some bugfixes and rescale filter
[melted] / src / modules / gtk2 / pixops.c
diff --git a/src/modules/gtk2/pixops.c b/src/modules/gtk2/pixops.c
new file mode 100644 (file)
index 0000000..9fe02b0
--- /dev/null
@@ -0,0 +1,781 @@
+/* GdkPixbuf library - Scaling and compositing functions
+ *
+ * Copyright (C) 1999 The Free Software Foundation
+ *
+ * Author: Owen Taylor <otaylor@redhat.com>
+ * Modified for YUV422 by Dan Dennedy <dan@dennedy.org>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 02111-1307, USA.
+ */
+#include <math.h>
+#include <glib.h>
+#include <stdio.h>
+
+#include "pixops.h"
+
+#define SUBSAMPLE_BITS 4
+#define SUBSAMPLE (1 << SUBSAMPLE_BITS)
+#define SUBSAMPLE_MASK ((1 << SUBSAMPLE_BITS)-1)
+#define SCALE_SHIFT 16
+
+typedef struct _PixopsFilter PixopsFilter;
+typedef struct _PixopsFilterDimension PixopsFilterDimension;
+
+struct _PixopsFilterDimension
+{
+       int n;
+       double offset;
+       double *weights;
+};
+
+struct _PixopsFilter
+{
+       PixopsFilterDimension x;
+       PixopsFilterDimension y;
+       double overall_alpha;
+};
+
+typedef guchar *( *PixopsLineFunc ) ( 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 );
+
+typedef void ( *PixopsPixelFunc ) ( guchar *dest, guint y1, guint cr, guint y2, guint cb );
+
+
+/* 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 );
+int pixops_have_mmx ( void );
+#endif
+
+static inline int
+get_check_shift ( int check_size )
+{
+       int check_shift = 0;
+       g_return_val_if_fail ( check_size >= 0, 4 );
+
+       while ( !( check_size & 1 ) )
+       {
+               check_shift++;
+               check_size >>= 1;
+       }
+
+       return check_shift;
+}
+
+static inline void
+pixops_scale_nearest ( guchar *dest_buf,
+                       int render_x0,
+                       int render_y0,
+                       int render_x1,
+                       int render_y1,
+                       int dest_rowstride,
+                       const guchar *src_buf,
+                       int src_width,
+                       int src_height,
+                       int src_rowstride,
+                       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;
+
+       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;
+               guchar *dest = dest_buf + i * dest_rowstride;
+
+               x = render_x0 * x_step + x_step / 2;
+
+               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 += x_step;
+               }
+       }
+}
+
+
+static inline guchar *
+scale_line ( 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;
+       int i, j;
+
+       while ( dest < dest_end )
+       {
+               int x_scaled = x >> SCALE_SHIFT;
+               int *pixel_weights;
+
+               pixel_weights = weights + ( ( x >> ( SCALE_SHIFT - SUBSAMPLE_BITS ) ) & SUBSAMPLE_MASK ) * n_x * n_y;
+
+               unsigned int y1 = 0, cb = 0, y2 = 0, cr = 0;
+               for ( i = 0; i < n_y; i++ )
+               {
+                       guchar *q = src[ i ] + x_scaled * 4;
+                       int *line_weights = pixel_weights + n_x * i;
+
+                       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;
+                       }
+               }
+
+               dest[ 0 ] = ( y1 + 0xffff ) >> 16;
+               dest[ 1 ] = ( cb + 0xffff ) >> 16;
+               dest[ 2 ] = ( y2 + 0xffff ) >> 16;
+               dest[ 3 ] = ( cr + 0xffff ) >> 16;
+
+               dest += 4;
+
+               x += x_step;
+       }
+
+       return dest;
+}
+
+#ifdef USE_MMX
+static inline guchar *
+scale_line_22_33_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 )
+{
+       guint32 mmx_weights[ 16 ][ 8 ];
+       int j;
+
+       for ( j = 0; j < 16; j++ )
+       {
+               mmx_weights[ j ][ 0 ] = 0x00010001 * ( weights[ 4 * j ] >> 8 );
+               mmx_weights[ j ][ 1 ] = 0x00010001 * ( weights[ 4 * j ] >> 8 );
+               mmx_weights[ j ][ 2 ] = 0x00010001 * ( weights[ 4 * j + 1 ] >> 8 );
+               mmx_weights[ j ][ 3 ] = 0x00010001 * ( weights[ 4 * j + 1 ] >> 8 );
+               mmx_weights[ j ][ 4 ] = 0x00010001 * ( weights[ 4 * j + 2 ] >> 8 );
+               mmx_weights[ j ][ 5 ] = 0x00010001 * ( weights[ 4 * j + 2 ] >> 8 );
+               mmx_weights[ j ][ 6 ] = 0x00010001 * ( weights[ 4 * j + 3 ] >> 8 );
+               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 );
+}
+#endif /* USE_MMX */
+
+static inline guchar *
+scale_line_22_33 ( 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 ];
+
