add logging

2026-03-03 13:25:25 -08:00
103 changed files with 607 additions and 1801 deletions
@@ -31,7 +31,7 @@ repositories {
 }

 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.70.1'
+    implementation 'com.google.android.filament:filament-android:1.69.5'
 }
 ```

@@ -50,7 +50,7 @@ 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', '~> 1.70.1'
+pod 'Filament', '~> 1.69.5'
 ```

 ## Documentation
@@ -7,9 +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).

-## v1.70.1
-
-
 ## v1.70.0

 - engine: fix crash when using variance shadow maps
@@ -200,7 +200,7 @@ subprojects {
        google()
    }

-    if (!name.startsWith("sample") && name != "filament-tools" && name != "gradle-plugin") {
+    if (!name.startsWith("sample") && name != "filament-tools") {
        apply plugin: 'com.android.library'

        android {
@@ -25,7 +25,7 @@ configured, the corresponding task will be disabled.

 ```groovy
 plugins {
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 filament {
@@ -0,0 +1,21 @@
+plugins {
+    id 'groovy-gradle-plugin'
+}
+
+gradlePlugin {
+    plugins {
+        create("filament-plugin") {
+            id = "filament-plugin"
+            implementationClass = "com.google.android.filament.gradle.FilamentPlugin"
+        }
+    }
+}
+
+repositories {
+    mavenCentral()
+    gradlePluginPortal()
+}
+
+dependencies {
+    implementation "com.google.gradle:osdetector-gradle-plugin:1.7.3"
+}
@@ -60,10 +60,9 @@ class ToolsLocator {
        def classifier =
                project.extensions.getByType(com.google.gradle.osdetector.OsDetector).classifier

-        // If com.google.android.filament.tools-dir is set to a non-empty string, we'll use it as
-        // the tool's base path.
+        // If com.google.android.filament.tools-dir is set, we'll use it as the tool's base path.
        def toolsDirProp = project.providers.gradleProperty("com.google.android.filament.tools-dir")
-        if (toolsDirProp.isPresent() && !toolsDirProp.get().trim().isEmpty()) {
+        if (toolsDirProp.isPresent()) {
            def toolsDir = toolsDirProp.get()
            def path = OperatingSystem.current().isWindows() ?
                "${toolsDir}/bin/${name}.exe" :
@@ -106,8 +106,8 @@ class ModelViewer(
    var skyboxCubemap: Texture? = null

    private lateinit var displayHelper: DisplayHelper
-    private var cameraManipulator: Manipulator? = null
-    private var gestureDetector: GestureDetector? = null
+    private lateinit var cameraManipulator: Manipulator
+    private lateinit var gestureDetector: GestureDetector
    private var surfaceView: SurfaceView? = null
    private var textureView: TextureView? = null

@@ -157,13 +157,15 @@ class ModelViewer(
            surfaceView: SurfaceView,
            engine: Engine = Engine.create(),
            uiHelper: UiHelper = UiHelper(UiHelper.ContextErrorPolicy.DONT_CHECK),
-            manipulator: Manipulator? = defaultCameraManipulator(surfaceView.width, surfaceView.height)
+            manipulator: Manipulator? = null
    ) : this(engine, uiHelper) {
+        cameraManipulator = manipulator ?: Manipulator.Builder()
+                .targetPosition(kDefaultObjectPosition.x, kDefaultObjectPosition.y, kDefaultObjectPosition.z)
+                .viewport(surfaceView.width, surfaceView.height)
+                .build(Manipulator.Mode.ORBIT)
+
        this.surfaceView = surfaceView
-        cameraManipulator = manipulator
-        cameraManipulator?.let { c ->
-            gestureDetector = GestureDetector(surfaceView, c)
-        }
+        gestureDetector = GestureDetector(surfaceView, cameraManipulator)
        displayHelper = DisplayHelper(surfaceView.context)
        uiHelper.renderCallback = SurfaceCallback()
        uiHelper.attachTo(surfaceView)
@@ -175,14 +177,15 @@ class ModelViewer(
            textureView: TextureView,
            engine: Engine = Engine.create(),
            uiHelper: UiHelper = UiHelper(UiHelper.ContextErrorPolicy.DONT_CHECK),
-            manipulator: Manipulator? = defaultCameraManipulator(textureView.width, textureView.height)
+            manipulator: Manipulator? = null
    ) : this(engine, uiHelper) {
-        cameraManipulator = manipulator
+        cameraManipulator = manipulator ?: Manipulator.Builder()
+                .targetPosition(kDefaultObjectPosition.x, kDefaultObjectPosition.y, kDefaultObjectPosition.z)
+                .viewport(textureView.width, textureView.height)
+                .build(Manipulator.Mode.ORBIT)
+
        this.textureView = textureView
-        cameraManipulator = manipulator
-        cameraManipulator?.let { c ->
-            gestureDetector = GestureDetector(textureView, c)
-        }
+        gestureDetector = GestureDetector(textureView, cameraManipulator)
        displayHelper = DisplayHelper(textureView.context)
        uiHelper.renderCallback = SurfaceCallback()
        uiHelper.attachTo(textureView)
@@ -299,13 +302,11 @@ class ModelViewer(
        asset?.let { populateScene(it) }

        // Extract the camera basis from the helper and push it to the Filament camera.
-        cameraManipulator?.let { cm ->
-            cm.getLookAt(eyePos, target, upward)
-            camera.lookAt(
+        cameraManipulator.getLookAt(eyePos, target, upward)
+        camera.lookAt(
                eyePos[0], eyePos[1], eyePos[2],
                target[0], target[1], target[2],
                upward[0], upward[1], upward[2])
-        }

        // Render the scene, unless the renderer wants to skip the frame.
        if (renderer.beginFrame(swapChain!!, frameTimeNanos)) {
@@ -397,7 +398,7 @@ class ModelViewer(
     * Handles a [MotionEvent] to enable one-finger orbit, two-finger pan, and pinch-to-zoom.
     */
    fun onTouchEvent(event: MotionEvent) {
-        gestureDetector?.onTouchEvent(event)
+        gestureDetector.onTouchEvent(event)
    }

    @SuppressWarnings("ClickableViewAccessibility")
@@ -450,7 +451,7 @@ class ModelViewer(

        override fun onResized(width: Int, height: Int) {
            view.viewport = Viewport(0, 0, width, height)
-            cameraManipulator?.setViewport(width, height)
+            cameraManipulator.setViewport(width, height)
            updateCameraProjection()
            synchronizePendingFrames(engine)
        }
@@ -467,11 +468,5 @@ class ModelViewer(

    companion object {
        private val kDefaultObjectPosition = Float3(0.0f, 0.0f, -4.0f)
-        private fun defaultCameraManipulator(width: Int, height: Int) : Manipulator {
-            return Manipulator.Builder()
-                .targetPosition(kDefaultObjectPosition.x, kDefaultObjectPosition.y, kDefaultObjectPosition.z)
-                .viewport(width, height)
-                .build(Manipulator.Mode.ORBIT)
-        }
    }
 }
@@ -1,31 +0,0 @@
-plugins {
-    id 'groovy-gradle-plugin'
-    id 'com.gradle.plugin-publish' version '2.1.0'
-}
-
-group = "com.google.android.filament"
-version = "0.1.0"
-
-gradlePlugin {
-    website = "https://github.com/google/filament/tree/main/android/gradle-plugin"
-    vcsUrl = "https://github.com/google/filament/tree/main/android/gradle-plugin"
-
-    plugins {
-        create("filament-tools") {
-            id = "com.google.android.filament-tools"
-            displayName = "Filament Tools Gradle Plugin"
-            description = "A plugin that helps integrate Filament into Android projects"
-            tags.addAll('android', 'graphics', 'rendering', 'filament', '3d', 'gltf', 'native')
-            implementationClass = "com.google.android.filament.gradle.FilamentPlugin"
-        }
-    }
-}
-
-repositories {
-    mavenCentral()
-    gradlePluginPortal()
-}
-
-dependencies {
-    implementation "com.google.gradle:osdetector-gradle-plugin:1.7.3"
-}
@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.70.1
+VERSION_NAME=1.69.5

 POM_DESCRIPTION=Real-time physically based rendering engine for Android.

@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,7 +1,7 @@
 plugins {
    id 'com.android.application'
    id 'kotlin-android'
-    id 'com.google.android.filament-tools'
+    id 'filament-plugin'
 }

 project.ext.isSample = true
@@ -1,8 +1,3 @@
-// Filament tools plugin
-pluginManagement {
-    includeBuild 'gradle-plugin'
-}
-
 // Libraries
 include ':filament-android'
 include ':filamat-android'
@@ -93,7 +93,6 @@ set(SRCS
        src/MorphTargetBuffer.cpp
        src/PostProcessManager.cpp
        src/ProgramSpecialization.cpp
-        src/LocalProgramCache.cpp
        src/RenderPass.cpp
        src/RenderPrimitive.cpp
        src/RenderTarget.cpp
@@ -187,7 +186,6 @@ set(PRIVATE_HDRS
        src/MaterialCache.h
        src/MaterialDefinition.h
        src/MaterialParser.h
-        src/LocalProgramCache.h
        src/MaterialInstanceManager.h
        src/PIDController.h
        src/PostProcessManager.h
@@ -600,7 +600,6 @@ if (FILAMENT_BUILD_TESTING)
            test/test_CircularBuffer.cpp
            test/test_CommandBufferQueue.cpp
            test/test_Template.cpp
-            test/test_Platform.cpp
        )
        if (FILAMENT_SUPPORTS_WEBGPU)
            list(APPEND BACKEND_TEST_SRC
@@ -731,4 +730,4 @@ if (FILAMENT_BUILD_TESTING)

        set_target_properties(metal_utils_test PROPERTIES FOLDER Tests)
    endif()
-endif()
+endif()
@@ -216,17 +216,6 @@ public:
        MULTIVIEW,
    };

-    /**
-     * Types of device/driver information that can be queried from the platform.
-     */
-    enum class DeviceInfoType {
-        OPENGL_RENDERER,    //!< glGetString(GL_RENDERER)
-        OPENGL_VENDOR,      //!< glGetString(GL_VENDOR)
-        VULKAN_DEVICE_NAME, //!< VkPhysicalDeviceProperties::deviceName
-        VULKAN_DRIVER_NAME, //!< VkPhysicalDeviceDriverProperties::driverName
-        VULKAN_DRIVER_INFO, //!< VkPhysicalDeviceDriverProperties::driverInfo
-    };
-
    /**
     * This controls the priority level for GPU work scheduling, which helps prioritize the
     * submitted GPU work and enables preemption.
@@ -327,7 +316,7 @@ public:
         */
        StereoscopicType stereoscopicType = StereoscopicType::NONE;

-        /**
+        /*
         * The number of eyes to render when stereoscopic rendering is enabled. Supported values are
         * between 1 and Engine::getMaxStereoscopicEyes() (inclusive).
         */
@@ -388,16 +377,6 @@ public:
     */
    virtual int getOSVersion() const noexcept = 0;

-    /**
-     * Queries device/driver information of the graphics API.
-     * @param infoType the type of information to query.
-     * @param driver a pointer to the current driver.
-     * @return a CString containing the requested information.
-     */
-    virtual utils::CString getDeviceInfo(DeviceInfoType infoType,
-            Driver* UTILS_NULLABLE driver) const noexcept = 0;
-
-
    /**
     * Creates and initializes the low-level API (e.g. an OpenGL context or Vulkan instance),
     * then creates the concrete Driver.
@@ -52,9 +52,6 @@ protected:

    ~OpenGLPlatform() noexcept override;

-    utils::CString getDeviceInfo(DeviceInfoType infoType,
-            Driver* UTILS_NULLABLE driver) const noexcept override;
-
 public:
    struct ExternalTexture {
        unsigned int target; // GLenum target
@@ -38,7 +38,6 @@ public:

    Driver* createDriver(void* sharedContext, const Platform::DriverConfig& driverConfig) override;
    int getOSVersion() const noexcept override { return 0; }
-    utils::CString getDeviceInfo(DeviceInfoType, Driver*) const noexcept override { return {}; }

    /**
     * Optionally initializes the Metal platform by acquiring resources necessary for rendering.
@@ -143,8 +143,6 @@ public:
        return 0;
    }

-    utils::CString getDeviceInfo(DeviceInfoType infoType, Driver* driver) const noexcept override;
-
    // ----------------------------------------------------
    // ---------- Platform Customization options ----------
    struct Customization {
@@ -177,12 +175,6 @@ public:
         * presentation. Default is true.
         */
        bool transitionSwapChainImageLayoutForPresent = true;
-
-        /**
-         * The number of frames before an unused framebuffer is evicted from the cache.
-         * Default is 3.
-         */
-        uint32_t timeBeforeEvictionFbo = 3;
    };

    /**
@@ -46,9 +46,6 @@ public:
    ~WebGPUPlatform() override = default;

    [[nodiscard]] int getOSVersion() const noexcept final { return 0; }
-    [[nodiscard]] utils::CString getDeviceInfo(DeviceInfoType, Driver*) const noexcept override {
-        return {};
-    }

    [[nodiscard]] wgpu::Instance& getInstance() noexcept { return mInstance; }

@@ -26,7 +26,6 @@ class PlatformNoop final : public Platform {
 public:

    int getOSVersion() const noexcept final { return 0; }
-    utils::CString getDeviceInfo(DeviceInfoType, Driver*) const noexcept override { return {}; }

    ~PlatformNoop() noexcept override = default;

@@ -341,44 +341,7 @@ void GLDescriptorSet::bind(
                if (arg.handle) {
                    GLTexture const* const t = handleAllocator.handle_cast<GLTexture*>(arg.handle);
                    gl.bindTexture(unit, t->gl.target, t->gl.id, t->gl.external);
-                    SamplerParams params = arg.params;
-
-#if defined(__EMSCRIPTEN__)
-                    // From https://registry.khronos.org/OpenGL-Refpages/es2.0/
-                    // GLES 2.0 will draw mipmapped samplers as black if the following conditions are not met:
-                    //
-                    // "...if the width or height of a texture image are not powers of two and either the
-                    // GL_TEXTURE_MIN_FILTER is set to one of the functions that requires mipmaps or the
-                    // GL_TEXTURE_WRAP_S or GL_TEXTURE_WRAP_T is not set to GL_CLAMP_TO_EDGE, then the
-                    // texture image unit will return (R, G, B, A) = (0, 0, 0, 1)."
-                    //
-                    // "If the texture has dimensions w × h , there are [floor(log2(max(w, h))) + 1] mipmap levels.
-                    // Level 0 is the original texture; level [floor(log2(max(w, h)))] is the final 1 × 1 mipmap."
-                    // "Suppose that a texture is accessed from a fragment shader or vertex shader and has set GL_TEXTURE_MIN_FILTER
-                    // to one of the functions that requires mipmaps. If either the dimensions of the texture images currently
-                    // defined (with previous calls to glTexImage2D, glCompressedTexImage2D, or glCopyTexImage2D) do not follow
-                    // the proper sequence for mipmaps (described above), or there are fewer texture images defined than are needed,
-                    // or the set of texture images were defined with different formats or types, then the texture image unit
-                    // will return (R, G, B, A) = (0, 0, 0, 1)."
-                    //
-                    // So we force the sampler to a valid state in those cases.
