Fix texture copy when using multiple canvas #1481
Conversation
| @@ -121,6 +121,9 @@ namespace Babylon::Graphics | |||
| TextureInfo GetTextureInfo(bgfx::TextureHandle handle); | |||
| static bx::AllocatorI& GetDefaultAllocator() { return m_allocator; } | |||
|
|
|||
| // call following method when a submit call is associated with the current acquired viewId | |||
| void SetViewAsUsed(); | |||
There was a problem hiding this comment.
I guess we have to expose this new function because FrameBuffer has to tell DeviceContext this needs to be done when a Submit happens. Is there any other case (other than Submit) where we would need to do this, or is this really tied to FrameBuffer::Submit?
If we need to expose this as a new function, would InvalidateView be a better name?
There was a problem hiding this comment.
Yes, Framebuffer tells device context the view contains draw call. I've added a call for Clear . InvalidateView works for me!
| @@ -1431,11 +1431,12 @@ namespace Babylon | |||
| { | |||
| const auto textureDestination = info[0].As<Napi::Pointer<Graphics::Texture>>().Get(); | |||
| const auto textureSource = info[1].As<Napi::Pointer<Graphics::Texture>>().Get(); | |||
|
|
|||
|
|
|||
| // Append a task instead of calling blit immediately or blit happens before drawcalls. | |||
| arcana::make_task(m_update.Scheduler(), *m_cancellationSource, [this, textureDestination, textureSource, cancellationSource = m_cancellationSource]() { | |||
There was a problem hiding this comment.
Did you try removing this? I think we might need to add this to the command queue for the ordering to be correct.
There was a problem hiding this comment.
Yes, I tried and copy doesn't happen in the correct order when removed.
There was a problem hiding this comment.
Can you try putting it in the command queue?
There is an issue when using multiple dynamic textures (Canvas). Only the 1st one is properly displayed.
GLTF Node NegativeScale (0)validation test looks like this:A Renderdoc capture gives this call trace:
Texture copies are all done at the same time in bgfx, at the end of sorted drawcalls of a viewid : https://github.com/bkaradzic/bgfx/blob/4602a01ce51611183d8ae8e320ca0f5305b79cb8/src/renderer_d3d11.cpp#L5739
All blit commands are submitted with the same ViewID. Which is confirmed when looking at this:
BabylonNative/Plugins/NativeEngine/Source/NativeEngine.cpp
Line 1438 in 9379cf3
Solution is incrementing ViewID in
Blitand moving the call into DeviceContext as it doesn't rely on FrameBuffer state.Result: