When Vulkan devices are enumerated, this also determines the available
present modes for each device, maps them to a vector, and gives
those options to the user.
OpenGL options are limited to On/Off.
Required creating a VkSurfaceKHR during device enumeration, which
may or may not be desireable. For the sake of a less confusing UI.
Also fixes a bug where if a graphics device disappears on the host, we
don't try and select the non-existant devices.
configure_graphics: Remove vsync runtime lock for Vulkan
configure_graphics: Recommend Mailbox present mode
configure_graphics: Fix type-limits warning
configure_graphics: Clean up includes
configure_graphics: Add tooltip
Uses mailbox, then immediate for unlocked framerate depending on
support for either. Also adds support for FIFO_RELAXED.
This function now assumes vsync_mode was originially configured to a value
that the driver supports.
vk_swapchain: ChooseSwapPresentMode determines updates
Simplifies swapchain a bit and allows us to change the present mode
during guest runtime.
vk_swapchain: Fix MSVC error
vk_swapchain: Enforce available present modes
Some frontends don't check the value of vsync_mode before comitting it.
Just as well, since a driver update or misconfiguration could problems
in the swap chain.
vk_swapchain: Silence warnings
Silences GCC warnings implicit-fallthrough and shadow, which apparently
are not enabled on clang.
Those vulkan settings do not correspond 1:1 to the swap intervals that
they set for OpenGL, so remove it.
bootmanager: Add missing include
I didn't add this log why did it break
Function is useful outside of bootmanager, so put it in a common place.
qt_common: Add missing include
qt_common: Add some newlines
qt_common: Add trailing newline
qt_common: Add trainline newline
We don't need the whole EmuWindow when creating a surface,
and it creates onerous requirements outside of typical usage for
creating a surface elsewhere.
Previously, yuzu would try and guess which vsync mode to use given
different scenarios, but apparently we didn't always get it right. This
exposes the separate modes in a drop-down the user can select.
If a mode isn't available in Vulkan, it defaults to FIFO.
Enables shadow-uncaptured-locals and implicit-fallthrough for Clang.
implicit-fallthrough is not enabled by default in -Wall or -Wextra, and
shadow-uncaptured-local doesn't seem to be enabled by default by
-Wshadow, even though GCC has both of these by their respective cases.
* During pipeline configure the function would acquire some payload space from the descriptor update queue,
write the descriptor data on the GPU thread and give the scheduler a pointer to the beginning of said space to update it later.
TickFrame resets the payload cursor, used to track acquires, back to the beginning of the buffer.
This wasn't a problem before since WaitWorker was called at the end of the frame but now it is.
If a frame writes to a cursor before the scheduler catches up, it will crash
* To fix this the payload buffer has been increased to account for the in flight frames that are allowed to exist now.
TickFrame will switch between the payload spaces instead of resetting
Intel's SPIR-V shader compiler is broken. For now, skip compiling any compute pipelines until they fix this issue.
This is not a perfect workaround, as there are a small subset of non-compute pipelines that still cause it to crash, but this should cover the majority of crashes.
It is unfortunate that even with a test case reported 6 months ago the issue has not been fixed in favor of fixing "the most popular games and apps".
Intel, you can do better than this.
On AMD a subpixel offset of 1/512 of the texel size is applied to the texture coordinates at a ImageGather call to ensure the rounding at the texel centers is done the same way as in Maxwell or other Nvidia architectures.
See https://www.reedbeta.com/blog/texture-gathers-and-coordinate-precision/ for more details why this might be necessary.
This should fix shadow artifacts at object edges in Zelda: Breath of the Wild (#9957, #6956).
Some games have very tight scheduling requirements for their audio which can't really be matched on the host, adding a constant to the reported value helps to provide some leeway.
