I've just completed very similar work on Skia's image encoders (for Android):
***Embedding ICC profiles
***Supporting both versions of unpremultiplication
***Supporting Float16 inputs
***SIMD optimizations for the above

I think it would be great to avoid duplicating the work and just maintain one set of encoders.  This seems like a good time to try to make the switch to Skia.

It looks like platform/image-encoders is unowned, so I guess I'm volunteering to take ownership.  I'm CCing a few of the platform/ OWNERS (not sure who else to add - there hasn't been much recent activity on the encoders).  Please feel free to add anyone that might be interested in this project.

The following revision refers to this bug:
  https://skia.googlesource.com/skia/+/c06f309cf52b885b1b1d98c6b045b120a09f5c54

commit c06f309cf52b885b1b1d98c6b045b120a09f5c54
Author: Leon Scroggins <scroggo@google.com>
Date: Tue May 02 17:08:35 2017

Revert "Add support for row-by-row jpeg encoding"

This reverts commit 9b848d5749c5e34b56f927a3a3374c8ebafbd9db.

Reason for revert: ASAN reports leaked memory [1]. Google3 reports a "delete size mismatch" [2], which I suspect is the same issue.

[1] https://chromium-swarm.appspot.com/task?id=35e2c9fa9eac6310&refresh=10&show_raw=1
[2] https://test.corp.google.com/ui#cl=154838904&flags=CAMQBQ==&id=OCL:154838904:BASE:154839043:1493741642370:9c96115f&t=//chrome/skia/dm_wrapper:dm_wrapper

Original change's description:
> Add support for row-by-row jpeg encoding
> 
> Bug:  713862 
> Change-Id: I787b7c49662a00b89ae0ef35845dfbd6be3e6fb1
> Reviewed-on: https://skia-review.googlesource.com/14641
> Commit-Queue: Matt Sarett <msarett@google.com>
> Reviewed-by: Leon Scroggins <scroggo@google.com>
> 

TBR=msarett@google.com,scroggo@google.com,reed@google.com
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true

Change-Id: Ic5a8d67e0d4a7733662586055ceff086a2ab335d
Reviewed-on: https://skia-review.googlesource.com/15140
Reviewed-by: Leon Scroggins <scroggo@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>

[delete] https://crrev.com/9b848d5749c5e34b56f927a3a3374c8ebafbd9db/src/images/SkJpegEncoder.h
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/gn/tests.gni
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/gm/encode-srgb.cpp
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/src/images/SkImageEncoderPriv.h
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/src/images/SkImageEncoder.cpp
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/src/images/SkJPEGImageEncoder.cpp
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/tests/ReadPixelsTest.cpp
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/tests/CodecTest.cpp
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/src/gpu/SkGr.cpp
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/src/core/SkImageInfoPriv.h
[modify] https://crrev.com/c06f309cf52b885b1b1d98c6b045b120a09f5c54/gm/encode-platform.cpp
[delete] https://crrev.com/9b848d5749c5e34b56f927a3a3374c8ebafbd9db/tests/EncodeTest.cpp

The following revision refers to this bug:
  https://skia.googlesource.com/skia/+/26b44df2333d5e8cec8aef11ab598d63b4ee05c7

commit 26b44df2333d5e8cec8aef11ab598d63b4ee05c7
Author: Matt Sarett <msarett@google.com>
Date: Tue May 02 20:40:10 2017

Add support for row-by-row jpeg encoding

Reland of:
https://skia-review.googlesource.com/c/14641/

Bug:  713862 
Change-Id: I9dca5ede4ebf569c5f80edcfb23a506b6cfa935e
Reviewed-on: https://skia-review.googlesource.com/15144
Reviewed-by: Leon Scroggins <scroggo@google.com>
Commit-Queue: Matt Sarett <msarett@google.com>

