Everything gets implemented in the shadows branch.
ShadowLOD
When a light casts shadows, it has to provide a ShadowLOD object. That object selects the level of detail needed by the cast shadows. There’re three possible values:
LowShadow
MediumShadow
HighShadow
If a light mustn’t cast shadow, there’s no ShadowLOD object to pass. Then, the interface is obviously Maybe ShadowLOD, with Nothing for… no cast shadows.
ShadowConf
ShadowConf is an object of the render side. It defines, for each level of detail, the configuration used in the renderer. That objects defines three parts – i.e. for low, medium and high shadows – which are each composed of two information:
the resolution of the shadowmaps to use
the limit number of available shadowmaps
For instance, ShadowConf 512 10 1024 5 2048 2 will spawn ten 512x512 low shadowmaps, five 1024x1024 medium shadowmaps and two 2048x2048 high shadowmaps.
Light data
Currently, lights have several information stored on the render side – in a SSBO:
position
color
radius
power
I think that’s pretty all
We are to add two new values to the list. The first one is the level of detail of the shadows to cast. That is a simple uint and choices are encoded that way:
Just LowShadow is 1
Just MediumShadow is 2
Just HighShadow is 3
Nothing is 0
The second value is the index of the shadowmap. Because we’ll have several shadows at the same time, we’ll need to know which one to sample from when dealing with a given light. That’s stored as a uint as well. If the LOD is 0, the shadowmap index is set to 0 and shouldn’t be read anyway to prevent fucks to happen.
Shader
The lighting shader will now have two additional information in the structure defining a light and three samplerCubeArray, for low, medium and high shadowmaps, respectively. Along with the above two information, we’ll be able to perform a correct lookup of the right shadowmap.
Shadowmap generation
Before sampling the shadowmaps, we’ll need the shadowmaps themselves. To generate them, we need to be able to bind the correct one. That can be done very easily be reading the Maybe ShadowLOD object. If we got Nothing, we won’t do anything and go on. If we got Just lod, depending on the lod, we’ll bind the corresponding cubemap array, and get in some kinda state the shadowmap index (then increment it in the state). That index will have to be passed to the depthmap generator shader to write to the correct layer-face. Pretty simple right.
Shadows
Everything gets implemented in the shadows branch.
ShadowLODWhen a light casts shadows, it has to provide a
ShadowLODobject. That object selects the level of detail needed by the cast shadows. There’re three possible values:LowShadowMediumShadowHighShadowIf a light mustn’t cast shadow, there’s no
ShadowLODobject to pass. Then, the interface is obviouslyMaybe ShadowLOD, withNothingfor… no cast shadows.ShadowConfShadowConfis an object of the render side. It defines, for each level of detail, the configuration used in the renderer. That objects defines three parts – i.e. for low, medium and high shadows – which are each composed of two information:For instance,
ShadowConf 512 10 1024 5 2048 2will spawn ten 512x512 low shadowmaps, five 1024x1024 medium shadowmaps and two 2048x2048 high shadowmaps.Light data
Currently, lights have several information stored on the render side – in a SSBO:
We are to add two new values to the list. The first one is the level of detail of the shadows to cast. That is a simple
uintand choices are encoded that way:Just LowShadowis1Just MediumShadowis2Just HighShadowis3Nothingis0The second value is the index of the shadowmap. Because we’ll have several shadows at the same time, we’ll need to know which one to sample from when dealing with a given light. That’s stored as a
uintas well. If the LOD is0, the shadowmap index is set to0and shouldn’t be read anyway to prevent fucks to happen.Shader
The lighting shader will now have two additional information in the structure defining a light and three
samplerCubeArray, for low, medium and high shadowmaps, respectively. Along with the above two information, we’ll be able to perform a correct lookup of the right shadowmap.Shadowmap generation
Before sampling the shadowmaps, we’ll need the shadowmaps themselves. To generate them, we need to be able to bind the correct one. That can be done very easily be reading the
Maybe ShadowLODobject. If we gotNothing, we won’t do anything and go on. If we gotJust lod, depending on the lod, we’ll bind the corresponding cubemap array, and get in some kinda state the shadowmap index (then increment it in the state). That index will have to be passed to the depthmap generator shader to write to the correct layer-face. Pretty simple right.