try filtering more to remove chroma problems

CVS patchset: 5060
CVS date: 2003/06/18 01:27:50

3 files changed, 98 insertions, 13 deletions

diff --git a/src/post/deinterlace/speedy.c b/src/post/deinterlace/speedy.c
index c782f873f..5dab5fa0d 100644
--- a/src/post/deinterlace/speedy.c
+++ b/src/post/deinterlace/speedy.c
@@ -70,7 +70,9 @@ void (*filter_luma_14641_packed422_inplace_scanline)( uint8_t *data, int width )
 unsigned int (*diff_factor_packed422_scanline)( uint8_t *cur, uint8_t *old, int width );
 unsigned int (*comb_factor_packed422_scanline)( uint8_t *top, uint8_t *mid,
                                                 uint8_t *bot, int width );
-void (*linearblend_chroma_packed422_scanline)( uint8_t *output, int width,
+void (*vfilter_chroma_121_packed422_scanline)( uint8_t *output, int width,
+                                               uint8_t *m, uint8_t *t, uint8_t *b);
+void (*vfilter_chroma_332_packed422_scanline)( uint8_t *output, int width,
                                                uint8_t *m, uint8_t *t, uint8_t *b);
 void (*kill_chroma_packed422_inplace_scanline)( uint8_t *data, int width );
 void (*mirror_packed422_inplace_scanline)( uint8_t *data, int width );
@@ -523,7 +525,7 @@ void packed422_to_packed444_rec601_scanline( uint8_t *dest, uint8_t *src, int wi
     }
 }
 
-static void linearblend_chroma_packed422_scanline_mmx( uint8_t *output, int width,
+static void vfilter_chroma_121_packed422_scanline_mmx( uint8_t *output, int width,
                                                        uint8_t *m, uint8_t *t, uint8_t *b)
 {
 #ifdef ARCH_X86
@@ -569,21 +571,99 @@ static void linearblend_chroma_packed422_scanline_mmx( uint8_t *output, int widt
         b += 8;
         m += 8;
     }
+    output++; t++; b++; m++;
+    while( width-- ) {
+        *output = (*t + *b + (*m << 1)) >> 2;
+        output+=2; t+=2; b+=2; m+=2;
+    }
+
+    emms();
+#endif
+}
+
+
+static void vfilter_chroma_121_packed422_scanline_c( uint8_t *output, int width,
+                                                     uint8_t *m, uint8_t *t, uint8_t *b)
+{
+    output++; t++; b++; m++;
+    while( width-- ) {
+        *output = (*t + *b + (*m << 1)) >> 2;
+        output+=2; t+=2; b+=2; m+=2;
+    }
+}
+
+
+static void vfilter_chroma_332_packed422_scanline_mmx( uint8_t *output, int width,
+                                                       uint8_t *m, uint8_t *t, uint8_t *b)
+{
+#ifdef ARCH_X86
+    int i;
+    const mmx_t ymask = { 0x00ff00ff00ff00ffULL };
+    const mmx_t cmask = { 0xff00ff00ff00ff00ULL };
+
+    // Get width in bytes.
+    width *= 2;
+    i = width / 8;
+    width -= i * 8;
+
+    movq_m2r( ymask, mm7 );
+    movq_m2r( cmask, mm6 );
+
+    while( i-- ) {
+        movq_m2r( *t, mm0 );
+        movq_m2r( *b, mm1 );
+        movq_m2r( *m, mm2 );
+
+        movq_r2r ( mm2, mm3 );
+        pand_r2r ( mm7, mm3 );
+
+        pand_r2r ( mm6, mm0 );
+        pand_r2r ( mm6, mm1 );
+        pand_r2r ( mm6, mm2 );
+
+        psrlq_i2r( 8, mm0 );
+        psrlq_i2r( 7, mm1 );
+        psrlq_i2r( 8, mm2 );
+
+        movq_r2r ( mm0, mm4 );
+        psllw_i2r( 1, mm4 );
+        paddw_r2r( mm4, mm0 );
+
+        movq_r2r ( mm2, mm4 );
+        psllw_i2r( 1, mm4 );
+        paddw_r2r( mm4, mm2 );
+
+        paddw_r2r( mm0, mm2 );
+        paddw_r2r( mm1, mm2 );
+
+        psllw_i2r( 5, mm2 );
+        pand_r2r( mm6, mm2 );
+
+        por_r2r ( mm3, mm2 );
+
+        movq_r2m( mm2, *output );
+        output += 8;
+        t += 8;
+        b += 8;
+        m += 8;
+    }
+    output++; t++; b++; m++;
     while( width-- ) {
-        output++; t++; b++; m++;
-        *output++ = (*t++ + *b++ + (*m++ << 1)) >> 2;
+        *output = (3 * *t + 3 * *m + 2 * *b) >> 3;
+        output+=2; t+=2; b+=2; m+=2;
     }
 
     emms();
 #endif
 }
 
-static void linearblend_chroma_packed422_scanline_c( uint8_t *output, int width,
+static void vfilter_chroma_332_packed422_scanline_c( uint8_t *output, int width,
                                                      uint8_t *m, uint8_t *t, uint8_t *b)
 {
+    output++; t++; b++; m++;
     while( width-- ) {
-        output++; t++; b++; m++;
-        *output++ = (*t++ + *b++ + (*m++ << 1)) >> 2;
+        *output = (3 * *t + 3 * *m + 2 * *b) >> 3;
+        output+=2; t+=2; b+=2; m+=2;
     }
 }
 
