Feedback Wanted: High Definition Render Pipeline

In my testing with HDRP it seems largely immune to problems like this due to camera relative rendering. What did you test that broke it?

 

Yes, it was a 3rd party asset for testing purposes. Thanks, the issue is resolved here: https://forum.unity.com/threads/hdrp-atmospheric-scattering.566239/#post-3902539

 

-edit- update at bottom
-edit2- working in forward but not deferred.

Quick question.

So I'm attempting to convert a project from LWRP to HDRP and I use a custom master node in shader graph for fog of war support in my project (it needs to modify/disable the lighting depending where in the world the pixel is which is not something I can do by default I don't believe.)

-edit- This bit solved

To do this, I mimic the HDLitMasterNode and HDLitSubShader classes and make them point to a different HDLitPass.template

So the first issue is, HDSubShaderUtilities.GenerateShaderPass templatePath is hardcoded to

string templatePath = Path.Combine(HDUtils.GetHDRenderPipelinePath(), "Editor/Material");

And I can't mimic the behavior (without a hell of a lot of reflection) as it references internal classes.
So is there any chance we could get an overload that takes template path so I don't have to store code in the actual HD renderpipeline directory? I'd much rather store it within the actual project (especially since the packages folder is not part of the repo.)

e.g.

Code (CSharp):

1.     public static class HDSubShaderUtilities
2.     {
3.         public static bool GenerateShaderPass(AbstractMaterialNode masterNode, Pass pass, GenerationMode mode, SurfaceMaterialOptions materialOptions, HashSet<string> activeFields, ShaderGenerator result, List<string> sourceAssetDependencyPaths, bool vertexActive)
4.         {
5.             string templatePath = Path.Combine(HDUtils.GetHDRenderPipelinePath(), "Editor/Material");
6.             return GenerateShaderPass(masterNode, templatePath, pass, mode, materialOptions, activeFields, result, sourceAssetDependencyPaths, vertexActive);
7.         }
8.  
9.         public static bool GenerateShaderPass(AbstractMaterialNode masterNode, string templatePath, Pass pass, GenerationMode mode, SurfaceMaterialOptions materialOptions, HashSet<string> activeFields, ShaderGenerator result, List<string> sourceAssetDependencyPaths, bool vertexActive)
10.         {
11.             string templateLocation = Path.Combine(Path.Combine(Path.Combine(templatePath, pass.MaterialName), "ShaderGraph"), pass.TemplateName);
12.             if (!File.Exists(templateLocation))
13.             {
14.                 // TODO: produce error here
15.                 Debug.LogError("Template not found: " + templateLocation);
16.                 return false;
17.             }
18.  
19.             ...
20.  

The next problem is more of an issue with my understanding how this style of shader works when trying to update the template - graphics and shader code are not my forte.

So in LWRP, within the

half4 frag (GraphVertexOutput IN ${FaceSign}) : SV_Target

I modified the template to

Code (CSharp):

1. float fow = FoWIntensity(IN.WorldSpacePosition);
2.         half4 color = fow == 0 ? half4(0, 0, 0, 1) : LightweightFragmentPBR(
3.             inputData,
4.             Albedo,
5.             Metallic,
6.             Specular,
7.             Smoothness,
8.             Occlusion,
9.             Emission,
10.             Alpha) * fow;

Should be straightforward enough with what I'm doing and this is all I need to do to get my fow working with LWRP.

But I'm having a hard time figuring out how to implement similar behavior in the HDRP as there is a lot more going on in the template.

Anyone have any ideas?

Thanks in advanced

-edit-

Having a closer look at how the shader is constructed and I think my approach is probably wrong in HDRP. The template does not look like I really need to modify it, instead I think I should look at modifying the passes (forward?).

Would also mean I don't need a override for GenerateShaderPass

-edit2-

Ok I got it working for forward rendering by using my own custom ShaderPassForward.hlsl with a simple override at the end

Code (CSharp):

1.         float3 worldSpacePosition = GetAbsolutePositionWS(input.positionRWS);
2.         float fow = FoWIntensity(worldSpacePosition);
3.         outColor = fow == 0 ? float4(0.0, 0.0, 0.0, outColor.a) : outColor * fow;

and setting Lit Shader mode to forward only"



Result:



So next question is, I don't exactly understand how the passes of deferred work. What (pass?) would I need to modify to add support for that as well?

