Optimized blur / drop shadow effects for the GL 2 paint engine.

Do the blur in half the resolution in software and then upload the
result as a texture and smooth-scale it on the GPU. This leads to stable
and decent performance regardless of the blur radius, and simplifies the
implementation quite a bit.

Reviewed-by: Bjørn Erik Nilsen

4 files changed, 39 insertions, 12 deletions

diff --git a/src/gui/effects/qgraphicseffect.cpp b/src/gui/effects/qgraphicseffect.cpp
index dafea52..a523bab 100644
--- a/src/gui/effects/qgraphicseffect.cpp
+++ b/src/gui/effects/qgraphicseffect.cpp
@@ -844,22 +844,19 @@ QRectF QGraphicsBlurEffect::boundingRectFor(const QRectF &rect) const
 void QGraphicsBlurEffect::draw(QPainter *painter)
 {
     Q_D(QGraphicsBlurEffect);
-    if (d->filter->radius() <= 0) {
+    if (d->filter->radius() < 1) {
         drawSource(painter);
         return;
     }
 
     PixmapPadMode mode = PadToEffectiveBoundingRect;
     if (painter->paintEngine()->type() == QPaintEngine::OpenGL2)
-        mode = PadToTransparentBorder;
+        mode = NoPad;
 
     // Draw pixmap in device coordinates to avoid pixmap scaling.
     QPoint offset;
-    const QPixmap pixmap = sourcePixmap(Qt::DeviceCoordinates, &offset, mode);
-    QTransform restoreTransform = painter->worldTransform();
-    painter->setWorldTransform(QTransform());
+    QPixmap pixmap = sourcePixmap(Qt::LogicalCoordinates, &offset, mode);
     d->filter->draw(painter, offset, pixmap);
-    painter->setWorldTransform(restoreTransform);
 }
 
 /*!
@@ -1040,7 +1037,7 @@ void QGraphicsDropShadowEffect::draw(QPainter *painter)
 
     PixmapPadMode mode = PadToEffectiveBoundingRect;
     if (painter->paintEngine()->type() == QPaintEngine::OpenGL2)
-        mode = PadToTransparentBorder;
+        mode = NoPad;
 
     // Draw pixmap in device coordinates to avoid pixmap scaling.
     QPoint offset;
diff --git a/src/gui/image/qpixmapfilter.cpp b/src/gui/image/qpixmapfilter.cpp
index 6289efe..9c7c28e 100644
--- a/src/gui/image/qpixmapfilter.cpp
+++ b/src/gui/image/qpixmapfilter.cpp
@@ -627,7 +627,7 @@ void expblur(QImage &img, qreal radius, bool improvedQuality = false, int transp
 {
     // halve the radius if we're using two passes
     if (improvedQuality)
-        radius *= 0.5;
+        radius *= qreal(0.5);
 
     Q_ASSERT(img.format() == QImage::Format_ARGB32_Premultiplied
              || img.format() == QImage::Format_RGB32);
@@ -638,7 +638,7 @@ void expblur(QImage &img, qreal radius, bool improvedQuality = false, int transp
     const qreal cutOffIntensity = 2;
     int alpha = radius <= qreal(1e-5)
         ? ((1 << aprec)-1)
-        : qRound((1<<aprec)*(1 - qPow(cutOffIntensity / 255, 1 / radius)));
+        : qRound((1<<aprec)*(1 - qPow(cutOffIntensity * (1 / qreal(255)), 1 / radius)));
 
     int img_height = img.height();
     for (int row = 0; row < img_height; ++row) {
@@ -873,7 +873,7 @@ Q_GUI_EXPORT void qt_blurImage(QPainter *p, QImage &blurImage, qreal radius, boo
     if (radius >= 4) {
         blurImage = qt_halfScaled(blurImage);
         scale = 2;
-        radius *= 0.5;
+        radius *= qreal(0.5);
     }
 
     if (alphaOnly)
@@ -896,6 +896,8 @@ Q_GUI_EXPORT void qt_blurImage(QImage &blurImage, qreal radius, bool quality, in
         expblur<12, 10, false>(blurImage, radius, quality, transposed);
 }
 
