CthughaNotes         ChangeLog and porting notes for OGRE 1.7.0 "Cthugha"

Welcome to the ChangeLog and porting notes for OGRE 1.7.0 "Cthugha" in progress

H.P. Lovecraft wrote:
Something was stirring deep within that fiery atmosphere; something monstrous that roared an insatiable anger against the chains of the Elder Gods which had bound it there for an eternity.... Unable to resist, utterly powerless to control his movements, he was diving headlong towards that ravening chaos, that age-old intelligence which was Cthugha.


ChangeLog

  • Compositor changes
    • Allow 'pooled' compositor surfaces.
      • Separate compositor instances using the same size & format surfaces can share them, saving memory
      • System detects the compositor instance chaining to avoid problems with interdependence
      • 'pooled' has to be explicitly enabled on texture definitions, it is not the default because once enabled you cannot necessarily rely on being able to see all the intermediate texture results (because they can be ping-ponging across shared textures); people may be relying on this
    • Manual switching between supported compositor techniques is now possible on the fly
      • Compositor techniques can now have 'scheme' names, you can manually switch between supported techniques using the scheme name to drop to different approaches for reasons other than hardware compatibility (performance, alternative tweaks etc)
      • You can keep & share the textures used by previously active techniques so switching back & forth is faster (must use 'pooled' option for this).
    • Compositors can now specify if they don't want to inherit the FSAA setting from the main target for texture definitions ('no_fsaa')
    • Compositors can now turn on sRGB gamma conversion per texture definition instead of just inheriting from the main surface ('gamma')
    • Cross-compositor communication (from SoC2009 Compositor)
      • Define a texture as accessible from other locations by the chain_scope or global_scope directive in the texture definition
      • Reference a texture from a different compositor in the chain (or in general) using the texture_ref directive
    • Compositor <-> code connection improvements (from SoC2009 Compositor)
      • It is now possible to define a custom composition pass (instead of just quad/scene/clear) in code and trigger it using the render_custom composition pass type.
      • You can now link between a compositor and related code (for example, a compositor listener) automatically using CompositorLogics.
  • SceneManager changes (from SoC2009 Compositor)
    • It is now possible to pause a frame's rendering mid-way (for example, during a callback called in the middle of the process) to trigger a side render and then resume rendering. See SceneManager::_pauseFrame and SceneManager::_resumeFrame.
    • Added an option to manually trigger the updating of shadow textures for specific lights. See SceneManager::prepareShadowTextures. The combination of the two new additions allow reusing a single texture for multiple shadowmaps.
  • Antialising changes
    • Support for Coverage Sampled AA (CSAA) - Dx9 & Dx10 only for now
    • Unified & simplified AA settings
      • on Root's config options the setting is now called 'FSAA' in all cases, and consist of samples and a hint string (separated by spaces)
      • on the miscParams to createRenderWindow you can supply 'FSAA' and 'FSAAHint'. The former is the number of samples, the latter any hinting (e.g. 'Quality')
  • Light changes
    • Near/far plane settings for shadow cameras can now be manually configured per light if required
    • You can now mask lights out per object by calling MovableObject::setLightMask - a renderable object's mask is bitwise and'ed with the light's mask and the light is excluded if the result is 0.
  • LOD changes
    • LOD no longer has to use distance as a metric for changing LOD levels
    • LodStrategy can now be set on both Material and Mesh, to either Distance or PixelCount (new strategies can also be added)
  • STL container changes
    • All STL containers now use custom memory allocators
  • Profiler changes
    • Allow milliseconds as well as percentage view - gives a better idea of absolute fluctuations.
    • Define profiler masks so that profiling can be added to core OGRE but still filtered out by categories (added some simple profiling to test)
    • Hierarchy of profiles is now inclusive instead of exclusive (children no longer subtract their time from parents). This is more useful in practice when doing breakdowns
    • Added numerical indicators instead of a scale with 'ticks' since its more useful
