- video_chroma: added support for IA-32/64 SSE2 accelaration (128 bit vector integer...
authorDamien Fouilleul <damienf@videolan.org>
Thu, 31 May 2007 09:56:25 +0000 (09:56 +0000)
committerDamien Fouilleul <damienf@videolan.org>
Thu, 31 May 2007 09:56:25 +0000 (09:56 +0000)
configure.ac
modules/video_chroma/Modules.am
modules/video_chroma/i420_yuy2.c
modules/video_chroma/i420_yuy2.h

index 471ff16..3d33cc4 100644 (file)
@@ -1269,22 +1269,14 @@ MMXEXT_MODULES="memcpymmxext"
 #MMXEXT_MODULES="${MMXEXT_MODULES} idctmmxext motionmmxext"
 THREEDNOW_MODULES="memcpy3dn"
 SSE_MODULES=""
+SSE2_MODULES=""
 ALTIVEC_MODULES="memcpyaltivec i420_yuy2_altivec"
 #ALTIVEC_MODULES="${ALTIVEC_MODULES} idctaltivec motionaltivec"
 
 if test "${enable_gprof}" != "yes" -a "${enable_debug}" != "yes"
 then
   MMX_MODULES="${MMX_MODULES} i420_yuy2_mmx"
-fi
-
-AC_CACHE_CHECK([if \$CC groks MMX inline assembly],
-    [ac_cv_mmx_inline],
-    [CFLAGS="${CFLAGS_save}"
-     AC_TRY_COMPILE(,[void *p;asm volatile("packuswb %%mm1,%%mm2"::"r"(p));],
-                    ac_cv_mmx_inline=yes, ac_cv_mmx_inline=no)])
-if test "${ac_cv_mmx_inline}" != "no"; then
-  AC_DEFINE(CAN_COMPILE_MMX, 1, Define if \$CC groks MMX inline assembly.)
-  ACCEL_MODULES="${ACCEL_MODULES} ${MMX_MODULES}"
+  SSE2_MODULES="${SSE2_MODULES} i420_yuy2_sse2"
 fi
 
 dnl  Check for fully workin MMX intrinsics
@@ -1312,6 +1304,41 @@ if test "${ac_cv_c_mmx_intrinsics}" != "no"; then
   VLC_ADD_CFLAGS([i420_rgb_mmx],[-mmmx])
 fi
 
+dnl  Check for fully workin SSE2 intrinsics
+dnl  We need support for -mmmx, we need <emmintrin.h>, and we also need a
+dnl  working compiler (http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23963)
+AC_CACHE_CHECK([if \$CC groks SSE2 intrinsics],
+    [ac_cv_c_sse2_intrinsics],
+    [CFLAGS="${CFLAGS_save} -O -msse2"
+     AC_TRY_COMPILE([#include <emmintrin.h>
+                     #include <stdint.h>
+                     uint64_t frobzor;],
+                    [__m128i a, b, c;
+                     a = b = c = _mm_set1_epi64((__m64)frobzor);
+                     a = _mm_slli_epi16(a, 3);
+                     a = _mm_adds_epi16(a, b);
+                     c = _mm_srli_epi16(c, 8);
+                     c = _mm_slli_epi16(c, 3);
+                     b = _mm_adds_epi16(b, c);
+                     a = _mm_unpacklo_epi8(a, b);
+                     frobzor = (uint64_t)_mm_movepi64_pi64(a);],
+                    [ac_cv_c_sse2_intrinsics=yes],
+                    [ac_cv_c_sse2_intrinsics=no])])
+if test "${ac_cv_c_sse2_intrinsics}" != "no"; then
+  AC_DEFINE(HAVE_SSE2_INTRINSICS, 1, Define if SSE2 intrinsics are available.)
+  dnl VLC_ADD_CFLAGS([i420_rgb_sse2],[-msse2])
+fi
+
+AC_CACHE_CHECK([if \$CC groks MMX inline assembly],
+    [ac_cv_mmx_inline],
+    [CFLAGS="${CFLAGS_save}"
+     AC_TRY_COMPILE(,[void *p;asm volatile("packuswb %%mm1,%%mm2"::"r"(p));],
+                    ac_cv_mmx_inline=yes, ac_cv_mmx_inline=no)])
+if test "${ac_cv_mmx_inline}" != "no"; then
+  AC_DEFINE(CAN_COMPILE_MMX, 1, Define if \$CC groks MMX inline assembly.)
+  ACCEL_MODULES="${ACCEL_MODULES} ${MMX_MODULES}"
+fi
+
 AC_CACHE_CHECK([if \$CC groks MMX EXT inline assembly],
     [ac_cv_mmxext_inline],
     [CFLAGS="${CFLAGS_save}"
@@ -1342,6 +1369,16 @@ if test "${ac_cv_sse_inline}" != "no" -a "${SYS}" != "solaris"; then
   ACCEL_MODULES="${ACCEL_MODULES} ${SSE_MODULES}"
 fi
 
