Pass the GLsync object from texture into our
command queue, and when executing the queue,
wait on the sync object the first time we
use its associated texture.
When adding mask nodes, I overlooked that
we have two separate functions for determining
what transforms a node supports without offlines.
Since we claim that mask nodes support general
transform, they must certainly support 2d transforms
as well.
They're not needed and GLES doesn't technically support them, even
though GTK had been using them via epoxy sneakily using the
GL_OES_vertex_array_object extension behind our back.
gsk_vulkan_render_download_target() currently resets the uploader
objects before downloading the image that it produces. This is
problematic because there might be unreleased buffers and images
in the command queue.
In particular, this can make validation layers complain about the
glyph atlas - of all things! - upload buffer being released while
still being used by the command queue.
Fix that by resetting the uploader after downloading the image.
The current implementation of the glyph cache deals with atlases by
padding them with 1 pixel at the beginning, at the end, and between
each glyph.
That's cool and all, however, there's a very subtle problem with
this approach: the contents of the atlas are garbage, so this padding
is filled with garbage memory!
Rework the Vulkan glyph cache to draw each and every glyph in a
surface that has 1 pixel border of padding around it. Ensure the
surface is completely black by drawing a rectangle before handing
it to Pango to draw the glyph. Update tx and ty to pick the texture
position adjusted to the 1 pixel padding. The atlas now starts at
position (0, 0), since each glyph individually contains its own padding.
To improve legibility, add a PADDING define and use it everywhere.
Vulkan renders text using VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA and
VK_BLEND_FACTOR_SRC_ALPHA, but that implies per-channel alpha
blending, which currently produces the wrong results when blending
glyphs with the images beneath them.
Use the default pipeline constructors, which implies using the
ONE and ONE_MINUS_SRC_ALPHA.
Basically what GL does, but without any debug or feature flag
to gatekeep it, since the Vulkan backend itself is experimental
already.
Ceil surface sizes, and floor coordinates, to the fractional scale
value.
The rects passed to the clip region are in buffer coordinates, and
must not be scaled. Consider the following scenario: Wayland, with
a 1024x768@2 window. That gives us a 2048x1536 raw image. To setup
the Vulkan render pass code, we'd scale 2048x1536 *again*, to an
unreasonable 4196x3072, which is (1) incorrect and (2) really
incorrect and (3) can lead to crashes at best, full GPU resets
at worst - and a GPU reset is incredibly not fun!
Now that we pass the right clip regions at the right coordinates
at all times, remove the extra scaling from the render pass.
This part of the Vulkan renderer is almost exactly equal to the GL
renderer, and the GL renderer already does that since at least
2a38cecd33. Copy that into the Vulkan renderer.
A nice side effect from this commit is that resizing a window now
actually works again.
Sneak in a trivial cleanup by using a variable to hold the draw
index.
This was a tricky one to figure out, but it's pretty simple to
understand (I hope!).
So, this AMD card I'm using requires buffer memory sizes to be
aligned to 16 bytes. Intel is aligned to 4 bytes I think, but
AMD - or at least this AMD model in particular - uses 16 bytes
for alignment.
When creating a a particular texture (I did not determin which one
specifically!) a buffer of size 1276 bytes is requested.
1276 / 16 = 79.75, which is clearly not aligned to the required
16 bytes.
We request Vulkan to create a buffer of 1276 bytes for us, it
figures out that it's not aligned, and creates a buffer of 1280
bytes, which is aligned. The extra 4 bytes are wasted, but that's
okay. We immediately query this buffer for this exact information,
using vkGetBufferMemoryRequirements(), and proceed to create actual
memory to back this buffer up.
The buffer tells us we must use 1280 bytes, so we pass 1280 bytes
and everyone is happy, right? Of course not. We pass 1276 bytes,
and Vulkan is subtly unhappy at us.
Fix that by passing the value that Vulkan asks us to use, i.e.,
the size returned by vkGetBufferMemoryRequirements().
This is what GL does, and for a reason: it can lead to width or
height for very small glyphs. Also, switch to dividing by a float
(1024.0) instead of an integer (1024).
This doesn't make any difference now, but will allow us to copy
subregions more easily. This is not obvious, but here's a quick
explanation:
Leaving 'bufferRowLength' and 'bufferImageHeight' implies that
Vulkan will assume the size passed in the 'imageExtent' field.
