Optimized scaled/transformed image blending for ARGB32PM and RGB16 on RGB16.

Before:
:/traces/qmlphoneconcept.trace, iterations: 5, frames: 48, min(ms):
1207, median(ms): 1212, stddev: 0,165153 %, max(fps): 39,768020

After:
traces/qmlphoneconcept.trace, iterations: 3, frames: 48, min(ms): 884,
median(ms): 886, stddev: 0,383097 %, max(fps): 54,298643

Task-number: QTBUG-6684
Reviewed-by: Gunnar Sletta

7 files changed, 890 insertions, 419 deletions

diff --git a/src/gui/painting/painting.pri b/src/gui/painting/painting.pri
index e61e07c..ed8ee76 100644
--- a/src/gui/painting/painting.pri
+++ b/src/gui/painting/painting.pri
@@ -91,6 +91,8 @@ SOURCES += \
 
         HEADERS +=                                      \
                 painting/qpaintengine_raster_p.h        \
+                painting/qdrawhelper_p.h                \
+                painting/qblendfunctions_p.h            \
                 painting/qrasterdefs_p.h                \
                 painting/qgrayraster_p.h
 
@@ -385,13 +387,13 @@ neon:*-g++* {
     SOURCES += painting/qdrawhelper_neon.cpp
     QMAKE_CXXFLAGS *= -mfpu=neon
 
-    PIXMAN_NEON_ASM_FILES = ../3rdparty/pixman/pixman-arm-neon-asm.S
+    DRAWHELPER_NEON_ASM_FILES = ../3rdparty/pixman/pixman-arm-neon-asm.S painting/qdrawhelper_neon_asm.S
 
     neon_compiler.commands = $$QMAKE_CXX -c
     neon_compiler.commands += $(CXXFLAGS) $(INCPATH) ${QMAKE_FILE_IN} -o ${QMAKE_FILE_OUT}
     neon_compiler.dependency_type = TYPE_C
     neon_compiler.output = ${QMAKE_VAR_OBJECTS_DIR}${QMAKE_FILE_BASE}$${first(QMAKE_EXT_OBJ)}
-    neon_compiler.input = PIXMAN_NEON_ASM_FILES
+    neon_compiler.input = DRAWHELPER_NEON_ASM_FILES
     neon_compiler.variable_out = OBJECTS
     neon_compiler.name = compiling[neon] ${QMAKE_FILE_IN}
     silent:neon_compiler.commands = @echo compiling[neon] ${QMAKE_FILE_IN} && $$neon_compiler.commands
diff --git a/src/gui/painting/qblendfunctions.cpp b/src/gui/painting/qblendfunctions.cpp
index 13bb260..24908ce 100644
--- a/src/gui/painting/qblendfunctions.cpp
+++ b/src/gui/painting/qblendfunctions.cpp
@@ -40,7 +40,7 @@
 ****************************************************************************/
 
 #include <qmath.h>
-#include "qdrawhelper_p.h"
+#include "qblendfunctions_p.h"
 
 QT_BEGIN_NAMESPACE
 
@@ -88,6 +88,8 @@ static inline quint16 convert_argb32_to_rgb16(quint32 spix)
 
 struct Blend_RGB16_on_RGB16_NoAlpha {
     inline void write(quint16 *dst, quint16 src) { *dst = src; }
+
+    inline void flush(void *) {}
 };
 
 struct Blend_RGB16_on_RGB16_ConstAlpha {
@@ -100,6 +102,8 @@ struct Blend_RGB16_on_RGB16_ConstAlpha {
         *dst = BYTE_MUL_RGB16(src, m_alpha) + BYTE_MUL_RGB16(*dst, m_ialpha);
     }
 
+    inline void flush(void *) {}
+
     quint32 m_alpha;
     quint32 m_ialpha;
 };
@@ -114,6 +118,8 @@ struct Blend_ARGB24_on_RGB16_SourceAlpha {
             *dst = s;
         }
     }
+
+    inline void flush(void *) {}
 };
 
 struct Blend_ARGB24_on_RGB16_SourceAndConstAlpha {
@@ -132,6 +138,8 @@ struct Blend_ARGB24_on_RGB16_SourceAndConstAlpha {
         }
     }
 
+    inline void flush(void *) {}
+
     quint32 m_alpha;
 };
 
@@ -145,6 +153,8 @@ struct Blend_ARGB32_on_RGB16_SourceAlpha {
             *dst = s;
         }
     }
+
+    inline void flush(void *) {}
 };
 
 struct Blend_ARGB32_on_RGB16_SourceAndConstAlpha {
@@ -163,99 +173,11 @@ struct Blend_ARGB32_on_RGB16_SourceAndConstAlpha {
         }
     }
 
+    inline void flush(void *) {}
+
     quint32 m_alpha;
 };
 
-template <typename SRC, typename T>
-void qt_scale_image_16bit(uchar *destPixels, int dbpl,
-                          const uchar *srcPixels, int sbpl,
-                          const QRectF &targetRect,
-                          const QRectF &srcRect,
-                          const QRect &clip,
-                          T blender)
-{
-    qreal sx = targetRect.width() / (qreal) srcRect.width();
-    qreal sy = targetRect.height() / (qreal) srcRect.height();
-
-    int ix = 0x00010000 / sx;
-    int iy = 0x00010000 / sy;
-
-//     qDebug() << "scale:" << endl
-//              << " - target" << targetRect << endl
-//              << " - source" << srcRect << endl
-//              << " - clip" << clip << endl
-//              << " - sx=" << sx << " sy=" << sy << " ix=" << ix << " iy=" << iy;
-
-    int cx1 = clip.x();
-    int cx2 = clip.x() + clip.width();
-    int cy1 = clip.top();
-    int cy2 = clip.y() + clip.height();
-
-    int tx1 = qRound(targetRect.left());
-    int tx2 = qRound(targetRect.right());
-    int ty1 = qRound(targetRect.top());
-    int ty2 = qRound(targetRect.bottom());
-
-    if (tx2 < tx1)
-        qSwap(tx2, tx1);
-
-    if (ty2 < ty1)
-        qSwap(ty2, ty1);
-
-    if (tx1 < cx1)
-        tx1 = cx1;
-
-    if (tx2 >= cx2)
-        tx2 = cx2;
-
-    if (tx1 >= tx2)
-        return;
-
-    if (ty1 < cy1)
-        ty1 = cy1;
-
-    if (ty2 >= cy2)
-       ty2 = cy2;
-
-    if (ty1 >= ty2)
-        return;
-
-    int h = ty2 - ty1;
-    int w = tx2 - tx1;
-
-
-    quint32 basex;
-    quint32 srcy;
-
-    if (sx < 0) {
-        int dstx = qFloor((tx1 + qreal(0.5) - targetRect.right()) * ix) + 1;
-        basex = quint32(srcRect.right() * 65536) + dstx;
-    } else {
-        int dstx = qCeil((tx1 + qreal(0.5) - targetRect.left()) * ix) - 1;
-        basex = quint32(srcRect.left() * 65536) + dstx;
-    }
-    if (sy < 0) {
-        int dsty = qFloor((ty1 + qreal(0.5) - targetRect.bottom()) * iy) + 1;
-        srcy = quint32(srcRect.bottom() * 65536) + dsty;
-    } else {
-        int dsty = qCeil((ty1 + qreal(0.5) - targetRect.top()) * iy) - 1;
-        srcy = quint32(srcRect.top() * 65536) + dsty;
-    }
-
-    quint16 *dst = ((quint16 *) (destPixels + ty1 * dbpl)) + tx1;
-
-    while (h--) {
-        const SRC *src = (const SRC *) (srcPixels + (srcy >> 16) * sbpl);
-        int srcx = basex;
-        for (int x=0; x<w; ++x) {
-            blender.write(&dst[x], src[srcx >> 16]);
-            srcx += ix;
-        }
-        dst = (quint16 *)(((uchar *) dst) + dbpl);
-        srcy += iy;
-    }
-}
-
 void qt_scale_image_rgb16_on_rgb16(uchar *destPixels, int dbpl,
                                    const uchar *srcPixels, int sbpl,
                                    const QRectF &targetRect,
@@ -643,6 +565,8 @@ void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
 
