Move yuv2rgb to postprocess

git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@2733 b3059339-0415-0410-9bf9-f77b7e298cf2

7 files changed, 936 insertions, 3 deletions

diff --git a/postproc/Makefile b/postproc/Makefile
index 8d16b53650..5f74289438 100644
--- a/postproc/Makefile
+++ b/postproc/Makefile
@@ -3,10 +3,14 @@ include ../config.mak
 
 LIBNAME = libpostproc.a
 
-SRCS=postprocess.c swscale.c rgb2rgb.c
+SRCS=postprocess.c swscale.c rgb2rgb.c yuv2rgb.c
 OBJS=$(SRCS:.c=.o)
 
-CFLAGS  = $(OPTFLAGS) -I. -I.. -Wall
+ifeq ($(TARGET_ARCH_X86),yes)
+SRCS += yuv2rgb_mmx.c
+endif
+
+CFLAGS  = $(OPTFLAGS) $(MLIB_INC) -I. -I.. -Wall
 # -I/usr/X11R6/include/
 
 .SUFFIXES: .c .o
diff --git a/postproc/rgb2rgb.c b/postproc/rgb2rgb.c
index ab28157e20..959988245b 100644
--- a/postproc/rgb2rgb.c
+++ b/postproc/rgb2rgb.c
@@ -1,6 +1,9 @@
 /*
  *
  *  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 stuff & yuv stuff by Michael
  */
diff --git a/postproc/rgb2rgb.h b/postproc/rgb2rgb.h
index 4851725703..771ccb8bb2 100644
--- a/postproc/rgb2rgb.h
+++ b/postproc/rgb2rgb.h
@@ -1,7 +1,9 @@
 /*
  *
  *  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
@@ -28,4 +30,16 @@ extern void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t
 	unsigned int width, unsigned int height,
 	unsigned int lumStride, unsigned int chromStride, unsigned int srcStride);
 
+#define MODE_RGB  0x1
+#define MODE_BGR  0x2
+
+typedef void (* yuv2rgb_fun) (uint8_t * image, uint8_t * py,
+			      uint8_t * pu, uint8_t * pv,
+			      int h_size, int v_size,
+			      int rgb_stride, int y_stride, int uv_stride);
+
+extern yuv2rgb_fun yuv2rgb;
+
+void yuv2rgb_init (int bpp, int mode);
+
 #endif
diff --git a/postproc/rgb2rgb_template.c b/postproc/rgb2rgb_template.c
index ab28157e20..959988245b 100644
--- a/postproc/rgb2rgb_template.c
+++ b/postproc/rgb2rgb_template.c
@@ -1,6 +1,9 @@
 /*
  *
  *  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 stuff & yuv stuff by Michael
  */
diff --git a/postproc/yuv2rgb.c b/postproc/yuv2rgb.c
new file mode 100644
index 0000000000..f0599618b5
--- /dev/null
+++ b/postproc/yuv2rgb.c
@@ -0,0 +1,422 @@
+/*
+ * yuv2rgb.c, Software YUV to RGB coverter
+ *
+ *  Copyright (C) 1999, Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
+ *  All Rights Reserved.
+ *
+ *  Functions broken out from display_x11.c and several new modes
+ *  added by Håkan Hjort <d95hjort@dtek.chalmers.se>
+ *
+ *  15 & 16 bpp support by Franck Sicard <Franck.Sicard@solsoft.fr>
+ *
+ *  This file is part of mpeg2dec, a free MPEG-2 video decoder
+ *
+ *  mpeg2dec 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, or (at your option)
+ *  any later version.