-                    auto isPowerOfTwo = [](uint32_t x) -> bool { return (x & (x - 1)) == 0 && x > 0; };
-                    bool textureIsPowerOfTwo = isPowerOfTwo(t->width) && isPowerOfTwo(t->height);
-
-                    uint32_t requiredMipLevels = (std::floor(std::log2(std::max(t->width, std::max(t->height, t->depth)))) + 1);
-                    bool textureHasRequiredMipLevels = t->levels == requiredMipLevels;
-
-                    bool samplerCanBeInvalid = params.filterMin > SamplerMinFilter::LINEAR;
-                    if (samplerCanBeInvalid && (!textureIsPowerOfTwo || !textureHasRequiredMipLevels)) {
-                        if (params.filterMin == SamplerMinFilter::NEAREST_MIPMAP_NEAREST
-                            || params.filterMin == SamplerMinFilter::LINEAR_MIPMAP_NEAREST) {
-                            params.filterMin = SamplerMinFilter::NEAREST;
-                        } else {
-                            params.filterMin = SamplerMinFilter::LINEAR;
-                        }
-                    }
-#endif
-                    
+                    SamplerParams const params = arg.params;
                    glTexParameteri(t->gl.target, GL_TEXTURE_MIN_FILTER,
                            (GLint)GLUtils::getTextureFilter(params.filterMin));
                    glTexParameteri(t->gl.target, GL_TEXTURE_MAG_FILTER,
@@ -45,19 +45,6 @@ Driver* OpenGLPlatform::createDefaultDriver(OpenGLPlatform* platform,

 OpenGLPlatform::~OpenGLPlatform() noexcept = default;

-utils::CString OpenGLPlatform::getDeviceInfo(DeviceInfoType infoType,
-        Driver* driver) const noexcept {
-    switch (infoType) {
-        case DeviceInfoType::OPENGL_RENDERER:
-            return getRendererString(driver);
-        case DeviceInfoType::OPENGL_VENDOR:
-            return getVendorString(driver);
-        default:
-            FILAMENT_CHECK_POSTCONDITION(false) << "Unsupported DeviceInfoType for OpenGLPlatform";
-            return {};
-    }
-}
-
 utils::CString OpenGLPlatform::getVendorString(Driver const* driver) {
    auto const p = static_cast<OpenGLDriverBase const*>(driver);
 #if UTILS_HAS_RTTI
@@ -706,11 +706,10 @@ bool PlatformEGL::setExternalImage(ExternalImageHandleRef externalImage,
 // -----------------------------------------------------------------------------------------------

 void PlatformEGL::initializeGlExtensions() noexcept {
+    // We're guaranteed to be on an ES platform, since we're using EGL
    const char* const extensions = (const char*)glGetString(GL_EXTENSIONS);
-    if (extensions) {
-        GLUtils::unordered_string_set const glExtensions = GLUtils::split(extensions);
-        ext.gl.OES_EGL_image_external_essl3 = glExtensions.has("GL_OES_EGL_image_external_essl3");
-    }
+    GLUtils::unordered_string_set const glExtensions = GLUtils::split(extensions);
+    ext.gl.OES_EGL_image_external_essl3 = glExtensions.has("GL_OES_EGL_image_external_essl3");
 }

 EGLContext PlatformEGL::getContextForType(ContextType const type) const noexcept {
@@ -118,6 +118,14 @@ void VulkanBufferCache::terminate() noexcept {
    }
 }

+size_t VulkanBufferCache::getSize() const noexcept {
+    size_t size = 0;
+    for (int i = 0; i < MAX_POOL_COUNT; i++) {
+        size += mGpuBufferPools[i].size();
+    }
+    return size;
+}
+
 void VulkanBufferCache::release(VulkanGpuBuffer const* gpuBuffer) noexcept {
    assert_invariant(gpuBuffer != nullptr);

@@ -48,6 +48,8 @@ public:
    // This should be called while the context's VkDevice is still alive.
    void terminate() noexcept;

+    size_t getSize() const noexcept;
+
 private:
    struct UnusedGpuBuffer {
        uint64_t lastAccessed;
@@ -158,6 +158,8 @@ struct CommandBufferPool {

    inline bool isRecording() const { return mRecording != INVALID; }

+    size_t getSize() const noexcept { return mBuffers.size(); }
+
 private:
    static constexpr int CAPACITY = FVK_MAX_COMMAND_BUFFERS;
    // int8 only goes up to 127, therefore capacity must be less than that.
@@ -255,6 +257,17 @@ public:
    // Updates the atomic "status" variable in every extant fence.
    void updateFences();

+    struct SizeInfo {
+        size_t regular;
+        size_t protectedPool;
+    };
+    SizeInfo getSize() const noexcept {
+        return {
+            mPool ? mPool->getSize() : 0,
+            mProtectedPool ? mProtectedPool->getSize() : 0
+        };
+    }
+
 #if FVK_ENABLED(FVK_DEBUG_GROUP_MARKERS)
    void pushGroupMarker(char const* str, VulkanGroupMarkers::Timestamp timestamp = {});
    void popGroupMarker();
@@ -137,14 +137,6 @@ public:
        return mFenceExportFlags;
    }

-    inline const char* getPhysicalDeviceName() const noexcept {
-        return mPhysicalDeviceProperties.properties.deviceName;
-    }
-
-    inline const char* getDriverName() const noexcept { return mDriverProperties.driverName; }
-
-    inline const char* getDriverInfo() const noexcept { return mDriverProperties.driverInfo; }
-
    inline bool isImageCubeArraySupported() const noexcept {
        return mPhysicalDeviceFeatures.features.imageCubeArray == VK_TRUE;
    }
@@ -220,16 +212,11 @@ public:
        return mGlobalPrioritySupported;
    }

-    inline bool isDriverPropertiesSupported() const noexcept { return mDriverPropertiesSupported; }
-
 private:
    VkPhysicalDeviceMemoryProperties mMemoryProperties = {};
    VkPhysicalDeviceProperties2 mPhysicalDeviceProperties = {
        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2,
    };
-    VkPhysicalDeviceDriverProperties mDriverProperties = {
-        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES,
-    };
    VkPhysicalDeviceVulkan11Features mPhysicalDeviceVk11Features = {
        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES,
    };
@@ -259,7 +246,6 @@ private:
    bool mProtectedMemorySupported = false;
    bool mVertexInputDynamicStateSupported = false;
    bool mGlobalPrioritySupported = false;
-    bool mDriverPropertiesSupported = false;

    // These are options that can be enabled or disabled at an application level.
    bool mAsyncPipelineCachePrewarmingEnabled = false;
@@ -119,6 +119,9 @@ public:
        return mCapacity;
    }

+    uint16_t size() const { return mSize; }
+    uint16_t unusedCount() const { return mUnusedCount; }
+
    // A convenience method for checking if this pool can allocate sets for a given layout.
    inline bool canAllocate(DescriptorCount const& count) {
        return count == mCount;
@@ -255,6 +258,16 @@ public:
        }
    }

+    VulkanDescriptorSetCache::SizeInfo getSize() const {
+        VulkanDescriptorSetCache::SizeInfo info = {};
+        info.poolCount = mPools.size();
+        for (auto& pool : mPools) {
+            info.totalSize += pool->size();
+            info.totalUnusedCount += pool->unusedCount();
+        }
+        return info;
+    }
+
 private:
    VkDevice mDevice;
    std::vector<std::unique_ptr<DescriptorPool>> mPools;
@@ -355,8 +368,35 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
 }

 void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
-        VkSampler sampler, VkDescriptorSetLayout externalSamplerLayout) noexcept {
+        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
+
+    // TODO: Fix the stream flow case base on the above comment!!
+    if (set->isAnExternalSamplerBound) {
+        auto layout = set->getLayout();
+        // Build a new descriptor set from the new layout
+        VkDescriptorSetLayout const genLayout = set->boundLayout;
+        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();
    VkImageSubresourceRange range = texture->getPrimaryViewRange();
    VkImageViewType const expectedType = texture->getViewType();
@@ -405,28 +445,6 @@ fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetCache::createSet
    return set;
 }

-void VulkanDescriptorSetCache::cloneSet(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        fvkutils::SamplerBitmask samplerMask) noexcept {
-    auto const& layout = set->getLayout();
-    // Build a new descriptor set from layout
-    VkDescriptorSetLayout const genLayout = set->boundLayout;
-    VkDescriptorSet const newSet = getVkSet(layout->count, genLayout);
-
-    // 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 ubo = layout->bitmask.ubo | layout->bitmask.dynamicUbo;
-    // Don't copy the samplers in the mask
-    Bitmask samplers = layout->bitmask.sampler ^ samplerMask;
-    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 VulkanDescriptorSetCache::getVkSet(DescriptorCount const& count,
        VkDescriptorSetLayout vklayout) {
    return mDescriptorPool->obtainSet(count, vklayout);
@@ -443,6 +461,10 @@ void VulkanDescriptorSetCache::manualRecycle(VulkanDescriptorSetLayout::Count co

 void VulkanDescriptorSetCache::gc() { mStashedSets = {}; }

+VulkanDescriptorSetCache::SizeInfo VulkanDescriptorSetCache::getSize() const noexcept {
+    return mDescriptorPool->getSize();
+}
+
 void VulkanDescriptorSetCache::copySet(VkDescriptorSet srcSet, VkDescriptorSet dstSet,
        fvkutils::SamplerBitmask bindings) const {
    // TODO: fix the size for better memory management
@@ -76,11 +76,6 @@ public:
    fvkmemory::resource_ptr<VulkanDescriptorSet> createSet(Handle<HwDescriptorSet> handle,
            fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);

-    // Create and set as current a new VkDescriptorSet using the `set` currently bound layout and
-    // copy all the bindings and ignoring the samplers bindings in the `samplerMask`.
-    void cloneSet(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-            fvkutils::SamplerBitmask samplerMask) noexcept;
-
    // This method is meant to be used with external samplers
    VkDescriptorSet getVkSet(DescriptorCount const& count, VkDescriptorSetLayout vklayout);

@@ -94,6 +89,13 @@ public:

    void resetCachedState() noexcept { mLastBoundInfo = {}; }

+    struct SizeInfo {
+        size_t poolCount;
+        size_t totalSize;
+        size_t totalUnusedCount;
+    };
+    SizeInfo getSize() const noexcept;
+
 private:
    void copySet(VkDescriptorSet srcSet, VkDescriptorSet destSet,
            fvkutils::SamplerBitmask copyBindings) const;
@@ -49,6 +49,8 @@ public:
            fvkutils::SamplerBitmask externalSamplers,
            utils::FixedCapacityVector<std::pair<uint64_t, VkSampler>> immutableSamplers = {});

+    size_t getSize() const noexcept { return mVkLayouts.size(); }
+
 private:
    VkDevice mDevice;
    fvkmemory::ResourceManager* mResourceManager;
@@ -49,6 +49,7 @@

 #include <chrono>
 #include <mutex>
+#include <stdio.h>

 using namespace bluevk;

@@ -260,8 +261,7 @@ VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext& context,
      mPipelineCache(*this, mPlatform->getDevice(), mContext),
      mStagePool(mAllocator, &mResourceManager, &mCommands, &mContext.getPhysicalDeviceLimits()),
      mBufferCache(mContext, mResourceManager, mAllocator),
-      mFramebufferCache(mPlatform->getDevice(),
-              mPlatform->getCustomization().timeBeforeEvictionFbo),
+      mFramebufferCache(mPlatform->getDevice()),
      mYcbcrConversionCache(mPlatform->getDevice()),
      mSamplerCache(mPlatform->getDevice()),
      mBlitter(mPlatform->getPhysicalDevice(), &mCommands),
@@ -446,6 +446,29 @@ void VulkanDriver::collectGarbage() {

    mResourceManager.gc();

+    auto dsSize = mDescriptorSetCache.getSize();
+    auto stageSize = mStagePool.getSize();
+    auto externalSize = mExternalImageManager.getSize();
+    auto commandsSize = mCommands.getSize();
+
+    fprintf(stderr, "Vulkan Cache Sizes:\n");
+    fprintf(stderr, "  Pipelines: %zu\n", mPipelineCache.getSize());
+    fprintf(stderr, "  Pipeline Layouts: %zu\n", mPipelineLayoutCache.getSize());
+    fprintf(stderr, "  Descriptor Set Layouts: %zu\n", mDescriptorSetLayoutCache.getSize());
+    fprintf(stderr, "  Descriptor Sets: %zu (pools: %zu, unused: %zu)\n", dsSize.totalSize, dsSize.poolCount, dsSize.totalUnusedCount);
+    fprintf(stderr, "  FBOs: %zu\n", mFramebufferCache.getFboCacheSize());
+    fprintf(stderr, "  Render Passes: %zu (refcount: %zu)\n",
+            mFramebufferCache.getRenderPassCacheSize(), mFramebufferCache.getRenderPassRefCountSize());
+    fprintf(stderr, "  Samplers: %zu\n", mSamplerCache.getSize());
+    fprintf(stderr, "  Buffers: %zu\n", mBufferCache.getSize());
+    fprintf(stderr, "  Stages: %zu, Free Images: %zu\n", stageSize.stages, stageSize.freeImages);
+    fprintf(stderr, "  YCbCr Conversions: %zu\n", mYcbcrConversionCache.getSize());
+    fprintf(stderr, "  External Images: %zu, Set Bindings: %zu\n", externalSize.images, externalSize.setBindings);
+    fprintf(stderr, "  Streamed Bindings: %zu\n", mStreamedImageManager.getSize());
+    fprintf(stderr, "  Command Buffers: %zu (regular: %zu, protected: %zu)\n",
+            commandsSize.regular + commandsSize.protectedPool, commandsSize.regular, commandsSize.protectedPool);
+    fprintf(stderr, "  Semaphores: %zu\n", mSemaphoreManager.getSize());
+
 #if FVK_ENABLED(FVK_DEBUG_RESOURCE_LEAK)
    mResourceManager.print();
 #endif
@@ -1990,7 +2013,7 @@ void VulkanDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassP
    rt->emitBarriersBeginRenderPass(*commandBuffer);

    fvkmemory::resource_ptr<VulkanFramebuffer> vkfb =
-            mFramebufferCache.getFramebuffer(fbkey, &mResourceManager, rt);
+            mFramebufferCache.getFramebuffer(fbkey, &mResourceManager);

 // Assign a label to the framebuffer for debugging purposes.
 #if FVK_ENABLED(FVK_DEBUG_GROUP_MARKERS | FVK_DEBUG_DEBUG_UTILS)
@@ -94,9 +94,6 @@ void VulkanExternalImageManager::updateSetAndLayout(
    fvkmemory::resource_ptr<VulkanDescriptorSetLayout> const& layout = set->getLayout();
    set->boundLayout = mDescriptorSetLayoutCache->getVkLayout(layout->bitmask,
            actualExternalSamplers, outSamplers);
-
-    mDescriptorSetCache->cloneSet(set, actualExternalSamplers);
-
    // Update the external samplers in the set
    for (auto& [binding, sampler, image]: samplerAndBindings) {
        // We cannot call updateSamplerForExternalSamplerSet because some samplers are non NULL
@@ -85,6 +85,12 @@ public:
    // - Update the bindings that use external samplers.
    void updateSetAndLayout(fvkmemory::resource_ptr<VulkanDescriptorSet> set);

+    struct SizeInfo {
+        size_t setBindings;
+        size_t images;
+    };
+    SizeInfo getSize() const noexcept { return { mSetBindings.size(), mImages.size() }; }
+
 private:

    VulkanSamplerCache* mSamplerCache;
@@ -20,7 +20,6 @@
 #include "VulkanHandles.h"
 #include "vulkan/utils/Image.h"

-#include <utils/compiler.h>
 #include <utils/Panic.h>

 // If any VkRenderPass or VkFramebuffer is unused for more than TIME_BEFORE_EVICTION frames, it
@@ -63,18 +62,16 @@ bool VulkanFboCache::FboKeyEqualFn::operator()(const FboKey& k1, const FboKey& k
    return true;
 }

-VulkanFboCache::VulkanFboCache(VkDevice device, uint32_t timeBeforeEvictionFbo)
-        : mDevice(device),
-          mTimeBeforeEvictionFbo(timeBeforeEvictionFbo) {}
+VulkanFboCache::VulkanFboCache(VkDevice device)
+    : mDevice(device) {}

 VulkanFboCache::~VulkanFboCache() {
    FILAMENT_CHECK_POSTCONDITION(mFramebufferCache.empty() && mRenderPassCache.empty())
            << "Please explicitly call terminate() while the VkDevice is still alive.";
 }

-fvkmemory::resource_ptr<VulkanFramebuffer> VulkanFboCache::getFramebuffer(FboKey const& config,
-        fvkmemory::ResourceManager* resManager,
-        fvkmemory::resource_ptr<VulkanRenderTarget> renderTarget) noexcept {
+fvkmemory::resource_ptr<VulkanFramebuffer> VulkanFboCache::getFramebuffer(
+        FboKey const& config, fvkmemory::ResourceManager* resManager) noexcept {
    FboMap::iterator iter = mFramebufferCache.find(config);
    if (UTILS_LIKELY(iter != mFramebufferCache.end())) {
        iter.value().timestamp = mCurrentTime;
@@ -123,9 +120,8 @@ fvkmemory::resource_ptr<VulkanFramebuffer> VulkanFboCache::getFramebuffer(FboKey
    FILAMENT_CHECK_POSTCONDITION(error == VK_SUCCESS) << "Unable to create framebuffer."
