How to create a smaller D3D12Resource from a specific area of another larger D3D12Resource

Question

I am trying to provide to a machine learning model a D3D12Resource of size 640x640. However, my input D3D12Resource can be of any size. If I have an input resource of 1024x1024 and I don't want to resize(downscale) the input resource to 640x640. How can I tile this resource into 640x640 D3D12Resources or how can I render a specific region of the 1024x1024 D3D12Resource to a 640x640 resource?

I have a shadder sampler and tried manipulating the PORTVIEW and SCISSORS RECT:

viewport = CD3DX12_VIEWPORT( static_cast<FLOAT>( 0 ), static_cast<FLOAT>( 0 ), static_cast<float>( 1024 ), static_cast<float>( 1024 ) );
scissorRect = CD3DX12_RECT( static_cast<LONG>( 0 ), static_cast<LONG>( 0 ), static_cast<LONG>( 640 ), static_cast<LONG>( 640 ) );

This will give me the quadrant (0,0, 640, 640 ). However, if I change the scissorRect coordinates to (640, 0, 1024, 640), the new render resource I cant get the new region rendered as expected to the new output resource. I tried also changing the viewport coordinates to match the scissor, but that also results in some shifted rendering.

I am not sure if I am trying the right approach for this problem.

Answer 1

ScissorRect and Viewport only define the output region (or Destination/Render Target region, whatever name you chose), it doesn't have a say on the input (source) region.

I don't know about machine learning or how you can use a render pipeline for it, the render pipeline was designed for rendering games after all.

Did you write the vertex and pixel shaders yourself, or did you use someone else's (third-party library)? Do you have access to its code, and do you have control over the input vertices data?

• in your vertex shader:

In your output, SV_POSITION (which is a float4 value), x is the horizontal coordinate, y is the vertical coordinate, and z is the depth coordinate.

Coordinates are related to the render target viewport, and X and Y coordinates range from -1.0 to +1.0, which means (0.0, 0.0) would indicate the center of your Viewport.

The positive direction of the X and Y axes are towards the right and the up, which means (-1.0, -1.0) would indicate the bottom left corner of the Viewport, and (+1.0, +1.0) would indicate the top right corner of the Viewport.

ScissorRect happens after the Viewport. On top of the Viewport, ScissorRect limits that only the pixels inside the defined Rect would be rendered.

• in your pixel shader:

the float2 coordinate (usually called UV coordinate) that you would typically use to sample a color from the texture, ranges from 0.0 to 1.0.

(0.0, 0.0) indicates the top left corner of the texture, (1.0, 1.0) indicates the bottom right corner of the texture.

• so generally:

In your code, I can see that you set the Viewport to left=0, top=0, right=1024, bottom=1024 and ScissorRect to left=0, top=0, right=640, bottom=640 and (if I understood correctly) in this way you mapped the source region pixels exactly 1:1 to the destination pixels, and clipped out the content beyond coordinate 640.

From this info, I can infer that you must have vertices like:

	SV_POSITION (destination coordinates in viewport)	UV (source coordinates in texture)
Vertex 0	(X=-1.0, Y=1.0, ...) (top left corner)	(U=0, V=0) (top left corner)
Vertex 1	(X=1.0, Y=1.0, ...) (top right corner)	(U=1, V=0) (top right corner)
Vertex 2	(X=-1.0, Y=-1.0, ...) (bottom left corner)	(U=0, V=1) (bottom left corner)
Vertex 3	(X=1.0, Y=-1.0, ...) (bottom right corner)	(U=1, V=1) (bottom right corner)

In this case, if you want to render from source region left=640, top=0, right=1024, bottom=640 to destination region left=0, top=0, right=384, bottom=640, the simplest approach would be to keep your vertices and shaders as it is, and adjust the Viewport and ScissorRects like this:

Viewport left=-640, top=0, right=384, bottom=640

ScissorRect left=0, top=0, right=384, bottom=640

---

Edit

Here, I drew a picture in MS Paint to help with understanding

In this picture, the two red rectangles are the regions you want to render from the source to the render target:

According to your current vertices/shader settings, you can consider the viewport as the source texture. We adjust the position of the source texture (viewport) so that the two desired regions overlap perfectly. Then we set the Scissor Rect to cover this region. Pixels within the Scissor Rect will be rendered.