+ *
+ *  mpeg2dec 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 GNU Make; see the file COPYING.  If not, write to
+ *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <inttypes.h>
+
+#include "config.h"
+//#include "video_out.h"
+#include "rgb2rgb.h"
+
+#ifdef HAVE_MLIB
+#include "yuv2rgb_mlib.c"
+#endif
+
+extern yuv2rgb_fun yuv2rgb_init_mmx (int bpp, int mode);
+
+
+uint32_t matrix_coefficients = 6;
+
+const int32_t Inverse_Table_6_9[8][4] = {
+    {117504, 138453, 13954, 34903}, /* no sequence_display_extension */
+    {117504, 138453, 13954, 34903}, /* ITU-R Rec. 709 (1990) */
+    {104597, 132201, 25675, 53279}, /* unspecified */
+    {104597, 132201, 25675, 53279}, /* reserved */
+    {104448, 132798, 24759, 53109}, /* FCC */
+    {104597, 132201, 25675, 53279}, /* ITU-R Rec. 624-4 System B, G */
+    {104597, 132201, 25675, 53279}, /* SMPTE 170M */
+    {117579, 136230, 16907, 35559}  /* SMPTE 240M (1987) */
+};
+
+static void yuv2rgb_c_init (int bpp, int mode);
+
+yuv2rgb_fun yuv2rgb;
+
+static void (* yuv2rgb_c_internal) (uint8_t *, uint8_t *,
+				    uint8_t *, uint8_t *,
+				    void *, void *, int);
+
+static void yuv2rgb_c (void * dst, uint8_t * py, 
+		       uint8_t * pu, uint8_t * pv, 
+		       int h_size, int v_size, 
+		       int rgb_stride, int y_stride, int uv_stride) 
+{
+    v_size >>= 1;
+
+    while (v_size--) {
+	yuv2rgb_c_internal (py, py + y_stride, pu, pv, dst, dst + rgb_stride,
+			    h_size);
+
+	py += 2 * y_stride;
+	pu += uv_stride;
+	pv += uv_stride;
+	dst += 2 * rgb_stride;
+    }
+}
+
+void yuv2rgb_init (int bpp, int mode) 
+{
+    yuv2rgb = NULL;
+#ifdef HAVE_MMX
+    if (yuv2rgb == NULL /*&& (config.flags & VO_MMX_ENABLE)*/) {
+        yuv2rgb = yuv2rgb_init_mmx (bpp, mode);
+        if (yuv2rgb != NULL)
+            printf ("Using MMX for colorspace transform\n");
+        else
+            printf ("Cannot init MMX colorspace transform\n");
+    }
+#endif
+#ifdef HAVE_MLIB
+    if (yuv2rgb == NULL /*&& (config.flags & VO_MLIB_ENABLE)*/) {
+	yuv2rgb = yuv2rgb_init_mlib (bpp, mode);
+	if (yuv2rgb != NULL)
+	    printf ("Using mlib for colorspace transform\n");
+    }
+#endif
+    if (yuv2rgb == NULL) {
+	printf ("No accelerated colorspace conversion found\n");
+	yuv2rgb_c_init (bpp, mode);
+	yuv2rgb = (yuv2rgb_fun)yuv2rgb_c;
+    }
+}
+
+void * table_rV[256];
+void * table_gU[256];
+int table_gV[256];
+void * table_bU[256];
+
+#define RGB(i)					\
+	U = pu[i];				\
+	V = pv[i];				\
+	r = table_rV[V];			\
+	g = table_gU[U] + table_gV[V];		\
+	b = table_bU[U];
+
+#define DST1(i)					\
+	Y = py_1[2*i];				\
+	dst_1[2*i] = r[Y] + g[Y] + b[Y];	\
+	Y = py_1[2*i+1];			\
+	dst_1[2*i+1] = r[Y] + g[Y] + b[Y];
+
+#define DST2(i)					\
+	Y = py_2[2*i];				\
+	dst_2[2*i] = r[Y] + g[Y] + b[Y];	\
+	Y = py_2[2*i+1];			\
+	dst_2[2*i+1] = r[Y] + g[Y] + b[Y];
+
+#define DST1RGB(i)							\
+	Y = py_1[2*i];							\
+	dst_1[6*i] = r[Y]; dst_1[6*i+1] = g[Y]; dst_1[6*i+2] = b[Y];	\
+	Y = py_1[2*i+1];						\
+	dst_1[6*i+3] = r[Y]; dst_1[6*i+4] = g[Y]; dst_1[6*i+5] = b[Y];
+
+#define DST2RGB(i)							\
+	Y = py_2[2*i];							\
+	dst_2[6*i] = r[Y]; dst_2[6*i+1] = g[Y]; dst_2[6*i+2] = b[Y];	\
+	Y = py_2[2*i+1];						\
