[Solved] Trouble with ComputeBuffer in OpenGL

Hi all! I'll try implement Order Independent Transparency with SM5.0 (using UAV and linked lists).
In first step, I create UAV buffers:

Code (CSharp):

1. private ComputeBuffer transparancyBuffer;
2. private RenderTexture renderTexture;
3. //....
4. renderTexture = new RenderTexture(targetCamera.pixelWidth,targetCamera.pixelHeight,0,RenderTextureFormat.RInt,RenderTextureReadWrite.Linear);
5. renderTexture.enableRandomWrite = true;
6. transparancyBuffer = new ComputeBuffer(12 * targetCamera.pixelWidth * targetCamera.pixelHeight, sizeof(uint) * 3, ComputeBufferType.Counter);
7.  

After it my shaders will fill it. In texture I save current head of linked list, in buffer I save elements of list. After all I compose this data in post process effect:
1) Per pixel sort of current list
2) Compose with opaque render.
In shaders I declare my buffers like this:

Code (CSharp):

1. struct NodeData {
2.     uint nextNode;
3.     uint packedColor;
4.     uint depth;
5. };
6.  
7. globallycoherent uniform RWTexture2D<uint> _NodeHeads : register(u1);
8. globallycoherent uniform RWStructuredBuffer<NodeData> _NodeList : register(u2) ;

In DX11 all work as expected, I see this picture:

But in OpenGL (4.5) its look like this:


In my opinion its happens because my RWStructuredBuffer will flushed between threads groups and results not shared... I cant find any solution, because my knowlege in OpenGL very small...
Maybe somebody can help me with it? Code or another information can be provided if it help.

 

So. Its solved. I get some very weird expirience with Unity OpenGL compiler...But, if anybody intrested in this theme, I can explain some key points and give some code for experements after optimisation will done.

 

Hi, how did you fix the bug? Can you share some key points?

 

Its was long time ago =) Write to me on andreich@holyshovelsoft.com, and I'll try find this code and explain details.