RGBA32 -> RGB565 dithering in Meego Graphicssystem,

Dithering in QMeeGoPixmapData::imageToEGLSharedImage is now done using
Ordered Dithering, not anymore Floyd-Steinberg dithering.
This provides better overall results.

Fixes: NB#208927 - X dithering looks better than meego graphicssystem one

Merge-request: 1079
Reviewed-by: Samuel Rødal <samuel.rodal@nokia.com>

1 files changed, 52 insertions, 95 deletions

diff --git a/src/plugins/graphicssystems/meego/dithering.cpp b/src/plugins/graphicssystems/meego/dithering.cpp
index 91e3337..2561c22 100644
--- a/src/plugins/graphicssystems/meego/dithering.cpp
+++ b/src/plugins/graphicssystems/meego/dithering.cpp
@@ -39,15 +39,22 @@
 **
 ****************************************************************************/
 
-// This is an implementation of the 32bit => 16bit Floyd-Steinberg dithering.
+// Implements two dithering methods:
+//
+//  * convertRGBA32_to_RGB565
+//
+// This is implemented using Ordered Bayer Dithering. The code has been adapted
+// from QX11PixmapData::fromImage. This method was originally implemented using
+// Floyd-Steinberg dithering but was later changed to Ordered Dithering because
+// of the better quality of the results.
+//
+//  * convertRGBA32_to_RGBA4444
+//
+// This is implemented using Floyd-Steinberg dithering.
+//
 // The alghorithm used here is not the fastest possible but it's prolly fast enough:
 // uses look-up tables, integer-only arthmetics and works in one pass on two lines
 // at a time. It's a high-quality dithering using 1/8 diffusion precission.
-// Two functions here to look at:
-//
-//   * convertRGBA32_to_RGB565
-//   * convertRGBA32_to_RGBA4444
-//
 // Each channel (RGBA) is diffused independently and alpha is dithered too.
 
 #include <string.h>
@@ -76,113 +83,63 @@
 // Converts incoming RGB32 (QImage::Format_RGB32) to RGB565. Returns the newly allocated data.
 unsigned short* convertRGB32_to_RGB565(const unsigned char *in, int width, int height, int stride)
 {
+    static bool thresholdMapInitialized = false;
+    static int thresholdMap[16][16];
+
+    if (!thresholdMapInitialized) {
+        int i;
+        int j;
+        int n;
+
+        thresholdMap[0][0] = 0;
+        thresholdMap[1][0] = 2;
+        thresholdMap[0][1] = 3;
+        thresholdMap[1][1] = 1;
+
+        for (n=2; n<16; n*=2) {
+            for (i=0; i<n; i++) {
+                for (j=0; j<n; j++) {
+                    thresholdMap[i][j]    *= 4;
+                    thresholdMap[i+n][j]   = thresholdMap[i][j] + 2;
+                    thresholdMap[i][j+n]   = thresholdMap[i][j] + 3;
+                    thresholdMap[i+n][j+n] = thresholdMap[i][j] + 1;
+                }
+            }
+        }
+
+        thresholdMapInitialized = true;
+    }
+
     // Output line stride. Aligned to 4 bytes.
     int alignedWidth = width;
     if (alignedWidth % 2 > 0)
         alignedWidth++;
 
     // Will store output
-    unsigned short *out = (unsigned short *) malloc(alignedWidth * height * 2);
-
-    // Lookup tables for the 8bit => 6bit and 8bit => 5bit conversion
-    unsigned char lookup_8bit_to_5bit[256];
-    short lookup_8bit_to_5bit_diff[256];
-    unsigned char lookup_8bit_to_6bit[256];
-    short lookup_8bit_to_6bit_diff[256];
-
-    // Macros for the conversion using the lookup table.
-    #define CONVERT_8BIT_TO_5BIT(v) (lookup_8bit_to_5bit[v])
-    #define DIFF_8BIT_TO_5BIT(v) (lookup_8bit_to_5bit_diff[v])
-
-    #define CONVERT_8BIT_TO_6BIT(v) (lookup_8bit_to_6bit[v])
-    #define DIFF_8BIT_TO_6BIT(v) (lookup_8bit_to_6bit_diff[v])
-
-    int i;
-    int x, y, c; // Pixel we're processing. c is component number (0, 1, 2 for r, b, b)
-    short component[3]; // Stores the new components (r, g, b) for pixel produced during conversion
-    short diff; // The difference between the converted value and the original one. To be accumulated.
-    QVarLengthArray <short> accumulatorData(3 * width * 2); // Data for three acumulators for r, g, b. Each accumulator is two lines.
-    short *accumulator[3]; // Helper for accessing the accumulator on a per-channel basis more easily.
-    accumulator[0] = accumulatorData.data();
-    accumulator[1] = accumulatorData.data() + width;
-    accumulator[2] = accumulatorData.data() + (width * 2);
-
-    // Produce the conversion lookup tables.
-    for (i = 0; i < 256; i++) {
-        lookup_8bit_to_5bit[i] = round(i / 8.0);
-
-        // Before bitshifts: (i * 8) - (... * 8 * 8)
-        lookup_8bit_to_5bit_diff[i] = (i << 3) - (lookup_8bit_to_5bit[i] << 6);
-        if (lookup_8bit_to_5bit[i] > 31)
-            lookup_8bit_to_5bit[i] -= 1;
-
-        lookup_8bit_to_6bit[i] = round(i / 4.0);
-
-        // Before bitshifts: (i * 8) - (... * 4 * 8)
-        lookup_8bit_to_6bit_diff[i] = (i << 3) - (lookup_8bit_to_6bit[i] << 5);
-        if (lookup_8bit_to_6bit[i] > 63)
-            lookup_8bit_to_6bit[i] -= 1;
-    }
+    unsigned short *out = (unsigned short *)malloc (alignedWidth * height * 2);
 
