Sets timestamps for `vkGetPastPresentationTimingEXT` for the remaining
present stages on headless:
- VK_PRESENT_STAGE_IMAGE_LATCHED_BIT_EXT
- VK_PRESENT_STAGE_IMAGE_FIRST_PIXEL_OUT_BIT_EXT
- VK_PRESENT_STAGE_IMAGE_FIRST_PIXEL_VISIBLE_BIT_EXT
These are all set to the same value: when swapchain::present_image is
called. It uses the best monotonic clock supported by the driver.
No need for slot member variables to be atomic any more because they
are only accessed while a lock on m_queue_mutex is held.
Signed-off-by: Alex Bates <alex.bates@arm.com>
Change-Id: I90ffc876890ee74620b7ba8da753794cae51dd36
- Refactor vkGetSwapchainTimeDomainPropertiesEXT to follow spec
- Ensure all time values are returned in nanoseconds
according to the active time domain
Signed-off-by: Maged Elnaggar <maged.elnaggar@arm.com>
Change-Id: Id4c6b86577bd509ec7bb5a946785dbe6642cfb1c
fully spec-compliant and avoid leaking unsupported symbols.
Extension enablement
* Only enable extensions promoted to core in Vulkan 1.1 when
the application’s API version is < 1.1.
* Filter the list passed to vkGet{Instance,Device}ProcAddr so
that only _application_-enabled extensions affect symbol lookup. Internally-added
extensions are ignored during GetProcAddr resolution.
Entrypoint exposure fixes
* getProcAddr now only returns pointers for entrypoints when _all_ of
their required extensions (or core version) are enabled:
* vkGetPhysicalDevicePresentRectanglesKHR
* vkAcquireNextImage2KHR
* vkGetDeviceGroupSurfacePresentModesKHR
* vkBindImageMemory2 / vkBindImageMemory2KHR
* vkGetPhysicalDeviceFeatures2KHR
* Removed all OR-conditions that previously exposed these functions prematurely,
replacing them with clear AND-checks per the spec.
Always-exposed device-level entrypoints
* Introduce INSTANCE_ENTRYPOINTS_LIST_EXPANSION for entrypoints that:
* Come from device-level extensions,
* Are looked up via instance APIs,
* Are _not_ in core Vulkan.
* Assign them an empty extension string so they’re always exposed, regardless
of which instance extensions the application requests.
Hiding non-intercepted functions
* Any entrypoint the layer lists internally but does _not_ intercept
is now hidden from applications, preventing calls into stubs.
These combined changes ensure the layer:
* advertises only the extensions it needs,
* obeys all cross-extension dependencies, and
* never leaks unsupported symbols via GetInstanceProcAddr/GetDeviceProcAddr.
Signed-off-by: Maged Elnaggar maged.elnaggar@arm.com
Change-Id: I7a5e5cb210e017f1aed76b187db1f40537010914
Fixes a few issues with the vkGetPastPresentationTimingsEXT
implementation. In particular:
- Entries with presentId = 0 are used to match entries in the
internal queue that didn't have a presentId supplied in the
presentation request.
- pPresentationTimings is returned to the application without any
holes.
- The appropriate return code is returned.
Adds an m_set member in swapchain_presentation_timing in order to allow
marking entries as completed that didn't have a timestamp, e.g.
a presentation request was rejected by the presentation engine.
Adds a mutex for guarding accesses to the internal queue by functions
that can be called by the application.
Minor documentation updates.
Change-Id: Id56bfd039b341e7f540a5ed028e954c490d00dd5
Signed-off-by: Iason Paraskevopoulos <iason.paraskevopoulos@arm.com>
Changes wait_for_pending_buffers to not call wait_for_free_buffer
if there are already free buffers. The reason is that
wait_for_free_buffer consumes the m_free_image_semaphore, which is ok
when it is done by acquire_next_image but wrong when it is done
by wait_for_pending_buffers which doesn't acquire any images. A side
effect of this could be that subsequent calls might
block if it they called after wait_for_pending_buffers and
there was only one free image remaining.
Renames wait_for_free_buffer to wait_and_get_free_buffer to better
communication that the semaphore is decremented after in the function.
