ALTIVEC
+               c->vYCoeffsBank = memalign (16, sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
+               c->vCCoeffsBank = memalign (16, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
+
+               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];
+               }
+
+               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];
+               }
+#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= MAX(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
+       c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
+       c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
+       //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++)
+               c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(16, 4000);
+       for(i=0; i<c->vChrBufSize; i++)
+               c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(16, 8000);
+
+       //try to avoid drawing green stuff between the right end and the stride end
+       for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
+       for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
+
+       ASSERT(c->chrDstH <= dstH)
+
+       if(flags&SWS_PRINT_INFO)
+       {
+#ifdef DITHER1XBPP
+               char *dither= " dithered";
+#else
+               char *dither= "";
+#endif
+               if(flags&SWS_FAST_BILINEAR)
+                       MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
+               else if(flags&SWS_BILINEAR)
+                       MSG_INFO("\nSwScaler: BILINEAR scaler, ");
+               else if(flags&SWS_BICUBIC)
+                       MSG_INFO("\nSwScaler: BICUBIC scaler, ");
+               else if(flags&SWS_X)
+                       MSG_INFO("\nSwScaler: Experimental scaler, ");
+               else if(flags&SWS_POINT)
+                       MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
+               else if(flags&SWS_AREA)
+                       MSG_INFO("\nSwScaler: Area Averageing scaler, ");
+               else if(flags&SWS_BICUBLIN)
+                       MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
+               else if(flags&SWS_GAUSS)
+                       MSG_INFO("\nSwScaler: Gaussian scaler, ");
+               else if(flags&SWS_SINC)
+                       MSG_INFO("\nSwScaler: Sinc scaler, ");
+               else if(flags&SWS_LANCZOS)
+                       MSG_INFO("\nSwScaler: Lanczos scaler, ");
+               else if(flags&SWS_SPLINE)
+                       MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
+               else
+                       MSG_INFO("\nSwScaler: ehh flags invalid?! ");
+
+               if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
+                       MSG_INFO("from %s to%s %s ", 
+                               vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
+               else
+                       MSG_INFO("from %s to %s ", 
+                               vo_format_name(srcFormat), vo_format_name(dstFormat));
+
+               if(flags & SWS_CPU_CAPS_MMX2)
+                       MSG_INFO("using MMX2\n");
+               else if(flags & SWS_CPU_CAPS_3DNOW)
+                       MSG_INFO("using 3DNOW\n");
+               else if(flags & SWS_CPU_CAPS_MMX)
+                       MSG_INFO("using MMX\n");
+               else if(flags & SWS_CPU_CAPS_ALTIVEC)
+                       MSG_INFO("using AltiVec\n");
+               else 
+                       MSG_INFO("using C\n");
+       }
+
+       if(flags & SWS_PRINT_INFO)
+       {
+               if(flags & SWS_CPU_CAPS_MMX)
+               {
+                       if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
+                               MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
+                       else
+                       {
+                               if(c->hLumFilterSize==4)
+                                       MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
+                               else if(c->hLumFilterSize==8)
+                                       MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
+                               else
+                                       MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
+
+                               if(c->hChrFilterSize==4)
+                                       MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
+                               else if(c->hChrFilterSize==8)
+                                       MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
+                               else
+                                       MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
+                       }
+               }
+               else
+               {
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+                       MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
+#else
+                       if(flags & SWS_FAST_BILINEAR)
+                               MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
+                       else
+                               MSG_V("SwScaler: using C scaler for horizontal scaling\n");
+#endif
+               }
+               if(isPlanarYUV(dstFormat))
+               {
+                       if(c->vLumFilterSize==1)
+                               MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+                       else
+                               MSG_V("SwScaler: 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)
+                               MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
+                                      "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+                       else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
+                               MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+                       else
+                               MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+               }
+
+               if(dstFormat==IMGFMT_BGR24)
+                       MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
+                               (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
+               else if(dstFormat==IMGFMT_BGR32)
+                       MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+               else if(dstFormat==IMGFMT_BGR16)
+                       MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+               else if(dstFormat==IMGFMT_BGR15)
+                       MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+
+               MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
+       }
+       if(flags & SWS_PRINT_INFO)
+       {
+               MSG_DBG2("SwScaler: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);
+               MSG_DBG2("SwScaler: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(flags);
+       return c;
+}
+
+/**
+ * swscale warper, so we don't need to export the SwsContext.