+       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;
+
+               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;
+               x += x_step;
+       }
+
+       return dest;
+}
+
+
+static inline void
+process_pixel ( int *weights, int n_x, int n_y,
+                guchar *dest, int dest_x, int dest_channels,
+                guchar **src, int src_channels,
+                int x_start, int src_width )
+{
+       unsigned int y1 = 0, cb = 0, y2 = 0, cr = 0;
+       int i, j;
+
+       for ( i = 0; i < n_y; i++ )
+       {
+               int *line_weights = weights + n_x * i;
+
+               for ( j = 0; j < n_x; j++ )
+               {
+                       unsigned int ta;
+                       guchar *q;
+
+                       if ( x_start + j < 0 )
+                               q = src[ i ];
+                       else if ( x_start + j < src_width )
+                               q = src[ i ] + ( x_start + j ) * src_channels;
+                       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 ];
+               }
+       }
+
+       dest[ 0 ] = ( y1 + 0xffffff ) >> 24;
+       dest[ 1 ] = ( cb + 0xffffff ) >> 24;
+       dest[ 2 ] = ( y2 + 0xffffff ) >> 24;
+       dest[ 3 ] = ( cr + 0xffffff ) >> 24;
+}
+
+
+static inline void
+correct_total ( int *weights,
+                int n_x,
+                int n_y,
+                int total,
+                double overall_alpha )
+{
+       int correction = ( int ) ( 0.5 + 65536 * overall_alpha ) - total;
+       int remaining, c, d, i;
+
+       if ( correction != 0 )
+       {
+               remaining = correction;
+               for ( d = 1, c = correction; c != 0 && remaining != 0; d++, c = correction / d )
+                       for ( i = n_x * n_y - 1; i >= 0 && c != 0 && remaining != 0; i-- )
+                               if ( *( weights + i ) + c >= 0 )
+                               {
+                                       *( weights + i ) += c;
+                                       remaining -= c;
+                                       if ( ( 0 < remaining && remaining < c ) ||
+                                               ( 0 > remaining && remaining > c ) )
+                                               c = remaining;
+                               }
+       }
+}
+
+
+static inline int *
+make_filter_table ( PixopsFilter *filter )
+{
+       int i_offset, j_offset;
+       int n_x = filter->x.n;
+       int n_y = filter->y.n;
+       int *weights = g_new ( int, SUBSAMPLE * SUBSAMPLE * n_x * n_y );
+
+       for ( i_offset = 0; i_offset < SUBSAMPLE; i_offset++ )
+               for ( j_offset = 0; j_offset < SUBSAMPLE; j_offset++ )
+               {
+                       double weight;
+                       int *pixel_weights = weights + ( ( i_offset * SUBSAMPLE ) + j_offset ) * n_x * n_y;
+                       int total = 0;
+                       int i, j;
+
+                       for ( i = 0; i < n_y; i++ )
+                               for ( j = 0; j < n_x; j++ )
+                               {
+                                       weight = filter->x.weights[ ( j_offset * n_x ) + j ] *
+                                                filter->y.weights[ ( i_offset * n_y ) + i ] *
+                                                filter->overall_alpha * 65536 + 0.5;
+
+                                       total += ( int ) weight;
+
+                                       *( pixel_weights + n_x * i + j ) = weight;
+                               }
+
+                       correct_total ( pixel_weights, n_x, n_y, total, filter->overall_alpha );
+               }
+
+       return weights;
+}
+
+
+static inline void
+pixops_process ( guchar *dest_buf,
+                 int render_x0,
+                 int render_y0,
+                 int render_x1,
+                 int render_y1,
+                 int dest_rowstride,
+                 int dest_channels,
+                 gboolean dest_has_alpha,
+                 const guchar *src_buf,
+                 int src_width,
+                 int src_height,
+                 int src_rowstride,
+                 int src_channels,
+                 gboolean src_has_alpha,
+                 double scale_x,
+                 double scale_y,
+                 int check_x,
+                 int check_y,
+                 int check_size,
+                 guint32 color1,
+                 guint32 color2,
+                 PixopsFilter *filter,
+                 PixopsLineFunc line_func )
+{
+       int i, j;
