Creating depth textures and getting back buffer in beta 2?

Creating depth textures and getting back buffer in beta 2?

• Sudheer Kandali

  Also, Shawn got to this thread right before I did. But what I lose in timeliness I make up for in volume! :)

  His answer about why you can't get the depth stencil buffer is much more correct than mine.

  
  
  

• Sandyee

  Rim van Wersch wrote:

  I'm not sure I understand your question. In MDX I mostly use a R32F render target as my shadow map, into which I render the depth from the light using a pixel shader. Then when rendering the scene, you can use this shadow map in another shader/technique to render the shadows. This is also demonstrated in the ShadowMap sample in the DX SDK. The SurfaceFormat.Single seems to correspond with the R32F format, so this should work out.

  
  
  I'm creating a SurfaceFormat.Color render target (RenderTarget2D) and a DepthFormat.Depth24Stencil8 depth stencil surface (DepthStencilBuffer). I render - or I would if I could get this to work - into both for my shadows.
  
  

  Rim van Wersch wrote:

  The DepthStencilSurface class only seems to hold info about the depth buffer and no data(inhereting only from IDisposable and exposing no useful members), so I guess the pixel shader approach is the only way to go in XNA.

  
  
  Regardless of what DepthStencilBuffer looks like, the class is more than just information - it's the buffer itself. The problem is that I don't understand how to get the texture from it, so there's no way I can set it into my shader. I didn't have this problem in beta 1 when the interface was totally different.
  
  
  

• GS80

  Also, ResolveBackBuffer (slowly) copies the contents of the current backbuffer for the given index into the provided texture, which you could use as an input elsewhere. You wouldn't need to "get the RenderTarget from it" - RenderTargets have Textures, so you can use the results of a rendering operation. If you want to use the Texture2D you passed into ResolveBackBuffer as an input to a subsequent rendering call, then you can just set it on a sampler (or effect parameter, etc.).
  
  

• HumteeDumbtee

  You can't read back depth buffer data into textures any more.
  
  We took out that ability because it doesn't work consistently on all cards: some graphics vendors have hacked the ability into their drivers, but this isn't an official part of the DX spec and people who use it tend to get in trouble when they try to run their program on a wider range of graphics cards and realise this doesn't work any more! It also doesn't work on Xbox.
  
  The more robust and cross platform way to do shadows is to render depth values into a floating point RenderTarget2D.
  
  
  
  

• Rod Yager

  I've managed to successfully convert my shadowing code to work with beta 2.
  
  To achieve colored/alpha blended shadows I use two render targets. One render target receives the shadow depth "color", which is just a single floating point value. The other render target receives the shadow color and alpha. In my main pixel shader I fetch the depth and color/alpha values and do a blend between the two.
  
  I will post a link to the project (which of course will include the source code) I'm working on when I'm done, which should be very soon.
  
  Thanks for the help everybody.
  
  
  

• mlgray

  Sorry for the multiple answers:

   

  You can create your own DepthStencilBuffer with it's constructor, and set it on the device.  When Draw (with the DepthBufferEnable render state turned on) or Clear (using the Color-only overload or specifying ClearOptions.DepthBuffer) are called, then the depth buffer must exist and match the resolution of the render target/back buffer set on that index.  If you don't want to bother with that, then set the DepthStencilBuffer on the device to null and turn off the render state (and make sure you use the full Clear overload and specify ClearOptions.Target only).

   

  Also, you can change the device-created depth buffer (or turn off EnableAutoDepthStencil).  You would add a method to handle the GraphicsDeviceManager's PreparingDeviceSettings event, and addthat method on the event in the game template constructor.  You can modify the arguments passed into that method, giving you more control while still letting the nice GraphicsDeviceManager handle everything else.  :)

   

  Finally, the big thing to wrap your head around with Resolve is that it clears the contents of GetTexture().  That means that you cannot accumulate a render target through multiple Sets.  Multiple Draws between SetRenderTarget and before ResolveRenderTarget accumulate, of course, but if you do:

  SetRenderTarget(0, rt1);

  DrawObject1();

  ResolveRenderTarget(0, rt1);  // after this, rt1.GetTexture has the results of the DrawObject1 call

  SetRenderTarget(0, rt2);

  DrawObject2();

  ResolveRenderTarget(0);

  SetRenderTarget(0, rt1);

  DrawObject3();

  ResolveRenderTarget(0); // after this, rt1.GetTexture only has the results of the DrawObject3 call!