[add] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/src/images/SkJpegEncoder.h
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/gn/tests.gni
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/gm/encode-srgb.cpp
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/src/images/SkImageEncoderPriv.h
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/src/images/SkImageEncoder.cpp
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/src/images/SkJPEGImageEncoder.cpp
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/tests/ReadPixelsTest.cpp
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/tests/CodecTest.cpp
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/src/gpu/SkGr.cpp
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/src/core/SkImageInfoPriv.h
[modify] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/gm/encode-platform.cpp
[add] https://crrev.com/26b44df2333d5e8cec8aef11ab598d63b4ee05c7/tests/EncodeTest.cpp

The following revision refers to this bug:
  https://skia.googlesource.com/skia/+/c367d03fb04fccdc63b1379cf74ddc1cab842039

commit c367d03fb04fccdc63b1379cf74ddc1cab842039
Author: Matt Sarett <msarett@google.com>
Date: Fri May 05 16:34:13 2017

Add support for row-by-row png encodes

Also adds a SkEncoder base class.

Bug:  713862 
Change-Id: Ia3f009cd9f376514f6c19396245fab3a43ae6536
Reviewed-on: https://skia-review.googlesource.com/15152
Reviewed-by: Leon Scroggins <scroggo@google.com>
Commit-Queue: Matt Sarett <msarett@google.com>

[add] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/src/images/SkPngEncoder.h
[add] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/src/images/SkEncoder.h
[modify] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/src/images/SkJpegEncoder.h
[modify] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/src/images/SkImageEncoderPriv.h
[modify] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/gm/encode-srgb.cpp
[modify] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/src/images/SkImageEncoder.cpp
[modify] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/src/images/SkJPEGImageEncoder.cpp
[modify] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/tests/CodecTest.cpp
[modify] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/tests/EncodeTest.cpp
[modify] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/gm/encode-platform.cpp
[modify] https://crrev.com/c367d03fb04fccdc63b1379cf74ddc1cab842039/src/images/SkPNGImageEncoder.cpp

The following revision refers to this bug:
  https://skia.googlesource.com/skia/+/be4c9b0a8b9a508cf0de66c0fa401fc1217a32db

commit be4c9b0a8b9a508cf0de66c0fa401fc1217a32db
Author: Matt Sarett <msarett@google.com>
Date: Mon May 08 16:43:44 2017

Add filtering and zlib-level options to SkPngEncoder

Bug:  skia:6409 
Bug:  713862 
Change-Id: If287e2bcad5af990fac11e9091305f45ec903dbf
Reviewed-on: https://skia-review.googlesource.com/15647
Commit-Queue: Matt Sarett <msarett@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>
Reviewed-by: Mike Klein <mtklein@chromium.org>

[modify] https://crrev.com/be4c9b0a8b9a508cf0de66c0fa401fc1217a32db/src/images/SkPngEncoder.h
[modify] https://crrev.com/be4c9b0a8b9a508cf0de66c0fa401fc1217a32db/src/images/SkPngEncoder.cpp
[modify] https://crrev.com/be4c9b0a8b9a508cf0de66c0fa401fc1217a32db/tests/EncodeTest.cpp

The following revision refers to this bug:
  https://skia.googlesource.com/skia/+/94fd06f074016e3ca6a82b88dfdc0ec61d24e67e

commit 94fd06f074016e3ca6a82b88dfdc0ec61d24e67e
Author: Matt Sarett <msarett@google.com>
Date: Tue May 09 17:46:30 2017

Move SkPngEncoder into public API

Bug:  713862 
Change-Id: I45068ed39affe41ffe0f29bf42c5ea1d9b0247ba
Reviewed-on: https://skia-review.googlesource.com/15897
Reviewed-by: Mike Reed <reed@google.com>
Reviewed-by: Mike Klein <mtklein@chromium.org>
Reviewed-by: Leon Scroggins <scroggo@google.com>
Commit-Queue: Matt Sarett <msarett@google.com>

