It wasn't just a matter of using smoothstep, I had to refactor the code to take a different approach to generating blend weights. We really have to avoid any kind of division for normalization of weights because that causes all the blend functions to become non-linear. So with these changes, the blend weights are calculated based on linear interpretation for displacement-based blending too (before only the non-displacement blending used linear interpolation). With that in place, smoothstep can now be used to give a smooth transition.
I'm not sure which of these edge cases may have existed before updating the property handling, and which were caused by it.
- Rearranged the per-layer parallax property groups because these are more general than just parallax. They can be used for displacement-based blending regardless of whether a parallax effect is being used.
-- Renamed "Parallax Mapping" to "Displacement" because these properties can be used for other things besides parallax, in particular the new displacement-based blend modes.
-- Removed the unnecessary per-layer "enable parallax" flags. This also allowed me to remove the StandardMultilayerPBR_ParallaxPerLayer.lua script and replace this with simply a UseTexture functor for each layer.
-- Made the "offset" property always available, so this can be used to adjust displacement for blending purposes even when there is no heightmap or parallax. The "factor" property still only shows up with a heightmap because its only purpose is to scale the heightmap.
-- In order to get the offset to work when there is no texture map, I had to fix the logic a bit in GetLayerDepthValues where it was ignoring the offset.
-- Had to rearrange the logic in StandardMultilayerPBR_Displacement.lua a bit to get this all working, particularly because the per-layer displacement properties are no longer hidden behind an enable flag.
- Change the displacementBlendFactor to displacementBlendDistance because it felt weird when sliding per-layer displacment offset values and seeing this impact the surface property transition. Using an absolute distance value feels more natural.
- Made the displacement blend mask push the displacement down *past* the min displacement value to address edge cases where blend mask 0 didn't actually make a layer disappear. (See GetSubMinDisplacement()).
- Inlined the GetBlendWeightsFromLayerDepthValues code into GetBlendWeights because I realized it was only being used there, and the code is easier to read this way IMO.
- Displacement-based blend weights weren't being normalized in cases where layerDepthBlendDistance is 0, which caused incorrect depth values where two layers meet.
Renamed StandardMultilayerPBR_Parallax.lua to StandardMultilayerPBR_Displacement.lua because it is used for more than just strictly parallax, it generally deals with displcament which can be used for blending even when parallax is disabled.
Updated the terminology (yet again) to hopefully be a bit more clear and consistent.
"Blend mask" means the R and G channels that mask layers 2-3. These can come from multipls source, including a "blend mask texture" or "blend mask vertex colors".
"Blend weights" are the final RGB channels that are multiplied and added with layer properties to do the final blend.
"Blend source" is the combination of data that is used to produce the blend weights, which could be a combination of displacement maps, blend mask texture, vertex colors, or others in the future.
Added another debug render mode, so now we can show either the blend mask and the final blend weights.
ATOM-14688 Disable Individual Layers
- Added new SetMaterialPropertyGroupVisibility functions to the material functors.
- Updated the MaterialFunctor::EditorContext to include parameters for handling material property group metadata.
- Updated the material inspector(s) to apply the property group visiblity changes from the material functor, to hide or show the property groups.
- Moved some code from MaterialPropertyDescriptor.h/cpp to a new MaterialDynamicMetadata.h/cpp, since these aren't really related to the MaterialPropertyDescriptor code. It's more for material functors to use.
- Also fixed the casing for the "GetMaterialPropertyValue_Image" lua function, since I was already in this code (ATOM-14793 "Fix Inconsistent Casing For LuaMaterialFunctorRuntimeContext")
Tested in MaterialEditor and in in the main Editor's MaterialComponent property override inspector.
Added flags for enabling StandardMultilayerPBR layers 2 and 3. This makes it easy to create a two-layer material, or to flip layers off and on for debugging.
Refactored StandardMultilayerPBR to collate all the code for each layer into a couple structs and utility functions. This makes the code easier to maintain, and in particular will make it easy for me to add Enable flags for the layers in a subsequent commit.
Also removed subsurface scattering and translucency from StandardMultilayerPBR, according to ATOM-4120 "Stabilize Standard PBR Regarding Subsurface and Translucency".
Squashed commit of the following:
commit a6052d6ad4f70183d0ce72e84c7dc5512dc24d5e
Author: Chris Santora <santorac@amazon.com>
Date: Tue May 11 16:32:15 2021 -0700
Got the refactor finally working. I had change it to blend the baseColor, spec factor, and metalness before converting to albedo and spec, in order to get exactly the same results as before.
commit 42d6da7f405097dea07b6ed0426d6a662b61440d
Author: Chris Santora <santorac@amazon.com>
Date: Tue May 11 15:58:38 2021 -0700
Fixed clear coat issue due to LightingData initialized too late.
commit 358194a5caf6f9eb99b0e5345ad5f7768b244a93
Author: Chris Santora <santorac@amazon.com>
Date: Tue May 11 15:18:30 2021 -0700
Fixed a couple issues.
commit adb431f8113b945057959db288a7ee2dd825dd69
Author: Chris Santora <santorac@amazon.com>
Date: Tue May 11 12:42:12 2021 -0700
WIP refactor of StandardMultilayerPBR to collate the code for each layer. Also removed subsurface scattering from multilayer.
Changed to lerp-based blending, with an implicit base layer.
Renamed some variables to be more clear (blendWeight instead of blendMask, since the weights could come from vertex colors instead of a texture).