  ... then you only get the results from the last Resolve.  It clears before Resolving (again, 360 platform behavior that is unified on both the PC and 360).

  
  
  

• jsdude99

  This seems very strange. It forces you to use RenderTarget2D now instead of Texture2Ds with render target usage. That doesnt seem too bad.

  So you want to get the backbuffer, and copy that into a texture.

  You do ResolveBackbuffer(texture, index) but how do you get the RenderTarget2D from the texture

  And how do you even copy between RenderTarget2D's

  
  
  

• bobbins

  Matthew Picioccio - MSFT wrote:

  There is no way to get the contents of the depth buffer with the current API. You could render that information into a regular color buffer, but that's definitely not ideal. This was mostly done for cross-platform compatibility. This does complicate a number of shadowing patterns, which we are aware of. If this is a major problem, then please provide feedback via the Connect tool described in the XNA F.A.Q.

  
  
  I appreciate the response, but I have to ask: if there's no way to get the depth buffer with the current API then how do you recommend shadow mapping be done   I was previously rendering into a color and depth buffer so that I could get transparent shadows (it looks roughly like the shadows being cast by light shining through a stained glass window).
  
  
  

• figuerres

  I'm not sure I understand your question.

  In MDX I mostly use a R32F render target as my shadow map, into which I render the depth from the light using a pixel shader. Then when rendering the final scene, you can use this shadow map as a texture in another shader/technique to render the shadows. This is also demonstrated in the ShadowMap sample in the DX SDK. The SurfaceFormat.Single seems to correspond with the R32F format, so this should work out.

  The DepthStencilSurface class only seems to hold info about the depth buffer and no data (inhereting only from IDisposable and exposing no useful members), so the shader/RenderTarget approach looks like the only way to go in XNA.

  And @X-tatic: I posted a suggestion to copy between render targets in your topic on them. Works for me, so I hope that helps

  
  
  

• arcliner

  I believe the DirectX SDK's ShadowMap sample uses the RenderTarget-has-shadow-map method that I described. I think. :)
  
  

• Douglas Penna

  You can create render targets with SurfaceFormat.Single, which still seems to work as it should. As for rendering to textures, I posted a snippet in this topic (sorry if I'm spamming about it, but I guess it's not easy to be found there).

  As for retrieving the backbuffer, I guess that's simply 'not done'. Calling GetRenderTarget(0) on a virgin device will return null. Restoring the backbuffer is apperently done by calling SetRenderTarget(0, null) once you're done with your target and called ResolveRenderTarget on it. It might be possible to copy the backbuffer onto a texture using the ResolveBackBuffer method, which takes a Texture2D as a parameter, but I haven't tested this yet.

  
  
  

• Sioln

  Rim van Wersch wrote:

  You can create render targets with SurfaceFormat.Single, which still seems to work as it should.

  If you create a render target as SurfaceFormat.Single, how do you get the DepthStencilBuffer that you use to set into the GraphicsDevice.DepthStencilBuffer
  

  
  
  

• Gary_MN

  null represents the implicit back buffer (i.e. the one created by the device). So, as previously answered, SetRenderTarget(index, null) will set the render target at that index to the implicit one.

  In general - Textures (2D, 3D, Cube) are inputs to shaders, RenderTargets (2D or Cube, specifying a face) are potential outputs from shaders. First you set the render target on the device, then you "draw" to it (Clear, Draw*, etc.). When you're done using the render target, you have to call ResolveRenderTarget(index). This is a requirement of the 360 implementation, and it is pretty clean once you get used to it. It also makes generating mipmaps for mipmapped render targets a lot easier (i.e. automatic). If you want to use the contents of a RenderTarget after it has been resolved, then you can call GetTexture on that object (it returns the same object each time, filled with the results of the draw calls on Resolve).

  There is no way to get the contents of the depth buffer with the current API. You could render that information into a regular color buffer, but that's definitely not ideal. This was mostly done for cross-platform compatibility. This does complicate a number of shadowing patterns, which we are aware of. If this is a major problem, then please provide feedback via the Connect tool described in the XNA F.A.Q.

  
  
  
• Creating depth textures and getting back buffer in beta 2?