X-Git-Url: http://research.m1stereo.tv/gitweb?a=blobdiff_plain;f=src%2Fmodules%2Fgtk2%2Fpixops.c;h=a835f704eabfaf1ac80571b1670482a317cd2127;hb=42e3082637998150a8dc0251c5b846f78e1d4772;hp=9fe02b0e6d90e87973e0545468df8f44ace45240;hpb=5f0547204a5a08185e9d8162016cc2ec9a670486;p=melted

diff --git a/src/modules/gtk2/pixops.c b/src/modules/gtk2/pixops.c
index 9fe02b0..a835f70 100644
--- a/src/modules/gtk2/pixops.c
+++ b/src/modules/gtk2/pixops.c
@@ -59,7 +59,7 @@ typedef void ( *PixopsPixelFunc ) ( guchar *dest, guint y1, guint cr, guint y2,
 
 /* mmx function declarations */
 #ifdef USE_MMX
-guchar *pixops_scale_line_22_33_mmx ( guint32 weights[ 16 ][ 8 ], guchar *p, guchar *q1, guchar *q2, int x_step, guchar *p_stop, int x_init );
+guchar *pixops_scale_line_22_yuv_mmx ( guint32 weights[ 16 ][ 8 ], guchar *p, guchar *q1, guchar *q2, int x_step, guchar *p_stop, int x_init, int destx );
 int pixops_have_mmx ( void );
 #endif
 
@@ -92,27 +92,22 @@ pixops_scale_nearest ( guchar *dest_buf,
                        double scale_x,
                        double scale_y )
 {
-	int i, j;
-	int x;
-	int x_step = ( 1 << SCALE_SHIFT ) / scale_x;
-	int y_step = ( 1 << SCALE_SHIFT ) / scale_y;
+	register int i, j;
+	register int x_step = ( 1 << SCALE_SHIFT ) / scale_x;
+	register int y_step = ( 1 << SCALE_SHIFT ) / scale_y;
+	register int x, x_scaled;
 
 	for ( i = 0; i < ( render_y1 - render_y0 ); i++ )
 	{
-		const guchar *src = src_buf + ( ( ( i + render_y0 ) * y_step + y_step / 2 ) >> SCALE_SHIFT ) * src_rowstride;
+		const guchar *src = src_buf + ( ( ( i + render_y0 ) * y_step + ( y_step >> 1 ) ) >> SCALE_SHIFT ) * src_rowstride;
 		guchar *dest = dest_buf + i * dest_rowstride;
-
-		x = render_x0 * x_step + x_step / 2;
-
+		x = render_x0 * x_step + ( x_step >> 1 );
+		
 		for ( j = 0; j < ( render_x1 - render_x0 ); j++ )
 		{
-			const guchar *p = src + ( x >> SCALE_SHIFT ) * 4;
-			guint32 *p32;
-
-			p32 = ( guint32 * ) dest;
-			*p32 = *( ( guint32 * ) p );
-
-			dest += 4;
+			x_scaled = x >> SCALE_SHIFT;
+			*dest++ = src[ x_scaled << 1 ];
+			*dest++ = src[ ( ( x_scaled >> 1 ) << 2 ) + ( ( j & 1 ) << 1 ) + 1 ];
 			x += x_step;
 		}
 	}
@@ -125,43 +120,37 @@ scale_line ( int *weights, int n_x, int n_y,
              guchar **src,
              int x_init, int x_step, int src_width )
 {
-	int x = x_init;
-	int i, j;
+	register int x = x_init;
+	register int i, j, x_scaled, y_index, uv_index;
 
 	while ( dest < dest_end )
 	{
-		int x_scaled = x >> SCALE_SHIFT;
-		int *pixel_weights;
+		unsigned int y = 0, uv = 0;
+		int *pixel_weights = weights + ( ( x >> ( SCALE_SHIFT - SUBSAMPLE_BITS ) ) & SUBSAMPLE_MASK ) * n_x * n_y;
 
-		pixel_weights = weights + ( ( x >> ( SCALE_SHIFT - SUBSAMPLE_BITS ) ) & SUBSAMPLE_MASK ) * n_x * n_y;
-
-		unsigned int y1 = 0, cb = 0, y2 = 0, cr = 0;
+		x_scaled = x >> SCALE_SHIFT;
+		y_index = x_scaled << 1;
+		uv_index = ( ( x_scaled >> 1 ) << 2 ) + ( ( dest_x & 1 ) << 1 ) + 1;
+		
 		for ( i = 0; i < n_y; i++ )
 		{
-			guchar *q = src[ i ] + x_scaled * 4;
 			int *line_weights = pixel_weights + n_x * i;
+			guchar *q = src[ i ];
 
