Use MapAsync and AllowProcessEvents for callback

NEW_RELEASE_NOTES.md

@@ -6,3 +6,7 @@
 appropriate header in [RELEASE_NOTES.md](./RELEASE_NOTES.md).
 
 ## Release notes for next branch cut
+
+- engine: add `View::getLastDynamicResolutionScale()` (b/457753622)
+
+- materials: Make Material Instances' UBO descriptor use dynamic offsets. [⚠️ **Recompile Materials**]

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.68.1'
+    implementation 'com.google.android.filament:filament-android:1.67.0'
 }
 ```
 
@@ -51,7 +51,7 @@ Here are all the libraries available in the group `com.google.android.filament`:
 iOS projects can use CocoaPods to install the latest release:
 
 ```shell
-pod 'Filament', '~> 1.68.1'
+pod 'Filament', '~> 1.67.0'
 ```
 
 ## Documentation

RELEASE_NOTES.md

@@ -7,15 +7,6 @@ A new header is inserted each time a *tag* is created.
 Instead, if you are authoring a PR for the main branch, add your release note to
 [NEW_RELEASE_NOTES.md](./NEW_RELEASE_NOTES.md).
 
-## v1.68.1
-
-
-## v1.68.0
-
-- engine: add `View::getLastDynamicResolutionScale()` (b/457753622)
-- Metal: report GPU errors to the platform via `debugUpdateStat` (b/431665753).
-- materials: Make Material Instances' UBO descriptor use dynamic offsets. [⚠️ **Recompile Materials**]
-
 ## v1.67.1
 
 - Metal: Add support for the `SwapChain::CONFIG_MSAA_4_SAMPLES` flag.

android/filament-utils-android/src/main/java/com/google/android/filament/utils/ModelViewer.kt

@@ -16,9 +16,6 @@
 
 package com.google.android.filament.utils
 
-import android.graphics.Bitmap
-import android.os.Handler
-import android.os.Looper
 import android.view.MotionEvent
 import android.view.Surface
 import android.view.SurfaceView
@@ -29,7 +26,6 @@ import com.google.android.filament.android.UiHelper
 import com.google.android.filament.gltfio.*
 import kotlinx.coroutines.*
 import java.nio.Buffer
-import java.nio.ByteBuffer
 
 private const val kNearPlane = 0.05f     // 5 cm
 private const val kFarPlane = 1000.0f    // 1 km
@@ -123,8 +119,6 @@ class ModelViewer(
     private val target = DoubleArray(3)
     private val upward = DoubleArray(3)
 
-    private var debugFrameCallback: ((Bitmap) -> Unit)? = null
-
     init {
         renderer = engine.createRenderer()
         scene = engine.createScene()
@@ -311,39 +305,10 @@ class ModelViewer(
         // Render the scene, unless the renderer wants to skip the frame.
         if (renderer.beginFrame(swapChain!!, frameTimeNanos)) {
             renderer.render(view)
-
-            debugFrameCallback?.let {
-                val viewport = view.viewport
-                val bitmap = Bitmap.createBitmap(viewport.width, viewport.height,
-                        Bitmap.Config.ARGB_8888)
-                val buffer = ByteBuffer.allocateDirect(viewport.width * viewport.height * 4)
-
-                val handler = Handler(Looper.getMainLooper())
-                val pixelBufferDescriptor = Texture.PixelBufferDescriptor(buffer,
-                        Texture.Format.RGBA, Texture.Type.UBYTE, 1, 0, 0, 0, handler) {
-                    buffer.rewind()
-                    bitmap.copyPixelsFromBuffer(buffer)
-                    it(bitmap)
-                }
-                renderer.readPixels(viewport.left, viewport.bottom, viewport.width,
-                                    viewport.height, pixelBufferDescriptor)
-                debugFrameCallback = null
-            }
-
             renderer.endFrame()
         }
     }
 
-    /*
-     * Sets a callback that will be invoked with the next rendered frame as a Bitmap. Note that this
-     * is a one-time callback.
-     *
-     *  @param callback   callback to be invoked with a rendered frame as [Bitmap]
-     */
-    fun debugGetNextFrameCallback(callback: (Bitmap) -> Unit) {
-        debugFrameCallback = callback
-    }
-
     private fun populateScene(asset: FilamentAsset) {
         val rcm = engine.renderableManager
         var count = 0

android/gradle.properties

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.68.1
+VERSION_NAME=1.67.0
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 

docs/dup/intro.html

@@ -181,7 +181,7 @@ important for <code>matc</code> (material compiler).</p>
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.67.1'
+    implementation 'com.google.android.filament:filament-android:1.66.2'
 }
 </code></pre>
 <p>Here are all the libraries available in the group <code>com.google.android.filament</code>:</p>
@@ -196,7 +196,7 @@ dependencies {
 </div>
 <h3 id="ios"><a class="header" href="#ios">iOS</a></h3>
 <p>iOS projects can use CocoaPods to install the latest release:</p>
-<pre><code class="language-shell">pod 'Filament', '~&gt; 1.67.1'
+<pre><code class="language-shell">pod 'Filament', '~&gt; 1.66.2'
 </code></pre>
 <h2 id="documentation"><a class="header" href="#documentation">Documentation</a></h2>
 <ul>

filament/backend/CMakeLists.txt

@@ -206,8 +206,6 @@ if (FILAMENT_SUPPORTS_VULKAN)
             src/vulkan/VulkanDriverFactory.h
             src/vulkan/VulkanExternalImageManager.cpp
             src/vulkan/VulkanExternalImageManager.h
-            src/vulkan/VulkanStreamedImageManager.cpp
-            src/vulkan/VulkanStreamedImageManager.h
             src/vulkan/VulkanFboCache.cpp
             src/vulkan/VulkanFboCache.h
             src/vulkan/VulkanHandles.cpp

filament/backend/include/backend/Platform.h

@@ -19,17 +19,13 @@
 #ifndef TNT_FILAMENT_BACKEND_PLATFORM_H
 #define TNT_FILAMENT_BACKEND_PLATFORM_H
 
-#include <utils/CString.h>
 #include <utils/compiler.h>
 #include <utils/Invocable.h>
-#include <utils/Mutex.h>
 
 #include <stddef.h>
 #include <stdint.h>
 
 #include <atomic>
-#include <memory>
-#include <mutex>
 
 namespace filament::backend {
 
@@ -74,9 +70,6 @@ public:
         ExternalImageHandle& operator=(ExternalImageHandle const& rhs) noexcept;
         ExternalImageHandle& operator=(ExternalImageHandle&& rhs) noexcept;
 
-        bool operator==(const ExternalImageHandle& rhs) const noexcept {
-            return mTarget == rhs.mTarget;
-        }
         explicit operator bool() const noexcept { return mTarget != nullptr; }
 
         ExternalImage* UTILS_NULLABLE get() noexcept { return mTarget; }
@@ -515,22 +508,13 @@ public:
     // --------------------------------------------------------------------------------------------
     // Debugging APIs
 
-    using DebugUpdateStatFunc = utils::Invocable<void(const char* UTILS_NONNULL key,
-            uint64_t intValue, utils::CString stringValue)>;
+    using DebugUpdateStatFunc = utils::Invocable<void(const char* UTILS_NONNULL key, uint64_t value)>;
 
     /**
      * Sets the callback function that the backend can use to update backend-specific statistics
      * to aid with debugging. This callback is guaranteed to be called on the Filament driver
      * thread.
      *
-     * The callback signature is (key, intValue, stringValue). Note that for any given call,
-     * only one of the value parameters (intValue or stringValue) will be meaningful, depending on
-     * the specific key.
-     *
-     * IMPORTANT_NOTE: because the callback is called on the driver thread, only quick, non-blocking
-     * work should be done inside it. Furthermore, no graphics API calls (such as GL calls) should
-     * be made, which could interfere with Filament's driver state.
-     *
      * @param debugUpdateStat   an Invocable that updates debug statistics
      */
     void setDebugUpdateStatFunc(DebugUpdateStatFunc&& debugUpdateStat) noexcept;
@@ -549,32 +533,15 @@ public:
      * This function is guaranteed to be called only on a single thread, the Filament driver
      * thread.
      *
-     * @param key           a null-terminated C-string with the key of the debug statistic
-     * @param intValue      the updated integer value of key (the string value passed to the
-     *                      callback will be empty)
+     * @param key          a null-terminated C-string with the key of the debug statistic
+     * @param value        the updated value of key
      */
-    void debugUpdateStat(const char* UTILS_NONNULL key, uint64_t intValue);
-
-    /**
-     * To track backend-specific statistics, the backend implementation can call the
-     * application-provided callback function debugUpdateStatFunc to associate or update a value
-     * with a given key. It is possible for this function to be called multiple times with the
-     * same key, in which case newer values should overwrite older values.
-     *
-     * This function is guaranteed to be called only on a single thread, the Filament driver
-     * thread.
-     *
-     * @param key           a null-terminated C-string with the key of the debug statistic
-     * @param stringValue   the updated string value of key (the integer value passed to the
-     *                      callback will be 0)
-     */
-    void debugUpdateStat(const char* UTILS_NONNULL key, utils::CString stringValue);
+    void debugUpdateStat(const char* UTILS_NONNULL key, uint64_t value);
 
 private:
-    std::shared_ptr<InsertBlobFunc> mInsertBlob;
-    std::shared_ptr<RetrieveBlobFunc> mRetrieveBlob;
-    std::shared_ptr<DebugUpdateStatFunc> mDebugUpdateStat;
-    mutable utils::Mutex mMutex;
+    InsertBlobFunc mInsertBlob;
+    RetrieveBlobFunc mRetrieveBlob;
+    DebugUpdateStatFunc mDebugUpdateStat;
 };
 
 } // namespace filament

filament/backend/include/backend/platforms/PlatformEGLAndroid.h

@@ -17,6 +17,8 @@
 #ifndef TNT_FILAMENT_BACKEND_OPENGL_OPENGL_PLATFORM_EGL_ANDROID_H
 #define TNT_FILAMENT_BACKEND_OPENGL_OPENGL_PLATFORM_EGL_ANDROID_H
 
+#include "AndroidSwapChainHelper.h"
+#include "AndroidFrameCallback.h"
 #include "AndroidNdk.h"
 
 #include <backend/AcquiredImage.h>
@@ -30,6 +32,8 @@
 
 #include <math/mat3.h>
 
+#include "AndroidNativeWindow.h"
+
 #include <chrono>
 
 #include <stddef.h>
@@ -167,10 +171,19 @@ protected:
             SwapChain* drawSwapChain,
             SwapChain* readSwapChain) override;
 
-private:
-    struct SwapChainEGLAndroid;
-    struct AndroidDetails;
+    struct SwapChainEGLAndroid : public SwapChainEGL {
+        SwapChainEGLAndroid(PlatformEGLAndroid const& platform,
+                void* nativeWindow, uint64_t flags);
+        SwapChainEGLAndroid(PlatformEGLAndroid const& platform,
+                uint32_t width, uint32_t height, uint64_t flags);
+        void terminate(PlatformEGLAndroid& platform);
+        bool setPresentFrameId(uint64_t frameId) const noexcept;
+        uint64_t getFrameId(uint64_t frameId) const noexcept;
+    private:
+        AndroidSwapChainHelper mImpl{};
+    };
 
+private:
     // prevent derived classes' implementations to call through
     [[nodiscard]] SwapChain* createSwapChain(void* nativeWindow, uint64_t flags) override;
     [[nodiscard]] SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) override;
@@ -190,14 +203,17 @@ private:
 
     int mOSVersion;
     ExternalStreamManagerAndroid& mExternalStreamManager;
-    AndroidDetails& mAndroidDetails;
     InitializeJvmForPerformanceManagerIfNeeded const mInitializeJvmForPerformanceManagerIfNeeded;
     utils::PerformanceHintManager mPerformanceHintManager;
     utils::PerformanceHintManager::Session mPerformanceHintSession;
+    SwapChainEGLAndroid* mCurrentDrawSwapChain{};
+
     using clock = std::chrono::high_resolution_clock;
     clock::time_point mStartTimeOfActualWork;
-    SwapChainEGLAndroid* mCurrentDrawSwapChain{};
+    AndroidProducerThrottling mProducerThrottling;
     bool mAssertNativeWindowIsValid = false;
+
+    AndroidFrameCallback mAndroidFrameCallback;
 };
 
 } // namespace filament::backend

filament/backend/include/backend/platforms/VulkanPlatformAndroid.h

@@ -17,13 +17,12 @@
 #ifndef TNT_FILAMENT_BACKEND_PLATFORMS_VULKAN_PLATFORM_ANDROID_H
 #define TNT_FILAMENT_BACKEND_PLATFORMS_VULKAN_PLATFORM_ANDROID_H
 
+#include "AndroidFrameCallback.h"
 #include "AndroidNdk.h"
 
 #include <backend/DriverEnums.h>
 #include <backend/platforms/VulkanPlatform.h>
 
-#include <utils/compiler.h>
-
 #include <android/hardware_buffer.h>
 
 namespace filament::backend {
@@ -92,8 +91,6 @@ protected:
     VkExternalFenceHandleTypeFlagBits getFenceExportFlags() const noexcept override;
 
 private:
-    struct AndroidDetails;
-
     struct ExternalImageVulkanAndroid : public ExternalImage {
         AHardwareBuffer* aHardwareBuffer = nullptr;
         bool sRGB = false;
@@ -102,7 +99,7 @@ private:
         ~ExternalImageVulkanAndroid() override;
     };
 
-    AndroidDetails& mAndroidDetails;
+    AndroidFrameCallback mAndroidFrameCallback;
     int mOSVersion{};
 };
 

filament/backend/src/AndroidNativeWindow.cpp

@@ -61,16 +61,6 @@ int NativeWindow::enableFrameTimestamps(ANativeWindow* anw, bool enable) {
     return pWindow->perform(anw, ENABLE_FRAME_TIMESTAMPS, enable);
 }
 
-int NativeWindow::frameTimestampsSupportsPresent(ANativeWindow* anw, bool* outSupportsPresent) {
-    NativeWindow const* pWindow = reinterpret_cast<NativeWindow const*>(anw);
-    int value = 0;
-    bool const success = pWindow->perform(anw, FRAME_TIMESTAMPS_SUPPORTS_PRESENT, &value);
-    if (success) {
-        *outSupportsPresent = bool(value);
-    }
-    return success;
-}
-
 int NativeWindow::getCompositorTiming(ANativeWindow* anw,
         int64_t* compositeDeadline, int64_t* compositeInterval,
         int64_t* compositeToPresentLatency) {

filament/backend/src/AndroidNativeWindow.h

@@ -32,7 +32,6 @@ struct NativeWindow {
     // is valid query enum value
     enum {
         IS_VALID                = 17,
-        FRAME_TIMESTAMPS_SUPPORTS_PRESENT = 18,
         GET_NEXT_FRAME_ID       = 24,
         ENABLE_FRAME_TIMESTAMPS = 25,
         GET_COMPOSITOR_TIMING   = 26,
@@ -52,7 +51,6 @@ struct NativeWindow {
 
     static int getNextFrameId(ANativeWindow* anw, uint64_t* frameId);
     static int enableFrameTimestamps(ANativeWindow* anw, bool enable);
-    static int frameTimestampsSupportsPresent(ANativeWindow* anw, bool* outSupportsPresent);
     static int getCompositorTiming(ANativeWindow* anw,
             int64_t* compositeDeadline, int64_t* compositeInterval,
             int64_t* compositeToPresentLatency);

filament/backend/src/AndroidNdk.cpp

@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 
-#include <backend/platforms/AndroidNdk.h>
+#include "AndroidNdk.h"
 
 #include <android/hardware_buffer.h>
 

filament/backend/src/AndroidSwapChainHelper.cpp

@@ -19,9 +19,6 @@
 
 #include <android/native_window.h>
 
-#include <utils/compiler.h>
-#include <utils/Logger.h>
-
 #include <cstddef>
 #include <cstdint>
 #include <limits>
@@ -39,23 +36,11 @@ bool AndroidSwapChainHelper::setPresentFrameId(
     int const status = NativeWindow::getNextFrameId(anw, &sysFrameId);
     if (status == 0) {
         std::lock_guard const lock(mLock);
-        // frameIds must be strictly monotonic, if that's not the case (i.e. the new frameId is
-        // less or equal to the last one in the map), we have to clear the map, because the
-        // map's find() assume sorted keys.
-        // This case can happen if two different filament::Renderer are used with the same
-        // ANativeWindow (the Renderer would have different frameIds). This is expected to
-        // be a rare case.
-        if (UTILS_UNLIKELY(!mFrameIdToSystemFrameId.empty() &&
-                frameId <= mFrameIdToSystemFrameId.back().first)) {
-            // this log is expected to happen very rarely
-            DLOG(INFO) << "clearing frame history anw=" << anw
-                    << ", frameId=" << frameId
-                    << ", previous=" << mFrameIdToSystemFrameId.back().first
-                    << ", sysFrameId=" << sysFrameId;
-            // clear the frame history
-            mFrameIdToSystemFrameId.clear();
+        auto const pos = mFrameIdToSystemFrameId.find(frameId);
+        if (pos && *pos != sysFrameId) {
+            // we're trying to associate the same frame id to a different frame!
+            return false;
         }
-
         // oldest entry is removed
         mFrameIdToSystemFrameId.insert(frameId, sysFrameId);
         return true;

filament/backend/src/Platform.cpp

@@ -139,73 +139,42 @@ bool Platform::queryFrameTimestamps(SwapChain const*, uint64_t, FrameTimestamps*
 }
 
 void Platform::setBlobFunc(InsertBlobFunc&& insertBlob, RetrieveBlobFunc&& retrieveBlob) noexcept {
-    std::lock_guard<decltype(mMutex)> lock(mMutex);
-    mInsertBlob = std::make_shared<InsertBlobFunc>(std::move(insertBlob));
-    mRetrieveBlob = std::make_shared<RetrieveBlobFunc>(std::move(retrieveBlob));
+    mInsertBlob = std::move(insertBlob);
+    mRetrieveBlob = std::move(retrieveBlob);
 }
 
 bool Platform::hasInsertBlobFunc() const noexcept {
-    std::lock_guard<decltype(mMutex)> lock(mMutex);
-    return mInsertBlob && bool(*mInsertBlob);
+    return bool(mInsertBlob);
 }
 
 bool Platform::hasRetrieveBlobFunc() const noexcept {
-    std::lock_guard<decltype(mMutex)> lock(mMutex);
-    return mRetrieveBlob && bool(*mRetrieveBlob);
+    return bool(mRetrieveBlob);
 }
 
 void Platform::insertBlob(void const* key, size_t keySize, void const* value, size_t valueSize) {
-    std::shared_ptr<InsertBlobFunc> callback;
-    {
-        std::unique_lock<decltype(mMutex)> lock(mMutex);
-        callback = mInsertBlob;
-    }
-    if (callback) {
-        (*callback)(key, keySize, value, valueSize);
+    if (mInsertBlob) {
+        mInsertBlob(key, keySize, value, valueSize);
     }
 }
 
 size_t Platform::retrieveBlob(void const* key, size_t keySize, void* value, size_t valueSize) {
-    std::shared_ptr<RetrieveBlobFunc> callback;
-    {
-        std::unique_lock<decltype(mMutex)> lock(mMutex);
-        callback = mRetrieveBlob;
-    }
-    if (callback) {
-        return (*callback)(key, keySize, value, valueSize);
+    if (mRetrieveBlob) {
+        return mRetrieveBlob(key, keySize, value, valueSize);
     }
     return 0;
 }
 
 void Platform::setDebugUpdateStatFunc(DebugUpdateStatFunc&& debugUpdateStat) noexcept {
-    std::lock_guard<decltype(mMutex)> lock(mMutex);
-    mDebugUpdateStat = std::make_shared<DebugUpdateStatFunc>(std::move(debugUpdateStat));
+    mDebugUpdateStat = std::move(debugUpdateStat);
 }
 
 bool Platform::hasDebugUpdateStatFunc() const noexcept {
-    std::lock_guard<decltype(mMutex)> lock(mMutex);
-    return mDebugUpdateStat && bool(*mDebugUpdateStat);
+    return bool(mDebugUpdateStat);
 }
 
-void Platform::debugUpdateStat(const char* key, uint64_t intValue) {
-    std::shared_ptr<DebugUpdateStatFunc> callback;
-    {
-        std::unique_lock<decltype(mMutex)> lock(mMutex);
-        callback = mDebugUpdateStat;
-    }
-    if (callback) {
-        (*callback)(key, intValue, "");
-    }
-}
-
-void Platform::debugUpdateStat(const char* key, utils::CString stringValue) {
-    std::shared_ptr<DebugUpdateStatFunc> callback;
-    {
-        std::unique_lock<decltype(mMutex)> lock(mMutex);
-        callback = mDebugUpdateStat;
-    }
-    if (callback) {
-        (*callback)(key, 0, stringValue);
+void Platform::debugUpdateStat(const char* key, uint64_t value) {
+    if (mDebugUpdateStat) {
+        mDebugUpdateStat(key, value);
     }
 }
 

filament/backend/src/metal/MetalBuffer.h

@@ -165,7 +165,7 @@ public:
          size_t size, bool forceGpuBuffer = false);
     ~MetalBuffer();
 
-    [[nodiscard]] bool wasAllocationSuccessful() const noexcept { return mBuffer; }
+    [[nodiscard]] bool wasAllocationSuccessful() const noexcept { return mBuffer || mCpuBuffer; }
 
     MetalBuffer(const MetalBuffer& rhs) = delete;
     MetalBuffer& operator=(const MetalBuffer& rhs) = delete;
@@ -185,12 +185,14 @@ public:
      * Denotes that this buffer is used for a draw call ensuring that its allocation remains valid
      * until the end of the current frame.
      *
-     * @return The MTLBuffer representing the current state of the buffer to bind, it never returns
-     * nil.
+     * @return The MTLBuffer representing the current state of the buffer to bind, or nil if there
+     * is no device allocation.
      *
      */
     id<MTLBuffer> getGpuBufferForDraw() noexcept;
 
+    void* getCpuBuffer() const noexcept { return mCpuBuffer; }
+
     void setLabel(const utils::ImmutableCString& label) {
 #if FILAMENT_METAL_DEBUG_LABELS
         if (label.empty()) {
@@ -233,6 +235,7 @@ private:
     UploadStrategy mUploadStrategy;
     TrackedMetalBuffer mBuffer;
     size_t mBufferSize = 0;
+    void* mCpuBuffer = nullptr;
     MetalContext& mContext;
 };
 

filament/backend/src/metal/MetalBuffer.mm

@@ -39,30 +39,34 @@ MetalBuffer::MetalBuffer(MetalContext& context, BufferObjectBinding bindingType,
         mUploadStrategy = UploadStrategy::POOL;
     }
 
-    MTLResourceOptions options = MTLResourceStorageModePrivate;
+    // If the buffer is less than 4K in size and is updated frequently, we don't use an explicit
+    // buffer. Instead, we use immediate command encoder methods like setVertexBytes:length:atIndex:.
+    // This won't work for SSBOs, since they are read/write.
 
-    // The buffer will be memory mapped for write operations.
-    if (any(usage & BufferUsage::SHARED_WRITE_BIT)) {
-#if defined(FILAMENT_IOS) || defined(__arm64__) || defined(__aarch64__)
-        // iOS and Apple Silicon devices use UMA (Unified Memory Architecture), so we use Shared memory.
-        options = MTLResourceStorageModeShared;
-#else
-        // Intel Macs require Managed memory for CPU/GPU synchronization.
-        options = MTLResourceStorageModeManaged;
-#endif
+    /*
+    if (size <= 4 * 1024 && bindingType != BufferObjectBinding::SHADER_STORAGE &&
+            usage == BufferUsage::DYNAMIC && !forceGpuBuffer) {
+        mBuffer = nil;
+        mCpuBuffer = malloc(size);
+        return;
     }
+    */
 
+    // Otherwise, we allocate a private GPU buffer.
     {
         ScopedAllocationTimer timer("generic");
-        mBuffer = { [context.device newBufferWithLength:size options:options],
+        mBuffer = { [context.device newBufferWithLength:size options:MTLResourceStorageModePrivate],
             TrackedMetalBuffer::Type::GENERIC };
     }
-
     // mBuffer might fail to be allocated. Clients can check for this by calling
     // wasAllocationSuccessful().
 }
 
-MetalBuffer::~MetalBuffer() = default;
+MetalBuffer::~MetalBuffer() {
+    if (mCpuBuffer) {
+        free(mCpuBuffer);
+    }
+}
 
 void MetalBuffer::copyIntoBuffer(
         void* src, size_t size, size_t byteOffset, TagResolver&& getHandleTag) {
@@ -79,6 +83,12 @@ void MetalBuffer::copyIntoBuffer(
     FILAMENT_CHECK_PRECONDITION(!(byteOffset & 0x3))
             << "byteOffset must be a multiple of 4, tag=" << getHandleTag();
 
+    // If we have a cpu buffer, we can directly copy into it.
+    if (mCpuBuffer) {
+        memcpy(static_cast<uint8_t*>(mCpuBuffer) + byteOffset, src, size);
+        return;
+    }
+
     switch (mUploadStrategy) {
         case UploadStrategy::BUMP_ALLOCATOR:
             uploadWithBumpAllocator(src, size, byteOffset, std::move(getHandleTag));
@@ -96,6 +106,11 @@ void MetalBuffer::copyIntoBufferUnsynchronized(
 }
 
 id<MTLBuffer> MetalBuffer::getGpuBufferForDraw() noexcept {
+    // If there's a CPU buffer, then we return nil here, as the CPU-side buffer will be bound
+    // separately.
+    if (mCpuBuffer) {
+        return nil;
+    }
     assert_invariant(mBuffer);
     return mBuffer.get();
 }
@@ -156,6 +171,41 @@ void MetalBuffer::bindBuffers(id<MTLCommandBuffer> cmdBuffer, id<MTLCommandEncod
                                                    offsets:metalOffsets.data()
                                                  withRange:bufferRange];
     }
+
+    for (size_t b = 0; b < count; b++) {
+        MetalBuffer* const buffer = buffers[b];
+        if (!buffer) {
+            continue;
+        }
+
+        const void* cpuBuffer = buffer->getCpuBuffer();
+        if (!cpuBuffer) {
+            continue;
+        }
+
+        const size_t bufferIndex = bufferStart + b;
+        const size_t offset = offsets[b];
+        auto* bytes = static_cast<const uint8_t*>(cpuBuffer);
+
+        if (stages & Stage::VERTEX) {
+            [(id<MTLRenderCommandEncoder>) encoder setVertexBytes:(bytes + offset)
+                                                           length:(buffer->getSize() - offset)
+                                                          atIndex:bufferIndex];
+        }
+        if (stages & Stage::FRAGMENT) {
+            [(id<MTLRenderCommandEncoder>) encoder setFragmentBytes:(bytes + offset)
+                                                             length:(buffer->getSize() - offset)
+                                                            atIndex:bufferIndex];
+        }
+        if (stages & Stage::COMPUTE) {
+            // TODO: using setBytes means the data is read-only, which currently isn't enforced.
+            // In practice this won't be an issue since MetalBuffer ensures all SSBOs are realized
+            // through actual id<MTLBuffer> allocations.
