LWRP 4.0 Oculus Go - Final Blit Performance. Is it needed?

Hi all,

I've been working on performance optimizations for our Oculus Go project for a bit and noticed a rather large performance drain in the FinalBlitPass that the LWRP enqueues. FinalBlit accounts for about half of the GPU time of our frame. I ended up disabling it by removing the "isTargetTexture2DArray" portion of the "RequireIntermediateColorTexture" function in ScriptableRenderer.cs, but I'm unsure why we need a intermediate color texture when using VR/Texture2DArrays in the first place.

The latest version of LWRP seems to keep this same behavior, though in renamed files.

Does anyone know why VR/using Texture2DArrays for the camera target requires the LWRP to use an intermediate texture and then blit to the final framebuffer? Though we've seen an awesome performance boost from disabling this behavior I'm weary of deleting things without understanding them.

Thanks!

 

IIRC it's a current issue with the core engine implementation, but I could be conflating that with one of the multiple other underlying core engine issues like MSAA and Depth.

 

I did see the comment in the LWRP about requiring a depth prepass for MSAA that you might be referencing, and funny enough have also disabled that when building for Android. MSAA still works on the Go so I assumed it was a desktop exclusive problem. I'm guessing the Texture2DArray FinalBlit behavior is the same type of thing.