Right now, this assumption is correct - the only user of this
function is the glyph cache, and it only copies and uploads
exact rects. Next commits will change that assumption, so we
must pass 'buffer*' fields, and tell Vulkan, "this part of the
buffer represents an image of width x height, and I want the
subregion (x, y, smallerWidth, smallerHeight) of this image".
When creating an image using gsk_vulkan_image_new_for_framebuffer(),
it passes VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL.
However, this is a mistake. The spec demands that the initial
layout must be either VK_IMAGE_LAYOUT_UNDEFINED or
VK_IMAGE_LAYOUT_PREINITIALIZED.
Apparently this was an oversight from commit b97fb75146, since the
commit message even documents that, and all other calls pass either
VK_IMAGE_LAYOUT_UNDEFINED or VK_IMAGE_LAYOUT_PREINITIALIZED.
Create framebuffer images using VK_IMAGE_LAYOUT_UNDEFINED, which is
what was originally expected.
Fractional scaling with the GL renderer is
experimental for now, so we disable it unless
GDK_DEBUG=gl-fractional is set.
This will give us time to work out the kinks.
This commit combines changes in the Wayland backend,
the GL context frontend, and the GL renderer to switch
them all to use the fractional scale.
In the Wayland backend, we now use the fractional scale
to size the EGL window.
In the GL frontend code, we use the fractional scale to
scale the damage region and surface in begin/end_frame.
And in the GL renderer, we replace gdk_surface_get_scale_factor()
with gdk_surface_get_scale().
Instead of tracking a single scale, track x and y scales separately.
Factor out gsk_vulkan_render_pass_new() into a private function that
receives both scales, and pass 'scale_factor' for both.
This is mostly a cosmetic change, and the goal is twofold:
1. Make it easier to spot unimplemented render node types; and
2. Prepare for a small rework
The implementation for each node now lives in specific functions,
like the GL renderer; unlike the GL renderer, however, we use a
node type vtable to map GskRenderNodeType → implementation. Render
node without an implementation map to NULL, and use the fallback
implementation. Render nodes that fail any check and return FALSE
also use fallback implementation.
If we encounter a node or texture the 1st time and they are going
to be used again, give them a name.
Then, when encountering them again, print them by name instead
of duplicating them.
We extend the syntax for nodes from:
<node-type> { ... }
to
<node-type> { ... }
<node-type> <string> { ... }
<string>;
where the first is the same as before, the 2nd defines a named node and
the last references a previously defined node.
Or to give an example:
color "node" {
bounds: 0 0 10 10;
color: red;
}
transform {
bounds: 20 0 10 10;
child: "node";
}
This will draw the red box twice, once at (0,0) and once at
(20,0).
The intended use for this is both shortening generated node files as
well as allowing to write tests that reuse nodes, in particular when
dealing with caches.
We extend the syntax for textures from just:
<url>
to
[<string>] <url>
<string>
where the first defines a named texture while the second references a
texture.
Or to give an example:
texture {
bounds: 0 0 10 10;
texture: "foo" url("foo.png");
}
texture {
bounds: 20 0 10 10;
texture: "foo";
}
This will draw the texture "foo.png" twice, once at (0,0) and once at
(20,0).
The intended use for this is both shortening generated node files as
well as allowing to write tests that reuse textures, in particular when
mixing them in texture and texture-scale nodes.
When the GL texture already has a mipmap, we don't
have to download and reupload it to generate one.
We differentiate the handling for texture scale nodes,
where we do want to force the mipmap creation even if
it requires us to reupload the GL texture, and plain
texture nodes, where we just take advantage of a
preexisting mipmap to allow trilinear filtering for
downscaling, or create one if we have to upload the
texture anyway.
Store texture coordinates for each slice
instead of assuming 0,0,1,1, and generate
overlapping slices to allow for proper mipmaps.
This almost fixes trilinear filtering with
sliced textures.
We cheat and just set the texture parameters instead and hope nothing
explodes.
So far it didn't.
This is only needed to support GLES 2.0 so it's quite a limited set of
hardware these days.
Instead of uploading a texture once per filter, ensure textures are
uploaded as little as possible and use samplers instead to switch
different filters.
Sometimes we have to reupload a texture unfortunately, when it is an
external one and we want to create mipmaps.