+	dst_2[6*i+3] = r[Y]; dst_2[6*i+4] = g[Y]; dst_2[6*i+5] = b[Y];
+
+#define DST1BGR(i)							\
+	Y = py_1[2*i];							\
+	dst_1[6*i] = b[Y]; dst_1[6*i+1] = g[Y]; dst_1[6*i+2] = r[Y];	\
+	Y = py_1[2*i+1];						\
+	dst_1[6*i+3] = b[Y]; dst_1[6*i+4] = g[Y]; dst_1[6*i+5] = r[Y];
+
+#define DST2BGR(i)							\
+	Y = py_2[2*i];							\
+	dst_2[6*i] = b[Y]; dst_2[6*i+1] = g[Y]; dst_2[6*i+2] = r[Y];	\
+	Y = py_2[2*i+1];						\
+	dst_2[6*i+3] = b[Y]; dst_2[6*i+4] = g[Y]; dst_2[6*i+5] = r[Y];
+
+static void yuv2rgb_c_32 (uint8_t * py_1, uint8_t * py_2,
+			  uint8_t * pu, uint8_t * pv,
+			  void * _dst_1, void * _dst_2, int h_size)
+{
+    int U, V, Y;
+    uint32_t * r, * g, * b;
+    uint32_t * dst_1, * dst_2;
+
+    h_size >>= 3;
+    dst_1 = _dst_1;
+    dst_2 = _dst_2;
+
+    while (h_size--) {
+	RGB(0);
+	DST1(0);
+	DST2(0);
+
+	RGB(1);
+	DST2(1);
+	DST1(1);
+
+	RGB(2);
+	DST1(2);
+	DST2(2);
+
+	RGB(3);
+	DST2(3);
+	DST1(3);
+
+	pu += 4;
+	pv += 4;
+	py_1 += 8;
+	py_2 += 8;
+	dst_1 += 8;
+	dst_2 += 8;
+    }
+}
+
+// This is very near from the yuv2rgb_c_32 code
+static void yuv2rgb_c_24_rgb (uint8_t * py_1, uint8_t * py_2,
+			      uint8_t * pu, uint8_t * pv,
+			      void * _dst_1, void * _dst_2, int h_size)
+{
+    int U, V, Y;
+    uint8_t * r, * g, * b;
+    uint8_t * dst_1, * dst_2;
+
+    h_size >>= 3;
+    dst_1 = _dst_1;
+    dst_2 = _dst_2;
+
+    while (h_size--) {
+	RGB(0);
+	DST1RGB(0);
+	DST2RGB(0);
+
+	RGB(1);
+	DST2RGB(1);
+	DST1RGB(1);
+
+	RGB(2);
+	DST1RGB(2);
+	DST2RGB(2);
+
+	RGB(3);
+	DST2RGB(3);
+	DST1RGB(3);
+
+	pu += 4;
+	pv += 4;
+	py_1 += 8;
+	py_2 += 8;
+	dst_1 += 24;
+	dst_2 += 24;
+    }
+}
+
+// only trivial mods from yuv2rgb_c_24_rgb
+static void yuv2rgb_c_24_bgr (uint8_t * py_1, uint8_t * py_2,
+			      uint8_t * pu, uint8_t * pv,
+			      void * _dst_1, void * _dst_2, int h_size)
+{
+    int U, V, Y;
+    uint8_t * r, * g, * b;
+    uint8_t * dst_1, * dst_2;
+
+    h_size >>= 3;
+    dst_1 = _dst_1;
+    dst_2 = _dst_2;
+
+    while (h_size--) {
+	RGB(0);
+	DST1BGR(0);
+	DST2BGR(0);
+
+	RGB(1);
+	DST2BGR(1);
+	DST1BGR(1);
+
+	RGB(2);
+	DST1BGR(2);
+	DST2BGR(2);
+
+	RGB(3);
+	DST2BGR(3);
+	DST1BGR(3);
+
+	pu += 4;
+	pv += 4;
+	py_1 += 8;
+	py_2 += 8;
+	dst_1 += 24;
+	dst_2 += 24;
+    }
+}
+
+// This is exactly the same code as yuv2rgb_c_32 except for the types of
+// r, g, b, dst_1, dst_2
+static void yuv2rgb_c_16 (uint8_t * py_1, uint8_t * py_2,
+			  uint8_t * pu, uint8_t * pv,
+			  void * _dst_1, void * _dst_2, int h_size)
+{
+    int U, V, Y;
+    uint16_t * r, * g, * b;
+    uint16_t * dst_1, * dst_2;
+
+    h_size >>= 3;
+    dst_1 = _dst_1;
+    dst_2 = _dst_2;
+
+    while (h_size--) {
+	RGB(0);
+	DST1(0);
+	DST2(0);
+
+	RGB(1);
+	DST2(1);
+	DST1(1);
+
+	RGB(2);
+	DST1(2);
+	DST2(2);
+
+	RGB(3);
+	DST2(3);
+	DST1(3);
+
+	pu += 4;
+	pv += 4;
+	py_1 += 8;
+	py_2 += 8;
+	dst_1 += 8;
+	dst_2 += 8;
+    }
+}
+
+static int div_round (int dividend, int divisor)
+{
+    if (dividend > 0)
+	return (dividend + (divisor>>1)) / divisor;
+    else
+	return -((-dividend + (divisor>>1)) / divisor);
+}
+
+static void yuv2rgb_c_init (int bpp, int mode)
+{  
+    int i;
+    uint8_t table_Y[1024];
+    uint32_t *table_32 = 0;
+    uint16_t *table_16 = 0;