 struct Blend_RGB32_on_RGB32_NoAlpha {
     inline void write(quint32 *dst, quint32 src) { *dst = src; }
+
+    inline void flush(void *) {}
 };
 
 struct Blend_RGB32_on_RGB32_ConstAlpha {
@@ -655,6 +579,8 @@ struct Blend_RGB32_on_RGB32_ConstAlpha {
         *dst = BYTE_MUL(src, m_alpha) + BYTE_MUL(*dst, m_ialpha);
     }
 
+    inline void flush(void *) {}
+
     quint32 m_alpha;
     quint32 m_ialpha;
 };
@@ -663,6 +589,8 @@ struct Blend_ARGB32_on_ARGB32_SourceAlpha {
     inline void write(quint32 *dst, quint32 src) {
         *dst = src + BYTE_MUL(*dst, qAlpha(~src));
     }
+
+    inline void flush(void *) {}
 };
 
 struct Blend_ARGB32_on_ARGB32_SourceAndConstAlpha {
@@ -676,98 +604,12 @@ struct Blend_ARGB32_on_ARGB32_SourceAndConstAlpha {
         *dst = src + BYTE_MUL(*dst, qAlpha(~src));
     }
 
+    inline void flush(void *) {}
+
     quint32 m_alpha;
     quint32 m_ialpha;
 };
 
-template <typename T> void qt_scale_image_32bit(uchar *destPixels, int dbpl,
-                                                const uchar *srcPixels, int sbpl,
-                                                const QRectF &targetRect,
-                                                const QRectF &srcRect,
-                                                const QRect &clip,
-                                                T blender)
-{
-    qreal sx = targetRect.width() / (qreal) srcRect.width();
-    qreal sy = targetRect.height() / (qreal) srcRect.height();
-
-    int ix = 0x00010000 / sx;
-    int iy = 0x00010000 / sy;
-
-//     qDebug() << "scale:" << endl
-//              << " - target" << targetRect << endl
-//              << " - source" << srcRect << endl
-//              << " - clip" << clip << endl
-//              << " - sx=" << sx << " sy=" << sy << " ix=" << ix << " iy=" << iy;
-
-    int cx1 = clip.x();
-    int cx2 = clip.x() + clip.width();
-    int cy1 = clip.top();
-    int cy2 = clip.y() + clip.height();
-
-    int tx1 = qRound(targetRect.left());
-    int tx2 = qRound(targetRect.right());
-    int ty1 = qRound(targetRect.top());
-    int ty2 = qRound(targetRect.bottom());
-
-    if (tx2 < tx1)
-        qSwap(tx2, tx1);
-
-    if (ty2 < ty1)
-        qSwap(ty2, ty1);
-
-    if (tx1 < cx1)
-        tx1 = cx1;
-
-    if (tx2 >= cx2)
-        tx2 = cx2;
-
-    if (tx1 >= tx2)
-        return;
-
-    if (ty1 < cy1)
-        ty1 = cy1;
-
-    if (ty2 >= cy2)
-       ty2 = cy2;
-
-    if (ty1 >= ty2)
-        return;
-
-    int h = ty2 - ty1;
-    int w = tx2 - tx1;
-
-    quint32 basex;
-    quint32 srcy;
-
-    if (sx < 0) {
-        int dstx = qFloor((tx1 + qreal(0.5) - targetRect.right()) * ix) + 1;
-        basex = quint32(srcRect.right() * 65536) + dstx;
-    } else {
-        int dstx = qCeil((tx1 + qreal(0.5) - targetRect.left()) * ix) - 1;
-        basex = quint32(srcRect.left() * 65536) + dstx;
-    }
-    if (sy < 0) {
-        int dsty = qFloor((ty1 + qreal(0.5) - targetRect.bottom()) * iy) + 1;
-        srcy = quint32(srcRect.bottom() * 65536) + dsty;
-    } else {
-        int dsty = qCeil((ty1 + qreal(0.5) - targetRect.top()) * iy) - 1;
-        srcy = quint32(srcRect.top() * 65536) + dsty;
-    }
-
-    quint32 *dst = ((quint32 *) (destPixels + ty1 * dbpl)) + tx1;
-
-    while (h--) {
-        const uint *src = (const quint32 *) (srcPixels + (srcy >> 16) * sbpl);
-        int srcx = basex;
-        for (int x=0; x<w; ++x) {
-            blender.write(&dst[x], src[srcx >> 16]);
-            srcx += ix;
-        }
-        dst = (quint32 *)(((uchar *) dst) + dbpl);
-        srcy += iy;
-    }
-}
-
 void qt_scale_image_rgb32_on_rgb32(uchar *destPixels, int dbpl,
                                    const uchar *srcPixels, int sbpl,
                                    const QRectF &targetRect,
@@ -818,244 +660,6 @@ void qt_scale_image_argb32_on_argb32(uchar *destPixels, int dbpl,
     }
 }
 
