All draw calls now use shaders instead of the fixed-function pipeline. I need to find a better way to organize the built-in shaders since right now the application has to manage them explicitly; I'm leaning towards the usual self-registration scheme. The next task is to have Drawlist look up shaders by name or use a default shader.

As an ( Read more... )

The console was upside down because it was using the wrong projection matrix. I ran a blank level, opened the console, and dumped the frame using GLIntercept. The console projection looked suspiciously like the virtual 640x480 overlay projection, and I confirmed it by running the game at a higher resolution. The console font scaled up instead of ( Read more... )

I got Drawlist and Font working with shaders this weekend, and (unlike my prior attempt) they run just as fast as the fixed-function versions. Anything that draws now requests a specific shader program and sets up uniforms and vertex attributes instead of assuming any particular values. I had to make the rendering system less sensitive to the order ( Read more... )

I found out why fixed-function geometry broke when I enabled the shader code path in Drawlist: the shader path enabled vertex attribute arrays but didn't disable them before switching back to fixed-function. Vertex attribute arrays apparently take precedence over fixed-function arrays and screw up rendering. Disabling them solved that problem.

I stopped the crash when using a shader program but dynamic geometry is completely b0rked. I must not be setting up something right when switching from shaders to fixed-function.

I spent the last three weeks or so bringing VideoVenture into the 21st century, upgrading from old-style OpenGL 1.1 to a more modern architecture. My self-imposed goal is to make the engine compatible with Core Profile or OpenGL ES 2.0 by avoiding deprecated features. I'm still a long way off from that goal but I'm closer than before and made ( Read more... )

I dove into the Expression system again last night intending to remove more template stuff but getting sidetracked by the SIMD implementation. In the end, I found a nicer way to combine four float values into an __m128 and splitting an __m128 into four float values.

ComponentBinary demonstrates both splitting and combining as it takes two __m128 ( Read more... )

Compared to the adapter implementation in my previous post, the current adapter is a big improvement. In particular, I removed the generic implementations that were never used and reused some common pieces.
( Better... )

Last night I made a big push to cleaning up gnarly template crap in the Expression system. Individual expression functions were reasonable but the "adapters" templates were nigh-impenetrable and ugly in use. Before posting the updated version, I wanted to show what is currently checked into the source repository.
( Scary C++ template code... )

I spent most of the evening encapsulating the physics engine within Collidable so nothing outside Collidable and PhysicsJoint depend on it or any of its data types. Everything now goes through the public Collidable interface instead of using Chipmunk (or Box2D) data types or functions. I was even able to remove the physics engine include file ( Read more... )