+AC_CACHE_CHECK([if \$CC groks SSE2 inline assembly],
+    [ac_cv_sse2_inline],
+    [CFLAGS="${CFLAGS_save}"
+     AC_TRY_COMPILE(,[void *p;asm volatile("punpckhqdq %%xmm1,%%xmm2"::"r"(p));],
+                    ac_cv_sse2_inline=yes, ac_cv_sse2_inline=no)])
+if test "${ac_cv_sse2_inline}" != "no" -a "${SYS}" != "solaris"; then
+  AC_DEFINE(CAN_COMPILE_SSE2, 1, Define if \$CC groks SSE2 inline assembly.)
+  ACCEL_MODULES="${ACCEL_MODULES} ${SSE2_MODULES}"
+fi
+
 if test "${SYS}" != "mingw32" -a "${SYS}" != "mingwce"; then
 AC_CACHE_CHECK([if \$CC groks AltiVec inline assembly],
     [ac_cv_altivec_inline],
@@ -1494,6 +1531,11 @@ then
     ARCH="${ARCH} mmx"
     VLC_ADD_BUILTINS([${ACCEL_MODULES}])
 fi
+if test "${host_cpu}" = "i686" -o "${host_cpu}" = "x86_64"
+then
+    ARCH="${ARCH} sse sse2"
+    VLC_ADD_BUILTINS([${ACCEL_MODULES}])
+fi
 
 dnl
 dnl  Memory usage
index a8f4abe..ae77124 100644 (file)
@@ -23,6 +23,11 @@ SOURCES_i420_yuy2_mmx = \
        i420_yuy2.h \
        $(NULL)
 
+SOURCES_i420_yuy2_sse2 = \
+       i420_yuy2.c \
+       i420_yuy2.h \
+       $(NULL)
+
 SOURCES_i420_yuy2_altivec = \
        i420_yuy2.c \
        i420_yuy2.h \
index 535488d..7fee066 100644 (file)
@@ -5,6 +5,7 @@
  * $Id$
  *
  * Authors: Samuel Hocevar <sam@zoy.org>
+ *          Damien Fouilleul <damien@videolan.org>
  *
  * 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
@@ -42,6 +43,8 @@
 #    define DEST_FOURCC "YUY2,YUNV,YVYU,UYVY,UYNV,Y422,IUYV,cyuv,Y211"
 #elif defined (MODULE_NAME_IS_i420_yuy2_mmx)
 #    define DEST_FOURCC "YUY2,YUNV,YVYU,UYVY,UYNV,Y422,IUYV,cyuv"
+#elif defined (MODULE_NAME_IS_i420_yuy2_sse2)
+#    define DEST_FOURCC "YUY2,YUNV,YVYU,UYVY,UYNV,Y422,IUYV,cyuv"
 #elif defined (MODULE_NAME_IS_i420_yuy2_altivec)
 #    define DEST_FOURCC "YUY2,YUNV,YVYU,UYVY,UYNV,Y422"
 #endif
@@ -63,8 +66,9 @@ static void I420_Y211           ( vout_thread_t *, picture_t *, picture_t * );
 #endif
 
 #ifdef MODULE_NAME_IS_i420_yuy2_mmx
-static uint64_t i_00ffw;
-static uint64_t i_80w;
+/* Initialize MMX-specific constants */
+static const uint64_t i_00ffw = 0x00ff00ff00ff00ffULL;
+static const uint64_t i_80w   = 0x0000000080808080ULL;
 #endif
 