                                                     << " error=" << static_cast<int32_t>(error);
    fvkmemory::resource_ptr<VulkanFramebuffer> fbh =
-            fvkmemory::resource_ptr<VulkanFramebuffer>::construct(resManager, mDevice, framebuffer,
-                    renderTarget);
-    mFramebufferCache[config] = { fbh, mCurrentTime };
+        fvkmemory::resource_ptr<VulkanFramebuffer>::construct(resManager, mDevice, framebuffer);
+    mFramebufferCache[config] = {fbh, mCurrentTime};
    return fbh;
 }

@@ -378,36 +374,31 @@ void VulkanFboCache::gc() noexcept {
    FVK_SYSTRACE_START("fbocache::gc");

    // If this is one of the first few frames, return early to avoid wrapping unsigned integers.
-    ++mCurrentTime;
+    if (++mCurrentTime <= TIME_BEFORE_EVICTION) {
+        return;
+    }
+    const uint32_t evictTime = mCurrentTime - TIME_BEFORE_EVICTION;

-    if (UTILS_UNLIKELY(mCurrentTime > mTimeBeforeEvictionFbo)) {
-        const uint32_t evictTimeFbo = mCurrentTime - mTimeBeforeEvictionFbo;
-        for (FboMap::iterator iter = mFramebufferCache.begin(); iter != mFramebufferCache.end();) {
-            const FboVal fbo = iter->second;
-            if (fbo.timestamp < evictTimeFbo && fbo.handle) {
-                mRenderPassRefCount[iter->first.renderPass]--;
+    for (FboMap::iterator iter = mFramebufferCache.begin(); iter != mFramebufferCache.end(); ) {
+        const FboVal fbo = iter->second;
+        if (fbo.timestamp < evictTime && fbo.handle) {
+            mRenderPassRefCount[iter->first.renderPass]--;

-                // erase(iterator) returns the iterator to the next element.
-                iter = mFramebufferCache.erase(iter);
-            } else {
-                ++iter;
-            }
+            // erase(iterator) returns the iterator to the next element.
+            iter = mFramebufferCache.erase(iter);
+        } else {
+            ++iter;
        }
    }

-    if (UTILS_UNLIKELY(mCurrentTime > TIME_BEFORE_EVICTION)) {
-        const uint32_t evictTimeRp = mCurrentTime - TIME_BEFORE_EVICTION;
-        for (RenderPassMap::iterator iter = mRenderPassCache.begin();
-                iter != mRenderPassCache.end();) {
-            const VkRenderPass handle = iter->second.handle->getVkRenderPass();
-            if (iter->second.timestamp < evictTimeRp && handle &&
-                    mRenderPassRefCount[handle] == 0) {
-                // erase(iterator) returns the iterator to the next element.
-                iter = mRenderPassCache.erase(iter);
-                mRenderPassRefCount.erase(handle);
-            } else {
-                ++iter;
-            }
+    for (RenderPassMap::iterator iter = mRenderPassCache.begin(); iter != mRenderPassCache.end(); ) {
+        const VkRenderPass handle = iter->second.handle->getVkRenderPass();
+        if (iter->second.timestamp < evictTime && handle && mRenderPassRefCount[handle] == 0) {
+            // erase(iterator) returns the iterator to the next element.
+            iter = mRenderPassCache.erase(iter);
+            mRenderPassRefCount.erase(handle);
+        } else {
+            ++iter;
        }
    }

@@ -100,13 +100,12 @@ public:
        bool operator()(const FboKey& k1, const FboKey& k2) const;
    };

-    explicit VulkanFboCache(VkDevice device, uint32_t timeBeforeEvictionFbo);
+    explicit VulkanFboCache(VkDevice device);
    ~VulkanFboCache();

    // Retrieves or creates a VkFramebuffer handle.
-    fvkmemory::resource_ptr<VulkanFramebuffer> getFramebuffer(FboKey const& config,
-            fvkmemory::ResourceManager* resManager,
-            fvkmemory::resource_ptr<VulkanRenderTarget> renderTarget) noexcept;
+    fvkmemory::resource_ptr<VulkanFramebuffer> getFramebuffer(
+        FboKey const& config, fvkmemory::ResourceManager* resManager) noexcept;

    // Retrieves or creates a VkRenderPass handle.
    fvkmemory::resource_ptr<VulkanRenderPass> getRenderPass(
@@ -121,6 +120,10 @@ public:
    // Frees all Vulkan objects. Call this during shutdown before the device is destroyed.
    void terminate() noexcept;

+    size_t getFboCacheSize() const noexcept { return mFramebufferCache.size(); }
+    size_t getRenderPassCacheSize() const noexcept { return mRenderPassCache.size(); }
+    size_t getRenderPassRefCountSize() const noexcept { return mRenderPassRefCount.size(); }
+
 private:
    VkDevice mDevice;
    using FboMap = tsl::robin_map<FboKey, FboVal, FboKeyHashFn, FboKeyEqualFn>;
@@ -130,7 +133,6 @@ private:
    RenderPassMap mRenderPassCache;
    tsl::robin_map<VkRenderPass, uint32_t> mRenderPassRefCount;
    uint32_t mCurrentTime = 0;
-    uint32_t mTimeBeforeEvictionFbo;
 };

 } // namespace filament::backend
@@ -715,11 +715,8 @@ VulkanRenderPrimitive::VulkanRenderPrimitive(PrimitiveType pt,
      vertexBuffer(vb),
      indexBuffer(ib) {}

-VulkanFramebuffer::VulkanFramebuffer(VkDevice device, VkFramebuffer framebuffer,
-        fvkmemory::resource_ptr<VulkanRenderTarget> renderTarget)
-        : mDevice(device),
-          mFramebuffer(framebuffer),
-          mRenderTarget(renderTarget) {}
+VulkanFramebuffer::VulkanFramebuffer(VkDevice device, VkFramebuffer framebuffer)
+    : mDevice(device), mFramebuffer(framebuffer) {}

 VulkanFramebuffer::~VulkanFramebuffer() {
    vkDestroyFramebuffer(mDevice, mFramebuffer, VKALLOC);
@@ -195,7 +195,7 @@ public:
    }

    // The current layout used by the descriptor set. This one will match the bindings, including
-    // external samplers data.
+    // external samplers data. 
    // This will not necessarilly be the same as `mLayout`.
    VkDescriptorSetLayout boundLayout = VK_NULL_HANDLE;

@@ -203,7 +203,7 @@ public:

    fvkutils::UniformBufferBitmask const& dynamicUboMask;
    uint8_t const uniqueDynamicUboCount;
-
+    
    // Flag to indicate if the current layout needs to be recreated or not.
    // This should only set to `true` when a external sampler image is bound to the descriptor set.
    bool isLayoutDirty = false;
@@ -586,9 +586,7 @@ struct VulkanRenderPrimitive : public HwRenderPrimitive, fvkmemory::Resource {
 };

 struct VulkanFramebuffer : public fvkmemory::Resource {
-    VulkanFramebuffer(VkDevice device,
-            VkFramebuffer framebuffer,
-            fvkmemory::resource_ptr<VulkanRenderTarget> renderTarget);
+    VulkanFramebuffer(VkDevice device, VkFramebuffer framebuffer);
    ~VulkanFramebuffer();

    inline VkFramebuffer getVkFramebuffer() const noexcept {
@@ -598,11 +596,6 @@ struct VulkanFramebuffer : public fvkmemory::Resource {
 private:
    VkDevice mDevice;
    VkFramebuffer mFramebuffer;
-
-    // We need to keep a reference to the renderTarget because the key of the framebuffer in the
-    // cache has references to the image views that are derived from the textures of the render
-    // target.
-    fvkmemory::resource_ptr<VulkanRenderTarget> mRenderTarget;
 };

 struct VulkanRenderPass : public fvkmemory::Resource {
@@ -136,6 +136,8 @@ public:
    void terminate() noexcept;
    void gc() noexcept;

+    size_t getSize() const noexcept { return mPipelines.size(); }
+
 private:
    // PIPELINE CACHE KEY
    // ------------------
@@ -59,6 +59,8 @@ public:
    VkPipelineLayout getLayout(DescriptorSetLayoutArray const& descriptorSetLayouts,
            fvkmemory::resource_ptr<VulkanProgram> program);

+    size_t getSize() const noexcept { return mPipelineLayouts.size(); }
+
 private:
    using Timestamp = uint64_t;
    struct PipelineLayoutCacheEntry {
@@ -40,6 +40,8 @@ public:
    explicit VulkanSamplerCache(VkDevice device);
    VkSampler getSampler(Params params);
    void terminate() noexcept;
+
+    size_t getSize() const noexcept { return mCache.size(); }
 private:
    VkDevice mDevice;

@@ -37,6 +37,8 @@ public:
    void terminate();
    Semaphore acquire();

+    size_t getSize() const noexcept { return mPool.size(); }
+
 private:
    friend struct VulkanSemaphore;
    void recycle(VkSemaphore semaphore);
@@ -223,6 +223,12 @@ public:
    // resource_ptrs, as this would lead to undefined behavior.
    void terminate() noexcept;

+    struct SizeInfo {
+        size_t stages;
+        size_t freeImages;
+    };
+    SizeInfo getSize() const noexcept { return { mStages.size(), mFreeImages.size() }; }
+
 private:
    VmaAllocator mAllocator;
    fvkmemory::ResourceManager* mResManager;
@@ -46,6 +46,8 @@ public:
    void onStreamAcquireImage(fvkmemory::resource_ptr<VulkanTexture> image,
            fvkmemory::resource_ptr<VulkanStream> stream);

+    size_t getSize() const noexcept { return mStreamedTexturesBindings.size(); }
+
 private:
    struct StreamedTextureBinding {
        uint8_t binding = 0;
@@ -42,6 +42,8 @@ public:
    VkSamplerYcbcrConversion getConversion(Params params);
    void terminate() noexcept;

+    size_t getSize() const noexcept { return mCache.size(); }
+
 private:
    VkDevice mDevice;

@@ -704,30 +704,6 @@ VulkanPlatform::VulkanPlatform() = default;

 VulkanPlatform::~VulkanPlatform() = default;

-utils::CString VulkanPlatform::getDeviceInfo(DeviceInfoType infoType,
-        Driver* driver) const noexcept {
-    if (mImpl->mPhysicalDevice == VK_NULL_HANDLE) {
-        return {};
-    }
-    auto& context = mImpl->mContext;
-    switch (infoType) {
-        case DeviceInfoType::VULKAN_DEVICE_NAME: {
-            return utils::CString(context.getPhysicalDeviceName());
-        }
-        case DeviceInfoType::VULKAN_DRIVER_NAME: {
-            return context.isDriverPropertiesSupported() ? utils::CString(context.getDriverName())
-                                                         : utils::CString();
-        }
-        case DeviceInfoType::VULKAN_DRIVER_INFO: {
-            return context.isDriverPropertiesSupported() ? utils::CString(context.getDriverInfo())
-                                                         : utils::CString();
-        }
-        default:
-            FILAMENT_CHECK_POSTCONDITION(false) << "Unsupported DeviceInfoType for VulkanPlatform";
-            return {};
-    }
-}
-
 VulkanPlatform::SwapChainBundle VulkanPlatform::getSwapChainBundle(SwapChainPtr handle) {
    return static_cast<VulkanPlatformSwapChainBase*>(handle)->getSwapChainBundle();
 }
@@ -976,11 +952,6 @@ void VulkanPlatform::queryAndSetDeviceFeatures(Platform::DriverConfig const& dri
        chainStruct(&context.mPhysicalDeviceFeatures, &globalPriorityFeatures);
    }

-    if (vkGetPhysicalDeviceProperties2) {
-        chainStruct(&context.mPhysicalDeviceProperties, &context.mDriverProperties);
-        context.mDriverPropertiesSupported = true;
-    }
-
    // Initialize the following fields: physicalDeviceProperties, memoryProperties,
    // physicalDeviceFeatures.
    vkGetPhysicalDeviceProperties2(mImpl->mPhysicalDevice, &context.mPhysicalDeviceProperties);
@@ -192,15 +192,8 @@ void WebGPUDriver::flush(int) {
 }

 void WebGPUDriver::finish(int /* dummy */) {
+    mReadPixelMapsCounter.waitForAllToFinish();
    mQueueManager.finish();
-
-    // We use polling to advance webgpu's callback counter until all the read backs have been
-    // processed. Note that blocking with mReadPixelMapsCounter.waitForAllToFinish will only
-    // deadlock since we could not advance the counter.
-    while (!mReadPixelMapsCounter.isIdle()) {
-        mAdapter.GetInstance().ProcessEvents();
-        std::this_thread::sleep_for(std::chrono::milliseconds(1));
-    }
 }

 void WebGPUDriver::destroyRenderPrimitive(Handle<HwRenderPrimitive> rph) {
@@ -947,6 +940,10 @@ size_t WebGPUDriver::getUniformBufferOffsetAlignment(){

 void WebGPUDriver::updateIndexBuffer(Handle<HwIndexBuffer> indexBufferHandle,
        BufferDescriptor&& bufferDescriptor, const uint32_t byteOffset) {
+    // make sure command elements (draws, etc.) prior to the buffer update are processed before the
+    // update on the GPU, otherwise the expected data may not be available at the time certain
+    // draw calls are made.
+    flush();
    handleCast<WebGPUIndexBuffer>(indexBufferHandle)
            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager, &mStagePool);
    scheduleDestroy(std::move(bufferDescriptor));
@@ -961,6 +958,10 @@ void WebGPUDriver::updateIndexBufferAsyncR(AsyncCallId jobId,

 void WebGPUDriver::updateBufferObject(Handle<HwBufferObject> bufferObjectHandle,
        BufferDescriptor&& bufferDescriptor, const uint32_t byteOffset) {
+    // make sure command elements (draws, etc.) prior to the buffer update are processed before the
+    // update on the GPU, otherwise the expected data may not be available at the time certain
+    // draw calls are made.
+    flush();
    handleCast<WebGPUBufferObject>(bufferObjectHandle)
            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager, &mStagePool);
    scheduleDestroy(std::move(bufferDescriptor));
@@ -1007,6 +1008,10 @@ void WebGPUDriver::update3DImage(Handle<HwTexture> textureHandle, const uint32_t
        const uint32_t xoffset, const uint32_t yoffset, const uint32_t zoffset,
        const uint32_t width, const uint32_t height, const uint32_t depth,
        PixelBufferDescriptor&& pixelBufferDescriptor) {
+    // one way or another this function writes texture(s), thus any commands (draw calls etc.) should
+    // get submitted prior to these updates so that subsequent commands/draws run with the
+    // image/texture(s) updated as expected.
+    flush();
    PixelBufferDescriptor* inputData{ &pixelBufferDescriptor };
    PixelBufferDescriptor reshapedData;
    if (reshape(pixelBufferDescriptor, reshapedData)) {
@@ -1178,6 +1183,11 @@ void WebGPUDriver::generateMipmaps(Handle<HwTexture> textureHandle) {
        return; // nothing to do
    }

+    // make sure command elements (draws, etc.) prior to the texture update are processed before the
+    // update on the GPU.