-    // Clear the accumulators
-    memset(accumulator[0], 0, width * 4);
-    memset(accumulator[1], 0, width * 4);
-    memset(accumulator[2], 0, width * 4);
+    int x;
+    int y;
+    int threshold;
 
     // For each line...
     for (y = 0; y < height; y++) {
 
-        // For each accumulator, move the second line (index 1) to replace the first line (index 0).
-        // Clear the second line (index 1)
-        memcpy(accumulator[0], accumulator[0] + width, width * 2);
-        memset(accumulator[0] + width, 0, width * 2);
-
-        memcpy(accumulator[1], accumulator[1] + width, width * 2);
-        memset(accumulator[1] + width, 0, width * 2);
-
-        memcpy(accumulator[2], accumulator[2] + width, width * 2);
-        memset(accumulator[2] + width, 0, width * 2);
-
         // For each column....
         for (x = 0; x < width; x++) {
 
-            // For each component (r, g, b)...
-            for (c = 0; c < 3; c++) {
+            int r = GET_RGBA_COMPONENT(in, x, y, stride, 0);
+            int g = GET_RGBA_COMPONENT(in, x, y, stride, 1);
+            int b = GET_RGBA_COMPONENT(in, x, y, stride, 2);
 
-                // Get the 8bit value from the original image
-                component[c] = GET_RGBA_COMPONENT(in, x, y, stride, c);
-
-                // Add the diffusion for this pixel we stored in the accumulator.
-                // >> 7 because the values in accumulator are stored * 128
-                if (x != 0 && x != (width - 1)) {
-                    if (accumulator[c][x] >> 7 != 0)
-                    component[c] += rand() % accumulator[c][x] >> 7;
-                }
-
-                // Make sure we're not over the boundaries.
-                CLAMP_256(component[c]);
-
-                // For green component we use 6 bits. Otherwise 5 bits.
-                // Store the difference from converting 8bit => 6 bit and the orig pixel.
-                // Convert 8bit => 6(5) bit.
-                if (c == 1) {
-                    diff = DIFF_8BIT_TO_6BIT(component[c]);
-                    component[c] = CONVERT_8BIT_TO_6BIT(component[c]);
-                } else {
-                    diff = DIFF_8BIT_TO_5BIT(component[c]);
-                    component[c] = CONVERT_8BIT_TO_5BIT(component[c]);
-                }
+            threshold = thresholdMap[x%16][y%16];
 
-                // Distribute the difference according to the matrix in the
-                // accumulation bufffer.
-                ACCUMULATE(accumulator[c], x + 1, 0, width, diff * 3);
-                ACCUMULATE(accumulator[c], x - 1, 1, width, diff * 5);
-                ACCUMULATE(accumulator[c], x, 1, width, diff * 5);
-                ACCUMULATE(accumulator[c], x + 1, 1, width, diff * 3);
-            }
+            if (r <= (255-(1<<3)) && ((r<<5) & 255) > threshold) r += (1<<3);
+            if (g <= (255-(1<<2)) && ((g<<6) & 255) > threshold) g += (1<<2);
+            if (b <= (255-(1<<3)) && ((b<<5) & 255) > threshold) b += (1<<3);
 
             // Write the newly produced pixel
-            PUT_565(out, x, y, alignedWidth, component[2], component[1], component[0]);
+            PUT_565(out, x, y, alignedWidth, ((b >> 3) & 0x1f), ((g >> 2) & 0x3f), ((r >> 3) & 0x1f));
         }
     }