+    uint8_t *table_8 = 0;
+    uint32_t entry_size = 0;
+    void *table_r = 0, *table_g = 0, *table_b = 0;
+
+    int crv = Inverse_Table_6_9[matrix_coefficients][0];
+    int cbu = Inverse_Table_6_9[matrix_coefficients][1];
+    int cgu = -Inverse_Table_6_9[matrix_coefficients][2];
+    int cgv = -Inverse_Table_6_9[matrix_coefficients][3];
+
+    for (i = 0; i < 1024; i++) {
+	int j;
+
+	j = (76309 * (i - 384 - 16) + 32768) >> 16;
+	j = (j < 0) ? 0 : ((j > 255) ? 255 : j);
+	table_Y[i] = j;
+    }
+
+    switch (bpp) {
+    case 32:
+	yuv2rgb_c_internal = yuv2rgb_c_32;
+
+	table_32 = malloc ((197 + 2*682 + 256 + 132) * sizeof (uint32_t));
+
+	entry_size = sizeof (uint32_t);
+	table_r = table_32 + 197;
+	table_b = table_32 + 197 + 685;
+	table_g = table_32 + 197 + 2*682;
+
+	for (i = -197; i < 256+197; i++)
+	    ((uint32_t *)table_r)[i] = table_Y[i+384] << ((mode==MODE_RGB) ? 16 : 0);
+	for (i = -132; i < 256+132; i++)
+	    ((uint32_t *)table_g)[i] = table_Y[i+384] << 8;
+	for (i = -232; i < 256+232; i++)
+	    ((uint32_t *)table_b)[i] = table_Y[i+384] << ((mode==MODE_RGB) ? 0 : 16);
+	break;
+
+    case 24:
+//	yuv2rgb_c_internal = (mode==MODE_RGB) ? yuv2rgb_c_24_rgb : yuv2rgb_c_24_bgr;
+	yuv2rgb_c_internal = (mode!=MODE_RGB) ? yuv2rgb_c_24_rgb : yuv2rgb_c_24_bgr;
+
+	table_8 = malloc ((256 + 2*232) * sizeof (uint8_t));
+
+	entry_size = sizeof (uint8_t);
+	table_r = table_g = table_b = table_8 + 232;
+
+	for (i = -232; i < 256+232; i++)
+	    ((uint8_t * )table_b)[i] = table_Y[i+384];
+	break;
+
+    case 15:
+    case 16:
+	yuv2rgb_c_internal = yuv2rgb_c_16;
+
+	table_16 = malloc ((197 + 2*682 + 256 + 132) * sizeof (uint16_t));
+
+	entry_size = sizeof (uint16_t);
+	table_r = table_16 + 197;
+	table_b = table_16 + 197 + 685;
+	table_g = table_16 + 197 + 2*682;
+
+	for (i = -197; i < 256+197; i++) {
+	    int j = table_Y[i+384] >> 3;
+
+	    if (mode == MODE_RGB)
+		j <<= ((bpp==16) ? 11 : 10);
+
+	    ((uint16_t *)table_r)[i] = j;
+	}
+	for (i = -132; i < 256+132; i++) {
+	    int j = table_Y[i+384] >> ((bpp==16) ? 2 : 3);
+
+	    ((uint16_t *)table_g)[i] = j << 5;
+	}
+	for (i = -232; i < 256+232; i++) {
+	    int j = table_Y[i+384] >> 3;
+
+	    if (mode == MODE_BGR)
+		j <<= ((bpp==16) ? 11 : 10);
+
+	    ((uint16_t *)table_b)[i] = j;
+	}
+	break;
+
+    default:
+	printf ("%ibpp not supported by yuv2rgb\n", bpp);
+	//exit (1);
+    }
+
+    for (i = 0; i < 256; i++) {
+	table_rV[i] = table_r + entry_size * div_round (crv * (i-128), 76309);
+	table_gU[i] = table_g + entry_size * div_round (cgu * (i-128), 76309);
+	table_gV[i] = entry_size * div_round (cgv * (i-128), 76309);
+	table_bU[i] = table_b + entry_size * div_round (cbu * (i-128), 76309);
+    }
+}
diff --git a/postproc/yuv2rgb_mlib.c b/postproc/yuv2rgb_mlib.c
new file mode 100644
index 0000000000..d8d3e0c4bc
--- /dev/null
+++ b/postproc/yuv2rgb_mlib.c
@@ -0,0 +1,84 @@
+/* 
+ * yuv2rgb_mlib.c, Software YUV to RGB coverter using mediaLib
+ *
+ *  Copyright (C) 2000, Håkan Hjort <d95hjort@dtek.chalmers.se>
+ *  All Rights Reserved.
+ *
+ *  This file is part of mpeg2dec, a free MPEG-2 video decoder
+ *
+ *  mpeg2dec 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, or (at your option)
+ *  any later version.
