Subversion Repository Public Repository


This repository has no backups
This repository's network speed is throttled to 100KB/sec

Entry Size Revision Updated
Framebuffer 710 Fri 20 May, 2016 16:24:40 +0000
PixelBuffer 648 Sun 21 Feb, 2016 16:32:52 +0000
ShaderBuffer 712 Tue 24 May, 2016 16:18:37 +0000
VertexBuffer 712 Tue 24 May, 2016 16:18:37 +0000

Commits for Divide-Framework/trunk/Source Code/Platform/Video/Buffers

Revision Author Commited Message
712 IonutCava picture IonutCava Tue 24 May, 2016 16:18:37 +0000

- Added the concept of “buffer” to be used by GenericVertexData and ShaderBuffer without caring if it’s persistent or not.
— Persistent buffers handle their own locking instead of relying on the parent class
- HiZ occlusion culling fixes and optimizations

710 IonutCava picture IonutCava Fri 20 May, 2016 16:24:40 +0000

- Code cleanup
- Initial work on Scene loading and unloading with and without unloading assets:
— Need to change AIManager from a Singleton to a per-scene component

709 IonutCava picture IonutCava Thu 19 May, 2016 16:21:46 +0000

- Massive rewrite of the resource system:
— ResourceCache now uses a map of weak pointers and passes shared pointers to the call point of CreateResource
— RemoveResource is no longer needed, but resource based smart pointers use a custom deleter that calls unload()
— Keep a shared_ptr of the resource in the entity that needs it and pass shared_ptr references from getter methods
-— All of this removed the need for calls to RemoveResource, REGISTER_TRACKED_DEPENDENCY and UNREGISTER_TRACKED_DEPENDENCY reducing the number of mutex locks and atomic exchanges
- Singleton cleanup:
— Removed ShaderManager singleton and merged it’s data and responsibilities in static methods in both ShaderProgram and Shader classes.
Explanation: Due to the complex interdependency chains in the system, copying, updating, tracking the cache in a thread safe way became a lot more slower than a straight forward smart pointer based implementation. (e.g. scene graph nodes have 3d objects and material instances, 3d objects have materials, materials have textures and shaders, etc)

675 IonutCava picture IonutCava Tue 12 Apr, 2016 14:23:59 +0000

- Create a “reflection target pool” that reflective nodes can use instead of having a reflection render target per node
- Update particles at fixed time intervals instead of each update loop. 33ms for now (30FPS) seems like a decent balance.
- Add a “parallel_for” call that uses the task pool, a function to call, a total count and a partition size
- Frustum culling does not create a new task per node, instead the partition size is now customizable with the new parallel_for system

673 IonutCava picture IonutCava Sun 10 Apr, 2016 19:26:18 +0000

- Use stl_allocator with hashmap
- Various memory allocation tweaks and bug fixes
- Initial Toksvig specular-AA shader code (from

671 IonutCava picture IonutCava Fri 08 Apr, 2016 16:26:49 +0000

- Custom memory allocation system part 1:
— Add
— Add custom allocators to strings and vectors
— Add custom allocators to a few test classes
— No support for aligned memory yet, so vec4<F32> and mat4<F32> vectors use the old allocation system
- Replace more std::async call with our custom task system

648 IonutCava picture IonutCava Sun 21 Feb, 2016 16:32:52 +0000

- Update copyright notice
- Move BoundingBox and BoundingSphere to a new BoundsComponent
— Add a temp hack in SceneGraphNode to update these

631 IonutCava picture IonutCava Sun 24 Jan, 2016 20:28:21 +0000

- Fix and optimize particle rendering system
- Temporarily disable Z_PRE_PASS system for testing
- Improve thread pool destruction system a bit
- Fix texture loading when used as fb attachment
- Forward+ renderer now takes the entire light buffer (no need to split up by light type)
- Start work on occlusion culling fixes:
— ToDo: Take into account batching system
— Fix gl_DrawIDARB usage

629 IonutCava picture IonutCava Thu 21 Jan, 2016 17:08:04 +0000

- New environment mapping system:
— Use previous frame’s list of nodes sorted front to back
— Update reflection for each if they meet the necessary criteria
- Cleanup particle system classes
- Cleanup GenericVertexData class
- Improve argument forwarding to frustum culling tasks

628 IonutCava picture IonutCava Wed 20 Jan, 2016 17:17:53 +0000

- Reworked GenericVertexData ring-buffering system and changed vertex attributes to the vertex format system
— Might be buggy
- Disabled bump-mapping for low-LoD level entities
- Removed a forgotten test line in BRDF shader that prevented lighting calculations to occur properly (an early return)
- Updated auto-reflection system for high shininess materials
- Converted a lot of ‘to_xyz’ calls to ‘to_const_xyz’ calls where appropriate to reduce runtime cost (‘to_const_xyz’ is evaluated at compile time)