XR_KHR_vulkan_enable2:
physicalDevice VkPhysicalDevice must match the device specified by xrGetVulkanGraphicsDevice2KHR
XR_KHR_vulkan_enable:
physicalDevice VkPhysicalDevice must match the device specified by xrGetVulkanGraphicsDeviceKHR
XR_KHR_vulkan_enable:
Add a trivial check that xrGetVulkanGraphicsDeviceKHR is called before xrCreateSession.
(our cached suggested device will be XR_NULL_HANDLE if it has not been called).
The XR_KHR_vulkan_enable2 code path already contains this check.
The problem:
* xrCreateVulkanDeviceKHR is passed a VkPhysicalDevice, but not a VkInstance.
* xrGetVulkanGraphicsDevice2KHR is passed a VkInstance and returns a VkPhysicalDevice
that is a child of that instance.
* xrCreateVulkanDeviceKHR must verify that the xrGetVulkanGraphicsDevice2KHR
has been called and that the passed VkPhysicalDevice matches the one returned
by xrGetVulkanGraphicsDevice2KHR.
We have to consider:
* xrCreateVulkanDeviceKHR has to work on the "correct" VkInstance, which the passed
VkPhysicalDevice is a child of.
The reqirement
> If the vulkanPhysicalDevice parameter does not match the output of
> xrGetVulkanGraphicsDeviceKHR, then the runtime must return XR_ERROR_HANDLE_INVALID.
is not 100% clear whether calling xrCreateVulkanInstance multiple times is allowed
and how a second call to xrGetVulkanGraphicsDevice2KHR with a dfferent VkInstance
should be handled.
For this implementation xrCreateVulkanDeviceKHR will only consider the most recent call
to xrGetVulkanGraphicsDevice2KHR and the VkInstance that was used for this call.
This enforces at least that VkPhysicalDevice is a child of the cached VkInstance when
xrCreateVulkanDeviceKHR is called, because using a different VkPhysicalDevice would be
an error.
As before, on the service side the GPU index the compositor runs on can be selected with
* XRT_COMPOSITOR_FORCE_GPU_INDEX=INDEX1
By default xrGetVulkanGraphicsDevice() will suggest the same GPU the compositor runs on.
It is also possible to override the GPU index suggested to applications with
* XRT_COMPOSITOR_FORCE_CLIENT_GPU_INDEX=INDEX2
The reason this is both done on the service side is that if compositor and client run
on different GPUs, the swapchains use linear tiling instead of optimal tiling.
To make chosen GPUs comparable across the compositor's and the client's vulkan instance,
VkPhysicalDeviceIDProperties.deviceUUID is used.
