Any plans to improve AS3 performance?

[ArmandN]

Lately I've been experimenting with porting my chess engine (SparkChess) from AS3 to Javascript via TypeScript.

The app has been in development for 6 years and its AS3 performance is fully optimized. All important functions were benchmarked and analysed at bytecode level, I discussed performance tips with Adobe engineers and I tested various approaches to find the fastest way.

I was quite surprised to see that the Javascript performance in Chrome is almost double than AS3. It starts slower but after a few iterations, the speed increases dramatically.

I'm not publishing numbers because it's not a 1:1 test yet, but the results so far are consistent.

Of course, Javascript has a number of disadvantages - Canvas-based UI is a kludge, script execution timeout varies across browsers and it's difficult to control (the engine needs time), support for some modern features is limited to newest browsers, code and assets are easy to steal, etc. A JS-based SparkChess would be a step backwards.

Still, I'd love to see some visible speed improvements in AS3. I know VM design is hard, I just hope there's still enough interest at Adobe in this area.

[ArmandN]

The final results are in.

I conducted a full deterministic 1:1 test, making sure that at every step the values are the same between engines.

Javascript in Chrome 42 is 254% faster than Flash Player 14.

My test uses a lot of Vector.<int> reading, bitwise operations and of course recursivity, but there's no complex Math functions involved.

[rewb0rn]

As far as I know, Chrome precompiles Javascript in the background, you will not be able to achieve that speed in ActionScript.

[flashdictionary]

I wish that Adobe could work on improving the performance!

ActionScript 4.0 would have been a great leap forward to improving performance. At this time, I believe there are no plans to improve the existing ActionScript bytecode execution performance in 2015. This assumption is based on the following facts: Adobe's existing backlog of the ActionScript runtime is way too large to focus on improving performance of custom ActionScript bytecode.

I believe that these are some of the other higher priority items on the existing backlog:

• iOS 8 support (this is getting close to finished).
• iOS 64-bit compiler and performance improvements (this is also close to getting finished).
• Still working on finishing up support for PPAPI for Chromium platform
  • Bug#3934097 - Program3D.upload is bottleneck when using chrome PPAPI
• Still working on bugs with Firefox and the NPAPI
• Mobile conflict of StageVideo and Context3D
  • To resolve this major conflict, Adobe has decided to work towards the VideoTexture object. But this work will still take months to finalize.
• Audio/Video Bugs and limitations
  • Feature#3694393 - Add support to NetStream to encode and publish audio using the HE-AAC v2 Audio Cod...
  • Feature#3728399 - Integrate native WebRTC audio/video on all platforms
  • Bug#3944393 - Flash in 16:9 video resolution shows green strip (noise) on right of video window
  • Bug#3928241 - Stream playback doesn't work in Safari for file from local filesystem
  • Bug#3937645 - VideoTexture will not play H.264 on iOS.
• Android video bugs, StageVideo crashing, normal video objects not rendering as expected, hardware acceleration failures.
  • (Hopefully these Android video bugs will naturally be resolved with the VideoTexture improvements.)
  • Feature#3648262 - Add support for Android to transmit RTMP stream with H264 codec.
  • Bug#3943423 - [Android] Some FLV videos don't work with AIR 15, 16 & 17
• Desktop/Android concurrency is not yet worthwhile due to several bugs floating around.
  • Feature#3851496 - New Feature: Add support to Camera and Microphone from within a Worker.
  • Bug#3851617 - Bug with Worker: Event.CHANNEL_MESSAGE is dispatched when there is no message availabl...
• Concurrency on iOS not yet available.
• Hardware Interfacing bugs, such as Camera, Microphone, ServerSocket, SecureSocket
  • https://bugbase.adobe.com/index.cfm?event=selectBug&CFGRIDKEY=3923435
  • Bug#3946519 - Crash on Video.attachCamera() on Alcatel OneTouch C7
• ByteArray security holes and performance limitations
  • Adobe has disabled the avm2.intrinsics.memory package if the ByteArray is set to shared due to security holes
    • Multiple Adobe Flash code execution vulnerabilities
    • Bug#3938269 - CrossBridge applications break after FP version <= 16
  • ByteArray compress and uncompress still needs to be re-factored to restore its performance
  • Bug#3660779 - [Requirement ByteArra size ]ByteArray size is rounded up to the next 24MB boundary
  • Bug#3298228 - [Platform_Windows]ByteArray exponentially slower with simultaneously executing Workers
• Windows 10 support needs attention.
• ActionScript Inline feature was never completed. As a result, the major ActionScript frameworks such as Starling, Away3D are not taking advantage of the Inline feature.
• Still many bugs in the ASC2.0 compiler
  • Bug#3935211 - ActionScript compiler 2.0 is mismatching interface getter for method
  • ActionScript 2.0 compiler bug: You can call super.anyMethod() and it will compile even though anyMet...
• Support for Android 5 needs more attention.
• Still a few bugs left with Adobe Scout