+    // this ensures subsequent draw calls are run after the mipmaps have been generated as expected
+    flush();
+
    const auto usage = wgpuTexture.GetUsage();
    FILAMENT_CHECK_PRECONDITION(usage & wgpu::TextureUsage::TextureBinding)
            << "Texture for mipmap generation must have TextureBinding usage.";
@@ -1366,9 +1376,6 @@ void WebGPUDriver::beginRenderPass(Handle<HwRenderTarget> renderTargetHandle,
            customDepthStencilMsaaSidecarTextureView);

    mRenderPassEncoder = commandEncoder.BeginRenderPass(&renderPassDescriptor);
-
-    // TODO: there's a bug here because the webgpu viewport has the origin at top-left, whereas
-    // filament expects it to be bottom left.
    mRenderPassEncoder.SetViewport(
            static_cast<float>(params.viewport.left),
            static_cast<float>(params.viewport.bottom),
@@ -1456,6 +1463,7 @@ void WebGPUDriver::stopCapture(int /* dummy */) {
 void WebGPUDriver::readPixels(Handle<HwRenderTarget> sourceRenderTargetHandle, const uint32_t x,
        const uint32_t y, const uint32_t width, const uint32_t height,
        PixelBufferDescriptor&& pixelBufferDescriptor) {
+    flush();
    const auto srcTarget{ handleCast<WebGPURenderTarget>(sourceRenderTargetHandle) };
    assert_invariant(srcTarget);

@@ -1513,7 +1521,10 @@ void WebGPUDriver::readTextureToBuffer(wgpu::Texture srcTexture, uint32_t level,
        return;
    }

-    auto commandEncoder = mQueueManager.getCommandEncoder();
+    const wgpu::CommandEncoderDescriptor commandEncoderDescriptor{
+        .label = "read_texture_to_buffer_command",
+    };
+    auto commandEncoder = mDevice.CreateCommandEncoder(&commandEncoderDescriptor);
    FILAMENT_CHECK_POSTCONDITION(commandEncoder)
            << "Failed to create command encoder for readTextureToBuffer?";

@@ -1635,6 +1646,9 @@ void WebGPUDriver::readTextureToBuffer(wgpu::Texture srcTexture, uint32_t level,
        .depthOrArrayLayers = 1,
    };
    commandEncoder.CopyTextureToBuffer(&source, &destination, &copySize);
+    wgpu::CommandBuffer commandBuffer = commandEncoder.Finish();
+    assert_invariant(commandBuffer);
+    mDevice.GetQueue().Submit(1, &commandBuffer);

    // Map the buffer to read the data
    struct UserData final {
@@ -1655,9 +1669,6 @@ void WebGPUDriver::readTextureToBuffer(wgpu::Texture srcTexture, uint32_t level,
    });

    mReadPixelMapsCounter.startTask();
-    // We need to flush here before we can readback from the staging buffer since the copy op is
-    // pending to be submitted.
-    mQueueManager.flush();
    userData->buffer.MapAsync(
            wgpu::MapMode::Read, 0, bufferSize, wgpu::CallbackMode::AllowSpontaneous,
            [](wgpu::MapAsyncStatus status, const char* message, UserData* userdata) {
@@ -1750,6 +1761,7 @@ void WebGPUDriver::blitDEPRECATED(TargetBufferFlags buffers,
    const wgpu::Extent2D destinationSize{ static_cast<uint32_t>(destinationViewport.width),
        static_cast<uint32_t>(destinationViewport.height) };

+    mQueueManager.flush();
    wgpu::CommandEncoder commandEncoder = mQueueManager.getCommandEncoder();
    const WebGPUBlitter::BlitArgs blitArgs{
        .source = { .texture = sourceTexture->getTexture(),
@@ -1769,6 +1781,8 @@ void WebGPUDriver::blitDEPRECATED(TargetBufferFlags buffers,
        .filter = filter,
    };
    mBlitter.blit(mDevice.GetQueue(), commandEncoder, blitArgs);
+
+    mQueueManager.flush();
 }

 void WebGPUDriver::resolve(Handle<HwTexture> destinationTextureHandle, const uint8_t sourceLevel,
@@ -1808,6 +1822,7 @@ void WebGPUDriver::blit(Handle<HwTexture> destinationTextureHandle, const uint8_
        Handle<HwTexture> sourceTextureHandle, const uint8_t destinationLevel,
        const uint8_t destinationLayer, const math::uint2 sourceOrigin, const math::uint2 size) {
    FWGPU_SYSTRACE_SCOPE();
+    mQueueManager.flush();
    wgpu::CommandEncoder commandEncoder = mQueueManager.getCommandEncoder();
    const auto sourceTexture{ handleCast<WebGPUTexture>(sourceTextureHandle) };
    const auto destinationTexture{ handleCast<WebGPUTexture>(destinationTextureHandle) };
@@ -1831,6 +1846,7 @@ void WebGPUDriver::blit(Handle<HwTexture> destinationTextureHandle, const uint8_
        .filter = SamplerMagFilter::NEAREST,
    };
    mBlitter.blit(mDevice.GetQueue(), commandEncoder, blitArgs);
+    mQueueManager.flush();
 }

 void WebGPUDriver::bindPipeline(PipelineState const& pipelineState) {
@@ -50,9 +50,6 @@ public:

    [[nodiscard]] bool isHeadless() const { return mType == SwapChainType::HEADLESS; }

-    [[nodiscard]] uint32_t getWidth() const { return mConfig.width; }
-    [[nodiscard]] uint32_t getHeight() const { return mConfig.height; }
-
    void present(DriverBase& driver);

    void setFrameScheduledCallback(CallbackHandler* handler, FrameScheduledCallback&& callback) {
@@ -41,10 +41,5 @@ void AsyncTaskCounter::waitForAllToFinish() {
    mFinishedCondition.wait(lock, [this] { return mTasksInProgress == 0; });
 }

-bool AsyncTaskCounter::isIdle() {
-    std::lock_guard<std::mutex> lock{ mMutex };
-    return mTasksInProgress == 0;
-}
-
 } // namespace filament::backend::webgpuutils

@@ -51,11 +51,6 @@ public:
     */
    void waitForAllToFinish();

-    /**
-     * Check if all tasks have finished (work counter is 0) (thread-safe)
-     */
-    bool isIdle();
-
 private:
    std::mutex mMutex;
    std::condition_variable mFinishedCondition;
@@ -35,10 +35,9 @@ test::NativeView getNativeView() {

 namespace {

-std::array<test::Backend, 3> const VALID_BACKENDS{
-    test::Backend::OPENGL,
-    test::Backend::VULKAN,
-    test::Backend::WEBGPU
+std::array<test::Backend, 2> const VALID_BACKENDS{
+        test::Backend::OPENGL,
+        test::Backend::VULKAN,
 };

 }// namespace
@@ -1,50 +0,0 @@
-/*
- * Copyright (C) 2026 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 <backend/Platform.h>
-#include <gtest/gtest.h>
-#include <private/backend/PlatformFactory.h>
-
-namespace filament::backend {
-
-TEST(PlatformTest, GetDeviceInfo) {
-    Backend backend = Backend::DEFAULT;
-    Platform* platform = PlatformFactory::create(&backend);
-    ASSERT_NE(platform, nullptr);
-
-    // Test valid queries for the current platform (will be either GL, Vulkan, or Metal depending on
-    // host)
-    if (backend == Backend::OPENGL) {
-        platform->getDeviceInfo(Platform::DeviceInfoType::OPENGL_RENDERER, nullptr);
-        platform->getDeviceInfo(Platform::DeviceInfoType::OPENGL_VENDOR, nullptr);
-
-        // Death tests for Vulkan info on OpenGL platform
-        EXPECT_DEATH(platform->getDeviceInfo(Platform::DeviceInfoType::VULKAN_DEVICE_NAME, nullptr),
-                "Unsupported DeviceInfoType");
-    } else if (backend == Backend::VULKAN) {
-        platform->getDeviceInfo(Platform::DeviceInfoType::VULKAN_DEVICE_NAME, nullptr);
-        platform->getDeviceInfo(Platform::DeviceInfoType::VULKAN_DRIVER_NAME, nullptr);
-        platform->getDeviceInfo(Platform::DeviceInfoType::VULKAN_DRIVER_INFO, nullptr);
-
-        // Death tests for OpenGL info on Vulkan platform
-        EXPECT_DEATH(platform->getDeviceInfo(Platform::DeviceInfoType::OPENGL_RENDERER, nullptr),
-                "Unsupported DeviceInfoType");
-    }
-
-    PlatformFactory::destroy(&platform);
-}
-
-} // namespace filament::backend
@@ -227,6 +227,7 @@ bool Froxelizer::prepare(
        const mat4f& projection, float const projectionNear, float const projectionFar,
        float4 const& clipTransform) noexcept {
    assert_invariant(projectionFar > projectionNear);
+    assert_invariant(projectionNear > 0);
    setViewport(viewport);
    setProjection(projection, projectionNear, projectionFar);

@@ -1,265 +0,0 @@
-/*
- * Copyright (C) 2026 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 "LocalProgramCache.h"
-#include "MaterialParser.h"
-
-#include "details/Engine.h"
-#include "details/Material.h"
-
-#include <backend/DriverApiForward.h>
-
-namespace filament {
-
-using namespace backend;
-using namespace utils;
-
-LocalProgramCache::LocalProgramCache(LocalProgramCache const& other)
-        : mMaterial(other.mMaterial),
-          mCachedPrograms(other.mCachedPrograms.size()),
-          mSpecializationConstants((other.mMaterial != nullptr)
-                                           ? other.mMaterial->getEngine()
-                                                     .getMaterialCache()
-                                                     .getSpecializationConstantsInternPool()
-                                                     .acquire(other.mSpecializationConstants)
-                                           : SpecializationConstants()) {}
-
-LocalProgramCache& LocalProgramCache::operator=(LocalProgramCache const& other) {
-    assert_invariant(mMaterial == nullptr);
-    assert_invariant(mCachedPrograms.empty());
-    assert_invariant(mSpecializationConstants.empty());
-
-    mMaterial = other.mMaterial;
-    if (mMaterial != nullptr) {
-        mCachedPrograms = FixedCapacityVector<Handle<HwProgram>>(other.mCachedPrograms.size());
-        mSpecializationConstants = other.mMaterial->getEngine()
-                .getMaterialCache()
-                .getSpecializationConstantsInternPool()
-                .acquire(other.mSpecializationConstants);
-    }
-
-    return *this;
-}
-
-void LocalProgramCache::initializeForMaterial(FEngine& engine, FMaterial const& material,
-        utils::FixedCapacityVector<backend::Program::SpecializationConstant>
-                specializationConstants) {
-    assert_invariant(mMaterial == nullptr);
-    assert_invariant(mCachedPrograms.empty());
-    assert_invariant(mSpecializationConstants.empty());
-
-    mMaterial = &material;
-
-    mSpecializationConstants =
-            engine.getMaterialCache().getSpecializationConstantsInternPool().acquire(
-                    std::move(specializationConstants));
-
-    size_t cachedProgramsSize;
-    switch (material.getMaterialDomain()) {
-        case filament::MaterialDomain::SURFACE:
-            cachedProgramsSize = 1 << VARIANT_BITS;
-            break;
-        case filament::MaterialDomain::POST_PROCESS:
-            cachedProgramsSize = 1 << POST_PROCESS_VARIANT_BITS;
-            break;
-        case filament::MaterialDomain::COMPUTE:
-            cachedProgramsSize = 1;
-            break;
-    }
-    mCachedPrograms = FixedCapacityVector<Handle<HwProgram>>(cachedProgramsSize);
-
-    material.getDefinition().acquirePrograms(engine, mCachedPrograms.as_slice(),
-            material.getMaterialParser(), mSpecializationConstants, material.isDefaultMaterial());
-}
-
-void LocalProgramCache::initializeForMaterialInstance(FEngine& engine, FMaterial const& material) {
-    assert_invariant(mMaterial == nullptr);
-    assert_invariant(mCachedPrograms.empty());
-    assert_invariant(mSpecializationConstants.empty());
-
-    mMaterial = &material;
-    LocalProgramCache const& programs = material.getPrograms();
-
-    mSpecializationConstants =
-            engine.getMaterialCache().getSpecializationConstantsInternPool().acquire(
-                    programs.getSpecializationConstants());
-
-    mCachedPrograms =
-            FixedCapacityVector<Handle<HwProgram>>(material.getPrograms().mCachedPrograms.size());
-}
-
-Handle<HwProgram> LocalProgramCache::prepareProgramSlow(DriverApi& driver, Variant const variant,
-        CompilerPriorityQueue const priorityQueue) const noexcept {
-    assert_invariant(mMaterial != nullptr);
-
-    FEngine& engine = mMaterial->getEngine();
-    if (mMaterial->isSharedVariant(variant)) {
-        FMaterial const* defaultMaterial = engine.getDefaultMaterial();
-        assert_invariant(defaultMaterial);
-        LocalProgramCache const& defaultPrograms = defaultMaterial->getPrograms();
-        Handle<HwProgram> program = defaultPrograms.mCachedPrograms[variant.key];
-        if (program) {
-            return mCachedPrograms[variant.key] = program;
-        }
-        return mCachedPrograms[variant.key] =
-                defaultPrograms.prepareProgram(driver, variant, priorityQueue);
-    }
-    return mCachedPrograms[variant.key] = mMaterial->getDefinition().prepareProgram(engine, driver,
-                   mMaterial->getMaterialParser(), getProgramSpecialization(variant), priorityQueue);
-}
-
-ProgramSpecialization LocalProgramCache::getProgramSpecialization(Variant variant) const noexcept {
-    assert_invariant(mMaterial != nullptr);
-
-    return ProgramSpecialization {
-        .materialCrc32 = mMaterial->getMaterialParser().getCrc32(),
-        .variant = variant,
-        .specializationConstants = mSpecializationConstants,
-    };
-}
-
-void LocalProgramCache::terminate(FEngine& engine) {
-    assert_invariant(mMaterial != nullptr);
-
-    mMaterial->getDefinition().releasePrograms(engine, mCachedPrograms.as_slice(),
-            mMaterial->getMaterialParser(), mSpecializationConstants,
-            mMaterial->isDefaultMaterial());
-    engine.getMaterialCache().releaseMaterial(engine, mMaterial->getDefinition());
-    engine.getMaterialCache().getSpecializationConstantsInternPool().release(
-            mSpecializationConstants);
-}
-
-void LocalProgramCache::clear(FEngine& engine) {
-    assert_invariant(mMaterial != nullptr);
-
-    mMaterial->getDefinition().releasePrograms(engine, mCachedPrograms.as_slice(),
-            mMaterial->getMaterialParser(), mSpecializationConstants,
-            mMaterial->isDefaultMaterial());
-}
-
-Variant LocalProgramCache::filterVariantForGetProgram(Variant variant) const noexcept {
-    if (UTILS_UNLIKELY(mMaterial->getEngine().features.material.enable_fog_as_postprocess)) {
-        // if the fog as post-process feature is enabled, we need to proceed "as-if" the material
-        // didn't have the FOG variant bit.
