Add support for encoding/decoding WebP images.

blimp/common/compositor/webp_decoder.cc

Index: blimp/common/compositor/webp_decoder.cc
diff --git a/blimp/common/compositor/webp_decoder.cc b/blimp/common/compositor/webp_decoder.cc
new file mode 100644
index 0000000000000000000000000000000000000000..108f13cc03c191d19351604f7e124b54b68a9366
--- /dev/null
+++ b/blimp/common/compositor/webp_decoder.cc
@@ -0,0 +1,77 @@
+// Copyright 2016 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "blimp/common/compositor/webp_decoder.h"
+
+#include "base/logging.h"
+#include "third_party/libwebp/webp/decode.h"
+#include "third_party/libwebp/webp/demux.h"
+#include "third_party/skia/include/core/SkBitmap.h"
+
+namespace {
+#if SK_B32_SHIFT  // Output little-endian RGBA pixels (Android).
+inline WEBP_CSP_MODE outputMode(bool hasAlpha) {
+  return hasAlpha ? MODE_rgbA : MODE_RGBA;
+}
+#else  // Output little-endian BGRA pixels.
+inline WEBP_CSP_MODE outputMode(bool hasAlpha) {
+  return hasAlpha ? MODE_bgrA : MODE_BGRA;
+}
+#endif
+
+}  // namespace
+
+namespace blimp {
+
+bool WebPDecoder(const void* input, size_t input_size, SkBitmap* bitmap) {
+  // Assuming only one frame.

[urvang, 2016/02/10 23:54:42]

A few choices here, based on the expected behavior:

(1) If you want to support animated WebP (now or in future), I'd suggest using
the WebPAnimDecoder API:
https://code.google.com/p/chromium/codesearch#chromium/src/third_party/libweb...
It's takes care of demuxing + decoding + frame disposal etc all internally.

[Chrome isn't using WebPAnimDecoder right now, because it needs to do a lot more
things: like caching some of the decoded frames, removing them from cache if
short on memory and re-decoding some frames etc.
But I think it will be suitable in your case, as, IIUC, you only  need to decode
all frames in sequence just once].

(2) If you are sure you will get non-animated (one frame) images only, then you
do NOT even need to create a WebPDemuxer object. You can directly decode by
   - WebPGetFeatures() to get width/height and flags
   - WebPINewDecoder() + WebPIUpdate() to decode the single-frame image directly
      [Or even just WebPDecode() if whole image data will always be available.
See comment below].

(3) If you want to decode the first frame only -- for both non-animated and
animated images -- then:
Using demuxer + decoder as you are doing right now would be most efficient. 

[nyquist, 2016/02/11 05:19:23]

Going for (2) for now, using WebPDecode. I'll definitely keep this in mind
though.

+  WebPData inputData = {reinterpret_cast<const uint8_t*>(input), input_size};
+  WebPDemuxState demuxState(WEBP_DEMUX_PARSING_HEADER);
+  WebPDemuxer* demux = WebPDemuxPartial(&inputData, &demuxState);
+
+  uint32_t width = WebPDemuxGetI(demux, WEBP_FF_CANVAS_WIDTH);
+  uint32_t height = WebPDemuxGetI(demux, WEBP_FF_CANVAS_HEIGHT);
+  // ... (Get information about the features present in the WebP file).
+  uint32_t flags = WebPDemuxGetI(demux, WEBP_FF_FORMAT_FLAGS);
+
+  DCHECK(bitmap);
+  bitmap->allocN32Pixels(width, height);
+
+  // TODO(nyquist): Iterate over all frames.
+  WebPIterator iter;
+  bool success = false;
+  if (WebPDemuxGetFrame(demux, 1, &iter)) {
+    // do {
+    //   // ... (Consume 'iter'; e.g. Decode 'iter.fragment' with WebPDecode(),
+    //   // ... and get other frame properties like width, height, offsets etc.
+    //   // ... see 'struct WebPIterator' below for more info).
+    // } while (WebPDemuxNextFrame(&iter));
+    WebPDecBuffer decoderBuffer;
+    WebPIDecoder* decoder = WebPINewDecoder(&decoderBuffer);

[urvang, 2016/02/10 23:54:43]

Note that WebPIDecoder is used for incremental decoding of WebP -- that is, if
you had image data downloading over a network gradually and wanted to decode as
many pixels as possible from data available so far, that's the way to go. In
that case, you basically keep calling WebPIUpdate() as you get more data.

But looking at your code, I think you have all the image data (in 'input') in
the first go itself, right? If so, you can use the simpler WebPDecode() call:
https://code.google.com/p/chromium/codesearch#chromium/src/third_party/libweb...

[nyquist, 2016/02/11 05:19:23]

Thanks! That was for sure simpler.
Updated to use WebPDecode now.

+    WebPInitDecBuffer(&decoderBuffer);
+    WEBP_CSP_MODE mode = outputMode((flags & ALPHA_FLAG) != 0);
+    decoderBuffer.colorspace = mode;
+    decoderBuffer.u.RGBA.stride = width * 4;
+    decoderBuffer.u.RGBA.size = decoderBuffer.u.RGBA.stride * height;
+    decoderBuffer.is_external_memory = 1;
+    decoderBuffer.u.RGBA.rgba =
+        reinterpret_cast<uint8_t*>(bitmap->getAddr32(0, 0));
+
+    success =
+        VP8_STATUS_OK ==
+        WebPIUpdate(decoder, static_cast<const uint8_t*>(input), input_size);
+    // applyPostProcessing(1);
+    WebPIDelete(decoder);
+    WebPDemuxReleaseIterator(&iter);
+  }
+  WebPDemuxDelete(demux);
+
+  // Set the bitmap's opaqueness based on what we saw.
+  bitmap->setAlphaType(flags & ALPHA_FLAG ? kPremul_SkAlphaType
+                                          : kOpaque_SkAlphaType);
+
+  return success;
+}
+
+}  // namespace blimp