More precise GPU selection and multi-monitor corrections

When a system has several GPUs, choosing by a display name can leave too much ambiguity. The January 2026 work added a way to resolve an adapter LUID from its PCI bus number, with the supporting build and device-property changes needed for that path.

The useful result is more specific adapter selection. Instead of assuming a friendly name uniquely identifies the device, the driver can connect the choice to hardware information. That builds on the earlier work to expose device identity in logs and configuration.

The spring brought corrections to HDR bit depth and multi-monitor swapchain threading. The latter came through a community contribution and was merged in May. A multi-monitor driver has to keep each monitor's work associated with the right resources and threads; a fix there addresses a different problem from simply offering more resolutions.

Build maintenance continued too, including moving the Windows Driver Frameworks headers into a Git submodule and updating CI configuration. That makes the source layout and required dependencies more explicit for contributors trying to reproduce a build.

The history also includes an initial shader experiment that was reverted in January. I'm leaving it out of the active feature list. An experiment is useful evidence of what was tried, but its introduction commit alone doesn't establish what remains in the driver.

Taken together, this stage is about being more exact: identify the intended GPU, describe the HDR format correctly, and keep separate virtual monitors from interfering with one another's rendering work. These are recorded source changes; a release artifact and the Windows configurations it passes still need their own evidence.