-        if (mMaterial->getMaterialDomain() == MaterialDomain::SURFACE) {
-            BlendingMode const blendingMode = mMaterial->getBlendingMode();
-            bool const hasScreenSpaceRefraction =
-                    mMaterial->getRefractionMode() == RefractionMode::SCREEN_SPACE;
-            bool const isBlendingCommand = !hasScreenSpaceRefraction &&
-                    (blendingMode != BlendingMode::OPAQUE && blendingMode != BlendingMode::MASKED);
-            if (!isBlendingCommand) {
-                variant.setFog(false);
-            }
-        }
-    }
-    return variant;
-}
-
-Program::SpecializationConstant LocalProgramCache::getConstantImpl(uint32_t id) const noexcept {
-    return mSpecializationConstants[id];
-}
-
-Program::SpecializationConstant LocalProgramCache::getConstantImpl(
-        std::string_view name) const noexcept {
-    assert_invariant(mMaterial != nullptr);
-
-    MaterialDefinition const& definition = mMaterial->getDefinition();
-    auto it = definition.specializationConstantsNameToIndex.find(name);
-    if (it != definition.specializationConstantsNameToIndex.cend()) {
-        return getConstantImpl(it->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS);
-    }
-
-    std::string name_cstring(name);
-    PANIC_PRECONDITION("No such constant exists: %s", name_cstring.c_str());
-    return {};
-}
-
-void LocalProgramCache::setConstants(
-        std::initializer_list<std::pair<uint32_t, Program::SpecializationConstant>>
-                constants) noexcept {
-    assert_invariant(mMaterial != nullptr);
-
-    auto newSpecializationConstants =
-            FixedCapacityVector<Program::SpecializationConstant>(mSpecializationConstants);
-
-    bool hasChanged = false;
-    for (const auto& [id, value] : constants) {
-        if (newSpecializationConstants[id] != value) {
-            newSpecializationConstants[id] = value;
-            hasChanged = true;
-        }
-    }
-
-    if (hasChanged) {
-        setConstantsImpl(std::move(newSpecializationConstants));
-    }
-}
-
-void LocalProgramCache::setConstants(
-        std::initializer_list<std::pair<std::string_view, Program::SpecializationConstant>>
-                constants) noexcept {
-    assert_invariant(mMaterial != nullptr);
-
-    auto newSpecializationConstants =
-            FixedCapacityVector<Program::SpecializationConstant>(mSpecializationConstants);
-
-    bool hasChanged = false;
-    for (const auto& [name, value] : constants) {
-        MaterialDefinition const& definition = mMaterial->getDefinition();
-        auto it = definition.specializationConstantsNameToIndex.find(name);
-        if (it != definition.specializationConstantsNameToIndex.cend()) {
-            uint32_t id = it->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS;
-            if (newSpecializationConstants[id] != value) {
-                newSpecializationConstants[id] = value;
-                hasChanged = true;
-            }
-        }
-    }
-
-    if (hasChanged) {
-        setConstantsImpl(std::move(newSpecializationConstants));
-    }
-}
-
-void LocalProgramCache::setConstantsImpl(
-        FixedCapacityVector<Program::SpecializationConstant> constants) noexcept {
-    FEngine& engine = mMaterial->getEngine();
-
-    auto& internPool = engine.getMaterialCache().getSpecializationConstantsInternPool();
-    MaterialParser const& materialParser = mMaterial->getMaterialParser();
-    MaterialDefinition const& definition = mMaterial->getDefinition();
-    const bool isDefaultMaterial = mMaterial->isDefaultMaterial();
-
-    // Release old resources...
-    definition.releasePrograms(engine, mCachedPrograms.as_slice(), materialParser,
-            mSpecializationConstants, isDefaultMaterial);
-    internPool.release(mSpecializationConstants);
-
-    // Then acquire new ones.
-    mSpecializationConstants = internPool.acquire(std::move(constants));
-    definition.acquirePrograms(engine, mCachedPrograms.as_slice(), materialParser,
-            mSpecializationConstants, isDefaultMaterial);
-}
-
-template int32_t LocalProgramCache::getConstant<int32_t>(uint32_t id) const noexcept;
-template float LocalProgramCache::getConstant<float>(uint32_t id) const noexcept;
-template bool LocalProgramCache::getConstant<bool>(uint32_t id) const noexcept;
-
-template int32_t LocalProgramCache::getConstant<int32_t>(std::string_view name) const noexcept;
-template float LocalProgramCache::getConstant<float>(std::string_view name) const noexcept;
-template bool LocalProgramCache::getConstant<bool>(std::string_view name) const noexcept;
-
-} // namespace filament
@@ -1,144 +0,0 @@
-/*
- * Copyright (C) 2026 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_LOCALPROGRAMCACHE_H
-#define TNT_FILAMENT_LOCALPROGRAMCACHE_H
-
-#include "MaterialDefinition.h"
-
-#include <backend/Handle.h>
-
-#include <private/filament/Variant.h>
-
-#include <backend/DriverEnums.h>
-#include <backend/DriverApiForward.h>
-#include <backend/Program.h>
-
-#include <utility>
-
-namespace filament {
-
-class FMaterial;
-
-// L0 cache for material programs. Manages recompiling them on-demand; owned by Material and
-// MaterialInstance.
-class LocalProgramCache {
-    template<typename T>
-    using is_supported_constant_parameter_t =
-            std::enable_if_t<std::is_same_v<int32_t, T> || std::is_same_v<float, T> ||
-                             std::is_same_v<bool, T>>;
-
-public:
-    using Programs = utils::Slice<const backend::Handle<backend::HwProgram>>;
-    using SpecializationConstants = utils::Slice<const backend::Program::SpecializationConstant>;
-
-    LocalProgramCache() = default;
-    LocalProgramCache(LocalProgramCache const& other);
-
-    LocalProgramCache& operator=(LocalProgramCache const& other);
-
-    // Initialize for use in a Material.
-    void initializeForMaterial(FEngine& engine, FMaterial const& material,
-            utils::FixedCapacityVector<backend::Program::SpecializationConstant>
-                    specializationConstants);
-
-    // Initialize for use in a MaterialInstance. Copies the set of spec constants currently in use
-    // from its Material.
-    void initializeForMaterialInstance(FEngine& engine, FMaterial const& material);
-
-    bool isInitialized() const noexcept { return mMaterial != nullptr; }
-
-    // prepareProgram creates the program for the material's given variant at the backend level.
-    // Must be called outside of backend render pass.
-    // Must be called before getProgram() below.
-    backend::Handle<backend::HwProgram> prepareProgram(backend::DriverApi& driver,
-            Variant const variant,
-            backend::CompilerPriorityQueue const priorityQueue) const noexcept {
-        backend::Handle<backend::HwProgram> program = mCachedPrograms[variant.key];
-        if (UTILS_LIKELY(program)) {
-            return program;
-        }
-        return prepareProgramSlow(driver, variant, priorityQueue);
-    }
-
-    // getProgram returns the backend program for the material's given variant.
-    // Must be called after prepareProgram().
-    [[nodiscard]]
-    backend::Handle<backend::HwProgram> getProgram(Variant variant) const noexcept {
-        variant = filterVariantForGetProgram(variant);
-        backend::Handle<backend::HwProgram> program = mCachedPrograms[variant.key];
-        assert_invariant(program);
-        return program;
-    }
-
-    SpecializationConstants getSpecializationConstants() const noexcept {
-        return mSpecializationConstants;
-    }
-
-    Programs getPrograms() const noexcept { return mCachedPrograms.as_slice(); }
-
-    // Free all engine resources associated with this instance.
-    void terminate(FEngine& engine);
-
-    // Clear all cached programs. Used primarily by matdbg to "sever" a Material's connection to the
-    // global material cache.
-    void clear(FEngine& engine);
-
-    // Get constant by ID.
-    template<typename T, typename = is_supported_constant_parameter_t<T>>
-    T getConstant(uint32_t id) const noexcept {
-        return std::get<T>(getConstantImpl(id));
-    }
-
-    // Get constant by name.
-    template<typename T, typename = is_supported_constant_parameter_t<T>>
-    T getConstant(std::string_view name) const noexcept {
-        return std::get<T>(getConstantImpl(name));
-    }
-
-    // Set constants by ID.
-    void setConstants(
-            std::initializer_list<std::pair<uint32_t, backend::Program::SpecializationConstant>>
-                    constants) noexcept;
-
-    // Set constants by name.
-    void setConstants(std::initializer_list<
-            std::pair<std::string_view, backend::Program::SpecializationConstant>>
-                    constants) noexcept;
-
-private:
-    backend::Handle<backend::HwProgram> prepareProgramSlow(backend::DriverApi& driver,
-            Variant const variant,
-            backend::CompilerPriorityQueue const priorityQueue) const noexcept;
-
-    ProgramSpecialization getProgramSpecialization(Variant variant) const noexcept;
-
-    Variant filterVariantForGetProgram(Variant const variant) const noexcept;
-
-    void setConstantsImpl(utils::FixedCapacityVector<backend::Program::SpecializationConstant>
-                    constants) noexcept;
-
-    backend::Program::SpecializationConstant getConstantImpl(uint32_t id) const noexcept;
-    backend::Program::SpecializationConstant getConstantImpl(std::string_view name) const noexcept;
-
-    FMaterial const* mMaterial = nullptr;
-    mutable utils::FixedCapacityVector<backend::Handle<backend::HwProgram>> mCachedPrograms;
-    SpecializationConstants mSpecializationConstants;
-};
-
-} // namespace filament
-
-#endif  // TNT_FILAMENT_LOCALPROGRAMCACHE_H
@@ -39,25 +39,12 @@ bool MaterialCache::MaterialKey::operator==(MaterialKey const& rhs) const noexce
    return parser == rhs.parser;
 }

-MaterialCache::MaterialCache()
-        : mDefinitions("MaterialCache::mDefinitions", 0),
-          mPrograms("MaterialCache::mPrograms", 0) {}
-
 MaterialCache::~MaterialCache() {
    assert_invariant(mDefinitions.empty());
    assert_invariant(mPrograms.empty());
    assert_invariant(mSpecializationConstantsInternPool.empty());
 }

-void MaterialCache::terminate(FEngine& engine) {
-    mPrograms.clearLruCache([&engine](backend::Handle<backend::HwProgram>& program) {
-        engine.getDriverApi().destroyProgram(program);
-    });
-    mDefinitions.clearLruCache([&engine](MaterialDefinition& definition) {
-        definition.terminate(engine);
-    });
-}
-
 MaterialDefinition* UTILS_NULLABLE MaterialCache::acquireMaterial(FEngine& engine,
        const void* UTILS_NONNULL data, size_t size) noexcept {
    std::unique_ptr<MaterialParser> parser = MaterialDefinition::createParser(engine.getBackend(),
@@ -56,13 +56,8 @@ public:
    using ProgramCache =
            utils::RefCountedMap<ProgramSpecialization, backend::Handle<backend::HwProgram>>;

-    MaterialCache();
    ~MaterialCache();

-    // All reference-counted resources should be freed by the time MaterialCache is destructed, but
-    // the LRU cache needs to be explicitly freed in addition.
-    void terminate(FEngine& engine);
-
    SpecializationConstantInternPool& getSpecializationConstantsInternPool() {
        return mSpecializationConstantsInternPool;
    }
@@ -49,7 +49,6 @@
 #include "fsr.h"
 #include "FrameHistory.h"
 #include "RenderPass.h"
-#include "ShadowMapManager.h"

 #include "details/Camera.h"
 #include "details/ColorGrading.h"
@@ -999,10 +998,13 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::screenSpaceAmbientOcclusion(
                auto& material = getPostProcessMaterial(materialName);

                FMaterial* ma = material.getMaterial(mEngine, driver);
-                ma->getPrograms().setConstants({
-                    { "useVisibilityBitmasks", options.gtao.useVisibilityBitmasks },
-                    { "linearThickness", options.gtao.linearThickness },
-                });
+                {
+                    FMaterial::SpecializationConstantsBuilder maConstants =
+                            ma->getSpecializationConstantsBuilder();
+                    maConstants.set("useVisibilityBitmasks", options.gtao.useVisibilityBitmasks);
+                    maConstants.set("linearThickness", options.gtao.linearThickness);
+                    ma->setSpecializationConstants(std::move(maConstants));
+                }

                ma = material.getMaterial(mEngine, driver);
                FMaterialInstance* const mi = getMaterialInstance(ma);
@@ -2915,18 +2917,20 @@ void PostProcessManager::configureTemporalAntiAliasingMaterial(backend::DriverAp
        TemporalAntiAliasingOptions const& taaOptions) noexcept {

    FMaterial* const ma = getPostProcessMaterial("taa").getMaterial(mEngine, driver);
-    ma->getPrograms().setConstants({
-        { "upscaling", taaOptions.upscaling > 1.0f },
-        { "historyReprojection", taaOptions.historyReprojection },
-        { "filterHistory", taaOptions.filterHistory },
-        { "filterInput", taaOptions.filterInput },
-        { "useYCoCg", taaOptions.useYCoCg },
-        { "hdr", taaOptions.hdr },
-        { "preventFlickering", taaOptions.preventFlickering },
-        { "boxType", int32_t(taaOptions.boxType) },
-        { "boxClipping", int32_t(taaOptions.boxClipping) },
-        { "varianceGamma", taaOptions.varianceGamma },
-    });
+    FMaterial::SpecializationConstantsBuilder maConstants = ma->getSpecializationConstantsBuilder();
+
+    maConstants.set("upscaling", taaOptions.upscaling > 1.0f);
+    maConstants.set("historyReprojection", taaOptions.historyReprojection);
+    maConstants.set("filterHistory", taaOptions.filterHistory);
+    maConstants.set("filterInput", taaOptions.filterInput);
+    maConstants.set("useYCoCg", taaOptions.useYCoCg);
+    maConstants.set("hdr", taaOptions.hdr);
+    maConstants.set("preventFlickering", taaOptions.preventFlickering);
+    maConstants.set("boxType", int32_t(taaOptions.boxType));
+    maConstants.set("boxClipping", int32_t(taaOptions.boxClipping));
+    maConstants.set("varianceGamma", taaOptions.varianceGamma);
+
+    ma->setSpecializationConstants(std::move(maConstants));
 }

 FMaterialInstance* PostProcessManager::configureColorGradingMaterial(backend::DriverApi& driver,
@@ -2934,10 +2938,13 @@ FMaterialInstance* PostProcessManager::configureColorGradingMaterial(backend::Dr
        ColorGradingConfig const& colorGradingConfig, VignetteOptions const& vignetteOptions,
        uint32_t const width, uint32_t const height) noexcept {
    FMaterial* ma = material.getMaterial(mEngine, driver);
-    ma->getPrograms().setConstants({
-        { "isOneDimensional", colorGrading->isOneDimensional() },
-        { "isLDR", colorGrading->isLDR() },
-    });
+    {
+        FMaterial::SpecializationConstantsBuilder maConstants =
+                ma->getSpecializationConstantsBuilder();
+        maConstants.set("isOneDimensional", colorGrading->isOneDimensional());
+        maConstants.set("isLDR", colorGrading->isLDR());
+        ma->setSpecializationConstants(std::move(maConstants));
+    }

    PostProcessVariant const variant = colorGradingConfig.translucent
                                               ? PostProcessVariant::TRANSLUCENT
@@ -3673,12 +3680,8 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::resolve(FrameGraph& fg,
    }

    // The Metal / Vulkan backends currently don't support depth/stencil resolve.
-    // TODO: Stencil resolve is actually *not* supported. Trying to resolve a stencil texture will
-    //     trigger an assert on debug builds. We need to investigate how this can be accomplished
-    //     through shaders or some other manipulation.
-    if ((isDepthFormat(inDesc.format) || isStencilFormat(inDesc.format)) &&
-            (!mDepthStencilResolveSupported)) {
-        return resolveDepthWithShader(fg, outputBufferName, input, outDesc);
+    if (isDepthFormat(inDesc.format) && (!mDepthStencilResolveSupported)) {
+        return resolveDepth(fg, outputBufferName, input, outDesc);
    }

    outDesc.width = inDesc.width;
@@ -3693,6 +3696,9 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::resolve(FrameGraph& fg,

    auto const& ppResolve = fg.addPass<ResolveData>("resolve",
            [&](FrameGraph::Builder& builder, auto& data) {
+                // we currently don't support stencil resolve.