Updated DefaultBlendMask_layers.png to better suit the new layering model. It has a black background and overlapping R, G, and B areas.
Updated some of the test materials UV transforms to better fit the new DefaultBlendMask_layers image.
Added a new test object, which is a plane that has painted vertices.
Note that I updated test criteria in AtomSampleViewer to account for these changes as well.
Added new "Displacment" blend source option that blends between layers based on which displaced height is higher. For now it simply picks the higher of the three. In subsequent commits I'll improve on the blending to allow for some transition.
Recactored GetLayerDepthValues(), GetBlendWeights(), and GetBlendWeightsFromLayerDepthValues() to optimize parallax searches so that the displacment maps are sampled once and used for both parallax and the calculating the blend weights.
Renamed several layer blending types and variables to be more clear.
Added Height Offset support to StandardMultilayerPBR.materialtype.
In order to make this work, I updated the GetDepth callback function to support the option of returning absolute depth values rather than relative depth values. Although I could have done transformations inside the GetDepth function, having this as absolute cleans things up a lot. StandardMultilayerPBR_Parallax.lua functor code now populates the MaterialSRG with displacement min/max values, instead of having to normalize the depth factors for each layer. I think this is easier to understand and work with.
Added Height Offset to each layer of StandardMultilayerPBR.
Updated the naming and description for the parallax factor in each layer, to match the other material types.
I removed the global "factor" material property because it doesn't seem applicable anymore since we have per-layer height offset. We can always add some form of this later if customers ask for it.
Squashed commit of the following:
commit 8df460800ff7058f9fbb01f995efdd5ab53d3d2c
Author: Chris Santora <santorac@amazon.com>
Date: Tue May 4 13:35:27 2021 -0700
Found a workaround for the DXC compiler bug
commit 5d81617285eb42bb7b48eb060234d2cb89249e34
Author: Chris Santora <santorac@amazon.com>
Date: Tue May 4 12:27:22 2021 -0700
Local WIP changes to get a DepthResult struct set up for the GetDepth functions.
Made sure the depth and shadow shaders enable parallax calculations when alpha clipping is enabled (instead of only when parallax POM is enabled).
Moved the alpha calculations to be *after* the parallax calculations.
Factored out ShouldHandleParallax() and ShouldHandleParallaxInDepthShaders() utility functions, which help us ensure consistent application of parallax calculations across the various shaders in each material type.
Removed some dead code in a couple shaders where dirToCamera was calculated but not used.
Also did ATOM-15034 "Remove Opacity From Multi-Layer Material Types For Now" rather than addressing whatever additional alpha cutout issues might be present on multilayer materials.
Testing:
AtomSampleViewer full test suite.
Added a new AtomSampleViewer screenshot test for alpha clipping with parallax (separate repo).
Addional testing of relevant properties in Matirial Editor.
Tested updated StandardMultilayerPBR in Editor.exe where I had shadows and clipping against other geometry.
Added parallax "Height Offset" properties to StandardPBR and EnhancedPBR.
- Refactored depth pass and shadow pass shaders to use the GetParallaxInput utility.
- Updated EnhancedPBR.materialtype to match the parallax controls and functors of StandardPBR.materialtype.
- Updated EnhancedPBR's shadow pass shader to apply a ShadowMapDepthBias because I noticed StandardPBR is doing this, so I made them match.
- Updated StandardPBR and EnhancedPBR to both use the depth offset property, as well as heightmap clipping debug view.
- Note that depth offset is not supported in StandardMultilayerPBR yet, it's hard-coded to 0 for now.
- Note that I plan to rename a lot of the "depth" terms to "heightmap" or "displacement" in an upcoming commit.
-Fix vulkan graphics pipeline creation failure having depth clamp off with devices not supporting depth clip.
-Added reason for the mobile version of screen space reflection
- Fixing cloth working with Actors by reading directly from ModelAsset instead from EMFX Mesh.
- Actor caches map from joint indices in skin metadata to skeleton indices so they can be query later.
- Actor cloth skinning reads indices and weights from Model Asset instead from EMFX Mesh.
- Actor cloth skinning with unlimited skinning bones.
- Sort out cloth unit tests by disabling them until there is a way to create an Atom mesh
- Addressing feedback.
Note there were significant conflicts in LightingModel.azsli and StandardPBR_ForwardPass.azsl. I only did basic resolves of these conflicts, mostly preserving the changes from main and distarding the changes in the local branch. The shaders are not compiling at this point, and of course the tests are not passing. I will fix up the code to make corrections in a separate commit, to make it easier to review those changes separate from this merge.
Added new input attachment to the fullscreen reflection pass to receive the "ambient" gbuffer.
Added diffuse and specular occlusion to the Skin material type; it was missing before.
ATOM-14040 Add Support for Cavity Maps
Testing:
AtomSampleViewer automation, with updates that will be submitted in that repo.
Manual testing in Material Editor.
Built all AtomTest assets and opened a few test levels, with updates that will be submitted in that repo.
Made local copies of the occlusion test materials and changed replaced the material types with EnhancedPBR and Skin, and got identical results (with clear coat disabled since it works different in EnhancedPBR and is absent in Skin).
moudgils
antonmic
mrieggeramzn
Chris Santora
gallowj
mriegger
AMZN-mnaumov
mnaumov
Terry Michaels
Michael Riegger
dmcdiar
jiaweig
moudgils
Peng
Chris Burel
Aaron Ruiz Mora
alexpete
Brian Herrera