6 replies to this topic

[Tim Coolman]

I am using a pixel shader to put some data into a texture. Typically, with a float4 formatted texture, you would output RGBA color data to the texture where each color component is a 0.0 - 1.0 float value. I'm trying to use the pixel shader to store non-color data. This texture is not meant for display. Instead, once the texture is filled, I convert the texture texels to a different binary format using a compute shader (due to the nature of the data, it makes sense for me to output this data with a pixel shader). When outputting to the texture from my pixel shader, I would like to store some uint values instead of floats in the Y, Z, W components. So here is an example of how I'm trying to return from the pixel shader:
[source lang="cpp"]return float4(floatValue, asfloat(firstUintValue), asfloat(secondUintValue), asfloat(thirdUintValue));[/source]
I do this because I don't want to cast the uint values to float, but rather maintain their binary equivalent.

However, when I read from the texture using my compute shader and convert these values back to uint using the asuint(texel.Y) function, they do not seem to be the same value I attempted to store in the first place. Actually, most the time I seem to get ZERO values out of this.

I know that I have supplied my compute shader with the texture as a shader resource properly, because I am able to retrieve the X component of the texels, which you'll notice above was a regular float (between 0.0 and 1.0).

Does the pixel shader require output to be 0.0 - 1.0 floats and do automatic adjustments otherwise?

Thanks you for your time and assistance.

[MJP]

The behavior depends on the format modifer of the DXGI format that you use for the render target. The different format modifiers and their behavior are listed in the documentation for the DXGI_FORMAT enumeration. In your case anything except for a 32-bit FLOAT format will mess up your values, because conversions will happen.

The easiest thing for you to do would be to just use two render targets: one that has a single FLOAT or UNORM channel for your floating-point value, and another that uses a UINT format so that you can just store your values as float's and uint's with no conversions necessary.

[Jason Z]

Have you considered writing to a UAV in your pixel shader? This should help you have a common way to reference the resource, and you can directly use the appropriate type in your shader code.

[Tim Coolman]

In your case anything except for a 32-bit FLOAT format will mess up your values, because conversions will happen.

What kind of conversions? I just figured that since I was using the asfloat() function to store my UINT values, the texture would accept it as a float - how would the texture know the difference that it is actually a binary representation of a UINT? Unless the texture requires that the value be a value color-component value between 0.0 and 1.0.

Have you considered writing to a UAV in your pixel shader?

I'll have to think about this. The reason I'm doing it this way is because I actually am storing graphical data - I still take advantage of the way the pixel shader projects the data onto my texture using transformation matrices, and I also need it to take care of depth buffering and resolution. However, I don't care about color - instead I have other data to keep track of, which is why I was trying to use the color-component values to store other information.

[MJP]

What kind of conversions?

The kind of conversions specified in the documentation for the DXGI_FORMAT enumeration.

I just figured that since I was using the asfloat() function to store my UINT values, the texture would accept it as a float - how would the texture know the difference that it is actually a binary representation of a UINT?

It won't know that it's a UINT, which can be a problem. The hardware will assume it's a 32-bit floating point value, and will treat it as such. If any format conversions are applied, it will apply floating point operations to your output value which will be totally invalid. For instance it might clamp to [0, 1] and convert to an unsigned integer, which will result in a bunch of garbage when you sample it an attempt to cast the value as an integer.

[Tim Coolman]

Okay. The values I'd like to store consist of one float and 3 uint values. Think using DXGI_FORMAT_R32G32B32A32_TYPELESS instead of DXGI_FORMAT_R32G32B32A32_FLOAT would prevent unexpected conversions from occurring?

[Tim Coolman]

Well, after trying a few other ways to do this, I put it back to how I had it and... now it works! Magic. I have no idea what changed since my first attempt, but is now working as I'd originally expected. I apologize, as I feel like I wasted your time with this question. But now using a DXGI_FORMAT_R32G32B32A32_FLOAT texture, I'm able to store UINT values using asfloat() and asuint() to convert back and forth between Pixel and Compute shaders.