+                assert_invariant(!isStencilFormat(inDesc.format));
+
                data.input = builder.read(input, FrameGraphTexture::Usage::BLIT_SRC);
                data.output = builder.createTexture(outputBufferName, outDesc);
                data.output = builder.write(data.output, FrameGraphTexture::Usage::BLIT_DST);
@@ -3717,7 +3723,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::resolve(FrameGraph& fg,
    return ppResolve->output;
 }

-FrameGraphId<FrameGraphTexture> PostProcessManager::resolveDepthWithShader(FrameGraph& fg,
+FrameGraphId<FrameGraphTexture> PostProcessManager::resolveDepth(FrameGraph& fg,
        utils::StaticString outputBufferName, FrameGraphId<FrameGraphTexture> const input,
        FrameGraphTexture::Descriptor outDesc) noexcept {

@@ -3742,8 +3748,11 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::resolveDepthWithShader(Frame
        FrameGraphId<FrameGraphTexture> output;
    };

-    auto const& ppResolve = fg.addPass<ResolveData>("resolveDepthWithShader",
+    auto const& ppResolve = fg.addPass<ResolveData>("resolveDepth",
            [&](FrameGraph::Builder& builder, auto& data) {
+                // we currently don't support stencil resolve
+                assert_invariant(!isStencilFormat(inDesc.format));
+
                data.input = builder.sample(input);
                data.output = builder.createTexture(outputBufferName, outDesc);
                data.output = builder.write(data.output, FrameGraphTexture::Usage::DEPTH_ATTACHMENT);
@@ -3752,9 +3761,6 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::resolveDepthWithShader(Frame
                        .clearFlags = TargetBufferFlags::DEPTH });
            },
            [=, this](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {
-                // we currently don't support stencil resolve
-                assert_invariant(!isStencilFormat(inDesc.format));
-
                bindPostProcessDescriptorSet(driver);
                bindPerRenderableDescriptorSet(driver);
                auto const& input = resources.getTexture(data.input);
@@ -3831,7 +3837,6 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::vsmMipmapPass(FrameGraph& fg
 }

 FrameGraphId<FrameGraphTexture> PostProcessManager::debugShadowCascades(FrameGraph& fg,
-        ShadowMapManager const& smm,
        FrameGraphId<FrameGraphTexture> const input,
        FrameGraphId<FrameGraphTexture> const depth) noexcept {

@@ -3849,31 +3854,17 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::debugShadowCascades(FrameGra
                data.output = builder.createTexture("Shadow Cascade Debug", desc);
                builder.declareRenderPass(data.output);
            },
-            [=, &smm, this](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {
+            [=, this](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {
                bindPostProcessDescriptorSet(driver);
                bindPerRenderableDescriptorSet(driver);
                auto color = resources.getTexture(data.color);
                auto depth = resources.getTexture(data.depth);
                auto const out = resources.getRenderPassInfo();
-
-                auto const& smu = smm.getShadowMappingUniforms();
-                mat4f lightFromWorldMatrix[4] = {};
-                float4 scissorNormalized[4] = {};
-                for (size_t i = 0, c = std::max(4u, (smu.cascades & 0xF)) ; i < c; i++) {
-                    auto const& csp = smm.getCascadeShaderParameters(i);
-                    lightFromWorldMatrix[i] = csp.lightSpace;
-                    scissorNormalized[i] = csp.scissorNormalized;
-                }
-
                auto const& material = getPostProcessMaterial("debugShadowCascades");
-                FMaterialInstance* const mi = getMaterialInstance(mEngine, driver, material);
+                FMaterialInstance* const mi =
+                        getMaterialInstance(mEngine, driver, material);
                mi->setParameter("color",  color, SamplerParams{});  // nearest
                mi->setParameter("depth",  depth, SamplerParams{});  // nearest
-                mi->setParameter("cascadeSplits",  smu.cascadeSplits);
-                mi->setParameter("cascadeCount",  smu.cascades & 0xF);
-                mi->setParameter("shadowAtlasResolution",  smu.atlasResolution);
-                mi->setParameter("lightFromWorldMatrix",  lightFromWorldMatrix, 4);
-                mi->setParameter("scissorNormalized",  scissorNormalized, 4);
                commitAndRenderFullScreenQuad(driver, out, mi);
                unbindAllDescriptorSets(driver);
            });
@@ -66,7 +66,6 @@ class FMaterialInstance;
 class FrameGraph;
 class RenderPass;
 class RenderPassBuilder;
-class ShadowMapManager;
 class UboManager;
 struct CameraInfo;

@@ -305,10 +304,9 @@ public:
            utils::StaticString outputBufferName, FrameGraphId<FrameGraphTexture> input,
            FrameGraphTexture::Descriptor outDesc) noexcept;

-    // Resolves base level of input and outputs a texture from outDesc using a shader instead of
-    // driver-implemented API.
+    // Resolves base level of input and outputs a texture from outDesc.
    // outDesc with, height, format and samples will be overridden.
-    FrameGraphId<FrameGraphTexture> resolveDepthWithShader(FrameGraph& fg,
+    FrameGraphId<FrameGraphTexture> resolveDepth(FrameGraph& fg,
            utils::StaticString outputBufferName, FrameGraphId<FrameGraphTexture> input,
            FrameGraphTexture::Descriptor outDesc) noexcept;

@@ -323,7 +321,6 @@ public:
            bool reinhard, size_t kernelWidth, float sigma) noexcept;

    FrameGraphId<FrameGraphTexture> debugShadowCascades(FrameGraph& fg,
-            ShadowMapManager const& smm,
            FrameGraphId<FrameGraphTexture> input,
            FrameGraphId<FrameGraphTexture> depth) noexcept;

@@ -54,7 +54,7 @@ RenderableManager::getInstance(Entity const e) const noexcept {
 }

 void RenderableManager::destroy(Entity const e) noexcept {
-    return downcast(this)->clientDestroy(e);
+    return downcast(this)->destroy(e);
 }

 void RenderableManager::setAxisAlignedBoundingBox(Instance const instance, const Box& aabb) {
@@ -681,7 +681,6 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
    ShadowTechnique shadowTechnique{};
    uint32_t directionalShadowsMask = 0;
    uint32_t cascadeHasVisibleShadows = 0;
-    float4 wsSplitPositionUniform = { -std::numeric_limits<float>::infinity() };

    if (hasVisibleShadows) {
        uint32_t const cascadeCount = cascadedShadowMaps.size();
@@ -705,8 +704,11 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
        // (which is the near plane, and doesn't need to be communicated to the shaders).
        static_assert(CONFIG_MAX_SHADOW_CASCADES <= 5,
                "At most, a float4 can fit 4 split positions for 5 shadow cascades");
+        float4 wsSplitPositionUniform{ -std::numeric_limits<float>::infinity() };
        std::copy(splits.begin() + 1, splits.end(), &wsSplitPositionUniform[0]);

+        mShadowMappingUniforms.cascadeSplits = wsSplitPositionUniform;
+
        // When computing the required bias we need a half-texel size, so we multiply by 0.5 here.
        // note: normalBias is set to zero for VSM
        const float normalBias = shadowMapInfo.vsm ? 0.0f : 0.5f * lcm.getShadowNormalBias(0);
@@ -731,8 +733,6 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
                    lightData, 0, cameraInfo, shadowMapInfo, sceneInfo,
                    canUseDepthClamp);

-            mCascadesShaderParameters[i] = shaderParameters;
-
            if (shadowMap.hasVisibleShadows()) {
                const size_t shadowIndex = shadowMap.getShadowIndex();
                assert_invariant(shadowIndex == i);
@@ -775,14 +775,10 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
    cascades |= uint32_t(cascadedShadowMaps.size());
    cascades |= cascadeHasVisibleShadows << 8u;

-    mShadowMappingUniforms = {
-        .cascadeSplits = wsSplitPositionUniform,
-        .ssContactShadowDistance = screenSpaceShadowDistance,
-        .directionalShadows = directionalShadowsMask,
-        .cascades = cascades,
-        .atlasResolution = { mTextureAtlasRequirements.size, 1.0f / float(mTextureAtlasRequirements.size) }
-    };
-    
+    mShadowMappingUniforms.directionalShadows = directionalShadowsMask;
+    mShadowMappingUniforms.ssContactShadowDistance = screenSpaceShadowDistance;
+    mShadowMappingUniforms.cascades = cascades;
+
    return shadowTechnique;
 }

@@ -48,7 +48,6 @@

 #include <math/mat4.h>
 #include <math/half.h>
-#include <math/vec2.h>
 #include <math/vec4.h>

 #include <algorithm>
@@ -76,7 +75,6 @@ struct ShadowMappingUniforms {
    float ssContactShadowDistance;
    uint32_t directionalShadows;
    uint32_t cascades;
-    math::float2 atlasResolution;
 };

 class ShadowMapManager {
@@ -172,10 +170,6 @@ public:
        return std::min(targetExponent, filterCeiling);
    }

-    ShadowMap::ShaderParameters const& getCascadeShaderParameters(size_t index) const noexcept {
-        return mCascadesShaderParameters[index];
-    }
-
 private:
    explicit ShadowMapManager(FEngine& engine);

@@ -259,8 +253,6 @@ private:

    ShadowMap::SceneInfo mSceneInfo;

-    ShadowMap::ShaderParameters mCascadesShaderParameters[4]{};
-
    // Inline storage for all our ShadowMap objects, we can't easily use a std::array<> directly.
    // Because ShadowMap doesn't have a default ctor, and we avoid out-of-line allocations.
    // Each ShadowMap is currently 88 bytes (total of ~12KB for 128 shadow maps)
@@ -85,6 +85,7 @@ void FCameraManager::destroy(FEngine& engine, Entity const e) noexcept {
        { // scope for camera -- it's invalid after this scope.
            FCamera* const camera = manager.elementAt<CAMERA>(i);
            assert_invariant(camera);
+            camera->terminate(engine);
            engine.getHeapAllocator().destroy(camera);

            // Remove the camera component
@@ -577,7 +577,7 @@ void FRenderableManager::create(
    FEngine::DriverApi& driver = engine.getDriverApi();

    if (UTILS_UNLIKELY(manager.hasComponent(entity))) {
-        destroy(entity, driver);
+        destroy(entity);
    }
    Instance const ci = manager.addComponent(entity);
    assert_invariant(ci);
@@ -733,42 +733,41 @@ void FRenderableManager::create(
 }

 // this destroys a single component from an entity
-void FRenderableManager::destroy(Entity const e, DriverApi& driver) noexcept {
+void FRenderableManager::destroy(Entity const e) noexcept {
    Instance const ci = getInstance(e);
    if (ci) {
-        destroyComponent(ci, driver);
+        destroyComponent(ci);
        mManager.removeComponent(e);
    }
 }

-void FRenderableManager::clientDestroy(utils::Entity e) noexcept {
-    destroy(e, mEngine.getDriverApi());
-}
-
 // this destroys all components in this manager
-void FRenderableManager::terminate(DriverApi& driver) noexcept {
+void FRenderableManager::terminate() noexcept {
    auto& manager = mManager;
    if (!manager.empty()) {
        DLOG(INFO) << "cleaning up " << manager.getComponentCount()
                   << " leaked Renderable components";
        while (!manager.empty()) {
            Instance const ci = manager.end() - 1;
-            destroyComponent(ci, driver);
+            destroyComponent(ci);
            manager.removeComponent(manager.getEntity(ci));
        }
    }
-    mHwRenderPrimitiveFactory.terminate(driver);
+    mHwRenderPrimitiveFactory.terminate(mEngine.getDriverApi());
 }

-void FRenderableManager::gc(EntityManager& em, DriverApi& driver) noexcept {
-    mManager.gc(em, [this, &driver](Entity const e) {
-        destroy(e, driver);
+void FRenderableManager::gc(EntityManager& em) noexcept {
+    mManager.gc(em, [this](Entity const e) {
+        destroy(e);
    });
 }

 // This is basically a Renderable's destructor.
-void FRenderableManager::destroyComponent(Instance const ci, DriverApi& driver) noexcept {
+void FRenderableManager::destroyComponent(Instance const ci) noexcept {
    auto& manager = mManager;
+    FEngine& engine = mEngine;
+
+    FEngine::DriverApi& driver = engine.getDriverApi();

    // See create(RenderableManager::Builder&, Entity)
    destroyComponentPrimitives(mHwRenderPrimitiveFactory, driver, manager[ci].primitives);
@@ -91,9 +91,9 @@ public:
    ~FRenderableManager();

    // free-up all resources
-    void terminate(backend::DriverApi& driver) noexcept;
+    void terminate() noexcept;

-    void gc(utils::EntityManager& em, backend::DriverApi& driver) noexcept;
+    void gc(utils::EntityManager& em) noexcept;

    /*
     * Component Manager APIs
@@ -125,11 +125,7 @@ public:

    void create(const Builder& builder, utils::Entity entity);

-    void destroy(utils::Entity e, backend::DriverApi& driver) noexcept;
-
-    // The client API should have taken an Engine&, but it doesn't, so that's a workaround
-    // so we can keep the API as it is.
-    void clientDestroy(utils::Entity e) noexcept;
+    void destroy(utils::Entity e) noexcept;

    inline void setAxisAlignedBoundingBox(Instance instance, const Box& aabb);

@@ -243,7 +239,7 @@ public:
    };

 private:
-    void destroyComponent(Instance ci, backend::DriverApi& driver) noexcept;
+    void destroyComponent(Instance ci) noexcept;
    static void destroyComponentPrimitives(
            HwRenderPrimitiveFactory& factory, backend::DriverApi& driver,
            utils::Slice<FRenderPrimitive> primitives) noexcept;
@@ -46,6 +46,9 @@ public:

    FCamera(FEngine& engine, utils::Entity e);

+    void terminate(FEngine&) noexcept { }
+
+
    // Sets the projection matrices (viewing and culling). The viewing matrice has infinite far.
    void setProjection(Projection projection,
                       double left, double right, double bottom, double top,
@@ -20,11 +20,6 @@
 #include "TextureCache.h"
 #include "RenderPrimitive.h"

-#include "components/CameraManager.h"
-#include "components/LightManager.h"
-#include "components/RenderableManager.h"
-#include "components/TransformManager.h"
-
 #include "details/BufferObject.h"
 #include "details/Camera.h"
 #include "details/Fence.h"
@@ -63,7 +58,6 @@
 #include <utils/CallStack.h>
 #include <utils/CString.h>
 #include <utils/Invocable.h>
-#include <utils/JobSystem.h>
 #include <utils/Logger.h>
 #include <utils/Panic.h>
 #include <utils/PrivateImplementation-impl.h>
@@ -544,10 +538,12 @@ void FEngine::init() {
            &debug.shadowmap.debug_directional_shadowmap, [this] {
                mMaterials.forEach([this](FMaterial* material) {
                    if (material->getMaterialDomain() == MaterialDomain::SURFACE) {
-                        material->getPrograms().setConstants({
-                            { +ReservedSpecializationConstants::CONFIG_DEBUG_DIRECTIONAL_SHADOWMAP,
-                              debug.shadowmap.debug_directional_shadowmap },
-                        });
+                        FMaterial::SpecializationConstantsBuilder constants =
+                                material->getSpecializationConstantsBuilder();
+                        constants.set(+ReservedSpecializationConstants::
+                                              CONFIG_DEBUG_DIRECTIONAL_SHADOWMAP,
+                                debug.shadowmap.debug_directional_shadowmap);
+                        material->setSpecializationConstants(std::move(constants));
                    }
                });
            });
@@ -556,10 +552,12 @@ void FEngine::init() {
            &debug.lighting.debug_froxel_visualization, [this] {
                mMaterials.forEach([this](FMaterial* material) {
                    if (material->getMaterialDomain() == MaterialDomain::SURFACE) {
-                        material->getPrograms().setConstants({
-                            { +ReservedSpecializationConstants::CONFIG_DEBUG_FROXEL_VISUALIZATION,
-                              debug.lighting.debug_froxel_visualization },
-                        });
+                        FMaterial::SpecializationConstantsBuilder constants =
+                                material->getSpecializationConstantsBuilder();
+                        constants.set(+ReservedSpecializationConstants::
+                                              CONFIG_DEBUG_FROXEL_VISUALIZATION,
+                                debug.lighting.debug_froxel_visualization);
+                        material->setSpecializationConstants(std::move(constants));
                    }
                });
            });
@@ -592,9 +590,6 @@ void FEngine::shutdown() {
    DLOG(INFO) << "CircularBuffer: High watermark " << wm / 1024 << " KiB (" << wmpct << "%)";
 #endif

-    /* Destroy any leftover items in the cache. */
-    mMaterialCache.terminate(*this);
-
    DriverApi& driver = getDriverApi();

    /*
@@ -605,7 +600,7 @@ void FEngine::shutdown() {
    mResourceAllocatorDisposer->terminate();
    mResourceAllocatorDisposer.reset();
    mDFG.terminate(*this);                  // free-up the DFG
-    mRenderableManager.terminate(driver);   // free-up all renderables
+    mRenderableManager.terminate();         // free-up all renderables
    mLightManager.terminate();              // free-up all lights
    mCameraManager.terminate(*this);        // free-up all cameras

@@ -769,30 +764,13 @@ void FEngine::prepare(DriverApi& driver) {
    });
 }

-void FEngine::gc() {
-    JobSystem& js = getJobSystem();
-    auto *rootJob = js.createJob();
-
-    js.run(
-            jobs::createJob(js, rootJob, [this] {
-                // These are safe to GC *sequentially* from a worker thread.