+            [(id<MTLComputeCommandEncoder>) encoder setBytes:(bytes + offset)
+                                                      length:(buffer->getSize() - offset)
+                                                     atIndex:bufferIndex];
+        }
+    }
 }
 
 void MetalBuffer::uploadWithPoolBuffer(

filament/backend/src/metal/MetalContext.h

@@ -17,7 +17,6 @@
 #ifndef TNT_METALCONTEXT_H
 #define TNT_METALCONTEXT_H
 
-#include "MetalErrorQueue.h"
 #include "MetalResourceTracker.h"
 #include "MetalShaderCompiler.h"
 #include "MetalState.h"
@@ -130,7 +129,6 @@ struct MetalContext {
     id<MTLCommandBuffer> pendingCommandBuffer = nil;
     id<MTLRenderCommandEncoder> currentRenderPassEncoder = nil;
     uint32_t currentFrame = 0;
-    MetalErrorQueue commandBufferErrors;
 
     std::atomic<bool> memorylessLimitsReached = false;
 

filament/backend/src/metal/MetalContext.mm

@@ -153,7 +153,6 @@ id<MTLCommandBuffer> getPendingCommandBuffer(MetalContext* context) {
 
         if (UTILS_UNLIKELY(errorCode != MTLCommandBufferErrorNone)) {
             logMTLCommandBufferError(errorCode);
-            context->commandBufferErrors.push(buffer.error);
         }
     }];
     FILAMENT_CHECK_POSTCONDITION(context->pendingCommandBuffer)

filament/backend/src/metal/MetalDriver.h

@@ -61,8 +61,6 @@ public:
 
     MetalContext* getContext() { return mContext; }
 
-    using DriverBase::scheduleDestroy;
-
 private:
 
     friend class MetalSwapChain;

filament/backend/src/metal/MetalDriver.mm

@@ -254,18 +254,6 @@ MetalDriver::~MetalDriver() noexcept {
 void MetalDriver::tick(int) {
     executeTickOps();
     executeDeferredOps();
-
-    // Notify platform of GPU errors.
-    auto& platform = mPlatform;
-    if (UTILS_UNLIKELY(!mContext->commandBufferErrors.isEmpty())) {
-        mContext->commandBufferErrors.flush([&platform](NSError* error) {
-            if (UTILS_VERY_UNLIKELY(!error)) {
-                return;
-            }
-            const utils::CString errorString(error.localizedDescription.UTF8String);
-            platform.debugUpdateStat("filament.metal.command_buffer_error", errorString);
-        });
-    }
 }
 
 void MetalDriver::beginFrame(int64_t monotonic_clock_ns,
@@ -2275,10 +2263,7 @@ MemoryMappedBufferHandle MetalDriver::mapBufferS() noexcept {
 void MetalDriver::mapBufferR(MemoryMappedBufferHandle mmbh,
         BufferObjectHandle boh, size_t offset,
         size_t size, MapBufferAccessFlags access, utils::ImmutableCString&& tag) {
-    assert_invariant(boh);
-    MetalBufferObject* bo = mHandleAllocator.handle_cast<MetalBufferObject*>(boh);
-    assert_invariant(bo);
-    construct_handle<MetalMemoryMappedBuffer>(mmbh, bo, offset, size, access);
+    construct_handle<MetalMemoryMappedBuffer>(mmbh, boh, offset, size, access);
     mHandleAllocator.associateTagToHandle(mmbh.getId(), std::move(tag));
 }
 
@@ -2286,16 +2271,21 @@ void MetalDriver::unmapBuffer(MemoryMappedBufferHandle mmbh) {
     if (UTILS_UNLIKELY(!mmbh)) {
         return;
     }
-
-    auto* mmb = handle_cast<MetalMemoryMappedBuffer>(mmbh);
-    mmb->unmap();
     destruct_handle<MetalMemoryMappedBuffer>(mmbh);
 }
 
 void MetalDriver::copyToMemoryMappedBuffer(MemoryMappedBufferHandle mmbh, size_t offset,
         BufferDescriptor&& data) {
-    auto* mmb = handle_cast<MetalMemoryMappedBuffer>(mmbh);
-    mmb->copy(*this, offset, std::move(data));
+    auto mmb = handle_cast<MetalMemoryMappedBuffer>(mmbh);
+
+    assert_invariant(any(mmb->access & MapBufferAccessFlags::WRITE_BIT));
+    assert_invariant(offset + data.size <= mmb->size);
+
+    // TODO: this isa zero-effort implementation of copyToMemoryMappedBuffer(), where we just
+    //       call updateBufferObject(). This could be a fallback implementation for when
+    //       shared memory is not available.
+    //       On UMA systems, this should just be a memcpy into the memory-mapped buffer.
+    updateBufferObject(mmb->boh, std::move(data), mmb->offset + offset);
 }
 
 // explicit instantiation of the Dispatcher

filament/backend/src/metal/MetalErrorQueue.h

@@ -1,66 +0,0 @@
-/*
- * Copyright (C) 2025 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef TNT_FILAMENT_DRIVER_METALERRORQUEUE_H
-#define TNT_FILAMENT_DRIVER_METALERRORQUEUE_H
-
-#import <Foundation/Foundation.h>
-
-#include <utils/compiler.h>
-
-#include <atomic>
-#include <functional>
-#include <mutex>
-#include <vector>
-
-class MetalErrorQueue {
-public:
-    bool isEmpty() const {
-        return !mHasErrors.load(std::memory_order_relaxed);
-    }
-
-    void push(NSError* error) {
-        std::lock_guard<std::mutex> lock(mMutex);
-        mErrors.push_back(error);
-        mHasErrors.store(true, std::memory_order_relaxed);
-    }
-
-    void flush(const std::function<void(NSError*)>& callback) {
-        if (UTILS_LIKELY(isEmpty())) {
-            return;
-        }
-        std::vector<NSError*> errors;
-        {
-            std::lock_guard<std::mutex> lock(mMutex);
-            std::swap(mErrors, errors);
-            mHasErrors.store(false, std::memory_order_relaxed);
-        }
-
-        for (const auto& error: errors) {
-            callback(error);
-        }
-    }
-
-private:
-    std::vector<NSError*> mErrors;
-    std::mutex mMutex;
-
-    // Optimization to avoid locking the mutex at each call to flush.
-    std::atomic<bool> mHasErrors;
-};
-
-
-#endif // TNT_FILAMENT_DRIVER_METALERRORQUEUE_H

filament/backend/src/metal/MetalHandles.h

@@ -569,22 +569,14 @@ struct MetalDescriptorSet : public HwDescriptorSet {
 
 
 struct MetalMemoryMappedBuffer : public HwMemoryMappedBuffer {
+    MetalMemoryMappedBuffer(BufferObjectHandle boh, size_t const offset,
+            size_t const size, MapBufferAccessFlags const access)
+        : boh(boh), access(access), size(size), offset(offset) {
+    }
+    BufferObjectHandle boh{};
     MapBufferAccessFlags access{};
-    struct {
-        MetalBufferObject* bo;
-        void* vaddr = nullptr;
-        uint32_t size = 0;
-        uint32_t offset = 0;
-    } mtl;
-
-    MetalMemoryMappedBuffer(MetalBufferObject* bo, size_t offset, size_t size,
-            MapBufferAccessFlags access) noexcept;
-
-    ~MetalMemoryMappedBuffer();
-
-    void unmap();
-
-    void copy(MetalDriver& mtld, size_t offset, BufferDescriptor&& data) const;
+    uint32_t size = 0;
+    uint32_t offset = 0;
 };
 
 } // namespace backend

filament/backend/src/metal/MetalHandles.mm

@@ -1664,44 +1664,5 @@ id<MTLBuffer> MetalDescriptorSet::finalizeAndGetBuffer(MetalDriver* driver, Shad
     return buffer.get();
 }
 
-MetalMemoryMappedBuffer::MetalMemoryMappedBuffer(MetalBufferObject* bo, size_t offset, size_t size,
-    MapBufferAccessFlags access) noexcept : access(access) {
-    MetalBuffer* buffer = bo->getBuffer();
-    assert_invariant(buffer);
-    id<MTLBuffer> mtlBuffer = buffer->getGpuBufferForDraw();
-
-    assert_invariant(offset + size <= bo->byteCount);
-    assert_invariant(mtlBuffer.storageMode != MTLStorageModePrivate);
-
-    mtl.bo = bo;
-    mtl.vaddr = static_cast<char*>(mtlBuffer.contents) + offset;
-    mtl.size = size;
-    mtl.offset = offset;
-}
-
-MetalMemoryMappedBuffer::~MetalMemoryMappedBuffer() = default;
-
-void MetalMemoryMappedBuffer::unmap() {
-#if !defined(FILAMENT_IOS) && defined(__x86_64__)
-    // Managed memory requires didModifyRange to synchronize changes to the GPU. This is specific to Intel Macs.
-    MetalBuffer* buffer = mtl.bo->getBuffer();
-    id<MTLBuffer> mtlBuffer = buffer->getGpuBufferForDraw();
-    if (mtlBuffer && mtlBuffer.storageMode == MTLStorageModeManaged) {
-        [mtlBuffer didModifyRange:NSMakeRange(mtl.offset, mtl.size)];
-    }
-#endif
-    // Shared memory on UMA systems is coherent; no explicit synchronization is required.
-}
-
-void MetalMemoryMappedBuffer::copy(MetalDriver& mtld, size_t offset, BufferDescriptor&& data) const {
-    assert_invariant(any(access & MapBufferAccessFlags::WRITE_BIT));
-    assert_invariant(offset + data.size <= mtl.size);
-    assert_invariant(mtl.vaddr);
-
-    memcpy(static_cast<char*>(mtl.vaddr) + offset, data.buffer, data.size);
-
-    mtld.scheduleDestroy(std::move(data));
-}
-
 } // namespace backend
 } // namespace filament

filament/backend/src/opengl/OpenGLContext.h

@@ -735,7 +735,7 @@ void OpenGLContext::bindVertexArray(RenderPrimitive const* p) noexcept {
         // - the nameVersion is out of date *and* we're on the protected context, in this case:
         //      - the name must be stale from a previous use of this context because we always
         //        destroy the protected context when we're done with it.
-        bool const recreateVaoName = vao != &mDefaultVAO &&
+        bool const recreateVaoName = p != &mDefaultVAO &&
                 ((vao->vao[contextIndex] == 0) ||
                         (vao->nameVersion != state.age && contextIndex == 1));
         if (UTILS_UNLIKELY(recreateVaoName)) {

filament/backend/src/opengl/platforms/PlatformEGLAndroid.cpp

@@ -28,13 +28,14 @@
 #include <private/backend/VirtualMachineEnv.h>
 
 #include "AndroidNativeWindow.h"
-#include "AndroidFrameCallback.h"
 #include "AndroidSwapChainHelper.h"
 #include "ExternalStreamManagerAndroid.h"
 
 #include <android/api-level.h>
 #include <android/native_window.h>
 #include <android/hardware_buffer.h>
+#include <android/choreographer.h>
+#include <android/looper.h>
 
 #include <utils/android/PerformanceHintManager.h>
 #include <utils/compiler.h>
@@ -98,27 +99,6 @@ using namespace glext;
 
 // ---------------------------------------------------------------------------------------------
 
-struct PlatformEGLAndroid::SwapChainEGLAndroid : public SwapChainEGL {
-    SwapChainEGLAndroid(PlatformEGLAndroid const& platform,
-            void* nativeWindow, uint64_t flags);
-    SwapChainEGLAndroid(PlatformEGLAndroid const& platform,
-            uint32_t width, uint32_t height, uint64_t flags);
-    void terminate(PlatformEGLAndroid& platform);
-    bool setPresentFrameId(uint64_t frameId) const noexcept;
-    uint64_t getFrameId(uint64_t frameId) const noexcept;
-    bool compositorTimingSupported = false;
-    bool frameTimestampsSupported = false;
-private:
-    AndroidSwapChainHelper mImpl{};
-};
-
-struct PlatformEGLAndroid::AndroidDetails {
-    AndroidProducerThrottling producerThrottling;
-    AndroidFrameCallback androidFrameCallback;
-};
-
-// ---------------------------------------------------------------------------------------------
-
 PlatformEGLAndroid::InitializeJvmForPerformanceManagerIfNeeded::InitializeJvmForPerformanceManagerIfNeeded() {
     // PerformanceHintManager() needs the calling thread to be a Java thread; so we need
     // to attach this thread to the JVM before we initialize PerformanceHintManager.
@@ -132,20 +112,17 @@ PlatformEGLAndroid::InitializeJvmForPerformanceManagerIfNeeded::InitializeJvmFor
 // ---------------------------------------------------------------------------------------------
 
 PlatformEGLAndroid::PlatformEGLAndroid() noexcept
-        : mExternalStreamManager(ExternalStreamManagerAndroid::create()),
-          mAndroidDetails(*(new(std::nothrow) AndroidDetails{})) {
+        : mExternalStreamManager(ExternalStreamManagerAndroid::create()) {
     mOSVersion = android_get_device_api_level();
     if (mOSVersion < 0) {
         mOSVersion = __ANDROID_API_FUTURE__;
     }
 }
 
-PlatformEGLAndroid::~PlatformEGLAndroid() noexcept {
-    delete &mAndroidDetails;
-}
+PlatformEGLAndroid::~PlatformEGLAndroid() noexcept = default;
 
 void PlatformEGLAndroid::terminate() noexcept {
-    mAndroidDetails.androidFrameCallback.terminate();
+    mAndroidFrameCallback.terminate();
     ExternalStreamManagerAndroid::destroy(&mExternalStreamManager);
     PlatformEGL::terminate();
 }
@@ -178,30 +155,20 @@ void PlatformEGLAndroid::beginFrame(
         int64_t const monotonic_clock_ns,
         int64_t refreshIntervalNs,
         uint32_t const frameId) noexcept {
-    // if frameId is 0, it means we're not associated to a particular frame, which is the case
-    // for standalone views. And in this case we skip the performance hint (since we wouldn't get
-    // the right timing anyway as well as the frame info timing collections)
+    // associate the user frameid with the system frame id
+    setPresentFrameId(mCurrentDrawSwapChain, frameId);
 
-    // associate the user frameid with the system frame id.
-    if (frameId && mCurrentDrawSwapChain) {
-        // mCurrentDrawSwapChain could be null if we're called from renderStandaloneView
-        setPresentFrameId(mCurrentDrawSwapChain, frameId);
-    }
-
-    if (frameId && mPerformanceHintSession.isValid()) {
+    if (mPerformanceHintSession.isValid()) {
         if (refreshIntervalNs <= 0) {
             // we're not provided with a target time, use the display period, if everything fails,
             // assume 16.67ms
             refreshIntervalNs = 16'666'667;
 
-            if (mCurrentDrawSwapChain) {
-                // mCurrentDrawSwapChain could be null if we're called from renderStandaloneView
-                CompositorTiming compositorTiming{};
-                bool const hasCompositorTiming =
-                        queryCompositorTiming(mCurrentDrawSwapChain, &compositorTiming);
-                if (hasCompositorTiming && compositorTiming.compositeInterval > 0) {
-                    refreshIntervalNs = compositorTiming.compositeInterval;
-                }
+            CompositorTiming compositorTiming{};
+            bool const hasCompositorTiming =
+                    queryCompositorTiming(mCurrentDrawSwapChain, &compositorTiming);
+            if (hasCompositorTiming && compositorTiming.compositeInterval > 0) {
+                refreshIntervalNs = compositorTiming.compositeInterval;
             }
         }
         mStartTimeOfActualWork = clock::time_point(std::chrono::nanoseconds(monotonic_clock_ns));
@@ -240,9 +207,9 @@ Driver* PlatformEGLAndroid::createDriver(void* sharedContext,
                     "eglGetNativeClientBufferANDROID"));
 
     if (ext.egl.ANDROID_presentation_time) {
-        eglPresentationTimeANDROID =
-                PFNEGLPRESENTATIONTIMEANDROIDPROC(eglGetProcAddress(
-                        "eglPresentationTimeANDROID"));
+        eglGetNativeClientBufferANDROID =
+                PFNEGLGETNATIVECLIENTBUFFERANDROIDPROC(eglGetProcAddress(
+                        "eglGetNativeClientBufferANDROID"));
     }
 
     if (ext.egl.ANDROID_get_frame_timestamps) {
@@ -273,7 +240,7 @@ Driver* PlatformEGLAndroid::createDriver(void* sharedContext,
 
     mAssertNativeWindowIsValid = driverConfig.assertNativeWindowIsValid;
 
-    mAndroidDetails.androidFrameCallback.init();
+    mAndroidFrameCallback.init();
 
     return driver;
 }
@@ -297,25 +264,15 @@ bool PlatformEGLAndroid::queryCompositorTiming(SwapChain const* swapchain,
     }
 
     AndroidFrameCallback::Timeline const preferredTimeline{
-            mAndroidDetails.androidFrameCallback.getPreferredTimeline() };
+            mAndroidFrameCallback.getPreferredTimeline() };
     outCompositorTiming->frameTime = preferredTimeline.frameTime;
     outCompositorTiming->expectedPresentTime = preferredTimeline.expectedPresentTime;
     outCompositorTiming->frameTimelineDeadline = preferredTimeline.frameTimelineDeadline;
-    outCompositorTiming->compositeDeadline = CompositorTiming::INVALID;
-    outCompositorTiming->compositeInterval = CompositorTiming::INVALID;
-    outCompositorTiming->compositeToPresentLatency = CompositorTiming::INVALID;
-
-    // From this point on, we always return "success" because some timings were returned.
-
-    if (!static_cast<SwapChainEGLAndroid const *>(swapchain)->compositorTimingSupported) {
-        // if this surface doesn't support it, don't attempt to query the values.
-        return true;
-    }
 
     if (UTILS_LIKELY(ext.egl.ANDROID_get_frame_timestamps)) {
         EGLSurface const sur = static_cast<SwapChainEGL const *>(swapchain)->sur;
         if (sur == EGL_NO_SURFACE) {
-            return true;
+            return false;
         }
 
         std::array<EGLnsecsANDROID, 3> values;
@@ -326,16 +283,26 @@ bool PlatformEGLAndroid::queryCompositorTiming(SwapChain const* swapchain,
         };
         EGLBoolean const success = eglGetCompositorTimingANDROID(getEglDisplay(), sur,
                 names.size(), names.data(), values.data());
-        if (UTILS_UNLIKELY(!success)) {
-            // reset current error to EGL_SUCCESS
-            eglGetError();
-        } else {
-            outCompositorTiming->compositeDeadline = values[0];
-            outCompositorTiming->compositeInterval = values[1];
-            outCompositorTiming->compositeToPresentLatency = values[2];
+        if (!success) {
+            return false;
         }
+        outCompositorTiming->compositeDeadline = values[0];
+        outCompositorTiming->compositeInterval = values[1];
+        outCompositorTiming->compositeToPresentLatency = values[2];
+        return true;
     }
-    return true;
+
+    // fallback to private APIs
+    auto const anw = static_cast<SwapChainEGL const *>(swapchain)->nativeWindow;
+    int const status = NativeWindow::getCompositorTiming(anw,
+            &outCompositorTiming->compositeDeadline,
+            &outCompositorTiming->compositeInterval,
+            &outCompositorTiming->compositeToPresentLatency);
+    if (status == 0) {
+        return true;
+    }
+
+    return PlatformEGL::queryCompositorTiming(swapchain, outCompositorTiming);
 }
 
 bool PlatformEGLAndroid::setPresentFrameId(SwapChain const* swapchain,
@@ -360,10 +327,6 @@ bool PlatformEGLAndroid::queryFrameTimestamps(SwapChain const* swapchain, uint64
         return false;
     }
 
-    if (!static_cast<SwapChainEGLAndroid const *>(swapchain)->frameTimestampsSupported) {
-        return false;
-    }
-
     if (UTILS_LIKELY(ext.egl.ANDROID_get_frame_timestamps)) {
         EGLSurface const sur = sc->sur;
         if (sur == EGL_NO_SURFACE) {
@@ -384,9 +347,7 @@ bool PlatformEGLAndroid::queryFrameTimestamps(SwapChain const* swapchain, uint64
         };
         EGLBoolean const success = eglGetFrameTimestampsANDROID(getEglDisplay(), sur, hwFrameId,
                 names.size(), names.data(), values.data());
-        if (UTILS_UNLIKELY(!success)) {
-            // reset current error to EGL_SUCCESS
-            eglGetError();
+        if (!success) {
             return false;
         }
         outFrameTimestamps->requestedPresentTime = values[0];
@@ -400,44 +361,28 @@ bool PlatformEGLAndroid::queryFrameTimestamps(SwapChain const* swapchain, uint64
         outFrameTimestamps->releaseTime = values[8];
         return true;
     }
+
+    // fallback to private APIs
+    auto const anw = sc->nativeWindow;
+    int const status = NativeWindow::getFrameTimestamps(anw, hwFrameId,
+            &outFrameTimestamps->requestedPresentTime,
+            &outFrameTimestamps->acquireTime,
+            &outFrameTimestamps->latchTime,
+            &outFrameTimestamps->firstCompositionStartTime,
+            &outFrameTimestamps->lastCompositionStartTime,
+            &outFrameTimestamps->gpuCompositionDoneTime,
+            &outFrameTimestamps->displayPresentTime,
+            &outFrameTimestamps->dequeueReadyTime,
+            &outFrameTimestamps->releaseTime);
+    if (status == 0) {
+        return true;
+    }
+
     return PlatformEGL::queryFrameTimestamps(swapchain, frameId, outFrameTimestamps);
 }
 
 Platform::SwapChain* PlatformEGLAndroid::createSwapChain(void* nativeWindow, uint64_t const flags) {
     auto* const sc = new(std::nothrow) SwapChainEGLAndroid(*this, nativeWindow, flags);
-    if (UTILS_LIKELY(ext.egl.ANDROID_get_frame_timestamps)) {
-        EGLDisplay const dpy = getEglDisplay();
-        sc->compositorTimingSupported =
-                eglGetCompositorTimingSupportedANDROID(dpy, sc->sur,
-                        EGL_COMPOSITE_DEADLINE_ANDROID) &&
-                eglGetCompositorTimingSupportedANDROID(dpy, sc->sur,
-                        EGL_COMPOSITE_INTERVAL_ANDROID) &&
-                eglGetCompositorTimingSupportedANDROID(dpy, sc->sur,
-                        EGL_COMPOSITE_TO_PRESENT_LATENCY_ANDROID);
-        sc->frameTimestampsSupported =
-                eglGetFrameTimestampSupportedANDROID(dpy, sc->sur,
-                        EGL_REQUESTED_PRESENT_TIME_ANDROID) &&
-                eglGetFrameTimestampSupportedANDROID(dpy, sc->sur,
-                        EGL_RENDERING_COMPLETE_TIME_ANDROID) &&
-                eglGetFrameTimestampSupportedANDROID(dpy, sc->sur,
-                        EGL_COMPOSITION_LATCH_TIME_ANDROID) &&
-                eglGetFrameTimestampSupportedANDROID(dpy, sc->sur,
-                        EGL_FIRST_COMPOSITION_START_TIME_ANDROID) &&
-                eglGetFrameTimestampSupportedANDROID(dpy, sc->sur,
-                        EGL_LAST_COMPOSITION_START_TIME_ANDROID) &&
-                eglGetFrameTimestampSupportedANDROID(dpy, sc->sur,
-                        EGL_FIRST_COMPOSITION_GPU_FINISHED_TIME_ANDROID) &&
-                eglGetFrameTimestampSupportedANDROID(dpy, sc->sur,
-                        EGL_DISPLAY_PRESENT_TIME_ANDROID) &&
-                eglGetFrameTimestampSupportedANDROID(dpy, sc->sur,
-                        EGL_DEQUEUE_READY_TIME_ANDROID) &&
-                eglGetFrameTimestampSupportedANDROID(dpy, sc->sur,
-                        EGL_READS_DONE_TIME_ANDROID);
-    }
-    // This is expected to be a low frequency log, only turned on in debug builds
-    DLOG(INFO) << "anw: " << nativeWindow
-            << ", compositorTimingSupported=" << sc->compositorTimingSupported
-            << ", frameTimestampsSupported=" << sc->frameTimestampsSupported;
     return sc;
 }
 
@@ -450,10 +395,6 @@ Platform::SwapChain* PlatformEGLAndroid::createSwapChain(
 void PlatformEGLAndroid::destroySwapChain(SwapChain* swapChain) noexcept {
     if (swapChain) {
         SwapChainEGLAndroid* const sc = static_cast<SwapChainEGLAndroid*>(swapChain);
-        if (mCurrentDrawSwapChain == sc) {
-            // don't keep a dangling pointer around
-            mCurrentDrawSwapChain = nullptr;
-        }
         sc->terminate(*this);
         delete sc;
     }
@@ -734,12 +675,12 @@ AcquiredImage PlatformEGLAndroid::transformAcquiredImage(AcquiredImage const sou
 
 
 bool PlatformEGLAndroid::isProducerThrottlingControlSupported() const {
-    return mAndroidDetails.producerThrottling.isSupported();
+    return mProducerThrottling.isSupported();
 }
 
 int32_t PlatformEGLAndroid::setProducerThrottlingEnabled(
     EGLNativeWindowType const nativeWindow, bool const enabled) const {
-    return mAndroidDetails.producerThrottling.setProducerThrottlingEnabled(nativeWindow, enabled);
+    return mProducerThrottling.setProducerThrottlingEnabled(nativeWindow, enabled);
 }
 
 // ---------------------------------------------------------------------------------------------
@@ -763,6 +704,8 @@ PlatformEGLAndroid::SwapChainEGLAndroid::SwapChainEGLAndroid(PlatformEGLAndroid
             // we ignore the result, it doesn't matter much if it fails
             eglSurfaceAttrib(platform.getEglDisplay(), sur, EGL_TIMESTAMPS_ANDROID, EGL_TRUE);
         }
+    } else {
+        NativeWindow::enableFrameTimestamps(EGLNativeWindowType(nativeWindow), true);
     }
 }
 

filament/backend/src/vulkan/VulkanDescriptorSetCache.cpp

@@ -19,7 +19,6 @@
 #include "VulkanCommands.h"
 #include "VulkanHandles.h"
 #include "VulkanConstants.h"
-#include "utils/compiler.h"
 
 #include <utils/FixedCapacityVector.h>
 #include <utils/Panic.h>
@@ -32,18 +31,6 @@ namespace filament::backend {
 
 namespace {
 
-using Bitmask = fvkutils::UniformBufferBitmask;
-static_assert(sizeof(Bitmask) * 8 == fvkutils::MAX_DESCRIPTOR_SET_BITMASK_BITS);
-
-Bitmask foldBitsInHalf(Bitmask bitset) {
-    Bitmask outBitset;
-    bitset.forEachSetBit([&](size_t index) {
-        constexpr size_t BITMASK_LOWER_BITS_LEN = sizeof(outBitset) * 4;
-        outBitset.set(index % BITMASK_LOWER_BITS_LEN);
-    });
-    return outBitset;
-}
-
 using DescriptorSetLayoutArray = VulkanDescriptorSetCache::DescriptorSetLayoutArray;
 using DescriptorCount = VulkanDescriptorSetCache::DescriptorCount;
 
@@ -317,7 +304,8 @@ void VulkanDescriptorSetCache::commit(VulkanCommandBuffer* commands,
         // This code actually binds the descriptor sets.
         auto set = updateSets[index];
         VkCommandBuffer const cmdbuffer = commands->buffer();
-        VkDescriptorSet const vkset = set->getVkSet();
+        VkDescriptorSet vkset = useExternalSamplers[index] ? set->getExternalSamplerVkSet() :
+            set->getVkSet();
         commands->acquire(set);
         vkCmdBindDescriptorSets(cmdbuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, index,
                 1, &vkset, set->uniqueDynamicUboCount, set->getOffsets()->data());
@@ -353,39 +341,17 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
         .pBufferInfo = &info,
     };
     vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
+
+    if (auto externalSamplerSet = set->getExternalSamplerVkSet();
+            externalSamplerSet != VK_NULL_HANDLE) {
+        descriptorWrite.dstSet = externalSamplerSet;
+        vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
+    }
     set->acquire(bufferObject);
 }
 
-void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
-        VkSampler sampler, VkDescriptorSetLayout externalSamplerLayout) noexcept {
-
-    // We have to update a bound set for two use cases
-    //   - streaming API (a changing feed of AHardwareBuffer)
-    //   - external samplers - potential changing of dataspace per-frame
-    if (UTILS_UNLIKELY(set->isBound())) {
-        auto layout = set->getLayout();
-        // Build a new descriptor set from the new layout
-        auto genLayout = externalSamplerLayout != VK_NULL_HANDLE ?
