Hey guys just wanted to let everyone know I finally figured out the problem with the deferred renderer demo and that is it seems the newer ATI video cards don’t like shaders accessing the depth buffer directly. I am not sure what is causing this but I am believing it may have something to do with changes in the HD series cards related to their hyper Z and related technologies.
What leads me to believe this is that my work desktop with its ATI Radeon 1100 Xpress works with the demo just fine reading from the depth buffer directly in the light.sha shader. This same thing however is broken on my ATI Radeon HD 3450 I have in my desktop at home and I read on ATI’s site about some of the changes they made in the HD 2000 series and higher cards. This is just a guess though.
For the fix I basically created another texture for the depth buffer and attached it using GraphicsOutput.RTMCopyTexture on the lightbuffer. So the texture and addRenderTexture calls look as follows now.
self.texDepth = Texture()
self.texDepthShader = Texture()
self.texAlbedo = Texture()
self.texNormal = Texture()
self.texFinal = Texture()
self.modelbuffer.addRenderTexture(self.texDepth,
GraphicsOutput.RTMBindOrCopy, GraphicsOutput.RTPDepth)
self.modelbuffer.addRenderTexture(self.texAlbedo,
GraphicsOutput.RTMBindOrCopy, GraphicsOutput.RTPColor)
self.modelbuffer.addRenderTexture(self.texNormal,
GraphicsOutput.RTMBindOrCopy, GraphicsOutput.RTPAuxRgba0)
self.lightbuffer.addRenderTexture(self.texDepth,
GraphicsOutput.RTMBindOrCopy, GraphicsOutput.RTPDepth)
self.lightbuffer.addRenderTexture(self.texDepthShader,
GraphicsOutput.RTMCopyTexture, GraphicsOutput.RTPDepth)
self.lightbuffer.addRenderTexture(self.texFinal,
GraphicsOutput.RTMBindOrCopy, GraphicsOutput.RTPColor)After doing this I had to change one line on the shader inputs for the light.sha shader to use the new texture copy.
tempnode = NodePath(PandaNode("temp node"))
tempnode.setShader(Shader.load(MYDIR+"/light.sha"))
tempnode.setShaderInput("texnormal",self.texNormal)
tempnode.setShaderInput("texalbedo",self.texAlbedo)
tempnode.setShaderInput("texdepth",self.texDepthShader)
tempnode.setShaderInput("proj",Vec4(proj_x,proj_y,proj_z,proj_w))
tempnode.setAttrib(ColorBlendAttrib.make(ColorBlendAttrib.MAdd,
ColorBlendAttrib.OOne, ColorBlendAttrib.OOne))
tempnode.setAttrib(CullFaceAttrib.make(CullFaceAttrib.MCullCounterClockwise))
tempnode.setAttrib(DepthTestAttrib.make(RenderAttrib.MGreaterEqual))
tempnode.setAttrib(DepthWriteAttrib.make(DepthWriteAttrib.MOff))
self.lightcam.node().setInitialState(tempnode.getState())Once I made these changes everything started working properly. The only downfall to this workaround is that it uses more video memory since it has to make a copy of the depth buffer for the shader to read from. One thing to note though is I didn’t see any loss in framerate implementing this workaround. I was still getting 22 FPS with around 300 lights using a 4 MRT setup using the Firefly tutorial code as a basis.
Of course this may not be the best way to fix this so if anyone can find a better way then please let me know as I am still pretty new to OpenGL graphics programming and Panda3D.