Texture.allowThreadedTextureCreation

Hi,

I've been trying to create Textures on a non-main thread without succes, even though Texture.allowThreadedTextureCreation is set to true. I can't find too much info on this property either. Is this supposed to be working yet, how should we use/interpret it?

Tnx,

Geert

 

Hi!

Tnx for replying!

I'm building for Windows, and allowThreadedTextureCreation is set to true (I've checked that at runtime). I was kind of hoping I could now "new Texture2D(width,height)" on a different thread, but that assumption is wrong I suppose?

The problem I'm trying to solve: I'm streaming 3D data over http, I've succeeded to a very large extent to have as much work done on other threads than the main thread to have very fluent rendering while streaming. The only problem remaining are large textures as I must allocate them on the main thread ...

Geert

 

I am experiencing the same issue on Mac. After setting Texture2D.allowThreadedTextureCreation and confirming that it returns true, calling "new Texture2D(width,height)" from a separate thread results in runtime errors that state that texture creation can only be done in Main Thread.

The documentation for Texture.allowThreadedTextureCreation is terse. Is there a prescribed technique to create textures from another thread?

jeff

 

@Unity Can you please show an example on how to create a texture on another thread, as we are also not able to create a texture even if allowThreadedTextureCreation is true:

Code (CSharp):

1.     private void CreateTextureAsync()
2.     {
3.         //UnityException: SupportsTextureFormatNative can only be called from the main thread.
4.         //Constructors and field initializers will be executed from the loading thread when loading a scene.
5.         //Don't use this function in the constructor or field initializers, instead move initialization code to the Awake or Start function.
6.         Texture2D tex = new Texture2D(256, 256);
7.     }
8.  
9.  
10.     private void StartCreateTexture()
11.     {
12.         Thread loadingThread = new Thread(CreateTextureAsync);
13.         loadingThread.Start();
14.     }

 

I still can't get it to work either, tried several combinations/variations on each new (minor) Unity version. I really need this, and I am quite amazed that not too many others are kinda waiting for this.

Texture.allowThreadedTextureCreation has been in the API for several months now. Anyone has any clue on how to use it or if it is possible in any way to create Textures or do a ImageConversion.LoadImage on another thread?

 

Ok.

Do you see any way of converting "raw png bytes" to "RawTextureData" on a non-main thread? (I currently simply use LoadImage for that). I suppose some kind of custom png decoder to start with?

Geert

 

Is it safe to use this pointer on another thread though or do we need to create a buffer copy of it before queueing it onto another thread?

edit: Further googling seems like using NativeArray created on the main thread on another thread probably isn't safe.

 

Having to call .Apply makes everything here redundant since I find that is the most expensive method by far. For example with a 1k by 1k texture I can get setting pixels down to 0.61ms but calling .Apply is taking a massive 49.49ms.

Are there any future plans to allow us to create textures on the fly on a different thread/job?

My end use case is that I am trying to build a Texture2DArray that contains a current list of textures needing rendered that I will be altering depending on what is in view. Without this, batching is impossible

 

If you are using SetPixels that is your problem. Any modern game should never use SetPixels. You should be using CopyTexture for your custom texture streaming solution.

To be clear, using Graphics.CopyTexture (or command buffer equivalent) will prevent you from having to call Apply().

 

Its not as simple as that. I'm downloading data from the network that I use to build a texture. I need to call Apply to get data to the GPU before I can play with things like Graphics.CopyTexture

 

What I do is call GetRawTextureData<byte>() on the texture, then write the data to that array, then call Apply() on the texture later. I get pretty good performance with that.

 

I'm encoding to JPG and storing the byte[].

Then I'm reading the byte[] and setting it into the texture array (works nice and fast).

But Apply was taking 50ms. This was only with a 1024 by 1024 with 10 entries! TextureArray is also RGB24

Edit: Edited to clarify that I'm editing an array, not just a Texture2D. My original wording wasn't clear on this

 

Maybe Apply will be faster if writing into the array from GetRawTextureData() is what I'm saying. You could also try writing only a bit per frame if that is allowable.

 

You have to write the whole mip at once with a Texture2DArray and you can't use GetRawTextureData() with them either. You can write pixel data with custom data which is fast but like I say, Apply is the problem.

The speed of Apply doesn't change depending on if you use SetPixels vs SetPixelData. Doubted it would cause its still just trying to go from CPU to GPU and those methods are only playing on CPU data