i422_yuy2: SSE2 improvements

[vlc] / modules / video_chroma / i422_yuy2.c

diff --git a/modules/video_chroma/i422_yuy2.c b/modules/video_chroma/i422_yuy2.c
index 4bc452891280c3b48f2638d6f256900da2d75745..2119755a2899d2835fec05889336e84459b6acc1 100644 (file)
--- a/modules/video_chroma/i422_yuy2.c
+++ b/modules/video_chroma/i422_yuy2.c
@@ -67,6 +67,10 @@ vlc_module_begin();
     set_description( _("MMX conversions from " SRC_FOURCC " to " DEST_FOURCC) );
     set_capability( "chroma", 100 );
     add_requirement( MMX );
+#elif defined (MODULE_NAME_IS_i422_yuy2_sse2)
+    set_description( _("SSE2 conversions from " SRC_FOURCC " to " DEST_FOURCC) );
+    set_capability( "chroma", 120 );
+    add_requirement( MMX );
 #endif
     set_callbacks( Activate, NULL );
 vlc_module_end();
@@ -143,17 +147,66 @@ static int Activate( vlc_object_t *p_this )
 static void I422_YUY2( vout_thread_t *p_vout, picture_t *p_source,
                                               picture_t *p_dest )
 {
-    uint8_t *p_pixels = p_dest->p->p_pixels;
-    int      i_pitch  = p_dest->p->i_pitch;
+    uint8_t *p_line = p_dest->p->p_pixels;
     uint8_t *p_y = p_source->Y_PIXELS;
     uint8_t *p_u = p_source->U_PIXELS;
     uint8_t *p_v = p_source->V_PIXELS;
 
     int i_x, i_y;
 
+    const int i_source_margin = p_source->p[0].i_pitch
+                                 - p_source->p[0].i_visible_pitch;
+    const int i_source_margin_c = p_source->p[1].i_pitch
+                                 - p_source->p[1].i_visible_pitch;
+    const int i_dest_margin = p_dest->p->i_pitch
+                               - p_dest->p->i_visible_pitch;
+
+#if defined (MODULE_NAME_IS_i422_yuy2_sse2)
+
+    if( 0 == (15 & (p_source->p[Y_PLANE].i_pitch|p_dest->p->i_pitch|
+        ((int)p_line|(int)p_y))) )
+    {
+        /* use faster SSE2 aligned fetch and store */
+        for( i_y = p_vout->render.i_height ; i_y-- ; )
+        {
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV422_YUYV_ALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV422_YUYV( p_line, p_y, p_u, p_v );
+            }
+            p_y += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line += i_dest_margin;
+        }
+    }
+    else {
+        /* use slower SSE2 unaligned fetch and store */
+        for( i_y = p_vout->render.i_height ; i_y-- ; )
+        {
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV422_YUYV_UNALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV422_YUYV( p_line, p_y, p_u, p_v );
+            }
+            p_y += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line += i_dest_margin;
+        }
+    }
+    SSE2_END;
+
+#else
+
     for( i_y = p_vout->render.i_height ; i_y-- ; )
     {
-        uint8_t *p_line = p_pixels;
         for( i_x = p_vout->render.i_width / 8 ; i_x-- ; )
         {
 #if defined (MODULE_NAME_IS_i422_yuy2)
@@ -165,12 +218,19 @@ static void I422_YUY2( vout_thread_t *p_vout, picture_t *p_source,
             MMX_CALL( MMX_YUV422_YUYV );
 #endif
         }
-        p_pixels += i_pitch;
+        for( i_x = ( p_vout->render.i_width % 8 ) / 2; i_x-- ; )
+        {
+            C_YUV422_YUYV( p_line, p_y, p_u, p_v );
+        }
+        p_y += i_source_margin;
+        p_u += i_source_margin_c;
+        p_v += i_source_margin_c;
+        p_line += i_dest_margin;
     }
 #if defined (MODULE_NAME_IS_i422_yuy2_mmx)
     MMX_END;
-#elif defined (MODULE_NAME_IS_i422_yuy2_sse2)
-    SSE2_END;
+#endif
+
 #endif
 }
 
@@ -180,17 +240,66 @@ static void I422_YUY2( vout_thread_t *p_vout, picture_t *p_source,
 static void I422_YVYU( vout_thread_t *p_vout, picture_t *p_source,
                                               picture_t *p_dest )
 {
-    uint8_t *p_pixels = p_dest->p->p_pixels;
-    int      i_pitch  = p_dest->p->i_pitch;
+    uint8_t *p_line = p_dest->p->p_pixels;
     uint8_t *p_y = p_source->Y_PIXELS;
     uint8_t *p_u = p_source->U_PIXELS;
     uint8_t *p_v = p_source->V_PIXELS;
 
     int i_x, i_y;
 