-                externalSamplerLayout : layout->getVkLayout();
-        VkDescriptorSet const newSet = getVkSet(layout->count, genLayout);
-        Bitmask const ubo = layout->bitmask.ubo | layout->bitmask.dynamicUbo;
-        Bitmask samplers = layout->bitmask.sampler;
-        samplers.unset(binding);
-
-        // Each bitmask denotes a binding index, and separated into two stages - vertex and buffer
-        // We fold the two stages into just the lower half of the bits to denote a combined set of
-        // bindings.
-        Bitmask const copyBindings = foldBitsInHalf(ubo | samplers);
-        VkDescriptorSet const srcSet = set->getVkSet();
-        copySet(srcSet, newSet, copyBindings);
-        set->addNewSet(newSet,
-                [this, layoutCount = layout->count, genLayout, newSet](VulkanDescriptorSet*) {
-                    this->manualRecycle(layoutCount, genLayout, newSet);
-                });
-    }
-
-    VkDescriptorSet const vkset = set->getVkSet();
+void VulkanDescriptorSetCache::updateSamplerImpl(VkDescriptorSet vkset, uint8_t binding,
+        fvkmemory::resource_ptr<VulkanTexture> texture, VkSampler sampler) noexcept {
     VkImageSubresourceRange range = texture->getPrimaryViewRange();
     VkImageViewType const expectedType = texture->getViewType();
     if (any(texture->usage & TextureUsage::DEPTH_ATTACHMENT) &&
@@ -410,6 +376,19 @@ void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescr
         .pImageInfo = &info,
     };
     vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
+}
+
+void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+        uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
+        VkSampler sampler) noexcept {
+    updateSamplerImpl(set->getVkSet(), binding, texture, sampler);
+    set->acquire(texture);
+}
+
+void VulkanDescriptorSetCache::updateSamplerForExternalSamplerSet(
+        fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t binding,
+        fvkmemory::resource_ptr<VulkanTexture> texture) noexcept {
+    updateSamplerImpl(set->getExternalSamplerVkSet(), binding, texture, VK_NULL_HANDLE);
     set->acquire(texture);
 }
 
@@ -426,7 +405,7 @@ fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetCache::createSet
     auto const& count = layout->count;
     auto const vklayout = layout->getVkLayout();
     auto set = fvkmemory::resource_ptr<VulkanDescriptorSet>::make(
-            mResourceManager, handle, layout,
+            mResourceManager, handle, layout->bitmask.dynamicUbo, layout->count.dynamicUbo,
             [vkSet, count, vklayout, this](
                     VulkanDescriptorSet*) { this->manualRecycle(count, vklayout, vkSet); },
             vkSet);
@@ -449,22 +428,4 @@ void VulkanDescriptorSetCache::manualRecycle(VulkanDescriptorSetLayout::Count co
 
 void VulkanDescriptorSetCache::gc() { mStashedSets = {}; }
 
-void VulkanDescriptorSetCache::copySet(VkDescriptorSet srcSet, VkDescriptorSet dstSet,
-        fvkutils::SamplerBitmask bindings) const {
-    // TODO: fix the size for better memory management
-    std::vector<VkCopyDescriptorSet> copies;
-    bindings.forEachSetBit([&](size_t index) {
-        copies.push_back({
-            .sType = VK_STRUCTURE_TYPE_COPY_DESCRIPTOR_SET,
-            .srcSet = srcSet,
-            .srcBinding = (uint32_t) index,
-            .dstSet = dstSet,
-            .dstBinding = (uint32_t) index,
-            .descriptorCount = 1,
-        });
-    });
-    vkUpdateDescriptorSets(mDevice, 0, nullptr, copies.size(), copies.data());
-}
-
-
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanDescriptorSetCache.h

@@ -55,8 +55,7 @@ public:
             VkDeviceSize size) noexcept;
 
     void updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t binding,
-            fvkmemory::resource_ptr<VulkanTexture> texture, VkSampler sampler,
-            VkDescriptorSetLayout externalSamplerLayout = VK_NULL_HANDLE) noexcept;
+            fvkmemory::resource_ptr<VulkanTexture> texture, VkSampler sampler) noexcept;
 
     void updateSamplerForExternalSamplerSet(fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t binding,
             fvkmemory::resource_ptr<VulkanTexture> texture) noexcept;
@@ -91,8 +90,8 @@ public:
     void resetCachedState() noexcept { mLastBoundInfo = {}; }
 
 private:
-    void copySet(VkDescriptorSet srcSet, VkDescriptorSet destSet,
-            fvkutils::SamplerBitmask copyBindings) const;
+    void updateSamplerImpl(VkDescriptorSet set, uint8_t binding,
+            fvkmemory::resource_ptr<VulkanTexture> texture, VkSampler sampler) noexcept;
 
     class DescriptorInfinitePool;
 

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -31,7 +31,6 @@
 #include "vulkan/memory/ResourcePointer.h"
 #include "vulkan/utils/Conversion.h"
 #include "vulkan/utils/Definitions.h"
-#include "vulkan/utils/Image.h"
 
 #include <backend/DriverEnums.h>
 #include <backend/platforms/VulkanPlatform.h>
@@ -150,31 +149,6 @@ static CallbackHandler::Callback syncCallbackWrapper = [](void* userData) {
     cbData->cb(cbData->sync, cbData->userData);
 };
 
-// This needs to be discussed because Vulkan has different default Matrix formats
-// right now this is following GL.
-static void copyMat3f(void* addr, size_t const offset, const math::mat3f& v) noexcept {
-    struct mat43 {
-        float v[3][4];
-    };
-
-    addr = static_cast<char*>(addr) + offset;
-    mat43& temp = *static_cast<mat43*>(addr);
-
-    temp.v[0][0] = v[0][0];
-    temp.v[0][1] = v[0][1];
-    temp.v[0][2] = v[0][2];
-
-    temp.v[1][0] = v[1][0];
-    temp.v[1][1] = v[1][1];
-    temp.v[1][2] = v[1][2];
-
-    temp.v[2][0] = v[2][0];
-    temp.v[2][1] = v[2][1];
-    temp.v[2][2] = v[2][2];
-
-    // don't store anything in temp.v[][3] because there could be uniforms packed there
-}
-
 }// anonymous namespace
 
 #if FVK_ENABLED(FVK_DEBUG_DEBUG_UTILS)
@@ -270,7 +244,6 @@ VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext& context,
       mQueryManager(mPlatform->getDevice()),
       mExternalImageManager(platform, &mSamplerCache, &mYcbcrConversionCache, &mDescriptorSetCache,
               &mDescriptorSetLayoutCache),
-      mStreamedImageManager(&mExternalImageManager),
       mIsSRGBSwapChainSupported(mPlatform->getCustomization().isSRGBSwapChainSupported),
       mIsMSAASwapChainSupported(false), // TODO: support MSAA swapchain
       mStereoscopicType(driverConfig.stereoscopicType) {
@@ -447,14 +420,12 @@ void VulkanDriver::beginFrame(int64_t monotonic_clock_ns,
         int64_t refreshIntervalNs, uint32_t frameId) {
     FVK_PROFILE_MARKER(PROFILE_NAME_BEGINFRAME);
 
-    // if frameId is 0, it means we're not associated to a particular frame, which is the case
-    // for standalone views. And in this case we skip the frame info timing collection.
-    if (frameId && mCurrentSwapChain) { // This should be guaranteed
+    if (mCurrentSwapChain) { // This should be guaranteed
         mPlatform->setPresentFrameId(mCurrentSwapChain->swapChain, frameId);
     }
 
     // Check if any command have finished and reset all its used resources. The resources
-    // won't be destroyed but their reference count will be decreased if the command is already
+    // wont be destroyed but their reference count will decreased if the command is already
     // completed.
     //
     // This will let us check if any VulkanBuffer is currently in flight or not.
@@ -508,9 +479,6 @@ void VulkanDriver::updateDescriptorSetTexture(
     if (UTILS_UNLIKELY(mExternalImageManager.isExternallySampledTexture(texture))) {
         mExternalImageManager.bindExternallySampledTexture(set, binding, texture, params);
         mAppState.hasBoundExternalImages = true;
-    } else if (bool(texture->getStream())) {
-        mStreamedImageManager.bindStreamedTexture(set, binding, texture, params);
-        mAppState.hasBoundExternalImages = true;
     } else {
         VulkanSamplerCache::Params cacheParams = {
             .sampler = params,
@@ -518,7 +486,6 @@ void VulkanDriver::updateDescriptorSetTexture(
         VkSampler const vksampler = mSamplerCache.getSampler(cacheParams);
         mDescriptorSetCache.updateSampler(set, binding, texture, vksampler);
         mExternalImageManager.clearTextureBinding(set, binding);
-        mStreamedImageManager.unbindStreamedTexture(set, binding);
     }
 }
 
@@ -954,6 +921,14 @@ void VulkanDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow,
         }
     }
 
+#if defined(__ANDROID__)
+    // on Android, disable producer throttling
+    if (mProducerThrottling.isSupported()) {
+        mProducerThrottling.setProducerThrottlingEnabled(
+                static_cast<ANativeWindow*>(nativeWindow), false);
+    }
+#endif
+
     auto swapChain = resource_ptr<VulkanSwapChain>::make(&mResourceManager, sch, mPlatform,
             mContext, &mResourceManager, mAllocator, &mCommands, mStagePool, nativeWindow, flags);
     swapChain.inc();
@@ -1141,12 +1116,6 @@ void VulkanDriver::destroySwapChain(Handle<HwSwapChain> sch) {
 }
 
 void VulkanDriver::destroyStream(Handle<HwStream> sh) {
-    FVK_SYSTRACE_SCOPE();
-    if (!sh) {
-        return;
-    }
-    auto stream = resource_ptr<VulkanStream>::cast(&mResourceManager, sh);
-    stream.dec();
 }
 
 void VulkanDriver::destroySync(Handle<HwSync> sh) {
@@ -1184,39 +1153,18 @@ void VulkanDriver::destroyDescriptorSet(Handle<HwDescriptorSet> dsh) {
 }
 
 Handle<HwStream> VulkanDriver::createStreamNative(void* nativeStream, utils::ImmutableCString tag) {
-    FILAMENT_CHECK_PRECONDITION(false) << "createStreamNative not supported in Vulkan.";
     return {};
 }
 
 Handle<HwStream> VulkanDriver::createStreamAcquired(utils::ImmutableCString tag) {
-    // @TODO This is still not thread-safe. We might have to revisit this question.
-    FVK_SYSTRACE_SCOPE();
-    auto handle = mResourceManager.allocHandle<VulkanStream>();
-    auto stream = resource_ptr<VulkanStream>::make(&mResourceManager, handle);
-    stream.inc();
-    return handle;
+    return {};
 }
 
 void VulkanDriver::setAcquiredImage(Handle<HwStream> sh, void* image, const math::mat3f& transform,
         CallbackHandler* handler, StreamCallback cb, void* userData) {
-    FVK_SYSTRACE_SCOPE();
-    auto stream = resource_ptr<VulkanStream>::cast(&mResourceManager, sh);
-    assert_invariant(stream->streamType == StreamType::ACQUIRED);
-    assert_invariant(image != nullptr);
-    // This covers the case where setAcquiredImage is called back to back (no updateStreams in between)
-    if (stream->previousNeedsRelease()) {
-        scheduleRelease(stream->takePrevious());
-    }
-    stream->acquire({ image, cb, userData, handler});
-    stream->setFrontEndTransform(transform);
-    mStreamsWithPendingAcquiredImage.push_back(stream);
 }
 
 void VulkanDriver::setStreamDimensions(Handle<HwStream> sh, uint32_t width, uint32_t height) {
-    FVK_SYSTRACE_SCOPE();
-    auto stream = resource_ptr<VulkanStream>::cast(&mResourceManager, sh);
-    stream->width = width;
-    stream->height = height;
 }
 
 int64_t VulkanDriver::getStreamTimestamp(Handle<HwStream> sh) {
@@ -1224,72 +1172,6 @@ int64_t VulkanDriver::getStreamTimestamp(Handle<HwStream> sh) {
 }
 
 void VulkanDriver::updateStreams(CommandStream* driver) {
-    FVK_SYSTRACE_SCOPE();
-    if (UTILS_UNLIKELY(!mStreamsWithPendingAcquiredImage.empty())) {
-        for (auto& stream: mStreamsWithPendingAcquiredImage) {
-            if (stream->previousNeedsRelease()) {
-                scheduleRelease(stream->takePrevious());
-            }
-
-            // This executes on the backend thread (updateStreams is synchonous which means it
-            // executes on the user thread) Note: stream is captured by copy which is fine, this is
-            // a copy of a resource_ptr<VulkanStream>. We only need it find the associated stream
-            // inside the mStreamedImageManager texture bindings
-            driver->queueCommand([this, stream, s = stream.get(),
-                                        image = stream->getAcquired().image,
-                                        transform = stream->getFrontEndTransform()]() {
-                auto texture = s->getTexture(image);
-                s->setBackendTransform(transform);
-                if (!texture) {
-                    auto externalImage = fvkutils::createExternalImageFromRaw(mPlatform, image, false);
-                    auto metadata = mPlatform->extractExternalImageMetadata(externalImage);
-                    auto imgData = mPlatform->createVkImageFromExternal(externalImage);
-
-                    assert_invariant(imgData.internal.valid() || imgData.external.valid());
-
-                    VkFormat vkformat = metadata.format;
-                    VkImage vkimage = VK_NULL_HANDLE;
-                    VkDeviceMemory memory = VK_NULL_HANDLE;
-                    if (imgData.internal.valid()) {
-                        metadata.externalFormat = 0;
-                        vkimage = imgData.internal.image;
-                        memory = imgData.internal.memory;
-                    } else {
-                        vkformat = VK_FORMAT_UNDEFINED;
-                        vkimage = imgData.external.image;
-                        memory = imgData.external.memory;
-                    }
-
-                    VkSamplerYcbcrConversion const conversion =
-                            mExternalImageManager.getVkSamplerYcbcrConversion(metadata);
-
-                    auto newTexture = resource_ptr<VulkanTexture>::construct(&mResourceManager,
-                            mContext, mPlatform->getDevice(), mAllocator, &mResourceManager,
-                            &mCommands, vkimage, memory, vkformat, conversion, metadata.samples,
-                            metadata.width, metadata.height, metadata.layers,
-                            metadata.filamentUsage, mStagePool);
-
-                    auto& commands = mCommands.get();
-                    // Unlike uploaded textures or swapchains, we need to explicit transition this
-                    // texture into the read layout.
-                    newTexture->transitionLayout(&commands, newTexture->getPrimaryViewRange(),
-                            VulkanLayout::FRAG_READ);
-
-                    if (imgData.external.valid()) {
-                        mExternalImageManager.addExternallySampledTexture(newTexture,
-                                externalImage);
-                        // Cache the AHB backed image. Acquires the image here.
-                        s->pushImage(image, newTexture);
-                    }
-
-                    texture = newTexture;
-                }
-
-                mStreamedImageManager.onStreamAcquireImage(texture, stream);
-            });
-        }
-        mStreamsWithPendingAcquiredImage.clear();
-    }
 }
 
 void VulkanDriver::destroyFence(Handle<HwFence> fh) {
@@ -1587,9 +1469,7 @@ void VulkanDriver::updateIndexBuffer(Handle<HwIndexBuffer> ibh, BufferDescriptor
 
 void VulkanDriver::registerBufferObjectStreams(Handle<HwBufferObject> boh,
         BufferObjectStreamDescriptor&& streams) {
-
-    auto bo = resource_ptr<VulkanBufferObject>::cast(&mResourceManager, boh);
-    mStreamUniformDescriptors[bo.get()] = std::move(streams);
+    // Noop
 }
 
 void VulkanDriver::updateBufferObject(Handle<HwBufferObject> boh, BufferDescriptor&& bd,
@@ -1598,34 +1478,6 @@ void VulkanDriver::updateBufferObject(Handle<HwBufferObject> boh, BufferDescript
 
     auto bo = resource_ptr<VulkanBufferObject>::cast(&mResourceManager, boh);
     commands.acquire(bo);
-
-    if (UTILS_UNLIKELY(!mStreamUniformDescriptors.empty())) {
-        auto streamDescriptors = mStreamUniformDescriptors.find(bo.get());
-        if (streamDescriptors != mStreamUniformDescriptors.end()) {
-            for (auto const& [offset, streamHandle, associationType]:
-                    streamDescriptors->second.mStreams) {
-                if (associationType == BufferObjectStreamAssociationType::TRANSFORM_MATRIX) {
-                    math::mat3f transform;
-                    if (streamHandle) {
-                        auto stream =
-                                resource_ptr<VulkanStream>::cast(&mResourceManager, streamHandle);
-                        if (stream->streamType == StreamType::NATIVE) {
-                            FILAMENT_CHECK_PRECONDITION(false)
-                                    << "Native Stream not supported in Vulkan.";
-                        } else {
-                            // Backend call since getStreamTransformMatrix is called from async
-                            // updateBufferObject.
-                            transform = stream->getBackEndTransform();
-                        }
-                    }
-
-                    copyMat3f(bd.buffer, offset, transform);
-                }
-            }
-            mStreamUniformDescriptors.erase(streamDescriptors);
-        }
-    }
-
     bo->loadFromCpu(commands, bd.buffer, byteOffset, bd.size);
 
     scheduleDestroy(std::move(bd));
@@ -1693,12 +1545,6 @@ TimerQueryResult VulkanDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint
 }
 
 void VulkanDriver::setExternalStream(Handle<HwTexture> th, Handle<HwStream> sh) {
-    FVK_SYSTRACE_SCOPE();
-    auto t = resource_ptr<VulkanTexture>::cast(&mResourceManager, th);
-    assert_invariant(t);
-    auto s = resource_ptr<VulkanStream>::cast(&mResourceManager, sh);
-    assert_invariant(s);
-    t->setStream(s);
 }
 
 void VulkanDriver::generateMipmaps(Handle<HwTexture> th) {

filament/backend/src/vulkan/VulkanDriver.h

@@ -34,14 +34,12 @@
 #include "vulkan/VulkanDescriptorSetCache.h"
 #include "vulkan/VulkanDescriptorSetLayoutCache.h"
 #include "vulkan/VulkanExternalImageManager.h"
-#include "vulkan/VulkanStreamedImageManager.h"
 #include "vulkan/VulkanPipelineLayoutCache.h"
 #include "vulkan/memory/ResourceManager.h"
 #include "vulkan/memory/ResourcePointer.h"
 #include "vulkan/utils/Definitions.h"
 
 #include "backend/DriverEnums.h"
-#include "backend/BufferObjectStreamDescriptor.h"
 
 #include "DriverBase.h"
 #include "private/backend/Driver.h"
@@ -50,6 +48,10 @@
 #include <utils/Allocator.h>
 #include <utils/compiler.h>
 
+#if defined(__ANDROID__)
+#include "AndroidNativeWindow.h"
+#endif
+
 namespace filament::backend {
 
 class VulkanPlatform;
@@ -159,12 +161,6 @@ private:
     VulkanDescriptorSetCache mDescriptorSetCache;
     VulkanQueryManager mQueryManager;
     VulkanExternalImageManager mExternalImageManager;
-    VulkanStreamedImageManager mStreamedImageManager;
-
-    // Stream transforms
-    std::unordered_map<VulkanBufferObject*, BufferObjectStreamDescriptor> mStreamUniformDescriptors;
-    math::mat3f getStreamTransformMatrix(Handle<HwStream> sh);
-
 
     // This maps a VulkanSwapchain to a native swapchain. VulkanSwapchain should have a copy of the
     // Platform::Swapchain pointer, but queryFrameTimestamps() and queryCompositorTiming() are
@@ -210,10 +206,9 @@ private:
     bool const mIsSRGBSwapChainSupported;
     bool const mIsMSAASwapChainSupported;
     backend::StereoscopicType const mStereoscopicType;
-
-    // setAcquiredImage is a DECL_DRIVER_API_SYNCHRONOUS_N which means we don't necessarily have the
-    // data to process it at call time. So we store it and process it during updateStreams.
-    std::vector<resource_ptr<VulkanStream>> mStreamsWithPendingAcquiredImage;
+#if defined(__ANDROID__)
+    AndroidProducerThrottling mProducerThrottling;
+#endif
 };
 
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanExternalImageManager.cpp

@@ -31,6 +31,9 @@ namespace filament::backend {
 
 namespace {
 
+using Bitmask = fvkutils::UniformBufferBitmask;
+static_assert(sizeof(Bitmask) * 8 == fvkutils::MAX_DESCRIPTOR_SET_BITMASK_BITS);
+
 template<typename T>
 void erasep(std::vector<T>& v, std::function<bool(T const&)> f) {
     auto newEnd = std::remove_if(v.begin(), v.end(), f);
@@ -47,6 +50,31 @@ ImageData& findImage(std::vector<ImageData>& images,
     return *itr;
 }
 
+void copySet(VkDevice device, VkDescriptorSet srcSet, VkDescriptorSet dstSet, Bitmask bindings) {
+    // TODO: fix the size for better memory management
+    std::vector<VkCopyDescriptorSet> copies;
+    bindings.forEachSetBit([&](size_t index) {
+        copies.push_back({
+            .sType = VK_STRUCTURE_TYPE_COPY_DESCRIPTOR_SET,
+            .srcSet = srcSet,
+            .srcBinding = (uint32_t) index,
+            .dstSet = dstSet,
+            .dstBinding = (uint32_t) index,
+            .descriptorCount = 1,
+        });
+    });
+    vkUpdateDescriptorSets(device, 0, nullptr, copies.size(), copies.data());
+}
+
+Bitmask foldBitsInHalf(Bitmask bitset) {
+    Bitmask outBitset;
+    bitset.forEachSetBit([&](size_t index) {
+        constexpr size_t BITMASK_LOWER_BITS_LEN = sizeof(outBitset) * 4;
+        outBitset.set(index % BITMASK_LOWER_BITS_LEN);
+    });
+    return outBitset;
+}
+
 }// namespace
 
 VulkanExternalImageManager::VulkanExternalImageManager(VulkanPlatform* platform,
@@ -113,9 +141,6 @@ void VulkanExternalImageManager::updateSetAndLayout(
             continue;
         }
         auto& imageData = findImage(mImages, bindingInfo.image);
-        // For non YUV images (some ext images are NOT ext FMT)
-        // getVkSamplerYcbcrConversion(metadata) will return NULL, and image->conversion will be
-        // null
         updateImage(&imageData);
 
         auto samplerParams = bindingInfo.samplerParams;
@@ -147,14 +172,38 @@ void VulkanExternalImageManager::updateSetAndLayout(
     std::for_each(samplerAndBindings.begin(), samplerAndBindings.end(),
             [&](auto const& b) { outSamplers.push_back(std::get<1>(b)); });
 
+    VkDescriptorSetLayout const oldLayout = layout->getExternalSamplerVkLayout();
     VkDescriptorSetLayout const newLayout = mDescriptorSetLayoutCache->getVkLayout(layout->bitmask,
             actualExternalSamplers, outSamplers);
-    layout->setExternalSamplerVkLayout(newLayout);
+
+    // Need to copy the set
+    VkDescriptorSet const oldSet = set->getExternalSamplerVkSet();
+    if (oldLayout != newLayout || oldSet == VK_NULL_HANDLE) {
+        // Build a new descriptor set from the new layout
+        VkDescriptorSet const newSet = mDescriptorSetCache->getVkSet(layout->count, newLayout);
+        auto const ubo = layout->bitmask.ubo | layout->bitmask.dynamicUbo;
+        auto const samplers = layout->bitmask.sampler & (~actualExternalSamplers);
+
+        // Each bitmask denotes a binding index, and separated into two stages - vertex and buffer
+        // We fold the two stages into just the lower half of the bits to denote a combined set of
+        // bindings.
+        Bitmask const copyBindings = foldBitsInHalf(ubo | samplers);
+        VkDescriptorSet const srcSet = oldSet != VK_NULL_HANDLE ? oldSet : set->getVkSet();
+        copySet(mPlatform->getDevice(), srcSet, newSet, copyBindings);
+
+        set->setExternalSamplerVkSet(newSet,
+                [&descriptorSetCache = mDescriptorSetCache, layoutCount = layout->count, newLayout,
+                        newSet](VulkanDescriptorSet*) {
+                    descriptorSetCache->manualRecycle(layoutCount, newLayout, newSet);
+                });
+        if (oldLayout != newLayout) {
+            layout->setExternalSamplerVkLayout(newLayout);
+        }
+    }
+
     // Update the external samplers in the set
     for (auto& [binding, sampler, image]: samplerAndBindings) {
-        // We cannot call updateSamplerForExternalSamplerSet because some samplers are non NULL
-        // (RGB) and we cannot do a combined update with a NULL sampler.
-        mDescriptorSetCache->updateSampler(set, binding, image, sampler, newLayout);
+        mDescriptorSetCache->updateSamplerForExternalSamplerSet(set, binding, image);
     }
 }
 
@@ -212,25 +261,13 @@ void VulkanExternalImageManager::bindExternallySampledTexture(
         fvkmemory::resource_ptr<VulkanTexture> image, SamplerParams samplerParams) {
     // Should we do duplicate validation here?
     auto& imageData = findImage(mImages, image);
-    auto itr = std::find_if(mSetBindings.begin(), mSetBindings.end(),
-            [&](SetBindingInfo const& binding) {
-                return (binding.set == set && binding.binding == bindingPoint);
-            });
-    if (itr == mSetBindings.end()) {
-        mSetBindings.push_back({ bindingPoint, imageData.image, set, samplerParams });
-    } else {
-        // override the image data in the binding point
-        itr->image = image;
-        itr->samplerParams = samplerParams;
-    }
+    mSetBindings.push_back({ bindingPoint, imageData.image, set, samplerParams });
 }
 
 void VulkanExternalImageManager::addExternallySampledTexture(
        fvkmemory::resource_ptr<VulkanTexture> image,
         Platform::ExternalImageHandleRef platformHandleRef) {
-    // By passing VK_NULL_HANDLE which is already there by default.
-    // We make it clear that all default images do NOT have a chroma conversion.
-    mImages.push_back({ image, platformHandleRef, false, VK_NULL_HANDLE });
+    mImages.push_back({ image, platformHandleRef, false });
 }
 
 void VulkanExternalImageManager::removeExternallySampledTexture(

filament/backend/src/vulkan/VulkanExternalImageManager.h

@@ -31,7 +31,7 @@ class VulkanSamplerCache;
 class VulkanDescriptorSetLayoutCache;
 class VulkanDescriptorSetCache;
 
-// Manages the logic of external images and their quirks wrt Vulkan.
+// Manages the logic of external images and their quirks wrt Vulikan.
 class VulkanExternalImageManager {
 public:
 

filament/backend/src/vulkan/VulkanHandles.cpp

@@ -40,11 +40,6 @@ namespace filament::backend {
 
 namespace {
 
-// We don't have a good estimate for this magic number. This is used to remove empty
-// slots from a descriptor set (a VulkanDescriptorset is backed by multiple
-// VkDescriptorSet.
