Unity + Quest: repeated VK_ERROR_INITIALIZATION_FAILED error

I've seen that VK_ERROR_INITIALIZATION_FAILED issue before on Quest Pro when moving a project to newer Unity + Oculus XR versions. In my experience it was mainly Vulkan unstable -- switching to OpenGLES3 in the Player Settings verify that immediately. It is also possible to clean out the shader cache, and then testing using Oculus XR plugin 3.2.x as 3.3.0 has been known to cause issues in conjunction with Vulkan on certain devices.

While I was working on troubleshooting my issues, I realized that I was having to go through endless trial-and-error loops and I began to track the duration using a basic work time calculator in order to keep an eye on the length of each build process was taking. It was a great way to stay in order when trying to figure out the cause.

Regards 
John Smith

Thanks, that makes sense. We're using Multi Pass -- for anyone else who's using Multi Pass, it sounds like disabling Symmetric Projection probably doesn't fix the issue if you're using Multi Pass unfortunately (based on the docs at https://docs.unity3d.com/Packages/com.unity.xr.oculus@2.0/manual/index.html it sounds like Symmetric Projection only has an effect when using Multiview).

A question for those who found that disabling Symmetric Projection fixed the issue -- what option do you have set for Stereo Rendering Mode?

Hmm. I tried in a test project with my old TextMesh Pro and didn't find any validation errors with a single simple text mesh, so there's more at play here.

Here are the validation errors that I'm seeing in Unity 2022.2.21f1. There is some overlap with my previous errors, but it seems that not all of them are specific to the GPU hang:

VULKAN: ERROR:  [Validation, 0]: Validation Error: [ VUID-vkCmdClearAttachments-rect-02682 ] Object 0: handle = 0xb4000078c9af0e20, type = VK_OBJECT_TYPE_COMMAND_BUFFER; | MessageID = 0xadbd476f | CmdClearAttachments(): pRects[1].rect.extent.width is zero. The Vulkan spec states: The rect member of each element of pRects must have an extent.width greater than 0 (https://www.khronos.org/registry/vulkan/specs/1.3-extensions/html/vkspec.html#VUID-vkCmdClearAttachments-rect-02682)

VULKAN: ERROR:  [Validation, 0]: Validation Error: [ VUID-VkImageViewCreateInfo-image-01762 ] Object 0: handle = 0x50c1c40000000176, type = VK_OBJECT_TYPE_IMAGE; | MessageID = 0x6516b437 | vkCreateImageView() format VK_FORMAT_R8G8B8A8_UNORM differs from VkImage 0x50c1c40000000176[] format VK_FORMAT_R8G8B8A8_SRGB.  Formats MUST be IDENTICAL unless VK_IMAGE_CREATE_MUTABLE_FORMAT BIT was set on image creation. The Vulkan spec states: If image was not created with the VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT flag, or if the format of the image is a multi-planar format and if subresourceRange.aspectMask is VK_IMAGE_ASPECT_COLOR_BIT, format must be identical to the format used to create image (https://www.khronos.org/registry/vulkan/specs/1.3-extensions/html/vkspec.html#VUID-VkImageViewCreateInfo-image-01762)

VULKAN: ERROR:  [Validation, 0]: Validation Error: [ VUID-vkResetFences-pFences-01123 ] Object 0: handle = 0x1d4e2e0000000062, type = VK_OBJECT_TYPE_FENCE; | MessageID = 0x68a5074e | VkFence 0x1d4e2e0000000062[] is in use. The Vulkan spec states: Each element of pFences must not be currently associated with any queue command that has not yet completed execution on that queue (https://www.khronos.org/registry/vulkan/specs/1.3-extensions/html/vkspec.html#VUID-vkResetFences-pFences-01123)

It seems that the Validation Error: [ VUID-vkResetFences-pFences-01123 ] is the only error that happened exclusively after the hang was observed, so my guess is that this is the problem error.

Hi there, Allen. I think we're seeing the same issue. We think the problem may be related to discarding fragments in the fragment shader.
Is it possible to see if removing discard or clip in your fragment shaders makes this problem go away?
I'm not saying that's a solution but it may help our Meta friends provide a real fix.

Turns out this problem also happens with Unity 2021.3.26f1 LTS.

And I realized that the actual error messages happen on a 0 process ID, so my logcat filter was removing them from my view. These are the actual errors that I'm seeing:

2023-05-26 15:50:34.502 0 0 Fatal kgsl-3d0 CP opcode error interrupt | opcode=0x00000000
2023-05-26 15:50:34.513 0 0 Fatal kgsl-3d0 MISC: GPU hang detected
2023-05-26 15:50:34.513 0 0 Error kgsl-3d0 dnobodyexplodes[8095]: gpu fault ctx 36 ctx_type VK ts 1235 status 00800005 rb 1124/1124 ib1 0000004003AE6000/0220 ib2 0000004003CE4000/0000
2023-05-26 15:50:34.513 0 0 Error kgsl-3d0 dnobodyexplodes[8095]: gpu fault rb 2 rb sw r/w 1124/1205
2023-05-26 15:50:34.513 0 0 Error platform 3d6a000.qcom,gmu Suspended GMU
2023-05-26 15:50:34.527 0 0 Fatal kgsl-3d0 CP opcode error interrupt | opcode=0x00000000
2023-05-26 15:50:34.536 0 0 Fatal kgsl-3d0 MISC: GPU hang detected
2023-05-26 15:50:34.537 0 0 Error platform 3d6a000.qcom,gmu Suspended GMU
2023-05-26 15:50:34.537 0 0 Warn healthd battery l=97 v=4272 t=30.2 h=2 st=4 c=-257324 fc=3519000 cc=57 chg=u
2023-05-26 15:50:34.546 8095 8168 Warn Adreno-GSL <log_gpu_snapshot:471>: panel.gpuSnapshotPath is not set.not generating user snapshot
2023-05-26 15:50:34.547 8095 8168 Warn Adreno-GSL <ioctl_kgsl_syncobj_create:5290>: (44, 24, 1238) fail 22 Invalid argument
2023-05-26 15:50:34.547 8095 8168 Warn Vulkan_Utils Vulkan error: vulkanLoader.vkQueueSubmit(VulkanContext.queue, 0, NULL, fenceUsed): 'VK_ERROR_INITIALIZATION_FAILED'(-3) at arvr/projects/xrruntime/mobile/VrRuntime/Client/VrRuntimeClient/VrRuntimeClient.cpp:2603
2023-05-26 15:50:34.541 0 0 Error kgsl-3d0 dnobodyexplodes[8095]: gpu failed ctx 36 ts 1236
...

This problem also happens with Unity 2022.2.21f1:

• Unity 2022.2.21f1
• Oculus XR Plugin 4.4.0
• Oculus Integration 53.1 (1.85.0)

I can also reproduce this exact same issue with a Quest 2:

• System Version 50543050028800150
• Version 54.0.0.107.344.474388434
• Runtime Version 54.0.0.110.348.475023458
• OS Version SQ3A.220605.009.A1