-struct QTransformImageVertex
-{
-    qreal x, y, u, v; // destination coordinates (x, y) and source coordinates (u, v)
-};
-
-template <class SrcT, class DestT, class Blender>
-void qt_transform_image_rasterize(DestT *destPixels, int dbpl,
-                                  const SrcT *srcPixels, int sbpl,
-                                  const QTransformImageVertex &topLeft, const QTransformImageVertex &bottomLeft,
-                                  const QTransformImageVertex &topRight, const QTransformImageVertex &bottomRight,
-                                  const QRect &sourceRect,
-                                  const QRect &clip,
-                                  qreal topY, qreal bottomY,
-                                  int dudx, int dvdx, int dudy, int dvdy, int u0, int v0,
-                                  Blender blender)
-{
-    int fromY = qMax(qRound(topY), clip.top());
-    int toY = qMin(qRound(bottomY), clip.top() + clip.height());
-    if (fromY >= toY)
-        return;
-
-    qreal leftSlope = (bottomLeft.x - topLeft.x) / (bottomLeft.y - topLeft.y);
-    qreal rightSlope = (bottomRight.x - topRight.x) / (bottomRight.y - topRight.y);
-    int dx_l = int(leftSlope * 0x10000);
-    int dx_r = int(rightSlope * 0x10000);
-    int x_l = int((topLeft.x + (0.5 + fromY - topLeft.y) * leftSlope + 0.5) * 0x10000);
-    int x_r = int((topRight.x + (0.5 + fromY - topRight.y) * rightSlope + 0.5) * 0x10000);
-
-    int fromX, toX, x1, x2, u, v, i, ii;
-    DestT *line;
-    for (int y = fromY; y < toY; ++y) {
-        line = reinterpret_cast<DestT *>(reinterpret_cast<uchar *>(destPixels) + y * dbpl);
-
-        fromX = qMax(x_l >> 16, clip.left());
-        toX = qMin(x_r >> 16, clip.left() + clip.width());
-        if (fromX < toX) {
-            // Because of rounding, we can get source coordinates outside the source image.
-            // Clamp these coordinates to the source rect to avoid segmentation fault and
-            // garbage on the screen.
-
-            // Find the first pixel on the current scan line where the source coordinates are within the source rect.
-            x1 = fromX;
-            u = x1 * dudx + y * dudy + u0;
-            v = x1 * dvdx + y * dvdy + v0;
-            for (; x1 < toX; ++x1) {
-                int uu = u >> 16;
-                int vv = v >> 16;
-                if (uu >= sourceRect.left() && uu < sourceRect.left() + sourceRect.width()
-                    && vv >= sourceRect.top() && vv < sourceRect.top() + sourceRect.height()) {
-                    break;
-                }
-                u += dudx;
-                v += dvdx;
-            }
-
-            // Find the last pixel on the current scan line where the source coordinates are within the source rect.
-            x2 = toX;
-            u = (x2 - 1) * dudx + y * dudy + u0;
-            v = (x2 - 1) * dvdx + y * dvdy + v0;
-            for (; x2 > x1; --x2) {
-                int uu = u >> 16;
-                int vv = v >> 16;
-                if (uu >= sourceRect.left() && uu < sourceRect.left() + sourceRect.width()
-                    && vv >= sourceRect.top() && vv < sourceRect.top() + sourceRect.height()) {
-                    break;
-                }
-                u -= dudx;
-                v -= dvdx;
-            }
-
-            // Set up values at the beginning of the scan line.
-            u = fromX * dudx + y * dudy + u0;
-            v = fromX * dvdx + y * dvdy + v0;
-            line += fromX;
-
-            // Beginning of the scan line, with per-pixel checks.
-            i = x1 - fromX;
-            while (i) {
-                int uu = qBound(sourceRect.left(), u >> 16, sourceRect.left() + sourceRect.width() - 1);
-                int vv = qBound(sourceRect.top(), v >> 16, sourceRect.top() + sourceRect.height() - 1);
-                blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + vv * sbpl)[uu]);
-                u += dudx;
-                v += dvdx;
-                ++line;
-                --i;
-            }
-
-            // Middle of the scan line, without checks.
-            // Manual loop unrolling.
-            i = x2 - x1;
-            ii = i >> 3;
-            while (ii) {
-                blender.write(&line[0], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
-                blender.write(&line[1], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
-                blender.write(&line[2], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
-                blender.write(&line[3], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
-                blender.write(&line[4], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
-                blender.write(&line[5], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
-                blender.write(&line[6], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
-                blender.write(&line[7], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
-                line += 8;
-                --ii;
-            }
-            switch (i & 7) {
-                case 7: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
-                case 6: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
-                case 5: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
-                case 4: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
-                case 3: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
-                case 2: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
-                case 1: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
-            }
-
-            // End of the scan line, with per-pixel checks.
-            i = toX - x2;
-            while (i) {
-                int uu = qBound(sourceRect.left(), u >> 16, sourceRect.left() + sourceRect.width() - 1);
-                int vv = qBound(sourceRect.top(), v >> 16, sourceRect.top() + sourceRect.height() - 1);
-                blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + vv * sbpl)[uu]);
-                u += dudx;
-                v += dvdx;
-                ++line;
-                --i;
-            }
-        }
-        x_l += dx_l;
-        x_r += dx_r;
-    }
-}
-
-template <class SrcT, class DestT, class Blender>
-void qt_transform_image(DestT *destPixels, int dbpl,
-                        const SrcT *srcPixels, int sbpl,
-                        const QRectF &targetRect,
-                        const QRectF &sourceRect,
-                        const QRect &clip,
-                        const QTransform &targetRectTransform,
-                        Blender blender)
-{
-    enum Corner
-    {
-        TopLeft,
-        TopRight,
-        BottomRight,
-        BottomLeft
-    };
-
-    // map source rectangle to destination.
-    QTransformImageVertex v[4];
-    v[TopLeft].u = v[BottomLeft].u = sourceRect.left();
-    v[TopLeft].v = v[TopRight].v = sourceRect.top();
-    v[TopRight].u = v[BottomRight].u = sourceRect.right();
-    v[BottomLeft].v = v[BottomRight].v = sourceRect.bottom();
-    targetRectTransform.map(targetRect.left(), targetRect.top(), &v[TopLeft].x, &v[TopLeft].y);
-    targetRectTransform.map(targetRect.right(), targetRect.top(), &v[TopRight].x, &v[TopRight].y);
-    targetRectTransform.map(targetRect.left(), targetRect.bottom(), &v[BottomLeft].x, &v[BottomLeft].y);
-    targetRectTransform.map(targetRect.right(), targetRect.bottom(), &v[BottomRight].x, &v[BottomRight].y);
-
-    // find topmost vertex.
-    int topmost = 0;
-    for (int i = 1; i < 4; ++i) {
-        if (v[i].y < v[topmost].y)
-            topmost = i;
-    }
-    // rearrange array such that topmost vertex is at index 0.
-    switch (topmost) {
-    case 1:
-        {
-            QTransformImageVertex t = v[0];
-            for (int i = 0; i < 3; ++i)
-                v[i] = v[i+1];
-            v[3] = t;
-        }
-        break;
-    case 2:
-        qSwap(v[0], v[2]);
-        qSwap(v[1], v[3]);
-        break;
-    case 3:
-        {
-            QTransformImageVertex t = v[3];
-            for (int i = 3; i > 0; --i)
-                v[i] = v[i-1];
-            v[0] = t;
-        }
-        break;
-    }
-
-    // if necessary, swap vertex 1 and 3 such that 1 is to the left of 3.
-    qreal dx1 = v[1].x - v[0].x;
-    qreal dy1 = v[1].y - v[0].y;
-    qreal dx2 = v[3].x - v[0].x;
-    qreal dy2 = v[3].y - v[0].y;
-    if (dx1 * dy2 - dx2 * dy1 > 0)
-        qSwap(v[1], v[3]);
-
-    QTransformImageVertex u = {v[1].x - v[0].x, v[1].y - v[0].y, v[1].u - v[0].u, v[1].v - v[0].v};
-    QTransformImageVertex w = {v[2].x - v[0].x, v[2].y - v[0].y, v[2].u - v[0].u, v[2].v - v[0].v};
-
-    qreal det = u.x * w.y - u.y * w.x;
-    if (det == 0)
-        return;
-
-    qreal invDet = 1.0 / det;
-    qreal m11, m12, m21, m22, mdx, mdy;
-
-    m11 = (u.u * w.y - u.y * w.u) * invDet;
-    m12 = (u.x * w.u - u.u * w.x) * invDet;
-    m21 = (u.v * w.y - u.y * w.v) * invDet;
-    m22 = (u.x * w.v - u.v * w.x) * invDet;
-    mdx = v[0].u - m11 * v[0].x - m12 * v[0].y;
-    mdy = v[0].v - m21 * v[0].x - m22 * v[0].y;
-
-    int dudx = int(m11 * 0x10000);
-    int dvdx = int(m21 * 0x10000);
-    int dudy = int(m12 * 0x10000);
-    int dvdy = int(m22 * 0x10000);
-    int u0 = qCeil((0.5 * m11 + 0.5 * m12 + mdx) * 0x10000) - 1;
-    int v0 = qCeil((0.5 * m21 + 0.5 * m22 + mdy) * 0x10000) - 1;
-
-    int x1 = qFloor(sourceRect.left());
-    int y1 = qFloor(sourceRect.top());
-    int x2 = qCeil(sourceRect.right());
-    int y2 = qCeil(sourceRect.bottom());
-    QRect sourceRectI(x1, y1, x2 - x1, y2 - y1);
-
-    // rasterize trapezoids.
-    if (v[1].y < v[3].y) {
-        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[0], v[1], v[0], v[3], sourceRectI, clip, v[0].y, v[1].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
-        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[1], v[2], v[0], v[3], sourceRectI, clip, v[1].y, v[3].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