-                // The JobSystem acts as a release/acquire operation.
-                mLightManager.gc(mEntityManager);
-                mTransformManager.gc(mEntityManager);
-                mCameraManager.gc(*this, mEntityManager);
-            }));
-
-    if (isAsynchronousModeEnabled()) {
-        // gcDeferredAsyncObjectDestruction() is thread-safe in this context because
-        // 1. JobSystem provides the release/acquire operation
-        // 2. it's only used from the main thread, which we are blocking during the gc
-        js.run(jobs::createJob(js, rootJob, &FEngine::gcDeferredAsyncObjectDestruction, this));
-    }
-
-    // RenderableManager cannot be gc'ed from a different thread, because it needs the DriverAPI
-    mRenderableManager.gc(mEntityManager, getDriverApi());
-
-    js.runAndWait(rootJob);
+void FEngine::gcManagers() {
+    // Note: this runs in a Job
+    auto& em = mEntityManager;
+    mRenderableManager.gc(em);
+    mLightManager.gc(em);
+    mTransformManager.gc(em);
+    mCameraManager.gc(*this, em);
 }

 void FEngine::gcDeferredAsyncObjectDestruction() {
@@ -820,7 +798,7 @@ void FEngine::submitFrame() {
 void FEngine::flush() {
    // flush the command buffer
    flushCommandBuffer(mCommandBufferQueue);
-
+    
    // In single-threaded mode, we have to call execute() to drain the command
    // buffer to really free up space
    if constexpr (!UTILS_HAS_THREADING) {
@@ -889,14 +867,8 @@ int FEngine::loop() {
    #endif
    if (portString != nullptr) {
        const int port = atoi(portString);
-
-        ShaderLanguage preferredLanguage = ShaderLanguage::UNSPECIFIED;
-        if (mBackend == Backend::METAL) {
-            preferredLanguage = ShaderLanguage::MSL;
-        }
-
        debug.server = new matdbg::DebugServer(mBackend,
-                mDriver->getShaderLanguages(preferredLanguage).front(),
+                mDriver->getShaderLanguages(ShaderLanguage::UNSPECIFIED).front(),
                matdbg::DbgShaderModel((uint8_t) mDriver->getShaderModel()), port);

        // Sometimes the server can fail to spin up (e.g. if the above port is already in use).
@@ -1427,7 +1399,7 @@ bool FEngine::destroy(const FMaterialInstance* p) {

 UTILS_NOINLINE
 void FEngine::destroy(Entity const e) {
-    mRenderableManager.destroy(e, getDriverApi());
+    mRenderableManager.destroy(e);
    mLightManager.destroy(e);
    mTransformManager.destroy(e);
    mCameraManager.destroy(*this, e);
@@ -456,7 +456,8 @@ public:
    }

    void prepare(backend::DriverApi& driver);
-    void gc();
+    void gcManagers();
+    void gcDeferredAsyncObjectDestruction();
    void submitFrame();

    using ShaderContent = utils::FixedCapacityVector<uint8_t>;
@@ -582,8 +583,6 @@ private:
    int loop();
    void flushCommandBuffer(backend::CommandBufferQueue& commandBufferQueue) const;

-    void gcDeferredAsyncObjectDestruction();
-
    template<typename T>
    bool isValid(const T* ptr, ResourceList<T> const& list) const;

@@ -176,7 +176,28 @@ FMaterial::FMaterial(FEngine& engine, const Builder& builder, MaterialDefinition
            driver.isWorkaroundNeeded(Workaround::DISABLE_DEPTH_PRECACHE_FOR_DEFAULT_MATERIAL);
    mDefaultMaterial = engine.getDefaultMaterial();

-    mPrograms.initializeForMaterial(engine, *this, processSpecializationConstants(builder));
+    FixedCapacityVector<Program::SpecializationConstant> specializationConstants =
+            processSpecializationConstants(builder);
+    mSpecializationConstants =
+            engine.getMaterialCache().getSpecializationConstantsInternPool().acquire(
+                    std::move(specializationConstants));
+
+    size_t cachedProgramsSize;
+    switch (mDefinition.materialDomain) {
+        case filament::MaterialDomain::SURFACE:
+            cachedProgramsSize = 1 << VARIANT_BITS;
+            break;
+        case filament::MaterialDomain::POST_PROCESS:
+            cachedProgramsSize = 1 << POST_PROCESS_VARIANT_BITS;
+            break;
+        case filament::MaterialDomain::COMPUTE:
+            cachedProgramsSize = 1;
+            break;
+    }
+    mCachedPrograms = FixedCapacityVector<backend::Handle<backend::HwProgram>>(cachedProgramsSize);
+
+    mDefinition.acquirePrograms(engine, mCachedPrograms.as_slice(), getMaterialParser(),
+            mSpecializationConstants, mIsDefaultMaterial);

 #if FILAMENT_ENABLE_MATDBG
    // Register the material with matdbg.
@@ -216,7 +237,11 @@ void FMaterial::terminate(FEngine& engine) {
    }
 #endif

-    mPrograms.terminate(engine);
+    mDefinition.releasePrograms(engine, mCachedPrograms.as_slice(),
+            getMaterialParser(), mSpecializationConstants, mIsDefaultMaterial);
+    engine.getMaterialCache().releaseMaterial(engine, mDefinition);
+    engine.getMaterialCache().getSpecializationConstantsInternPool().release(
+            mSpecializationConstants);
 }

 filament::DescriptorSetLayout const& FMaterial::getPerViewDescriptorSetLayout(
@@ -281,6 +306,22 @@ void FMaterial::compile(CompilerPriorityQueue const priority,
    }
 }

+Handle<HwProgram> FMaterial::prepareProgramSlow(DriverApi& driver, Variant const variant,
+        CompilerPriorityQueue const priorityQueue) const noexcept {
+    if (isSharedVariant(variant)) {
+        FMaterial const* defaultMaterial = mEngine.getDefaultMaterial();
+        FILAMENT_CHECK_PRECONDITION(defaultMaterial);
+        Handle<HwProgram> program = defaultMaterial->mCachedPrograms[variant.key];
+        if (program) {
+            return mCachedPrograms[variant.key] = program;
+        }
+        return mCachedPrograms[variant.key] =
+                defaultMaterial->prepareProgram(driver, variant, priorityQueue);
+    }
+    return mCachedPrograms[variant.key] = mDefinition.prepareProgram(mEngine, driver,
+            getMaterialParser(), getProgramSpecialization(variant), priorityQueue);
+}
+
 FMaterialInstance* FMaterial::createInstance(const char* name) const noexcept {
    if (mDefaultMaterialInstance) {
        // if we have a default instance, use it to create a new one
@@ -315,6 +356,14 @@ BufferInterfaceBlock::FieldInfo const* FMaterial::reflect(
    return mDefinition.uniformInterfaceBlock.getFieldInfo(name);
 }

+ProgramSpecialization FMaterial::getProgramSpecialization(Variant const variant) const noexcept {
+    return ProgramSpecialization {
+        .materialCrc32 = mDefinition.getMaterialParser().getCrc32(),
+        .variant = variant,
+        .specializationConstants = mSpecializationConstants,
+    };
+}
+
 size_t FMaterial::getParameters(ParameterInfo* parameters, size_t count) const noexcept {
    count = std::min(count, getParameterCount());

@@ -381,7 +430,8 @@ void FMaterial::updateActiveProgramsForMatdbg(Variant const variant) const noexc
 void FMaterial::applyPendingEdits() noexcept {
    const char* name = mDefinition.name.c_str();
    DLOG(INFO) << "Applying edits to " << (name ? name : "(untitled)");
-    mPrograms.clear(mEngine);
+    mDefinition.releasePrograms(mEngine, mCachedPrograms.as_slice(), getMaterialParser(),
+            mSpecializationConstants, mIsDefaultMaterial);
    latchPendingEdits();
 }

@@ -430,6 +480,26 @@ void FMaterial::onQueryCallback(void* userdata, VariantList* pActiveVariants) {

 #endif // FILAMENT_ENABLE_MATDBG

+void FMaterial::setSpecializationConstants(SpecializationConstantsBuilder&& builder) noexcept {
+    if (builder.mConstants.empty()) {
+        // Nothing was changed.
+        return;
+    }
+
+    auto& internPool = mEngine.getMaterialCache().getSpecializationConstantsInternPool();
+    MaterialParser const& materialParser = getMaterialParser();
+
+    // Release old resources...
+    mDefinition.releasePrograms(mEngine, mCachedPrograms.as_slice(), materialParser,
+            mSpecializationConstants, mIsDefaultMaterial);
+    internPool.release(mSpecializationConstants);
+
+    // Then acquire new ones.
+    mSpecializationConstants = internPool.acquire(std::move(builder.mConstants));
+    mDefinition.acquirePrograms(mEngine, mCachedPrograms.as_slice(), materialParser,
+            mSpecializationConstants, mIsDefaultMaterial);
+}
+
 FixedCapacityVector<Program::SpecializationConstant> FMaterial::processSpecializationConstants(
        Builder const& builder) {
    FixedCapacityVector<Program::SpecializationConstant> specializationConstants =
@@ -490,4 +560,12 @@ const char* FMaterial::getParameterTransformName(std::string_view samplerName) c
    return info->transformName.c_str();
 }

+template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<int32_t>(uint32_t id, int32_t value) noexcept;
+template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<float>(uint32_t id, float value) noexcept;
+template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<bool>(uint32_t id, bool value) noexcept;
+
+template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<int32_t>(std::string_view name, int32_t value) noexcept;
+template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<float>(std::string_view name, float value) noexcept;
+template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<bool>(std::string_view name, bool value) noexcept;
+
 } // namespace filament
@@ -18,7 +18,6 @@
 #define TNT_FILAMENT_DETAILS_MATERIAL_H

 #include "downcast.h"
-#include "LocalProgramCache.h"

 #include "details/MaterialInstance.h"

@@ -46,6 +45,7 @@
 #include <utils/Invocable.h>
 #include <utils/Mutex.h>

+#include <optional>
 #include <string_view>

 #include <stddef.h>
@@ -72,6 +72,62 @@ public:
        DefaultMaterialBuilder();
    };

+    // Used to change specialization constants at runtime internally by Filament.
+    //
+    // Call Material::getSpecializationConstantsBuilder() to return an instance of
+    // SpecializationConstantsBuilder. The lifetime of this object must not exceed the lifetime of
+    // the corresponding Material. After spec constants are changed with calls to set(), apply the
+    // changes to the original Material via setSpecializationConstants().
+    //
+    // It would be nice to reuse Builder for this, but Builder allocates an entire map and asserts
+    // if specified constants don't exist. In constrast, SpecializationConstantsBuilder is
+    // copy-on-write and allows non-existent specialization constants to be named.
+    class SpecializationConstantsBuilder {
+        friend class FMaterial;
+
+    public:
+        SpecializationConstantsBuilder(SpecializationConstantsBuilder const& rhs) = delete;
+        SpecializationConstantsBuilder& operator=(
+                SpecializationConstantsBuilder const& rhs) = delete;
+
+        SpecializationConstantsBuilder(SpecializationConstantsBuilder&& rhs) = default;
+        SpecializationConstantsBuilder& operator=(SpecializationConstantsBuilder&& rhs) = default;
+
+        template<typename T, typename = Builder::is_supported_constant_parameter_t<T>>
+        SpecializationConstantsBuilder& set(uint32_t id, T value) noexcept {
+            // Don't allocate if we can help it.
+            if (mConstants.empty()) {
+                if (std::get<T>(mDefaultConstants[id]) == value) {
+                    return *this;
+                }
+                mConstants = utils::FixedCapacityVector<backend::Program::SpecializationConstant>(
+                        mDefaultConstants);
+            }
+            mConstants[id] = value;
+            return *this;
+        }
+
+        template<typename T, typename = Builder::is_supported_constant_parameter_t<T>>
+        SpecializationConstantsBuilder& set(std::string_view name, T value) noexcept {
+            auto it = mDefinition->specializationConstantsNameToIndex.find(name);
+            if (it != mDefinition->specializationConstantsNameToIndex.cend()) {
+                set(it->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS, value);
+            }
+            return *this;
+        }
+
+    private:
+        SpecializationConstantsBuilder(MaterialDefinition const& definition,
+                utils::Slice<const backend::Program::SpecializationConstant> defaultConstants)
+                : mDefinition(&definition),
+                  mDefaultConstants(defaultConstants) {}
+
+        MaterialDefinition const* mDefinition;
+        utils::Slice<const backend::Program::SpecializationConstant> mDefaultConstants;
+        // Copy-on-write vector.
+        utils::FixedCapacityVector<backend::Program::SpecializationConstant> mConstants;
+    };
+
    void terminate(FEngine& engine);

    // return the uniform interface block for this material
@@ -131,7 +187,11 @@ public:
    backend::Handle<backend::HwProgram> prepareProgram(backend::DriverApi& driver,
            Variant const variant,
            backend::CompilerPriorityQueue const priorityQueue) const noexcept {
-        return mPrograms.prepareProgram(driver, variant, priorityQueue);
+        backend::Handle<backend::HwProgram> program = mCachedPrograms[variant.key];
+        if (UTILS_LIKELY(program)) {
+            return program;
+        }
+        return prepareProgramSlow(driver, variant, priorityQueue);
    }

    // getProgram returns the backend program for the material's given variant.
@@ -156,7 +216,9 @@ public:
 #if FILAMENT_ENABLE_MATDBG
        updateActiveProgramsForMatdbg(variant);
 #endif
-        return mPrograms.getProgram(variant);
+        backend::Handle<backend::HwProgram> program = mCachedPrograms[variant.key];
+        assert_invariant(program);
+        return program;
    }

    // MaterialInstance::use() binds descriptor sets before drawing. For shared variants,
@@ -213,8 +275,6 @@ public:
    float getSpecularAntiAliasingVariance() const noexcept { return mDefinition.specularAntiAliasingVariance; }
    float getSpecularAntiAliasingThreshold() const noexcept { return mDefinition.specularAntiAliasingThreshold; }

-    bool isDefaultMaterial() const noexcept { return mIsDefaultMaterial; }
-
    backend::descriptor_binding_t getSamplerBinding(
            std::string_view const& name) const;

@@ -237,7 +297,12 @@ public:

    uint32_t generateMaterialInstanceId() const noexcept { return mMaterialInstanceId++; }

-    LocalProgramCache& getPrograms() noexcept { return mPrograms; }
+    // Update specialization constants.