[modify] https://crrev.com/94fd06f074016e3ca6a82b88dfdc0ec61d24e67e/BUILD.gn
[rename] https://crrev.com/94fd06f074016e3ca6a82b88dfdc0ec61d24e67e/include/encode/SkEncoder.h
[rename] https://crrev.com/94fd06f074016e3ca6a82b88dfdc0ec61d24e67e/include/encode/SkPngEncoder.h

The following revision refers to this bug:
  https://skia.googlesource.com/skia/+/fe3190846c9af2cf19a76d7ab9799d7fa69c2369

commit fe3190846c9af2cf19a76d7ab9799d7fa69c2369
Author: Matt Sarett <msarett@google.com>
Date: Tue May 09 18:37:10 2017

Add downsampling option to SkJpegEncoder

Bug:  713862 
Change-Id: Ibad6ecf836ccfd355499e1bace7bcd4ba772a97a
Reviewed-on: https://skia-review.googlesource.com/15891
Commit-Queue: Matt Sarett <msarett@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>

[modify] https://crrev.com/fe3190846c9af2cf19a76d7ab9799d7fa69c2369/src/images/SkJpegEncoder.cpp
[modify] https://crrev.com/fe3190846c9af2cf19a76d7ab9799d7fa69c2369/tests/EncodeTest.cpp
[modify] https://crrev.com/fe3190846c9af2cf19a76d7ab9799d7fa69c2369/src/images/SkJpegEncoder.h

The following revision refers to this bug:
  https://skia.googlesource.com/skia/+/6a4dc66990f73b9443816e69054f101386197593

commit 6a4dc66990f73b9443816e69054f101386197593
Author: Matt Sarett <msarett@google.com>
Date: Thu May 11 14:36:56 2017

Add SkJpegEncoder and SkWebpEncoder to public API

Bug:  713862 
Change-Id: I05d8b3a506f817d63b0bc3ab3f9233e24bbe3be0
Reviewed-on: https://skia-review.googlesource.com/16233
Commit-Queue: Matt Sarett <msarett@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>
Reviewed-by: Mike Reed <reed@google.com>

[rename] https://crrev.com/6a4dc66990f73b9443816e69054f101386197593/include/encode/SkWebpEncoder.h
[modify] https://crrev.com/6a4dc66990f73b9443816e69054f101386197593/src/images/SkPngEncoder.cpp
[modify] https://crrev.com/6a4dc66990f73b9443816e69054f101386197593/src/images/SkJpegEncoder.cpp
[rename] https://crrev.com/6a4dc66990f73b9443816e69054f101386197593/include/encode/SkJpegEncoder.h
[modify] https://crrev.com/6a4dc66990f73b9443816e69054f101386197593/include/encode/SkPngEncoder.h

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8

commit b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8
Author: msarett <msarett@google.com>
Date: Fri May 12 18:42:47 2017

Compile Skia image encoders

BUG= 713862 

Review-Url: https://codereview.chromium.org/2865363002
Cr-Commit-Position: refs/heads/master@{#471382}

[modify] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/content/renderer/gpu/gpu_benchmarking_extension.cc
[modify] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/skia/BUILD.gn
[delete] https://crrev.com/460d2176609a29589b5428b447a8f6b7d0a1fef4/skia/ext/skia_encode_image.cc
[delete] https://crrev.com/460d2176609a29589b5428b447a8f6b7d0a1fef4/skia/ext/skia_encode_image.h
[modify] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/third_party/libwebp/BUILD.gn
[modify] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/third_party/libwebp/README.chromium
[add] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/third_party/libwebp/mux/anim_encode.c
[add] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/third_party/libwebp/mux/animi.h
[add] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/third_party/libwebp/mux/muxedit.c
[add] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/third_party/libwebp/mux/muxi.h
[add] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/third_party/libwebp/mux/muxinternal.c
[add] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/third_party/libwebp/mux/muxread.c
[modify] https://crrev.com/b06d81fdf9fc3a85fe95aa3ef9cd1d7de09758c8/ui/gfx/codec/BUILD.gn
[delete] https://crrev.com/460d2176609a29589b5428b447a8f6b7d0a1fef4/ui/gfx/codec/skia_image_encoder_adapter.cc
[delete] https://crrev.com/460d2176609a29589b5428b447a8f6b7d0a1fef4/ui/gfx/codec/skia_image_encoder_adapter.h