+       int x, y;                       /* X and Y position in source (fixed_point) */
+
+       guchar **line_bufs = g_new ( guchar *, filter->y.n );
+       int *filter_weights = make_filter_table ( filter );
+
+       int x_step = ( 1 << SCALE_SHIFT ) / scale_x; /* X step in source (fixed point) */
+       int y_step = ( 1 << SCALE_SHIFT ) / scale_y; /* Y step in source (fixed point) */
+
+       int check_shift = check_size ? get_check_shift ( check_size ) : 0;
+
+       int scaled_x_offset = floor ( filter->x.offset * ( 1 << SCALE_SHIFT ) );
+
+       /* Compute the index where we run off the end of the source buffer. The furthest
+        * source pixel we access at index i is:
+        *
+        *  ((render_x0 + i) * x_step + scaled_x_offset) >> SCALE_SHIFT + filter->x.n - 1
+        *
+        * So, run_end_index is the smallest i for which this pixel is src_width, i.e, for which:
+        *
+        *  (i + render_x0) * x_step >= ((src_width - filter->x.n + 1) << SCALE_SHIFT) - scaled_x_offset
+        *
+        */
+#define MYDIV(a,b) ((a) > 0 ? (a) / (b) : ((a) - (b) + 1) / (b))    /* Division so that -1/5 = -1 */
+
+       int run_end_x = ( ( ( src_width - filter->x.n + 1 ) << SCALE_SHIFT ) - scaled_x_offset );
+       int run_end_index = MYDIV ( run_end_x + x_step - 1, x_step ) - render_x0;
+       run_end_index = MIN ( run_end_index, render_x1 - render_x0 );
+
+       y = render_y0 * y_step + floor ( filter->y.offset * ( 1 << SCALE_SHIFT ) );
+       for ( i = 0; i < ( render_y1 - render_y0 ); i++ )
+       {
+               int dest_x;
+               int y_start = y >> SCALE_SHIFT;
+               int x_start;
+               int *run_weights = filter_weights +
+                                  ( ( y >> ( SCALE_SHIFT - SUBSAMPLE_BITS ) ) & SUBSAMPLE_MASK ) *
+                                  filter->x.n * filter->y.n * SUBSAMPLE;
+               guchar *new_outbuf;
+               guint32 tcolor1, tcolor2;
+
+               guchar *outbuf = dest_buf + dest_rowstride * i;
+               guchar *outbuf_end = outbuf + dest_channels * ( render_x1 - render_x0 );
+
+               if ( ( ( i + check_y ) >> check_shift ) & 1 )
+               {
+                       tcolor1 = color2;
+                       tcolor2 = color1;
+               }
+               else
+               {
+                       tcolor1 = color1;
+                       tcolor2 = color2;
+               }
+
+               for ( j = 0; j < filter->y.n; j++ )
+               {
+                       if ( y_start < 0 )
+                               line_bufs[ j ] = ( guchar * ) src_buf;
+                       else if ( y_start < src_height )
+                               line_bufs[ j ] = ( guchar * ) src_buf + src_rowstride * y_start;
+                       else
+                               line_bufs[ j ] = ( guchar * ) src_buf + src_rowstride * ( src_height - 1 );
+
+                       y_start++;
+               }
+
+               dest_x = check_x;
+               x = render_x0 * x_step + scaled_x_offset;
+               x_start = x >> SCALE_SHIFT;
+
+               while ( x_start < 0 && outbuf < outbuf_end )
+               {
+                       process_pixel ( run_weights + ( ( x >> ( SCALE_SHIFT - SUBSAMPLE_BITS ) ) & SUBSAMPLE_MASK ) * ( filter->x.n * filter->y.n ),
+                                       filter->x.n, filter->y.n,
+                                       outbuf, dest_x, dest_channels,
+                                       line_bufs, src_channels,
+                                       x >> SCALE_SHIFT, src_width );
+
+                       x += x_step;
+                       x_start = x >> SCALE_SHIFT;
+                       dest_x++;
+                       outbuf += dest_channels;
+               }
+
+               new_outbuf = ( *line_func ) ( run_weights, filter->x.n, filter->y.n,
+                                             outbuf, dest_x,
+                                             dest_buf + dest_rowstride * i + run_end_index * dest_channels,
+                                             line_bufs,
+                                             x, x_step, src_width );
+
+               dest_x += ( new_outbuf - outbuf ) / dest_channels;
+
+               x = ( dest_x - check_x + render_x0 ) * x_step + scaled_x_offset;
+               outbuf = new_outbuf;
+
+               while ( outbuf < outbuf_end )
+               {
+                       process_pixel ( run_weights + ( ( x >> ( SCALE_SHIFT - SUBSAMPLE_BITS ) ) & SUBSAMPLE_MASK ) * ( filter->x.n * filter->y.n ),