-        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[1], v[2], v[3], v[2], sourceRectI, clip, v[3].y, v[2].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
-    } else {
-        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[0], v[1], v[0], v[3], sourceRectI, clip, v[0].y, v[3].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
-        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[0], v[1], v[3], v[2], sourceRectI, clip, v[3].y, v[1].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
-        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[1], v[2], v[3], v[2], sourceRectI, clip, v[1].y, v[2].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
-    }
-}
-
 void qt_transform_image_rgb16_on_rgb16(uchar *destPixels, int dbpl,
                                        const uchar *srcPixels, int sbpl,
                                        const QRectF &targetRect,
diff --git a/src/gui/painting/qblendfunctions_p.h b/src/gui/painting/qblendfunctions_p.h
new file mode 100644
index 0000000..ad754b0
--- /dev/null
+++ b/src/gui/painting/qblendfunctions_p.h
@@ -0,0 +1,497 @@
+/****************************************************************************
+**
+** Copyright (C) 2010 Nokia Corporation and/or its subsidiary(-ies).
+** All rights reserved.
+** Contact: Nokia Corporation (qt-info@nokia.com)
+**
+** This file is part of the QtGui module of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** No Commercial Usage
+** This file contains pre-release code and may not be distributed.
+** You may use this file in accordance with the terms and conditions
+** contained in the Technology Preview License Agreement accompanying
+** this package.
+**
+** GNU Lesser General Public License Usage
+** Alternatively, this file may be used under the terms of the GNU Lesser
+** General Public License version 2.1 as published by the Free Software
+** Foundation and appearing in the file LICENSE.LGPL included in the
+** packaging of this file.  Please review the following information to
+** ensure the GNU Lesser General Public License version 2.1 requirements
+** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
+**
+** In addition, as a special exception, Nokia gives you certain additional
+** rights.  These rights are described in the Nokia Qt LGPL Exception
+** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
+**
+** If you have questions regarding the use of this file, please contact
+** Nokia at qt-info@nokia.com.
+**
+**
+**
+**
+**
+**
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#ifndef QBLENDFUNCTIONS_P_H
+#define QBLENDFUNCTIONS_P_H
+
+#include <qmath.h>
+#include "qdrawhelper_p.h"
+
+QT_BEGIN_NAMESPACE
+
+//
+//  W A R N I N G
+//  -------------
+//
+// This file is not part of the Qt API.  It exists purely as an
+// implementation detail.  This header file may change from version to
+// version without notice, or even be removed.
+//
+// We mean it.
+//
+
+template <typename SRC, typename T>
+void qt_scale_image_16bit(uchar *destPixels, int dbpl,
+                          const uchar *srcPixels, int sbpl,
+                          const QRectF &targetRect,
+                          const QRectF &srcRect,
+                          const QRect &clip,
+                          T blender)
+{
+    qreal sx = targetRect.width() / (qreal) srcRect.width();
+    qreal sy = targetRect.height() / (qreal) srcRect.height();
+
+    int ix = 0x00010000 / sx;
+    int iy = 0x00010000 / sy;
+
+//     qDebug() << "scale:" << endl
+//              << " - target" << targetRect << endl
+//              << " - source" << srcRect << endl
+//              << " - clip" << clip << endl
+//              << " - sx=" << sx << " sy=" << sy << " ix=" << ix << " iy=" << iy;
+
+    int cx1 = clip.x();
+    int cx2 = clip.x() + clip.width();
+    int cy1 = clip.top();
+    int cy2 = clip.y() + clip.height();
+
+    int tx1 = qRound(targetRect.left());
+    int tx2 = qRound(targetRect.right());
+    int ty1 = qRound(targetRect.top());
+    int ty2 = qRound(targetRect.bottom());
+
+    if (tx2 < tx1)
+        qSwap(tx2, tx1);
+
+    if (ty2 < ty1)
+        qSwap(ty2, ty1);
+
+    if (tx1 < cx1)
+        tx1 = cx1;
+
+    if (tx2 >= cx2)
+        tx2 = cx2;
+
+    if (tx1 >= tx2)
+        return;
+
+    if (ty1 < cy1)
+        ty1 = cy1;
+
+    if (ty2 >= cy2)
+       ty2 = cy2;
+
+    if (ty1 >= ty2)
+        return;
+
+    int h = ty2 - ty1;
+    int w = tx2 - tx1;
+
+
+    quint32 basex;
+    quint32 srcy;
+
+    if (sx < 0) {
+        int dstx = qFloor((tx1 + qreal(0.5) - targetRect.right()) * ix) + 1;
+        basex = quint32(srcRect.right() * 65536) + dstx;
+    } else {
+        int dstx = qCeil((tx1 + qreal(0.5) - targetRect.left()) * ix) - 1;
+        basex = quint32(srcRect.left() * 65536) + dstx;
+    }
+    if (sy < 0) {
+        int dsty = qFloor((ty1 + qreal(0.5) - targetRect.bottom()) * iy) + 1;
+        srcy = quint32(srcRect.bottom() * 65536) + dsty;
+    } else {
+        int dsty = qCeil((ty1 + qreal(0.5) - targetRect.top()) * iy) - 1;
+        srcy = quint32(srcRect.top() * 65536) + dsty;
+    }
+
+    quint16 *dst = ((quint16 *) (destPixels + ty1 * dbpl)) + tx1;
+
+    while (h--) {
+        const SRC *src = (const SRC *) (srcPixels + (srcy >> 16) * sbpl);
+        int srcx = basex;
+        int x = 0;
+        for (; x<w-7; x+=8) {
+            blender.write(&dst[x], src[srcx >> 16]); srcx += ix;
+            blender.write(&dst[x+1], src[srcx >> 16]); srcx += ix;
+            blender.write(&dst[x+2], src[srcx >> 16]); srcx += ix;
+            blender.write(&dst[x+3], src[srcx >> 16]); srcx += ix;
+            blender.write(&dst[x+4], src[srcx >> 16]); srcx += ix;
+            blender.write(&dst[x+5], src[srcx >> 16]); srcx += ix;
+            blender.write(&dst[x+6], src[srcx >> 16]); srcx += ix;
+            blender.write(&dst[x+7], src[srcx >> 16]); srcx += ix;
+        }
+        for (; x<w; ++x) {
+            blender.write(&dst[x], src[srcx >> 16]);
+            srcx += ix;
+        }
+        blender.flush(&dst[x]);
+        dst = (quint16 *)(((uchar *) dst) + dbpl);
+        srcy += iy;
+    }
+}
+
+template <typename T> void qt_scale_image_32bit(uchar *destPixels, int dbpl,
+                                                const uchar *srcPixels, int sbpl,
+                                                const QRectF &targetRect,
+                                                const QRectF &srcRect,
+                                                const QRect &clip,
+                                                T blender)
+{
+    qreal sx = targetRect.width() / (qreal) srcRect.width();
+    qreal sy = targetRect.height() / (qreal) srcRect.height();
+
+    int ix = 0x00010000 / sx;
+    int iy = 0x00010000 / sy;
+
+//     qDebug() << "scale:" << endl
+//              << " - target" << targetRect << endl
+//              << " - source" << srcRect << endl
+//              << " - clip" << clip << endl
+//              << " - sx=" << sx << " sy=" << sy << " ix=" << ix << " iy=" << iy;
+
+    int cx1 = clip.x();
+    int cx2 = clip.x() + clip.width();
+    int cy1 = clip.top();
+    int cy2 = clip.y() + clip.height();
+
+    int tx1 = qRound(targetRect.left());
+    int tx2 = qRound(targetRect.right());
+    int ty1 = qRound(targetRect.top());
+    int ty2 = qRound(targetRect.bottom());
+
+    if (tx2 < tx1)
+        qSwap(tx2, tx1);
+
+    if (ty2 < ty1)
+        qSwap(ty2, ty1);
+
+    if (tx1 < cx1)
+        tx1 = cx1;
+
+    if (tx2 >= cx2)
+        tx2 = cx2;
+
+    if (tx1 >= tx2)
+        return;
+
+    if (ty1 < cy1)
+        ty1 = cy1;
+
+    if (ty2 >= cy2)
+       ty2 = cy2;
+
+    if (ty1 >= ty2)
+        return;
+
+    int h = ty2 - ty1;
+    int w = tx2 - tx1;
+
+    quint32 basex;
+    quint32 srcy;
+
+    if (sx < 0) {
+        int dstx = qFloor((tx1 + qreal(0.5) - targetRect.right()) * ix) + 1;
+        basex = quint32(srcRect.right() * 65536) + dstx;
+    } else {
+        int dstx = qCeil((tx1 + qreal(0.5) - targetRect.left()) * ix) - 1;
+        basex = quint32(srcRect.left() * 65536) + dstx;
+    }
+    if (sy < 0) {
+        int dsty = qFloor((ty1 + qreal(0.5) - targetRect.bottom()) * iy) + 1;
+        srcy = quint32(srcRect.bottom() * 65536) + dsty;
+    } else {
+        int dsty = qCeil((ty1 + qreal(0.5) - targetRect.top()) * iy) - 1;
+        srcy = quint32(srcRect.top() * 65536) + dsty;
+    }
+
+    quint32 *dst = ((quint32 *) (destPixels + ty1 * dbpl)) + tx1;
+
+    while (h--) {
+        const uint *src = (const quint32 *) (srcPixels + (srcy >> 16) * sbpl);
+        int srcx = basex;
+        int x = 0;
+        for (; x<w; ++x) {
+            blender.write(&dst[x], src[srcx >> 16]);
+            srcx += ix;
+        }
+        blender.flush(&dst[x]);
+        dst = (quint32 *)(((uchar *) dst) + dbpl);
+        srcy += iy;
+    }
+}
+
+struct QTransformImageVertex
+{
+    qreal x, y, u, v; // destination coordinates (x, y) and source coordinates (u, v)
+};
+
+template <class SrcT, class DestT, class Blender>
+void qt_transform_image_rasterize(DestT *destPixels, int dbpl,
+                                  const SrcT *srcPixels, int sbpl,
+                                  const QTransformImageVertex &topLeft, const QTransformImageVertex &bottomLeft,
+                                  const QTransformImageVertex &topRight, const QTransformImageVertex &bottomRight,
+                                  const QRect &sourceRect,
+                                  const QRect &clip,
+                                  qreal topY, qreal bottomY,
+                                  int dudx, int dvdx, int dudy, int dvdy, int u0, int v0,
+                                  Blender blender)
+{
+    int fromY = qMax(qRound(topY), clip.top());
+    int toY = qMin(qRound(bottomY), clip.top() + clip.height());
+    if (fromY >= toY)
+        return;
+
+    qreal leftSlope = (bottomLeft.x - topLeft.x) / (bottomLeft.y - topLeft.y);
+    qreal rightSlope = (bottomRight.x - topRight.x) / (bottomRight.y - topRight.y);
+    int dx_l = int(leftSlope * 0x10000);
+    int dx_r = int(rightSlope * 0x10000);