-			for ( j = 0; j < n_x; j++ )
+			for ( j = 0; j < n_x; j ++ )
 			{
 				unsigned int ta = line_weights[ j ];
 
-				y1 += ta * q[ 0 ];
-				cb += ta * q[ 1 ];
-				y2 += ta * q[ 2 ];
-				cr += ta * q[ 3 ];
-
-				q += 4;
+				y  += ta * q[ y_index ];
+				uv += ta * q[ uv_index ];
 			}
 		}
 
-		dest[ 0 ] = ( y1 + 0xffff ) >> 16;
-		dest[ 1 ] = ( cb + 0xffff ) >> 16;
-		dest[ 2 ] = ( y2 + 0xffff ) >> 16;
-		dest[ 3 ] = ( cr + 0xffff ) >> 16;
-
-		dest += 4;
+		*dest++ = ( y  + 0xffff ) >> SCALE_SHIFT;
+		*dest++ = ( uv + 0xffff ) >> SCALE_SHIFT;
 
 		x += x_step;
+		dest_x++;
 	}
 
 	return dest;
@@ -169,7 +158,7 @@ scale_line ( int *weights, int n_x, int n_y,
 
 #ifdef USE_MMX
 static inline guchar *
-scale_line_22_33_mmx_stub ( int *weights, int n_x, int n_y,
+scale_line_22_yuv_mmx_stub ( int *weights, int n_x, int n_y,
                             guchar *dest, int dest_x, guchar *dest_end,
                             guchar **src,
                             int x_init, int x_step, int src_width )
@@ -189,65 +178,59 @@ scale_line_22_33_mmx_stub ( int *weights, int n_x, int n_y,
 		mmx_weights[ j ][ 7 ] = 0x00010001 * ( weights[ 4 * j + 3 ] >> 8 );
 	}
 
-	return pixops_scale_line_22_33_mmx ( mmx_weights, dest, src[ 0 ], src[ 1 ], x_step, dest_end, x_init );
+	return pixops_scale_line_22_yuv_mmx ( mmx_weights, dest, src[ 0 ], src[ 1 ], x_step, dest_end, x_init, dest_x );
 }
 #endif /* USE_MMX */
 
 static inline guchar *
-scale_line_22_33 ( int *weights, int n_x, int n_y,
+scale_line_22_yuv ( int *weights, int n_x, int n_y,
                    guchar *dest, int dest_x, guchar *dest_end,
                    guchar **src,
                    int x_init, int x_step, int src_width )
 {
-	int x = x_init;
-	guchar *src0 = src[ 0 ];
-	guchar *src1 = src[ 1 ];
+	register int x = x_init;
+	register guchar *src0 = src[ 0 ];
+	register guchar *src1 = src[ 1 ];
+	register unsigned int p;
+	register guchar *q0, *q1;
+	register int w1, w2, w3, w4;
+	register int x_scaled, x_aligned, uv_index;
 
 	while ( dest < dest_end )
 	{
-		unsigned int y1, cb, y2, cr;
-		int x_scaled = x >> SCALE_SHIFT;
-		int *pixel_weights;
-		guchar *q0, *q1;
-		int w1, w2, w3, w4;
-
-		q0 = src0 + x_scaled * 4;
-		q1 = src1 + x_scaled * 4;
-
-		pixel_weights = weights + ( ( x >> ( SCALE_SHIFT - SUBSAMPLE_BITS ) ) & SUBSAMPLE_MASK ) * 4;
+		int *pixel_weights = weights + ( ( x >> ( SCALE_SHIFT - SUBSAMPLE_BITS ) ) & SUBSAMPLE_MASK ) * 4;
+		
+		x_scaled = x >> SCALE_SHIFT;
 
 		w1 = pixel_weights[ 0 ];
 		w2 = pixel_weights[ 1 ];
 		w3 = pixel_weights[ 2 ];
 		w4 = pixel_weights[ 3 ];
 