Change-Id: Ie88fa435ba5805066d7bf13fce9fb2ce4c966cef
Signed-off-by: Iason Paraskevopoulos <iason.paraskevopoulos@arm.com>
If the time domain used by the compositor matches one of the time domains supported by the ICD:
* Advertise `VK_PRESENT_STAGE_IMAGE_FIRST_PIXEL_VISIBLE_BIT_EXT` in `VkPresentTimingSurfaceCapabilitiesEXT`
* List `VK_PRESENT_STAGE_IMAGE_FIRST_PIXEL_VISIBLE_BIT_EXT` in `swapchain_time_domains`
Also handles the case where neither `VK_TIME_DOMAIN_CLOCK_MONOTONIC_KHR` or `VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_KHR` is supported for all backends.
Additionally fixes the issue where Wayland protocol was not dispatched correctly which resulted in Clock_ID event not being communicated back to the layer.
Signed-off-by: Alex Bates <alex.bates@arm.com>
Signed-off-by: Dennis Tsiang <dennis.tsiang@arm.com>
Signed-off-by: Normunds Rieksts <normunds.rieksts@arm.com>
Set CMake minimum required version range to 3.4.3...4.0
to silence compatibility errors in CMake 4.0
by explicitly opting into all policies up to 4.0
Signed-off-by: Maged Elnaggar <maged.elnaggar@arm.com>
Change-Id: I2e0527dde4e764e9c17f519fc0ddd3c0e382fa31
Fixes the following issues:
* For Wayland backend, populates the presentation feedback listener with all callbacks as it is considered a fault by the protocol to not implement those
* For Wayland backend presentation feedback listener, forwards the feedback_discarded event to present ID as there could be situations where previously submitted buffers are discarded, such as when swapchains are using the MAILBOX presentation mode.
* For all swapchains, communicate critical errors back to present wait extension as otherwise, all callers waiting that are waiting on present to be delivered in vkWaitForPresentKHR call will never exit.
* Expanded the implementation in vkWaitForPresentKHR to be able to return critical swapchain errors if they have occured during the wait time.
* Fix issues in present ID where infinite waits could result in the vkWaitForPresentKHR call returning immediately due to UINT64_MAX timeout resulting in overflowing the system clock used in std::condition_variable
Change-Id: I1e475c3073c05394db259657eae1da21764a5a5c
Signed-off-by: Normunds Rieksts <normunds.rieksts@arm.com>
Signed-off-by: Alex Bates <alex.bates@arm.com>
In this change, handling for the present timings stages are improved.
Signed-off-by: Ginu Jacob <ginu.jacob@arm.com>
Change-Id: I68e9ae67fa94d39e23504890d922a85bdce938cd
Tags the fences created by the layer with the WsiLayerFence name
when instrumentation has been enabled.
Change-Id: I90ad71e5073d80cd650c6ba3be406abd67185376
Signed-off-by: Iason Paraskevopoulos <iason.paraskevopoulos@arm.com>
Enabling VK_PRESENT_MODE_FIFO_LATEST_READY_EXT for headless & wayland
This extension is handled the same as the MAILBOX ext.
Signed-off-by: Nir Ekhauz <nir.ekhauz@arm.com>
Change-Id: I010c3c9474a6d9c28964806ba5aa63549b622454
The check that determines device extension entrypoint visibility
was changed in order to fix CTS failures.
Signed-off-by: Fufu Fang <fufu.fang@arm.com>
Use the standard GitLab workflows to ensure the build job is run on
both Merge Requests and the main branch.
Signed-off-by: Dennis Tsiang <dennis.tsiang@arm.com>
Change-Id: I0469fd48200cb258ec6fea77bf9e817506d3570f
In this change, instead of checking the available version for
vkGetCalibratedTimestamps between KHR and EXT, unconditionally use the
KHR version in VK_EXT_present_timing as the extension has a dependency
on VK_KHR_calibrated_timestamps extension.
Signed-off-by: Ginu Jacob <ginu.jacob@arm.com>
Change-Id: I279182ae6d5311613c430546a8054dee0276c499