+                                       filter->x.n, filter->y.n,
+                                       outbuf, dest_x, dest_channels,
+                                       line_bufs, src_channels,
+                                       x >> SCALE_SHIFT, src_width );
+
+                       x += x_step;
+                       dest_x++;
+                       outbuf += dest_channels;
+               }
+
+               y += y_step;
+       }
+
+       g_free ( line_bufs );
+       g_free ( filter_weights );
+}
+
+
+/* Compute weights for reconstruction by replication followed by
+ * sampling with a box filter
+ */
+static inline void
+tile_make_weights ( PixopsFilterDimension *dim,
+                    double scale )
+{
+       int n = ceil ( 1 / scale + 1 );
+       double *pixel_weights = g_new ( double, SUBSAMPLE * n );
+       int offset;
+       int i;
+
+       dim->n = n;
+       dim->offset = 0;
+       dim->weights = pixel_weights;
+
+       for ( offset = 0; offset < SUBSAMPLE; offset++ )
+       {
+               double x = ( double ) offset / SUBSAMPLE;
+               double a = x + 1 / scale;
+
+               for ( i = 0; i < n; i++ )
+               {
+                       if ( i < x )
+                       {
+                               if ( i + 1 > x )
+                                       * ( pixel_weights++ ) = ( MIN ( i + 1, a ) - x ) * scale;
+                               else
+                                       *( pixel_weights++ ) = 0;
+                       }
+                       else
+                       {
+                               if ( a > i )
+                                       * ( pixel_weights++ ) = ( MIN ( i + 1, a ) - i ) * scale;
+                               else
+                                       *( pixel_weights++ ) = 0;
+                       }
+               }
+       }
+}
+
+/* Compute weights for a filter that, for minification
+ * is the same as 'tiles', and for magnification, is bilinear
+ * reconstruction followed by a sampling with a delta function.
+ */
+static inline void
+bilinear_magnify_make_weights ( PixopsFilterDimension *dim,
+                                double scale )
+{
+       double * pixel_weights;
+       int n;
+       int offset;
+       int i;
+
+       if ( scale > 1.0 )              /* Linear */
+       {
+               n = 2;
+               dim->offset = 0.5 * ( 1 / scale - 1 );
+       }
+       else                          /* Tile */
+       {
+               n = ceil ( 1.0 + 1.0 / scale );
+               dim->offset = 0.0;
+       }
+
+       dim->n = n;
+       dim->weights = g_new ( double, SUBSAMPLE * n );
+
+       pixel_weights = dim->weights;
+
+       for ( offset = 0; offset < SUBSAMPLE; offset++ )
+       {
+               double x = ( double ) offset / SUBSAMPLE;
+
+               if ( scale > 1.0 )        /* Linear */
+               {
+                       for ( i = 0; i < n; i++ )
+                               *( pixel_weights++ ) = ( ( ( i == 0 ) ? ( 1 - x ) : x ) / scale ) * scale;
+               }
+               else                  /* Tile */
+               {
+                       double a = x + 1 / scale;
+
+                       /*           x
+                        * ---------|--.-|----|--.-|-------  SRC
+                        * ------------|---------|---------  DEST
+                        */
+                       for ( i = 0; i < n; i++ )
+                       {
+                               if ( i < x )
+                               {
+                                       if ( i + 1 > x )
+                                               * ( pixel_weights++ ) = ( MIN ( i + 1, a ) - x ) * scale;
+                                       else
+                                               *( pixel_weights++ ) = 0;
+                               }
+                               else
+                               {
+                                       if ( a > i )
+                                               * ( pixel_weights++ ) = ( MIN ( i + 1, a ) - i ) * scale;
+                                       else
+                                               *( pixel_weights++ ) = 0;
+                               }
+                       }
+               }
+       }
+}
+
+/* Computes the integral from b0 to b1 of
+ *
+ * f(x) = x; 0 <= x < 1
+ * f(x) = 0; otherwise
+ *
+ * We combine two of these to compute the convolution of
+ * a box filter with a triangular spike.
