Drop frames when the LGMP queue is full instead of blocking
while holding an IddCx surface. Always finish every
acquired frame, including duplicate frame numbers, so
replug teardown cannot trigger the release-frame watchdog.
Render ImGui to an intermediate SDR framebuffer and map it into the
native PQ or scRGB output using the Wayland surface reference white.
Preserve premultiplied alpha during conversion and limit composition to
damaged overlay regions.
Generate the EDID once during adapter initialization and update only the
live mode list when resolutions change.
Changing the EDID during a monitor replug causes Windows to identify it
as a different monitor, losing the existing display configuration.
Reusing the original EDID preserves the monitor identity while still
allowing dynamic modes to be advertised through IddCx.
LGMP session validation may wait 500 ms for the shared-memory heartbeat
to advance. Run it on a worker so Wayland input remains responsive while
preserving stale-data detection.
Match the EGL client’s DRM formats, use implicit linear modifiers, and
reset cached imports safely across format changes and restarts. Fall
back to CPU upload if either DMA buffer cannot be imported.
Submit the D3D12 copy before exposing the frame through LGMP, preventing
clients from waiting on work that has not yet been queued. Track the
last published buffer separately for safe subscriber resends and handle
submission failures.
Build the active filter chain only when its configuration changes,
including filter settings or order, input format, dimensions, output
size, and DMA mode.
Unchanged frames now execute only the cached active filters, avoiding
repeated setup, preparation, and output-resolution queries.
Reconfiguration forces a full-frame pass to prevent stale intermediate
framebuffer contents.
Also refresh CAS and FSR uniforms when loading presets and move
downscale pass selection out of the per-frame path.
Track context-local GL bindings to avoid redundant program, framebuffer,
buffer, texture, sampler, viewport, blend, and scissor state changes.
Invalidate the cache across context switches, ImGui rendering, and
shared object changes. Preserve CPU-backed texture upload correctness by
explicitly clearing pixel unpack buffer bindings where required.
This reduces hot-path driver overhead without adding waits or GPU
synchronization.
Track PBO fences per upload slot and wait only when a slot is reused,
allowing rendering to proceed without a CPU stall. Use GPU-side waits
for cross-context DMABUF synchronization.
Also propagate KVMFR write-completion failures so incomplete
shared-memory frames are never submitted, while retaining incremental
damage-copy waits to avoid added latency.
Replace the per-frame uniform list with stable, name-cached handles that
retain their values across shader recompilation. Track dirty uniforms so
only changed values are uploaded, reducing redundant GL calls and buffer
management.
Add a reusable effect-pass pipeline with per-pass shaders, samplers,
formats, resolutions, and intermediate render targets. This makes
external effects easier to integrate and supports arbitrary multipass
chains.
Migrate the desktop, cursor, damage, 24-bit, downscale, CAS, and FSR
rendering paths to the new APIs. Also ensure framebuffer objects are
released correctly.
While it would make much more sense to use these then a full replug to
change modes, Microsoft have not properly implemented the API to clear
the cached monitor mode states internally, making these calls useless.
Revert to just replugging the device on mode change
Also refactored the versioning code to be more modular.
The old registry code was turned into a fallback and simplified, since
C++11 and newer guarantees the byte past the end of the std::string is
valid for null-terminator writes.
There is no reason to call a bunch of stuff before the xdg_surface is
configured, so let's not call them and cause early painting to happen
should the compositor decide to send the `preferred_scale` message early.
