[Interest] Qt3D Framegraphs

[Andy]

On Mon, Sep 3, 2018 at 9:25 AM Paul Lemire <paul.lemire at kdab.com> wrote:

> Glad to hear that, hopefully things are starting to make more sense now.
>

Getting there - thank you!

On 09/03/2018 02:54 PM, Andy wrote:
>
> Progress! Here's my current framegraph:
>
> [snip]
>
> Question:
>
>    1) I am using an RGBAFormat for my texture. I changed the alpha in the
> clear colour from 0x80 to 0xEE and I now see an alpha cleared background in
> the offscreen (see image). I can just use RGB for my purposes right now,
> but I'm curious why the onscreen clearing is not using the alpha channel? I
> can confirm this by changing the clear colour to #FF000000 - I just get
> solid black.
>
> Well I believe that this depends on the format of your back buffer
> (usually it is RGB). You can try to query it with
> QSurfaceFormat::defaultFormat() and looking for the alphaBuffer size (or
> apitrace also gives you the format when you select a draw call that renders
> to screen).
>

Got it! If I setAlphaBufferSize( 8 ) on my default format it works.

>
> Problem:
>
>    1) The resulting scene isn't the same in the offscreen capture:
>       - the yellow cube is on top of everything
>       - the red & blue arrows aren't clipped by the plane
>
> I suspect that this is caused by the fact that you have no depth
> attachment on your RenderTarget so that depth testing isn't performed
> properly. You would need to create another RenderTargetOutput that you bind
> to the attachment point Depth with a suitable Texture2D texture with format
> (D32, D24 ...).
>

Bingo. That fixes it.


>       - it isn't antialiased
>
> That's likely caused by a) not having a high resolution enough for your
> attachments b) using a Texture2D instead of a Texture2DMultisample (though
> I'm not sure RenderCapture would work with the latter).
>
Have you tried going for a 2048/2408 texture instead of 512/512 assuming
> you have no memory constraints? Then you can always scale back the QImage
> you capture to 512/512 if need be.
>

Texture2DMultisample does indeed make it better. Once I set "samples" to
the same as my QSurfaceFormat::defaultFormat(), I get decent results. Not
100% the same, but very close. (So it does work w/RenderCapture!)

My "final" frame graph (for those following along):

RenderSurfaceSelector:
  Viewport:
    ClearBuffers:
      buffers: ColorDepthBuffer
      clearColor: "#faebd7"
      NoDraw: {}
    FrustumCulling:
      # OnScreen
      CameraSelector:
        objectName: onScreenCameraSelector
        RenderCapture:
          objectName: onScreenCapture
      # OffScreen
      CameraSelector:
        objectName: offScreenCameraSelector
        RenderTargetSelector:
          target:
            RenderTarget:
              attachments:
              - RenderTargetOutput:
                  attachmentPoint: Color0
                  texture:
                    Texture2DMultisample:
                      objectName: offScreenTexture
                      width: 1024
                      height: 768
                      format: RGBFormat
                      samples: 8
              - RenderTargetOutput:
                  attachmentPoint: Depth
                  texture:
                    Texture2DMultisample:
                      width: 1024
                      height: 768
                      format: D24
                      samples: 8
          ClearBuffers:
            buffers: ColorDepthBuffer
            clearColor: "#faebd7"
            NoDraw: {}
          RenderCapture:
            objectName: offScreenCapture


If anyone is interested in the code to read framegraphs as YAML like this,
please get in touch and I can clean it up & put on gitlab (sometime next
month). It makes it a lot easier to iterate on building a framegraph. It
also drastically reduces the amount of boilerplate code, you can include &
read them as resources, and you don't have to bring in all of QML.

Now that I have the basics working... I'll need to dig into the multipass
shader stuff to get the effects I want.

Thank you for your patience!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.qt-project.org/pipermail/interest/attachments/20180904/8b95f0e2/attachment.html>