 /*****************************************************************************
@@ -78,9 +82,10 @@ vlc_module_begin();
     set_description( _("MMX conversions from " SRC_FOURCC " to " DEST_FOURCC) );
     set_capability( "chroma", 100 );
     add_requirement( MMX );
-    /* Initialize MMX-specific constants */
-    i_00ffw = 0x00ff00ff00ff00ffULL;
-    i_80w   = 0x0000000080808080ULL;
+#elif defined (MODULE_NAME_IS_i420_yuy2_sse2)
+    set_description( _("SSE2 conversions from " SRC_FOURCC " to " DEST_FOURCC) );
+    set_capability( "chroma", 120 );
+    add_requirement( SSE2 );
 #elif defined (MODULE_NAME_IS_i420_yuy2_altivec)
     set_description(
             _("AltiVec conversions from " SRC_FOURCC " to " DEST_FOURCC) );
@@ -125,7 +130,6 @@ static int Activate( vlc_object_t *p_this )
                 case VLC_FOURCC('Y','4','2','2'):
                     p_vout->chroma.pf_convert = I420_UYVY;
                     break;
-
 #if !defined (MODULE_NAME_IS_i420_yuy2_altivec)
                 case VLC_FOURCC('I','U','Y','V'):
                     p_vout->chroma.pf_convert = I420_IUYV;
@@ -256,6 +260,7 @@ static void I420_YUY2( vout_thread_t *p_vout, picture_t *p_source,
     const int i_dest_margin = p_dest->p->i_pitch
                                - p_dest->p->i_visible_pitch;
 
+#if !defined(MODULE_NAME_IS_i420_yuy2_sse2)
     for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
     {
         p_line1 = p_line2;
@@ -265,20 +270,23 @@ static void I420_YUY2( vout_thread_t *p_vout, picture_t *p_source,
         p_y2 += p_source->p[Y_PLANE].i_pitch;
 
 #if !defined (MODULE_NAME_IS_i420_yuy2_mmx)
-        for( i_x = p_vout->render.i_width / ; i_x-- ; )
+        for( i_x = p_vout->render.i_width / 8; i_x-- ; )
         {
             C_YUV420_YUYV( );
+            C_YUV420_YUYV( );
+            C_YUV420_YUYV( );
+            C_YUV420_YUYV( );
         }
 #else
         for( i_x = p_vout->render.i_width / 8 ; i_x-- ; )
         {
             MMX_CALL( MMX_YUV420_YUYV );
         }
+#endif
         for( i_x = ( p_vout->render.i_width % 8 ) / 2; i_x-- ; )
         {
             C_YUV420_YUYV( );
         }
-#endif
 
         p_y1 += i_source_margin;
         p_y2 += i_source_margin;
@@ -288,9 +296,77 @@ static void I420_YUY2( vout_thread_t *p_vout, picture_t *p_source,
         p_line2 += i_dest_margin;
     }
 
+#if defined (MODULE_NAME_IS_i420_yuy2_mmx)
+    __asm__ __volatile__("emms" :: );
+#endif
+
 #if defined (MODULE_NAME_IS_i420_yuy2_altivec)
     }
 #endif
+
+#else // defined(MODULE_NAME_IS_i420_yuy2_sse2)
+    /*
+    ** SSE2 128 bytes fetch/store instructions are faster 
+    ** if memory access is 16 bytes aligned
+    */
+    if( 0 == (15 & (p_source->p[Y_PLANE].i_pitch|p_dest->p->i_pitch|
+        ((int)p_line2|(int)p_y2))) )
+    {
+        /* use faster SSE2 aligned fetch and store */
+        for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
+        {
+            p_line1 = p_line2;
+            p_line2 += p_dest->p->i_pitch;
+
+            p_y1 = p_y2;
+            p_y2 += p_source->p[Y_PLANE].i_pitch;
+
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV420_YUYV_ALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV420_YUYV( );
+            }
+
+            p_y1 += i_source_margin;
+            p_y2 += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line1 += i_dest_margin;
+            p_line2 += i_dest_margin;
+        }
+    }
+    else
+    {
+        /* use slower SSE2 unaligned fetch and store */
+        for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
+        {
+            p_line1 = p_line2;
+            p_line2 += p_dest->p->i_pitch;
+
+            p_y1 = p_y2;
+            p_y2 += p_source->p[Y_PLANE].i_pitch;
+
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV420_YUYV_UNALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV420_YUYV( );
+            }
+
+            p_y1 += i_source_margin;
+            p_y2 += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line1 += i_dest_margin;
+            p_line2 += i_dest_margin;
+        }
+    }
+#endif // defined(MODULE_NAME_IS_i420_yuy2_sse2)
 }
 
