HLSL compute shader FViewUniformShaderParameters not populated.

Development Rendering
, , ,
1

Hey, I’m running some hlsl that uses the View.ScreenPositionScaleBias to cast a ray from the camera position, but it seems to be 0.0, 0.0 every time.
I’m using the code implemented in Temaran’s compute shader plugin example to render out to a texture based on a raycast from the camera into world space, my hlsl looks like

  #include "Common.usf"

  void MainPixelShader( in float4 uv : TEXCOORD0, 
                        out float4 OutColor : SV_Target0 )
  {
    OutColor = world_space_color( uv_projection(uv) );
  }

  float2 uv_projection( float4 uv )
  {
    float2 screen_uv = uv.xy / uv.w * View.ScreenPositionScaleBias.xy + View.ScreenPositionScaleBias.wz;
    float  scene_depth = CalcSceneDepth(screen_uv);
    float4 h_world_pos = mul(float4(uv.xy * uv.w * scene_depth, scene_depth, 1), View.ScreenToWorld);
    float3 world_pos = h_world_pos.xyz / h_world_pos.w;
    float3 camera_dir = normalize(world_pos - View.ViewOrigin.xyz);
    return intersect_ray_plane( world_pos, camera_dir);
  }

  float2 intersect_ray_plane(float3 origin, float3 direction)
  {
    float3 planePos = float3( 0,0,0 );
    float3 planeNormal = float3( 0,0,1 );
    float denom =  dot(direction, planeNormal);
    if (abs(denom) < 0.0000001 )
    {
      return float2(0.0, 0.0);
    }
    float t = dot((planePos - origin), planeNormal) / denom;
    return (origin + t * direction).xy;
  }

but View.ScreenPositionScaleBias.xy is always (0.0, 0.0) at which point everything messes up.
Im using the ENQUEUE_UNIQUE_RENDER_COMMAND_ONEPARAMETER to add the shader to the render thread, does it just mean that I happen to be rendering before the FViewUniformShaderParameters has been populated, or do I need to bind the data to my pixel shader when it executes somehow within its threaded function, atm that is

	FRHICommandListImmediate& RHICmdList = GRHICommandList.GetImmediateCommandList();

  SetRenderTarget( RHICmdList, current_render_target_->GetRenderTargetResource()->GetRenderTargetTexture(), FTextureRHIRef() );
  RHICmdList.SetBlendState( TStaticBlendState<>::GetRHI() );
  RHICmdList.SetRasterizerState( TStaticRasterizerState<>::GetRHI() );
  RHICmdList.SetDepthStencilState( TStaticDepthStencilState<false, CF_Always>::GetRHI() );

  static FGlobalBoundShaderState BoundShaderState;
  TShaderMapRef<FBasicVertexShader> VertexShader( GetGlobalShaderMap( feature_level_ ) );
  TShaderMapRef<FTerrainShader> PixelShader( GetGlobalShaderMap( feature_level_ ) );

  SetGlobalBoundShaderState( RHICmdList, feature_level_, BoundShaderState, GTextureVertexDeclaration.VertexDeclarationRHI, *VertexShader, *PixelShader );

  PixelShader->SetUniformBuffers( RHICmdList, constant_parameters_, variable_parameters_ );

  FTextureVertex verts[ 4 ];
  verts[ 0 ].Position = FVector4( -1.0f, 1.0f, 0, 1.0f );
  verts[ 1 ].Position = FVector4( 1.0f, 1.0f, 0, 1.0f );
  verts[ 2 ].Position = FVector4( -1.0f, -1.0f, 0, 1.0f );
  verts[ 3 ].Position = FVector4( 1.0f, -1.0f, 0, 1.0f );
  verts[ 0 ].UV = FVector2D( 0, 0 );
  verts[ 1 ].UV = FVector2D( 1, 0 );
  verts[ 2 ].UV = FVector2D( 0, 1 );
  verts[ 3 ].UV = FVector2D( 1, 1 );

  DrawPrimitiveUP( RHICmdList, PT_TriangleStrip, 2, verts, sizeof( verts[ 0 ] ) );

  PixelShader->UnbindBuffers( RHICmdList );

the result I am then using in a material to

84546-capture.png

so that my procedural texture is then projected onto our terrain with the material system.
I can see that it is the View structure that doesn’t have any data if I debug the shader by outputing different colors for different issues and it all traced back to this structure not having the data I need. Do I need to just populate out some of the data myself, or is this a render order issue (like how the decal system is deferred?)

2

Ok nm I figured it all out, shaders have a way of making me feel stupid a lot of the time, all I really needed was the inverse View Projection matrix to do the ray cast so I bound that up in my shader variables (BEGIN_UNIFORM_BUFFER_STRUCT) and passed that through and did my raytrace with a bit of

float2 uv_projection( float4 uv )
{
  float4 clip;
  clip.x = 2.0 * uv.x - 1.0;
  clip.y = -2.0 * uv.y + 1.0;
  clip.z = 0.5;
  clip.w = 1.0;

  float4 world_pos = mul(clip,TerrainVariables.screen_to_world);
  world_pos.w = 1.0 / world_pos.w;
  world_pos.xyz *= world_pos.w;

  return intersect_ray_plane( world_pos.xyz, normalize( world_pos.xyz - TerrainVariables.view_origin ) );
}

I did end up poking around trying to figure out why my shader had an empty View object and saw that it was because I was not binding the the FViewUniformShaderParameters with SetUniformBufferParameter to my specific shader.
Had some issues with a delay on the viewprj matrix being out of sync with the camera so it would lag on the projection. Managed to fix it up by reconstructing it using a thread safe set of code from this post but am still having some issues with the material update/projection occuring before we have rendered the asset from the active camera. Not sure how/if I can adjust when my render thread task occurs, if I can make it a higher priority?