Skip to content

fix(🍏): run CAMetalLayer operation on main thread #280

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wcandillon wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

fix(🍏): run CAMetalLayer operation on main thread #280

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions apps/example/ios/Podfile.lock
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1865,7 +1865,7 @@ PODS:
- ReactCommon/turbomodule/core
- SocketRocket
- Yoga
- react-native-wgpu (0.3.0):
- react-native-wgpu (0.3.1):
- boost
- DoubleConversion
- fast_float
Expand Down Expand Up @@ -2903,7 +2903,7 @@ SPEC CHECKSUMS:
React-microtasksnativemodule: 75b6604b667d297292345302cc5bfb6b6aeccc1b
react-native-safe-area-context: c6e2edd1c1da07bdce287fa9d9e60c5f7b514616
react-native-skia: 5bf2b2107cd7f2d806fd364f5e16b1c7554ed3cd
react-native-wgpu: 6704c6ad8ef91fdbbad58543089ac50c0e6f3fa7
react-native-wgpu: e35c1eb10f65d23f3cee907d44a3ca0a3d7ccbbd
React-NativeModulesApple: 879fbdc5dcff7136abceb7880fe8a2022a1bd7c3
React-oscompat: 93b5535ea7f7dff46aaee4f78309a70979bdde9d
React-perflogger: 5536d2df3d18fe0920263466f7b46a56351c0510
Expand Down
35 changes: 35 additions & 0 deletions packages/webgpu/apple/MainThreadUtils.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,35 @@
#pragma once

#ifdef __APPLE__

#include <dispatch/dispatch.h>
#include <functional>
#include <pthread.h>

namespace rnwgpu {

namespace detail {
inline void RunOnMainThreadSyncInvoke(void *context) {
auto *funcPtr = static_cast<std::function<void()> *>(context);
(*funcPtr)();
}
} // namespace detail

inline void RunOnMainThreadSync(const std::function<void()> &task) {
if (!task) {
return;
}

if (pthread_main_np() != 0) {
task();
return;
}

auto callable = task;
dispatch_sync_f(dispatch_get_main_queue(), &callable,
detail::RunOnMainThreadSyncInvoke);
}

} // namespace rnwgpu

#endif
25 changes: 19 additions & 6 deletions packages/webgpu/cpp/rnwgpu/SurfaceRegistry.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,12 +1,17 @@
#pragma once

#include <functional>
#include <memory>
#include <shared_mutex>
#include <unordered_map>
#include <utility>

#include "webgpu/webgpu_cpp.h"

#ifdef __APPLE__
#include "../../apple/MainThreadUtils.h"
#endif

namespace rnwgpu {

struct NativeInfo {
Expand All @@ -31,7 +36,7 @@ class SurfaceInfo {
std::unique_lock<std::shared_mutex> lock(_mutex);
config.width = newWidth;
config.height = newHeight;
_configure();
_invokeOnMainThread([this]() { _applyConfigurationLocked(); });
}

void configure(wgpu::SurfaceConfiguration &newConfig) {
Expand All @@ -40,7 +45,7 @@ class SurfaceInfo {
config.width = width;
config.height = height;
config.presentMode = wgpu::PresentMode::Fifo;
_configure();
_invokeOnMainThread([this]() { _applyConfigurationLocked(); });
}

void unconfigure() {
Expand Down Expand Up @@ -78,7 +83,7 @@ class SurfaceInfo {
// surface
if (texture != nullptr) {
config.usage = config.usage | wgpu::TextureUsage::CopyDst;
_configure();
_invokeOnMainThread([this]() { _applyConfigurationLocked(); });
// We flush the offscreen texture to the onscreen one
// TODO: there is a faster way to do this without validation?
wgpu::CommandEncoderDescriptor encoderDesc;
Expand All @@ -102,7 +107,7 @@ class SurfaceInfo {
wgpu::CommandBuffer commands = encoder.Finish();
wgpu::Queue queue = device.GetQueue();
queue.Submit(1, &commands);
surface.Present();
_invokeOnMainThread([this]() { surface.Present(); });
texture = nullptr;
}
}
Expand All @@ -116,7 +121,7 @@ class SurfaceInfo {
void present() {
std::unique_lock<std::shared_mutex> lock(_mutex);
if (surface) {
surface.Present();
_invokeOnMainThread([this]() { surface.Present(); });
}
}

Expand Down Expand Up @@ -152,7 +157,15 @@ class SurfaceInfo {
}

private:
void _configure() {
void _invokeOnMainThread(const std::function<void()> &task) {
#ifdef __APPLE__
RunOnMainThreadSync(task);
Copy link
Contributor

@bhamiltoncx bhamiltoncx Feb 25, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is probably not entirely safe. The caller holds _mutex, but the main thread might itself be blocked waiting for _mutex.

Example, just after a background thread takes _mutex and then wants to dispatch to the main thread:

  1. RN updates view props
  2. RN calls (on the UI thread) -[WebGPUView updateProps:oldProps:]
  3. WebGPUView calls -[MetalView configure]
  4. -[MetalView configure] calls SurfaceRegistry.switchToOnscreen()
  5. SurfaceRegistry.switchToOnscreen() tries to take _mutex and blocks

In this scenario, the dispatch_sync_f() to the main thread will deadlock.

I think this needs to move to a queue-based system where changes to the surface and/or configuration need to be queued up and then applied in order (which must happen on the main thread on Apple platforms).

#else
task();
#endif
}

void _applyConfigurationLocked() {
if (surface) {
surface.Configure(&config);
} else {
Expand Down
2 changes: 1 addition & 1 deletion packages/webgpu/package.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
{
"name": "react-native-wgpu",
"version": "0.3.1",
"version": "0.3.2",
"description": "React Native WebGPU",
"main": "lib/commonjs/index",
"module": "lib/module/index",
Expand Down
Loading