 /*****************************************************************************
@@ -393,6 +469,7 @@ static void I420_YVYU( vout_thread_t *p_vout, picture_t *p_source,
     const int i_dest_margin = p_dest->p->i_pitch
                                - p_dest->p->i_visible_pitch;
 
+#if !defined(MODULE_NAME_IS_i420_yuy2_sse2)
     for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
     {
         p_line1 = p_line2;
@@ -420,9 +497,78 @@ static void I420_YVYU( vout_thread_t *p_vout, picture_t *p_source,
         p_line1 += i_dest_margin;
         p_line2 += i_dest_margin;
     }
+
+#if defined (MODULE_NAME_IS_i420_yuy2_mmx)
+    __asm__ __volatile__("emms" :: );
+#endif
+
 #if defined (MODULE_NAME_IS_i420_yuy2_altivec)
     }
 #endif
+
+#else // defined(MODULE_NAME_IS_i420_yuy2_sse2)
+    /*
+    ** SSE2 128 bytes fetch/store instructions are faster 
+    ** if memory access is 16 bytes aligned
+    */
+    if( 0 == (15 & (p_source->p[Y_PLANE].i_pitch|p_dest->p->i_pitch|
+        ((int)p_line2|(int)p_y2))) )
+    {
+        /* use faster SSE2 aligned fetch and store */
+        for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
+        {
+            p_line1 = p_line2;
+            p_line2 += p_dest->p->i_pitch;
+
+            p_y1 = p_y2;
+            p_y2 += p_source->p[Y_PLANE].i_pitch;
+
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV420_YVYU_ALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV420_YVYU( );
+            }
+
+            p_y1 += i_source_margin;
+            p_y2 += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line1 += i_dest_margin;
+            p_line2 += i_dest_margin;
+        }
+    }
+    else
+    {
+        /* use slower SSE2 unaligned fetch and store */
+        for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
+        {
+            p_line1 = p_line2;
+            p_line2 += p_dest->p->i_pitch;
+
+            p_y1 = p_y2;
+            p_y2 += p_source->p[Y_PLANE].i_pitch;
+
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV420_YVYU_UNALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV420_YVYU( );
+            }
+
+            p_y1 += i_source_margin;
+            p_y2 += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line1 += i_dest_margin;
+            p_line2 += i_dest_margin;
+        }
+    }
+#endif // defined(MODULE_NAME_IS_i420_yuy2_sse2)
 }
 
 /*****************************************************************************
@@ -525,6 +671,7 @@ static void I420_UYVY( vout_thread_t *p_vout, picture_t *p_source,
     const int i_dest_margin = p_dest->p->i_pitch
                                - p_dest->p->i_visible_pitch;
 
+#if !defined(MODULE_NAME_IS_i420_yuy2_sse2)
     for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
     {
         p_line1 = p_line2;
@@ -564,6 +711,70 @@ static void I420_UYVY( vout_thread_t *p_vout, picture_t *p_source,
 #if defined (MODULE_NAME_IS_i420_yuy2_altivec)
     }
 #endif
+
+#else // defined(MODULE_NAME_IS_i420_yuy2_sse2)
+    /*
+    ** SSE2 128 bytes fetch/store instructions are faster 
+    ** if memory access is 16 bytes aligned
+    */
+    if( 0 == (15 & (p_source->p[Y_PLANE].i_pitch|p_dest->p->i_pitch|
+        ((int)p_line2|(int)p_y2))) )
+    {
+        /* use faster SSE2 aligned fetch and store */
+        for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
+        {
+            p_line1 = p_line2;
+            p_line2 += p_dest->p->i_pitch;
+
+            p_y1 = p_y2;
+            p_y2 += p_source->p[Y_PLANE].i_pitch;
+
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV420_UYVY_ALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV420_UYVY( );
+            }
+
+            p_y1 += i_source_margin;
+            p_y2 += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line1 += i_dest_margin;
+            p_line2 += i_dest_margin;
+        }
+    }
+    else
+    {
+        /* use slower SSE2 unaligned fetch and store */
+        for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
+        {
+            p_line1 = p_line2;
+            p_line2 += p_dest->p->i_pitch;
+
+            p_y1 = p_y2;
+            p_y2 += p_source->p[Y_PLANE].i_pitch;
+
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV420_UYVY_UNALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV420_UYVY( );
+            }
+
+            p_y1 += i_source_margin;
+            p_y2 += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line1 += i_dest_margin;
+            p_line2 += i_dest_margin;
+        }
+    }
+#endif // defined(MODULE_NAME_IS_i420_yuy2_sse2)
 }
 