+    int x_l = int((topLeft.x + (0.5 + fromY - topLeft.y) * leftSlope + 0.5) * 0x10000);
+    int x_r = int((topRight.x + (0.5 + fromY - topRight.y) * rightSlope + 0.5) * 0x10000);
+
+    int fromX, toX, x1, x2, u, v, i, ii;
+    DestT *line;
+    for (int y = fromY; y < toY; ++y) {
+        line = reinterpret_cast<DestT *>(reinterpret_cast<uchar *>(destPixels) + y * dbpl);
+
+        fromX = qMax(x_l >> 16, clip.left());
+        toX = qMin(x_r >> 16, clip.left() + clip.width());
+        if (fromX < toX) {
+            // Because of rounding, we can get source coordinates outside the source image.
+            // Clamp these coordinates to the source rect to avoid segmentation fault and
+            // garbage on the screen.
+
+            // Find the first pixel on the current scan line where the source coordinates are within the source rect.
+            x1 = fromX;
+            u = x1 * dudx + y * dudy + u0;
+            v = x1 * dvdx + y * dvdy + v0;
+            for (; x1 < toX; ++x1) {
+                int uu = u >> 16;
+                int vv = v >> 16;
+                if (uu >= sourceRect.left() && uu < sourceRect.left() + sourceRect.width()
+                    && vv >= sourceRect.top() && vv < sourceRect.top() + sourceRect.height()) {
+                    break;
+                }
+                u += dudx;
+                v += dvdx;
+            }
+
+            // Find the last pixel on the current scan line where the source coordinates are within the source rect.
+            x2 = toX;
+            u = (x2 - 1) * dudx + y * dudy + u0;
+            v = (x2 - 1) * dvdx + y * dvdy + v0;
+            for (; x2 > x1; --x2) {
+                int uu = u >> 16;
+                int vv = v >> 16;
+                if (uu >= sourceRect.left() && uu < sourceRect.left() + sourceRect.width()
+                    && vv >= sourceRect.top() && vv < sourceRect.top() + sourceRect.height()) {
+                    break;
+                }
+                u -= dudx;
+                v -= dvdx;
+            }
+
+            // Set up values at the beginning of the scan line.
+            u = fromX * dudx + y * dudy + u0;
+            v = fromX * dvdx + y * dvdy + v0;
+            line += fromX;
+
+            // Beginning of the scan line, with per-pixel checks.
+            i = x1 - fromX;
+            while (i) {
+                int uu = qBound(sourceRect.left(), u >> 16, sourceRect.left() + sourceRect.width() - 1);
+                int vv = qBound(sourceRect.top(), v >> 16, sourceRect.top() + sourceRect.height() - 1);
+                blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + vv * sbpl)[uu]);
+                u += dudx;
+                v += dvdx;
+                ++line;
+                --i;
+            }
+
+            // Middle of the scan line, without checks.
+            // Manual loop unrolling.
+            i = x2 - x1;
+            ii = i >> 3;
+            while (ii) {
+                blender.write(&line[0], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
+                blender.write(&line[1], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
+                blender.write(&line[2], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
+                blender.write(&line[3], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
+                blender.write(&line[4], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
+                blender.write(&line[5], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
+                blender.write(&line[6], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
+                blender.write(&line[7], reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx;
+
+                line += 8;
+
+                --ii;
+            }
+            switch (i & 7) {
+                case 7: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
+                case 6: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
+                case 5: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
+                case 4: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
+                case 3: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
+                case 2: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
+                case 1: blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + (v >> 16) * sbpl)[u >> 16]); u += dudx; v += dvdx; ++line;
+            }
+
+            // End of the scan line, with per-pixel checks.
+            i = toX - x2;
+            while (i) {
+                int uu = qBound(sourceRect.left(), u >> 16, sourceRect.left() + sourceRect.width() - 1);
+                int vv = qBound(sourceRect.top(), v >> 16, sourceRect.top() + sourceRect.height() - 1);
+                blender.write(line, reinterpret_cast<const SrcT *>(reinterpret_cast<const uchar *>(srcPixels) + vv * sbpl)[uu]);
+                u += dudx;
+                v += dvdx;
+                ++line;
+                --i;
+            }
+
+            blender.flush(line);
+        }
+        x_l += dx_l;
+        x_r += dx_r;
+    }
+}
+
+template <class SrcT, class DestT, class Blender>
+void qt_transform_image(DestT *destPixels, int dbpl,
+                        const SrcT *srcPixels, int sbpl,
+                        const QRectF &targetRect,
+                        const QRectF &sourceRect,
+                        const QRect &clip,
+                        const QTransform &targetRectTransform,
+                        Blender blender)
+{
+    enum Corner
+    {
+        TopLeft,
+        TopRight,
+        BottomRight,
+        BottomLeft
+    };
+
+    // map source rectangle to destination.
+    QTransformImageVertex v[4];
+    v[TopLeft].u = v[BottomLeft].u = sourceRect.left();
+    v[TopLeft].v = v[TopRight].v = sourceRect.top();
+    v[TopRight].u = v[BottomRight].u = sourceRect.right();
+    v[BottomLeft].v = v[BottomRight].v = sourceRect.bottom();
+    targetRectTransform.map(targetRect.left(), targetRect.top(), &v[TopLeft].x, &v[TopLeft].y);
+    targetRectTransform.map(targetRect.right(), targetRect.top(), &v[TopRight].x, &v[TopRight].y);
+    targetRectTransform.map(targetRect.left(), targetRect.bottom(), &v[BottomLeft].x, &v[BottomLeft].y);
+    targetRectTransform.map(targetRect.right(), targetRect.bottom(), &v[BottomRight].x, &v[BottomRight].y);
+
+    // find topmost vertex.
+    int topmost = 0;
+    for (int i = 1; i < 4; ++i) {
+        if (v[i].y < v[topmost].y)
+            topmost = i;
+    }
+    // rearrange array such that topmost vertex is at index 0.
+    switch (topmost) {
+    case 1:
+        {
+            QTransformImageVertex t = v[0];
+            for (int i = 0; i < 3; ++i)
+                v[i] = v[i+1];
+            v[3] = t;
+        }
+        break;
+    case 2:
+        qSwap(v[0], v[2]);
+        qSwap(v[1], v[3]);
+        break;
+    case 3:
+        {
+            QTransformImageVertex t = v[3];
+            for (int i = 3; i > 0; --i)
+                v[i] = v[i-1];
+            v[0] = t;
+        }
+        break;
+    }
+
+    // if necessary, swap vertex 1 and 3 such that 1 is to the left of 3.
+    qreal dx1 = v[1].x - v[0].x;
+    qreal dy1 = v[1].y - v[0].y;
+    qreal dx2 = v[3].x - v[0].x;
+    qreal dy2 = v[3].y - v[0].y;
+    if (dx1 * dy2 - dx2 * dy1 > 0)
+        qSwap(v[1], v[3]);
+
+    QTransformImageVertex u = {v[1].x - v[0].x, v[1].y - v[0].y, v[1].u - v[0].u, v[1].v - v[0].v};
+    QTransformImageVertex w = {v[2].x - v[0].x, v[2].y - v[0].y, v[2].u - v[0].u, v[2].v - v[0].v};
+
+    qreal det = u.x * w.y - u.y * w.x;
+    if (det == 0)
+        return;
+
+    qreal invDet = 1.0 / det;
+    qreal m11, m12, m21, m22, mdx, mdy;
+
+    m11 = (u.u * w.y - u.y * w.u) * invDet;
+    m12 = (u.x * w.u - u.u * w.x) * invDet;
+    m21 = (u.v * w.y - u.y * w.v) * invDet;
+    m22 = (u.x * w.v - u.v * w.x) * invDet;
+    mdx = v[0].u - m11 * v[0].x - m12 * v[0].y;
+    mdy = v[0].v - m21 * v[0].x - m22 * v[0].y;
+
+    int dudx = int(m11 * 0x10000);
+    int dvdx = int(m21 * 0x10000);
+    int dudy = int(m12 * 0x10000);
+    int dvdy = int(m22 * 0x10000);
+    int u0 = qCeil((0.5 * m11 + 0.5 * m12 + mdx) * 0x10000) - 1;
+    int v0 = qCeil((0.5 * m21 + 0.5 * m22 + mdy) * 0x10000) - 1;
+
+    int x1 = qFloor(sourceRect.left());
+    int y1 = qFloor(sourceRect.top());
+    int x2 = qCeil(sourceRect.right());
+    int y2 = qCeil(sourceRect.bottom());
+    QRect sourceRectI(x1, y1, x2 - x1, y2 - y1);
+
+    // rasterize trapezoids.
+    if (v[1].y < v[3].y) {
+        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[0], v[1], v[0], v[3], sourceRectI, clip, v[0].y, v[1].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
+        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[1], v[2], v[0], v[3], sourceRectI, clip, v[1].y, v[3].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
+        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[1], v[2], v[3], v[2], sourceRectI, clip, v[3].y, v[2].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
+    } else {
+        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[0], v[1], v[0], v[3], sourceRectI, clip, v[0].y, v[3].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
+        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[0], v[1], v[3], v[2], sourceRectI, clip, v[3].y, v[1].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
+        qt_transform_image_rasterize(destPixels, dbpl, srcPixels, sbpl, v[1], v[2], v[3], v[2], sourceRectI, clip, v[1].y, v[2].y, dudx, dvdx, dudy, dvdy, u0, v0, blender);
+    }
+}
+
+QT_END_NAMESPACE
+
+#endif // QBLENDFUNCTIONS_P_H
diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index 130270b..dc3b79b 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -7964,6 +7964,12 @@ void qInitDrawhelperAsm()
             qBlendFunctions[QImage::Format_RGB16][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_rgb16_neon;
             qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB16] = qt_blend_rgb16_on_argb32_neon;
 