-size_t DESCRIPTOR_SET_GC_LIMIT = 10;
-
 inline VulkanBufferBinding getBufferObjectBinding(BufferObjectBinding bindingType) noexcept {
     switch (bindingType) {
         case BufferObjectBinding::VERTEX:
@@ -197,70 +192,15 @@ VulkanDescriptorSetLayout::Bitmask VulkanDescriptorSetLayout::Bitmask::fromLayou
     return fromBackendLayout(layout);
 }
 
-VulkanDescriptorSet::~VulkanDescriptorSet() {
-    for (auto setBundle: mSets) {
-        if (setBundle.onRecycleFn) {
-            setBundle.onRecycleFn(this);
-        }
-    }
-}
-
-VulkanDescriptorSet::VulkanDescriptorSet(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout,
-        OnRecycle&& onRecycleFn, VkDescriptorSet vkSet)
-    : dynamicUboMask(layout->bitmask.dynamicUbo),
-      uniqueDynamicUboCount(layout->count.dynamicUbo),
-      mLayout(layout),
-      mCurrentSetIndex(0) {
-    addNewSet(vkSet, std::move(onRecycleFn));
-}
-
+// This method will store an age associated with this command buffer into the VulkanBuffer, which
+// will allow us to determine whether a barrier is necessary or not.
 void VulkanDescriptorSet::referencedBy(VulkanCommandBuffer& commands) {
-    // This will store an age associated with this command buffer into the VulkanBuffer, which
-    // will allow us to determine whether a barrier is necessary or not.
     mUboMask.forEachSetBit([this, &commands](size_t index) {
         auto& res = mResources[index];
         fvkmemory::resource_ptr<VulkanBufferObject> bo =
                 fvkmemory::resource_ptr<VulkanBufferObject>::cast((VulkanBufferObject*) res.get());
         bo->referencedBy(commands);
     });
-    mSets[mCurrentSetIndex].fenceStatus = commands.getFenceStatus();
-}
-
-void VulkanDescriptorSet::gc() {
-    size_t empty = 0;
-    for (auto& setBundle: mSets) {
-        assert_invariant(setBundle.onRecycleFn);
-        if (setBundle.vkSet == VK_NULL_HANDLE) {
-            empty++;
-            continue;
-        }
-        if (setBundle.fenceStatus && setBundle.fenceStatus->getStatus() == VK_SUCCESS) {
-            setBundle.onRecycleFn(this);
-            setBundle.vkSet = VK_NULL_HANDLE;
-            setBundle.fenceStatus = {};
-            empty++;
-        }
-    }
-
-    if (empty > DESCRIPTOR_SET_GC_LIMIT) {
-        std::vector<InternalVkSet> retainedSets;
-        for (auto& setBundle: mSets) {
-            if (setBundle.vkSet != VK_NULL_HANDLE) {
-                retainedSets.push_back({
-                    setBundle.vkSet,
-                    std::move(setBundle.onRecycleFn),
-                    std::move(setBundle.fenceStatus)
-                });
-            }
-        }
-        std::swap(mSets, retainedSets);
-    }
-}
-
-void VulkanDescriptorSet::addNewSet(VkDescriptorSet vkSet, OnRecycle&& onRecycleFn) {
-    gc();
-    mCurrentSetIndex = mSets.size();
-    mSets.push_back({ vkSet, std::move(onRecycleFn) });
 }
 
 PushConstantDescription::PushConstantDescription(backend::Program const& program) {

filament/backend/src/vulkan/VulkanHandles.h

@@ -28,7 +28,6 @@
 #include "VulkanTexture.h"
 #include "vulkan/VulkanCommands.h"
 #include "vulkan/memory/Resource.h"
-#include "vulkan/memory/ResourcePointer.h"
 #include "vulkan/utils/Definitions.h"
 #include "vulkan/utils/StaticVector.h"
 
@@ -58,8 +57,6 @@ inline uint8_t collapsedCount(Bitmask const& mask) {
 
 } // anonymous namespace
 
-class VulkanDescriptorSetCache;
-
 struct VulkanBufferObject;
 
 struct VulkanDescriptorSetLayout : public HwDescriptorSetLayout, fvkmemory::Resource {
@@ -162,14 +159,38 @@ public:
     using OnRecycle = std::function<void(VulkanDescriptorSet*)>;
 
     VulkanDescriptorSet(
-            fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout,
-            OnRecycle&& onRecycleFn, VkDescriptorSet vkSet);
+            fvkutils::UniformBufferBitmask const& dynamicUboMask,
+            uint8_t uniqueDynamicUboCount,
+            OnRecycle&& onRecycleFn, VkDescriptorSet vkSet)
+        : dynamicUboMask(dynamicUboMask),
+          uniqueDynamicUboCount(uniqueDynamicUboCount),
+          mVkSet(vkSet),
+          mOnRecycleFn(std::move(onRecycleFn)) {}
 
     // NOLINTNEXTLINE(bugprone-exception-escape)
-    ~VulkanDescriptorSet();
+    ~VulkanDescriptorSet() {
+        if (mOnRecycleFn) {
+            mOnRecycleFn(this);
+        }
+        if (mOnRecycleExternalSamplerFn) {
+            mOnRecycleExternalSamplerFn(this);
+        }
+    }
 
     VkDescriptorSet getVkSet() const noexcept {
-        return mSets[mCurrentSetIndex].vkSet;
+        return mVkSet;
+    }
+
+    VkDescriptorSet getExternalSamplerVkSet() const noexcept {
+        return mExternalSamplerVkSet;
+    }
+
+    void setExternalSamplerVkSet(VkDescriptorSet vkset, OnRecycle onRecycle) {
+        mExternalSamplerVkSet = vkset;
+        if (mOnRecycleExternalSamplerFn) {
+            mOnRecycleExternalSamplerFn(this);
+        }
+        mOnRecycleExternalSamplerFn = onRecycle;
     }
 
     void setOffsets(backend::DescriptorSetOffsetArray&& offsets) noexcept {
@@ -190,33 +211,17 @@ public:
 
     void referencedBy(VulkanCommandBuffer& commands);
 
-    bool isBound() const {
-        return bool(mSets[mCurrentSetIndex].fenceStatus);
-    }
-
-    fvkmemory::resource_ptr<VulkanDescriptorSetLayout> getLayout() const { return mLayout; }
-
-    fvkutils::UniformBufferBitmask const& dynamicUboMask;
+    fvkutils::UniformBufferBitmask const dynamicUboMask;
     uint8_t const uniqueDynamicUboCount;
 
 private:
-    friend class VulkanDescriptorSetCache;
+    VkDescriptorSet const mVkSet;
+    VkDescriptorSet mExternalSamplerVkSet = VK_NULL_HANDLE;
 
-    void addNewSet(VkDescriptorSet vkSet, OnRecycle&& onRecycleFn);
-
-    void gc();
-
-    struct InternalVkSet {
-        VkDescriptorSet vkSet = {};
-        OnRecycle onRecycleFn;
-        std::shared_ptr<VulkanCmdFence> fenceStatus;
-    };
-
-    fvkmemory::resource_ptr<VulkanDescriptorSetLayout> mLayout;
     backend::DescriptorSetOffsetArray mOffsets;
     std::vector<fvkmemory::resource_ptr<fvkmemory::Resource>> mResources;
-    uint8_t mCurrentSetIndex;
-    std::vector<InternalVkSet> mSets;
+    OnRecycle mOnRecycleFn;
+    OnRecycle mOnRecycleExternalSamplerFn;
     fvkutils::UniformBufferBitmask mUboMask;
 };
 

filament/backend/src/vulkan/VulkanPipelineCache.cpp

@@ -191,12 +191,9 @@ VulkanPipelineCache::PipelineCacheEntry* VulkanPipelineCache::createPipeline() n
         .alphaToCoverageEnable = raster.alphaToCoverageEnable,
         .alphaToOneEnable = VK_FALSE,
     };
-    bool const enableDepthTest =
-        raster.depthCompareOp != SamplerCompareFunc::A ||
-        raster.depthWriteEnable;
     VkPipelineDepthStencilStateCreateInfo vkDs = {
         .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
-        .depthTestEnable = enableDepthTest ? VK_TRUE : VK_FALSE,
+        .depthTestEnable = VK_TRUE,
         .depthWriteEnable = raster.depthWriteEnable,
         .depthCompareOp = fvkutils::getCompareOp(raster.depthCompareOp),
         .depthBoundsTestEnable = VK_FALSE,

filament/backend/src/vulkan/VulkanStreamedImageManager.cpp

@@ -1,60 +0,0 @@
-/*
- * Copyright (C) 2025 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include "VulkanStreamedImageManager.h"
-
-#include "VulkanDescriptorSetCache.h"
-#include "VulkanExternalImageManager.h"
-#include "VulkanDescriptorSetLayoutCache.h"
-#include "VulkanSamplerCache.h"
-
-namespace filament::backend {
-VulkanStreamedImageManager::VulkanStreamedImageManager(VulkanExternalImageManager* manager)
-        : mExternalImageManager(manager) {}
-
-VulkanStreamedImageManager::~VulkanStreamedImageManager() = default;
-
-void VulkanStreamedImageManager::terminate() { mStreamedTexturesBindings.clear(); }
-
-void VulkanStreamedImageManager::bindStreamedTexture(
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        uint8_t bindingPoint, fvkmemory::resource_ptr<VulkanTexture> image,
-        SamplerParams samplerParams) {
-    mStreamedTexturesBindings.push_back({ bindingPoint, image, set, samplerParams });
-}
-
-void VulkanStreamedImageManager::unbindStreamedTexture(
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        uint8_t bindingPoint) {
-    auto iter = std::remove_if(mStreamedTexturesBindings.begin(), mStreamedTexturesBindings.end(),
-            [&](StreamedTextureBinding& binding) {
-                return ((binding.set == set) && (binding.binding == bindingPoint));
-            });
-    mStreamedTexturesBindings.erase(iter, mStreamedTexturesBindings.end());
-}
-
-void VulkanStreamedImageManager::onStreamAcquireImage(fvkmemory::resource_ptr<VulkanTexture> image,
-        fvkmemory::resource_ptr<VulkanStream> stream) {
-    for (StreamedTextureBinding const& data: mStreamedTexturesBindings) {
-        // Find the right stream
-        if (data.image->getStream() == stream) {
-            mExternalImageManager->bindExternallySampledTexture(data.set, data.binding, image,
-                    data.samplerParams);
-        }
-    }
-}
-
-} // namespace filament::backend

filament/backend/src/vulkan/VulkanStreamedImageManager.h

@@ -1,64 +0,0 @@
-/*
- * Copyright (C) 2025 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef TNT_FILAMENT_BACKEND_CACHING_VULKANSTREAMEDIMAGEMANAGER_H
-#define TNT_FILAMENT_BACKEND_CACHING_VULKANSTREAMEDIMAGEMANAGER_H
-
-#include "VulkanHandles.h"
-
-#include <backend/DriverEnums.h>
-
-#include <array>
-#include <vector>
-
-namespace filament::backend {
-
-class VulkanExternalImageManager;
-class VulkanDescriptorSetCache;
-class VulkanSamplerCache;
-
-// Manages the logic of streamed images.
-class VulkanStreamedImageManager {
-public:
-    VulkanStreamedImageManager(
-            VulkanExternalImageManager* manager);
-    ~VulkanStreamedImageManager();
-    void terminate();
-
-public:
-    void bindStreamedTexture(fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t bindingPoint,
-            fvkmemory::resource_ptr<VulkanTexture> image, SamplerParams samplerParams);
-    void unbindStreamedTexture(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-            uint8_t bindingPoint);
-    void onStreamAcquireImage(fvkmemory::resource_ptr<VulkanTexture> image,
-            fvkmemory::resource_ptr<VulkanStream> stream);
-
-private:
-    struct StreamedTextureBinding {
-        uint8_t binding = 0;
-        fvkmemory::resource_ptr<VulkanTexture> image;
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set;
-        SamplerParams samplerParams;
-    };
-    // keep track of all the stream bindings
-    std::vector<StreamedTextureBinding> mStreamedTexturesBindings;
-
-    VulkanExternalImageManager* mExternalImageManager;
-};
-
-} // namespace filament::backend
-
-#endif // TNT_FILAMENT_BACKEND_CACHING_VULKANSTREAMEDIMAGEMANAGER_H

filament/backend/src/vulkan/VulkanTexture.h

@@ -36,49 +36,6 @@ namespace filament::backend {
 
 struct VulkanTexture;
 
-struct VulkanStream : public HwStream, fvkmemory::Resource {
-
-    //-- These methods are only called from the frontend
-    void acquire(const AcquiredImage& image) {
-        user_thread.mPrevious = user_thread.mAcquired;
-        user_thread.mAcquired = image;
-    }
-    bool previousNeedsRelease() const { return (user_thread.mPrevious.image != nullptr); }
-    // this function will null the previous once the caller takes it.
-    // It ensures we don't schedule for release twice.
-    AcquiredImage takePrevious() {
-        AcquiredImage previous = user_thread.mPrevious;
-        user_thread.mPrevious = {};
-        return previous;
-    }
-    const AcquiredImage& getAcquired() const { return user_thread.mAcquired; }
-    void setFrontEndTransform(const math::mat3f& transform) { user_thread.mTransform = transform; }
-    const math::mat3f& getFrontEndTransform() const { return user_thread.mTransform; }
-
-    //-- These methods are only called from the backend thread
-    void setBackendTransform(const math::mat3f& transform) { mTransform = transform; }
-    const math::mat3f& getBackEndTransform() const { return mTransform; }
-    fvkmemory::resource_ptr<VulkanTexture> getTexture(void* ahb) {
-        if (auto itr = mTextures.find(ahb); itr != mTextures.end()) {
-            return itr->second;
-        }
-        return {};
-    }
-    void pushImage(void* ahb, fvkmemory::resource_ptr<VulkanTexture> tex) { mTextures[ahb] = tex; }
-
-private:
-    // These are only called from the frontend
-    struct {
-        AcquiredImage mAcquired;
-        AcquiredImage mPrevious;
-        math::mat3f mTransform;
-    } user_thread;
-
-    // #TODO b/442937292
-    math::mat3f mTransform;
-    std::unordered_map<void*, fvkmemory::resource_ptr<VulkanTexture>> mTextures;
-};
-
 struct VulkanTextureState : public fvkmemory::Resource {
     VulkanTextureState(VulkanStagePool& stagePool, VulkanCommands* commands, VmaAllocator allocator,
             VkDevice device, VkImage image, VkDeviceMemory deviceMemory, VkFormat format,
@@ -117,8 +74,6 @@ private:
 
     // The texture with the sidecar owns the sidecar.
     fvkmemory::resource_ptr<VulkanTexture> mSidecarMSAA;
-    // The stream this texture is associated with.
-    fvkmemory::resource_ptr<VulkanStream> mStream;
 
     VkImage const mTextureImage;
     VkDeviceMemory const mTextureImageMemory;
@@ -150,7 +105,6 @@ private:
     using ImageViewHash = utils::hash::MurmurHashFn<ImageViewKey>;
     std::unordered_map<ImageViewKey, VkImageView, ImageViewHash> mCachedImageViews;
 
-
     friend struct VulkanTexture;
 };
 
@@ -228,14 +182,6 @@ struct VulkanTexture : public HwTexture, fvkmemory::Resource {
         return mState->mSidecarMSAA;
     }
 
-    void setStream(fvkmemory::resource_ptr<VulkanStream> stream) {
-        mState->mStream = stream;
-    }
-
-    fvkmemory::resource_ptr<VulkanStream> getStream() const {
-        return mState->mStream;
-    }
-
     bool isTransientAttachment() const {
         return mState->mUsage & VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT;
     }

filament/backend/src/vulkan/memory/Resource.cpp

@@ -41,7 +41,6 @@ template ResourceType getTypeEnum<VulkanBuffer>() noexcept;
 template ResourceType getTypeEnum<VulkanSync>() noexcept;
 template ResourceType getTypeEnum<VulkanMemoryMappedBuffer>() noexcept;
 template ResourceType getTypeEnum<VulkanSemaphore>() noexcept;
-template ResourceType getTypeEnum<VulkanStream>() noexcept;
 
 template<typename D>
 ResourceType getTypeEnum() noexcept {
@@ -105,9 +104,6 @@ ResourceType getTypeEnum() noexcept {
     if constexpr (std::is_same_v<D, VulkanSemaphore>) {
         return ResourceType::SEMAPHORE;
     }
-    if constexpr (std::is_same_v<D, VulkanStream>) {
-        return ResourceType::STREAM;
-    }
     return ResourceType::UNDEFINED_TYPE;
 }
 
@@ -153,8 +149,6 @@ std::string_view getTypeStr(ResourceType type) {
             return "VulkanMemoryMappedBuffer";
         case ResourceType::SEMAPHORE:
             return "Semaphore";
-        case ResourceType::STREAM:
-            return "VulkanStream";
         case ResourceType::UNDEFINED_TYPE:
             return "";
     }

filament/backend/src/vulkan/memory/Resource.h

@@ -55,8 +55,7 @@ enum class ResourceType : uint8_t {
     SYNC = 17,
     MEMORY_MAPPED_BUFFER = 18,
     SEMAPHORE = 19,
-    STREAM = 20,
-    UNDEFINED_TYPE = 21,    // Must be the last enum because we use it for iterating over the enums.
+    UNDEFINED_TYPE = 20,    // Must be the last enum because we use it for iterating over the enums.
 };
 
 template<typename D>

filament/backend/src/vulkan/memory/ResourceManager.cpp

@@ -123,9 +123,6 @@ void ResourceManager::destroyWithType(ResourceType type, HandleId id) {
         case ResourceType::SEMAPHORE:
             destruct<VulkanSemaphore>(Handle<VulkanSemaphore>(id));
             break;
-        case ResourceType::STREAM:
-            destruct<VulkanStream>(Handle<VulkanStream>(id));
-            break;
         case ResourceType::UNDEFINED_TYPE:
             break;
     }

filament/backend/src/vulkan/platform/VulkanPlatformAndroid.cpp

@@ -13,7 +13,6 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-
 #include <backend/platforms/VulkanPlatformAndroid.h>
 
 #include "vulkan/VulkanConstants.h"
@@ -21,8 +20,6 @@
 #include "vulkan/platform/VulkanPlatformSwapChainImpl.h"
 #include "vulkan/utils/Image.h"
 
-#include "AndroidFrameCallback.h"
-
 #include <private/backend/BackendUtilsAndroid.h>
 
 #include <backend/DriverEnums.h>
@@ -35,7 +32,6 @@
 #include <android/hardware_buffer.h>
 #include <android/native_window.h>
 
-#include <new>
 #include <utility>
 
 using namespace bluevk;
@@ -204,14 +200,6 @@ uint32_t selectMemoryTypeForExternalImage(VkPhysicalDevice physicalDevice, VkDev
 
 } // namespace
 
-// ---------------------------------------------------------------------------------------------
-
-struct VulkanPlatformAndroid::AndroidDetails {
-    AndroidFrameCallback androidFrameCallback;
-};
-
-// ---------------------------------------------------------------------------------------------
-
 VulkanPlatformAndroid::ExternalImageVulkanAndroid::~ExternalImageVulkanAndroid() {
     if (__builtin_available(android 26, *)) {
         if (aHardwareBuffer) {
@@ -220,19 +208,14 @@ VulkanPlatformAndroid::ExternalImageVulkanAndroid::~ExternalImageVulkanAndroid()
     }
 }
 
-// ---------------------------------------------------------------------------------------------
-
-VulkanPlatformAndroid::VulkanPlatformAndroid() :
-        mAndroidDetails(*(new(std::nothrow) AndroidDetails{})) {
+VulkanPlatformAndroid::VulkanPlatformAndroid() {
     mOSVersion = android_get_device_api_level();
     if (mOSVersion < 0) {
         mOSVersion = __ANDROID_API_FUTURE__;
     }
 }
 
-VulkanPlatformAndroid::~VulkanPlatformAndroid() noexcept {
-    delete &mAndroidDetails;
-}
+VulkanPlatformAndroid::~VulkanPlatformAndroid() noexcept = default;
 
 Platform::ExternalImageHandle VulkanPlatformAndroid::createExternalImage(
         AHardwareBuffer const* buffer, bool sRGB) noexcept {
@@ -514,14 +497,14 @@ int VulkanPlatformAndroid::getOSVersion() const noexcept {
 }
 
 void VulkanPlatformAndroid::terminate() {
-    mAndroidDetails.androidFrameCallback.terminate();
+    mAndroidFrameCallback.terminate();
     VulkanPlatform::terminate();
 }
 
 Driver* VulkanPlatformAndroid::createDriver(void* sharedContext, DriverConfig const& driverConfig) {
     Driver* driver = VulkanPlatform::createDriver(sharedContext, driverConfig);
     if (driver) {
-        mAndroidDetails.androidFrameCallback.init();
+        mAndroidFrameCallback.init();
     }
     return driver;
 }
@@ -537,19 +520,13 @@ bool VulkanPlatformAndroid::queryCompositorTiming(SwapChain const* swapchain,
     }
 
     AndroidFrameCallback::Timeline const preferredTimeline{
-        mAndroidDetails.androidFrameCallback.getPreferredTimeline() };
+        mAndroidFrameCallback.getPreferredTimeline() };
     outCompositorTiming->frameTime = preferredTimeline.frameTime;
     outCompositorTiming->expectedPresentTime = preferredTimeline.expectedPresentTime;
     outCompositorTiming->frameTimelineDeadline = preferredTimeline.frameTimelineDeadline;
-    outCompositorTiming->compositeDeadline = CompositorTiming::INVALID;
-    outCompositorTiming->compositeInterval = CompositorTiming::INVALID;
-    outCompositorTiming->compositeToPresentLatency = CompositorTiming::INVALID;
-
-    // From this point on, we always return "success" because some timings were returned.
 
     auto vulkanSwapchain = static_cast<VulkanPlatformSwapChainBase const *>(swapchain);
-    vulkanSwapchain->queryCompositorTiming(outCompositorTiming);
-    return true;
+    return vulkanSwapchain->queryCompositorTiming(outCompositorTiming);
 }
 
 bool VulkanPlatformAndroid::setPresentFrameId(SwapChain const* swapchain, uint64_t frameId) noexcept {

filament/backend/src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp

@@ -165,11 +165,6 @@ VulkanPlatformSurfaceSwapChain::~VulkanPlatformSurfaceSwapChain() {
 VkResult VulkanPlatformSurfaceSwapChain::create() {
 #ifdef __ANDROID__
     NativeWindow::enableFrameTimestamps(static_cast<ANativeWindow*>(mNativeWindow), true);
-    // on Android, disable producer throttling
-    if (mProducerThrottling.isSupported()) {
-        mProducerThrottling.setProducerThrottlingEnabled(
-                static_cast<ANativeWindow*>(mNativeWindow), false);
-    }
 #endif
 
     VkSurfaceFormatKHR surfaceFormat = {};
@@ -361,15 +356,13 @@ bool VulkanPlatformSurfaceSwapChain::queryCompositorTiming(
         CompositorTiming* outCompositorTiming) const {
 #ifdef __ANDROID__
     // fallback to private APIs
-    if (UTILS_VERY_LIKELY(mNativeWindow)) {
-        int const status = NativeWindow::getCompositorTiming(
-                static_cast<ANativeWindow*>(mNativeWindow),
-                &outCompositorTiming->compositeDeadline,
-                &outCompositorTiming->compositeInterval,
-                &outCompositorTiming->compositeToPresentLatency);
-        if (status == 0) {
-            return true;
-        }
+    int const status = NativeWindow::getCompositorTiming(
+            static_cast<ANativeWindow*>(mNativeWindow),
+            &outCompositorTiming->compositeDeadline,
+            &outCompositorTiming->compositeInterval,
+            &outCompositorTiming->compositeToPresentLatency);
+    if (status == 0) {
+        return true;
     }
 #endif
     return VulkanPlatformSwapChainBase::queryCompositorTiming(outCompositorTiming);

filament/backend/src/vulkan/platform/VulkanPlatformSwapChainImpl.h

@@ -25,8 +25,7 @@
 #include <bluevk/BlueVK.h>
 
 #ifdef __ANDROID__
-#include "AndroidSwapChainHelper.h"
-#include "AndroidNativeWindow.h"
+#include <AndroidSwapChainHelper.h>
 #endif
 
 #include <unordered_map>
@@ -123,7 +122,6 @@ private:
 
 #ifdef __ANDROID__
     AndroidSwapChainHelper mImpl{};
-    AndroidProducerThrottling mProducerThrottling;
 #endif
 };
 

filament/backend/src/vulkan/utils/Image.cpp

@@ -18,12 +18,6 @@
 
 #include "vulkan/VulkanTexture.h"
 
-#if defined(__ANDROID__)
-#include "backend/platforms/VulkanPlatformAndroid.h"
-#include <android/hardware_buffer.h>
-#include <android/native_window.h>
-#endif
-
 #include <utils/Panic.h>
 #include <utils/algorithm.h>
 #include <utils/debug.h>
@@ -240,19 +234,6 @@ uint8_t reduceSampleCount(uint8_t sampleCount, VkSampleCountFlags mask) {
     return mostSignificantBit((sampleCount - 1) & mask);
 }
 
-filament::backend::Platform::ExternalImageHandle createExternalImageFromRaw(
-        filament::backend::VulkanPlatform* platform, void* image, bool sRGB) {
-
-#if defined(__ANDROID__)
-    if (__builtin_available(android 26, *)) {
-        return static_cast<filament::backend::VulkanPlatformAndroid*>(platform)
-                ->createExternalImage(reinterpret_cast<AHardwareBuffer const*>(image), sRGB);
-    }
-#endif
-    return {};
-
-}
-
 } // namespace filament::backend::fvkutils
 
 bool operator<(const VkImageSubresourceRange& a, const VkImageSubresourceRange& b) {

filament/backend/src/vulkan/utils/Image.h

@@ -18,7 +18,6 @@
 #define TNT_FILAMENT_BACKEND_VULKAN_UTILS_IMAGE_H
 
 #include <backend/DriverEnums.h>
-#include <backend/platforms/VulkanPlatform.h>
 
 #include <utils/Log.h>
 
@@ -107,9 +106,6 @@ VkImageAspectFlags getImageAspect(VkFormat format);
 
 uint8_t reduceSampleCount(uint8_t sampleCount, VkSampleCountFlags mask);
 
-Platform::ExternalImageHandle createExternalImageFromRaw(filament::backend::VulkanPlatform* platform,
-        void* image,
-        bool sRGB);
 } // namespace fvkutils
 
 } // namespace filament::backend

filament/backend/src/webgpu/WebGPUBufferBase.cpp

@@ -63,7 +63,7 @@ WebGPUBufferBase::WebGPUBufferBase(wgpu::Device const& device, const wgpu::Buffe
 // WebGPU requires that the size of the data copied from the staging buffer to the GPU buffer is a
 // multiple of 4. This function handles cases where the buffer descriptor's size is not a multiple
 // of 4 by padding with zeros.
-void WebGPUBufferBase::updateGPUBuffer(BufferDescriptor const& bufferDescriptor,
+void WebGPUBufferBase::updateGPUBuffer(BufferDescriptor&& bufferDescriptor,
         const uint32_t byteOffset, wgpu::Device const& device,
         WebGPUQueueManager* const webGPUQueueManager) {
     FILAMENT_CHECK_PRECONDITION(bufferDescriptor.buffer)
@@ -89,24 +89,49 @@ void WebGPUBufferBase::updateGPUBuffer(BufferDescriptor const& bufferDescriptor,
     wgpu::BufferDescriptor descriptor{
         .label = "Filament WebGPU Staging Buffer",
         .usage = wgpu::BufferUsage::MapWrite | wgpu::BufferUsage::CopySrc,
-        .size = stagingBufferSize,
-        .mappedAtCreation = true };
+        .size = stagingBufferSize};
     wgpu::Buffer stagingBuffer = device.CreateBuffer(&descriptor);
 
-    void* mappedRange = stagingBuffer.GetMappedRange();
-    memcpy(mappedRange, bufferDescriptor.buffer, bufferDescriptor.size);
-
-    // Make sure the padded memory is set to 0 to have deterministic behaviors
-    if (remainder != 0) {
-        uint8_t* paddingStart = static_cast<uint8_t*>(mappedRange) + bufferDescriptor.size;
-        memset(paddingStart, 0, FILAMENT_WEBGPU_BUFFER_SIZE_MODULUS - remainder);
-    }
-
-    stagingBuffer.Unmap();
-
-    // Copy the staging buffer contents to the destination buffer.