+    const int i_source_margin = p_source->p[0].i_pitch
+                                 - p_source->p[0].i_visible_pitch;
+    const int i_source_margin_c = p_source->p[1].i_pitch
+                                 - p_source->p[1].i_visible_pitch;
+    const int i_dest_margin = p_dest->p->i_pitch
+                               - p_dest->p->i_visible_pitch;
+
+#if defined (MODULE_NAME_IS_i422_yuy2_sse2)
+
+    if( 0 == (15 & (p_source->p[Y_PLANE].i_pitch|p_dest->p->i_pitch|
+        ((int)p_line|(int)p_y))) )
+    {
+        /* use faster SSE2 aligned fetch and store */
+        for( i_y = p_vout->render.i_height ; i_y-- ; )
+        {
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV422_YVYU_ALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV422_YVYU( p_line, p_y, p_u, p_v );
+            }
+            p_y += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line += i_dest_margin;
+        }
+    }
+    else {
+        /* use slower SSE2 unaligned fetch and store */
+        for( i_y = p_vout->render.i_height ; i_y-- ; )
+        {
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV422_YVYU_UNALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV422_YVYU( p_line, p_y, p_u, p_v );
+            }
+            p_y += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line += i_dest_margin;
+        }
+    }
+    SSE2_END;
+
+#else
+
     for( i_y = p_vout->render.i_height ; i_y-- ; )
     {
-        uint8_t *p_line = p_pixels;
         for( i_x = p_vout->render.i_width / 8 ; i_x-- ; )
         {
 #if defined (MODULE_NAME_IS_i422_yuy2)
@@ -202,12 +311,19 @@ static void I422_YVYU( vout_thread_t *p_vout, picture_t *p_source,
             MMX_CALL( MMX_YUV422_YVYU );
 #endif
         }
-        p_pixels += i_pitch;
+        for( i_x = ( p_vout->render.i_width % 8 ) / 2; i_x-- ; )
+        {
+            C_YUV422_YVYU( p_line, p_y, p_u, p_v );
+        }
+        p_y += i_source_margin;
+        p_u += i_source_margin_c;
+        p_v += i_source_margin_c;
+        p_line += i_dest_margin;
     }
 #if defined (MODULE_NAME_IS_i422_yuy2_mmx)
     MMX_END;
-#elif defined (MODULE_NAME_IS_i422_yuy2_sse2)
-    SSE2_END;
+#endif
+
 #endif
 }
 
@@ -217,17 +333,66 @@ static void I422_YVYU( vout_thread_t *p_vout, picture_t *p_source,
 static void I422_UYVY( vout_thread_t *p_vout, picture_t *p_source,
                                               picture_t *p_dest )
 {
-    uint8_t *p_pixels = p_dest->p->p_pixels;
-    int      i_pitch  = p_dest->p->i_pitch;
+    uint8_t *p_line = p_dest->p->p_pixels;
     uint8_t *p_y = p_source->Y_PIXELS;
     uint8_t *p_u = p_source->U_PIXELS;
     uint8_t *p_v = p_source->V_PIXELS;
 
     int i_x, i_y;
 
+    const int i_source_margin = p_source->p[0].i_pitch
+                                 - p_source->p[0].i_visible_pitch;
+    const int i_source_margin_c = p_source->p[1].i_pitch
+                                 - p_source->p[1].i_visible_pitch;
+    const int i_dest_margin = p_dest->p->i_pitch
+                               - p_dest->p->i_visible_pitch;
+
+#if defined (MODULE_NAME_IS_i422_yuy2_sse2)
+
+    if( 0 == (15 & (p_source->p[Y_PLANE].i_pitch|p_dest->p->i_pitch|
+        ((int)p_line|(int)p_y))) )
+    {
+        /* use faster SSE2 aligned fetch and store */
+        for( i_y = p_vout->render.i_height ; i_y-- ; )
+        {
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV422_UYVY_ALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV422_UYVY( p_line, p_y, p_u, p_v );
+            }
+            p_y += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line += i_dest_margin;
+        }
+    }
+    else {
+        /* use slower SSE2 unaligned fetch and store */
+        for( i_y = p_vout->render.i_height ; i_y-- ; )
+        {
+            for( i_x = p_vout->render.i_width / 16 ; i_x-- ; )
+            {
+                SSE2_CALL( SSE2_YUV422_UYVY_UNALIGNED );
+            }
+            for( i_x = ( p_vout->render.i_width % 16 ) / 2; i_x-- ; )
+            {
+                C_YUV422_UYVY( p_line, p_y, p_u, p_v );
+            }
+            p_y += i_source_margin;
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+            p_line += i_dest_margin;
+        }
+    }
+    SSE2_END;
+
+#else
+
     for( i_y = p_vout->render.i_height ; i_y-- ; )
     {
-        uint8_t *p_line = p_pixels;
         for( i_x = p_vout->render.i_width / 8 ; i_x-- ; )
         {
 #if defined (MODULE_NAME_IS_i422_yuy2)
@@ -239,12 +404,19 @@ static void I422_UYVY( vout_thread_t *p_vout, picture_t *p_source,
             MMX_CALL( MMX_YUV422_UYVY );
 #endif
         }
-        p_pixels += i_pitch;
+        for( i_x = ( p_vout->render.i_width % 8 ) / 2; i_x-- ; )
+        {
+            C_YUV422_UYVY( p_line, p_y, p_u, p_v );
+        }
+        p_y += i_source_margin;
+        p_u += i_source_margin_c;
+        p_v += i_source_margin_c;
+        p_line += i_dest_margin;
     }
 #if defined (MODULE_NAME_IS_i422_yuy2_mmx)
     MMX_END;
-#elif defined (MODULE_NAME_IS_i422_yuy2_sse2)
-    SSE2_END;
+#endif
+
 #endif
 }