When filtering changes for an already-cached
texture, we need to clear the render data
before setting the new one, otherwise it
does not take and we end up reuploading
the texture every frame.
Allow to set max texture size using the
GSK_MAX_TEXTURE_SIZE environment variable.
We only allow to lower the max (for obvious
reasons), and we don't allow values smaller
than 512 (since our atlases use that size).
This allows dropping or copy/pasting rendernodes into apps that accept
SVGs.
Not sure how useful this is because we advertise text/plain from
rendernodes already and we prefer that.
In certain scenarios, address the issue where gnome.compile_resources
fails to transmit the present source directory. This is most notably
visible with MSBuild.
Use the same approach and only create an offscreen
that is big enough for the clipped part of the scaled
texture.
If the clipped part is still too large for a single
texture, we give up and just render the texture without
filters (using the regular texture rendering code path
which supports slicing).
The following commit will add the texture-scale-magnify-10000x
test which fails without this fix.
Scale nodes can use large scale factors and we don't want to create
insanely huge Cairo surfaces.
A subsequent commit will add the texture-scale-magnify-10000x
test which fails without this fix.
Cairo surfaces are created transparent.
And even if they weren't, overdrawing with transparency wouldn't erase
what's in the surface because it's a no-op.
It would require CAIRO_OPERATOR_CLEAR or CAIRO_OPERATOR_SOURCE.
The API docs outline why quite well.
This should make it possible to do saving of textures to image files
without any private API with the same featureset that GTK uses.
Also remove the gsktextureprivate.h include where
gdk_texture_get_format() was the only reason for it.
When we truncate the command queue because it
is too big, we were messing up our state accounting
and running into criticals as a consequence.
This can be reproduced by opening a well-populated
fishbowl demo in the inspectors recorder.
Fixes: #5188
Add GskMaskNode, and support it in the render node
parser, in the inspector and in GtkSnapshot.
The rendering is just fallback for now.
Based on old work by Timm Bäder.
By dividing the blur radius to obtain the clip radius, we may end up
with halved values that result in an overshunk clip mask. Extend this
so that we ensure to cover the last pixel.
Fixes artifacts seen with the cairo renderer in X11 when resizing
windows horizontally, a black 1px high line would be seen in the
top of the window due to these outset bounds being used in clipping.
More mysteriously, also seems to fix resize lag in the GL renderer
(also X11), if e.g. the bottom-right corner of a window is resized
diagonally in bottom-left -> top-right direction, or
bottom-right -> top-left.
Related: https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2175#note_1599335
Instead of asserting only in debug builds (which are generally not
shipped in distributions) we should deliver a critical log-level message
so that these can be found sooner when not developing with jhbuild,
Flatpak, etc.
Also assert that we've setup the state correctly when realizing the
GskGLRenderer object.
Fixes#4625
Those property features don't seem to be in use anywhere.
They are redundant since the docs cover the same information
and more. They also created unnecessary translation work.
Closes#4904
Having the initial layout set to VK_IMAGE_LAYOUT_GENERAL causes issues
when going from the final layout to the initial layout since the image
layout is expected to be the general layout. Setting the initial layout
to undefined doesn't have this restriction.
We now collect this information during node
construction, so use it here.
The concrete change here is that we now avoid
offscreens for container nodes with multiple children,
as long as they don't overlap. In particular, this
avoid offscreens for ellipsized dim labels.
This fixes two issues with the offscreen rendering code for nodes with
bounds not aligned with the pixel grid:
1.) When drawing to an offscreen buffer the size of the offscreen buffer
was rounded up, but then later when used as texture the vertices
correspond to the original bounds with the unrounded size. This could
then result in the offscreen texture being drawn onscreen at a slightly
smaller size, which then lead to it being visually shifted and blurry.
This is fixed by adjusting the u/v coordinates to ignore the padding
region in the offscreen texture that got added by the size increase from
rounding.
2.) The viewport used when rendering to the offscreen buffer was not
aligned with the pixel grid for nodes at coordinates not aligned with
the pixel grid. Then because the content of the offscreen buffer is not
aligned with the pixel grid and later when used as textures sampling
from it will result in interpolated values for an onscreen pixel. This
could also result in shifting and blurriness, especially for nested
offscreen rendering at different offsets.