-    webGPUQueueManager->getCommandEncoder().CopyBufferToBuffer(stagingBuffer, 0, mBuffer,
-            byteOffset, stagingBufferSize);
+    struct UserData final {
+        uint32_t byteOffset;
+        size_t stagingBufferSize;
+        size_t remainder;
+        BufferDescriptor srcBufferDescriptor;
+        wgpu::Buffer stagingBuffer;
+        WebGPUQueueManager* const webGPUQueueManager;
+        wgpu::Buffer dstBuffer;
+    };
+    auto userData = std::make_unique<UserData>(UserData{
+        .byteOffset = byteOffset,
+        .stagingBufferSize = stagingBufferSize,
+        .remainder = remainder,
+        .srcBufferDescriptor = std::move(bufferDescriptor),
+        .stagingBuffer = stagingBuffer,
+        .webGPUQueueManager = webGPUQueueManager,
+        .dstBuffer = mBuffer});
+    stagingBuffer.MapAsync(
+            wgpu::MapMode::Write, 0, stagingBufferSize, wgpu::CallbackMode::AllowProcessEvents,
+            [](wgpu::MapAsyncStatus status, const char* message, UserData* userdata) {
+                std::unique_ptr<UserData> data(static_cast<UserData*>(userdata));
+                if (UTILS_LIKELY(status == wgpu::MapAsyncStatus::Success)) {
+                    void* mappedRange = data->stagingBuffer.GetMappedRange();
+                    memcpy(mappedRange, data->srcBufferDescriptor.buffer,
+                            data->srcBufferDescriptor.size);
+                    if (data->remainder != 0) {
+                        uint8_t* paddingStart =
+                                static_cast<uint8_t*>(mappedRange) + data->srcBufferDescriptor.size;
+                        memset(paddingStart, 0,
+                                FILAMENT_WEBGPU_BUFFER_SIZE_MODULUS - data->remainder);
+                    }
+                    data->stagingBuffer.Unmap();
+                    data->webGPUQueueManager->getCommandEncoder().CopyBufferToBuffer(
+                            data->stagingBuffer, 0, data->dstBuffer, data->byteOffset,
+                            data->stagingBufferSize);
+                } else {
+                    FWGPU_LOGE << "Failed to map staging buffer for readPixels: " << message;
+                }
+            },
+            userData.release());
 }
 
 } // namespace filament::backend

filament/backend/src/webgpu/WebGPUBufferBase.h

@@ -39,7 +39,7 @@ public:
      * happen after draw commands encoded in the encoder. Submitting any commands up to this point
      * ensures the calls happen in the expected sequence.
      */
-    void updateGPUBuffer(BufferDescriptor const&, uint32_t byteOffset, wgpu::Device const& device,
+    void updateGPUBuffer(BufferDescriptor&&, uint32_t byteOffset, wgpu::Device const& device,
             WebGPUQueueManager* const webGPUQueueManager);
 
     [[nodiscard]] wgpu::Buffer const& getBuffer() const { return mBuffer; }

filament/backend/src/webgpu/WebGPUDriver.cpp

@@ -856,7 +856,7 @@ void WebGPUDriver::updateIndexBuffer(Handle<HwIndexBuffer> indexBufferHandle,
     // draw calls are made.
     flush();
     handleCast<WebGPUIndexBuffer>(indexBufferHandle)
-            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager);
+            ->updateGPUBuffer(std::move(bufferDescriptor), byteOffset, mDevice, &mQueueManager);
     scheduleDestroy(std::move(bufferDescriptor));
 }
 
@@ -867,14 +867,14 @@ void WebGPUDriver::updateBufferObject(Handle<HwBufferObject> bufferObjectHandle,
     // draw calls are made.
     flush();
     handleCast<WebGPUBufferObject>(bufferObjectHandle)
-            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager);
+            ->updateGPUBuffer(std::move(bufferDescriptor), byteOffset, mDevice, &mQueueManager);
     scheduleDestroy(std::move(bufferDescriptor));
 }
 
 void WebGPUDriver::updateBufferObjectUnsynchronized(Handle<HwBufferObject> bufferObjectHandle,
         BufferDescriptor&& bufferDescriptor, const uint32_t byteOffset) {
     handleCast<WebGPUBufferObject>(bufferObjectHandle)
-            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager);
+            ->updateGPUBuffer(std::move(bufferDescriptor), byteOffset, mDevice, &mQueueManager);
     scheduleDestroy(std::move(bufferDescriptor));
 }
 

filament/backend/src/webgpu/WebGPUQueueManager.cpp

@@ -21,6 +21,7 @@
 #include <chrono>
 #include <cstdint>
 #include <thread>
+#include <iostream>
 
 namespace filament::backend {
 
@@ -58,6 +59,7 @@ WebGPUQueueManager::WebGPUQueueManager(wgpu::Device const& device)
 WebGPUQueueManager::~WebGPUQueueManager() = default;
 
 wgpu::CommandEncoder WebGPUQueueManager::getCommandEncoder() {
+    // std::unique_lock<std::mutex> lock(mLock);
     if (!mCommandEncoder) {
         wgpu::CommandEncoderDescriptor commandEncoderDescriptor = {
             .label = "Filament Command Encoder",
@@ -90,7 +92,9 @@ std::shared_ptr<WebGPUSubmissionState> WebGPUQueueManager::getLatestSubmissionSt
 }
 
 void WebGPUQueueManager::submit() {
+    // std::unique_lock<std::mutex> lock(mLock);
     if (!mCommandEncoder) {
+        std::cout << "Run Yu: no mCommandEncoder found!\n";
         return;
     }
 

filament/backend/src/webgpu/WebGPUQueueManager.h

@@ -74,6 +74,7 @@ private:
     wgpu::Queue mQueue;
     wgpu::CommandEncoder mCommandEncoder;
     std::shared_ptr<WebGPUSubmissionState> mLatestSubmissionState;
+    std::mutex mLock;
 };
 
 } // namespace filament::backend

filament/include/filament/Renderer.h

@@ -113,10 +113,8 @@ public:
     /**
      * Retrieve a history of frame timing information. The maximum frame history size is
      * given by getMaxFrameHistorySize().
-     * All or part of the history can be lost when using a different SwapChain in beginFrame().
      * @param historySize requested history size. The returned vector could be smaller.
      * @return A vector of FrameInfo.
-     * @see beginFrame()
      */
     utils::FixedCapacityVector<FrameInfo> getFrameInfoHistory(
             size_t historySize = 1) const noexcept;
@@ -328,8 +326,6 @@ public:
      *                                 or 0 if unknown. This value should be the timestamp of
      *                                 the last h/w vsync. It is expressed in the
      *                                 std::chrono::steady_clock time base.
-     *                                 On Android this should be the frame time received from
-     *                                 a Choreographer.
      * @param swapChain A pointer to the SwapChain instance to use.
      *
      * @return
@@ -341,8 +337,6 @@ public:
      *
      * @note
      * All calls to render() must happen *after* beginFrame().
-     * It is recommended to use the same swapChain for every call to beginFrame, failing to do
-     * so can result is losing all or part of the FrameInfo history.
      *
      * @see
      * endFrame()

filament/src/details/BufferAllocator.cpp

@@ -16,7 +16,6 @@
 
 #include "details/BufferAllocator.h"
 
-#include <private/utils/Tracing.h>
 #include <utils/Panic.h>
 #include <utils/debug.h>
 
@@ -170,7 +169,6 @@ void BufferAllocator::releaseGpu(AllocationId id) {
 }
 
 void BufferAllocator::releaseFreeSlots() {
-    FILAMENT_TRACING_CALL(FILAMENT_TRACING_CATEGORY_FILAMENT);
     if (!mHasPendingFrees) {
         return;
     }

filament/src/details/Engine.cpp

@@ -718,7 +718,8 @@ void FEngine::prepare() {
 
     if (useUboBatching) {
         assert_invariant(mUboManager != nullptr);
-        mUboManager->beginFrame(driver);
+
+        mUboManager->beginFrame(driver, mMaterialInstances);
     }
 
     UboManager* uboManager = mUboManager;
@@ -757,7 +758,7 @@ void FEngine::gc() {
 void FEngine::submitFrame() {
     if (isUboBatchingEnabled()) {
         DriverApi& driver = getDriverApi();
-        getUboManager()->endFrame(driver);
+        getUboManager()->endFrame(driver, getMaterialInstanceResourceList());
     }
 }
 
@@ -1284,6 +1285,11 @@ UTILS_NOINLINE
 bool FEngine::destroy(const FMaterialInstance* p) {
     if (p == nullptr) return true;
 
+    if (p->isUsingUboBatching()) {
+        assert_invariant(isUboBatchingEnabled());
+        mUboManager->retireSlot(p->getAllocationId());
+    }
+
     // Check that the material instance we're destroying is not in use in the RenderableManager
     // To do this, we currently need to inspect all render primitives in the RenderableManager
     EntityManager const& em = mEntityManager;

filament/src/details/Engine.h

@@ -780,7 +780,7 @@ public:
                 bool assert_texture_can_generate_mipmap = CORRECTNESS_ASSERTION_DEFAULT;
             } debug;
             struct {
-                bool disable_gpu_frame_complete_metric = true;
+                bool disable_gpu_frame_complete_metric = false;
             } frame_info;
         } engine;
         struct {

filament/src/details/MaterialInstance.cpp

@@ -95,7 +95,6 @@ FMaterialInstance::FMaterialInstance(FEngine& engine, FMaterial const* material,
 
     if (mUseUboBatching) {
         mUboData = BufferAllocator::UNALLOCATED;
-        engine.getUboManager()->manageMaterialInstance(this);
     } else {
         mUboData = driver.createBufferObject(mUniforms.getSize(), BufferObjectBinding::UNIFORM,
                 BufferUsage::STATIC, ImmutableCString{ material->getName().c_str_safe() });
@@ -168,7 +167,6 @@ FMaterialInstance::FMaterialInstance(FEngine& engine,
 
     if (mUseUboBatching) {
         mUboData = BufferAllocator::UNALLOCATED;
-        engine.getUboManager()->manageMaterialInstance(this);
     } else {
         mUboData = driver.createBufferObject(mUniforms.getSize(), BufferObjectBinding::UNIFORM,
                 BufferUsage::DYNAMIC, ImmutableCString{ material->getName().c_str_safe() });
@@ -213,10 +211,6 @@ FMaterialInstance::~FMaterialInstance() noexcept = default;
 void FMaterialInstance::terminate(FEngine& engine) {
     FEngine::DriverApi& driver = engine.getDriverApi();
     mDescriptorSet.terminate(driver);
-    if (mUseUboBatching) {
-        engine.getUboManager()->unmanageMaterialInstance(this);
-    }
-
     auto* ubHandle = std::get_if<Handle<HwBufferObject>>(&mUboData);
     if (ubHandle){
         driver.destroyBufferObject(*ubHandle);

filament/src/details/Renderer.cpp

@@ -571,6 +571,7 @@ void FRenderer::renderStandaloneView(FView const* view) {
 
     if (UTILS_LIKELY(view->getScene())) {
         mPreviousRenderTargets.clear();
+        mFrameId++;
 
         // ask the engine to do what it needs to (e.g. updates light buffer, materials...)
         FEngine& engine = mEngine;
@@ -580,7 +581,8 @@ void FRenderer::renderStandaloneView(FView const* view) {
         driver.beginFrame(
                 steady_clock::now().time_since_epoch().count(),
                 mDisplayInfo.refreshRate == 0.0 ? 0 : int64_t(
-                        1'000'000'000.0 / mDisplayInfo.refreshRate), 0);
+                        1'000'000'000.0 / mDisplayInfo.refreshRate),
+                mFrameId);
 
         // because we don't have a "present" call, we use flush so the driver can submit
         // the command buffer; we do this before driver.endFrame() to mimic what would
@@ -589,7 +591,7 @@ void FRenderer::renderStandaloneView(FView const* view) {
 
         engine.submitFrame();
 
-        driver.endFrame(0);
+        driver.endFrame(mFrameId);
 
         // engine.flush() has already been called by renderInternal(), but we need an extra one
         // for endFrame() above. This operation in actually not too heavy, it just kicks the

filament/src/details/Renderer.h

@@ -209,7 +209,7 @@ private:
     backend::Handle<backend::HwRenderTarget> mRenderTargetHandle;
     FSwapChain* mSwapChain = nullptr;
     size_t mCommandsHighWatermark = 0;
-    uint32_t mFrameId = 1; // id 0 is reserved for standalone views
+    uint32_t mFrameId = 0;
     FrameInfoManager mFrameInfoManager;
     backend::TextureFormat mHdrTranslucent;
     backend::TextureFormat mHdrQualityMedium;

filament/src/details/UboManager.cpp

@@ -39,7 +39,7 @@ using allocation_size_t = BufferAllocator::allocation_size_t;
 // FenceManager
 // ------------------------------------------------------------------------------------------------
 
-void UboManager::FenceManager::track(DriverApi& driver, AllocationIdContainer&& allocationIds) {
+void UboManager::FenceManager::track(DriverApi& driver, std::unordered_set<AllocationId>&& allocationIds) {
     if (allocationIds.empty()) {
         return;
     }
@@ -48,7 +48,6 @@ void UboManager::FenceManager::track(DriverApi& driver, AllocationIdContainer&&
 
 void UboManager::FenceManager::reclaimCompletedResources(DriverApi& driver,
         std::function<void(AllocationId)> const& onReclaimed) {
-    FILAMENT_TRACING_CALL(FILAMENT_TRACING_CATEGORY_FILAMENT);
     uint32_t signaledCount = 0;
     bool seenSignaledFence = false;
 
@@ -116,7 +115,8 @@ UboManager::UboManager(DriverApi& driver, allocation_size_t defaultSlotSizeInByt
     reallocate(driver, defaultTotalSizeInBytes);
 }
 
-void UboManager::beginFrame(DriverApi& driver) {
+void UboManager::beginFrame(DriverApi& driver,
+        const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>& materialInstances) {
     FILAMENT_TRACING_CALL(FILAMENT_TRACING_CATEGORY_FILAMENT);
     // Check finished frames and decrement GPU count accordingly.
     mFenceManager.reclaimCompletedResources(driver,
@@ -126,7 +126,7 @@ void UboManager::beginFrame(DriverApi& driver) {
     mAllocator.releaseFreeSlots();
 
     // Traverse all MIs and see which of them need slot allocation.
-    if (allocateOnDemand() == SUCCESS) {
+    if (allocateOnDemand(materialInstances) == SUCCESS) {
         // No need to grow the buffer, so we can just map the buffer for writing and return.
         mMemoryMappedBufferHandle = driver.mapBuffer(mUbHandle, 0, mUboSize, MapBufferAccessFlags::WRITE_BIT,
                 "UboManager");
@@ -135,19 +135,25 @@ void UboManager::beginFrame(DriverApi& driver) {
     }
 
     // Calculate the required size and grow the Ubo.
-    const allocation_size_t requiredSize = calculateRequiredSize();
+    const allocation_size_t requiredSize = calculateRequiredSize(materialInstances);
     reallocate(driver, requiredSize);
 
     // Allocate slots for each MI on the new Ubo.
-    allocateAllInstances();
+    allocateAllInstances(materialInstances);
 
     // Map the buffer so that we can write to it
     mMemoryMappedBufferHandle =
             driver.mapBuffer(mUbHandle, 0, mUboSize, MapBufferAccessFlags::WRITE_BIT, "UboManager");
 
     // Invalidate the migrated MIs, so that next commit() call must be triggered.
-    for (const auto* mi : mManagedInstances) {
-        mi->getUniformBuffer().invalidate();
+    for (const auto& materialInstance : materialInstances) {
+        materialInstance.second.forEach([](const FMaterialInstance* mi) {
+            if (!mi->isUsingUboBatching()) {
+                return;
+            }
+
+            mi->getUniformBuffer().invalidate();
+        });
     }
 }
 
@@ -158,17 +164,24 @@ void UboManager::finishBeginFrame(DriverApi& driver) {
     }
 }
 
-void UboManager::endFrame(DriverApi& driver) {
-    auto allocationIds =
-            FenceManager::AllocationIdContainer::with_capacity(mManagedInstances.size());
-    for (const auto* mi: mManagedInstances) {
-        const AllocationId id = mi->getAllocationId();
-        if (UTILS_UNLIKELY(!BufferAllocator::isValid(id))) {
-            continue;
-        }
+void UboManager::endFrame(DriverApi& driver,
+        const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>& materialInstances) {
+    BufferAllocator& allocator = mAllocator;
+    std::unordered_set<AllocationId> allocationIds;
+    for (const auto& materialInstance : materialInstances) {
+        materialInstance.second.forEach([&allocator, &allocationIds](const FMaterialInstance* mi) {
+            if (!mi->isUsingUboBatching()) {
+                return;
+            }
 
-        mAllocator.acquireGpu(id);
-        allocationIds.push_back(id);
+            const AllocationId id = mi->getAllocationId();
+            if (!BufferAllocator::isValid(id)) {
+                return;
+            }
+
+            allocator.acquireGpu(id);
+            allocationIds.insert(id);
+        });
     }
 
     mFenceManager.track(driver, std::move(allocationIds));
@@ -181,90 +194,76 @@ void UboManager::terminate(DriverApi& driver) {
 
 void UboManager::updateSlot(DriverApi& driver, AllocationId id,
         BufferDescriptor bufferDescriptor) const {
-    if (!mMemoryMappedBufferHandle) {
+    if (!mMemoryMappedBufferHandle)
         return;
-    }
 
     const allocation_size_t offset = mAllocator.getAllocationOffset(id);
     driver.copyToMemoryMappedBuffer(mMemoryMappedBufferHandle, offset, std::move(bufferDescriptor));
 }
 
-void UboManager::manageMaterialInstance(FMaterialInstance* instance) {
-    mPendingInstances.insert(instance);
+void UboManager::retireSlot(BufferAllocator::AllocationId id) {
+    if (!BufferAllocator::isValid(id))
+        return;
+    mAllocator.retire(id);
 }
 
-void UboManager::unmanageMaterialInstance(FMaterialInstance* materialInstance) {
-    AllocationId id = materialInstance->getAllocationId();
-    mPendingInstances.erase(materialInstance);
-    mManagedInstances.erase(materialInstance);
-
-    if (!BufferAllocator::isValid(id)) {
-        return;
+UboManager::AllocationResult UboManager::allocateOnDemand(
+        const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>&
+                materialInstances) {
+    // Collect all MIs that need allocation into two groups.
+    std::vector<FMaterialInstance*> newInstances;
+    std::vector<FMaterialInstance*> existingInstances;
+    for (const auto& [_, miList] : materialInstances) {
+        miList.forEach([&](FMaterialInstance* mi) {
+            if (!mi->isUsingUboBatching()) {
+                return;
+            }
+            if (BufferAllocator::isValid(mi->getAllocationId())) {
+                existingInstances.push_back(mi);
+            } else {
+                newInstances.push_back(mi);
+            }
+        });
     }
 
-    mAllocator.retire(id);
-    materialInstance->assignUboAllocation(mUbHandle, BufferAllocator::UNALLOCATED, 0);
-}
-
-UboManager::AllocationResult UboManager::allocateOnDemand() {
-    FILAMENT_TRACING_CALL(FILAMENT_TRACING_CATEGORY_FILAMENT);
     bool reallocationNeeded = false;
 
     // Pass 1: Allocate slots for new material instances (that don't have a slot yet).
-    for (auto* mi : mPendingInstances) {
-        mManagedInstances.insert(mi);
+    for (FMaterialInstance* mi : newInstances) {
         auto [newId, newOffset] = mAllocator.allocate(mi->getUniformBuffer().getSize());
-
-        // Even if the newId is not valid, we assign it to the MI so that the following process knows
-        // this material instance was not allocated successfully. Then we can calculate the new
-        // required UBO size properly.
         mi->assignUboAllocation(mUbHandle, newId, newOffset);
-
         if (!BufferAllocator::isValid(newId)) {
             reallocationNeeded = true;
         }
     }
-    mPendingInstances.clear();
 
     // Pass 2: Allocate slots for existing material instances that need to be orphaned.
-    for (auto* mi: mManagedInstances) {
-        if (!BufferAllocator::isValid(mi->getAllocationId())) {
-            continue;
-        }
-
-        // This instance doesn't need orphaning.
-        if (!mi->getUniformBuffer().isDirty() || !mAllocator.isLockedByGpu(mi->getAllocationId())) {
-            continue;
-        }
-
-        mAllocator.retire(mi->getAllocationId());
-
-        // If the space is already not sufficient, we don't need to give another try on allocation.
-        if (reallocationNeeded) {
-            mi->assignUboAllocation(mUbHandle, REALLOCATION_REQUIRED, 0);
-            continue;
-        }
-
-        auto [newId, newOffset] = mAllocator.allocate(mi->getUniformBuffer().getSize());
-
-        // Even if the newId is not valid, we assign it to the MI so that the following process knows
-        // this material instance was not allocated successfully. Then we can calculate the new
-        // required UBO size properly.
-        mi->assignUboAllocation(mUbHandle, newId, newOffset);
-
-        if (!BufferAllocator::isValid(newId)) {
-            reallocationNeeded = true;
+    for (FMaterialInstance* mi : existingInstances) {
+        if (mi->getUniformBuffer().isDirty() && mAllocator.isLockedByGpu(mi->getAllocationId())) {
+            mAllocator.retire(mi->getAllocationId());
+            auto [newId, newOffset] = mAllocator.allocate(mi->getUniformBuffer().getSize());
+            mi->assignUboAllocation(mUbHandle, newId, newOffset);
+            if (!BufferAllocator::isValid(newId)) {
+                reallocationNeeded = true;
+            }
         }
     }
 
     return reallocationNeeded ? REALLOCATION_REQUIRED : SUCCESS;
 }
 
-void UboManager::allocateAllInstances() {
-    for (auto* mi: mManagedInstances) {
-        auto [newId, newOffset] = mAllocator.allocate(mi->getUniformBuffer().getSize());
-        assert_invariant(BufferAllocator::isValid(newId));
-        mi->assignUboAllocation(mUbHandle, newId, newOffset);
+void UboManager::allocateAllInstances(
+        const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>&
+                materialInstances) {
+    for (const auto& [_, miList] : materialInstances) {
+        miList.forEach([this](FMaterialInstance* mi) {
+            if (!mi->isUsingUboBatching()) {
+                return;
+            }
+            auto [newId, newOffset] = mAllocator.allocate(mi->getUniformBuffer().getSize());
+            assert_invariant(BufferAllocator::isValid(newId));
+            mi->assignUboAllocation(mUbHandle, newId, newOffset);
+        });
     }
 }
 
@@ -289,19 +288,28 @@ void UboManager::reallocate(DriverApi& driver, allocation_size_t requiredSize) {
             BufferUsage::DYNAMIC | BufferUsage::SHARED_WRITE_BIT);
 }
 
-allocation_size_t UboManager::calculateRequiredSize() {
+allocation_size_t UboManager::calculateRequiredSize(
+        const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>&
+                materialInstances) {
+    BufferAllocator& allocator = mAllocator;
     allocation_size_t newBufferSize = 0;
-    for (const auto* mi: mManagedInstances) {
-        const AllocationId allocationId = mi->getAllocationId();
-        if (allocationId == BufferAllocator::REALLOCATION_REQUIRED) {
-            // For MIs whose parameters have been updated, aside from the slot it is being
-            // occupied by the GPU, we need to preserve an additional slot for it.
-            newBufferSize += 2 * mAllocator.alignUp(mi->getUniformBuffer().getSize());
-        } else {
-            newBufferSize += mAllocator.alignUp(mi->getUniformBuffer().getSize());
-        }
+    for (const auto& materialInstance: materialInstances) {
+        materialInstance.second.forEach([&newBufferSize, &allocator](const FMaterialInstance* mi) {
+            if (!mi->isUsingUboBatching()) {
+                return;
+            }
+
+            const AllocationId allocationId = mi->getAllocationId();
+            if (allocationId == BufferAllocator::REALLOCATION_REQUIRED) {
+                // For MIs whose parameters have been updated, aside from the slot it is being
+                // occupied by the GPU, we need to preserve an additional slot for it.
+                newBufferSize += 2 * allocator.alignUp(mi->getUniformBuffer().getSize());
+            } else {
+                newBufferSize += allocator.alignUp(mi->getUniformBuffer().getSize());
+            }
+        });
     }
-    return mAllocator.alignUp(newBufferSize * BUFFER_SIZE_GROWTH_MULTIPLIER);
+    return allocator.alignUp(newBufferSize * BUFFER_SIZE_GROWTH_MULTIPLIER);
 }
 
 } // namespace filament

filament/src/details/UboManager.h

@@ -17,6 +17,7 @@
 #ifndef TNT_FILAMENT_DETAILS_UBOMANAGER_H
 #define TNT_FILAMENT_DETAILS_UBOMANAGER_H
 
+#include "ResourceList.h"
 #include "backend/DriverApiForward.h"
 
 #include "details/BufferAllocator.h"
@@ -28,8 +29,6 @@
 #include <unordered_set>
 #include <vector>
 
-class UboManagerTest;
-
 namespace filament {
 
 class FMaterial;
@@ -55,7 +54,6 @@ public:
     class FenceManager {
     public:
         using AllocationId = BufferAllocator::AllocationId;
-        using AllocationIdContainer = utils::FixedCapacityVector<AllocationId>;
 
         FenceManager() = default;
         ~FenceManager() = default;
@@ -66,7 +64,7 @@ public:
 
         // Creates a new fence to track a set of allocation IDs for the current frame.
         // This marks the beginning of GPU's usage of these resources.
-        void track(backend::DriverApi& driver, AllocationIdContainer&& allocationIds);
+        void track(backend::DriverApi& driver, std::unordered_set<AllocationId>&& allocationIds);
 
 
         // Checks all tracked fences and invokes a callback for resources associated with
@@ -81,7 +79,7 @@ public:
     private:
         // Not ideal, but we need to know which slots to decrement gpuUseCount for each frame.
         using FenceAndAllocations =
-                std::pair<backend::Handle<backend::HwFence>, AllocationIdContainer>;
+                std::pair<backend::Handle<backend::HwFence>, std::unordered_set<AllocationId>>;
         std::vector<FenceAndAllocations> mFenceAllocationList;
     };
 
@@ -98,8 +96,9 @@ public:
     //    instances with modified uniforms).
     // 3. Reallocating a larger shared UBO if the current one is insufficient.
     // 4. Mapping the shared UBO into CPU-accessible memory to prepare for uniform data writes.
-    // Note that it must happen before committing all MIs.
-    void beginFrame(backend::DriverApi& driver);
+    void beginFrame(backend::DriverApi& driver,
+            const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>&
+                    materialInstances);
 
     // Unmap the buffer here
     void finishBeginFrame(backend::DriverApi& driver);
@@ -107,31 +106,23 @@ public:
     // Create a fence and associate it with a set of allocation ids.
