Windows vulkaninfo sometimes aborts with STATUS_INTEGER_DIVIDE_BY_ZERO

[lunarpapillo]

This first started appearing around loader 1.3.260. It's quite rare, only appearing a few times, and so far only on two CI test machines tcwinamd3 and tcwinnvi6. It's likely tied to the particular version of vulkaninfo or the loader, though the fact that it appears in two different SDKs is concerning.

The symptoms appear as a failure to run vulkaninfoSDK, which is used on every Vulkan CI job run to determine the configuraiton of the running machine:

INFO: executing #1: C:\VulkanSDK\1.3.275.0\Bin32\vulkaninfoSDK.exe -j
ERROR: execute() call failed after 0.28 seconds (0.28 seconds): returncode=3221225620 (0xc0000094=STATUS_INTEGER_DIVIDE_BY_ZERO); execution log index: 1, working directory: C:\j\workspace\setuptest\BITS\32\label\tcwinnvi6\vulkantest-results
ERROR: failure during run setup: execute(C:\VulkanSDK\1.3.275.0\Bin32\vulkaninfoSDK.exe -j) failed after 0.28 seconds (0.28 seconds) with returncode=3221225620 (0xc0000094=STATUS_INTEGER_DIVIDE_BY_ZERO)

So far it has appeared at least in these instances (it may have appeared in more developer runs that were never reported):

• 20240119 tcwinamd3 1.3.260
• 20240109 tcwinamd3 1.3.275
• 20240715 tcwinnvi6 http://tcubuser.lunarg.localdomain:8080/job/Vulkan-ValidationLayers/17075/BITS=64,BUILD_MODE=Debug,label=Windows-NVIDIA-20XX-rolling-NDA/console
• 20240729 tcwinnvi6 http://tcubuser.lunarg.localdomain:8080/view/Administration/job/setuptest/448/BITS=32,label=tcwinnvi6/console
• 20241210 tcwinnvi6 http://tcubuser.lunarg.localdomain:8080/job/Vulkan-ValidationLayers/18284/BITS=64,BUILD_MODE=Debug,label=Windows-NVIDIA-20XX-rolling-NDA/console

[lunarpapillo]

Note that we have the option to load a newer SDK on the affected systems and hope the (rare) CI problem goes away... but if we do so we lose the ability to reproduce this bug and potentially figure out and address why it crashes.