The following revision refers to this bug:
  https://skia.googlesource.com/skia/+/d5a16911bb0da6b03daee6e9dd896429c7f8e8a1

commit d5a16911bb0da6b03daee6e9dd896429c7f8e8a1
Author: Matt Sarett <msarett@google.com>
Date: Wed May 17 02:24:29 2017

Add support for webp lossless compression

Bug:  713862 
Change-Id: I8dcc6506338f3c54fb14a78620e7daaadadfedde
Reviewed-on: https://skia-review.googlesource.com/17073
Reviewed-by: Mike Reed <reed@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>
Commit-Queue: Matt Sarett <msarett@google.com>

[modify] https://crrev.com/d5a16911bb0da6b03daee6e9dd896429c7f8e8a1/include/encode/SkWebpEncoder.h
[modify] https://crrev.com/d5a16911bb0da6b03daee6e9dd896429c7f8e8a1/src/images/SkWebpEncoder.cpp
[modify] https://crrev.com/d5a16911bb0da6b03daee6e9dd896429c7f8e8a1/include/core/SkImageEncoder.h
[modify] https://crrev.com/d5a16911bb0da6b03daee6e9dd896429c7f8e8a1/tests/EncodeTest.cpp
[modify] https://crrev.com/d5a16911bb0da6b03daee6e9dd896429c7f8e8a1/src/images/SkImageEncoder.cpp

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/86074971b9640dbf59869071b4a18bec51c61424

commit 86074971b9640dbf59869071b4a18bec51c61424
Author: msarett <msarett@google.com>
Date: Thu May 18 12:25:27 2017

Use SkJpegEncoder in WebKit platform

This also makes CanvasAsyncBlobCreator use SkPngEncoder, since it is
natural to land jpeg and png together here.

The motivations for using Skia encoders are:
***Support for embedding ICC profiles
***Support for linear unpremultiplication and blending
***Support for multiple color types (including F16)
***Support for multiple alpha types (possible follow-up optimization)

As a long term refactoring goal, we should be able to delete
the WebKit image encoders and the ui/gfx image encoders, thus
leaving one set of image encoders to maintain.

Incidentally, I think this CL is also a nice simplication for
CanvasAsyncBlobCreator.

BUG= 713862 

Review-Url: https://codereview.chromium.org/2878333004
Cr-Commit-Position: refs/heads/master@{#472781}

[modify] https://crrev.com/86074971b9640dbf59869071b4a18bec51c61424/third_party/WebKit/Source/core/html/canvas/CanvasAsyncBlobCreator.cpp
[modify] https://crrev.com/86074971b9640dbf59869071b4a18bec51c61424/third_party/WebKit/Source/core/html/canvas/CanvasAsyncBlobCreator.h
[modify] https://crrev.com/86074971b9640dbf59869071b4a18bec51c61424/third_party/WebKit/Source/core/html/canvas/CanvasAsyncBlobCreatorTest.cpp
[modify] https://crrev.com/86074971b9640dbf59869071b4a18bec51c61424/third_party/WebKit/Source/platform/BUILD.gn
[modify] https://crrev.com/86074971b9640dbf59869071b4a18bec51c61424/third_party/WebKit/Source/platform/graphics/ImageBuffer.cpp
[add] https://crrev.com/86074971b9640dbf59869071b4a18bec51c61424/third_party/WebKit/Source/platform/image-encoders/ImageEncoder.cpp
[add] https://crrev.com/86074971b9640dbf59869071b4a18bec51c61424/third_party/WebKit/Source/platform/image-encoders/ImageEncoder.h
[delete] https://crrev.com/8b69ce5fc7a7e267f2f893df6954c3787d5f2d2f/third_party/WebKit/Source/platform/image-encoders/JPEGImageEncoder.cpp
[delete] https://crrev.com/8b69ce5fc7a7e267f2f893df6954c3787d5f2d2f/third_party/WebKit/Source/platform/image-encoders/JPEGImageEncoder.h
[delete] https://crrev.com/8b69ce5fc7a7e267f2f893df6954c3787d5f2d2f/third_party/WebKit/Source/platform/image-encoders/JPEGImageEncoderTest.cpp
[delete] https://crrev.com/8b69ce5fc7a7e267f2f893df6954c3787d5f2d2f/third_party/WebKit/Source/platform/image-encoders/RGBAtoRGB.h