  • Optimisations
    • Fixed-function light state is now handled more intelligently, leading to better performance with large numbers of objects
    • Shader parameters are now updated more selectively, reducing unnecessary updates
  • GpuProgramParameters changes
    • Support added for shared parameter sets which allow you to define / update shader variables for many programs and materials in one place. See GpuProgramManager::createSharedParamerers
    • Use shared_params, shared_param_named and shared_params_ref in scripts to define and reference shared parameter sets.
    • Parameters are now automatically migrated when the program they are based on is changed and reloaded; any parameters which still apply are merged into the new parameters
  • Archive changes
    • Archive now supports create and remove of files (FileSystem only implemented for now)
  • DataStream changes
    • Writeable data streams are now supported (FileSystem only implemented for now)
  • File handling changes
    • New class StreamSerialiser is the new way to read & write binary chunk-based formats
  • Build changes
    • Cmake is now used to generate project files, separate explicitly maintained build systems are being removed. See Building Ogre
  • RenderWindow changes
    • miscParams now supports 'vsyncInterval' option, allowing you to sync to a multiple of the refresh rate if you want (and the hardware supports it)
  • Viewport changes
    • Added clear method to manually clear any combination of colour/depth/stencil to a specified value without performing an update.
  • Image changes
    • Added loadTwoImagesAsRGBA and combineTwoImagesAsRGBA to make it easier to construct combined normal/height and diffuse/specular images etc
  • New Paging Component
    • SceneManager-independent, separate optional component (OGRE_HOME/Components/Paging)
    • Pluggable strategy components to control paging strategy for a section of the scene
    • Pluggable content components to control paging of content
    • Plugggable collection components so that paged elements can be composed or selected between in different ways (e.g. paging many -LOD levels within a page)
  • New Terrain Component
    • SceneManager-independent, separate optional component (OGRE_HOME/Components/Terrain)
    • Inherently editable
    • Hierarchical geometry batching; batch counts reduce at lower LODs as well as vertex count. At the lowest level of detail, the entire terrain page is a single batch.
    • -LOD now adapts in real-time to camera settings (viewport sizes & -LOD bias) so you can use the same terrain with multiple views efficiently
    • Skirts are used instead of stitching to avoid cracks in geometry; this means fewer indexing arrangements & lower overall index buffer usage
    • Saving & loading of terrain built in, including loading / processing in a background thread
    • In-built support for splatting layers, configurable sampler inputs and pluggable material generators
    • Support for generating global normal maps and light maps, in a background thread
  • New Property Component
    • Separate optional component (OGRE_HOME/Components/Property)
    • boost::bind based property system to make it easier to expose reflected properties from your objects
  • Threading changes
    • WorkQueue added to accept generalised work items to be executed in background worker threads
    • WorkQueue starts the number of workers based on hardware, or can be told to start a different number
    • Main Ogre WorkQueue is in Root::getWorkQueue. You can also subclass WorkQueue and provide your own if you want
    • ResourceBackgroundQueue now uses WorkQueue instead of using its own queue
    • All Ogre thread support will use WorkQueue in future
  • Resource changes
    • ResourcePool added as a place for other application components to shelve & re-use resources
  • Material changes
    • scene_blend_op and separate_scene_blend_op added to passes, to change the default '+' operator between source & dest blending params
    • Material listeners can now register to listen to a specific scheme, to allow for none-competing scheme handlers for different schemes.
  • iPhone OS port
    • Synced GLESRenderSystem with GLRenderSystem and added some more extension detection
    • Added support for ARM architecture and CPU feature detection
    • iPhone OGRE SDK disc image and sdk package script
  • OS X improvements
    • Moved code and resources from the Mac directory to OgreMain
    • Eliminated deprecation warnings when targeting Mac OS X 10.5/10.6
    • General organisation and compatibility improvements