     // The gpuUseCount of these allocations will be incremented, and they will be decremented
     // After the corresponding frame has been done.
-    void endFrame(backend::DriverApi& driver);
+    void endFrame(backend::DriverApi& driver,
+            const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>&
+                    materialInstances);
 
     void terminate(backend::DriverApi& driver);
 
     void updateSlot(backend::DriverApi& driver, BufferAllocator::AllocationId id,
             backend::BufferDescriptor bufferDescriptor) const;
 
-    // Call this to register a new material instance to UboManager.
-    void manageMaterialInstance(FMaterialInstance* instance);
-
-    // Call this when a material instance is destroyed.
-    void unmanageMaterialInstance(FMaterialInstance* materialInstance);
+    // Call this when a material instance is no longer holding a slot. e.g. it is destroyed.
+    void retireSlot(BufferAllocator::AllocationId id);
 
     // Returns the size of the actual UBO. Note that when there's allocation failed, it will be
     // reallocated to a bigger size at the next frame.
     [[nodiscard]] BufferAllocator::allocation_size_t getTotalSize() const noexcept;
 
-    // For testing
-    [[nodiscard]] backend::MemoryMappedBufferHandle getMemoryMappedBufferHandle() const noexcept {
-        return mMemoryMappedBufferHandle;
-    }
-
 private:
-    friend class ::UboManagerTest;
-
     constexpr static float BUFFER_SIZE_GROWTH_MULTIPLIER = 1.5f;
 
     enum AllocationResult {
@@ -143,19 +134,23 @@ private:
     [[nodiscard]] BufferAllocator::allocation_size_t getAllocationOffset(
             BufferAllocator::AllocationId id) const;
 
-    AllocationResult allocateOnDemand();
+    AllocationResult allocateOnDemand(
+            const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>&
+                    materialInstances);
 
-    void allocateAllInstances();
+    void allocateAllInstances(
+            const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>&
+                    materialInstances);
 
     void reallocate(backend::DriverApi& driver, BufferAllocator::allocation_size_t requiredSize);
 
-    BufferAllocator::allocation_size_t calculateRequiredSize();
+    BufferAllocator::allocation_size_t calculateRequiredSize(
+            const std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>>&
+                    materialInstances);
 
     backend::Handle<backend::HwBufferObject> mUbHandle;
     backend::MemoryMappedBufferHandle mMemoryMappedBufferHandle;
     BufferAllocator::allocation_size_t mUboSize{};
-    std::unordered_set<FMaterialInstance*> mPendingInstances;
-    std::unordered_set<FMaterialInstance*> mManagedInstances;
 
     FenceManager mFenceManager;
     BufferAllocator mAllocator;

filament/src/details/VertexBuffer.cpp

@@ -285,45 +285,53 @@ FVertexBuffer::FVertexBuffer(FEngine& engine, const Builder& builder)
 
     // calculate buffer sizes
     size_t bufferSizes[MAX_VERTEX_BUFFER_COUNT] = {};
-
-    auto shouldCreateBuffer = [this](size_t attributeIndex) {
-        const uint8_t slot = mAttributes[attributeIndex].buffer;
-        return mDeclaredAttributes[attributeIndex] && slot != Attribute::BUFFER_UNUSED &&
-                !mBufferObjects[slot];
-    };
-    auto updateBufferSize = [&bufferSizes, this](size_t attributeIndex) {
-        const uint32_t offset = mAttributes[attributeIndex].offset;
-        const uint8_t stride = mAttributes[attributeIndex].stride;
-        const uint8_t slot = mAttributes[attributeIndex].buffer;
-        const size_t end = offset + mVertexCount * stride;
-        assert_invariant(slot < MAX_VERTEX_BUFFER_COUNT);
-        bufferSizes[slot] = std::max(bufferSizes[slot], end);
-    };
-
-    if (!mBufferObjectsEnabled) {
-        #pragma nounroll
-        for (size_t i = 0, n = mAttributes.size(); i < n; ++i) {
-            if (shouldCreateBuffer(i)) {
-                updateBufferSize(i);
+    #pragma nounroll
+    for (size_t i = 0, n = mAttributes.size(); i < n; ++i) {
+        if (mDeclaredAttributes[i]) {
+            const uint32_t offset = mAttributes[i].offset;
+            const uint8_t stride = mAttributes[i].stride;
+            const uint8_t slot = mAttributes[i].buffer;
+            const size_t end = offset + mVertexCount * stride;
+            if (slot != Attribute::BUFFER_UNUSED) {
+                assert_invariant(slot < MAX_VERTEX_BUFFER_COUNT);
+                bufferSizes[slot] = std::max(bufferSizes[slot], end);
             }
         }
-    } else if (mAdvancedSkinningEnabled) {
-        // For advanced skinning mode, only relevant buffers (BONE_INDICES & BONE_WEIGHTS) are
-        // created. We manually populated the relevant attributes for those buffers above.
-        updateBufferSize(BONE_INDICES);
-        updateBufferSize(BONE_WEIGHTS);
     }
 
-    // create buffers
-    for (size_t i = 0; i < MAX_VERTEX_BUFFER_COUNT; ++i) {
-        if (bufferSizes[i] == 0 || mBufferObjects[i]) {
-            continue;
+    if (!mBufferObjectsEnabled) {
+        // If buffer objects are not enabled at the API level, then we create them internally.
+        #pragma nounroll
+        for (size_t index = 0; index < MAX_VERTEX_ATTRIBUTE_COUNT; ++index) {
+            size_t const i = mAttributes[index].buffer;
+            if (i != Attribute::BUFFER_UNUSED) {
+                assert_invariant(bufferSizes[i] > 0);
+                if (!mBufferObjects[i]) {
+                    BufferObjectHandle const bo = driver.createBufferObject(bufferSizes[i],
+                            BufferObjectBinding::VERTEX, BufferUsage::STATIC,
+                            utils::ImmutableCString{ builder.getName() });
+                    driver.setVertexBufferObject(mHandle, i, bo);
+                    mBufferObjects[i] = bo;
+                }
+            }
+        }
+    } else {
+        // in advanced skinning mode, we manage the BONE_INDICES and BONE_WEIGHTS arrays ourselves,
+        // so we have to set the corresponding buffer objects.
+        if (mAdvancedSkinningEnabled) {
+            for (auto const index : { BONE_INDICES, BONE_WEIGHTS }) {
+                size_t const i = mAttributes[index].buffer;
+                assert_invariant(i != Attribute::BUFFER_UNUSED);
+                assert_invariant(bufferSizes[i] > 0);
+                if (!mBufferObjects[i]) {
+                    BufferObjectHandle const bo = driver.createBufferObject(bufferSizes[i],
+                            BufferObjectBinding::VERTEX, BufferUsage::STATIC,
+                            utils::ImmutableCString{ builder.getName() });
+                    driver.setVertexBufferObject(mHandle, i, bo);
+                    mBufferObjects[i] = bo;
+                }
+            }
         }
-        BufferObjectHandle const bo = driver.createBufferObject(bufferSizes[i],
-                BufferObjectBinding::VERTEX, BufferUsage::STATIC,
-                utils::ImmutableCString{ builder.getName() });
-        driver.setVertexBufferObject(mHandle, i, bo);
-        mBufferObjects[i] = bo;
     }
 }
 

filament/test/CMakeLists.txt

@@ -50,17 +50,14 @@ if (TNT_DEV)
             test_BufferAllocatorStress.cpp
             test_CircularQueue.cpp
             test_FenceManager.cpp
-            test_UboManager.cpp
             filament_test_exposure.cpp
             filament_rendering_test.cpp
             filament_bimap_test.cpp
             filament_framegraph_test.cpp
-            filament_test.cpp
-            ${RESGEN_SOURCE})
+            filament_test.cpp)
 
     target_link_libraries(test_${TARGET} PRIVATE filament gtest)
     target_compile_options(test_${TARGET} PRIVATE ${COMPILER_FLAGS})
-    target_include_directories(test_${TARGET} PRIVATE ${RESOURCE_DIR})
     set_target_properties(test_${TARGET} PROPERTIES FOLDER Tests)
 
     add_executable(test_depth depth_test.cpp)

filament/test/test_UboManager.cpp

@@ -1,366 +0,0 @@
-/*
- * Copyright (C) 2025 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include <gmock/gmock.h>
-#include <gtest/gtest.h>
-
-#include "MockDriver.h"
-#include "details/MaterialInstance.h"
-#include "details/UboManager.h"
-
-#include <private/backend/CommandBufferQueue.h>
-#include <private/backend/CommandStream.h>
-#include <private/backend/Driver.h>
-
-#include "filament_test_resources.h"
-
-namespace {
-using namespace filament;
-using namespace backend;
-
-using ::testing::NiceMock;
-
-using AllocationId = BufferAllocator::AllocationId;
-using allocation_size_t = BufferAllocator::allocation_size_t;
-} // anonymous namespace
-
-class UboManagerTest : public ::testing::Test {
-protected:
-    static constexpr size_t MIN_COMMAND_BUFFERS_SIZE = 1 * 1024 * 1024;
-    static constexpr size_t COMMAND_BUFFERS_SIZE = 3 * MIN_COMMAND_BUFFERS_SIZE;
-    static constexpr BufferAllocator::allocation_size_t DEFAULT_SLOT_SIZE = 64;
-    static constexpr BufferAllocator::allocation_size_t DEFAULT_TOTAL_SIZE = 1024;
-
-    UboManagerTest()
-            : mCommandBufferQueue(MIN_COMMAND_BUFFERS_SIZE, COMMAND_BUFFERS_SIZE, false),
-              mCommandStream(mMockDriver, mCommandBufferQueue.getCircularBuffer()),
-
-              mDriverApi(mCommandStream),
-              // The constructor will call reallocate, which calls createBufferObject.
-              // MockDriver's default ...S() implementation returns an incrementing handle.
-              // So, the first handle will be 1.
-              mUboManager(mDriverApi, DEFAULT_SLOT_SIZE, DEFAULT_TOTAL_SIZE),
-              mPendingInstances(mUboManager.mPendingInstances),
-              mManagedInstances(mUboManager.mManagedInstances),
-              mUbHandle(mUboManager.mUbHandle),
-              mAllocator(mUboManager.mAllocator) {
-        mEngine = Engine::Builder()
-                          .feature("material.enable_material_instance_uniform_batching", true)
-                          .backend(Backend::NOOP)
-                          .build();
-
-        mMaterial = Material::Builder()
-                            .package(FILAMENT_TEST_RESOURCES_TEST_MATERIAL_DATA,
-                                    FILAMENT_TEST_RESOURCES_TEST_MATERIAL_SIZE)
-                            .build(*mEngine);
-    }
-
-    void TearDown() override {
-        mEngine->destroy(mMaterial);
-        Engine::destroy(&mEngine);
-    }
-
-    // The engine is only for creating materials/material instances, we're not using the UboManager
-    // inside for testing.
-    Engine* mEngine = nullptr;
-    NiceMock<MockDriver> mMockDriver;
-    CommandBufferQueue mCommandBufferQueue;
-    CommandStream mCommandStream;
-    DriverApi& mDriverApi;
-    UboManager mUboManager;
-    Material const* mMaterial;
-    std::unordered_set<FMaterialInstance*>& mPendingInstances;
-    std::unordered_set<FMaterialInstance*>& mManagedInstances;
-    Handle<HwBufferObject>& mUbHandle;
-    BufferAllocator& mAllocator;
-};
-
-TEST_F(UboManagerTest, InitialState) {
-    EXPECT_EQ(mUboManager.getTotalSize(), DEFAULT_TOTAL_SIZE);
-    EXPECT_EQ(mMockDriver.nextFakeHandle, 2);
-    EXPECT_NE(mUbHandle.getId(), HandleBase::nullid);
-}
-
-TEST_F(UboManagerTest, BeginFrameWithoutReallocate) {
-    BufferAllocator::allocation_size_t originalBufferSize = mUboManager.getTotalSize();
-    auto mi1 = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-    EXPECT_EQ(mi1->getAllocationId(), BufferAllocator::UNALLOCATED);
-    ASSERT_TRUE(mi1->isUsingUboBatching());
-
-    // The mi1 should be put in the pending list.
-    mUboManager.manageMaterialInstance(mi1);
-    EXPECT_TRUE(mPendingInstances.contains(mi1));
-    EXPECT_FALSE(mManagedInstances.contains(mi1));
-
-    mUboManager.beginFrame(mDriverApi);
-
-    // The mi1 should be moved to managed list after beginFrame.
-    EXPECT_FALSE(mPendingInstances.contains(mi1));
-    EXPECT_TRUE(mManagedInstances.contains(mi1));
-    // Should have allocation after beginFrame.
-    EXPECT_TRUE(BufferAllocator::isValid(mi1->getAllocationId()));
-
-    // Reallocation is not triggered under this case.
-    EXPECT_EQ(mUboManager.getTotalSize(), originalBufferSize);
-    EXPECT_NE(mUboManager.getMemoryMappedBufferHandle().getId(), HandleBase::nullid);
-
-    mUboManager.finishBeginFrame(mDriverApi);
-    EXPECT_EQ(mUboManager.getMemoryMappedBufferHandle().getId(), HandleBase::nullid);
-
-    mUboManager.endFrame(mDriverApi);
-    EXPECT_TRUE(mAllocator.isLockedByGpu(mi1->getAllocationId()));
-
-    // We're not using the UboManager inside mEngine, so we need to unmanage the material instance
-    // by ourselves.
-    mUboManager.unmanageMaterialInstance(mi1);
-    EXPECT_FALSE(mPendingInstances.contains(mi1));
-    EXPECT_FALSE(mManagedInstances.contains(mi1));
-
-    mUboManager.terminate(mDriverApi);
-    mEngine->destroy(mi1);
-}
-
-TEST_F(UboManagerTest, BeginFrameWithReallocate) {
-    const allocation_size_t originalBufferSize = mUboManager.getTotalSize();
-    const Handle<HwBufferObject> originalBufferHandle = mUbHandle;
-
-    // Create enough material instances to trigger a reallocation.
-    constexpr size_t numInstances = (DEFAULT_TOTAL_SIZE / DEFAULT_SLOT_SIZE) + 1;
-    std::vector<FMaterialInstance*> instances;
-    instances.reserve(numInstances);
-
-    for (size_t i = 0; i < numInstances; ++i) {
-        auto mi = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-        instances.push_back(mi);
-        mUboManager.manageMaterialInstance(mi);
-    }
-
-    // Before beginFrame, all instances should be pending.
-    EXPECT_EQ(mPendingInstances.size(), numInstances);
-    EXPECT_TRUE(mManagedInstances.empty());
-
-    mUboManager.beginFrame(mDriverApi);
-
-    // After beginFrame, reallocation should have occurred.
-    EXPECT_NE(mUbHandle.getId(), originalBufferHandle.getId());
-    EXPECT_GT(mUboManager.getTotalSize(), originalBufferSize);
-
-    // All instances should now be managed and have valid allocations.
-    EXPECT_TRUE(mPendingInstances.empty());
-    EXPECT_EQ(mManagedInstances.size(), numInstances);
-    for (const auto* mi: instances) {
-        EXPECT_TRUE(mManagedInstances.contains(const_cast<FMaterialInstance*>(mi)));
-        EXPECT_TRUE(BufferAllocator::isValid(mi->getAllocationId()));
-    }
-
-    mUboManager.finishBeginFrame(mDriverApi);
-    mUboManager.terminate(mDriverApi);
-
-    for (auto* mi: instances) {
-        // We're not using the UboManager inside mEngine, so we need to unmanage the material instance
-        // by ourselves.
-        mUboManager.unmanageMaterialInstance(mi);
-        mEngine->destroy(mi);
-    }
-}
-
-TEST_F(UboManagerTest, RecycleSlot) {
-    auto mi1 = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-    mUboManager.manageMaterialInstance(mi1);
-
-    // Frame 1: mi1 gets an allocation.
-    mUboManager.beginFrame(mDriverApi);
-    const AllocationId mi1AllocationId = mi1->getAllocationId();
-    const allocation_size_t mi1AllocationOffset =
-            mAllocator.getAllocationOffset(mi1AllocationId);
-    EXPECT_TRUE(BufferAllocator::isValid(mi1AllocationId));
-    mUboManager.finishBeginFrame(mDriverApi);
-    mUboManager.endFrame(mDriverApi); // Locks mi1's allocation.
-
-    // Now, unmanage mi1. The slot should be retired but not yet released.
-    mUboManager.unmanageMaterialInstance(mi1);
-    EXPECT_TRUE(mAllocator.isLockedByGpu(mi1AllocationId));
-
-    // Frame 2: The slot for mi1 is still locked by the GPU.
-    // We expect getFenceStatus to be called for the fence from frame 1.
-    // We'll mock it to return TIMEOUT_EXPIRED, so the resource is not reclaimed.
-    EXPECT_CALL(mMockDriver, getFenceStatus(_)).WillOnce(Return(FenceStatus::TIMEOUT_EXPIRED));
-    mUboManager.beginFrame(mDriverApi);
-    mUboManager.finishBeginFrame(mDriverApi);
-    mUboManager.endFrame(mDriverApi);
-
-    // Frame 3: Now, we'll simulate that the fence from frame 1 has signaled.
-    // The resource for mi1 should be reclaimed.
-    EXPECT_CALL(mMockDriver, getFenceStatus(_)).WillOnce(Return(FenceStatus::CONDITION_SATISFIED));
-
-    auto mi2 = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-    mUboManager.manageMaterialInstance(mi2);
-
-    mUboManager.beginFrame(mDriverApi);
-
-    // mi2 should now have a valid allocation, and it should reuse the slot from mi1.
-    EXPECT_TRUE(BufferAllocator::isValid(mi2->getAllocationId()));
-    EXPECT_EQ(mAllocator.getAllocationOffset(mi2->getAllocationId()), mi1AllocationOffset);
-
-    mUboManager.finishBeginFrame(mDriverApi);
-    mUboManager.unmanageMaterialInstance(mi2);
-    mUboManager.terminate(mDriverApi);
-
-    mEngine->destroy(mi1);
-    mEngine->destroy(mi2);
-}
-
-TEST_F(UboManagerTest, OrphanSlot) {
-    auto mi1 = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-    mUboManager.manageMaterialInstance(mi1);
-
-    // Frame 1: mi1 gets an allocation.
-    mUboManager.beginFrame(mDriverApi);
-    const AllocationId alloc1 = mi1->getAllocationId();
-    EXPECT_TRUE(BufferAllocator::isValid(alloc1));
-    mUboManager.finishBeginFrame(mDriverApi);
-    mUboManager.endFrame(mDriverApi); // Locks alloc1.
-
-    // Frame 2: Mark the instance as dirty and begin a new frame.
-    // This should trigger orphaning.
-    mi1->getUniformBuffer().invalidate();
-    EXPECT_CALL(mMockDriver, getFenceStatus(_)).WillOnce(Return(FenceStatus::TIMEOUT_EXPIRED));
-    mUboManager.beginFrame(mDriverApi);
-
-    const AllocationId alloc2 = mi1->getAllocationId();
-    EXPECT_TRUE(BufferAllocator::isValid(alloc2));
-    EXPECT_NE(alloc1, alloc2); // Should have a new allocation.
-
-    mUboManager.finishBeginFrame(mDriverApi);
-    mUboManager.endFrame(mDriverApi); // Locks alloc2.
-
-    // Frame 3: The fence for alloc1 should now be signaled.
-    EXPECT_CALL(mMockDriver, getFenceStatus(_))
-            .WillOnce(Return(FenceStatus::TIMEOUT_EXPIRED))      // For alloc2's fence
-            .WillOnce(Return(FenceStatus::CONDITION_SATISFIED)); // For alloc1's fence
-    mUboManager.beginFrame(mDriverApi);
-    mUboManager.finishBeginFrame(mDriverApi);
-    mUboManager.unmanageMaterialInstance(mi1);
-    mUboManager.terminate(mDriverApi);
-
-    mEngine->destroy(mi1);
-}
-
-TEST_F(UboManagerTest, DoubleManage) {
-    auto mi1 = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-    mUboManager.manageMaterialInstance(mi1);
-    EXPECT_EQ(mPendingInstances.size(), 1);
-
-    // Managing the same instance again should be a no-op.
-    mUboManager.manageMaterialInstance(mi1);
-    EXPECT_EQ(mPendingInstances.size(), 1);
-
-    mUboManager.terminate(mDriverApi);
-    mEngine->destroy(mi1);
-}
-
-TEST_F(UboManagerTest, ManageAndUnmanageBeforeBeginFrame) {
-    auto mi1 = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-    mUboManager.manageMaterialInstance(mi1);
-    EXPECT_TRUE(mPendingInstances.contains(mi1));
-
-    mUboManager.unmanageMaterialInstance(mi1);
-    EXPECT_FALSE(mPendingInstances.contains(mi1));
-
-    // After beginFrame, the instance should not be in any list.
-    mUboManager.beginFrame(mDriverApi);
-    EXPECT_FALSE(mPendingInstances.contains(mi1));
-    EXPECT_FALSE(mManagedInstances.contains(mi1));
-    EXPECT_EQ(mi1->getAllocationId(), BufferAllocator::UNALLOCATED);
-
-    mUboManager.terminate(mDriverApi);
-    mEngine->destroy(mi1);
-}
-
-TEST_F(UboManagerTest, UnmanageUnmanaged) {
-    auto mi1 = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-
-    // Unmanaging an instance that was never managed should not cause any issues.
-    mUboManager.unmanageMaterialInstance(mi1);
-    EXPECT_FALSE(mPendingInstances.contains(mi1));
-    EXPECT_FALSE(mManagedInstances.contains(mi1));
-
-    mUboManager.terminate(mDriverApi);
-    mEngine->destroy(mi1);
-}
-
-TEST_F(UboManagerTest, AllAllocationsLockedAfterEndFrame) {
-    constexpr size_t numInstances = 5;
-    std::vector<FMaterialInstance*> instances;
-    instances.reserve(numInstances);
-
-    for (size_t i = 0; i < numInstances; ++i) {
-        auto mi = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-        instances.push_back(mi);
-        mUboManager.manageMaterialInstance(mi);
-    }
-
-    mUboManager.beginFrame(mDriverApi);
-    mUboManager.finishBeginFrame(mDriverApi);
-    mUboManager.endFrame(mDriverApi);
-
-    for (const auto* mi: instances) {
-        EXPECT_TRUE(mAllocator.isLockedByGpu(mi->getAllocationId()));
-    }
-
-    mUboManager.terminate(mDriverApi);
-    for (auto* mi: instances) {
-        // We're not using the UboManager inside mEngine, so we need to unmanage the material instance
-        // by ourselves.
-        mUboManager.unmanageMaterialInstance(mi);
-        mEngine->destroy(mi);
-    }
-}
-
-TEST_F(UboManagerTest, AllAllocationsLockedAfterEndFrameWithInvalidIdInBetween) {
-    constexpr size_t numInstances = 5;
-    std::vector<FMaterialInstance*> instances;
-    instances.reserve(numInstances);
-
-    for (size_t i = 0; i < numInstances; ++i) {
-        auto mi = static_cast<FMaterialInstance*>(mMaterial->createInstance());
-        instances.push_back(mi);
-        mUboManager.manageMaterialInstance(mi);
-    }
-
-    mUboManager.beginFrame(mDriverApi);
-    mUboManager.finishBeginFrame(mDriverApi);
-
-    // It should rarely happen, but we want to make sure all other instances are locked properly.
-    instances[2]->assignUboAllocation(mUbHandle, BufferAllocator::REALLOCATION_REQUIRED, 0);
-    mUboManager.endFrame(mDriverApi);
-
-    for (const auto* mi: instances) {
-        if (BufferAllocator::isValid(mi->getAllocationId())) {
-            EXPECT_TRUE(mAllocator.isLockedByGpu(mi->getAllocationId()));
-        }
-    }
-
-    mUboManager.terminate(mDriverApi);
-    for (auto* mi: instances) {
-        // We're not using the UboManager inside mEngine, so we need to unmanage the material instance
-        // by ourselves.