+ * assumes planar YUV to be in YUV order instead of YVU
+ */
+int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
+                           int srcSliceH, uint8_t* dst[], int dstStride[]){
+       if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
+           MSG_ERR("swScaler: slices start in the middle!\n");
+           return 0;
+       }
+       if (c->sliceDir == 0) {
+           if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
+       }
+
+       // copy strides, so they can safely be modified
+       if (c->sliceDir == 1) {
+           // slices go from top to bottom
+           int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
+           int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
+           return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
+       } else {
+           // slices go from bottom to top => we flip the image internally
+           uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
+                              src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
+                              src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
+           };
+           uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
+                              dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
+                              dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
+           int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
+           int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
+           
+           return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
+       }
+}
+
+/**
+ * swscale warper, so we don't need to export the SwsContext
+ */
+int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
+                           int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
+       int srcStride[3];
+       int dstStride[3];
+       uint8_t *src[3];
+       uint8_t *dst[3];
+       sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
+       sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
+//printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
+
+       return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
+}
+
+SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 
+                               float lumaSharpen, float chromaSharpen,
+                               float chromaHShift, float chromaVShift,
+                               int verbose)
+{
+       SwsFilter *filter= malloc(sizeof(SwsFilter));
+
+       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_printVec(filter->chrH);
+       if(verbose) sws_printVec(filter->lumH);
+
+        return filter;
+}
+
+/**
+ * returns a normalized gaussian curve used to filter stuff
+ * quality=3 is high quality, lowwer is lowwer quality
+ */
+SwsVector *sws_getGaussianVec(double variance, double quality){
+       const int length= (int)(variance*quality + 0.5) | 1;
+       int i;
+       double *coeff= memalign(sizeof(double), length*sizeof(double));
+       double middle= (length-1)*0.5;
+       SwsVector *vec= malloc(sizeof(SwsVector));
+
+       vec->coeff= coeff;
+       vec->length= length;
+
+       for(i=0; i<length; i++)
+       {
+               double dist= i-middle;
+               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;
+       double *coeff= memalign(sizeof(double), length*sizeof(double));
+       SwsVector *vec= malloc(sizeof(SwsVector));
+
+       vec->coeff= coeff;
+       vec->length= length;
+
+       for(i=0; i<length; i++)
+               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;
+       double *coeff= memalign(sizeof(double), length*sizeof(double));
+       int i, j;
+       SwsVector *vec= malloc(sizeof(SwsVector));
+
+       vec->coeff= coeff;
+       vec->length= length;
+
+       for(i=0; i<length; i++) coeff[i]= 0.0;
+
+       for(i=0; i<a->length; i++)
+       {
+               for(j=0; j<b->length; j++)
+               {
+                       coeff[i+j]+= a->coeff[i]*b->coeff[j];
+               }
+       }
+
+       return vec;
+}
+
+static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
+       int length= MAX(a->length, b->length);
+       double *coeff= memalign(sizeof(double), length*sizeof(double));
+       int i;
+       SwsVector *vec= malloc(sizeof(SwsVector));
+
+       vec->coeff= coeff;
+       vec->length= length;
+
+       for(i=0; i<length; i++) coeff[i]= 0.0;
+
+       for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
+       for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
+
+       return vec;
+}
+
+static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
+       int length= MAX(a->length, b->length);
+       double *coeff= memalign(sizeof(double), length*sizeof(double));
+       int i;
+       SwsVector *vec= malloc(sizeof(SwsVector));
+
+       vec->coeff= coeff;
+       vec->length= length;
+
+       for(i=0; i<length; i++) coeff[i]= 0.0;
