Repository navigation
[Impeller] Guard the sampler library caches with a mutex - #193738
Conversation
Flutter GPU binds textures from the UI thread (InternalFlutterGpu_RenderPass_BindTexture calls GetSampler) while Impeller requests samplers on the raster thread (for example in FramebufferBlendContents::Render). SamplerLibraryMTL, SamplerLibraryVK and SamplerLibraryGLES scan and push_back a plain vector with no lock, so a scan can read a buffer that a concurrent push_back is reallocating, and two concurrent inserts can corrupt the vector or drop a sampler that was already handed out. Guard samplers_ with a mutex in all three backends. On Metal and Vulkan the lock covers the lookup and the insert but not sampler creation, and the insert re-checks the cache so concurrent requests share one sampler. GLES creates no device object and holds the lock for the whole lookup. Add RendererTest.SamplerLibraryAllowsConcurrentGetSampler, which requests the same set of samplers from four threads and checks that every thread gets the same valid sampler for each descriptor.
There was a problem hiding this comment.
Code Review
This pull request introduces thread safety to the sampler libraries across GLES, Metal, and Vulkan backends in Impeller. It adds mutexes to guard the sampler caches (samplers_) and implements helper methods to safely retrieve existing samplers under lock. Additionally, a concurrent unit test has been added to verify that GetSampler can be safely called from multiple threads simultaneously. There are no review comments, so no feedback is provided.
|
fyi @bdero |
bdero
left a comment
There was a problem hiding this comment.
LGTM, thanks. The returned raw_ptr stays valid after the lock drops since the vector holds shared_ptrs and entries are never removed, and the recheck on insert keeps concurrent misses on one sampler.
…ller-sampler-library-lock # Conflicts: # engine/src/flutter/impeller/renderer/renderer_unittests.cc
|
Resolved the conflicts. |
…#13220) Manual roll Flutter from d03768eef055 to 4249ccbc18a3 (158 revisions) Manual roll requested by [email protected] flutter/flutter@d03768e...4249ccb 2026-10-12 [email protected] [AGP 9.1.0 Migration #8] Unify add-to-app module wiring on the variant API (flutter/flutter#193717) 2026-10-12 [email protected] [master] Update Flutter DEPS to Dart 5933cb486b4aef2f8cf6b9f89a5651c0d5b0955e (flutter/flutter#193982) 2026-10-12 [email protected] [Android] Fix use-after-free of SurfaceControl.Transaction in HCPP (flutter/flutter#192969) 2026-10-12 [email protected] [windows] Add OpenGL compositor tests against ANGLE at D3D11 feature level 10_0 (flutter/flutter#194114) 2026-10-12 [email protected] Fix HDR image packing for ios (flutter/flutter#192526) 2026-10-12 [email protected] Bump customer tests to pick up material_ui and cupertino_ui (flutter/flutter#194236) 2026-10-12 [email protected] Remove no-shuffle from text_field_test.dart (flutter/flutter#193550) 2026-10-12 [email protected] [web] Clip semantics DOM text labels to stop stealing taps (flutter/flutter#194011) 2026-10-12 [email protected] Roll Skia from 9b6ceba4728d to 17f6986c226c (6 revisions) (flutter/flutter#194232) 2026-10-12 [email protected] [web] Skip non-focusable containers in Focusable.focusAsRouteDefault (flutter/flutter#194138) 2026-10-12 [email protected] [web_paragraph] Honor kTextHeightNone in TextStyle and ParagraphStyle (flutter/flutter#193733) 2026-10-12 [email protected] Roll Skia from c88b49d1d6c7 to 9b6ceba4728d (4 