+            qScaleFunctions[QImage::Format_RGB16][QImage::Format_ARGB32_Premultiplied] = qt_scale_image_argb32_on_rgb16_neon;
+            qScaleFunctions[QImage::Format_RGB16][QImage::Format_RGB16] = qt_scale_image_rgb16_on_rgb16_neon;
+
+            qTransformFunctions[QImage::Format_RGB16][QImage::Format_ARGB32_Premultiplied] = qt_transform_image_argb32_on_rgb16_neon;
+            qTransformFunctions[QImage::Format_RGB16][QImage::Format_RGB16] = qt_transform_image_rgb16_on_rgb16_neon;
+
             qDrawHelper[QImage::Format_RGB16].alphamapBlit = qt_alphamapblit_quint16_neon;
         }
 #endif
diff --git a/src/gui/painting/qdrawhelper_neon.cpp b/src/gui/painting/qdrawhelper_neon.cpp
index ef1b85c..ca1d85f 100644
--- a/src/gui/painting/qdrawhelper_neon.cpp
+++ b/src/gui/painting/qdrawhelper_neon.cpp
@@ -40,6 +40,8 @@
 ****************************************************************************/
 
 #include <private/qdrawhelper_p.h>
+#include <private/qblendfunctions_p.h>
+#include <private/qmath_p.h>
 