+ *
+ *  mpeg2dec 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 GNU Make; see the file COPYING.  If not, write to
+ *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "config.h"
+#include "video_out.h"
+#include "yuv2rgb.h"
+
+#include <mlib_types.h>
+#include <mlib_status.h>
+#include <mlib_sys.h>
+#include <mlib_video.h>
+
+static void mlib_YUV2ARGB420_32(uint8_t* image, uint8_t* py, 
+			 uint8_t* pu, uint8_t* pv, 
+			 int h_size, int v_size, 
+			 int rgb_stride, int y_stride, int uv_stride)
+{
+  mlib_VideoColorYUV2ARGB420(image, py, pu, pv, h_size,
+			     v_size, rgb_stride, y_stride, uv_stride);
+}
+
+static void mlib_YUV2ABGR420_32(uint8_t* image, uint8_t* py, 
+			 uint8_t* pu, uint8_t* pv, 
+			 int h_size, int v_size, 
+			 int rgb_stride, int y_stride, int uv_stride)
+{
+  mlib_VideoColorYUV2ABGR420(image, py, pu, pv, h_size,
+			     v_size, rgb_stride, y_stride, uv_stride);
+}
+
+static void mlib_YUV2RGB420_24(uint8_t* image, uint8_t* py, 
+			 uint8_t* pu, uint8_t* pv, 
+			 int h_size, int v_size, 
+			 int rgb_stride, int y_stride, int uv_stride)
+{
+  mlib_VideoColorYUV2RGB420(image, py, pu, pv, h_size,
+			    v_size, rgb_stride, y_stride, uv_stride);
+}
+
+
+yuv2rgb_fun yuv2rgb_init_mlib(int bpp, int mode) 
+{  
+
+	if( bpp == 24 ) 
+	{
+		if( mode == MODE_RGB )
+			return mlib_YUV2RGB420_24;
+  }
+
+	if( bpp == 32 ) 
+	{
+		if( mode == MODE_RGB )
+			return mlib_YUV2ARGB420_32;
+		else if( mode == MODE_BGR )
+			return mlib_YUV2ABGR420_32;
+	}
+  
+	return NULL;
+}
+
diff --git a/postproc/yuv2rgb_template.c b/postproc/yuv2rgb_template.c
new file mode 100644
index 0000000000..851182b974
--- /dev/null
+++ b/postproc/yuv2rgb_template.c
@@ -0,0 +1,403 @@
+
+/*
+ * yuv2rgb_mmx.c, Software YUV to RGB coverter with Intel MMX "technology"
+ *
+ * Copyright (C) 2000, Silicon Integrated System Corp.
+ * All Rights Reserved.
+ *
+ * Author: Olie Lho <ollie@sis.com.tw>
+ *
+ * This file is part of mpeg2dec, a free MPEG-2 video decoder
+ *
+ * mpeg2dec 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, or (at your option)
+ * any later version.
+ *
+ * mpeg2dec 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 GNU Make; see the file COPYING. If not, write to
+ * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "../config.h"
+
+//#include "libmpeg2/mpeg2.h"
+//#include "libmpeg2/mpeg2_internal.h"
+#include <inttypes.h>
+
+#include "rgb2rgb.h"
+
+/* hope these constant values are cache line aligned */
+uint64_t mmx_80w = 0x0080008000800080;
+uint64_t mmx_10w = 0x1010101010101010;
+uint64_t mmx_00ffw = 0x00ff00ff00ff00ff;
+uint64_t mmx_Y_coeff = 0x253f253f253f253f;
+
+/* hope these constant values are cache line aligned */
+uint64_t mmx_U_green = 0xf37df37df37df37d;
+uint64_t mmx_U_blue = 0x4093409340934093;
+uint64_t mmx_V_red = 0x3312331233123312;
+uint64_t mmx_V_green = 0xe5fce5fce5fce5fc;
+
+/* hope these constant values are cache line aligned */
+uint64_t mmx_redmask = 0xf8f8f8f8f8f8f8f8;
+uint64_t mmx_grnmask = 0xfcfcfcfcfcfcfcfc;
+uint64_t mmx_grnshift = 0x03;
+uint64_t mmx_blueshift = 0x03;
+
+#ifdef HAVE_MMX2
+/* use this for K7 and p3 only */
+#define MOVNTQ "movntq"
+#else
+/* for MMX-only processors */
+#define MOVNTQ "movq"
+#endif
+
+#if !defined( HAVE_MMX2) && defined( HAVE_3DNOW)
+/* for K6 2/2+/3 */
+#define EMMS "femms;"
+#else
+#define EMMS "emms;"
+#endif
+
+static void yuv420_rgb16_mmx (uint8_t * image, uint8_t * py,
+			      uint8_t * pu, uint8_t * pv,
+			      int h_size, int v_size,
+			      int rgb_stride, int y_stride, int uv_stride)
+{
+    int even = 1;
+    int x, y;
+
+    __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
+
+    for (y = v_size; --y >= 0; ) {
+	uint8_t *_image = image;
+	uint8_t *_py = py;
+	uint8_t *_pu = pu;
+	uint8_t *_pv = pv;
+
+	/* load data for start of next scan line */
+	__asm__ __volatile__ (
+		 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */ 
+		 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */ 
+		 "movq (%0), %%mm6;" /* Load 8  Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
+
+		 : : "r" (_py), "r" (_pu), "r" (_pv));
+
