Converting the LCMS2 transform to a 3D LUT early causes issues:
- It's a lossy process, the consumer will not be able to pick a
3D LUT size on their own.
- It requires unnecessary conversions and allocations: an intermediate
3D LUT is allocated, but the renderer already allocates one.
- It makes it harder to support arbitrary color transforms in the
renderer, because each type needs to be handled differently.
Instead, expose a function to evaluate a color transform, and use
that to build the 3D LUT in the renderer.
Perform a primitive garbage collection of buffers that have not been
used in the past 10 seconds, an arbitrarily selected number.
As garbage collection also makes span buffer allocation happen much more
often, logging on allocation activity leads to a lot of log noise so get
rid of that while at it.
Color transform can have multiple types and these different types
want to store different metadata. We previously stored this metadata
directly on wlr_color_transform even for transforms that don't use it.
Instead, let's take the prior art from wlr_scene where each scene node
is built on a base node. Notice how wlr_color_transform_lut3d now has
a `struct wlr_color_transform base`. This is advantageous in multiple
ways:
1. We don't allocate memory for metadata that will never be used.
2. This is more type safe: Compositors can pass around a
struct wlr_color_transform_lut3d if they know they only want to use a
3d_lut.
3. This is more scalable. As we add more transform types, we don't have
to keep growing a monolithic struct.
The spec for VkMemoryPropertyFlagBits says:
> device coherent accesses may be slower than equivalent accesses
> without device coherence [...] it is generally inadvisable to
> use device coherent or device uncached memory except when really
> needed
We don't really need coherent memory so let's not require it and
invalidate the memory range after mapping instead.
Closes: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/3868
The mapping is shared between all users of the stage span, so it should
always map the whole thing and apply the allocation offset to the mapped
pointer.
The use of stage spans for 3dluts was missed when the new cached
mappings were introduced, meaning that it would try to map and unmap
memory that might already have a cached mapping.
Vulkan does not support mapping memory multiple times, so make sure the
3dlut code also uses the cached mapping to avoid segfaults after unmap.
UNDEFINED when used as source layout means that the contents of
the underlying memory becomes undefined. This isn't what we want
here: we don't want to mutate the imported pixel data.
The Vulkan spec isn't really clear what the proper value should be
here, but after discussing with driver developers [1] it seems like
UNDEFINED isn't the right one. The recommendation is to use GENERAL
instead.
[1]: https://github.com/ValveSoftware/gamescope/issues/356
This change makes it possible to support both the direct srgb-format
pipeline and indirect (color-managed, or non-srgb-format) pipeline
for the same render buffer.
Instead of having separate getters for shm formats and DMA-BUF
formats, use the same pattern as wlr_output_impl.get_primary_formats
with a single function which takes buffer caps as input.
Pre-multipled sRGB values need to be un-multiplied before conversion
to linear and then re-multiplied after. Compare shaders/texture.frag.
This fixes an issue in labwc where titlebar corners (rendered as
ARGB textures) did not match the rest of the titlebar (rendered as
a solid wlr_scene_rect).
Note: 0.17 has other instances in render/vulkan/renderer.c that need
the same fix.
