Speed up png decoding by eliminating branches from the core loop that fills ImageFrame with pixels

Source/core/platform/image-decoders/png/PNGImageDecoder.cpp

 /*
  * Copyright (C) 2006 Apple Computer, Inc.
  * Copyright (C) Research In Motion Limited 2009-2010. All rights reserved.
  *
  * Portions are Copyright (C) 2001 mozilla.org
  *
  * Other contributors:
  *   Stuart Parmenter <stuart@mozilla.com>
  *
  * This library is free software; you can redistribute it and/or
@@ skipping 471 matching lines @@
 
 #if USE(QCMSLIB)
     if (qcms_transform* transform = m_reader->colorTransform()) {
         qcms_transform_data(transform, row, m_reader->rowBuffer(), size().width());
         row = m_reader->rowBuffer();
     }
 #endif
 
     // Write the decoded row pixels to the frame buffer.
     ImageFrame::PixelData* address = buffer.getAddr(0, y);
-    bool nonTrivialAlpha = false;
+    unsigned alphaMask = 255;
     int width = size().width();
 
+    // Do not merge the loops below. They are hand rolled for each case for maximal performance.
     png_bytep pixel = row;
-    for (int x = 0; x < width; ++x, pixel += colorChannels) {
-        unsigned alpha = hasAlpha ? pixel[3] : 255;
-        buffer.setRGBA(address++, pixel[0], pixel[1], pixel[2], alpha);
-        nonTrivialAlpha |= alpha < 255;
+    if (hasAlpha) {
+        if (buffer.premultiplyAlpha()) {
+            for (int x = 0; x < width; ++x, pixel += 4) {
+                buffer.setRGBAPremultiply(address + x, pixel[0], pixel[1], pixel[2], pixel[3]);

[Noel Gordon, 2013/06/26 05:15:26]

address++ not working for you?  Why address + x or is there some measured
benefit (I don't see any mention of it in the change log).

+                alphaMask &= pixel[3];
+            }
+        } else {
+            for (int x = 0; x < width; ++x, pixel += 4) {
+                buffer.setRGBARaw(address + x, pixel[0], pixel[1], pixel[2], pixel[3]);
+                alphaMask &= pixel[3];
+            }
+        }
+    } else {
+        for (int x = 0; x < width; ++x, pixel += 3)
+            buffer.setRGB(address + x, pixel[0], pixel[1], pixel[2]);
     }
 
-    if (nonTrivialAlpha && !buffer.hasAlpha())
-        buffer.setHasAlpha(nonTrivialAlpha);
+    bool hasNonTrivialAlpha = alphaMask != 255;
+    if (!buffer.hasAlpha())
+        buffer.setHasAlpha(hasNonTrivialAlpha);
 }
 
 void PNGImageDecoder::pngComplete()
 {
     if (!m_frameBufferCache.isEmpty())
         m_frameBufferCache.first().setStatus(ImageFrame::FrameComplete);
 }
 
 void PNGImageDecoder::decode(bool onlySize)
 {
     if (failed())
         return;
 
     if (!m_reader)
         m_reader = adoptPtr(new PNGImageReader(this));
 
     // If we couldn't decode the image but we've received all the data, decoding
     // has failed.
     if (!m_reader->decode(*m_data, onlySize) && isAllDataReceived())
         setFailed();
     // If we're done decoding the image, we don't need the PNGImageReader
     // anymore.  (If we failed, |m_reader| has already been cleared.)
     else if (isComplete())
         m_reader.clear();
 }
 
 } // namespace WebCore