The implementation comes from mesa, which came from Berkeley SoftFloat
3e Library.
The software float implementation will be used to convert from and to
half floating point pixel format bitmaps, when no the CPU isn't capable
of the conversion.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3441>
The type of the intermediate medium for storing pixel channels is
changed from "is 8 or 16 bit" to an enum, and switch cases. This doesn't
add support for anything, but will make adding a "float" medium type
less intrusive.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3441>
There is no internal fp16 format which has no alpha which means we would
get garbage alpha when reading the framebuffer directly. We have to use
the packing/unpacking to always get the alpha of 1.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3441>
Which gltype and glformat are allowed for a given gl framebuffer depends
on the internal gl framebuffer format. Either the format we want to read
into matches the gltype, glformat and doesn't need CPU packing from
another format or we have to use a tmp buffer with a format that matches
the GL requirements.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3441>
Sometimes it's useful to combine the description from multiple places.
For example, a subsequent commit will add the output name as the first
thing to the frame clock span descriptions, leaving the rest of the
description with complex checks unchanged.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3417>
Tracy can filter its statistics by user text, in our case by span
description. It's useful to filter by actor type and name, and not so
much by the pointer. So, remove it, and also reduce the amount of
punctuation.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3417>
Transfer none was achieved using a stack GArray in the stage which
would get resized to 0 at the end of every frame to "free" it.
In the case of direct scanout however, painting the next frame only
happens after leaving fullscreen again. Until then the array just kept
growing and because GArrays don't reallocate when shrunk, this memory
remained allocated even after leaving fullscreen.
There is no cache benefit from storing paint volumes this way, because
nothing accesses them after their immediate use in the calling code.
Also the reduced overhead from avoiding malloc calls seems negligible as
according to heaptrack this only makes up about 2-3% of the temporary
allocations.
Changing this to transfer full and removing the stack array simplifies
the code and fixes the "leak".
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3191
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3442>
It was for a failed expriment that tried to mmap() dmabuf memory and
find damaged regions to decrease the amount that was eventually used to
write to an onscreen, but mmap:ing is only fast enough on intel, and
it's only relevant on various server GPUs. For it to be achievable, we
need to render to system memory in a way that we don't need to copy it
out of OpenGL, but that's currently not possible.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/3443>
This is a step in cleaning up the Clutter context management. By making
it a GObject it's easier to add e.g. properties and features that helps
with introspection.
For now, this means the context creation is changed to go via a
"constructor" (clutter_create_context()). This is so that the global
context singleton can be mantained outsid of ClutterContext, until it
can be removed.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2016>
The original purpose of being able to report errors is no longer
relevant, since the Clutter backend is now practically a thin wrapper
around the actual backend, which has already dealt with error reporting.
Thus move this to the regular constructor path.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2016>
