【Beta】Enable Vulkan Renderer - #1032
Closed
LFRon wants to merge 7541 commits into
Closed
Conversation
This constant can be useful whenever one has to deal with timespecs, so let's move it to the header.
Perform a primitive garbage collection of buffers that have not been used in the past 10 seconds, an arbitrarily selected number. As garbage collection also makes span buffer allocation happen much more often, logging on allocation activity leads to a lot of log noise so get rid of that while at it.
This is the preferred name since Meson 1.1.
Signed-off-by: xurui <xurui@kylinos.cn>
Signed-off-by: xurui <xurui@kylinos.cn>
Signed-off-by: xurui <xurui@kylinos.cn>
Signed-off-by: xurui <xurui@kylinos.cn>
If we are trying directly scan-out a shm buffer, we create a temporary buffer pool to instantiate a wl_buffer from, created in accordance to our buffers size. If the buffer has an offset, it will end up out of bounds of the created pool. Extend the temporary pool by the buffer offset to compensate. Matching the original pool size does matter unless we want to optimize away the temporary pool to reduce mappings in the parent compositor.
Passing the wlr_client_buffer directly has a downsides because a fresh wlr_buffer pointer is passed each output commit instead of cycling through existing wlr_buffer objects: - The FDs are re-imported each time in the backend. - Any import failure is logged every output commit [1]. - The Wayland backend cannot handle import failures without roundtripping each output commit [2]. Instead, extract the source buffer from the wlr_client_buffer and pass that to the backend. [1]: https://gitlab.freedesktop.org/wlroots/wlroots/-/merge_requests/4836 [2]: https://gitlab.freedesktop.org/wlroots/wlroots/-/merge_requests/4841
create_immed() is a bit dangerous because on failure, either the connection is closed or the buffer is silently ignored. Use create() with a roundtrip to figure out whether the import succeeded.
We don't need to use the private header in these files.
We don't need any custom behavior since d8c0707 ("backend/drm: return secondary backend DRM FD").
This is now unused.
We assumed that all atomic backends supported syncobj, but gud does not. Instead, query DRM_CAP_SYNCOBJ_TIMELINE when using the atomic backend.
wlr_drm_format_set_remove lets you remove a modifier from the specified format, useful for filtering implicit modifiers.
If init_drm_renderer failed, it would destroy the renderer but would not set it to NULL, leading to use-after-free. NULL the renderer after destroying it.
We create a renderer for the sole purpose of blitting buffers from a primary renderer that we might not be able to scan-out from. If we end up with the pixman renderer, it either won't work becuase it cannot import dmabufs from the primary renderer, or won't have any effect because the primary renderer already uses dumb buffers. We test for DMA-BUF capabilities specifically to make it clear what our interest is, rather than focusing too much on the pixman renderer.
The presence of the renderer is what matters with respect to blitting. Having a parent without the need to blit will be allowed later.
This helps readability a bit and will make error handling in a coming commit easier.
If an mgpu device does not have a renderer, continue without one rather than ignoring it entirely. It is not guaranteed that we will be able to scan out to it in any particular configuration, but that is true for any output regardless, and having the output visible is not harmful even if it cannot light up. To proceed safely, we strip implicit modifier support from all planes, while avoiding duplication of the same logic for mdgpu_formats. This helps GUD and DisplayLink scenarios.
wlr_renderer and wlr_allocator will print out errors as they go, and end with a final error if they fail to create anything. The caller of this function will also log when it fails. Skip the redundant errors emitted here.
Client buffers backed by wl_shm is aggressively released, in which case we are not allowed to access it. Locking an already released buffer and later unlocking it will also re-trigger release, confusing clients. As a quick workaround, guard the unwrap by checking if the buffer is locked, which will be the case for non-wl_shm buffers.
wl_buffer.release event delivery becomes undefined when using the linux-drm-syncobj-v1 protocol, so we need to wait for buffer release via a timeline point instead. The protocol requires both wait and signal timelines to be set, so we need to create one when the compositor only supplies a wait timeline.
We were always creating a custom mode object when the output didn't have a fixed mode. This is important to handle two cases: - Virtual outputs with no concept of fixed modes. - DRM outputs with a list of fixed modes but with a custom mode set. However, in the case where an output didn't have a fixed mode and was disabled, we were also creating the custom mode object. Clients would then see a "ghost" mode: a mode object with no properties at all. Fix this by only creating the custom mode object if the output is enabled. Fixes: 5de9e1a ("wlr-output-management: Send custom modes to clients") Closes: swaywm/sway#8420
It got removed in 3ed1268 ("render: Nuke old read pixels API") Signed-off-by: Guido Günther <agx@sigxcpu.org>
It's not required.
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
There was a problem hiding this comment.
