User:AlexK/Gsoc2013/vse

From BlenderWiki

Scheduling

• We will use OpenCL kernels and cpu threads
  • So, there is cpu and gpu platform memory space
    • So, we can share images between gpu-s in the same platform
    • We would have queue for each memory space.
  • A bucket is the list of instruction which is assign to a single device
    • A bucket is dependent on outputs of other buckets
    • Later we can extend the bucket system to work on tiles instead of whole images
    • By making tiles smaller, we can have fewer frames in processing == more interactive
• Frame ID
  • When we update a strip parameter, we drop the frame associated with it (if we have time)
  • The composition starts all over again
• Scene Strips
  • For now we would make a copy to avoid time jumping problems (we would process them in main thread anyway)

Sequencer

• Images of a strip are stored as cache 2
  • They are associated with clip ad stored in array liked structures
  • We can prefect frames in advanced (videos)
• We make a copy of a scene to avoid update conflicts
• coping is also done for OpenGL preview rendering
• Later, we can make not full copies, and cut corners for faster performance

Plugins

• Plugins will be in OpenCL
  • With python api
  • It can be used for sound
• Using python for math can be too slow
• for C we don't have compiler, plus we need to handle sigsegv

Strips

• Size
  • Each strip will have position, scaling, rotation
    • Internally, they are all modifiers
    • Moreover, users will be able to manipulate the position in a view
    • Selecting a strip by selecting an object in the view should be possible, but it is not a priority
• The cache is used for storing the images. The level 2 cache is associated and is stored in a strip. A image in cache will have have "in use" flag.
• Strip will have offset time
• the current implementation of movie clips is fine
  • closer integration can be possible later

Effects

• Effects
  • One input effects
    • These effects can be used as modifiers for a strips
    • For current modifiers, they can be used as effects strips
    • The effects won't be restricted to original clip (for this we will have modifiers)
    • Also, effects will have a (default) option for a sequence below as an input
  • Other effects
    • Are mainly blends
    • They all can be used as blend mode for a strip
  • Internally, modifiers/effects/blends will be handled the same way by engine

Views

• By design, the engine will support multiple views
• The engine will calculate the target resolution for outputs
• The outputs can be different types: original strip, left/right, waveform, histogram
• The engine will have the list of all target outputs for optimization

Quality

• The quality level can be selected
  • filters can be turned off, or put to the draft quality
• For draft, we will first pre-scale the images to the target output size .
• Each kernel will be compiled as char, half (gpu only), float. The quality settings are applied to the whole sequencer.
• The whole track can be muted (audio and video)

Colorspaces

• Blender will convert images into rgb type color space with OCIO
• All operations will be done in internal color space
• The output image will be converted to display colorspace with OCIO

DNA

• Pretty much will stay the same
• Will be extended for new modifiers/effects
• Position/size might be move to modifier stack