+	for (x = h_size >> 3; --x >= 0; ) {
+	    /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
+	       pixels in each iteration */
+
+	    __asm__ __volatile__ (
+		     /* Do the multiply part of the conversion for even and odd pixels,
+			register usage:
+			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
+			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
+			mm6 -> Y even, mm7 -> Y odd */
+		     /* convert the chroma part */
+		     "punpcklbw %%mm4, %%mm0;" /* scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 */ 
+		     "punpcklbw %%mm4, %%mm1;" /* scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 */ 
+
+		     "psubsw mmx_80w, %%mm0;" /* Cb -= 128 */ 
+		     "psubsw mmx_80w, %%mm1;" /* Cr -= 128 */ 
+
+		     "psllw $3, %%mm0;" /* Promote precision */ 
+		     "psllw $3, %%mm1;" /* Promote precision */ 
+
+		     "movq %%mm0, %%mm2;" /* Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 */ 
+		     "movq %%mm1, %%mm3;" /* Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 */ 
+
+		     "pmulhw mmx_U_green, %%mm2;" /* Mul Cb with green coeff -> Cb green */ 
+		     "pmulhw mmx_V_green, %%mm3;" /* Mul Cr with green coeff -> Cr green */ 
+
+		     "pmulhw mmx_U_blue, %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */ 
+		     "pmulhw mmx_V_red, %%mm1;" /* Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 */ 
+
+		     "paddsw %%mm3, %%mm2;" /* Cb green + Cr green -> Cgreen */ 
+
+		     /* convert the luma part */
+		     "psubusb mmx_10w, %%mm6;" /* Y -= 16 */ 
+
+		     "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
+		     "pand mmx_00ffw, %%mm6;" /* get Y even 00 Y6 00 Y4 00 Y2 00 Y0 */ 
+
+		     "psrlw $8, %%mm7;" /* get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 */ 
+
+		     "psllw $3, %%mm6;" /* Promote precision */ 
+		     "psllw $3, %%mm7;" /* Promote precision */ 
+
+		     "pmulhw mmx_Y_coeff, %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */ 
+		     "pmulhw mmx_Y_coeff, %%mm7;" /* Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 */ 
+
+		     /* Do the addition part of the conversion for even and odd pixels,
+			register usage:
+			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
+			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
+			mm6 -> Y even, mm7 -> Y odd */
+		     "movq %%mm0, %%mm3;" /* Copy Cblue */ 
+		     "movq %%mm1, %%mm4;" /* Copy Cred */ 
+		     "movq %%mm2, %%mm5;" /* Copy Cgreen */ 
+
+		     "paddsw %%mm6, %%mm0;" /* Y even + Cblue 00 B6 00 B4 00 B2 00 B0 */ 
+		     "paddsw %%mm7, %%mm3;" /* Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 */ 
+
+		     "paddsw %%mm6, %%mm1;" /* Y even + Cred 00 R6 00 R4 00 R2 00 R0 */ 
+		     "paddsw %%mm7, %%mm4;" /* Y odd + Cred 00 R7 00 R5 00 R3 00 R1 */ 
+
+		     "paddsw %%mm6, %%mm2;" /* Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 */ 
+		     "paddsw %%mm7, %%mm5;" /* Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 */ 
+
+		     /* Limit RGB even to 0..255 */
+		     "packuswb %%mm0, %%mm0;" /* B6 B4 B2 B0  B6 B4 B2 B0 */ 
+		     "packuswb %%mm1, %%mm1;" /* R6 R4 R2 R0  R6 R4 R2 R0 */ 
+		     "packuswb %%mm2, %%mm2;" /* G6 G4 G2 G0  G6 G4 G2 G0 */ 
+
+		     /* Limit RGB odd to 0..255 */
+		     "packuswb %%mm3, %%mm3;" /* B7 B5 B3 B1  B7 B5 B3 B1 */ 
+		     "packuswb %%mm4, %%mm4;" /* R7 R5 R3 R1  R7 R5 R3 R1 */ 
+		     "packuswb %%mm5, %%mm5;" /* G7 G5 G3 G1  G7 G5 G3 G1 */ 
+
+		     /* Interleave RGB even and odd */
+		     "punpcklbw %%mm3, %%mm0;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
+		     "punpcklbw %%mm4, %%mm1;" /* R7 R6 R5 R4 R3 R2 R1 R0 */ 
+		     "punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */ 
+
+		     /* mask unneeded bits off */
+		     "pand mmx_redmask, %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */ 
+		     "pand mmx_grnmask, %%mm2;" /* g7g6g5g4 g3g2_0_0 g7g6g5g4 g3g2_0_0 */ 
+		     "pand mmx_redmask, %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */ 
+
+		     "psrlw mmx_blueshift,%%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */ 
+		     "pxor %%mm4, %%mm4;" /* zero mm4 */ 
+