Sorry @LFRon, your pull request is larger than the review limit of 150000 diff characters
Contributor
Author
|
现在的这个实现我在Mesa RADV/ANV驱动上都检查过除了Blur效果外没有问题了 |
LFRon
changed the title
LFRon
force-pushed
the
feat-and-refactor/enable-vulkan-renderer
branch
5 times, most recently
from
July 17, 2026 04:25
42c5204 to
fca7de4
Compare
|
TAG Bot New tag: 0.8.15 |
LFRon
force-pushed
the
feat-and-refactor/enable-vulkan-renderer
branch
from
July 17, 2026 07:46
fca7de4 to
94efc4f
Compare
|
CLA Assistant Lite bot: 0 out of 12 committers have signed the CLA. |
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: LFRon The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
… patched wlroots) Select Qt Quick Vulkan when WLR_RENDERER=vulkan is requested while keeping the default OpenGL/GLES path unchanged. Wire linux-dmabuf to the wlroots renderer and use the patched wlroots Vulkan helpers to render Qt Quick directly into compositor-owned output buffers. Keep client buffers on the wlroots texture path: surfaces still become wlroots-provided QSGTextures for read-only Qt sampling, with per-pass sampling preparation and provider lifetime fixes. Do not add syncobj support or a Treeland-owned client dmabuf cache/fence/direct-import pipeline. Acquire and release wlroots render buffers around Qt RenderControl frames. Disable Vulkan backdrop blits that copy from the output buffer, use a software cursor for the first Vulkan path, and create a separate preserve render target for overlay passes so Vulkan load/store state is fixed before QRhi creates the render pass. With the patched wlroots 0.19 helpers installed, the explicit Vulkan path now displays DDM, dde-shell, DTK and ordinary clients correctly.
Remove the branch-local VulkanSamplingPolicy API and the Raw/ShmUpload classification in WSGTextureProvider. All valid wlroots Vulkan textures are now declared to Qt as shader-read sampled textures, while per-render-pass sampling still uses wlroots' Vulkan helper to prepare and finish the active texture set. Keep the established Vulkan correctness boundaries intact: client buffers remain wlroots-provided QSGTextures, output rendering still uses wlroots render-buffer acquire/release, and primary output renderers are not exposed as texture providers on Vulkan. Also remove high-frequency success debug logs from the Vulkan sampling/cache paths and keep the Vulkan RenderBufferBlitter fallback as an explicit backdrop source policy.
Add a waylib WPresentation wrapper around wlroots wp_presentation and attach it for the Vulkan backend. Track surfaces actually sampled by the Qt Quick render pass, then submit textured-on-output feedback before committing the matching wlroots output buffer. Fix the qwlroots wlroots-0.19 presentation create signature, keep the presentation global lifetime owned by wlroots, and add the centralized waylib.protocols.presentation logging category.
Keep cursor contents on the existing QtQuick/software cursor path, but stop binding cursor item geometry to the live cursor position when it is rendered through OutputLayer. Schedule output frames on cursor motion and compute cursor layer placement from wlroots cursor/output coordinates during layer rendering. This avoids dirtying the main QML scene on cursor movement while preserving the existing cursor content generation and fallback behavior.
Pair Vulkan texture sampling prepare with a matching per-pass finish. Release prepared wlroots textures in LIFO order after Qt records the pass commands, continue best-effort on failures, and fail the render pass if any release cannot be recorded. Defer destruction of Vulkan render resources that may still be referenced by the current QRhi frame. Let WOutput swapchain configuration optionally return the replaced swapchain, retire old Vulkan swapchains in WBufferRenderer, and retire old WRenderHelper BufferData/render target wrappers until several completed frames after rc()->endFrame(). Keep non-Vulkan paths on the existing immediate cleanup behavior. This prevents stale foreign texture ownership and avoids destroying wlroots VkImages while Qt/Vulkan command buffers may still reference them.
When wrapping wlroots Vulkan textures for Qt Quick sampling, create an exact VkImageView using the original format instead of Qt's mapped format. Qt's QSGRhiSupport maps both VK_FORMAT_A2R10G10B10_UNORM_PACK32 and VK_FORMAT_A2B10G10R10_UNORM_PACK32 to a single QRhiTexture::RGB10A2 format, but the Vulkan backend creates an A2B10G10R10 view first. This causes channel order mismatches for A2R10 images (e.g. BGRA buffers from Electron apps like Youdao Note) resulting in color corruption. The fix: - Create a candidate QRhiTexture wrapper for the external VkImage - Replace Qt's default image view with an exact view using the wlroots texture's actual format - Apply correct alpha swizzle based on wlroots texture's has_alpha flag - Reject A2R10G10B10 format that cannot be precisely represented without MUTABLE_FORMAT Also includes: - Add qw_vulkan namespace for type-safe wlroots Vulkan API wrappers - Add WVulkanTrace debug tracing system (enabled via WAYLIB_VULKAN_TRACE=1) - Improve texture provider lifecycle with 4-frame delayed cleanup for Vulkan resources - Add helper methods to qw_client_buffer for texture/source access Fixes color corruption in Electron-based applications (e.g. Youdao Note) running under the Vulkan renderer backend.
