I'm using the test case available at: http://codepen.io/Savago/pen/VPeQaX

Follow a screenshot of a trace collected in a Nexus 4.

Deleted: adler32_277ms_191ms.png 260 KB

	adler32_277ms_191ms.png 260 KB View Download

The trace files.

Deleted: trace_vanilla_798ms.json.gz 187 KB

trace_vanilla_798ms.json.gz 187 KB Download

Deleted: trace_zlib_888ms.json.gz 204 KB

trace_zlib_888ms.json.gz 204 KB Download

Disclaimers: This is only 1 data point (in 1 device) in 1 large PNG file (850KB).

YMMV.

Comparing the CPU self time:
>>> 1 - (188.463/232.320)
0.18877840909090904

Or 18% improvement.

Comparing the 'Wall Duration' we have: 
>>> 1 - (191.821/277.552)
0.30888265982590657

Or 30% improvement. 

It will definitely also vary depending on how the image was encoded (i.e. how many times the checksum will be called and the length of the byte array being checked).

Not mention if the code is running in the big or little core, CPU freq, thermal, behavior of EAS (Energy Aware Scheduler), etc, etc.

Ideally I would like to repeat the test in an ARMv8 device (e.g. Pixel).

Initial patch at:
https://codereview.chromium.org/2676493007/

For reference, using all the 3 patches linked to this issue will yield a performance boost around 40% in PNG image decoding speed (using a Nexus 6).

The image shows the traces for a test page (http://codepen.io/Savago/pen/VPeQaX) where we compare Chromium m59 vanilla X patched (using the 3 optimizations resulting from the PNG investigation: Adler32, inflate_fast, palette).

It is interesting to see that the time spent decoding the image dropped from 116.187ms to 73.844ms (an improvement of around 40%).

It is interesting to see that now GPUImageDecodeCache::DecodeImage() will take longer to execute (94.025ms) than actually decoding the image in ImageFrameGenerator::decode().

Is the image cache compressed? I wonder if we could make it faster?

The Adler32 optimization was submitted to zlib-ng and is waiting for review/merge.

Deleted: all_patches_vanilla.png 287 KB

	all_patches_vanilla.png 287 KB View Download

Update: the optimization was submitted to Canonical zlib on https://github.com/madler/zlib/pull/251

zlib-ng has merged the optimization in their development branch:
https://github.com/Dead2/zlib-ng/commit/ec02ecf104e1d3f1836a908a359f20aa93494df5

Traces collected in a Pixel (Snapdragon 821).

Deleted: adler_traces.png 268 KB

	adler_traces.png 268 KB View Download

All patches combined (Pixel SnapDragon 821).

Deleted: all_patches_trace.png 182 KB

	all_patches_trace.png 182 KB View Download

Updated patch:
https://chromium-review.googlesource.com/c/611492

Merged with:
https://bugs.chromium.org/p/chromium/issues/detail?id=762564

Code landed on:
https://chromium.googlesource.com/chromium/src/+/09b784fd12f255a9da38107ac6e0386f4dde6d68