Ganesh: images upload to GPU performance fix (skip copying encoded data)

third_party/WebKit/Source/platform/graphics/ImageFrameGenerator.cpp

Index: third_party/WebKit/Source/platform/graphics/ImageFrameGenerator.cpp
diff --git a/third_party/WebKit/Source/platform/graphics/ImageFrameGenerator.cpp b/third_party/WebKit/Source/platform/graphics/ImageFrameGenerator.cpp
index b332fd9d54f678d45843a7c906dc1d3ff6805cb7..ec26d177b71fff524042605d3a1243116de798a7 100644
--- a/third_party/WebKit/Source/platform/graphics/ImageFrameGenerator.cpp
+++ b/third_party/WebKit/Source/platform/graphics/ImageFrameGenerator.cpp
@@ -117,12 +117,23 @@ void ImageFrameGenerator::setData(PassRefPtr<SharedBuffer> data, bool allDataRec
     m_data.setData(data.get(), allDataReceived);
 }
 
-void ImageFrameGenerator::copyData(RefPtr<SharedBuffer>* data, bool* allDataReceived)
+static void sharedBufSkDataReleaseProc(const void* addr, void* ctx)
+{
+    RefPtr<SharedBuffer>* pbuffer = static_cast<RefPtr<SharedBuffer>*>(ctx);
+    ASSERT(pbuffer && (*pbuffer)->data() == addr);
+    delete pbuffer;

[Stephen White, 2015/12/02 02:44:54]

It might be clearer to pass a bare pointer though the callback, and use an
explicit ref() at the call site and derefIfNotNull() here, and forget about
allocating/deleting the RefPtr. Would that work? (I don't normally recommend
using bare ptrs in Blink, but this callback situation is unusual.)

+}
+
+SkData* ImageFrameGenerator::refSkData()
 {
     SharedBuffer* buffer = 0;
-    m_data.data(&buffer, allDataReceived);
-    if (buffer)
-        *data = buffer->copy();
+    bool allDataReceived = false;
+    m_data.data(&buffer, &allDataReceived);

[chrishtr, 2015/12/01 21:56:45]

Please change ThreadSafeDataTransport::data to return a PassRefPtr, and change
the name to something like getData(). That would make this CL easier to read and
more obviously correct.

[Stephen White, 2015/12/02 02:50:39]

Even if we fix that, we still need something to pass through the callback. I've
recommended a bare pointer above, since I find this RefPtr* really odd, but
maybe that's going against your suggestion here.

[aleksandar.stojiljkovic, 2015/12/02 11:04:16]

Thanks, both chrishtr and you have a point here about readability - looks
cleaner now without allocating RefPtr on heap to do deref.

+    if (!allDataReceived)
+        return nullptr;
+
+    RefPtr<SharedBuffer>* pbuffer = new RefPtr<SharedBuffer>(buffer);
+    return SkData::NewWithProc(buffer->data(), buffer->size(), sharedBufSkDataReleaseProc, pbuffer);
 }
 
 bool ImageFrameGenerator::decodeAndScale(const SkImageInfo& info, size_t index, void* pixels, size_t rowBytes)