Decode PQ/BT.2020 desktop pixels into an FP16 linear scRGB target before
compositing the color cursor, ImGui, damage overlay, and letterboxing.
Encode the damaged result to PQ once at the surface boundary. Keep logical
AND/XOR cursor masks in the encoded domain.
This preserves premultiplied cursor edges, prevents encoded-domain
blending, uses the Wayland output reference white for local overlays, and
keeps guest cursor white and display calibration independent from frame
SDR white. It also corrects HDR-to-SDR absolute luminance scaling, gamut
matrices, black handling, and buffer-age repair for the extra pass.
Suppress overlays when their conversion framebuffer is unavailable rather
than allowing unconverted sRGB draws into an HDR target.
Use named transfer constants for cursor shader state. Convert clipped
surface rectangles explicitly before using the frame-damage merge helper,
which avoids mixing its unsigned type with the EGL surface rectangle type.
Only advertise native PQ when the FP16 composition target is available.
SDR and native scRGB retain their direct single-pass paths.
Request true floating-point EGL configs for scRGB and expose the native
encodings supported by the selected surface. Fall back to software HDR
mapping when the surface cannot represent the incoming wire format.
Resample PQ in linear light, keep SDR-only effects out of native PQ
chains, and consume the cursor-specific white level without changing
binary monochrome cursor mask operations.
Honor the IddCx 3x4 XYZ matrix and post-transfer LUT in the IDD
conversion pipeline. Carry transform changes to the client so EGL applies
the same calibration to hardware cursors, and invalidate the full frame
whenever calibration changes.
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.
According to Erik @ NVidia the open source NVidia driver will not
create a EGLImage from a DMABUF if the target is not
GL_TEXTURE_EXTERNAL_OES. This change set converts the dmabuf texture
from GL_TEXTURE_2D to GL_TEXTURE_EXTERNAL_OES and at runtime performs a
global search & replace on fragment shaders as needed to remain
compatible, replacing `sampler2D` with `samplerExternalOES`.
Ref: https://github.com/NVIDIA/open-gpu-kernel-modules/discussions/243#discussioncomment-3283415
This fixes a rounding issue on certain hardware (NVidia) which actually
implement mediump as half precision (16-bit) float. It's safe to assume
`highp` is available as if the GPU does not support it, then the shader
compiler will try to find a lower precision that is supported by the GPU
This saves a lot of GPU power for partial updates. Running testufo with
lanczos downscaling and FSR upscaling consumed over 90 W, but with this
commit, consumed only 75 W.
Using 4x4 means that some pixels will be outside of the lanczos window. The
ideal lanczos function should in fact be zero in those areas, so we shouldn't
waste time processing those pixels.
I can't notice any difference in the results.
gl_Position is expected to be using homogeneous coordinates, which requires
w to be a coordinate scale factor, usually 1.0. z should also be set in order
for depth to be well-defined. Therefore, we should set gl_Position.zw to
vec2(0.0, 1.0).