+
+       for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
+       for(i=0; i<b->length; i++) 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 + ABS(shift)*2;
+       double *coeff= memalign(sizeof(double), length*sizeof(double));
+       int i;
+       SwsVector *vec= malloc(sizeof(SwsVector));
+
+       vec->coeff= coeff;
+       vec->length= length;
+
+       for(i=0; i<length; i++) coeff[i]= 0.0;
+
+       for(i=0; i<a->length; i++)
+       {
+               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);
+       free(a->coeff);
+       a->coeff= shifted->coeff;
+       a->length= shifted->length;
+       free(shifted);
+}
+
+void sws_addVec(SwsVector *a, SwsVector *b){
+       SwsVector *sum= sws_sumVec(a, b);
+       free(a->coeff);
+       a->coeff= sum->coeff;
+       a->length= sum->length;
+       free(sum);
+}
+
+void sws_subVec(SwsVector *a, SwsVector *b){
+       SwsVector *diff= sws_diffVec(a, b);
+       free(a->coeff);
+       a->coeff= diff->coeff;
+       a->length= diff->length;
+       free(diff);
+}
+
+void sws_convVec(SwsVector *a, SwsVector *b){
+       SwsVector *conv= sws_getConvVec(a, b);
+       free(a->coeff);  
+       a->coeff= conv->coeff;
+       a->length= conv->length;
+       free(conv);
+}
+
+SwsVector *sws_cloneVec(SwsVector *a){
+       double *coeff= memalign(sizeof(double), a->length*sizeof(double));
+       int i;
+       SwsVector *vec= malloc(sizeof(SwsVector));
+
+       vec->coeff= coeff;
+       vec->length= a->length;
+
+       for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
+
+       return vec;
+}
+
+void sws_printVec(SwsVector *a){
+       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);
+               MSG_DBG2("%1.3f ", a->coeff[i]);
+               for(;x>0; x--) MSG_DBG2(" ");
+               MSG_DBG2("|\n");
+       }
+}
+
+void sws_freeVec(SwsVector *a){
+       if(!a) return;
+       if(a->coeff) free(a->coeff);
+       a->coeff=NULL;
+       a->length=0;
+       free(a);
+}
+
+void sws_freeFilter(SwsFilter *filter){
+       if(!filter) return;
+
+       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);
+       free(filter);
+}
+
+
+void sws_freeContext(SwsContext *c){
+       int i;
+       if(!c) return;
+
+       if(c->lumPixBuf)
+       {
+               for(i=0; i<c->vLumBufSize; i++)
+               {
+                       if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
+                       c->lumPixBuf[i]=NULL;
+               }
+               free(c->lumPixBuf);
+               c->lumPixBuf=NULL;
+       }
+
+       if(c->chrPixBuf)
+       {
+               for(i=0; i<c->vChrBufSize; i++)
+               {
+                       if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
+                       c->chrPixBuf[i]=NULL;
+               }
+               free(c->chrPixBuf);
+               c->chrPixBuf=NULL;
+       }
+
+       if(c->vLumFilter) free(c->vLumFilter);
+       c->vLumFilter = NULL;
+       if(c->vChrFilter) free(c->vChrFilter);
+       c->vChrFilter = NULL;
+       if(c->hLumFilter) free(c->hLumFilter);
+       c->hLumFilter = NULL;
+       if(c->hChrFilter) free(c->hChrFilter);
+       c->hChrFilter = NULL;
+#ifdef HAVE_ALTIVEC
+       if(c->vYCoeffsBank) free(c->vYCoeffsBank);
+       c->vYCoeffsBank = NULL;
+       if(c->vCCoeffsBank) free(c->vCCoeffsBank);
+       c->vCCoeffsBank = NULL;
+#endif
+
+       if(c->vLumFilterPos) free(c->vLumFilterPos);
+       c->vLumFilterPos = NULL;
+       if(c->vChrFilterPos) free(c->vChrFilterPos);
+       c->vChrFilterPos = NULL;
+       if(c->hLumFilterPos) free(c->hLumFilterPos);
+       c->hLumFilterPos = NULL;
+       if(c->hChrFilterPos) free(c->hChrFilterPos);
+       c->hChrFilterPos = NULL;
+
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+#ifdef MAP_ANONYMOUS
+       if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
+       if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
+#else
+       if(c->funnyYCode) free(c->funnyYCode);
+       if(c->funnyUVCode) free(c->funnyUVCode);
+#endif
+       c->funnyYCode=NULL;
+       c->funnyUVCode=NULL;
+#endif
+
+       if(c->lumMmx2Filter) free(c->lumMmx2Filter);
+       c->lumMmx2Filter=NULL;
+       if(c->chrMmx2Filter) free(c->chrMmx2Filter);
+       c->chrMmx2Filter=NULL;
+       if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
+       c->lumMmx2FilterPos=NULL;
+       if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
+       c->chrMmx2FilterPos=NULL;
+       if(c->yuvTable) free(c->yuvTable);
+       c->yuvTable=NULL;
+
+       free(c);
+}
+
diff --git a/libswscale/swscale.h b/libswscale/swscale.h
new file mode 100644 (file)
index 0000000..e589e7d
--- /dev/null
@@ -0,0 +1,129 @@
+/*
+    Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
+
+    This program 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.
+
+    This program is distributed in the hope that it will be useful,