@@ -2082,7 +2162,8 @@ void setup_speedy_calls( uint32_t accel, int verbose )
     filter_luma_14641_packed422_inplace_scanline = filter_luma_14641_packed422_inplace_scanline_c;
     comb_factor_packed422_scanline = 0;
     diff_factor_packed422_scanline = diff_factor_packed422_scanline_c;
-    linearblend_chroma_packed422_scanline = linearblend_chroma_packed422_scanline_c;
+    vfilter_chroma_121_packed422_scanline = vfilter_chroma_121_packed422_scanline_c;
+    vfilter_chroma_332_packed422_scanline = vfilter_chroma_332_packed422_scanline_c;
     kill_chroma_packed422_inplace_scanline = kill_chroma_packed422_inplace_scanline_c;
     mirror_packed422_inplace_scanline = mirror_packed422_inplace_scanline_c;
     halfmirror_packed422_inplace_scanline = halfmirror_packed422_inplace_scanline_c;
@@ -2105,7 +2186,8 @@ void setup_speedy_calls( uint32_t accel, int verbose )
         composite_packed4444_alpha_to_packed422_scanline = composite_packed4444_alpha_to_packed422_scanline_mmxext;
         composite_alphamask_to_packed4444_scanline = composite_alphamask_to_packed4444_scanline_mmxext;
         premultiply_packed4444_scanline = premultiply_packed4444_scanline_mmxext;
-        linearblend_chroma_packed422_scanline = linearblend_chroma_packed422_scanline_mmx;
+        vfilter_chroma_121_packed422_scanline = vfilter_chroma_121_packed422_scanline_mmx;
+        vfilter_chroma_332_packed422_scanline = vfilter_chroma_332_packed422_scanline_mmx;
         kill_chroma_packed422_inplace_scanline = kill_chroma_packed422_inplace_scanline_mmx;
         blend_packed422_scanline = blend_packed422_scanline_mmxext;
         diff_factor_packed422_scanline = diff_factor_packed422_scanline_mmx;
@@ -2123,7 +2205,8 @@ void setup_speedy_calls( uint32_t accel, int verbose )
         blit_packed422_scanline = blit_packed422_scanline_mmx;
         diff_factor_packed422_scanline = diff_factor_packed422_scanline_mmx;
         comb_factor_packed422_scanline = comb_factor_packed422_scanline_mmx;
-        linearblend_chroma_packed422_scanline = linearblend_chroma_packed422_scanline_mmx;
+        vfilter_chroma_121_packed422_scanline = vfilter_chroma_121_packed422_scanline_mmx;
+        vfilter_chroma_332_packed422_scanline = vfilter_chroma_332_packed422_scanline_mmx;
         kill_chroma_packed422_inplace_scanline = kill_chroma_packed422_inplace_scanline_mmx;
         diff_packed422_block8x8 = diff_packed422_block8x8_mmx;
         speedy_memcpy = speedy_memcpy_mmx;
diff --git a/src/post/deinterlace/speedy.h b/src/post/deinterlace/speedy.h
index b84615584..78fa42468 100644
--- a/src/post/deinterlace/speedy.h
+++ b/src/post/deinterlace/speedy.h
@@ -110,7 +110,9 @@ extern void (*filter_luma_14641_packed422_inplace_scanline)( uint8_t *data, int
 extern unsigned int (*diff_factor_packed422_scanline)( uint8_t *cur, uint8_t *old, int width );
 extern unsigned int (*comb_factor_packed422_scanline)( uint8_t *top, uint8_t *mid,
                                                        uint8_t *bot, int width );
-extern void (*linearblend_chroma_packed422_scanline)( uint8_t *output, int width,
+extern void (*vfilter_chroma_121_packed422_scanline)( uint8_t *output, int width,
+                                                      uint8_t *m, uint8_t *t, uint8_t *b);
+extern void (*vfilter_chroma_332_packed422_scanline)( uint8_t *output, int width,
                                                       uint8_t *m, uint8_t *t, uint8_t *b);
 extern void (*kill_chroma_packed422_inplace_scanline)( uint8_t *data, int width );
 extern void (*mirror_packed422_inplace_scanline)( uint8_t *data, int width );
diff --git a/src/post/deinterlace/xine_plugin.c b/src/post/deinterlace/xine_plugin.c
index c27473978..deef56239 100644
--- a/src/post/deinterlace/xine_plugin.c
+++ b/src/post/deinterlace/xine_plugin.c
@@ -17,7 +17,7 @@
  * along with this program; if not, write to the Free Software
  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA
  *
- * $Id: xine_plugin.c,v 1.5 2003/06/17 17:14:13 miguelfreitas Exp $
+ * $Id: xine_plugin.c,v 1.6 2003/06/18 01:27:50 miguelfreitas Exp $
  *
  * advanced video deinterlacer plugin
  * Jun/2003 by Miguel Freitas
@@ -515,7 +515,7 @@ static void apply_chroma_filter( uint8_t *data, int stride, int width, int heigh
    * at all, we just want a kind of filter here.
    */
   for( i = 0; i < height; i++, data += stride ) {
-    linearblend_chroma_packed422_scanline( data, width,
+    vfilter_chroma_332_packed422_scanline( data, width,
                                            data, 
                                            (i) ? (data - stride) : data,
                                            (i < height-1) ? (data + stride) : data );