-        mUboManager.unmanageMaterialInstance(mi);
-        mEngine->destroy(mi);
-    }
-}
-
-// TODO: Add more tests for the beginFrame flow

ios/CocoaPods/Filament.podspec

@@ -1,12 +1,12 @@
 Pod::Spec.new do |spec|
   spec.name = "Filament"
-  spec.version = "1.68.1"
+  spec.version = "1.67.0"
   spec.license = { :type => "Apache 2.0", :file => "LICENSE" }
   spec.homepage = "https://google.github.io/filament"
   spec.authors = "Google LLC."
   spec.summary = "Filament is a real-time physically based rendering engine for Android, iOS, Windows, Linux, macOS, and WASM/WebGL."
   spec.platform = :ios, "11.0"
-  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.68.1/filament-v1.68.1-ios.tgz" }
+  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.67.0/filament-v1.67.0-ios.tgz" }
 
   # Fix linking error with Xcode 12; we do not yet support the simulator on Apple silicon.
   spec.pod_target_xcconfig = {

libs/filabridge/include/filament/MaterialEnums.h

@@ -28,7 +28,7 @@
 namespace filament {
 
 // update this when a new version of filament wouldn't work with older materials
-static constexpr size_t MATERIAL_VERSION = 68;
+static constexpr size_t MATERIAL_VERSION = 67;
 
 /**
  * Supported shading models

libs/filagui/src/ImGuiHelper.cpp

@@ -190,47 +190,6 @@ void ImGuiHelper::render(float timeStepInSeconds, Callback imguiCommands) {
 void ImGuiHelper::processImGuiCommands(ImDrawData* commands, const ImGuiIO& io) {
     ImGui::SetCurrentContext(mImGuiContext);
 
-    if (commands->Textures != nullptr) {
-        for (ImTextureData* tex : *commands->Textures) {
-            if (tex->Status == ImTextureStatus_OK) {
-                continue;
-            } else if (tex->Status == ImTextureStatus_WantCreate) {
-                IM_ASSERT(tex->TexID == 0 && tex->BackendUserData == nullptr);
-                Texture* ftex = Texture::Builder()
-                        .width(tex->Width)
-                        .height(tex->Height)
-                        .levels(1)
-                        .format(Texture::InternalFormat::RGBA8)
-                        .sampler(Texture::Sampler::SAMPLER_2D)
-                        .build(*mEngine);
-
-                IM_ASSERT(tex->Format == ImTextureFormat_RGBA32);
-                const int size = tex->Width * tex->Height * 4;
-                Texture::PixelBufferDescriptor pb(tex->GetPixels(), size,
-                                                Texture::Format::RGBA,
-                                                Texture::Type::UBYTE);
-                ftex->setImage(*mEngine, 0, std::move(pb));
-
-                tex->SetTexID((ImTextureID)ftex);
-                tex->SetStatus(ImTextureStatus_OK);
-            } else if (tex->Status == ImTextureStatus_WantUpdates) {
-                const int size = tex->Width * tex->Height * 4;
-                Texture::PixelBufferDescriptor pb(tex->GetPixels(), size,
-                                                Texture::Format::RGBA,
-                                                Texture::Type::UBYTE);
-                filament::Texture* ftex = (filament::Texture*)tex->TexID;
-                ftex->setImage(*mEngine, 0, std::move(pb));
-                tex->SetStatus(ImTextureStatus_OK);
-            } else if (tex->Status == ImTextureStatus_WantDestroy &&
-                       tex->UnusedFrames > 0) {
-              filament::Texture* ftex = (filament::Texture*)tex->TexID;
-              mEngine->destroy(ftex);
-              tex->SetTexID(ImTextureID_Invalid);
-              tex->SetStatus(ImTextureStatus_Destroyed);
-            }
-        }
-    }
-
     mHasSynced = false;
     auto& rcm = mEngine->getRenderableManager();
 
@@ -269,7 +228,7 @@ void ImGuiHelper::processImGuiCommands(ImDrawData* commands, const ImGuiIO& io)
             if (pcmd.UserCallback) {
                 pcmd.UserCallback(cmds, &pcmd);
             } else {
-                auto texture = (Texture const*)pcmd.GetTexID();
+                auto texture = (Texture const*)pcmd.TextureId;
                 MaterialInstance* materialInstance;
 #ifdef __ANDROID__
                 if (texture && texture->getTarget() == Texture::Sampler::SAMPLER_EXTERNAL) {

libs/filamentapp/CMakeLists.txt

@@ -37,13 +37,9 @@ set(SRCS
         src/IcoSphere.cpp
         src/MeshAssimp.cpp
         src/Sphere.cpp
-        src/KeyInputConversion.h
+        src/VulkanPlatformHelper.cpp
 )
 
-if (FILAMENT_SUPPORTS_VULKAN)
-    list(APPEND SRCS src/VulkanPlatformHelper.cpp)
-endif()
-
 set(LIBS
         assimp
         camutils
@@ -83,11 +79,7 @@ endif()
 
 if (WIN32)
     list(APPEND SRCS src/NativeWindowHelperWindows.cpp)
-
-    if (FILAMENT_SUPPORTS_VULKAN)
-        list(APPEND SRCS src/VulkanPlatformHelperWindows.cpp)
-    endif()
-
+    list(APPEND SRCS src/VulkanPlatformHelperWindows.cpp)
     list(APPEND LIBS sdl2main)
 endif()
 

libs/filamentapp/src/FilamentApp.cpp

@@ -16,8 +16,6 @@
 
 #include <filamentapp/FilamentApp.h>
 
-#include "KeyInputConversion.h"
-
 #if defined(WIN32)
 #    include <SDL_syswm.h>
 #    include <utils/unwindows.h>
@@ -213,6 +211,27 @@ void FilamentApp::run(const Config& config, SetupCallback setupCallback,
             SDL_GetWindowWMInfo(window->getSDLWindow(), &wmInfo);
             ImGui::GetMainViewport()->PlatformHandleRaw = wmInfo.info.win.window;
         #endif
+        io.KeyMap[ImGuiKey_Tab] = SDL_SCANCODE_TAB;
+        io.KeyMap[ImGuiKey_LeftArrow] = SDL_SCANCODE_LEFT;
+        io.KeyMap[ImGuiKey_RightArrow] = SDL_SCANCODE_RIGHT;
+        io.KeyMap[ImGuiKey_UpArrow] = SDL_SCANCODE_UP;
+        io.KeyMap[ImGuiKey_DownArrow] = SDL_SCANCODE_DOWN;
+        io.KeyMap[ImGuiKey_PageUp] = SDL_SCANCODE_PAGEUP;
+        io.KeyMap[ImGuiKey_PageDown] = SDL_SCANCODE_PAGEDOWN;
+        io.KeyMap[ImGuiKey_Home] = SDL_SCANCODE_HOME;
+        io.KeyMap[ImGuiKey_End] = SDL_SCANCODE_END;
+        io.KeyMap[ImGuiKey_Insert] = SDL_SCANCODE_INSERT;
+        io.KeyMap[ImGuiKey_Delete] = SDL_SCANCODE_DELETE;
+        io.KeyMap[ImGuiKey_Backspace] = SDL_SCANCODE_BACKSPACE;
+        io.KeyMap[ImGuiKey_Space] = SDL_SCANCODE_SPACE;
+        io.KeyMap[ImGuiKey_Enter] = SDL_SCANCODE_RETURN;
+        io.KeyMap[ImGuiKey_Escape] = SDL_SCANCODE_ESCAPE;
+        io.KeyMap[ImGuiKey_A] = SDL_SCANCODE_A;
+        io.KeyMap[ImGuiKey_C] = SDL_SCANCODE_C;
+        io.KeyMap[ImGuiKey_V] = SDL_SCANCODE_V;
+        io.KeyMap[ImGuiKey_X] = SDL_SCANCODE_X;
+        io.KeyMap[ImGuiKey_Y] = SDL_SCANCODE_Y;
+        io.KeyMap[ImGuiKey_Z] = SDL_SCANCODE_Z;
         io.SetClipboardTextFn = [](void*, const char* text) {
             SDL_SetClipboardText(text);
         };
@@ -286,19 +305,15 @@ void FilamentApp::run(const Config& config, SetupCallback setupCallback,
                         io.AddInputCharactersUTF8(event->text.text);
                         break;
                     }
-                    case SDL_KEYUP:
-                    case SDL_KEYDOWN: {
-                        SDL_Scancode const scancode = event->key.keysym.scancode;
-                        SDL_Keycode const keycode = event->key.keysym.sym;
-
-                        auto modState = SDL_GetModState();
-                        io.AddKeyEvent(ImGuiMod_Ctrl, (modState & KMOD_CTRL) != 0);
-                        io.AddKeyEvent(ImGuiMod_Shift, (modState & KMOD_SHIFT) != 0);
-                        io.AddKeyEvent(ImGuiMod_Alt, (modState & KMOD_ALT) != 0);
-                        io.AddKeyEvent(ImGuiMod_Super, (modState & KMOD_GUI) != 0);
-                        io.AddKeyEvent(
-                                filamentapp_utils::ImGui_ImplSDL2_KeyEventToImGuiKey(keycode, scancode),
-                                event->type == SDL_KEYDOWN);
+                    case SDL_KEYDOWN:
+                    case SDL_KEYUP: {
+                        int key = event->key.keysym.scancode;
+                        IM_ASSERT(key >= 0 && key < IM_ARRAYSIZE(io.KeysDown));
+                        io.KeysDown[key] = (event->type == SDL_KEYDOWN);
+                        io.KeyShift = ((SDL_GetModState() & KMOD_SHIFT) != 0);
+                        io.KeyAlt = ((SDL_GetModState() & KMOD_ALT) != 0);
+                        io.KeyCtrl = ((SDL_GetModState() & KMOD_CTRL) != 0);
+                        io.KeySuper = ((SDL_GetModState() & KMOD_GUI) != 0);
                         break;
                     }
                 }

libs/filamentapp/src/KeyInputConversion.h

@@ -1,297 +0,0 @@
-/*
- * Copyright (C) 2025 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef TNT_FILAMENT_FILAMENTAPP_KEYINPUTCONVERSION_H
-#define TNT_FILAMENT_FILAMENTAPP_KEYINPUTCONVERSION_H
-
-#include <imgui.h>
-#include <SDL.h>
-
-namespace filamentapp_utils {
-
-// Copied from
-// https://github.com/ocornut/imgui/blob/940627d008b8f0584b7f50d24574537cf24a32e1/backends/imgui_impl_sdl2.cpp#L206
-ImGuiKey ImGui_ImplSDL2_KeyEventToImGuiKey(SDL_Keycode keycode, SDL_Scancode scancode) {
-    switch (keycode) {
-        case SDLK_TAB:
-            return ImGuiKey_Tab;
-        case SDLK_LEFT:
-            return ImGuiKey_LeftArrow;
-        case SDLK_RIGHT:
-            return ImGuiKey_RightArrow;
-        case SDLK_UP:
-            return ImGuiKey_UpArrow;
-        case SDLK_DOWN:
-            return ImGuiKey_DownArrow;
-        case SDLK_PAGEUP:
-            return ImGuiKey_PageUp;
-        case SDLK_PAGEDOWN:
-            return ImGuiKey_PageDown;
-        case SDLK_HOME:
-            return ImGuiKey_Home;
-        case SDLK_END:
-            return ImGuiKey_End;
-        case SDLK_INSERT:
-            return ImGuiKey_Insert;
-        case SDLK_DELETE:
-            return ImGuiKey_Delete;
-        case SDLK_BACKSPACE:
-            return ImGuiKey_Backspace;
-        case SDLK_SPACE:
-            return ImGuiKey_Space;
-        case SDLK_RETURN:
-            return ImGuiKey_Enter;
-        case SDLK_ESCAPE:
-            return ImGuiKey_Escape;
-        // case SDLK_QUOTE: return ImGuiKey_Apostrophe;
-        case SDLK_COMMA:
-            return ImGuiKey_Comma;
-        // case SDLK_MINUS: return ImGuiKey_Minus;
-        case SDLK_PERIOD:
-            return ImGuiKey_Period;
-        // case SDLK_SLASH: return ImGuiKey_Slash;
-        case SDLK_SEMICOLON:
-            return ImGuiKey_Semicolon;
-        // case SDLK_EQUALS: return ImGuiKey_Equal;
-        // case SDLK_LEFTBRACKET: return ImGuiKey_LeftBracket;
-        // case SDLK_BACKSLASH: return ImGuiKey_Backslash;
-        // case SDLK_RIGHTBRACKET: return ImGuiKey_RightBracket;
-        // case SDLK_BACKQUOTE: return ImGuiKey_GraveAccent;
-        case SDLK_CAPSLOCK:
-            return ImGuiKey_CapsLock;
-        case SDLK_SCROLLLOCK:
-            return ImGuiKey_ScrollLock;
-        case SDLK_NUMLOCKCLEAR:
-            return ImGuiKey_NumLock;
-        case SDLK_PRINTSCREEN:
-            return ImGuiKey_PrintScreen;
-        case SDLK_PAUSE:
-            return ImGuiKey_Pause;
-        case SDLK_KP_0:
-            return ImGuiKey_Keypad0;
-        case SDLK_KP_1:
-            return ImGuiKey_Keypad1;
-        case SDLK_KP_2:
-            return ImGuiKey_Keypad2;
-        case SDLK_KP_3:
-            return ImGuiKey_Keypad3;
-        case SDLK_KP_4:
-            return ImGuiKey_Keypad4;
-        case SDLK_KP_5:
-            return ImGuiKey_Keypad5;
-        case SDLK_KP_6:
-            return ImGuiKey_Keypad6;
-        case SDLK_KP_7:
-            return ImGuiKey_Keypad7;
-        case SDLK_KP_8:
-            return ImGuiKey_Keypad8;
-        case SDLK_KP_9:
-            return ImGuiKey_Keypad9;
-        case SDLK_KP_PERIOD:
-            return ImGuiKey_KeypadDecimal;
-        case SDLK_KP_DIVIDE:
-            return ImGuiKey_KeypadDivide;
-        case SDLK_KP_MULTIPLY:
-            return ImGuiKey_KeypadMultiply;
-        case SDLK_KP_MINUS:
-            return ImGuiKey_KeypadSubtract;
-        case SDLK_KP_PLUS:
-            return ImGuiKey_KeypadAdd;
-        case SDLK_KP_ENTER:
-            return ImGuiKey_KeypadEnter;
-        case SDLK_KP_EQUALS:
-            return ImGuiKey_KeypadEqual;
-        case SDLK_LCTRL:
-            return ImGuiKey_LeftCtrl;
-        case SDLK_LSHIFT:
-            return ImGuiKey_LeftShift;
-        case SDLK_LALT:
-            return ImGuiKey_LeftAlt;
-        case SDLK_LGUI:
-            return ImGuiKey_LeftSuper;
-        case SDLK_RCTRL:
-            return ImGuiKey_RightCtrl;
-        case SDLK_RSHIFT:
-            return ImGuiKey_RightShift;
-        case SDLK_RALT:
-            return ImGuiKey_RightAlt;
-        case SDLK_RGUI:
-            return ImGuiKey_RightSuper;
-        case SDLK_APPLICATION:
-            return ImGuiKey_Menu;
-        case SDLK_0:
-            return ImGuiKey_0;
-        case SDLK_1:
-            return ImGuiKey_1;
-        case SDLK_2:
-            return ImGuiKey_2;
-        case SDLK_3:
-            return ImGuiKey_3;
-        case SDLK_4:
-            return ImGuiKey_4;
-        case SDLK_5:
-            return ImGuiKey_5;
-        case SDLK_6:
-            return ImGuiKey_6;
-        case SDLK_7:
-            return ImGuiKey_7;
-        case SDLK_8:
-            return ImGuiKey_8;
-        case SDLK_9:
-            return ImGuiKey_9;
-        case SDLK_a:
-            return ImGuiKey_A;
-        case SDLK_b:
-            return ImGuiKey_B;
-        case SDLK_c:
-            return ImGuiKey_C;
-        case SDLK_d:
-            return ImGuiKey_D;
-        case SDLK_e:
-            return ImGuiKey_E;
-        case SDLK_f:
-            return ImGuiKey_F;
-        case SDLK_g:
-            return ImGuiKey_G;
-        case SDLK_h:
-            return ImGuiKey_H;
-        case SDLK_i:
-            return ImGuiKey_I;
-        case SDLK_j:
-            return ImGuiKey_J;
-        case SDLK_k:
-            return ImGuiKey_K;
-        case SDLK_l:
-            return ImGuiKey_L;
-        case SDLK_m:
-            return ImGuiKey_M;
-        case SDLK_n:
-            return ImGuiKey_N;
-        case SDLK_o:
-            return ImGuiKey_O;
-        case SDLK_p:
-            return ImGuiKey_P;
-        case SDLK_q:
-            return ImGuiKey_Q;
-        case SDLK_r:
-            return ImGuiKey_R;
-        case SDLK_s:
-            return ImGuiKey_S;
-        case SDLK_t:
-            return ImGuiKey_T;
-        case SDLK_u:
-            return ImGuiKey_U;
-        case SDLK_v:
-            return ImGuiKey_V;
-        case SDLK_w:
-            return ImGuiKey_W;
-        case SDLK_x:
-            return ImGuiKey_X;
-        case SDLK_y:
-            return ImGuiKey_Y;
-        case SDLK_z:
-            return ImGuiKey_Z;
-        case SDLK_F1:
-            return ImGuiKey_F1;
-        case SDLK_F2:
-            return ImGuiKey_F2;
-        case SDLK_F3:
-            return ImGuiKey_F3;
-        case SDLK_F4:
-            return ImGuiKey_F4;
-        case SDLK_F5:
-            return ImGuiKey_F5;
-        case SDLK_F6:
-            return ImGuiKey_F6;
-        case SDLK_F7:
-            return ImGuiKey_F7;
-        case SDLK_F8:
-            return ImGuiKey_F8;
-        case SDLK_F9:
-            return ImGuiKey_F9;
-        case SDLK_F10:
-            return ImGuiKey_F10;
-        case SDLK_F11:
-            return ImGuiKey_F11;
-        case SDLK_F12:
-            return ImGuiKey_F12;
-        case SDLK_F13:
-            return ImGuiKey_F13;
-        case SDLK_F14:
-            return ImGuiKey_F14;
-        case SDLK_F15:
-            return ImGuiKey_F15;
-        case SDLK_F16:
-            return ImGuiKey_F16;
-        case SDLK_F17:
-            return ImGuiKey_F17;
-        case SDLK_F18:
-            return ImGuiKey_F18;
-        case SDLK_F19:
-            return ImGuiKey_F19;
-        case SDLK_F20:
-            return ImGuiKey_F20;
-        case SDLK_F21:
-            return ImGuiKey_F21;
-        case SDLK_F22:
-            return ImGuiKey_F22;
-        case SDLK_F23:
-            return ImGuiKey_F23;
-        case SDLK_F24:
-            return ImGuiKey_F24;
-        case SDLK_AC_BACK:
-            return ImGuiKey_AppBack;
-        case SDLK_AC_FORWARD:
-            return ImGuiKey_AppForward;
-        default:
-            break;
-    }
-
-    // Fallback to scancode
-    switch (scancode) {
-        case SDL_SCANCODE_GRAVE:
-            return ImGuiKey_GraveAccent;
-        case SDL_SCANCODE_MINUS:
-            return ImGuiKey_Minus;
-        case SDL_SCANCODE_EQUALS:
-            return ImGuiKey_Equal;
-        case SDL_SCANCODE_LEFTBRACKET:
-            return ImGuiKey_LeftBracket;
-        case SDL_SCANCODE_RIGHTBRACKET:
-            return ImGuiKey_RightBracket;
-        case SDL_SCANCODE_NONUSBACKSLASH:
-            return ImGuiKey_Oem102;
-        case SDL_SCANCODE_BACKSLASH:
-            return ImGuiKey_Backslash;
-        case SDL_SCANCODE_SEMICOLON:
-            return ImGuiKey_Semicolon;
-        case SDL_SCANCODE_APOSTROPHE:
-            return ImGuiKey_Apostrophe;
-        case SDL_SCANCODE_COMMA:
-            return ImGuiKey_Comma;
-        case SDL_SCANCODE_PERIOD:
-            return ImGuiKey_Period;
-        case SDL_SCANCODE_SLASH:
-            return ImGuiKey_Slash;
-        default:
-            break;
-    }
-    return ImGuiKey_None;
-}
-
-} // namespace filamentapp_utils
-
-#endif // TNT_FILAMENT_FILAMENTAPP_KEYINPUTCONVERSION_H

libs/utils/include/utils/Invocable.h

@@ -65,7 +65,6 @@ public:
     // Creates an Invocable that does not contain a functor.
     // Will evaluate to false.
     Invocable() = default;
-    Invocable(std::nullptr_t) noexcept {}
 
     ~Invocable() noexcept;
 
@@ -78,7 +77,6 @@ public:
 
     Invocable& operator=(const Invocable&) = delete;
     Invocable& operator=(Invocable&& rhs) noexcept;
-    Invocable& operator=(std::nullptr_t) noexcept;
 
     // Invokes the invocable with the args passed in.
     // If the Invocable is empty, this will assert.
@@ -143,17 +141,6 @@ Invocable<R(Args...)>& Invocable<R(Args...)>::operator=(Invocable&& rhs) noexcep
     return *this;
 }
 
-template<typename R, typename... Args>
-Invocable<R(Args...)>& Invocable<R(Args...)>::operator=(std::nullptr_t) noexcept {
-    if (mDeleter) {
-        mDeleter(mInvocable);
-    }
-    mInvocable = nullptr;
-    mDeleter = nullptr;
-    mInvoker = nullptr;
-    return *this;
-}
-
 template<typename R, typename... Args>
 template<typename... OperatorArgs>
 R Invocable<R(Args...)>::operator()(OperatorArgs&& ... args) {

libs/utils/include/utils/MonotonicRingMap.h

@@ -56,9 +56,6 @@ public:
     //! Returns true if the map is full.
     bool full() const noexcept { return mSize == N; }
 
-    //! Clears the map entirely.
-    void clear() noexcept { mSize = 0; mHead = 0; }
-
     /**
      * Inserts a new key-value pair.
      * The key must be greater than the key of the last inserted element.
@@ -68,7 +65,7 @@ public:
      */
     UTILS_NOINLINE void insert(key_type key, mapped_type value) {
         assert(empty() || key > back().first); // assert monotonic
-        if (UTILS_LIKELY(full())) {
+        if (full()) {
             // container is full, replace the oldest element
             mStorage[mHead] = { key, value };
             mHead = (mHead + 1) % N;

libs/utils/src/AsyncJobQueue.cpp

@@ -55,10 +55,7 @@ AsyncJobQueue::AsyncJobQueue(const char* name, Priority priority) {
 }
 
 AsyncJobQueue::~AsyncJobQueue() noexcept {
-    // wait for all pending callbacks to be called & terminate the thread
-    drainAndExit();
 #if !defined(__EMSCRIPTEN__)
-    assert_invariant(!mThread.joinable());
     assert_invariant(mQueue.empty());
 #endif
 }
@@ -97,9 +94,6 @@ bool AsyncJobQueue::isValid() const noexcept {
 void AsyncJobQueue::drainAndExit() {
 #if !defined(__EMSCRIPTEN__)
     std::unique_lock lock(mLock);
-    if (mExitRequested) {
-        return;
-    }
     // we request the service thread to exit, but we're guaranteed that it'll only exit
     // after all current callbacks are processed. In addition, once mExitRequested is set,
     // no new jobs can be added, so we can join the thread.

libs/viewer/src/ViewerGui.cpp

@@ -41,36 +41,6 @@ using namespace filament::math;
 namespace filament {
 namespace viewer {
 
-namespace {
-
-ImGuiKey convertKey(int keycode) {
-
-    // The following table uses normal ANSI codes, which is consistent with the keyCode that
-    // comes from a web "keydown" event.
-    switch (keycode) {
-        case 9: return ImGuiKey_Tab;
-        case 37: return ImGuiKey_LeftArrow;
-        case 39: return ImGuiKey_RightArrow;
-        case 38: return ImGuiKey_UpArrow;
-        case 40: return ImGuiKey_DownArrow;
-        case 36: return ImGuiKey_Home;
-        case 35: return ImGuiKey_End;
-        case 46: return ImGuiKey_Delete;
-        case 8: return ImGuiKey_Backspace;
-        case 13: return ImGuiKey_Enter;
-        case 27: return ImGuiKey_Escape;
-        case 65: return ImGuiKey_A;
-        case 67: return ImGuiKey_C;
-        case 86: return ImGuiKey_V;
-        case 88: return ImGuiKey_X;
-        case 89: return ImGuiKey_Y;
-        case 90: return ImGuiKey_Z;
-        default: return ImGuiKey_None;
-    }
-}
-
-}
-
 mat4f fitIntoUnitCube(const Aabb& bounds, float zoffset) {
     float3 minpt = bounds.min;
     float3 maxpt = bounds.max;
@@ -593,6 +563,27 @@ void ViewerGui::renderUserInterface(float timeStepInSeconds, View* guiView, floa
         mImGuiHelper = new ImGuiHelper(mEngine, guiView, "");
 
         auto& io = ImGui::GetIO();
+
+        // The following table uses normal ANSI codes, which is consistent with the keyCode that
+        // comes from a web "keydown" event.
+        io.KeyMap[ImGuiKey_Tab] = 9;
+        io.KeyMap[ImGuiKey_LeftArrow] = 37;
+        io.KeyMap[ImGuiKey_RightArrow] = 39;
+        io.KeyMap[ImGuiKey_UpArrow] = 38;
+        io.KeyMap[ImGuiKey_DownArrow] = 40;
+        io.KeyMap[ImGuiKey_Home] = 36;
+        io.KeyMap[ImGuiKey_End] = 35;
+        io.KeyMap[ImGuiKey_Delete] = 46;
+        io.KeyMap[ImGuiKey_Backspace] = 8;
+        io.KeyMap[ImGuiKey_Enter] = 13;
+        io.KeyMap[ImGuiKey_Escape] = 27;
+        io.KeyMap[ImGuiKey_A] = 65;
+        io.KeyMap[ImGuiKey_C] = 67;
+        io.KeyMap[ImGuiKey_V] = 86;
+        io.KeyMap[ImGuiKey_X] = 88;
+        io.KeyMap[ImGuiKey_Y] = 89;
+        io.KeyMap[ImGuiKey_Z] = 90;
+
         // TODO: this is not the best way to handle high DPI in ImGui, but it is fine when using the
         // proggy font. Users need to refresh their window when dragging between displays with
         // different pixel ratios.
@@ -618,28 +609,20 @@ void ViewerGui::mouseEvent(float mouseX, float mouseY, bool mouseButton, float m
         io.MouseDown[0] = mouseButton != 0;
         io.MouseDown[1] = false;
         io.MouseDown[2] = false;
-        io.AddKeyEvent(ImGuiMod_Ctrl, control);
+        io.KeyCtrl = control;
     }
 }
 
 void ViewerGui::keyDownEvent(int keyCode) {
-    if (!mImGuiHelper) {
-        return;
+    if (mImGuiHelper && keyCode < IM_ARRAYSIZE(ImGui::GetIO().KeysDown)) {
+        ImGui::GetIO().KeysDown[keyCode] = true;
     }
-    ImGuiIO& io = ImGui::GetIO();
-    io.AddKeyEvent(
-            convertKey(keyCode),
-            true);
 }
 
 void ViewerGui::keyUpEvent(int keyCode) {
-    if (!mImGuiHelper) {
-        return;
+    if (mImGuiHelper && keyCode < IM_ARRAYSIZE(ImGui::GetIO().KeysDown)) {
+        ImGui::GetIO().KeysDown[keyCode] = false;
     }
-    ImGuiIO& io = ImGui::GetIO();
-    io.AddKeyEvent(
-            convertKey(keyCode),
-            false);
 }
 
 void ViewerGui::keyPressEvent(int charCode) {

shaders/src/surface_shading_lit.fs

@@ -302,16 +302,14 @@ vec4 evaluateLights(const MaterialInputs material) {
     return vec4(color, computeDiffuseAlpha(material.baseColor.a));
 }
 
-vec3 emissive(const MaterialInputs material, float alpha) {
+void addEmissive(const MaterialInputs material, inout vec4 color) {
 #if defined(MATERIAL_HAS_EMISSIVE)
     highp vec4 emissive = material.emissive;
     highp float attenuation = mix(1.0, getExposure(), emissive.w);
 #if defined(BLEND_MODE_TRANSPARENT) || defined(BLEND_MODE_FADE)
-    attenuation *= alpha;
+    attenuation *= color.a;
 #endif
-    return emissive.rgb * attenuation;
-#else
-    return vec3(0.0);
+    color.rgb += emissive.rgb * attenuation;
 #endif
 }
 
@@ -323,7 +321,7 @@ vec3 emissive(const MaterialInputs material, float alpha) {
  */
 vec4 evaluateMaterial(const MaterialInputs material) {
     vec4 color = evaluateLights(material);
-    color.rgb += emissive(material, color.a);
+    addEmissive(material, color);
     // kill +inf generated by strong highlights
     color.rgb = min(color.rgb, MEDIUMP_FLT_MAX);
     return color;

test/renderdiff/src/viewer_requirements.txt

@@ -10,4 +10,4 @@ MarkupSafe==3.0.2
 requests==2.32.5
 urllib3==2.5.0
 waitress==3.0.2
-Werkzeug==3.1.4
+Werkzeug==3.1.3

third_party/benchmark/.clang-format

@@ -1,5 +1,5 @@
 ---
 Language:        Cpp
 BasedOnStyle:  Google
-PointerAlignment: Left
 ...