+		     "movq %%mm0, %%mm5;" /* Copy B7-B0 */ 
+		     "movq %%mm2, %%mm7;" /* Copy G7-G0 */ 
+
+		     /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
+		     "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */ 
+		     "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */ 
+
+		     "psllw mmx_blueshift,%%mm2;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */ 
+		     "por %%mm2, %%mm0;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */ 
+
+		     "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
+		     MOVNTQ " %%mm0, (%3);" /* store pixel 0-3 */ 
+
+		     /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
+		     "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */ 
+		     "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */ 
+
+		     "psllw mmx_blueshift,%%mm7;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */ 
+		     "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */ 
+
+		     "por %%mm7, %%mm5;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */ 
+		     "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */ 
+
+		     MOVNTQ " %%mm5, 8 (%3);" /* store pixel 4-7 */ 
+		     : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
+
+	    _py += 8;
+	    _pu += 4;
+	    _pv += 4;
+	    _image += 16;
+	}
+
+	if (!even) {
+	    pu += uv_stride;
+	    pv += uv_stride;
+	}
+
+	py += y_stride;
+	image += rgb_stride;
+
+	even = (!even);
+    }
+
+    __asm__ __volatile__ (EMMS);
+}
+
+static void yuv420_argb32_mmx (uint8_t * image, uint8_t * py,
+			       uint8_t * pu, uint8_t * pv,
+			       int h_size, int v_size,
+			       int rgb_stride, int y_stride, int uv_stride)
+{
+    int even = 1;
+    int x, y;
+
+    __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
+
+    for (y = v_size; --y >= 0; ) {
+	uint8_t *_image = image;
+	uint8_t *_py = py;
+	uint8_t *_pu = pu;
+	uint8_t *_pv = pv;
+
+	/* load data for start of next scan line */
+	__asm__ __volatile__ 
+	    (
+	     "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */ 
+	     "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */ 
+	     "movq (%0), %%mm6;" /* Load 8  Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
+	     : : "r" (_py), "r" (_pu), "r" (_pv)
+	     );
+
+	for (x = h_size >> 3; --x >= 0; ) {
+	    /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
+	       pixels in each iteration */
+	    __asm__ __volatile__ (
+		     /* Do the multiply part of the conversion for even and odd pixels,
+			register usage:
+			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
+			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
+			mm6 -> Y even, mm7 -> Y odd */
+
+		     /* convert the chroma part */
+		     "punpcklbw %%mm4, %%mm0;" /* scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 */ 
+		     "punpcklbw %%mm4, %%mm1;" /* scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 */ 
+
+		     "psubsw mmx_80w, %%mm0;" /* Cb -= 128 */ 
+		     "psubsw mmx_80w, %%mm1;" /* Cr -= 128 */ 
+
+		     "psllw $3, %%mm0;" /* Promote precision */ 
+		     "psllw $3, %%mm1;" /* Promote precision */ 
+
+		     "movq %%mm0, %%mm2;" /* Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 */ 
+		     "movq %%mm1, %%mm3;" /* Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 */ 
+
+		     "pmulhw mmx_U_green, %%mm2;" /* Mul Cb with green coeff -> Cb green */ 
+		     "pmulhw mmx_V_green, %%mm3;" /* Mul Cr with green coeff -> Cr green */ 
+
+		     "pmulhw mmx_U_blue, %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */ 
+		     "pmulhw mmx_V_red, %%mm1;" /* Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 */ 
+
+		     "paddsw %%mm3, %%mm2;" /* Cb green + Cr green -> Cgreen */ 
+
+		     /* convert the luma part */
+		     "psubusb mmx_10w, %%mm6;" /* Y -= 16 */ 
+
+		     "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
+		     "pand mmx_00ffw, %%mm6;" /* get Y even 00 Y6 00 Y4 00 Y2 00 Y0 */ 
+
+		     "psrlw $8, %%mm7;" /* get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 */ 
+
+		     "psllw $3, %%mm6;" /* Promote precision */ 
+		     "psllw $3, %%mm7;" /* Promote precision */ 
+