revisions) (flutter/flutter#194226) 2026-10-12 [email protected] Migrate remaining test to macOS 26 (flutter/flutter#193825) 2026-10-12 [email protected] Roll Dart SDK from 047ca322aa9c to 221363557163 (1 revision) (flutter/flutter#194219) 2026-10-12 [email protected] [flutter_tools] Include dev_dependencies hooks for release integration tests (flutter/flutter#193436) 2026-10-12 [email protected] Roll Skia from d8afad1b1829 to c88b49d1d6c7 (1 revision) (flutter/flutter#194198) 2026-10-12 [email protected] Roll Skia from b23ddbff1265 to d8afad1b1829 (4 revisions) (flutter/flutter#194195) 2026-10-11 [email protected] Roll Skia from a168863e44ec to b23ddbff1265 (1 revision) (flutter/flutter#194185) 2026-10-11 [email protected] Roll Skia from 0c1d93abfa04 to a168863e44ec (1 revision) (flutter/flutter#194181) 2026-10-11 [email protected] iOS: Re-enable rendering on willEnterForeground (flutter/flutter#191350) 2026-10-11 [email protected] Roll Skia from 93c9d3ffda25 to 0c1d93abfa04 (1 revision) (flutter/flutter#194179) 2026-10-11 [email protected] Remove more instances of deprecated `withOpacity` (flutter/flutter#193559) 2026-10-11 [email protected] Fix another `parameter_assignments` warning (flutter/flutter#193669) 2026-10-11 [email protected] Roll Skia from 3136abef8f58 to 93c9d3ffda25 (1 revision) (flutter/flutter#194169) 2026-10-11 [email protected] [Impeller] Guard the sampler library caches with a mutex (flutter/flutter#193738) 2026-10-11 [email protected] Roll Dart SDK from e9c89888bcc1 to 047ca322aa9c (1 revision) (flutter/flutter#194157) 2026-10-10 [email protected] Roll Dart SDK from 787baa8aae78 to e9c89888bcc1 (2 revisions) (flutter/flutter#194141) 2026-10-10 [email protected] Fix OverlayPortal layout in shrink-wrapped overlays (flutter/flutter#193079) 2026-10-10 [email protected] Roll Skia from 353076e27b1c to 3136abef8f58 (1 revision) (flutter/flutter#194130) 2026-10-10 [email protected] Roll Dart SDK from 88a07589b1ff to 787baa8aae78 (1 revision) (flutter/flutter#194124) 2026-10-10 [email protected] Route windowing view lookups through WidgetsBinding (flutter/flutter#192666) 2026-10-10 [email protected] Roll Dart SDK from 2a06cab34715 to 88a07589b1ff (1 revision) (flutter/flutter#194117) 2026-10-10 [email protected] Roll Skia from f9043b7aa798 to 353076e27b1c (2 revisions) (flutter/flutter#194115) 2026-10-10 [email protected] Refactored PipelineLibraryGLES (flutter/flutter#193985) 2026-10-09 [email protected] add shorthands for scroll physics subclasses in framework (flutter/flutter#193082) 2026-10-09 [email protected] Roll Dart SDK from ba889e3cee8a to 2a06cab34715 (1 revision) (flutter/flutter#194105) 2026-10-09 [email protected] Roll Skia from 5ea03d8a53a6 to f9043b7aa798 (2 revisions) (flutter/flutter#194099) 2026-10-09 [email protected] Add docs for using bb and led with LUCI (flutter/flutter#193859) 2026-10-09 [email protected] Roll Skia from e329bb617820 to 5ea03d8a53a6 (5 revisions) (flutter/flutter#194087) 2026-10-09 [email protected] Roll Packages from 54706bd to d8d9e97 (5 revisions) (flutter/flutter#194080) 2026-10-09 [email protected] Move the Vulkan wrapped-texture helper out of the GPU surface (flutter/flutter#193001) 2026-10-09 [email protected] Roll Skia from 3e54bcf4c963 to e329bb617820 (1 revision) (flutter/flutter#194078) 2026-10-09 [email protected] Roll Dart SDK from 23f0308f9af8 to ba889e3cee8a (2 revisions) (flutter/flutter#194077) 2026-10-09 [email protected] Bump dartdoc to 9.0.11 (flutter/flutter#194076) ...