+bool qt_scaleForTransform(const QTransform &transform, qreal *scale);
+
 /*!
     \internal
 */
@@ -914,11 +916,16 @@ void QPixmapBlurFilter::draw(QPainter *painter, const QPointF &p, const QPixmap
         return;
     }
 
+    qreal scaledRadius = radiusScale * d->radius;
+    qreal scale;
+    if (qt_scaleForTransform(painter->transform(), &scale))
+        scaledRadius /= scale;
+
     QPixmapFilter *filter = painter->paintEngine() && painter->paintEngine()->isExtended() ?
         static_cast<QPaintEngineEx *>(painter->paintEngine())->pixmapFilter(type(), this) : 0;
     QPixmapBlurFilter *blurFilter = static_cast<QPixmapBlurFilter*>(filter);
     if (blurFilter) {
-        blurFilter->setRadius(radiusScale * d->radius);
+        blurFilter->setRadius(scaledRadius);
         blurFilter->setBlurHints(d->hints);
         blurFilter->draw(painter, p, src, srcRect);
         return;
@@ -936,7 +943,7 @@ void QPixmapBlurFilter::draw(QPainter *painter, const QPointF &p, const QPixmap
 
     QTransform transform = painter->worldTransform();
     painter->translate(p);
-    qt_blurImage(painter, srcImage, radiusScale * d->radius, (d->hints & QGraphicsBlurEffect::QualityHint), false);
+    qt_blurImage(painter, srcImage, scaledRadius, (d->hints & QGraphicsBlurEffect::QualityHint), false);
     painter->setWorldTransform(transform);
 }
 
diff --git a/src/gui/painting/qmemrotate.cpp b/src/gui/painting/qmemrotate.cpp
index 67dc2c7..e3a6f78 100644
--- a/src/gui/painting/qmemrotate.cpp
+++ b/src/gui/painting/qmemrotate.cpp
@@ -572,5 +572,26 @@ QT_IMPL_MEMROTATE(quint32, qrgb_generic16)
 QT_IMPL_MEMROTATE(quint16, qrgb_generic16)
 #endif
 
+struct qrgb_gl_rgba
+{
+public:
+    inline qrgb_gl_rgba(quint32 v) {
+        if (QSysInfo::ByteOrder == QSysInfo::LittleEndian)
+            data = ((v << 16) & 0xff0000) | ((v >> 16) & 0xff) | (v & 0xff00ff00);
+        else
+            data = (v << 8) | ((v >> 24) & 0xff);
+    }
+
+    inline operator quint32() const { return data; }
+
+private:
+    quint32 data;
+} Q_PACKED;
+
+void Q_GUI_EXPORT qt_memrotate90_gl(const quint32 *src, int srcWidth, int srcHeight, int srcStride,
+                                    quint32 *dest, int dstStride)
+{
+    qt_memrotate90_template(src, srcWidth, srcHeight, srcStride, reinterpret_cast<qrgb_gl_rgba *>(dest), dstStride);
+}
 
 QT_END_NAMESPACE
diff --git a/src/gui/painting/qmemrotate_p.h b/src/gui/painting/qmemrotate_p.h
index 676a880..8aee575 100644
--- a/src/gui/painting/qmemrotate_p.h
+++ b/src/gui/painting/qmemrotate_p.h
@@ -81,6 +81,8 @@ QT_BEGIN_NAMESPACE
     void Q_GUI_QWS_EXPORT qt_memrotate180(const srctype*, int, int, int, desttype*, int); \
     void Q_GUI_QWS_EXPORT qt_memrotate270(const srctype*, int, int, int, desttype*, int)
 
+void Q_GUI_EXPORT qt_memrotate90(const quint32*, int, int, int, quint32*, int);
+
 QT_DECL_MEMROTATE(quint32, quint32);
 QT_DECL_MEMROTATE(quint32, quint16);
 QT_DECL_MEMROTATE(quint16, quint32);