 #if !defined (MODULE_NAME_IS_i420_yuy2_altivec)
@@ -601,6 +812,7 @@ static void I420_cyuv( vout_thread_t *p_vout, picture_t *p_source,
     const int i_dest_margin = p_dest->p->i_pitch
                                - p_dest->p->i_visible_pitch;
 
+#if !defined(MODULE_NAME_IS_i420_yuy2_sse2)
     for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
     {
         p_line1 -= 3 * p_dest->p->i_pitch;
@@ -611,7 +823,7 @@ static void I420_cyuv( vout_thread_t *p_vout, picture_t *p_source,
 
         for( i_x = p_vout->render.i_width / 8 ; i_x-- ; )
         {
-#if defined (MODULE_NAME_IS_i420_yuy2)
+#if !defined (MODULE_NAME_IS_i420_yuy2_mmx)
             C_YUV420_UYVY( );
             C_YUV420_UYVY( );
             C_YUV420_UYVY( );
@@ -628,6 +840,74 @@ static void I420_cyuv( vout_thread_t *p_vout, picture_t *p_source,
         p_line1 += i_dest_margin;
         p_line2 += i_dest_margin;
     }
+
+#if defined (MODULE_NAME_IS_i420_yuy2_mmx)
+    __asm__ __volatile__("emms" :: );
+#endif
+
+#else // defined(MODULE_NAME_IS_i420_yuy2_sse2)
+    /*
+    ** SSE2 128 bytes fetch/store instructions are faster 
+    ** if memory access is 16 bytes aligned
+    */
+    if( 0 == (15 & (p_source->p[Y_PLANE].i_pitch|p_dest->p->i_pitch|
+        ((int)p_line2|(int)p_y2))) )
+    {
+        /* use faster SSE2 aligned fetch and store */
+        for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
+        {
+            p_line1 = p_line2;
+            p_line2 += p_dest->p->i_pitch;
+
+            p_y1 = p_y2;
+            p_y2 += p_source->p[Y_PLANE].i_pitch;
+
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV420_UYVY_ALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV420_UYVY( );
+            }
+
+            p_y1 += i_source_margin;
+            p_y2 += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line1 += i_dest_margin;
+            p_line2 += i_dest_margin;
+        }
+    }
+    else
+    {
+        /* use slower SSE2 unaligned fetch and store */
+        for( i_y = p_vout->render.i_height / 2 ; i_y-- ; )
+        {
+            p_line1 = p_line2;
+            p_line2 += p_dest->p->i_pitch;
+
+            p_y1 = p_y2;
+            p_y2 += p_source->p[Y_PLANE].i_pitch;
+
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV420_UYVY_UNALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV420_UYVY( );
+            }
+
+            p_y1 += i_source_margin;
+            p_y2 += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line1 += i_dest_margin;
+            p_line2 += i_dest_margin;
+        }
+    }
+#endif // defined(MODULE_NAME_IS_i420_yuy2_sse2)
 }
 #endif // !defined (MODULE_NAME_IS_i420_yuy2_altivec)
 
@@ -675,4 +955,3 @@ static void I420_Y211( vout_thread_t *p_vout, picture_t *p_source,
     }
 }
 #endif
-
index c5bcf9b..19ea8b0 100644 (file)
@@ -5,6 +5,7 @@
  * $Id$
  *
  * Authors: Samuel Hocevar <sam@zoy.org>
+ *          Damien Fouilleul <damien@videolan.org>
  *
  * 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
@@ -32,7 +33,7 @@
         : "r" (p_line1),  "r" (p_line2),  "r" (p_y1),  "r" (p_y2),          \
           "r" (p_u), "r" (p_v) );                                           \
     p_line1 += 16; p_line2 += 16; p_y1 += 8; p_y2 += 8; p_u += 4; p_v += 4; \
-    } while(0);                                                             \
+    } while(0)
 