 

Is there still no way to add custom renderpass in HDRP?
did you guys just add Raytracing support to HDRP?

 

Few observations using latest master and 2019.1.0a9:

- SSR and MSAA don't both work together, you can only have one at a time (I know SSR isn't supported in VR but I'd love to have forward+msaa+ssr on desktop).

- Even when everything is setup for HDRP VR as instructed here, you get upside down image at least with OpenVR. (edit: flipping is now fixed on latest master).

I hope these will get addressed eventually, I know 2019.1 is at alpha so currently just reporting these here

 

Anyone else crashing with HDRP in 2018.3 with Enable SRP Batcher enabled when going into play mode or is it just me?

 

Runs fine for me, with batcher on

 

I submitted a bug report, but it's been ongoing for a month now so I guess it's really just me (tm).

Case number and fix: 1103712

I solved it by making sure HDRP scene did not have any materials using HDRenderPipeline/Unlit

Unlit shader causes crash with batcher enabled. Was surprised.

 

Upgraded to 4.3 (using 2018.3.0b11), can no longer compile to windows standalone (other platforms untested, not targeted)

Code (CSharp):

1. Error building Player: Shader error in 'Hidden/HDRenderPipeline/GGXConvolve': failed to open source file: 'GGXConvolution.cs.hlsl' at line 24 (on d3d11)
2.  
3. Compiling Vertex program with USE_MIS
4. Platform defines: UNITY_ENABLE_REFLECTION_BUFFERS UNITY_USE_DITHER_MASK_FOR_ALPHABLENDED_SHADOWS UNITY_PBS_USE_BRDF1 UNITY_SPECCUBE_BOX_PROJECTION UNITY_SPECCUBE_BLENDING UNITY_ENABLE_DETAIL_NORMALMAP SHADER_API_DESKTOP UNITY_LIGHT_PROBE_PROXY_VOLUME UNITY_LIGHTMAP_FULL_HDR
5.  

-edit-

started build again, succeeded this time without changing anything. Not sure what was up.

 

I will try to do so soon, sorry about this have deadline at the moment I am working with multi-scene at the moment with a realtime GI setup.

 

Hello everyone !​

I'm in the process of porting my project on the HDRP. I could move all my custom shaders via custom nodes and I mostly got a better look with all the post process effects etc... But I'm still stuck on some things :

• While looking on the generated HLSL code of shader graph, I found out that nothing goes on the vertex part of the shader. Even when modifying the position before the HD Lit master. There's a bunch of things I could put on there to speed things up. Instead, I'm stuck with putting everything on the frag, which is bad practice.

• I'm using DrawMeshInstanced with MaterialPropertyBlock. So, the instanced part works, but I cannot access my properties. I tried some work around by putting the unity macros :

UNITY_INSTANCING_BUFFER_START(Props)
UNITY_DEFINE_INSTANCED_PROP(float4, _Properties)
#define _Properties_arr Props
UNITY_INSTANCING_BUFFER_END(Props)

after the "{}" of a custom node HLSL code, but unfortunately, I cannot find a way to put the​

UNITY_VERTEX_INPUT_INSTANCE_ID
UNITY_SETUP_INSTANCE_ID(v);
UNITY_TRANSFER_INSTANCE_ID(v, o);

in the vertex shader / struct of the node. There's also the fact that my code would be put on the frag in the generated code instead of the vertex where it should have been, which would be very bad for performance. Obviously, I could take the generated shader, copy it and modifying it so that it's working correctly...​

• Refraction is not working in my case. Before, I was using the grabpass principle. Now, I tried the rough refraction, but it seems it's just working with refraction probes in realtime or baked. I'm procedurally generating a civ-like world and in this case, HDRP refraction is doing absolutly nothing (the water is not transparent). Will there be a solution ? I suppose I could do a customRender derivation of the HDRP on a camera (already did this for some post process fx, it's actually quite powerful without having to modify the HDRP), copy a part of the HDRP render inside, execute before the main camera in a render texture, and use the result in a custom node. But I would have expected that such a feature would be more or less working out of the box with the HDRP.