This is fixed by adding similar padding at the beginning of the
texture and also adjusting the u/v coordinates to ignore this region.
Fixes: https://gitlab.gnome.org/GNOME/gtk/-/issues/3833
This moves a lot of the texture atlas control out of the driver and into
the various texture libraries through their base GskGLTextureLibrary class.
Additionally, this gives more control to libraries on allocating which can
be necessary for some tooling such as future Glyphy integration.
As part of this, the 1x1 pixel initialization is moved to the Glyph library
which is the only place where it is actually needed.
The compact vfunc now is responsible for compaction and it allows for us
to iterate the atlas hashtable a single time instead of twice as we were
doing previously.
The init_atlas vfunc is used to do per-library initialization such as
adding a 1x1 pixel in the Glyph cache used for coloring lines.
The allocate vfunc purely allocates but does no upload. This can be useful
for situations where a library wants to reuse the allocator from the
base class but does not want to actually insert a key/value entry. The
glyph library uses this for it's 1x1 pixel.
In the future, we will also likely want to decouple the rectangle packing
implementation from the atlas structure, or at least move it into a union
so that we do not allocate unused memory for alternate allocators.
This removes the sharing of atlases across various texture libraries. Doing
so is necessary so that atlases can have different semantics for how they
allocate within the texture as well as potentially allowing for different
formats of texture data.
For example, in the future we might store non-pixel data in the textures
such as Glyphy or even keep glyphs with color content separate from glyphs
which do not and can use alpha channel only.
This allows the gskglprograms.defs a bit more control over how a shader
will get generated and if it needs to combine sources. Currently, none of
the built-in shaders do that, but upcoming shaders which come from external
libraries will need the ability to inject additional sources in-between
layers.
If the max_entry_size is zero, then assume we can add anything to the
atlas. This allows for situations where we might be uploading an arc list
to the atlas instead of pixel data for GPU font rendering.
When large viewports are passed to gsk_renderer_render_texture(), don't
fail (or even return NULL).
Instead, draw multiple tiles and assemble them into a memory texture.
Tests added to the testsuite for this.
There are situations where our "default framebuffer" is not actually
zero, yet we still want to apply a scissor rect.
Generally, 0 is the default framebuffer. But on platforms where we need
to bind a platform-specific feature to a GL_FRAMEBUFFER, we might have a
default that is not 0. For example, on macOS we bind an IOSurfaceRef to
a GL_TEXTURE_RECTANGLE which then is assigned as the backing store for a
framebuffer. This is different than using gsk_gl_renderer_render_texture()
in that we don't want to incur an extra copy to the destination surface
nor do we even have a way to pass a texture_id into render_texture().
If the rendering operation is over an opaque region, we can potentially
avoid clearing a large section of the framebuffer destination. Some cases
you do want to clear, such as when clearing the whole contents as some
drivers have fast paths for that to avoid bringing data back into the
framebuffer.
Since now we have the shaders working on Windows under GLES with libANGLE using
a 3.0+ context, drop the check to fall back to the Cairo renderer when GLES is
being used.
Various transforms are normalized with their next transform, and if they
end up being the identity transform will return NULL.
For example a translation by (x,y,z) and followed by (-x,-y,-z) will
result in NULL.
Document the return value and more importantly, specify that a call to
`gsk_renderer_realize()` needs to be matched with a call
`gsk_renderer_unrealize()`.
Prevents issues like https://gitlab.gnome.org/GNOME/gtk/-/issues/4625
Instead of just passing major/minor, pass them twice, once for GL and
once for GLES. This way, we don't need to check for GL and GLES
separately.
If something is supported unconditionally, passing 0/0 works fine.
That said, I'd like to group the arguments somehow, because otherwise
it's just a confusing list of numbers - but I have no idea how to do
that.
Limit the diff region to 30 rectangles (randomly chosen because it
looked big enough to not trigger by accident and small enough to not
cause performance issues).
If the diff region gets more complicated, we abort to the parent node
and use its bounds as the diff region instead and then continue diffing
the rest of the node tree.
Fixes: #4560Fixes: #2396
For libANGLE to work with our shaders, we must use "300 es" for
the #version directive in our shaders, as well as using the non-legacy/
non-GLES codepath in the shaders. In order to check whether we are
using the GLSL 300 es shaders, we check whether we are using a GLES 3.0+
context. As a result, make ->glsl_version a const char* and make sure
the existing shader version macros are defined apprpriately, and add a
new macro for the "300 es" shader version string.
