Bugfix #18733 & #18609 (revisited)
authorTon Roosendaal <ton@blender.org>
Mon, 18 May 2009 10:34:26 +0000 (10:34 +0000)
committerTon Roosendaal <ton@blender.org>
Mon, 18 May 2009 10:34:26 +0000 (10:34 +0000)
New imbuf scaling code, advertised as "quick and quality" gave
inacceptable noise and rounding errors.

Check this bugreport for images that show it well:
http://projects.blender.org/tracker/?func=detail&atid=125&aid=18609&group_id=9

For release, better disable this code and fall back on perfectly
working old code. :)

source/blender/imbuf/intern/scaling.c

index b308915cd62a93cdc4c413e34fb3b1a83bd01599..1ccdad05debae32264ff78ad25aa035df935c6a0 100644 (file)
@@ -596,12 +596,12 @@ static void shrink_picture_byte(
        y_counter = 65536;
        for (y_src = 0; y_src < src_height; y_src++) {
                unsigned char* line = src + y_src * 4 * src_width;
-               uintptr_t weight1y = 65536 - (y_dst & 0xffff);
-               uintptr_t weight2y = 65536 - weight1y;
+               uintptr_t weight1y = 65535 - (y_dst & 0xffff);
+               uintptr_t weight2y = 65535 - weight1y;
                x_dst = 0;
                for (x_src = 0; x_src < src_width; x_src++) {
-                       uintptr_t weight1x = 65536 - (x_dst & 0xffff);
-                       uintptr_t weight2x = 65536 - weight1x;
+                       uintptr_t weight1x = 65535 - (x_dst & 0xffff);
+                       uintptr_t weight2x = 65535 - weight1x;
 
                        uintptr_t x = x_dst >> 16;
 
@@ -609,7 +609,7 @@ static void shrink_picture_byte(
 
                        w = (weight1y * weight1x) >> 16;
 
-                       /* ensure correct rounding, without this you get ugly banding (ton) */
+                       /* ensure correct rounding, without this you get ugly banding, or too low color values (ton) */
                        dst_line1[x].r += (line[0] * w + 32767) >> 16;
                        dst_line1[x].g += (line[1] * w + 32767) >> 16;
                        dst_line1[x].b += (line[2] * w + 32767) >> 16;
@@ -647,18 +647,18 @@ static void shrink_picture_byte(
                y_dst += dy_dst;
                y_counter -= dy_dst;
                if (y_counter < 0) {
+                       int val;
                        uintptr_t x;
                        struct scale_outpix_byte * temp;
 
                        y_counter += 65536;
                        
                        for (x=0; x < dst_width; x++) {
-                               uintptr_t f = 0x80000000UL
-                                       / dst_line1[x].weight;
-                               *dst++ = (dst_line1[x].r * f) >> 15;
-                               *dst++ = (dst_line1[x].g * f) >> 15;
-                               *dst++ = (dst_line1[x].b * f) >> 15;
-                               *dst++ = (dst_line1[x].a * f) >> 15;
+                               uintptr_t f =  0x80000000UL / dst_line1[x].weight;
+                               *dst++ = (val= (dst_line1[x].r * f) >> 15) > 255 ? 255: val;
+                               *dst++ = (val= (dst_line1[x].g * f) >> 15) > 255 ? 255: val;
+                               *dst++ = (val= (dst_line1[x].b * f) >> 15) > 255 ? 255: val;
+                               *dst++ = (val= (dst_line1[x].a * f) >> 15) > 255 ? 255: val;
                        }
                        memset(dst_line1, 0, dst_width *
                               sizeof(struct scale_outpix_byte));
@@ -668,13 +668,14 @@ static void shrink_picture_byte(
                }
        }
        if (dst - dst_begin < dst_width * dst_height * 4) {
+               int val;
                uintptr_t x;
                for (x = 0; x < dst_width; x++) {
                        uintptr_t f = 0x80000000UL / dst_line1[x].weight;
-                       *dst++ = (dst_line1[x].r * f) >> 15;
-                       *dst++ = (dst_line1[x].g * f) >> 15;
-                       *dst++ = (dst_line1[x].b * f) >> 15;
-                       *dst++ = (dst_line1[x].a * f) >> 15;
+                       *dst++ = (val= (dst_line1[x].r * f) >> 15) > 255 ? 255: val;
+                       *dst++ = (val= (dst_line1[x].g * f) >> 15) > 255 ? 255: val;
+                       *dst++ = (val= (dst_line1[x].b * f) >> 15) > 255 ? 255: val;
+                       *dst++ = (val= (dst_line1[x].a * f) >> 15) > 255 ? 255: val;
                }
        }
        MEM_freeN(dst_line1);
@@ -912,6 +913,8 @@ static void q_scale_float(float* in, float* out, int in_width,
    Should be comparable in speed to the ImBuf ..._fast functions at least 
    for byte-buffers.
 
+   NOTE: disabled, due to inacceptable inaccuracy and quality loss, see bug #18609 (ton)
+
 */
 static int q_scale_linear_interpolation(
        struct ImBuf *ibuf, int newx, int newy)
@@ -1584,7 +1587,8 @@ struct ImBuf *IMB_scaleImBuf(struct ImBuf * ibuf, short newx, short newy)
        scalefast_Z_ImBuf(ibuf, newx, newy);
 
        /* try to scale common cases in a fast way */
-       if (q_scale_linear_interpolation(ibuf, newx, newy)) {
+       /* disabled, quality loss is inacceptable, see report #18609  (ton) */
+       if (0 && q_scale_linear_interpolation(ibuf, newx, newy)) {
                return ibuf;
        }