Openmax-decoding-of-images

OpenMAX decoding of images

Description

Add OpenMAX IL as a backend for image decoding. Use hardware decoder to directly decode into graphics-memory (zero-copy-decoding).
Additional we need to keep Openmax backend for mm kit in mind, as this is the next logical step in improving performance/features.

Sub tasks

1. Investigate OpenMAX IL API and concept
2. Implement a prove of concept in application space
3. First, small benchmark application to compare implementation with "normal" decoding
4. Integrate implementation into Qt image-loading code
5. Improve benchmark application; Run sophisticated tests
6. Upstream
7. white-paper

Optional sub tasks

This sub tasks might or might not be covered, based on the findings and decision happening.
* Implement Qt-OpenMAX_IL-wrapper
* build/port OpenMAX AL for the rpi
* build/port gst-openmax
This sub tasks are in relation with Openmax backend for mm kit

Logbook

28. February - Thomas Senyk

• Reading OpenMAX IL Specification (and building my first Qt5 for rpi)
• Searching for example code

29. February - Thomas Senyk

• Further searching for example code … found gst-openmax, but gst code isn't great as a reference ;)
• Got some special code underhand :)

1. March - Thomas Senyk

• First look at the handed OpenMAX IL example indicates the following:
  • OpenMAX IL is a micro-controll API (as suspected)
  • OpenMAX IL is a async c api, with callback, state-control, ect. (as suspected)

• The conclusion is that there are 3 options to continue:
  # implement a image-decoder-plugin which directly uses openmax il, don't do much abstraction, just use it.
  # write a proper Qt/c++ API to access basic OMX IL interfaces (components, tunnels, callbacks, ect.). … and based on that, write a image decoder plugin.
  # As there is already a "convenience" API "for" OpenMAX IL: Try to get OpenMAX AL running.
  Option 3 seams to be the most effective and cleanest approach, it's preferable but depends on how easy it is to build/intergrate or of rpi will eventually support it out-of-the-box

2. March - Thomas Senyk

• Further investigation of options
• Started proof-of-concept implementation

13. March - Thomas Senyk

• Finished implementation. It's "running", but it's only producing errors so far.
• The problem now is to understand the behavior, expectation and communication with OpenMAX IL.
• It could take a while to figurer everything out and get something to the screen

14. March - Thomas Senyk

• Got more understanding of the behavior and expectations of OMX IL
• Sorted a few setup things out
• reimplemented the thread-sync code
• got to a state where a got a few specific errors which need so be solved
• THE main error I got is that I can't set: OMX_SetParameter(, OMX_IndexParamPortDefinition, &portSettings) on any of my 2 components (decoder and renderer) … which sucks because I eventually need to set the EGL_Display
  Next thing to do: a lot of cleaning, reading/removing my left-over 'TODO's
• Then: fix problems
• After that: world domination!

15. March - Thomas Senyk

• A lot of cleanup
• Better understanding of what happens when/how/where
• Code seams to be ok
• Had successful run of image-decoding … seams to be very unstable … probably some check/wait/additional property missing …
• couldn't get anything to screen … I hate opengl es 2.0 … need to fix painting, then retry

16. March - Thomas Senyk

• very quickly fixed the painting … was a stupid error
• clean up of the code
• created repository
• it's seams to work now, but texture doesn't change … I need to reboot everytime I start the application … at the 2nd execute (without reboot) the program hangs on OMX_FillThisBuffer(mRenderHandle, eglBuffer); … the 3rd time it can't even init EGL.

I need help to fix this … going to get in contact with the xbmc guys.

20. March - Thomas Senyk

• some communication with xbmc guys … but the tips so far didn't help
• still stuck … no clue how to proceed

21. March - Thomas Senyk

• xbmc guys still silent
• posted on khronos forum: http://www.khronos.org/message_boards/viewtopic.php?f=31&t=4851 … hope that brings some light into the problem