 #define MMX_YUV420_YUYV "                                                 \n\
 movq       (%2), %%mm0  # Load 8 Y            y7 y6 y5 y4 y3 y2 y1 y0     \n\
@@ -111,7 +112,135 @@ punpcklbw %%mm2, %%mm1  #                     v2 Y6 u2 Y4 v0 Y2 u0 Y0     \n\
 movq      %%mm1, (%1)   # Store YUYV                                      \n\
 "
 
-#else
+#elif defined( MODULE_NAME_IS_i420_yuy2_sse2 )
+
+/* SSE2 */
+
+#define SSE2_CALL(SSE2_INSTRUCTIONS)   \
+    do {                                                                    \
+    __asm__ __volatile__(                                                   \
+        ".p2align 3 \n\t"                                                   \
+        SSE2_INSTRUCTIONS                                                   \
+        :                                                                   \
+        : "r" (p_line1),  "r" (p_line2),  "r" (p_y1),  "r" (p_y2),          \
+          "r" (p_u), "r" (p_v) );                                           \
+        p_line1 += 32; p_line2 += 32; p_y1 += 16; p_y2 += 16;               \
+        p_u += 8; p_v += 8;                                                 \
+    } while(0)
+
+#define SSE2_YUV420_YUYV_ALIGNED "                                        \n\
+movdqa      (%2), %%xmm0  # Load 16 Y         y15 y14 y13 .. y2 y1 y0     \n\
+movq        (%4), %%xmm1  # Load 8 Cb         u7 u6 u5 u4 u3 u2 u1 u0     \n\
+movq        (%5), %%xmm2  # Load 8 Cr         v7 06 v5 v4 v3 v2 v1 v0     \n\
+punpcklbw %%xmm2, %%xmm1  #                   v7 u7 v6 u6 .. u1 v0 u0     \n\
+movdqa    %%xmm0, %%xmm2  #                   y15 y14 y13 .. y2 y1 y0     \n\
+punpcklbw %%xmm1, %%xmm2  #                   v3 y7 u3 .. v0 y1 u0 y0     \n\
+movdqa    %%xmm2, (%0)    # Store low YUYV                                \n\
+punpckhbw %%xmm1, %%xmm0  #                   v3 y7 u3 y6 v2 y5 u2 y4     \n\
+movdqa    %%xmm0, 16(%0)  # Store high YUYV                               \n\
+movdqa      (%3), %%xmm0  # Load 8 Y          Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+movdqa    %%xmm0, %%xmm2  #                   Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+punpcklbw %%xmm1, %%xmm2  #                   v1 Y3 u1 Y2 v0 Y1 u0 Y0     \n\
+movdqa    %%xmm2, (%1)    # Store low YUYV                                \n\
+punpckhbw %%xmm1, %%xmm0  #                   v3 Y7 u3 Y6 v2 Y5 u2 Y4     \n\
+movdqa    %%xmm0, 16(%1)  # Store high YUYV                               \n\
+"
+
+#define SSE2_YUV420_YUYV_UNALIGNED "                                      \n\
+movdqu      (%2), %%xmm0  # Load 16 Y         y7 y6 y5 y4 y3 y2 y1 y0     \n\
+movq        (%4), %%xmm1  # Load 8 Cb         00 00 00 00 u3 u2 u1 u0     \n\
+movq        (%5), %%xmm2  # Load 8 Cr         00 00 00 00 v3 v2 v1 v0     \n\
+punpcklbw %%xmm2, %%xmm1  #                   v3 u3 v2 u2 v1 u1 v0 u0     \n\
+movdqa    %%xmm0, %%xmm2  #                   y7 y6 y5 y4 y3 y2 y1 y0     \n\
+punpcklbw %%xmm1, %%xmm2  #                   v1 y3 u1 y2 v0 y1 u0 y0     \n\
+movdqu    %%xmm2, (%0)    # Store low YUYV                                \n\
+punpckhbw %%xmm1, %%xmm0  #                   v3 y7 u3 y6 v2 y5 u2 y4     \n\
+movdqu    %%xmm0, 16(%0)  # Store high YUYV                               \n\