+		     "pmulhw mmx_Y_coeff, %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */ 
+		     "pmulhw mmx_Y_coeff, %%mm7;" /* Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 */ 
+
+		     /* Do the addition part of the conversion for even and odd pixels,
+			register usage:
+			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
+			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
+			mm6 -> Y even, mm7 -> Y odd */
+
+		     "movq %%mm0, %%mm3;" /* Copy Cblue */ 
+		     "movq %%mm1, %%mm4;" /* Copy Cred */ 
+		     "movq %%mm2, %%mm5;" /* Copy Cgreen */ 
+
+		     "paddsw %%mm6, %%mm0;" /* Y even + Cblue 00 B6 00 B4 00 B2 00 B0 */ 
+		     "paddsw %%mm7, %%mm3;" /* Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 */ 
+
+		     "paddsw %%mm6, %%mm1;" /* Y even + Cred 00 R6 00 R4 00 R2 00 R0 */ 
+		     "paddsw %%mm7, %%mm4;" /* Y odd + Cred 00 R7 00 R5 00 R3 00 R1 */ 
+
+		     "paddsw %%mm6, %%mm2;" /* Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 */ 
+		     "paddsw %%mm7, %%mm5;" /* Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 */ 
+
+		     /* Limit RGB even to 0..255 */
+		     "packuswb %%mm0, %%mm0;" /* B6 B4 B2 B0 B6 B4 B2 B0 */ 
+		     "packuswb %%mm1, %%mm1;" /* R6 R4 R2 R0 R6 R4 R2 R0 */ 
+		     "packuswb %%mm2, %%mm2;" /* G6 G4 G2 G0 G6 G4 G2 G0 */ 
+
+		     /* Limit RGB odd to 0..255 */
+		     "packuswb %%mm3, %%mm3;" /* B7 B5 B3 B1 B7 B5 B3 B1 */ 
+		     "packuswb %%mm4, %%mm4;" /* R7 R5 R3 R1 R7 R5 R3 R1 */ 
+		     "packuswb %%mm5, %%mm5;" /* G7 G5 G3 G1 G7 G5 G3 G1 */ 
+
+		     /* Interleave RGB even and odd */
+		     "punpcklbw %%mm3, %%mm0;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
+		     "punpcklbw %%mm4, %%mm1;" /* R7 R6 R5 R4 R3 R2 R1 R0 */ 
+		     "punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */ 
+
+		     /* convert RGB plane to RGB packed format, 
+			mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
+			mm4 -> GB, mm5 -> AR pixel 4-7,
+			mm6 -> GB, mm7 -> AR pixel 0-3 */
+		     "pxor %%mm3, %%mm3;" /* zero mm3 */ 
+
+		     "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
+		     "movq %%mm1, %%mm7;" /* R7 R6 R5 R4 R3 R2 R1 R0 */ 
+
+		     "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
+		     "movq %%mm1, %%mm5;" /* R7 R6 R5 R4 R3 R2 R1 R0 */ 
+
+		     "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */ 
+		     "punpcklbw %%mm3, %%mm7;" /* 00 R3 00 R2 00 R1 00 R0 */ 
+
+		     "punpcklwd %%mm7, %%mm6;" /* 00 R1 B1 G1 00 R0 B0 G0 */ 
+		     MOVNTQ " %%mm6, (%3);" /* Store ARGB1 ARGB0 */ 
+
+		     "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
+		     "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */ 
+
+		     "punpckhwd %%mm7, %%mm6;" /* 00 R3 G3 B3 00 R2 B3 G2 */ 
+		     MOVNTQ " %%mm6, 8 (%3);" /* Store ARGB3 ARGB2 */ 
+
+		     "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */ 
+		     "punpckhbw %%mm3, %%mm5;" /* 00 R7 00 R6 00 R5 00 R4 */ 
+				
+		     "punpcklwd %%mm5, %%mm4;" /* 00 R5 B5 G5 00 R4 B4 G4 */ 
+		     MOVNTQ " %%mm4, 16 (%3);" /* Store ARGB5 ARGB4 */ 
+
+		     "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
+		     "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */ 
+
+		     "punpckhwd %%mm5, %%mm4;" /* 00 R7 G7 B7 00 R6 B6 G6 */ 
+		     MOVNTQ " %%mm4, 24 (%3);" /* Store ARGB7 ARGB6 */ 
+
+		     "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */ 
+		     "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */ 
+
+		     "pxor %%mm4, %%mm4;" /* zero mm4 */ 
+		     "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
+
+		     : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
+
+	    _py += 8;
+	    _pu += 4;
+	    _pv += 4;
+	    _image += 32;
+	}
+
+	if (!even) {
+	    pu += uv_stride;
+	    pv += uv_stride;
+	}
+
+	py += y_stride;
+	image += rgb_stride;
+
+	even = (!even);
+    }
+
+    __asm__ __volatile__ (EMMS);
+}
+
+yuv2rgb_fun yuv2rgb_init_mmx (int bpp, int mode)
+{
+//    if (bpp == 15 || bpp == 16) {
+    if (bpp == 16 && mode == MODE_RGB) return yuv420_rgb16_mmx;
+    if (bpp == 32 && mode == MODE_RGB) return yuv420_argb32_mmx;
+    return NULL; // Fallback to C.
+}
+