How do I get the raw pixels of a RFloat RenderTexture on the CPU?

@burtles I'm afraid not. I ended up doing something incredibly inefficient (using an ARGBFloat32 and copying value to all 4 channels) as my use case was more about it working than perf or mem.

If perf / memory are an issue then there are some options I thought about:

In some cases you can workaround the problem by having each shader process 4 pixels instead of 1 and then packing your RFloat values into an ARGBFloat texture with a quarter of width. Whether this turns out to be inefficient then depends on the complexity of your shader but there are scenarios where it would be almost as fast as doing it with RFloats.

Instead of using RFloat use an integer format like ARGB32. Then when writing it out encode your float into 4 colour components. Might be tricky to make it completely accurate but should be possible.

Perhaps the reason Unity doesn't support this is due to limitations on some platforms, if you need to solve it on a specific platform it may be possible to use a native plug-in.

In my case I needed to access the pixels to save them to disk, sometimes one can rework a design to avoid the need to bring the data back to the CPU, in many cases even if you can do it, that copy is pretty inefficient.

The slightly frustrating thing is that Unity clearly can do it internally as you can get a preview of an RFloat texture in the editor.

Good luck

Did you ever figure this out? As far as I can tell it's impossible.

You can just add the extra copy using this:

 RenderTexture tempRT = RenderTexture.GetTemporary( rTex.width, rTex.height, 0, RenderTextureFormat.ARGB32, RenderTextureReadWrite.Linear );
 Graphics.Blit( rTex, tempRT );
 

You can use a shader to write the texture to a ComputeBuffer and then GetData from that buffer. Double device memory consumption implied, of course. Notice that under DX11 you can do this trick on following uavs only: R32_FLOAT R32_UINT R32_SINT

You can get the raw pixels of a RFloat Texture on the CPU, next convert Texture to RenderTexture, for example:

 const int maxGridSideLength = 50;
 float[] data = new float[maxGridSideLength  * maxGridSideLength  * maxGridSideLength ];//this is your raw data
 
 //set texture3D
 Texture3D texture = new Texture3D(maxGridSideLength, maxGridSideLength, maxGridSideLength, TextureFormat.RFloat, false);
 texture.filterMode = FilterMode.Bilinear;
 texture.SetPixelData(data, 0);
 
 //convert texture3D to RenderTexture
 RenderTexture renderTex = new RenderTexture(maxGridSideLength, maxGridSideLength, 0, RenderTextureFormat.RFloat);
 renderTex.enableRandomWrite = true; 
 renderTex.dimension = UnityEngine.Rendering.TextureDimension.Tex3D; 
 renderTex.volumeDepth = maxGridSideLength; 
 renderTex.filterMode = FilterMode.Bilinear;
 renderTex.wrapMode = TextureWrapMode.Repeat;
 renderTex.useMipMap = false;
 renderTex.Create();
 Graphics.CopyTexture(tex3D, renderTex);
 
 

But ,there may be a small problem that when multiple rendertextures exist, only the last rendertexture conversion can succeed. But I don't know why this is.