+movdqu      (%3), %%xmm0  # Load 16 Y         Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+movdqa    %%xmm0, %%xmm2  #                   Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+punpcklbw %%xmm1, %%xmm2  #                   v1 Y3 u1 Y2 v0 Y1 u0 Y0     \n\
+movdqu    %%xmm2, (%1)    # Store low YUYV                                \n\
+punpckhbw %%xmm1, %%xmm0  #                   v3 Y7 u3 Y6 v2 Y5 u2 Y4     \n\
+movdqu    %%xmm0, 16(%1)  # Store high YUYV                               \n\
+"
+
+#define SSE2_YUV420_YVYU_ALIGNED "                                          \n\
+movdqa      (%2), %%xmm0  # Load 16 Y           y7 y6 y5 y4 y3 y2 y1 y0     \n\
+movq        (%4), %%xmm2  # Load 8 Cb           00 00 00 00 u3 u2 u1 u0     \n\
+movq        (%5), %%xmm1  # Load 8 Cr           00 00 00 00 v3 v2 v1 v0     \n\
+punpcklbw %%xmm2, %%xmm1  #                     u3 v3 u2 v2 u1 v1 u0 v0     \n\
+movdqa    %%xmm0, %%xmm2  #                     y7 y6 y5 y4 y3 y2 y1 y0     \n\
+punpcklbw %%xmm1, %%xmm2  #                     u1 y3 v1 y2 u0 y1 v0 y0     \n\
+movdqa    %%xmm2, (%0)    # Store low YUYV                                  \n\
+punpckhbw %%xmm1, %%xmm0  #                     u3 y7 v3 y6 u2 y5 v2 y4     \n\
+movdqa    %%xmm0, 16(%0)  # Store high YUYV                                 \n\
+movdqa      (%3), %%xmm0  # Load 16 Y           Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+movdqa    %%xmm0, %%xmm2  #                     Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+punpcklbw %%xmm1, %%xmm2  #                     u1 Y3 v1 Y2 u0 Y1 v0 Y0     \n\
+movdqa    %%xmm2, (%1)    # Store low YUYV                                  \n\
+punpckhbw %%xmm1, %%xmm0  #                     u3 Y7 v3 Y6 u2 Y5 v2 Y4     \n\
+movdqa    %%xmm0, 16(%1)  # Store high YUYV                                 \n\
+"
+
+#define SSE2_YUV420_YVYU_UNALIGNED "                                        \n\
+movdqu      (%2), %%xmm0  # Load 16 Y           y7 y6 y5 y4 y3 y2 y1 y0     \n\
+movq        (%4), %%xmm2  # Load 8 Cb           00 00 00 00 u3 u2 u1 u0     \n\
+movq        (%5), %%xmm1  # Load 8 Cr           00 00 00 00 v3 v2 v1 v0     \n\
+punpcklbw %%xmm2, %%xmm1  #                     u3 v3 u2 v2 u1 v1 u0 v0     \n\
+movdqu    %%xmm0, %%xmm2  #                     y7 y6 y5 y4 y3 y2 y1 y0     \n\
+punpcklbw %%xmm1, %%xmm2  #                     u1 y3 v1 y2 u0 y1 v0 y0     \n\
+movdqu    %%xmm2, (%0)    # Store low YUYV                                  \n\
+punpckhbw %%xmm1, %%xmm0  #                     u3 y7 v3 y6 u2 y5 v2 y4     \n\
+movdqu    %%xmm0, 16(%0)  # Store high YUYV                                 \n\
+movdqu      (%3), %%xmm0  # Load 16 Y           Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+movdqu    %%xmm0, %%xmm2  #                     Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+punpcklbw %%xmm1, %%xmm2  #                     u1 Y3 v1 Y2 u0 Y1 v0 Y0     \n\
+movdqu    %%xmm2, (%1)    # Store low YUYV                                  \n\
+punpckhbw %%xmm1, %%xmm0  #                     u3 Y7 v3 Y6 u2 Y5 v2 Y4     \n\
+movdqu    %%xmm0, 16(%1)  # Store high YUYV                                 \n\
+"
+
+#define SSE2_YUV420_UYVY_ALIGNED "                                          \n\
+movdqa      (%2), %%xmm0  # Load 16 Y           y7 y6 y5 y4 y3 y2 y1 y0     \n\
