vo: remove qimage member in qpainter renderer

src/QtAV/QPainterRenderer.h

@@ -35,7 +35,7 @@ public:
     QPainterRenderer();
     bool isSupported(VideoFormat::PixelFormat pixfmt) const Q_DECL_OVERRIDE;
 protected:
-    bool prepareFrame(const VideoFrame& frame);
+    bool preparePixmap(const VideoFrame& frame);
     void drawBackground() Q_DECL_OVERRIDE;
     //draw the current frame using the current paint engine. called by paintEvent()
     void drawFrame() Q_DECL_OVERRIDE;

src/output/video/GraphicsItemRenderer.cpp

@@ -124,7 +124,7 @@ bool GraphicsItemRenderer::receiveFrame(const VideoFrame& frame)
     } else
 #endif
     {
-        prepareFrame(frame);
+        preparePixmap(frame);
     }
     scene()->update(sceneBoundingRect()); //TODO: thread?
     //update(); //does not cause an immediate paint. my not redraw.

src/output/video/QPainterRenderer.cpp

@@ -43,34 +43,30 @@ bool QPainterRenderer::isSupported(VideoFormat::PixelFormat pixfmt) const
     return VideoFormat::imageFormatFromPixelFormat(pixfmt) != QImage::Format_Invalid;
 }
 
-bool QPainterRenderer::prepareFrame(const VideoFrame &frame)
+bool QPainterRenderer::preparePixmap(const VideoFrame &frame)
 {
     DPTR_D(QPainterRenderer);
-    /*
-     * QImage constructed from memory do not deep copy the data, data should be available throughout
-     * image's lifetime and not be modified. painting image happens in main thread, we must ensure the
-     * image data is not changed if not use the original frame size, so we need the lock.
-     * In addition, the data passed by ref, so even if we lock, the original data may changed in decoder
-     * (ImageConverter), so we need a copy of this data(not deep copy) to ensure the image's data not changes before it
-     * is painted.
-     * But if we use the fixed original frame size, the data address and size always the same, so we can
-     * avoid the lock and use the ref data directly and safely
-     */
     // already locked in a larger scope of receive()
-    //If use d.data.data() it will eat more cpu, deep copy?
-    if (frame.constBits(0))
+    QImage::Format imgfmt = frame.imageFormat();
+    if (frame.constBits(0)) {
         d.video_frame = frame;
-    else
-        d.video_frame = frame.to(frame.pixelFormat()); // assume frame format is supported
-    // DO NOT use frameData().data() because it's temp ptr while QImage does not deep copy the data
-    QImage::Format imgfmt = d.video_frame.imageFormat();
+    } else {
+        if (imgfmt == QImage::Format_Invalid) {
+            d.video_frame = frame.to(VideoFormat::Format_RGB32);
+            imgfmt = d.video_frame.imageFormat();
+        } else {
+            d.video_frame = frame.to(frame.pixelFormat());
+        }
+    }
     const bool swapRGB = (int)imgfmt < 0;
     if (swapRGB) {
         imgfmt = (QImage::Format)(-imgfmt);
     }
-    d.image = QImage((uchar*)d.video_frame.constBits(), d.video_frame.width(), d.video_frame.height(), d.video_frame.bytesPerLine(), imgfmt);
+    // DO NOT use frameData().data() because it's temp ptr while d.image does not deep copy the data
+    QImage image = QImage((uchar*)d.video_frame.constBits(), d.video_frame.width(), d.video_frame.height(), d.video_frame.bytesPerLine(), imgfmt);
     if (swapRGB)
-        d.image = d.image.rgbSwapped();
+        image = image.rgbSwapped();
+    d.pixmap = QPixmap::fromImage(image);
     //Format_RGB32 is fast. see document
     return true;
 }
@@ -101,20 +97,17 @@ void QPainterRenderer::drawFrame()
     DPTR_D(QPainterRenderer);
     if (!d.painter)
         return;
-    if (d.image.isNull())
+    if (d.pixmap.isNull())
         return;
     QRect roi = realROI();
     if (orientation() == 0) {
         //assume that the image data is already scaled to out_size(NOT renderer size!)
         if (roi.size() == d.out_rect.size()) {
-            d.painter->drawPixmap(d.out_rect.topLeft(), QPixmap::fromImage(d.image), roi);
-            //d.painter->drawImage(d.out_rect.topLeft(), d.image, roi);
+            d.painter->drawPixmap(d.out_rect.topLeft(), d.pixmap, roi);
         } else {
-            // drawPixmap() is faster for on screen painting
-            d.painter->drawPixmap(d.out_rect, QPixmap::fromImage(d.image), roi);
-            //d.painter->drawImage(d.out_rect, d.image, roi);
+            d.painter->drawPixmap(d.out_rect, d.pixmap, roi);
             //what's the difference?
-            //d.painter->drawImage(QPoint(), image.scaled(d.renderer_width, d.renderer_height));
+            //d.painter->drawPixmap(QPoint(), d.pixmap.scaled(d.renderer_width, d.renderer_height));
         }
         return;
     }
@@ -129,8 +122,7 @@ void QPainterRenderer::drawFrame()
         d.painter->scale((qreal)d.out_rect.width()/(qreal)rendererWidth(), (qreal)d.out_rect.height()/(qreal)rendererHeight());
     d.painter->rotate(orientation());
     d.painter->translate(-rendererWidth()/2, -rendererHeight()/2);
-    d.painter->drawPixmap(QRect(0, 0, rendererWidth(), rendererHeight()), QPixmap::fromImage(d.image), roi);
-    //d.painter->drawImage(QRect(0, 0, rendererWidth(), rendererHeight()), d.image, roi);
+    d.painter->drawPixmap(QRect(0, 0, rendererWidth(), rendererHeight()), d.pixmap, roi);
     d.painter->restore();
 }
 

src/output/video/WidgetRenderer.cpp

@@ -78,7 +78,7 @@ WidgetRenderer::WidgetRenderer(WidgetRendererPrivate &d, QWidget *parent, Qt::Wi
 
 bool WidgetRenderer::receiveFrame(const VideoFrame &frame)
 {
-    prepareFrame(frame);
+    preparePixmap(frame);
     updateUi();
     /*
      * workaround for the widget not updated if has parent. don't know why it works and why update() can't