I also found strange unity editor behaviours :

• I'm using uv2 for reading a biome world map texture to determine what biome I should put on a hexagon. It seems that after some time in the editor play mode (going back and forth to the scene / game, tweaking things, etc...), the uv2 gets overwritten and I see the biome on the world being like camera relative : I move the camera, the biomes on the hexagons are changing. This happen randomly and I have another custom shader much more complex but with the same code embedded inside and this one is not moving (it's using uv2 and uv3).
  
• When I add or remove a post process from a volume during play mode, all the textures I put on my materials via script (SetTexture()) are removed.
  
• SRP batching is breaking all my shaders. I have a uniform color everywhere with it.

Overall, it was not so hard to port the project. The problems I have are more current limitations of the HDRP that didn't exist in the legacy system.

 

I'm a developer of ambient occlusion asset. I don't know whether HDRP or Post Processing Stack v2 has an injection point for these kinds of effects that needs to be rendered before direct lighting. When I check the source code of HDRP, it seems to be hard coding for MSVO in post processing stack v2. I wish there can be a injection point (postprocessevent) or some other solutions to let us inject the code precisely to some render stage.

 

The same needed for beforeTransparent effects.

 

I also get this problem building standalone for Windows approximately every other time I build. I saw it in b11 / 4.2.0-preview and it still happens in b12 / 4.3.0-preview

 

fbx with embedded mat color and textures is not supported by hdrp due to a bad mat conversion.
i hope this get fixed.

 

How can I set light layer for Unity Terrain?

 

How is the support for Vulkan and DX12 currently with HDRP 4.3.0 ?

 

@Remy_Unity is HDRP tesselation supported in shadergraph?

 

Haven't seen Tessellation or Displacement yet. Been checking the GitHub branches every now and then to see if any info shows up.

 

-edit-

turned out to be an issue with shader graph. moved to a more relevant topic.

 

Hi,

I have been experiencing an issue with the HD Render Pipeline and Collab. I have recently upgraded a project to Unity 2018.3 in order to try the HDRP volume lights which are great. Every time we sync with Collab though, after starting Unity no meshes have been rendering/displaying both editor/run time. Anyone else experienced this that could shed a light on?
All materials seem to be properly set to HDRP/Lit. No geometry is displayed, wireframe or shader mode.
The error displayed in the console is:

Code (CSharp):

1. NullReferenceException: Object reference not set to an instance of an object
2. UnityEngine.Experimental.Rendering.HDPipeline.HDRenderPipeline.UpgradeResourcesIfNeeded () (at Library/PackageCache/com.unity.render-pipelines.high-definition@4.3.0-preview/Runtime/RenderPipeline/HDRenderPipeline.cs:351)
3. UnityEngine.Experimental.Rendering.HDPipeline.HDRenderPipeline..ctor (UnityEngine.Experimental.Rendering.HDPipeline.HDRenderPipelineAsset asset) (at Library/PackageCache/com.unity.render-pipelines.high-definition@4.3.0-preview/Runtime/RenderPipeline/HDRenderPipeline.cs:244)
4. UnityEngine.Experimental.Rendering.HDPipeline.HDRenderPipelineAsset.InternalCreatePipeline () (at Library/PackageCache/com.unity.render-pipelines.high-definition@4.3.0-preview/Runtime/RenderPipeline/HDRenderPipelineAsset.cs:23)
5. UnityEngine.Experimental.Rendering.RenderPipelineAsset.CreatePipeline () (at C:/buildslave/unity/build/Runtime/Export/RenderPipeline/RenderPipelineAsset.cs:19)
6. UnityEngine.Experimental.Rendering.RenderPipelineManager.PrepareRenderPipeline (IRenderPipelineAsset pipe) (at C:/buildslave/unity/build/Runtime/Export/RenderPipeline/RenderPipelineManager.cs:55)
7. UnityEngine.Experimental.Rendering.RenderPipelineManager.DoRenderLoop_Internal (IRenderPipelineAsset pipe, UnityEngine.Camera[] cameras, IntPtr loopPtr) (at C:/buildslave/unity/build/Runtime/Export/RenderPipeline/RenderPipelineManager.cs:28)
8. UnityEngine.GUIUtility:ProcessEvent(Int32, IntPtr)

Before the Collab sync everything seems to work properly.
Any pointers would be appreciated. I have deleted the library folder and generated a new one. The HDRenderAsset settings and resources are now displaying like this:


Thanks,

G.