This will allow the gtk4 programs to run under Windows using EGL via
libANGLE. Some of the GL demos won't work for now, but at least this
makes things a lot better for using GL-accelerated graphics under Windows
for those that want to or need to use libANGLE (such as those with
graphics drivers that aren't capable of our Desktop (W)GL requirements in
GTK.
On Windows with nVidia drivers at least, when we create a legacy context
via wglCreateContext(), we may still get a (W)GL 4.x context. Allow
such contexts to also use GLSL version 130 instead of 110, so that
things do continue to work.
But don't call it too early, we only want to call it once we have
prepared the target.
This way, we guarantee that a GL context is always available and that it
is bound to the correct target.
Don't pass texture + rect, but instead have
gdk_memory_texture_new_subtexture()
and use it to generate subtextures and pass them.
This has the advantage of downloading the a too large texture only once
instead of N times.
It does not belong in GdkGLContext, it's a renderer thing.
It's also the only user of that API.
Introduce gdk_gl_context_check_version() private API to make version
checks simpler.
It turns out glReadPixels() cannot convert pixels and you are only
allowed to pass a single value into the function arguments. You need to
know which ones or things will explode.
GL is great.
Pass a format do GdkTextureClass::download(). That way we can download
data in any format.
Also replace gdk_texture_download_texture() with
gdk_memory_texture_from_texture() which again takes a format.
The old functionality is still there for code that wants it: Just pass
gdk_texture_get_format (texture) as the format argument.
Make a deep texture, if the render nodes have
high depth content.
For now, we use 32F here for the deep format,
since using 16F causes small rounding errors
that break the memorytexture roundtrip tests.
Look at the framebuffer and the rendernode to
determine what format to use for intermediate
textures.
Our preference here is to use fp16, if we have it
and it makes sense for the framebuffer we're given.
Add private api to find out if the content
of a render node should be considered 'deep'.
The information is collected at creation time,
so there is no tree-walking involved when we
are using this information in the renderer.
Currently, this comes down to whether there are
any texture nodes with high depth textures in the subtree.
In the future, we may want to allow marking gradient
nodes in this way as well.
For MemoryTexture, this is a simple change.
For GLTexture, we need to query the format at texture creation. This
sounds like a bad idea and extra work until one realizes that we'd
need to do that anyway when using the texure the first time - either
when downloading, or when trying to use it in a rendernode, where we
will soon need that information to determine if the texture prefers high
depth.
Also, now make gdk_memory_convert() the only conversion functions
and allow conversions between any 2 formats by going via a float[4].
This could be optimized via fast-paths, but so far it isn't.
We never put large icons into the icon cache,
so all its items are always atlased, but we do
put large glyphs in to the glyph cache, and we
were never freeing those items, even when they
go unused. Fix that.
This reverts commit 87af45403a.
I've found that this change is needed to ensure that the
bounding boxes of text nodes encompass all the glyphd drawing.
Without it, we overdraw the widget boundaries and cut off
glyphs.
We are rendering the glyphs on a larger surface,
and we should avoid introducing unnecessary
rounding errors here. Also, I've found that
we always need to enlarge the surface by one
pixels in each direction to avoid cutting off
the tops of large glyphs.
For 2D transforms, we can read the scale
factors more directly off the matrix.
This should eventually be moved out into a
function to decompose a 2D transform into
scale + rotation + skew + translation.
We avoid an offscreen if we know the child node
can 'handle' the transform. Shadow nodes can if their
child node does - either the child node is a text node
in which case the shortcuts we take for shadow nodes
will work fine with the transform (we just render the
text node offset), or the child is not a text node,
in which case we render the shadow to an offscreen
anyway.
This change makes the label-shadows reftest pass with
the GL renderer, not by fixing the issue but by avoiding
it.
For shadow nodes, we try pretty hard to avoid
rendering shadows, and and we have a shortcut
that just renders text offset, but we can try
harder to do nothing - if the text is offset
by zero, we don't need to draw it at all.
We need to use an offscreen whenever there is overlapping
children somewhere in the tree below, just checking the
direct child of the opacity node is not enough.
Fixes: #4261