-		y1 = w1 * q0[ 0 ];
-		cb = w1 * q0[ 1 ];
-		y2 = w1 * q0[ 2 ];
-		cr = w1 * q0[ 3 ];
-
-		y1 += w2 * q0[ 4 ];
-		cb += w2 * q0[ 5 ];
-		y2 += w2 * q0[ 6 ];
-		cr += w2 * q0[ 7 ];
-
-		y1 += w3 * q1[ 0 ];
-		cb += w3 * q1[ 1 ];
-		y2 += w3 * q1[ 2 ];
-		cr += w3 * q1[ 3 ];
-
-		y1 += w4 * q1[ 4 ];
-		cb += w4 * q1[ 5 ];
-		y2 += w4 * q1[ 6 ];
-		cr += w4 * q1[ 7 ];
-
-		dest[ 0 ] = ( y1 + 0x8000 ) >> 16;
-		dest[ 1 ] = ( cb + 0x8000 ) >> 16;
-		dest[ 2 ] = ( y2 + 0x8000 ) >> 16;
-		dest[ 3 ] = ( cr + 0x8000 ) >> 16;
-
-		dest += 4;
+		/* process Y */
+		q0 = src0 + ( x_scaled << 1 );
+		q1 = src1 + ( x_scaled << 1 );
+		p  = w1 * q0[ 0 ];
+		p += w2 * q0[ 2 ];
+		p += w3 * q1[ 0 ];
+		p += w4 * q1[ 2 ];
+		*dest++ = ( p + 0x8000 ) >> SCALE_SHIFT;
+
+		/* process U/V */
+		x_aligned = ( ( x_scaled >> 1 ) << 2 );
+		uv_index = ( ( dest_x & 1 ) << 1 ) + 1;
+		
+		q0 = src0 + x_aligned;
+		q1 = src1 + x_aligned;
+		p  = w1 * q0[ uv_index ];
+		p += w3 * q1[ uv_index ];
+		p += w2 * q0[ uv_index ];
+		p += w4 * q1[ uv_index ];
+		
 		x += x_step;
+		dest_x ++;
+
+		*dest++ = ( p + 0x8000 ) >> SCALE_SHIFT;
 	}
 
 	return dest;
@@ -260,8 +243,9 @@ process_pixel ( int *weights, int n_x, int n_y,
                 guchar **src, int src_channels,
                 int x_start, int src_width )
 {
-	unsigned int y1 = 0, cb = 0, y2 = 0, cr = 0;
-	int i, j;
+	register unsigned int y = 0, uv = 0;
+	register int i, j;
+	int uv_index = ( ( dest_x & 1 ) << 1 ) + 1;
 
 	for ( i = 0; i < n_y; i++ )
 	{
@@ -269,29 +253,28 @@ process_pixel ( int *weights, int n_x, int n_y,
 
 		for ( j = 0; j < n_x; j++ )
 		{
-			unsigned int ta;
-			guchar *q;
+			unsigned int ta = 0xff * line_weights[ j ];
 
 			if ( x_start + j < 0 )
-				q = src[ i ];
+			{
+				y  += ta * src[ i ][ 0 ];
+				uv += ta * src[ i ][ uv_index ];
+			}
 			else if ( x_start + j < src_width )
-				q = src[ i ] + ( x_start + j ) * src_channels;
+			{
+				y  += ta * src[ i ][ ( x_start + j ) << 1 ];
+				uv += ta * src[ i ][ ( ( ( x_start + j ) >> 1 ) << 2) + uv_index ];
+			}
 			else
-				q = src[ i ] + ( src_width - 1 ) * src_channels;
-
-			ta = 0xff * line_weights[ j ];
-
-			y1 += ta * q[ 0 ];
-			cb += ta * q[ 1 ];
-			y2 += ta * q[ 2 ];
-			cr += ta * q[ 3 ];
+			{
+				y  += ta * src[ i ][ ( src_width - 1 ) << 1 ];
+				uv += ta * src[ i ][ ( ( ( src_width - 1 ) >> 1 ) << 2) + uv_index ];
+			}
 		}
 	}
 
-	dest[ 0 ] = ( y1 + 0xffffff ) >> 24;
-	dest[ 1 ] = ( cb + 0xffffff ) >> 24;
-	dest[ 2 ] = ( y2 + 0xffffff ) >> 24;
-	dest[ 3 ] = ( cr + 0xffffff ) >> 24;
+	*dest++ = ( y  + 0xffffff ) >> 24;
+	*dest++ = ( uv + 0xffffff ) >> 24;
 }
 
 
@@ -749,22 +732,24 @@ yuv422_scale ( guchar *dest_buf,
 		                       dest_rowstride,
 		                       src_buf, src_width, src_height, src_rowstride,
 		                       scale_x, scale_y );
-		return ;
+		return;
 	}
 
 	filter.overall_alpha = 1.0;
 	make_weights ( &filter, interp_type, scale_x, scale_y );
 
-fprintf( stderr, "RESCALE: %d %d\n", filter.x.n, filter.y.n );
 	if ( filter.x.n == 2 && filter.y.n == 2 )
 	{
 #ifdef USE_MMX
-		if ( 0 && found_mmx )
-			line_func = scale_line_22_33_mmx_stub;
+		if ( found_mmx )
+		{
+			//fprintf( stderr, "rescale: using mmx\n" );
+			line_func = scale_line_22_yuv_mmx_stub;
+		}
 		else
 #endif
 
-			line_func = scale_line_22_33;
+			line_func = scale_line_22_yuv;
 	}
 	else
 		line_func = scale_line;