+ */
+static inline double
+linear_box_half ( double b0, double b1 )
+{
+       double a0, a1;
+       double x0, x1;
+
+       a0 = 0.;
+       a1 = 1.;
+
+       if ( a0 < b0 )
+       {
+               if ( a1 > b0 )
+               {
+                       x0 = b0;
+                       x1 = MIN ( a1, b1 );
+               }
+               else
+                       return 0;
+       }
+       else
+       {
+               if ( b1 > a0 )
+               {
+                       x0 = a0;
+                       x1 = MIN ( a1, b1 );
+               }
+               else
+                       return 0;
+       }
+
+       return 0.5 * ( x1 * x1 - x0 * x0 );
+}
+
+/* Compute weights for reconstructing with bilinear
+ * interpolation, then sampling with a box filter
+ */
+static inline void
+bilinear_box_make_weights ( PixopsFilterDimension *dim,
+                            double scale )
+{
+       int n = ceil ( 1 / scale + 2.0 );
+       double *pixel_weights = g_new ( double, SUBSAMPLE * n );
+       double w;
+       int offset, i;
+
+       dim->offset = -1.0;
+       dim->n = n;
+       dim->weights = pixel_weights;
+
+       for ( offset = 0 ; offset < SUBSAMPLE; offset++ )
+       {
+               double x = ( double ) offset / SUBSAMPLE;
+               double a = x + 1 / scale;
+
+               for ( i = 0; i < n; i++ )
+               {
+                       w = linear_box_half ( 0.5 + i - a, 0.5 + i - x );
+                       w += linear_box_half ( 1.5 + x - i, 1.5 + a - i );
+
+                       *( pixel_weights++ ) = w * scale;
+               }
+       }
+}
+
+
+static inline void
+make_weights ( PixopsFilter *filter,
+               PixopsInterpType interp_type,
+               double scale_x,
+               double scale_y )
+{
+       switch ( interp_type )
+       {
+       case PIXOPS_INTERP_NEAREST:
+               g_assert_not_reached ();
+               break;
+
+       case PIXOPS_INTERP_TILES:
+               tile_make_weights ( &filter->x, scale_x );
+               tile_make_weights ( &filter->y, scale_y );
+               break;
+
+       case PIXOPS_INTERP_BILINEAR:
+               bilinear_magnify_make_weights ( &filter->x, scale_x );
+               bilinear_magnify_make_weights ( &filter->y, scale_y );
+               break;
+
+       case PIXOPS_INTERP_HYPER:
+               bilinear_box_make_weights ( &filter->x, scale_x );
+               bilinear_box_make_weights ( &filter->y, scale_y );
+               break;
+       }
+}
+
+
+void
+yuv422_scale ( guchar *dest_buf,
+               int render_x0,
+               int render_y0,
+               int render_x1,
+               int render_y1,
+               int dest_rowstride,
+               int dest_channels,
+               gboolean dest_has_alpha,
+               const guchar *src_buf,
+               int src_width,
+               int src_height,
+               int src_rowstride,
+               int src_channels,
+               gboolean src_has_alpha,
+               double scale_x,
+               double scale_y,
+               PixopsInterpType interp_type )
+{
+       PixopsFilter filter;
+       PixopsLineFunc line_func;
+
+#ifdef USE_MMX
+       gboolean found_mmx = pixops_have_mmx();
+#endif
+
+       //g_return_if_fail ( !( dest_channels == 3 && dest_has_alpha ) );
+       //g_return_if_fail ( !( src_channels == 3 && src_has_alpha ) );
+       //g_return_if_fail ( !( src_has_alpha && !dest_has_alpha ) );
+
+       if ( scale_x == 0 || scale_y == 0 )
+               return ;
+
+       if ( interp_type == PIXOPS_INTERP_NEAREST )
+       {
+               pixops_scale_nearest ( dest_buf, render_x0, render_y0, render_x1, render_y1,
+                                      dest_rowstride,
+                                      src_buf, src_width, src_height, src_rowstride,
+                                      scale_x, scale_y );
+               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;
+               else
+#endif
+
+                       line_func = scale_line_22_33;
+       }
+       else
+               line_func = scale_line;
+
+       pixops_process ( dest_buf, render_x0, render_y0, render_x1, render_y1,
+                        dest_rowstride, dest_channels, dest_has_alpha,
+                        src_buf, src_width, src_height, src_rowstride, src_channels,
+                        src_has_alpha, scale_x, scale_y, 0, 0, 0, 0, 0,
+                        &filter, line_func );
+
+       g_free ( filter.x.weights );
+       g_free ( filter.y.weights );
+}
+