Porting notes

  • Build changes
    • Cmake is now used to generate project files, separate explicitly maintained build systems are being removed. See Building Ogre
    • Dependencies have been updated, please see the official release downloads and / or this forum thread
  • Root changes
    • getAvailableRenderers now returns a const RenderSystemList& instead of a RenderSystemList*, in keeping with code standards
  • MovableObject changes
    • Misspelled 'detatchFromParent' fixed to 'detachFromParent'
  • Anti Aliasing changes
    • 'Anti Aliasing' config option in D3D rendersystems renamed to 'FSAA' for consistency
    • Non-maskable AA no longer supported
  • LOD changes
    • Mesh::getLodIndexSquaredDepth and Material::getLodIndexSquaredDepth no longer exist
    • Instead of MeshLodUsage::fromDepthSquared, use MeshLodUsage::userValue which is an UNsquared distance. Each strategy is responsible for transforming the userValue into a value which is to be used at runtime
    • Renamed LodDistanceList to LodValueList and LodDistanceIterator to LodValueIterator. The values are post-transformation by LodStrategy::transformUserValue
    • Material::getLodIndex and Mesh::getLodIndex now require the passed in value to be the transformed value as per the currently assigned strategy. To obtain this, take a userValue (e.g. a distance or pixelcount), and use getLodStrategy()->transformUserValue(userValue) to obtain the lookup value.
    • lod_distances in Material scripts is deprecated (but still supported); use lod_values instead
    • Both the XML and binary mesh formats have been updated to support [[LOD]] strategy and to use generic 'value' entries instead of squared distances. Old formats are still supported but new meshes will use the new format.
  • STL container changes
    • Containers are now defined with custom allocators
    • Your interfaces should change automatically so long as you used the typedefs
    • If not, use vector<T>::type instead of std::vector<T>, and similarly for other types to get the same definition
  • SceneManager changes
    • _populateLightList has an extra parameter, 'lightMask', be sure to add it if you override this method
  • Node changes
    • Node no longer subclasses Renderable; if you need to get at the debug display use getDebugRenderable()
    • Node's API is no longer polluted with Renderable methods
  • SceneNode changes
    • findLights has an extra parameter, 'lightMask', be sure to add it if you override this method
  • RenderSystem changes
    • bindGpuProgramParameters has changed signature to include a 'variabilityMask', which should be used to only update necessary shader values
  • ResourceBackgroundQueue changes
    • RBQ no longer starts its own thread, it uses the centralised WorkQueue in Root::getWorkQueue. If you were previously calling RBQ to change whether it started its own thread, or to process queue items using your own thread, you should now do all this via WorkQueue instead. It may be easiest for you to create a WorkQueue subclass and provide that to Root if you want to customise how threads are managed and to unify Ogre's threading with your own thread management.
  • ScriptCompilerListener changes
    • The ScriptCompilerListener interface has changed from having handleEvent and createObject to just handleEvent. This function now takes a derived class of ScriptCompilerEvent. There are several built-in event types defined in OgreScriptCompiler.h and users with custom script processing may create and throw their own events. Customized resource allocation that used to occur in the createObject function of ScriptCompilerListener is now handled using specific ScriptCompilerEvent classes such as CreateMaterialScriptCompilerEvent.
  • Compositor changes
    • Instead of 'shared' you should used 'pooled' in both scripts and C++ code. The name change was to clarify that it is the surface memory that is re-used, and not the content; since the compositor enhancements now allow texture contents to be referenced across compositors
  • UserDefinedObject changes
    • This class has been deprecated for some time and has now been removed. Instead of subclassing this marker interface you can wrap any of your own classes in an Ogre::Any to associate with Ogre objects via setUserAny.
  • Sample changes
    • Samples now use a unified [[SoC2009_Samples|Sample Browser]]
    • OgreCore.zip has been removed (new core media for the new browser is in SdkTrays.zip)

Annotated roadmap

  • Full DirectX 10 support [delayed to the next release (1.8)]
    • Geometry shaders
    • Constant buffers
    • Compiling Pass state into state objects
    • Texture arrays
    • Stream out
    • Buffer 'views'
    • GL implementations of those that are implemented as extensions
  • Object naming
    • Make names optional wherever possible, no mapping of names if blank
    • Reference count these objects where naming / mapping is optional to preserve auto-shutdown
    • Hash service which detects collisions in debug mode for required naming
  • Skeletal animation bounding box updates
    • One 'update' BB which covers the mesh at every animation point
    • Dynamic AABB which updates based on bone extents, plus padding factor
  • Shader parameters
    • Allow global parameter pools shared by many materials [done]
  • Resource loading
    • Be more tolerant of duplicate resources and allow configurable behavior [done]
  • Separate & generalise concept of debug rendering from isVisible on Camera etc
  • Debug geometry (nodes, cameras etc)
  • Merge remaining (core) Google Summer of Code 2008 projects
    • LOD techniques [done]
    • Motion synthesis

Merge (core) Google Summer of Code 2009 projects

    • Improve Ogre's Compositor Framework [done]
    • Unified Samples Framework & Browser
  • Sampler matching with Cg / HLSL
    Currently you need to use ': register(s0)' etc to match up samplers and texture_unit entries reliably. Allow the option of matching based on texture_unit name instead?
  • Allow parent node to be specified for StaticGeometry / InstancedGeometry sections
    http://www.ogre3d.org/phpBB2/viewtopic.php?p=318372#318372
  • Core paging support [in progress]
    • Loading & unloading only, not coordinate system re-basing
    • New terrain system
  • multi-monitor support for the DirectX 9 render system [done]
  • RT shader generator [in progress]
    • Full fixed function emulation [done]
    • Per-pixel lighting [done]
    • Create a material that demonstrates using bump map lighting model of the RT Shader system [done]
    • Support material script extensions, both in parsing stage and serialization stage [done]
    • Photoshop blending modes
  • Make OGRE compile and run on the iPhone [done]
  • Create CMake build system [in progress]
  • OpenGL ES 1.x render system [done]