The following revision refers to this bug:
  https://skia.googlesource.com/skia/+/2f687877835b60497a48537b940cd634088eedf4

commit 2f687877835b60497a48537b940cd634088eedf4
Author: Matt Sarett <msarett@google.com>
Date: Fri May 19 23:35:46 2017

SkWebpEncoder: use bgra for lossless and yuv for lossy

Previosuly, we would (accidentally) always use just yuv.

Bug:  713862 
Change-Id: I00acc6ca2841ba0636494119b7b4f46a9deee401
Reviewed-on: https://skia-review.googlesource.com/17406
Commit-Queue: Matt Sarett <msarett@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>

[modify] https://crrev.com/2f687877835b60497a48537b940cd634088eedf4/tests/EncodeTest.cpp
[modify] https://crrev.com/2f687877835b60497a48537b940cd634088eedf4/src/images/SkWebpEncoder.cpp

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/217d81bc49377c2235b8325a2f03dfce9ef45f95

commit 217d81bc49377c2235b8325a2f03dfce9ef45f95
Author: msarett <msarett@google.com>
Date: Tue May 23 18:19:42 2017

Use SkPngEncoder and SkWebpEncoder in WebKit platform

The motivations for using Skia encoders are:
***Support for embedding ICC profiles
***Support for linear unpremultiplication and blending
***Support for multiple color types (including F16)
***Support for multiple alpha types

As a long term refactoring goal, we should be able to delete
the WebKit image encoders and the ui/gfx image encoders, thus
leaving one set of image encoders to maintain.

BUG= 713862 

Review-Url: https://codereview.chromium.org/2891373002
Cr-Commit-Position: refs/heads/master@{#473985}

[modify] https://crrev.com/217d81bc49377c2235b8325a2f03dfce9ef45f95/third_party/WebKit/Source/core/inspector/InspectorLayerTreeAgent.cpp
[modify] https://crrev.com/217d81bc49377c2235b8325a2f03dfce9ef45f95/third_party/WebKit/Source/platform/BUILD.gn
[modify] https://crrev.com/217d81bc49377c2235b8325a2f03dfce9ef45f95/third_party/WebKit/Source/platform/graphics/ImageBuffer.cpp
[modify] https://crrev.com/217d81bc49377c2235b8325a2f03dfce9ef45f95/third_party/WebKit/Source/platform/graphics/LoggingCanvas.cpp
[modify] https://crrev.com/217d81bc49377c2235b8325a2f03dfce9ef45f95/third_party/WebKit/Source/platform/graphics/PictureSnapshot.cpp
[modify] https://crrev.com/217d81bc49377c2235b8325a2f03dfce9ef45f95/third_party/WebKit/Source/platform/image-encoders/ImageEncoder.cpp
[modify] https://crrev.com/217d81bc49377c2235b8325a2f03dfce9ef45f95/third_party/WebKit/Source/platform/image-encoders/ImageEncoder.h
[delete] https://crrev.com/e635c3e1ed9ed9a0838beba6893b996a63c3e6a1/third_party/WebKit/Source/platform/image-encoders/PNGImageEncoder.cpp
[delete] https://crrev.com/e635c3e1ed9ed9a0838beba6893b996a63c3e6a1/third_party/WebKit/Source/platform/image-encoders/PNGImageEncoder.h
[delete] https://crrev.com/e635c3e1ed9ed9a0838beba6893b996a63c3e6a1/third_party/WebKit/Source/platform/image-encoders/WEBPImageEncoder.cpp
[delete] https://crrev.com/e635c3e1ed9ed9a0838beba6893b996a63c3e6a1/third_party/WebKit/Source/platform/image-encoders/WEBPImageEncoder.h