 

Feedback on removal of Camera, Clear Flags -> Depth Only.

This essentially removes the ability to use multiple cameras with their culling mask set to different layers, in order to render objects differently within the scene.

I really want to understand the decision made here. Someone else in this thread described it as

" Remove multi-camera support. LWRP and HDRP will not support multi-camera layered rendering.".

Whilst I'm sure there are technical reasons for removing the ability to do this, I see and am experiencing a few problems because of it.

In first person games it's expected by the player that they can change their FOV. FOV beyond 60 usually means that any object close to the camera will appear elongated and stretched. FPS games, mine included, uses two cameras so that a wide FOV can be used on the scene generally and then objects on a different layer, such as weapons that are close to the camera, are rendered using a different camera. With the removal of the depth only clear flag parameter on cameras, I can't see how the same functionality / problem solution can be achieved. It's also useful as a feature because one can prevent weapons clipping through walls by using layers on the second camera.

I am pursuing converting to the HDRP because of the improved lighting, massively improved BRDF and excellent built in shaders but with functionality like depth only clear flags removed, I can't see how I can do it. Am I overlooking a solution others are using for the problem I've described? Is this something Unity is willing to reconsider or explain?

 

This is probably not a right place to ask but I am not sure where to get my question answered. My only concern is whether there would be a support for more than 8 textures in TerrainLit shader in some of the future releases? I have converted my project into HDRP with really satisfying results, but terrain textures limit is preventing me from continuing. Thank you for an answer.

 

There is a thread about this subject, I doubt I that any magic answers but at least you can find other people wrestling with the same or similar issue over a period of time:

HDRP Weapon shader?

I only reskimmed that thread briefly just now, but the closest thing to a solution seems to involve studying how this was done in the FPS sample that Unity provided to much fanfare some weeks back.

 

Awesome, thank you very much for sharing this. So I understand a little better now why a two camera solution is not ideal and can see from the thread that with a bit of work it's not quite the disaster I thought it might be

 

I asked recently on postprocessing thread about apparently what is v3 PP on HDRP here:
https://forum.unity.com/threads/new-post-processing-stack-pre-release.435581/page-20#post-3937846

Looking at recent commit and PR comments, it's pretty clear that the plan is to completely remove the ability to run PP stack v2 on HDRP. I'm having bunch of concerns though.

First, is it going to be extendable like v2 so people can hook in their own PP effects? We just got Amplify Occlusion to support v2 but that seems to be a short lived joy now.

Secondly, is there going to be some usability pass later on? The whole "lets stuff this all around volume settings" approach is really tedious to use (anyone who's tried this in practice will know what I mean). It's miles behind how easy PP stack v2 was on usability.

And thirdly, I really hope you don't leave SMAA behind, it's not there yet so I'm worried. SMAA is currently the best compromise on not getting everything blurred like on FXAA or ghosting etc artifacts from TAA.

 

Could anyone explain how to setup SSR in HDRP? I enabled all checkboxes in pipeline asset and added SSR volume component and nothing works. Using hdrp 4.3 and Unity 2018.3

 

Make sure your smoothness > 0.9

 

@iamarugin make sure you have smooth materials that can reflect things + don't have MSAA enabled. At least on recent 5.x.x HDRP, SSR and MSAA don't work at the same time, could be similar limitation on 4.3.0. If I remember right, in past you also had to have at least one reflection probe in the scene for some reason (you still clearly saw SSR effect when you enabled or disabled SSR so it wasn't just reflection from probe).

 

Anybody know the best alternative to the inbuilt terrain for HDRP? cant get grass etc to work on it, so what is everyone else doing? I tried using polybrush to scatter grass meshes on a plane but it takes forever and doesnt seem to be able to make a brush that covers a large enough area, when compared to the terrain grass brushes

 

I'll quote a post from another thread here as many of the HDRP users must be interested on current targets:

 

@SebLagarde or @Remy_Unity : Can we get an option to use gradient Color/Texture into the volumetric fog density volume, with the option to use the Depth,Height,FogAnisotropyValue, or ViewDirection as the gradient coordinate (also option to use the coordinate on X or Y texture coordinate)?

 

Hey, I've got a problem with materials. One of my meshes is black, reflections are overall buggy (black artifacts), any new material doesn't change anything. Everything changed when I created UVs 2 for lightmaps, that's what i think at least. Did someone had something similiar?



