Only initialize the Vulkan or EGL parts where possible.
When dmabufs or dmabuf formats are actually used, we still
initialize fully by creating both a Vulkan and EGL downloader.
This shortens the time to first commit from 149ms to 108ms.
Just use two individual fields, so we can track if we've already
created each one. This also matches the individual fields we have
for the dmabuf formats.
And change preference order of downloaders
Previously, our order was mmap > vulkan > egl.
But depending on the hw (discrete vs integrated gpu), mmap
can be catastrophically slower (on the order of 20ms vs 1.5s).
So, change the order to egl > vulkan > mmap.
Note that this currently has less effect than we'd like to,
since we don't let the downloaders claim linear formats.
This one got added in 66ba1f76ba but didn’t end
the sentence with a dot, and didn’t have its enum name between parentheses so
that people can debug more easily which error code got generated.
`thread_current_context` might be holding the last reference to
`previous`, in which case `gdk_gl_context_make_current` on the new
context will free `previous`, leaving it a dangling pointer.
Avoid this by making sure to hold a reference.
Fixes: 41cd0c6f "gl: Fix initial EGL context creation on X11"
Resolves: https://gitlab.gnome.org/GNOME/gtk/-/issues/6995
Signed-off-by: Simon McVittie <smcv@debian.org>
Set the alpha channel to "undefined" in this case.
Gimp doesn't seem to like this when opening the image and insists to
doing something with it, that's a bit of a bummer.
But it allows GTK to load RGBx textures.
It seems Mesa doesn't support that yet, but having it doesn't hurt.
And it allows drivers to allocate less memory for the swapchains,
because we don't need all the 4 images we request in minImageCount.
But drivers tend to take that minimum image count as gospel, so we need
to use a higher number to not cause lag in corner cases.
Removed via regex and grep.
The following were intentionally not removed:
- GtkImage:file: (attributes org.gtk.Property.set=gtk_image_set_from_file)
- GtkImage:resource: (attributes org.gtk.Property.set=gtk_image_set_from_resource)
As they have no getter and (setter PROP) without a (getter PROP) crash
gobject-introspection. This is fixed by
ad3118eb51.
The annotations should only be set when the name of the setter or getter
for a property "GtkClassName:prop-name" is not gtk_class_name_g(s)et_property_name.
linear will average all the pixels for the lod, nearest will just pick
one (using the same method as OpenGL/Vulkan, picking bottom right
center).
This doesn't really make linear/nearest filtering work as it should
(because it's still a form of mipmaps), but it has 2 advantages:
1. it gets closer to the desired effect
2. it is a lot faster
Because only 1 pixel is chosen from the original image, instead of
averaging all pixels, a lot less memory needs to be accessed, and
because memory access is the bottleneck for large images, the speedup is
almost linear with the number of pixels not accessed.
And that means that even for lot level 3, aka 1/8th scale, only 1/64 of
the pixels need to be accessed, and everything is 50x faster.
Switching gtk4-demo --run=image_scaling to linear/nearest makes all the
lag go away for me, even with a 64k x 64k image.
We have fast conversion functions, use those directly instead of calling
into gdk_memory_convert().
This is useful because as mentioned before, the main optimization here
is RGB8 => RGBA8 and we have a fastpath for that.
Why do we need this? Because RGB images are provided in RGB format but
GPUs can't handle RGB, only RGBA, so we need to convert.
And we need to do that without allocating too much memory, because
allocating memory is slow. Which means in aprticular we need to do the
conversion after mipmapping, not before (like we were doing).
This allows uploading less memory but requires computing lod levels on
the CPU which is slow because it reads through all of the memory and so
far entirely not optimized.
However, it uses significantly less VRAM.
This is done by adding a gdk_memory_mipmap() function that does this
task.
The texture upload op now accepts a lod level and if that is >0 it uses
gdk_memory_mipmap() on the source texture.
This is just the API. Users will come later.
I considered putting it into gdkmemoryformat.c because it's likely gonna
be the only user and this one function is so little code, but it didn't
fit at all.
So now it's a new file.
Depth of a rendernode should be determined by the textures used and the
compositing colorstate requirements.
Colors influence the colorstate choice, so they indirectly influence the
depth, but they should not influence the depth directly.
Otherwise a single color in a border being rec2100-pq would make us
switch to 16bit float.
Also remove gdk_color_get_depth(), because it was only used here and
because again: Colors should not influence depth decisions.
When prepare_gl fails in the right way (or the wrong way?), we
end up creating the leader window twice, and as a side effect,
creating two instances of the "Virtual core pointer" device, which
is bad news for grabs.
Fixes: #6840