This is fixed.  The skia encoding APIs are listed here for convenience.
https://cs.chromium.org/chromium/src/third_party/skia/include/encode/SkJpegEncoder.h
https://cs.chromium.org/chromium/src/third_party/skia/include/encode/SkPngEncoder.h
https://cs.chromium.org/chromium/src/third_party/skia/include/encode/SkWebpEncoder.h

When calling into the Skia encoders, a color space will be embedded as long as the SkColorSpace on the input SkPixmap is non-null.  It's worth mentioning that this is currently only supported for color spaces with "numerical" transfer functions.

The unpremultiplication/blending behavior can also be controlled using the SkTransferFunctionBehavior on the various Options structs.  Note that that color correct option (kRespect) is only supported for sRGB and linear transfer functions.

It's also cool to point out that the new encoders support all (valid and logical) SkImageInfos.  So F16, kPremul, kBGRA, etc. are all now supported.  This may particularly useful in the JPEG case, where it appears that we currently unpremultiply before calling the encoder... and then the jpeg encoder redoes the premultiply in order to implement the "kBlendOnToBlack" behavior.  This is not necessary anymore.

Reza, feel free to direct further questions about the API to myself and Leon.

"This may particularly useful in the JPEG case, where it appears that we currently unpremultiply before calling the encoder... 

Not quite following.  The web would break if we did what you suggest we are doing, no?  

Before the changes herein, we would send premultipled pixels to the JPEG image encoder, and encode those pixels.  Encoding premultipled pixels forms a composite-over-black as required by the <canvas> spec.

For the WebGL <canvas>, we receive unpremultiplied pixels on readback from the GPU.  And so we must premultiply those pixels before sending them to the JPEG image encoder, again to meet the composite-over-black spec requirement.

Was any of the above changed during the course of this work?

There was no behavior change.

IIUC, as it is currently written, the pixels sent to the ImageEncoders are unpremultiplied.  Blending unpremultiplied pixels onto black is equivalent to doing a premultiply.  Ex:

SrcOver: s*a + d*(1-a) = s*a + 0*(1-a) = s*a

What I was suggesting is that it is now possible the send premultiplied pixels to the encoder.  Which means that blending onto black is equivalent to doing nothing.  I'm suggesting this as a possible optimization because I imagine that we are doing work on readback in order to unpremultiply the pixels.

> There was no behavior change.

Ah good (there are layout tests covering this).

>IIUC, as it is currently written, the pixels sent to the ImageEncoders are unpremultiplied.

My understanding was that we were always sending premultiplied pixels to the JPEG encoder (see webkit link below).

> What I was suggesting is that it is now possible the send premultiplied pixels to the encoder.  Which means that blending onto black is equivalent to doing nothing.

Good suggestion, but we have been doing the suggested do-nothing optimization to form a composite-over-black for quite some time [1]

[1] https://bugs.webkit.org/attachment.cgi?id=76977&action=prettypatch

> I'm suggesting this as a possible optimization because I imagine that we are doing work on readback in order to unpremultiply the pixels.

It depends on what type of canvas we are reading-back from.

Based on the above link, when we readback from canvas 2d, we receive premultiplied pixels, and we would do the do-nothing optimization: pass the premultiplied pixels directly to the JPEG encoder. Covered by [2] at least.

[2] LayoutTests/canvas/philip/tests/toDataURL.jpeg.alpha.html

However, when we readback from canvas 3d, we receive unpremultiplied pixels.  So we had to premutliply the pixels first, and send the result to the encoder due to the composite over black spec requirement.  Not sure if we have GL tests covering this case?

 Issue 536926  has been merged into this issue.