) `SamplerLibraryMTL`, `SamplerLibraryVK` and `SamplerLibraryGLES` cache samplers in a plain vector that `GetSampler` scans and then appends to, with no lock. Flutter GPU calls `GetSampler` on the UI thread (`RenderPass.bindTexture`) while Impeller calls it on the raster thread (for example `FramebufferBlendContents::Render`). A scan can read a buffer that a concurrent `push_back` is reallocating, and two concurrent inserts can corrupt the vector or drop a sampler that was already returned. This PR guards `samplers_` with a mutex in all three backends. On Metal and Vulkan, the lock covers the lookup and the insert but not sampler creation, and the insert checks the cache again so concurrent requests share one sampler. GLES creates no device object, so it holds the lock for the whole lookup. The new test, `RendererTest.SamplerLibraryAllowsConcurrentGetSampler`, starts four threads on a latch, has each request the same 36 samplers, and checks that every thread gets the same valid sampler for each descriptor. Verified on master d03768e: Metal on macOS (M4 Max), and Vulkan (SwiftShader) and OpenGL ES (Mesa, Xvfb) on Linux x64 with the CI environment: | | Without the lock | With the lock | | --- | --- | --- | | New test, 40 runs (Metal / Vulkan / GLES) | 0 / 1 / 0 pass | 40/40 on each | | New test under TSan | sampler races on all three | no sampler race | | Repro app from the issue under TSan (Metal) | race in 4 of 6 runs | none in 6 runs | The macOS `impeller_unittests` CI filter (2426 passed, 0 failed) and the Metal backend tests pass, and the format and clang-tidy checks are clean. On Linux, the full `impeller_unittests` run was stopped early with 1154 passed and 0 failed, and the runs used the previous revision of the test, which started its threads with a spin barrier instead of a latch. Related: flutter#193721 / flutter#193722 (the same race in `PipelineLibraryMTL`). Both PRs add a test to `renderer_unittests.cc`; merging them conflicts only on the `#include <...>` lines. Fixes flutter#193737 ## Pre-launch Checklist - [x] I read the [Contributor Guide] and followed the process outlined there for submitting PRs. - [x] I read the [AI contribution guidelines] and understand my responsibilities, or I am not using AI tools. - [x] I read the [Tree Hygiene] wiki page, which explains my responsibilities. - [x] I read and followed the [Flutter Style Guide], including [Features we expect every widget to implement]. - [x] I signed the [CLA]. - [x] I listed at least one issue that this PR fixes in the description above. - [x] I updated/added relevant in-code documentation (doc comments with `///`). - [x] If this PR introduces a new feature or capability, I created and linked a website documentation issue or PR in [flutter/website] (or verified none is needed). - [x] I added new tests to check the change I am making, or this PR is [test-exempt]. - [x] I followed the [breaking change policy] and added [Data Driven Fixes] where supported. - [x] All existing and new tests are passing. <!-- Links --> [Contributor Guide]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md#overview [AI contribution guidelines]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md#ai-contribution-guidelines [Tree Hygiene]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md [test-exempt]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md#tests [Flutter Style Guide]: https://github.com/flutter/flutter/blob/main/docs/contributing/Style-guide-for-Flutter-repo.md [Features we expect every widget to implement]: https://github.com/flutter/flutter/blob/main/docs/contributing/Style-guide-for-Flutter-repo.md#features-we-expect-every-widget-to-implement [CLA]: https://cla.developers.google.com/ [flutter/website]: https://github.com/flutter/website [breaking change policy]: https://github.com/flutter/flutter/blob/main/docs/contributing/Tree-hygiene.md#handling-breaking-changes [Data Driven Fixes]: https://github.com/flutter/flutter/blob/main/docs/contributing/Data-driven-Fixes.md
SamplerLibraryMTL,SamplerLibraryVKandSamplerLibraryGLEScache samplers in a plain vector thatGetSamplerscans and then appends to, with no lock. Flutter GPU callsGetSampleron the UI thread (RenderPass.bindTexture) while Impeller calls it on the raster thread (for exampleFramebufferBlendContents::Render). A scan can read a buffer that a concurrentpush_backis reallocating, and two concurrent inserts can corrupt the vector or drop a sampler that was already returned.This PR guards
samplers_with a mutex in all three backends. On Metal and Vulkan, the lock covers the lookup and the insert but not sampler creation, and the insert checks the cache again so concurrent requests share one sampler. GLES creates no device object, so it holds the lock for the whole lookup.The new test,
RendererTest.SamplerLibraryAllowsConcurrentGetSampler, starts four threads on a latch, has each request the same 36 samplers, and checks that every thread gets the same valid sampler for each descriptor.Verified on master d03768e: Metal on macOS (M4 Max), and Vulkan (SwiftShader) and OpenGL ES (Mesa, Xvfb) on Linux x64 with the CI environment:
The macOS
impeller_unittestsCI filter (2426 passed, 0 failed) and the Metal backend tests pass, and the format and clang-tidy checks are clean. On Linux, the fullimpeller_unittestsrun was stopped early with 1154 passed and 0 failed, and the runs used the previous revision of the test, which started its threads with a spin barrier instead of a latch.Related: #193721 / #193722 (the same race in
PipelineLibraryMTL). Both PRs add a test torenderer_unittests.cc; merging them conflicts only on the#include <...>lines.Fixes #193737
Pre-launch Checklist
///).