Avoid QRhi texture readback for treeland interactive capture. The Vulkan backend wraps wlroots-owned images for rendering/sampling, not for Qt-side transfer readback, which can leave portal screenshot clients waiting forever. Route one-shot capture through qw_buffer/qw_renderer-based WBufferDumper, use the output viewport's last buffer for output/region sources, and report frame failures when image creation or client-buffer copy fails.
Cursor output layers cache their rendered buffer and previously only redrew when geometry or render state changed. With the Vulkan render path, cursor shape/image updates can keep the same geometry, so the compositor could keep committing a stale cursor layer until another output repaint happened. Track WQuickCursor content revisions and mark cursor layers dirty when the cursor content changes. Schedule the owning output conservatively: reuse an already-dirty output when possible, otherwise force an output update so software-composited Vulkan cursors repaint in the same frame. Avoid duplicate dirty notifications when WCursorImage already emitted imageChanged. Also route Vulkan textures created during updatePaintNode() through the current render pass prepare list, so late-created cursor textures are finished together with the rest of the pass instead of being prepared out of band.
Keep QRhi's native output layout in sync with the GENERAL layout recorded by the wlroots release barrier, so reused render buffers get the required transition back to COLOR_ATTACHMENT_OPTIMAL. Treat Vulkan rendering as an all-or-nothing frame transaction. Preserve QRhi begin/end results, reject output commits after acquire, release, submission, or device-loss failures, and terminate the compositor on unrecoverable scene graph errors. Stage sampled texture candidates until preparation succeeds and keep the previous valid texture on failure. Restrict active provider scans to the Vulkan path so GLES2 and software rendering retain their existing behavior. Replace the four-frame Vulkan cleanup delay with exact retirement after successful synchronous offscreen completion and output state handling. Destroy QRhi wrappers before owned wlroots textures and buffer locks, and remove the forced updates previously used to advance cleanup. Also drop unsupported async-offscreen claims from Vulkan tracing. This keeps the existing wlroots texture -> Qt Quick composition -> wlroots output buffer architecture without modifying Qt or wlroots.
Vulkan renderers may expose DMA-BUF import formats only through explicit modifiers. Advertising the legacy wl_drm global unconditionally lets clients select the implicit DMA-BUF path even when the renderer cannot import DRM_FORMAT_MOD_INVALID buffers. Add qwlroots helpers to inspect the renderer's DMA-BUF texture format set and check for DRM_FORMAT_MOD_INVALID. Only create the legacy wl_drm global when the renderer actually supports implicit DMA-BUF import; otherwise keep zwp_linux_dmabuf_v1 as the DMA-BUF path exposed to clients. This preserves the GLES2 compatibility case while avoiding a misleading legacy global for the Vulkan renderer. wl_drm remains owned and destroyed by wlroots as before.
LFRon
force-pushed
the
feat-and-refactor/enable-vulkan-renderer
branch
from
July 17, 2026 09:18
d815183 to
8e9532d
Compare
Treeland now builds against the vendored wlroots under 3rdparty/wlroots instead of the previously patched wlroots-upstream tree. The Vulkan render path still needs the wlroots Vulkan helper entry points used by qwlroots and waylib to bridge Qt Quick RHI rendering with wlroots-owned buffers and textures. Port the existing Vulkan helper patch into the vendored wlroots tree: - expose render-buffer VkImage attributes for compositor-owned output buffers - record acquire/release barriers around Qt rendering into wlroots buffers - prepare wlroots Vulkan textures before Qt samples them - flush pending stage uploads for shm/pixel textures - wait DMA-BUF fences and return foreign texture ownership after sampling The change is limited to the Vulkan renderer. GLES2 behavior is unchanged, and this does not enable the linux-drm-syncobj protocol or add a separate Treeland-side client buffer import/cache pipeline. Verification: - cmake --build --preset default - ctest --test-dir build --output-on-failure - nm -D build/wlroots/libwaylib-wlroots.so confirms the new helpers are exported
Contributor
|
Can you guys fix your CLA bot? It mentions every wlroots contributor every time you guys force push a PR with the wlroots tree, I have over 50 email notifications for this |
Contributor
Author
I will fix it tomorrow with force rebase |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
这个PR让treeland内调用的Qt RHI Backend切换到Vulkan, 然后通过wlroots Vulkan后端进行上屏
注意: 该PR现在需要patched wlroots, 需要合并的补丁在 Better-Deepin-Repo/wlroots-0.19-deepin@c9240cd