+

third_party/benchmark/.clang-tidy

@@ -1,37 +0,0 @@
----
-Checks: >
-  abseil-*,
-  bugprone-*,
-  clang-analyzer-*,
-  cppcoreguidelines-*,
-  google-*,
-  misc-*,
-  performance-*,
-  readability-*,
-  -clang-analyzer-deadcode*,
-  -clang-analyzer-optin*,
-  -readability-identifier-length
-WarningsAsErrors: ''
-HeaderFilterRegex: ''
-FormatStyle:     none
-CheckOptions:
-  llvm-else-after-return.WarnOnConditionVariables: 'false'
-  modernize-loop-convert.MinConfidence: reasonable
-  modernize-replace-auto-ptr.IncludeStyle: llvm
-  cert-str34-c.DiagnoseSignedUnsignedCharComparisons: 'false'
-  google-readability-namespace-comments.ShortNamespaceLines: '10'
-  cert-err33-c.CheckedFunctions: '::aligned_alloc;::asctime_s;::at_quick_exit;::atexit;::bsearch;::bsearch_s;::btowc;::c16rtomb;::c32rtomb;::calloc;::clock;::cnd_broadcast;::cnd_init;::cnd_signal;::cnd_timedwait;::cnd_wait;::ctime_s;::fclose;::fflush;::fgetc;::fgetpos;::fgets;::fgetwc;::fopen;::fopen_s;::fprintf;::fprintf_s;::fputc;::fputs;::fputwc;::fputws;::fread;::freopen;::freopen_s;::fscanf;::fscanf_s;::fseek;::fsetpos;::ftell;::fwprintf;::fwprintf_s;::fwrite;::fwscanf;::fwscanf_s;::getc;::getchar;::getenv;::getenv_s;::gets_s;::getwc;::getwchar;::gmtime;::gmtime_s;::localtime;::localtime_s;::malloc;::mbrtoc16;::mbrtoc32;::mbsrtowcs;::mbsrtowcs_s;::mbstowcs;::mbstowcs_s;::memchr;::mktime;::mtx_init;::mtx_lock;::mtx_timedlock;::mtx_trylock;::mtx_unlock;::printf_s;::putc;::putwc;::raise;::realloc;::remove;::rename;::scanf;::scanf_s;::setlocale;::setvbuf;::signal;::snprintf;::snprintf_s;::sprintf;::sprintf_s;::sscanf;::sscanf_s;::strchr;::strerror_s;::strftime;::strpbrk;::strrchr;::strstr;::strtod;::strtof;::strtoimax;::strtok;::strtok_s;::strtol;::strtold;::strtoll;::strtoul;::strtoull;::strtoumax;::strxfrm;::swprintf;::swprintf_s;::swscanf;::swscanf_s;::thrd_create;::thrd_detach;::thrd_join;::thrd_sleep;::time;::timespec_get;::tmpfile;::tmpfile_s;::tmpnam;::tmpnam_s;::tss_create;::tss_get;::tss_set;::ungetc;::ungetwc;::vfprintf;::vfprintf_s;::vfscanf;::vfscanf_s;::vfwprintf;::vfwprintf_s;::vfwscanf;::vfwscanf_s;::vprintf_s;::vscanf;::vscanf_s;::vsnprintf;::vsnprintf_s;::vsprintf;::vsprintf_s;::vsscanf;::vsscanf_s;::vswprintf;::vswprintf_s;::vswscanf;::vswscanf_s;::vwprintf_s;::vwscanf;::vwscanf_s;::wcrtomb;::wcschr;::wcsftime;::wcspbrk;::wcsrchr;::wcsrtombs;::wcsrtombs_s;::wcsstr;::wcstod;::wcstof;::wcstoimax;::wcstok;::wcstok_s;::wcstol;::wcstold;::wcstoll;::wcstombs;::wcstombs_s;::wcstoul;::wcstoull;::wcstoumax;::wcsxfrm;::wctob;::wctrans;::wctype;::wmemchr;::wprintf_s;::wscanf;::wscanf_s;'
-  cert-oop54-cpp.WarnOnlyIfThisHasSuspiciousField: 'false'
-  cert-dcl16-c.NewSuffixes: 'L;LL;LU;LLU'
-  google-readability-braces-around-statements.ShortStatementLines: '1'
-  cppcoreguidelines-non-private-member-variables-in-classes.IgnoreClassesWithAllMemberVariablesBeingPublic: 'true'
-  google-readability-namespace-comments.SpacesBeforeComments: '2'
-  modernize-loop-convert.MaxCopySize: '16'
-  modernize-pass-by-value.IncludeStyle: llvm
-  modernize-use-nullptr.NullMacros: 'NULL'
-  llvm-qualified-auto.AddConstToQualified: 'false'
-  modernize-loop-convert.NamingStyle: CamelCase
-  llvm-else-after-return.WarnOnUnfixable: 'false'
-  google-readability-function-size.StatementThreshold: '800'
-...
-

third_party/benchmark/.clang-tidy.ignore

@@ -1 +0,0 @@
-.*third_party/.*

third_party/benchmark/.github/ISSUE_TEMPLATE/bug_report.md

@@ -1,32 +0,0 @@
----
-name: Bug report
-about: Create a report to help us improve
-title: "[BUG]"
-labels: ''
-assignees: ''
-
----
-
-**Describe the bug**
-A clear and concise description of what the bug is.
-
-**System**
-Which OS, compiler, and compiler version are you using:
-  - OS: 
-  - Compiler and version: 
-
-**To reproduce**
-Steps to reproduce the behavior:
-1. sync to commit ...
-2. cmake/bazel...
-3. make ...
-4. See error
-
-**Expected behavior**
-A clear and concise description of what you expected to happen.
-
-**Screenshots**
-If applicable, add screenshots to help explain your problem.
-
-**Additional context**
-Add any other context about the problem here.

third_party/benchmark/.github/ISSUE_TEMPLATE/feature_request.md

@@ -1,20 +0,0 @@
----
-name: Feature request
-about: Suggest an idea for this project
-title: "[FR]"
-labels: ''
-assignees: ''
-
----
-
-**Is your feature request related to a problem? Please describe.**
-A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]
-
-**Describe the solution you'd like**
-A clear and concise description of what you want to happen.
-
-**Describe alternatives you've considered**
-A clear and concise description of any alternative solutions or features you've considered.
-
-**Additional context**
-Add any other context or screenshots about the feature request here.

third_party/benchmark/.github/install_bazel.sh

@@ -1,12 +0,0 @@
-if ! bazel version; then
-  arch=$(uname -m)
-  if [ "$arch" == "aarch64" ]; then
-    arch="arm64"
-  fi
-  echo "Downloading $arch Bazel binary from GitHub releases."
-  curl -L -o $HOME/bin/bazel --create-dirs "https://github.com/bazelbuild/bazel/releases/download/8.2.0/bazel-8.2.0-linux-$arch"
-  chmod +x $HOME/bin/bazel
-else
-  # Bazel is installed for the correct architecture
-  exit 0
-fi

third_party/benchmark/.github/libcxx-setup.sh

@@ -1,35 +0,0 @@
-#!/usr/bin/env bash
-
-set -e
-
-# Checkout LLVM sources
-git clone --filter=blob:none --depth=1 --branch llvmorg-19.1.6 --no-checkout https://github.com/llvm/llvm-project.git llvm-project
-cd llvm-project
-git sparse-checkout set --cone
-git checkout llvmorg-19.1.6
-git sparse-checkout set cmake llvm/cmake runtimes libcxx libcxxabi
-cd ..
-
-## Setup libc++ options
-if [ -z "$BUILD_32_BITS" ]; then
-  export BUILD_32_BITS=OFF && echo disabling 32 bit build
-fi
-
-## Build and install libc++ (Use unstable ABI for better sanitizer coverage)
-mkdir llvm-build && cd llvm-build
-cmake -GNinja                                   \
-      -DCMAKE_C_COMPILER=${CC}                  \
-      -DCMAKE_CXX_COMPILER=${CXX}               \
-      -DCMAKE_BUILD_TYPE=RelWithDebInfo         \
-      -DCMAKE_INSTALL_PREFIX=/usr               \
-      -DLIBCXX_ABI_UNSTABLE=OFF                 \
-      -DLLVM_USE_SANITIZER=${LIBCXX_SANITIZER}  \
-      -DLLVM_BUILD_32_BITS=${BUILD_32_BITS}     \
-      -DLIBCXXABI_USE_LLVM_UNWINDER=OFF         \
-      -DLLVM_INCLUDE_TESTS=OFF                  \
-      -DLIBCXX_INCLUDE_TESTS=OFF                \
-      -DLIBCXX_INCLUDE_BENCHMARKS=OFF           \
-      -DLLVM_ENABLE_RUNTIMES='libcxx;libcxxabi' \
-      ../llvm-project/runtimes/
-cmake --build . -- cxx cxxabi
-cd ..

third_party/benchmark/.github/workflows/bazel.yml

@@ -1,37 +0,0 @@
-name: bazel
-
-on:
-  push: {}
-  pull_request: {}
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  build_and_test_default:
-    name: bazel.${{ matrix.os }}
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
-    steps:
-    - uses: actions/checkout@v4
-
-    - name: mount bazel cache
-      uses: actions/cache@v4
-      env:
-        cache-name: bazel-cache
-      with:
-        path: "~/.cache/bazel"
-        key: ${{ env.cache-name }}-${{ matrix.os }}-${{ github.ref }}
-        restore-keys: |
-          ${{ env.cache-name }}-${{ matrix.os }}-main
-
-    - name: build
-      run: |
-        bazel build //:benchmark //:benchmark_main //test/...
-
-    - name: test
-      run: |
-        bazel test --test_output=all //test/...

third_party/benchmark/.github/workflows/build-and-test-min-cmake.yml

@@ -1,49 +0,0 @@
-name: build-and-test-min-cmake
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  job:
-    name: ${{ matrix.os }}.min-cmake
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        os: [ubuntu-latest, macos-latest]
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: lukka/get-cmake@latest
-        with:
-          cmakeVersion: 3.13.0
-
-      - name: create build environment
-        run: cmake -E make_directory ${{ runner.workspace }}/_build
-
-      - name: setup cmake initial cache
-        run: touch compiler-cache.cmake
-
-      - name: configure cmake
-        env:
-          CXX: ${{ matrix.compiler }}
-        shell: bash
-        working-directory: ${{ runner.workspace }}/_build
-        run: >
-          cmake -C ${{ github.workspace }}/compiler-cache.cmake
-          $GITHUB_WORKSPACE
-          -DBENCHMARK_DOWNLOAD_DEPENDENCIES=ON
-          -DCMAKE_CXX_VISIBILITY_PRESET=hidden
-          -DCMAKE_VISIBILITY_INLINES_HIDDEN=ON
-
-      - name: build
-        shell: bash
-        working-directory: ${{ runner.workspace }}/_build
-        run: cmake --build .

third_party/benchmark/.github/workflows/build-and-test-perfcounters.yml

@@ -1,54 +0,0 @@
-name: build-and-test-perfcounters
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  job:
-    # TODO(dominic): Extend this to include compiler and set through env: CC/CXX.
-    name: ${{ matrix.os }}.${{ matrix.build_type }}
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        os: [ubuntu-latest]
-        build_type: ['Release', 'Debug']
-    steps:
-    - uses: actions/checkout@v4
-
-    - name: install libpfm
-      run: |
-        sudo apt update
-        sudo apt -y install libpfm4-dev
-
-    - name: create build environment
-      run: cmake -E make_directory ${{ runner.workspace }}/_build
-
-    - name: configure cmake
-      shell: bash
-      working-directory: ${{ runner.workspace }}/_build
-      run: >
-        cmake $GITHUB_WORKSPACE
-        -DBENCHMARK_ENABLE_LIBPFM=1
-        -DBENCHMARK_DOWNLOAD_DEPENDENCIES=ON
-        -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
-
-    - name: build
-      shell: bash
-      working-directory: ${{ runner.workspace }}/_build
-      run: cmake --build . --config ${{ matrix.build_type }}
-
-    # Skip testing, for now. It seems perf_event_open does not succeed on the
-    # hosting machine, very likely a permissions issue.
-    # TODO(mtrofin): Enable test.
-    # - name: test
-    #   shell: bash
-    #   working-directory: ${{ runner.workspace }}/_build
-    #   run: ctest -C ${{ matrix.build_type }} --rerun-failed --output-on-failure
-

third_party/benchmark/.github/workflows/build-and-test.yml

@@ -1,151 +0,0 @@
-name: build-and-test
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  # TODO: add 32-bit builds (g++ and clang++) for ubuntu
-  #   (requires g++-multilib and libc6:i386)
-  # TODO: add coverage build (requires lcov)
-  # TODO: add clang + libc++ builds for ubuntu
-  job:
-    name: ${{ matrix.os }}.${{ matrix.build_type }}.${{ matrix.lib }}.${{ matrix.compiler }}
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        os: [ubuntu-24.04, ubuntu-22.04, ubuntu-24.04-arm, macos-latest]
-        build_type: ['Release', 'Debug']
-        compiler: ['g++', 'clang++']
-        lib: ['shared', 'static']
-
-    steps:
-      - name: Install dependencies (macos)
-        if: runner.os == 'macOS'
-        run: brew install ninja
-
-      - uses: actions/checkout@v4
-
-      - name: build
-        uses: threeal/cmake-action@v2.1.0
-        with:
-          build-dir: ${{ runner.workspace }}/_build
-          cxx-compiler: ${{ matrix.compiler }}
-          options: |
-            BENCHMARK_DOWNLOAD_DEPENDENCIES=ON
-            BUILD_SHARED_LIBS=${{ matrix.lib == 'shared' }}
-            CMAKE_BUILD_TYPE=${{ matrix.build_type }}
-            CMAKE_CXX_COMPILER=${{ matrix.compiler }}
-            CMAKE_CXX_VISIBILITY_PRESET=hidden
-            CMAKE_VISIBILITY_INLINES_HIDDEN=ON
-
-      - name: test
-        shell: bash
-        working-directory: ${{ runner.workspace }}/_build
-        run: ctest -C ${{ matrix.build_type }} -VV
-
-  msvc:
-    name: ${{ matrix.os }}.${{ matrix.build_type }}.${{ matrix.lib }}.${{ matrix.msvc }}
-    runs-on: ${{ matrix.os }}
-    defaults:
-        run:
-            shell: powershell
-    strategy:
-      fail-fast: false
-      matrix:
-        msvc:
-          - VS-16-2019
-          - VS-17-2022
-        build_type:
-          - Debug
-          - Release
-        lib:
-          - shared
-          - static
-        include:
-          - msvc: VS-16-2019
-            os: windows-2019
-            generator: 'Visual Studio 16 2019'
-          - msvc: VS-17-2022
-            os: windows-2022
-            generator: 'Visual Studio 17 2022'
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: lukka/get-cmake@latest
-
-      - name: configure cmake
-        run: >
-          cmake -S . -B ${{ runner.workspace }}/_build/
-          -G "${{ matrix.generator }}"
-          -DBENCHMARK_DOWNLOAD_DEPENDENCIES=ON
-          -DBUILD_SHARED_LIBS=${{ matrix.lib == 'shared' }}
-
-      - name: build
-        run: cmake --build ${{ runner.workspace }}/_build/ --config ${{ matrix.build_type }}
-
-      - name: test
-        run: ctest --test-dir ${{ runner.workspace }}/_build/ -C ${{ matrix.build_type }} -VV
-
-  msys2:
-    name: ${{ matrix.os }}.${{ matrix.build_type }}.${{ matrix.lib }}.${{ matrix.msys2.msystem }}
-    runs-on: ${{ matrix.os }}
-    defaults:
-        run:
-            shell: msys2 {0}
-    strategy:
-      fail-fast: false
-      matrix:
-        os: [ windows-latest ]
-        msys2:
-          - { msystem: MINGW64,    arch: x86_64,  family: GNU,  compiler: g++ }
-          - { msystem: CLANG64,    arch: x86_64,  family: LLVM, compiler: clang++ }
-          - { msystem: UCRT64,     arch: x86_64,  family: GNU,  compiler: g++ }
-        build_type:
-          - Debug
-          - Release
-        lib:
-          - shared
-          - static
-
-    steps:
-      - name: setup msys2
-        uses: msys2/setup-msys2@v2
-        with:
-          cache: false
-          msystem: ${{ matrix.msys2.msystem }}
-          update: true
-          install: >-
-            git
-            base-devel
-          pacboy: >-
-            gcc:p
-            clang:p
-            cmake:p
-            ninja:p
-
-      - uses: actions/checkout@v4
-
-      # NOTE: we can't use cmake actions here as we need to do everything in msys2 shell.
-      - name: configure cmake
-        env:
-          CXX: ${{ matrix.msys2.compiler }}
-        run: >
-          cmake -S . -B _build/
-          -GNinja
-          -DBENCHMARK_DOWNLOAD_DEPENDENCIES=ON
-          -DBUILD_SHARED_LIBS=${{ matrix.lib == 'shared' }}
-
-      - name: build
-        run: cmake --build _build/ --config ${{ matrix.build_type }}
-
-      - name: test
-        working-directory: _build
-        run: ctest -C ${{ matrix.build_type }} -VV

third_party/benchmark/.github/workflows/clang-format-lint.yml

@@ -1,19 +0,0 @@
-name: clang-format-lint
-on:
-  push: {}
-  pull_request: {}
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  job:
-    name: check-clang-format
-    runs-on: ubuntu-latest
-
-    steps:
-    - uses: actions/checkout@v4
-    - uses: DoozyX/clang-format-lint-action@v0.18.2
-      with:
-        source: './include/benchmark ./src ./test'
-        clangFormatVersion: 18

third_party/benchmark/.github/workflows/clang-tidy-lint.yml

@@ -1,41 +0,0 @@
-name: clang-tidy
-
-on:
-  push: {}
-  pull_request: {}
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  job:
-    name: run-clang-tidy
-    runs-on: ubuntu-latest
-    strategy:
-      fail-fast: false
-    steps:
-    - uses: actions/checkout@v4
-
-    - name: install clang-tidy
-      run: sudo apt update && sudo apt -y install clang-tidy
-
-    - name: create build environment
-      run: cmake -E make_directory ${{ github.workspace }}/_build
-
-    - name: configure cmake
-      shell: bash
-      working-directory: ${{ github.workspace }}/_build
-      run: >
-        cmake $GITHUB_WORKSPACE
-        -DBENCHMARK_ENABLE_ASSEMBLY_TESTS=OFF
-        -DBENCHMARK_ENABLE_LIBPFM=OFF
-        -DBENCHMARK_DOWNLOAD_DEPENDENCIES=ON
-        -DCMAKE_C_COMPILER=clang
-        -DCMAKE_CXX_COMPILER=clang++
-        -DCMAKE_EXPORT_COMPILE_COMMANDS=ON
-        -DGTEST_COMPILE_COMMANDS=OFF
-
-    - name: run
-      shell: bash
-      working-directory: ${{ github.workspace }}/_build
-      run: run-clang-tidy -config-file=$GITHUB_WORKSPACE/.clang-tidy

third_party/benchmark/.github/workflows/doxygen.yml

@@ -1,31 +0,0 @@
-name: doxygen
-
-on:
-  push:
-    branches: [main]
-  pull_request:
-    branches: [main]
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  build-and-deploy:
-    name: Build HTML documentation
-    runs-on: ubuntu-latest
-    steps:
-    - name: Fetching sources
-      uses: actions/checkout@v4
-
-    - name: Installing build dependencies
-      run: |
-        sudo apt update
-        sudo apt install doxygen gcc git
-
-    - name: Creating build directory
-      run: mkdir build
-
-    - name: Building HTML documentation with Doxygen
-      run: |
-        cmake -S . -B build -DBENCHMARK_ENABLE_TESTING:BOOL=OFF -DBENCHMARK_ENABLE_DOXYGEN:BOOL=ON -DBENCHMARK_INSTALL_DOCS:BOOL=ON
-        cmake --build build --target benchmark_doxygen

third_party/benchmark/.github/workflows/pre-commit.yml

@@ -1,41 +0,0 @@
-name: python + Bazel pre-commit checks
-
-on:
-  push:
-    branches: [ main ]
-  pull_request:
-    branches: [ main ]
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  pre-commit:
-    runs-on: ubuntu-latest
-    env:
-      MYPY_CACHE_DIR: "${{ github.workspace }}/.cache/mypy"
-      RUFF_CACHE_DIR: "${{ github.workspace }}/.cache/ruff"
-      PRE_COMMIT_HOME: "${{ github.workspace }}/.cache/pre-commit"
-
-    steps:
-    - uses: actions/checkout@v4
-      with:
-        fetch-depth: 0
-    - name: Set up Python
-      uses: actions/setup-python@v5
-      with:
-        python-version: 3.11
-        cache: pip
-        cache-dependency-path: pyproject.toml
-    - name: Install dependencies
-      run: python -m pip install ".[dev]"
-    - name: Cache pre-commit tools
-      uses: actions/cache@v4
-      with:
-        path: |
-          ${{ env.MYPY_CACHE_DIR }}
-          ${{ env.RUFF_CACHE_DIR }}
-          ${{ env.PRE_COMMIT_HOME }}
-        key: ${{ runner.os }}-${{ hashFiles('.pre-commit-config.yaml') }}-linter-cache
-    - name: Run pre-commit checks
-      run: pre-commit run --all-files --verbose --show-diff-on-failure

third_party/benchmark/.github/workflows/sanitizer.yml

@@ -1,97 +0,0 @@
-name: sanitizer
-
-on:
-  push: {}
-  pull_request: {}
-
-env:
-  CMAKE_GENERATOR: Ninja
-  UBSAN_OPTIONS: "print_stacktrace=1"
-
-jobs:
-  job:
-    name: ${{ matrix.sanitizer }}.${{ matrix.build_type }}
-    runs-on: ubuntu-latest
-    strategy:
-      fail-fast: false
-      matrix:
-        build_type: ['Debug', 'RelWithDebInfo']
-        sanitizer: ['asan', 'ubsan', 'tsan', 'msan']
-
-    steps:
-    - uses: actions/checkout@v4
-
-    - name: configure msan env
-      if: matrix.sanitizer == 'msan'
-      run: |
-        echo "EXTRA_FLAGS=-g -O2 -fno-omit-frame-pointer -fsanitize=memory -fsanitize-memory-track-origins" >> $GITHUB_ENV
-        echo "LIBCXX_SANITIZER=MemoryWithOrigins" >> $GITHUB_ENV
-
-    - name: configure ubsan env
-      if: matrix.sanitizer == 'ubsan'
-      run: |
-        echo "EXTRA_FLAGS=-g -O2 -fno-omit-frame-pointer -fsanitize=undefined -fno-sanitize-recover=all" >> $GITHUB_ENV
-        echo "LIBCXX_SANITIZER=Undefined" >> $GITHUB_ENV
-
-    - name: configure asan env
-      if: matrix.sanitizer == 'asan'
-      run: |
-        echo "EXTRA_FLAGS=-g -O2 -fno-omit-frame-pointer -fsanitize=address -fno-sanitize-recover=all" >> $GITHUB_ENV
-        echo "LIBCXX_SANITIZER=Address" >> $GITHUB_ENV
-
-    - name: configure tsan env
-      if: matrix.sanitizer == 'tsan'
-      run: |
-        echo "EXTRA_FLAGS=-g -O2 -fno-omit-frame-pointer -fsanitize=thread -fno-sanitize-recover=all" >> $GITHUB_ENV
-        echo "LIBCXX_SANITIZER=Thread" >> $GITHUB_ENV
-
-    - name: fine-tune asan options
-      # in asan we get an error from std::regex. ignore it.
-      if: matrix.sanitizer == 'asan'
-      run: |
-        echo "ASAN_OPTIONS=alloc_dealloc_mismatch=0" >> $GITHUB_ENV
-
-    - name: setup clang
-      uses: egor-tensin/setup-clang@v1
-      with:
-        version: latest
-        platform: x64
-
-    - name: configure clang
-      run: |
-        echo "CC=cc" >> $GITHUB_ENV
-        echo "CXX=c++" >> $GITHUB_ENV
-
-    - name: build libc++ (non-asan)
-      if: matrix.sanitizer != 'asan'
-      run: |
-        "${GITHUB_WORKSPACE}/.github/libcxx-setup.sh"
-        echo "EXTRA_CXX_FLAGS=-stdlib=libc++ -L${GITHUB_WORKSPACE}/llvm-build/lib -lc++abi -I${GITHUB_WORKSPACE}/llvm-build/include/c++/v1 -Isystem${GITHUB_WORKSPACE}/llvm-build/include/c++/v1 -Wl,-rpath,${GITHUB_WORKSPACE}/llvm-build/lib" >> $GITHUB_ENV
-
-    - name: create build environment
-      run: cmake -E make_directory ${{ runner.workspace }}/_build
-
-    - name: configure cmake
-      shell: bash
-      working-directory: ${{ runner.workspace }}/_build
-      run: >
-        VERBOSE=1
-        cmake -GNinja $GITHUB_WORKSPACE
-        -DBENCHMARK_ENABLE_ASSEMBLY_TESTS=OFF
-        -DBENCHMARK_ENABLE_LIBPFM=OFF
-        -DBENCHMARK_DOWNLOAD_DEPENDENCIES=ON
-        -DCMAKE_C_COMPILER=${{ env.CC }}
-        -DCMAKE_CXX_COMPILER=${{ env.CXX }}
-        -DCMAKE_C_FLAGS="${{ env.EXTRA_FLAGS }}"
-        -DCMAKE_CXX_FLAGS="${{ env.EXTRA_FLAGS }} ${{ env.EXTRA_CXX_FLAGS }}"
-        -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
-
-    - name: build
-      shell: bash
-      working-directory: ${{ runner.workspace }}/_build
-      run: cmake --build . --config ${{ matrix.build_type }}
-
-    - name: test
-      shell: bash
-      working-directory: ${{ runner.workspace }}/_build
-      run: ctest -C ${{ matrix.build_type }} -VV

third_party/benchmark/.github/workflows/test_bindings.yml

@@ -1,33 +0,0 @@
-name: test-bindings
-
-on:
-  push:
-    branches: [main]
-  pull_request:
-    branches: [main]
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  python_bindings:
-    name: Test GBM Python ${{ matrix.python-version }} bindings on ${{ matrix.os }}
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        os: [ ubuntu-latest, macos-latest, windows-latest ]
-        python-version: [ "3.10", "3.11", "3.12", "3.13" ]
-
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-      - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v5
-        with:
-          python-version: ${{ matrix.python-version }}
-      - name: Install GBM Python bindings on ${{ matrix.os }}
-        run: python -m pip install .
-      - name: Run example on ${{ matrix.os }} under Python ${{ matrix.python-version }}
-        run: python bindings/python/google_benchmark/example.py

third_party/benchmark/.github/workflows/wheels.yml

@@ -1,83 +0,0 @@
-name: Build and upload Python wheels
-
-on:
-  workflow_dispatch:
-  release:
-    types:
-      - published
-
-env:
-  CMAKE_GENERATOR: Ninja
-
-jobs:
-  build_sdist:
-    name: Build source distribution
-    runs-on: ubuntu-latest
-    steps:
-      - name: Check out repo
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-      - name: Install Python 3.12
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.12"
-      - run: python -m pip install build
-      - name: Build sdist
-        run: python -m build --sdist
-      - uses: actions/upload-artifact@v4
-        with:
-          name: dist-sdist
-          path: dist/*.tar.gz
-
-  build_wheels:
-    name: Build Google Benchmark wheels on ${{ matrix.os }}
-    runs-on: ${{ matrix.os }}
-    strategy:
-      matrix:
-        os: [ubuntu-latest, ubuntu-24.04-arm, macos-13, macos-14, windows-latest]
-    steps:
-      - name: Check out Google Benchmark
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-
-      - uses: actions/setup-python@v5
-        name: Install Python 3.12
-        with:
-          python-version: "3.12"
-      - run: pip install --upgrade pip uv
-
-      - name: Build wheels on ${{ matrix.os }} using cibuildwheel
-        uses: pypa/cibuildwheel@v2.23.2
-        env:
-          CIBW_BUILD: "cp310-* cp311-* cp312-*"
-          CIBW_BUILD_FRONTEND: "build[uv]"
-          CIBW_SKIP: "*-musllinux_*"
-          CIBW_ARCHS: auto64
-          CIBW_BEFORE_ALL_LINUX: bash .github/install_bazel.sh
-          # Grab the rootless Bazel installation inside the container.
-          CIBW_ENVIRONMENT_LINUX: PATH=$PATH:$HOME/bin
-          CIBW_TEST_COMMAND: python {project}/bindings/python/google_benchmark/example.py
-          # unused by Bazel, but needed explicitly by delocate on MacOS.
-          MACOSX_DEPLOYMENT_TARGET: "10.14"
-
-      - name: Upload Google Benchmark ${{ matrix.os }} wheels
-        uses: actions/upload-artifact@v4
-        with:
-          name: dist-${{ matrix.os }}
-          path: wheelhouse/*.whl
-
-  pypi_upload:
-    name: Publish google-benchmark wheels to PyPI
-    needs: [build_sdist, build_wheels]
-    runs-on: ubuntu-latest
-    permissions:
-      id-token: write
-    steps:
-      - uses: actions/download-artifact@v4
-        with:
-          path: dist
-          pattern: dist-*
-          merge-multiple: true
-      - uses: pypa/gh-action-pypi-publish@release/v1