 #ifdef QT_HAVE_NEON
 
@@ -310,6 +312,143 @@ void qt_alphamapblit_quint16_neon(QRasterBuffer *rasterBuffer,
     pixman_composite_over_n_8_0565_asm_neon(mapWidth, mapHeight, dest, destStride, color, 0, mask, mapStride);
 }
 
+extern "C" void blend_8_pixels_argb32_on_rgb16_neon(quint16 *dst, const quint32 *src, int const_alpha);
+extern "C" void blend_8_pixels_rgb16_on_rgb16_neon(quint16 *dst, const quint16 *src, int const_alpha);
+
+template <typename SRC, typename BlendFunc>
+struct Blend_on_RGB16_SourceAndConstAlpha_Neon {
+    Blend_on_RGB16_SourceAndConstAlpha_Neon(BlendFunc blender, int const_alpha)
+        : m_index(0)
+        , m_blender(blender)
+        , m_const_alpha(const_alpha)
+    {
+    }
+
+    inline void write(quint16 *dst, quint32 src)
+    {
+        srcBuffer[m_index++] = src;
+
+        if (m_index == 8) {
+            m_blender(dst - 7, srcBuffer, m_const_alpha);
+            m_index = 0;
+        }
+    }
+
+    inline void flush(quint16 *dst)
+    {
+        if (m_index > 0) {
+            quint16 dstBuffer[8];
+            for (int i = 0; i < m_index; ++i)
+                dstBuffer[i] = dst[i - m_index];
+
+            m_blender(dstBuffer, srcBuffer, m_const_alpha);
+
+            for (int i = 0; i < m_index; ++i)
+                dst[i - m_index] = dstBuffer[i];
+
+            m_index = 0;
+        }
+    }
+
+    SRC srcBuffer[8];
+
+    int m_index;
+    BlendFunc m_blender;
+    int m_const_alpha;
+};
+
+template <typename SRC, typename BlendFunc>
+Blend_on_RGB16_SourceAndConstAlpha_Neon<SRC, BlendFunc>
+Blend_on_RGB16_SourceAndConstAlpha_Neon_create(BlendFunc blender, int const_alpha)
+{
+    return Blend_on_RGB16_SourceAndConstAlpha_Neon<SRC, BlendFunc>(blender, const_alpha);
+}
+
+void qt_scale_image_argb32_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                         const uchar *srcPixels, int sbpl,
+                                         const QRectF &targetRect,
+                                         const QRectF &sourceRect,
+                                         const QRect &clip,
+                                         int const_alpha)
+{
+    if (const_alpha == 0)
+        return;
+
+    qt_scale_image_16bit<quint32>(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip,
+        Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint32>(blend_8_pixels_argb32_on_rgb16_neon, const_alpha));
+}
+
+void qt_scale_image_rgb16_on_rgb16(uchar *destPixels, int dbpl,
+                                   const uchar *srcPixels, int sbpl,
+                                   const QRectF &targetRect,
+                                   const QRectF &sourceRect,
+                                   const QRect &clip,
+                                   int const_alpha);
+
+void qt_scale_image_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                        const uchar *srcPixels, int sbpl,
+                                        const QRectF &targetRect,
+                                        const QRectF &sourceRect,
+                                        const QRect &clip,
+                                        int const_alpha)
+{
+    if (const_alpha == 0)
+        return;
+
+    if (const_alpha == 256) {
+        qt_scale_image_rgb16_on_rgb16(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip, const_alpha);
+        return;
+    }
+
+    qt_scale_image_16bit<quint16>(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip,
+        Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint16>(blend_8_pixels_rgb16_on_rgb16_neon, const_alpha));
+}
+
+extern void qt_transform_image_rgb16_on_rgb16(uchar *destPixels, int dbpl,
+                                              const uchar *srcPixels, int sbpl,
+                                              const QRectF &targetRect,
+                                              const QRectF &sourceRect,
+                                              const QRect &clip,
+                                              const QTransform &targetRectTransform,
+                                              int const_alpha);
+
+void qt_transform_image_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                            const uchar *srcPixels, int sbpl,
+                                            const QRectF &targetRect,
+                                            const QRectF &sourceRect,
+                                            const QRect &clip,
+                                            const QTransform &targetRectTransform,
+                                            int const_alpha)
+{
+    if (const_alpha == 0)
+        return;
+
+    if (const_alpha == 256) {
+        qt_transform_image_rgb16_on_rgb16(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip, targetRectTransform, const_alpha);
+        return;
+    }
+
+    qt_transform_image(reinterpret_cast<quint16 *>(destPixels), dbpl,
+                       reinterpret_cast<const quint16 *>(srcPixels), sbpl, targetRect, sourceRect, clip, targetRectTransform,
+        Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint16>(blend_8_pixels_rgb16_on_rgb16_neon, const_alpha));
+}
+
+void qt_transform_image_argb32_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                             const uchar *srcPixels, int sbpl,
+                                             const QRectF &targetRect,
+                                             const QRectF &sourceRect,
+                                             const QRect &clip,
+                                             const QTransform &targetRectTransform,
+                                             int const_alpha)
+{
+    if (const_alpha == 0)
+        return;
+
+    qt_transform_image(reinterpret_cast<quint16 *>(destPixels), dbpl,
+                       reinterpret_cast<const quint32 *>(srcPixels), sbpl, targetRect, sourceRect, clip, targetRectTransform,
+        Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint32>(blend_8_pixels_argb32_on_rgb16_neon, const_alpha));
+}
+
 QT_END_NAMESPACE
 