Edit: it seems like it's some kind of a problem with my FBX, obj works well, don't know why

 

Updated to HDRP & Shadergraph 4.6 and unlit graphs are now broken - nothing in the changelog indicated this might be expected, any info?

 

Hi!
Some llittle questions
1) There is a known HD Pipeline roadmap on the features / improvements we can expect along the next year?
2) About the "Fabric" shader / "Hair" shader... can you share more info? There is some sample or print you could share?
3) There is other specific shaders being worked, like "skin", "eye" or "water" (LW pipeline has its own water shader )?
4) HD pipe will get its own postprocessing features... it includes specific TAA? How it will be different from the current implementation?

Thanks a lot... the HD team work is really awesome!

 

Hey,

Post FX:
- The color from the ambient occlusion post effect does not work. Also if you turn AO on, it turns off until you click it again
- A bool to only show Fog in game and not in scene would help a lot when overviewing your environment

Lighting:
- I miss the 3 layered skybox color from SDRP, right now it uses 2 colors (ground and sky), of which only one has effect on the environment lighting, (sky tint does nothing on the environment) before you could have 3 different colors for shadows, ground, and ambient or so, which had great artistic effect and the lack of those I would consider one of the biggest downgrades.
(Gradient sky does not have any parameters, and procedural sky is always shown despite nor active, looks like a UI bug)

Shaders
- The shader for the terrain is so extremely simplistic its basically unusable for real production work, and its not possible to use shader graph to do something there. Amplify shader even allowed making a terrain material.
The terrain material is either for absolute beginners or full shader programmers, all others are left out totally.

- Shader graph: It is very cumbersome having to set each texture / shader exposure node individually, all other editors/assets I know create them automatically after a new node. Also the reference name is not automatically set to your node name, so you also write the name double. This makes it a real drag creating a new shader quickly. It feels like I require double as long as in ASE and is far less enjoyable.

- Standard shader: If we are being real, most unity/game projects don't require the die-hard optimization of converting all masks into one channel, and never see release, especially not with the new texture crunching method. Having AO, Gloss and alpha all cut into different channels of one texture, same as with the old shader, makes quick iteration and tests insanely slow and cumbersome (and surely confusing for many), and I find the old and new standard shader terrible as a cause. Offer an optional one with all textures as separate input and one optimized one like it is now.

Nice to have:
- I would really love having a choice of different BRDFs in shader graph or somewhere, that would help not making all games having the same look (as with unreal 4 per example) I assume GGX is used, it would be great to have an exposed falloff for more artistic choice.

(Ah also: since the new update, my shader graph materials are broken "Incorrect GPU", and im sure nothing is wrong with them. Someone wrote something with "." was replaced with "," by mistake or so, but not sure.)

 

with 2018.3 are released, is 2019 are using the latest HDRP now?

 

It looks like the docs (on GitHub) are a bit behind with respect to properties in the HD Render Pipeline Asset. I'm trying to decide whether I should enable certain features (ideally disabling those I don't need for better performance), but for some of these I'm just not sure what they do, and what impact it will have if I turn them off. (It's not visually obvious that something happened when changing some of these...) So I was hoping someone might have a sense of whether I can turn some of these off.

• Render Pipeline Supported Features
  • Transparent Backface
  • Transparent Depth Prepass/Postpass
• Rendering Passes
  • Transparent Prepass/Postpass
  • Rough Refraction
• Rendering Settings
  • Depth Prepass with Deferred Rendering
• Async Compute Settings
  • What does any of this stuff do? Is it always better to enable all of these?
• Light Loop Settings
  • I don't know what any of the settings here do.
• Reflections
  • Compress Reflection Probe cache: Does this lead to worse performance as it needs to compress/decompress to use the reflection data?

 

There is an option to do realtime updates of the Skybox, but it takes ~30ms to copy back the image from the graphics card, making it completely unusable. I found this note regarding it from last year:

Currently can’t do dynamic time of day as Enlighten do a synchronized readback of textures which syncs with GPU (Edit: Need to use the new async readback feature)

Are there any plans to update this? It makes a huge difference in visual quality for my game.

 

The only Refresh Mode now available on Reflection Probes set to Realtime is "Every Frame", meaning we can no longer calculate on the first frame only.