[fluorite] Fixes AMD iGPU SEGFAULT

[kerberjg]

Relates to:

• https://github.com/toyota-connected/fluorite/issues/1

Tentative fix to SEGFAULT crashes when creating VBOs for meshes on AMD iGPU in toyota-connected/fluorite

@jwinarske Needs testing on following devices:

• AMD Ryzen 9 + AMD Radeon Graphics

[jwinarske]

nanopc-t6 (rk3588) - max API 1.2

FEngine (64 bits) created at 0xffff700052d0 (threading is enabled)
FEngine resolved backend: Vulkan
Postcondition
in selectPhysicalDevice:601
reason: Unable to find suitable device.
#0	homescreen                          0xaaaae4d19088 (null) + 187650960101512
#1	homescreen                          0xaaaae4d18dc8 (null) + 187650960100808
#2	homescreen                          0xaaaae4d18ccc (null) + 187650960100556
#3	homescreen                          0xaaaae4c182bc (null) + 187650959049404
#4	homescreen                          0xaaaae4c18048 (null) + 187650959048776
#5	homescreen                          0xaaaae4c0f988 (null) + 187650959014280
#6	homescreen                          0xaaaae4c15188 (null) + 187650959036808
#7	libc.so.6                           0xffffac411078 (null) + 281473571688568


libc++abi: terminating due to uncaught exception of type utils::PostconditionPanic
Aborted                    (core dumped) homescreen -b /usr/share/flutter/toyota-connected-tcna-packages-filament-scene-fluorite-examples-demo/3.32.8/release/ -f

vulkaninfo snippet

GPU0:
VkPhysicalDeviceProperties:
---------------------------
	apiVersion        = 1.2.311 (4202807)
	driverVersion     = 25.1.6 (104861702)
	vendorID          = 0x13b5
	deviceID          = 0xa8670000
	deviceType        = PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
	deviceName        = Mali-G610
	pipelineCacheUUID = f09e9b4d-67a8-7061-6e00-000000000000

Switching patch to be 1.2 it ends up with the same segfault.

[kerberjg]

nanopc-t6 (rk3588) - max API 1.2

@jwinarske There's a chance it might not be compatible :(

RaspberryPi 5 supports Vulkan 1.3, we might have better luck there. If not, I do have a second suspect but I'd rather wait for the test results

[jwinarske]

@kerberjg raspberrypi5

FEngine (64 bits) created at 0x7fff580052d0 (threading is enabled)
FEngine resolved backend: Vulkan
Vulkan device driver: V3DV Mesa Mesa 25.1.6
Selected physical device 'V3D 7.1.7.0' from 1 physical devices. (vendor 0x14e4, device 0x55701c33, driver 0x6401006, api 1.3)
Backend feature level: 2
FEngine feature level: 1
[15:00:42.104297] [I] All systems initialized
...
[15:00:46.458830] [I] flutter: Camera 12 enabled
[15:00:46.458855] [I] From SetActiveCamera You're calling to get a system from an off thread, undefined experience! Use a message to do your work or grab the ecsystemmanager strand and do your work.
MESA: error: Failed to allocate device memory for BO
MESA: error: Failed to allocate device memory for BO
Postcondition
in VulkanTexture:474
reason: Unable to allocate image memory. error=-2

vulkaninfo snippet

GPU0:
VkPhysicalDeviceProperties:
---------------------------
	apiVersion        = 1.3.311 (4206903)
	driverVersion     = 25.1.6 (104861702)
	vendorID          = 0x14e4
	deviceID          = 0x55701c33
	deviceType        = PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
	deviceName        = V3D 7.1.7.0
	pipelineCacheUUID = 6cd98fa4-ad43-94fa-a479-b44b255aa35d

[kerberjg]

@jwinarske Bummer, I really thought I had it :( well, back to the drawing board for ARM

Does it mitigate the crash on AMD iGPU at least? It oughta, the specific crash we got was because a Vulkan v1.3 function was being called on v1.1

[jwinarske]

@kerberjg the hit rate on Ryzen9 for me; rare

[kerberjg]

@jwinarske I went from 90% crashes to 0%. Let's talk tomorrow and see what we can do

[kerberjg]

@jwinarske Bad news, repro'd even with fix enabled, identical stack trace. Made more progress on researching the issue though