 #endif // QT_HAVE_NEON
diff --git a/src/gui/painting/qdrawhelper_neon_asm.S b/src/gui/painting/qdrawhelper_neon_asm.S
new file mode 100644
index 0000000..9992817
--- /dev/null
+++ b/src/gui/painting/qdrawhelper_neon_asm.S
@@ -0,0 +1,192 @@
+/****************************************************************************
+**
+** Copyright (C) 2010 Nokia Corporation and/or its subsidiary(-ies).
+** All rights reserved.
+** Contact: Nokia Corporation (qt-info@nokia.com)
+**
+** This file is part of the QtGui module of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** No Commercial Usage
+** This file contains pre-release code and may not be distributed.
+** You may use this file in accordance with the terms and conditions
+** contained in the Technology Preview License Agreement accompanying
+** this package.
+**
+** GNU Lesser General Public License Usage
+** Alternatively, this file may be used under the terms of the GNU Lesser
+** General Public License version 2.1 as published by the Free Software
+** Foundation and appearing in the file LICENSE.LGPL included in the
+** packaging of this file.  Please review the following information to
+** ensure the GNU Lesser General Public License version 2.1 requirements
+** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
+**
+** In addition, as a special exception, Nokia gives you certain additional
+** rights.  These rights are described in the Nokia Qt LGPL Exception
+** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
+**
+** If you have questions regarding the use of this file, please contact
+** Nokia at qt-info@nokia.com.
+**
+**
+**
+**
+**
+**
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+/* Prevent the stack from becoming executable for no reason... */
+#if defined(__linux__) && defined(__ELF__)
+.section .note.GNU-stack,"",%progbits
+#endif
+
+.text
+.fpu neon
+.arch armv7a
+.altmacro
+
+/* void blend_8_pixels_argb32_on_rgb16_neon(quint16 *dst, const quint32 *src, int const_alpha) */
+
+    .func blend_8_pixels_argb32_on_rgb16_neon
+    .global blend_8_pixels_argb32_on_rgb16_neon
+    /* For ELF format also set function visibility to hidden */
+#ifdef __ELF__
+    .hidden blend_8_pixels_argb32_on_rgb16_neon
+    .type blend_8_pixels_argb32_on_rgb16_neon, %function
+#endif
+blend_8_pixels_argb32_on_rgb16_neon:
+    vld4.8      { d0, d1, d2, d3 }, [r1]
+    vld1.16     { d4, d5 }, [r0]
+
+    cmp         r2, #256
+    beq         .blend_32_inner
+
+    vdup.8      d6, r2
+
+    /* multiply by const_alpha */
+    vmull.u8    q8,   d6, d0
+    vmull.u8    q9,   d6, d1
+    vmull.u8    q10,  d6, d2
+    vmull.u8    q11,  d6, d3
+
+    vshrn.u16   d0,  q8, #8
+    vshrn.u16   d1,  q9, #8
+    vshrn.u16   d2, q10, #8
+    vshrn.u16   d3, q11, #8
+
+.blend_32_inner:
+    /* convert 8 r5g6b5 pixel data from {d4, d5} to planar 8-bit format
+       and put data into d6 - red, d7 - green, d30 - blue */
+    vshrn.u16   d6, q2, #8
+    vshrn.u16   d7, q2, #3
+    vsli.u16    q2, q2, #5
+    vsri.u8     d6, d6, #5
+    vmvn.8      d3, d3
+    vsri.u8     d7, d7, #6
+    vshrn.u16   d30, q2, #2
+
+    pld [r0, #128]
+
+    /* now do alpha blending, storing results in 8-bit planar format
+       into d16 - red, d19 - green, d18 - blue */
+    vmull.u8    q10, d3, d6
+    vmull.u8    q11, d3, d7
+    vmull.u8    q12, d3, d30
+    vrshr.u16   q13, q10, #8
+    vrshr.u16   q3,  q11, #8
+    vrshr.u16   q15, q12, #8
+    vraddhn.u16 d20, q10, q13
+    vraddhn.u16 d23, q11, q3
+    vraddhn.u16 d22, q12, q15
+    vqadd.u8    d16, d2, d20
+    vqadd.u8    q9, q0, q11
+    /* convert the result to r5g6b5 and store it into {d28, d29} */
+    vshll.u8    q14, d16, #8
+    vshll.u8    q8, d19, #8
+    vshll.u8    q9, d18, #8
+    vsri.u16    q14, q8, #5
+    vsri.u16    q14, q9, #11
+
+    vst1.16     { d28, d29 }, [r0]
+
+    bx          lr
+
+    .endfunc
+
+/* void blend_8_pixels_rgb16_on_rgb16_neon(quint16 *dst, const quint16 *src, int const_alpha) */
+
+    .func blend_8_pixels_rgb16_on_rgb16_neon
+    .global blend_8_pixels_rgb16_on_rgb16_neon
+    /* For ELF format also set function visibility to hidden */
+#ifdef __ELF__
+    .hidden blend_8_pixels_rgb16_on_rgb16_neon
+    .type blend_8_pixels_rgb16_on_rgb16_neon, %function
+#endif
+blend_8_pixels_rgb16_on_rgb16_neon:
+    vld1.16     { d0, d1 }, [r0]
+    vld1.16     { d2, d3 }, [r1]
+
+    rsb         r3, r2, #256
+    vdup.8      d4, r2
+    vdup.8      d5, r3
+
+    /* convert 8 r5g6b5 pixel data from {d0, d1} to planar 8-bit format
+       and put data into d6 - red, d7 - green, d30 - blue */
+    vshrn.u16   d6,  q0,  #8
+    vshrn.u16   d7,  q0,  #3
+    vsli.u16    q0,  q0,  #5
+    vsri.u8     d6,  d6,  #5
+    vsri.u8     d7,  d7,  #6
+    vshrn.u16   d30, q0,  #2
+
+    /* same from {d2, d3} into {d26, d27, d28} */
+    vshrn.u16   d26, q1,  #8
+    vshrn.u16   d27, q1,  #3
+    vsli.u16    q1,  q1,  #5
+    vsri.u8     d26, d26, #5
+    vsri.u8     d27, d27, #6
+    vshrn.u16   d28, q1,  #2
+
+    /* multiply dst by inv const_alpha */
+    vmull.u8    q10, d5,  d6
+    vmull.u8    q11, d5,  d7
+    vmull.u8    q12, d5,  d30
+
+    vshrn.u16   d6,  q10, #8
+    vshrn.u16   d7,  q11, #8
+    vshrn.u16   d30, q12, #8
+
+    /* multiply src by const_alpha */
+    vmull.u8    q10,  d4, d26
+    vmull.u8    q11,  d4, d27
+    vmull.u8    q12,  d4, d28
+
+    vshrn.u16   d26, q10, #8
+    vshrn.u16   d27, q11, #8
+    vshrn.u16   d28, q12, #8
+
+    /* preload dst + 128 */
+    pld [r0, #128]
+
+    /* add components, storing results in 8-bit planar format
+       into d16 - red, d19 - green, d18 - blue */
+    vadd.u8     d16, d26, d6
+    vadd.u8     d19, d27, d7
+    vadd.u8     d18, d28, d30
+
+    /* convert the result to r5g6b5 and store it into {d28, d29} */
+    vshll.u8    q14, d16, #8
+    vshll.u8    q8,  d19, #8
+    vshll.u8    q9,  d18, #8
+    vsri.u16    q14,  q8, #5
+    vsri.u16    q14,  q9, #11
+
+    vst1.16     { d28, d29 }, [r0]
+
+    bx          lr
+
+    .endfunc
diff --git a/src/gui/painting/qdrawhelper_neon_p.h b/src/gui/painting/qdrawhelper_neon_p.h
index 6f9604f..6f25243 100644
--- a/src/gui/painting/qdrawhelper_neon_p.h
+++ b/src/gui/painting/qdrawhelper_neon_p.h
@@ -84,6 +84,37 @@ void qt_alphamapblit_quint16_neon(QRasterBuffer *rasterBuffer,
                                   const uchar *bitmap,
                                   int mapWidth, int mapHeight, int mapStride,
                                   const QClipData *clip);
+
+void qt_scale_image_argb32_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                         const uchar *srcPixels, int sbpl,
+                                         const QRectF &targetRect,
+                                         const QRectF &sourceRect,
+                                         const QRect &clip,
+                                         int const_alpha);
+
+void qt_scale_image_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                        const uchar *srcPixels, int sbpl,
+                                        const QRectF &targetRect,
+                                        const QRectF &sourceRect,
+                                        const QRect &clip,
+                                        int const_alpha);
+
+void qt_transform_image_argb32_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                             const uchar *srcPixels, int sbpl,
+                                             const QRectF &targetRect,
+                                             const QRectF &sourceRect,
+                                             const QRect &clip,
+                                             const QTransform &targetRectTransform,
+                                             int const_alpha);
+
+void qt_transform_image_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl,
+                                            const uchar *srcPixels, int sbpl,
+                                            const QRectF &targetRect,
+                                            const QRectF &sourceRect,
+                                            const QRect &clip,
+                                            const QTransform &targetRectTransform,
+                                            int const_alpha);
+
 #endif // QT_HAVE_NEON
 
 QT_END_NAMESPACE