+movdqa      (%3), %%xmm3  # Load 16 Y           Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+movq        (%4), %%xmm1  # Load 8 Cb           00 00 00 00 u3 u2 u1 u0     \n\
+movq        (%5), %%xmm2  # Load 8 Cr           00 00 00 00 v3 v2 v1 v0     \n\
+punpcklbw %%xmm2, %%xmm1  #                     v3 u3 v2 u2 v1 u1 v0 u0     \n\
+movdqa    %%xmm1, %%xmm2  #                     v3 u3 v2 u2 v1 u1 v0 u0     \n\
+punpcklbw %%xmm0, %%xmm2  #                     y3 v1 y2 u1 y1 v0 y0 u0     \n\
+movdqa    %%xmm2, (%0)    # Store low UYVY                                  \n\
+movdqa    %%xmm1, %%xmm2  #                     u3 v3 u2 v2 u1 v1 u0 v0     \n\
+punpckhbw %%xmm0, %%xmm2  #                     y3 v1 y2 u1 y1 v0 y0 u0     \n\
+movdqa    %%xmm2, 16(%0)  # Store high UYVY                                 \n\
+movdqa    %%xmm1, %%xmm2  #                     u3 v3 u2 v2 u1 v1 u0 v0     \n\
+punpcklbw %%xmm3, %%xmm2  #                     Y3 v1 Y2 u1 Y1 v0 Y0 u0     \n\
+movdqa    %%xmm2, (%1)    # Store low UYVY                                  \n\
+punpckhbw %%xmm3, %%xmm1  #                     Y7 v3 Y6 u3 Y5 v2 Y4 u2     \n\
+movdqa    %%xmm1, 16(%1)  # Store high UYVY                                 \n\
+"
+
+#define SSE2_YUV420_UYVY_UNALIGNED "                                        \n\
+movdqu      (%2), %%xmm0  # Load 16 Y           y7 y6 y5 y4 y3 y2 y1 y0     \n\
+movdqu      (%3), %%xmm3  # Load 16 Y           Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0     \n\
+movq        (%4), %%xmm1  # Load 8 Cb           00 00 00 00 u3 u2 u1 u0     \n\
+movq        (%5), %%xmm2  # Load 8 Cr           00 00 00 00 v3 v2 v1 v0     \n\
+punpcklbw %%xmm2, %%xmm1  #                     v3 u3 v2 u2 v1 u1 v0 u0     \n\
+movdqu    %%xmm1, %%xmm2  #                     v3 u3 v2 u2 v1 u1 v0 u0     \n\
+punpcklbw %%xmm0, %%xmm2  #                     y3 v1 y2 u1 y1 v0 y0 u0     \n\
+movdqu    %%xmm2, (%0)    # Store low UYVY                                  \n\
+movdqu    %%xmm1, %%xmm2  #                     u3 v3 u2 v2 u1 v1 u0 v0     \n\
+punpckhbw %%xmm0, %%xmm2  #                     y3 v1 y2 u1 y1 v0 y0 u0     \n\
+movdqu    %%xmm2, 16(%0)  # Store high UYVY                                 \n\
+movdqu    %%xmm1, %%xmm2  #                     u3 v3 u2 v2 u1 v1 u0 v0     \n\
+punpcklbw %%xmm3, %%xmm2  #                     Y3 v1 Y2 u1 Y1 v0 Y0 u0     \n\
+movdqu    %%xmm2, (%1)    # Store low UYVY                                  \n\
+punpckhbw %%xmm3, %%xmm1  #                     Y7 v3 Y6 u3 Y5 v2 Y4 u2     \n\
+movdqu    %%xmm1, 16(%1)  # Store high UYVY                                 \n\
+"
+
+#endif
+
+/* Used in both accelerated and C modules */
 
 #define C_YUV420_YVYU( )                                                    \
     *(p_line1)++ = *(p_y1)++; *(p_line2)++ = *(p_y2)++;                     \
@@ -127,9 +256,7 @@ movq      %%mm1, (%1)   # Store YUYV                                      \n\
     *(p_line2)++ = *(p_y2); p_y2 += 2;                                      \
     *(p_line1)++ = *(p_line2)++ = *(p_v) - 0x80; p_v += 2;                  \
 
-#endif
 
-/* Used in both MMX and C modules */
 #define C_YUV420_YUYV( )                                                    \
     *(p_line1)++ = *(p_y1)++; *(p_line2)++ = *(p_y2)++;                     \
     *(p_line1)++ =            *(p_line2)++ = *(p_u)++;                      \