--- trunk/xvidcore/src/image/image.c	2005/12/17 11:24:32	1664
+++ trunk/xvidcore/src/image/image.c	2006/10/13 11:26:18	1736
@@ -19,7 +19,7 @@
  *  along with this program ; if not, write to the Free Software
  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  *
- * $Id: image.c,v 1.33 2005-12-17 11:24:32 syskin Exp $
+ * $Id: image.c,v 1.38 2006-10-13 11:26:18 Skal Exp $
  *
  ****************************************************************************/
 
@@ -34,6 +34,7 @@
 #include "interpolate8x8.h"
 #include "../utils/mem_align.h"
 #include "../motion/sad.h"
+#include "../utils/emms.h"
 
 #include "font.h"		/* XXX: remove later */
 
@@ -236,12 +237,11 @@
 	}
 }
 
-/* bframe encoding requires image-based u,v interpolation */
 void
-image_interpolate(const IMAGE * refn,
-				  IMAGE * refh,
-				  IMAGE * refv,
-				  IMAGE * refhv,
+image_interpolate(const uint8_t * refn,
+				  uint8_t * refh,
+				  uint8_t * refv,
+				  uint8_t * refhv,
 				  uint32_t edged_width,
 				  uint32_t edged_height,
 				  uint32_t quarterpel,
@@ -250,13 +250,13 @@
 	const uint32_t offset = EDGE_SIZE2 * (edged_width + 1); /* we only interpolate half of the edge area */
 	const uint32_t stride_add = 7 * edged_width;
 
-	uint8_t *n_ptr, *h_ptr, *v_ptr, *hv_ptr;
+	uint8_t *n_ptr;
+	uint8_t *h_ptr, *v_ptr, *hv_ptr;
 	uint32_t x, y;
 
-
-	n_ptr = refn->y;
-	h_ptr = refh->y;
-	v_ptr = refv->y;
+	n_ptr = (uint8_t*)refn;
+	h_ptr = refh;
+	v_ptr = refv;
 
 	n_ptr -= offset;
 	h_ptr -= offset;
@@ -285,8 +285,8 @@
 			n_ptr += stride_add;
 		}
 
-		h_ptr = refh->y + (edged_height - EDGE_SIZE - EDGE_SIZE2)*edged_width - EDGE_SIZE2;
-		hv_ptr = refhv->y + (edged_height - EDGE_SIZE - EDGE_SIZE2)*edged_width - EDGE_SIZE2;
+		h_ptr = refh + (edged_height - EDGE_SIZE - EDGE_SIZE2)*edged_width - EDGE_SIZE2;
+		hv_ptr = refhv + (edged_height - EDGE_SIZE - EDGE_SIZE2)*edged_width - EDGE_SIZE2;
 
 		for (y = 0; y < (edged_height - EDGE_SIZE); y = y + 8) {
 			hv_ptr -= stride_add;
@@ -302,7 +302,7 @@
 		}
 	} else {
 
-		hv_ptr = refhv->y;
+		hv_ptr = refhv;
 		hv_ptr -= offset;
 
 		for (y = 0; y < (edged_height - EDGE_SIZE); y += 8) {
@@ -1015,3 +1015,70 @@
 		p += edged_width/2;
 	}
 }
+
+/****************************************************************************/
+
+static void (*deintl_core)(uint8_t *, int width, int height, const int stride) = 0;
+extern void xvid_deinterlace_sse(uint8_t *, int width, int height, const int stride);
+
+#define CLIP_255(x)   ( ((x)&~255) ? ((-(x)) >> (8*sizeof((x))-1))&0xff : (x) )
+
+static void deinterlace_c(uint8_t *pix, int width, int height, const int bps)
+{
+  pix += bps;
+  while(width-->0)
+  {
+    int p1 = pix[-bps];
+    int p2 = pix[0];
+    int p0 = p2;
+    int j = (height>>1) - 1;
+    int V;
+    unsigned char *P = pix++;
+    while(j-->0)
+    {
+      const int  p3 = P[  bps];
+      const int  p4 = P[2*bps];
+      V =  ((p1+p3+1)>>1) + ((p2 - ((p0+p4+1)>>1)) >> 2);
+      P[0] = CLIP_255( V );
+      p0 = p2;
+      p1 = p3;
+      p2 = p4;
+      P += 2*bps;
+    }
+    V =  ((p1+p1+1)>>1) + ((p2 - ((p0+p2+1)>>1)) >> 2);
+    P[0] = CLIP_255( V );
+  }
+}
+#undef CLIP_255
+
+int xvid_image_deinterlace(xvid_image_t* img, int width, int height, int bottom_first)
+{
+	if (height&1)
+		return 0;
+	if (img->csp!=XVID_CSP_PLANAR && img->csp!=XVID_CSP_I420 && img->csp!=XVID_CSP_YV12)
+		return 0;       /* not yet supported */	
+	if (deintl_core==0) {
+		deintl_core = deinterlace_c;
+#ifdef ARCH_IS_IA32
+		{
+			int cpu_flags = check_cpu_features();
+			if (cpu_flags & XVID_CPU_MMX)
+				deintl_core = xvid_deinterlace_sse;
+		}
+#endif			
+	}
+	if (!bottom_first) {
+		deintl_core(img->plane[0], width,    height,    img->stride[0]);
+		deintl_core(img->plane[1], width>>1, height>>1, img->stride[1]);
+		deintl_core(img->plane[2], width>>1, height>>1, img->stride[2]);
+	}
+	else {
+		deintl_core((uint8_t *)img->plane[0] + ( height    -1)*img->stride[0], width,    height,    -img->stride[0]);
+		deintl_core((uint8_t *)img->plane[1] + ((height>>1)-1)*img->stride[1], width>>1, height>>1, -img->stride[1]);
+		deintl_core((uint8_t *)img->plane[2] + ((height>>1)-1)*img->stride[2], width>>1, height>>1, -img->stride[2]);
+	}
+	emms();
+
+	return 1;
+}
+