+    SpecializationConstantsBuilder getSpecializationConstantsBuilder() const noexcept {
+        return SpecializationConstantsBuilder(mDefinition, mSpecializationConstants);
+    }
+
+    void setSpecializationConstants(SpecializationConstantsBuilder&& builder) noexcept;

    uint8_t getPerViewLayoutIndex() const noexcept {
        return mDefinition.perViewLayoutIndex;
@@ -251,30 +316,6 @@ public:
        return mDefinition.source.c_str_safe();
    }

-    inline bool isSharedVariant(Variant const variant) const {
-        // HACK: The default material "should" have MNT | DEP, but then we'd have to compile it as a
-        // lit material, which would increase binary size. Perhaps we could specially compile it
-        // with this variant, but with the shader program cache in active development, the days of
-        // the default material are numbered anyway.
-        constexpr Variant::type_t vsmAndDep = Variant::MNT | Variant::DEP;
-        return mDefinition.materialDomain == MaterialDomain::SURFACE && !mIsDefaultMaterial &&
-               !mDefinition.hasCustomDepthShader && Variant::isValidDepthVariant(variant) &&
-               (variant.key & vsmAndDep) != vsmAndDep;
-    }
-
-    MaterialParser const& getMaterialParser() const noexcept {
-#if FILAMENT_ENABLE_MATDBG
-        if (mEditedMaterialParser) {
-            return *mEditedMaterialParser;
-        }
-#endif
-        return mDefinition.getMaterialParser();
-    }
-
-    MaterialDefinition const& getDefinition() const noexcept { return mDefinition; }
-
-    LocalProgramCache const& getPrograms() const noexcept { return mPrograms; }
-
 #if FILAMENT_ENABLE_MATDBG
    void applyPendingEdits() noexcept;

@@ -308,6 +349,21 @@ public:
 #endif

 private:
+    MaterialParser const& getMaterialParser() const noexcept {
+#if FILAMENT_ENABLE_MATDBG
+        if (mEditedMaterialParser) {
+            return *mEditedMaterialParser;
+        }
+#endif
+        return mDefinition.getMaterialParser();
+    }
+
+    ProgramSpecialization getProgramSpecialization(Variant const variant) const noexcept;
+
+    backend::Handle<backend::HwProgram> prepareProgramSlow(backend::DriverApi& driver,
+            Variant const variant,
+            backend::CompilerPriorityQueue const priorityQueue) const noexcept;
+
    utils::FixedCapacityVector<backend::Program::SpecializationConstant>
            processSpecializationConstants(Builder const& builder);

@@ -315,6 +371,24 @@ private:

    void createAndCacheProgram(backend::DriverApi& driver, backend::Program&& p, Variant variant) const noexcept;

+    backend::DescriptorSetLayout const& getPerViewDescriptorSetLayoutDescription(
+            Variant variant, bool useVsmDescriptorSetLayout) const noexcept;
+
+    backend::DescriptorSetLayout const& getDescriptorSetLayoutDescription(
+            Variant variant = {}) const noexcept;
+
+    bool isSharedVariant(Variant const variant) const {
+        // HACK: The default material "should" have MNT | DEP, but then we'd have to compile it as a
+        // lit material, which would increase binary size. Perhaps we could specially compile it
+        // with this variant, but with the shader program cache in active development, the days of
+        // the default material are numbered anyway.
+        constexpr Variant::type_t vsmAndDep = Variant::MNT | Variant::DEP;
+        return mDefinition.materialDomain == MaterialDomain::SURFACE && !mIsDefaultMaterial &&
+               !mDefinition.hasCustomDepthShader && Variant::isValidDepthVariant(variant) &&
+               (variant.key & vsmAndDep) != vsmAndDep;
+    }
+
+    mutable utils::FixedCapacityVector<backend::Handle<backend::HwProgram>> mCachedPrograms;
    MaterialDefinition const& mDefinition;

    bool mIsDefaultMaterial = false;
@@ -329,6 +403,9 @@ private:
    // reserve some space to construct the default material instance
    mutable FMaterialInstance* mDefaultMaterialInstance = nullptr;

+    // current specialization constants for the HwProgram
+    utils::Slice<const backend::Program::SpecializationConstant> mSpecializationConstants;
+
 #if FILAMENT_ENABLE_MATDBG
    matdbg::MaterialKey mDebuggerId;
    mutable utils::Mutex mActiveProgramsLock;
@@ -346,8 +423,6 @@ private:
    FEngine& mEngine;
    const uint32_t mMaterialId;
    mutable uint32_t mMaterialInstanceId = 0;
-
-    LocalProgramCache mPrograms;
 };


@@ -310,7 +310,13 @@ void FRenderer::skipFrame(uint64_t vsyncSteadyClockTimeNano) {
    engine.flush();     // flush command stream

    // Run GC
-    engine.gc();
+    JobSystem& js = engine.getJobSystem();
+    auto *rootJob = js.createJob();
+    js.run(jobs::createJob(js, rootJob, &FEngine::gcManagers, &engine)); // gc all component managers
+    if (engine.isAsynchronousModeEnabled()) {
+        js.run(jobs::createJob(js, rootJob, &FEngine::gcDeferredAsyncObjectDestruction, &engine));
+    }
+    js.runAndWait(rootJob);

    mFrameSkipper.frameSkipped();
 }
@@ -512,7 +518,13 @@ void FRenderer::endFrame() {
    engine.flush();     // flush command stream

    // Run GC
-    engine.gc();
+    JobSystem& js = engine.getJobSystem();
+    auto *rootJob = js.createJob();
+    js.run(jobs::createJob(js, rootJob, &FEngine::gcManagers, &engine)); // gc all component managers
+    if (engine.isAsynchronousModeEnabled()) {
+        js.run(jobs::createJob(js, rootJob, &FEngine::gcDeferredAsyncObjectDestruction, &engine));
+    }
+    js.runAndWait(rootJob);
 }

 void FRenderer::readPixels(
@@ -1362,7 +1374,7 @@ void FRenderer::renderJob(DriverApi& driver, RootArenaScope& rootArenaScope, FVi
    // Debug: CSM visualisation
    if (UTILS_UNLIKELY(engine.debug.shadowmap.visualize_cascades &&
                       view.hasShadowing() && view.hasDirectionalLighting())) {
-        input = ppm.debugShadowCascades(fg, view.getShadowMapManager(), input, depth);
+        input = ppm.debugShadowCascades(fg, input, depth);
    }

    // TODO: DoF should be applied here, before TAA -- but if we do this it'll result in a lot of
@@ -506,8 +506,6 @@ public:

    MaterialGlobals getMaterialGlobals() const { return mMaterialGlobals; }

-    ShadowMapManager const& getShadowMapManager() const noexcept { return *mShadowMapManager; }
-
 private:
    struct FPickingQuery : public PickingQuery {
    private:
@@ -18,24 +18,9 @@ material {
            precision: high
        },
        {
-            type : int,
+            type : uint,
            name : cascadeCount,
            precision: high
-        },
-        {
-            type : float2,
-            name : shadowAtlasResolution,
-            precision: high
-        },
-        {
-            type : mat4[4],
-            name : lightFromWorldMatrix,
-            precision: high
-        },
-        {
-            type : float4[4],
-            name : scissorNormalized,
-            precision: high
        }
    ],
    variables : [
@@ -74,10 +59,9 @@ fragment {
        return vec3(0.0, 0.0, 0.0);
    }

-    int getShadowCascade(highp vec4 position) {
-        highp float z = position.z / position.w;
-        ivec4 greaterZ = ivec4(greaterThan(materialParams.cascadeSplits, vec4(z)));
-        int cascadeCount = materialParams.cascadeCount;
+    int getShadowCascade(highp float z) {
+        ivec4 greaterZ = ivec4(greaterThan(frameUniforms.cascadeSplits, vec4(z)));
+        int cascadeCount = frameUniforms.cascades & 0xF;
        return clamp(greaterZ.x + greaterZ.y + greaterZ.z + greaterZ.w, 0, cascadeCount - 1);
    }

@@ -86,13 +70,11 @@ fragment {

        // depth from the depth buffer
        highp float depth = textureLod(materialParams_depth, variable_vertex.xy, 0.0).r;
-
-        highp vec2 clip = variable_vertex.xy * 2.0 - vec2(1.0);
-
        // convert to view-space (linear z).
-        highp vec4 p = mulMat4x4Float3(getViewFromClipMatrix(), vec3(clip, depth));
+        highp vec4 p = mulMat4x4Float3(getViewFromClipMatrix(), vec3(0, 0, depth));
+        highp float z = p.z / p.w;

-        color.rgb *= uintToColorDebug(getShadowCascade(p));
+        color.rgb *= uintToColorDebug(getShadowCascade(z));
        postProcess.color = color;
    }
 }
@@ -1,12 +1,12 @@
 Pod::Spec.new do |spec|
  spec.name = "Filament"
-  spec.version = "1.70.1"
+  spec.version = "1.69.5"
  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/v1.70.1/filament-v1.70.1-ios.tgz" }
+  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.69.5/filament-v1.69.5-ios.tgz" }

  spec.libraries = 'c++'

@@ -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 = 70;
+static constexpr size_t MATERIAL_VERSION = 69;

 // Those are the api levels that are used in the source material file (.mat)
 //
@@ -39,10 +39,6 @@ using namespace utils;

 namespace image {

-// Define maximum sane image dimensions to prevent integer overflows and memory exhaustion (DoS).
-constexpr size_t MAX_IMAGE_DIMENSION = 65536;
-constexpr size_t MAX_IMAGE_PIXELS = 16384 * 16384;
-
 const char HDRDecoder::sigRadiance[] = { '#', '?', 'R', 'A', 'D', 'I', 'A', 'N', 'C', 'E', 0xa };
 const char HDRDecoder::sigRGBE[]     = { '#', '?', 'R', 'G', 'B', 'E', 0xa };

@@ -63,16 +59,16 @@ HDRDecoder::HDRDecoder(std::istream& stream)
 HDRDecoder::~HDRDecoder() = default;

 LinearImage HDRDecoder::decode() {
-    float gamma = 1.0f;
-    float exposure = 1.0f;
-    char sy = 0, sx = 0;
+    float gamma;
+    float exposure;
+    char sy, sx;
    unsigned int height = 0, width = 0;
    {
-        char buf[1024] = {}; // Initialize buffer to prevent reading stack garbage
-        // Check mStream status in the loop condition to prevent an infinite loop on EOF or read error
-        while (mStream.getline(buf, sizeof(buf), 0xa)) {
+        char buf[1024];
+        do {
+            char format[128];
+            mStream.getline(buf, sizeof(buf), 0xa);
            if (buf[0] == '#') continue;
-            char format[128] = {0};
            sscanf(buf, "FORMAT=%127s", format); // NOLINT
            sscanf(buf, "GAMMA=%f", &gamma); // NOLINT
            sscanf(buf, "EXPOSURE=%f", &exposure); // NOLINT
@@ -80,29 +76,14 @@ LinearImage HDRDecoder::decode() {
                (sscanf(buf, "%cX %u %cY %u", &sx, &width, &sy, &height) == 4)) {  // NOLINT
                break;
            }
-        }
-
-        // Verify that dimensions were successfully found before proceeding
-        if (width == 0 || height == 0) {
-            slog.e << "Invalid or missing image dimensions" << io::endl;
-            return {};
-        }
+        } while (true);
    }
-
-    // Enforce sane limits on dimensions to prevent integer overflow and massive allocations
-    if (width > MAX_IMAGE_DIMENSION || height > MAX_IMAGE_DIMENSION ||
-        (size_t(width) * size_t(height) > MAX_IMAGE_PIXELS)) {
-        slog.e << "Image dimensions exceed safety limits" << io::endl;
-        return {};
-    }
-
    LinearImage image(width, height, 3);

    if (sx == '-') image = (image);
    if (sy == '+') image = verticalFlip(image);

    // Allocate memory to hold one row of decoded pixel data.
-    // width is now validated, so width * 4 will not integer overflow.
    std::unique_ptr<uint8_t[]> rgbe(new uint8_t[width * 4]);

    // First, test for non-RLE images.
@@ -146,37 +127,21 @@ LinearImage HDRDecoder::decode() {
                size_t num_bytes = 0;
                while (num_bytes < width) {
                    uint8_t rle_count;
-                    if (!mStream.read((char*) &rle_count, 1)) {
-                        slog.e << "Unexpected EOF during RLE read" << io::endl;
-                        return {};
-                    }
-
+                    mStream.read((char*) &rle_count, 1);
                    if (rle_count > 128) {
-                        size_t run_length = rle_count - 128;
-                        // Prevent Heap Buffer Overflow by checking bounds
-                        if (num_bytes + run_length > width) {
-                            slog.e << "RLE buffer overflow detected" << io::endl;
-                            return {};
-                        }
                        char v;
                        mStream.read(&v, 1);
-                        memset(d, v, run_length);
-                        d += run_length;
-                        num_bytes += run_length;
+                        memset(d, v, size_t(rle_count - 128));
+                        d += rle_count - 128;
+                        num_bytes += rle_count - 128;
                    } else {
                        if (rle_count == 0) {
                            slog.e << "run length is zero" << io::endl;
                            return {};
                        }
-                        size_t run_length = rle_count;
-                        // Prevent Heap Buffer Overflow by checking bounds
-                        if (num_bytes + run_length > width) {
-                            slog.e << "RLE buffer overflow detected" << io::endl;
-                            return {};
-                        }
-                        mStream.read(d, run_length);
-                        d += run_length;
-                        num_bytes += run_length;
+                        mStream.read(d, rle_count);
+                        d += rle_count;
+                        num_bytes += rle_count;
                    }
                }
            }
@@ -31,7 +31,6 @@ set(DIST_HDRS
        ${PUBLIC_HDR_DIR}/${TARGET}/Invocable.h
        ${PUBLIC_HDR_DIR}/${TARGET}/Log.h
        ${PUBLIC_HDR_DIR}/${TARGET}/Logger.h
-        ${PUBLIC_HDR_DIR}/${TARGET}/LruCache.h
        ${PUBLIC_HDR_DIR}/${TARGET}/memalign.h
        ${PUBLIC_HDR_DIR}/${TARGET}/MonotonicRingMap.h
        ${PUBLIC_HDR_DIR}/${TARGET}/Mutex.h
@@ -182,7 +181,6 @@ if (FILAMENT_BUILD_TESTING)
            test/test_FixedCircularBuffer.cpp
            test/test_Hash.cpp
            test/test_InternPool.cpp
-            test/test_LruCache.cpp
            test/test_JobSystem.cpp
            test/test_MonotonicRingMap.cpp
            test/test_QuadTreeArray.cpp
@@ -1,322 +0,0 @@
-/*
- * Copyright (C) 2026 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_UTILS_LRUCACHE_H
-#define TNT_UTILS_LRUCACHE_H
-
-#include <utils/Allocator.h>
-#include <utils/compiler.h>
-
-#include <tsl/robin_map.h>
-
-#include <cassert>
-#include <cstddef>
-#include <functional>
-#include <optional>
-#include <utility>
-
-namespace utils {
-
-/**
- * A fixed-capacity Least Recently Used (LRU) cache.
- *
- * This container allows O(1) access, insertion, and eviction. It uses an Arena with an
- * ObjectPoolAllocator to store the nodes of a doubly-linked list representing the LRU order,
- * and a robin_map for fast key lookups.
- *
- * This class is not thread-safe. Additionally, the pop() and put() methods have special concerns
- * with regards to object lifetimes.
- *
- * @tparam Key The type of the keys.
- * @tparam T The type of the values.
- * @tparam Hash The hasher for the keys.
- */
-template<typename Key,
-         typename T,
-         typename Hash = std::hash<Key>,
-         typename KeyEqual = std::equal_to<Key>>
-class LruCache {
-    // If Key is bigger than a pointer, robin map keys are references to Node::key. Otherwise,
-    // they're full copies.
-    using MapKey = std::conditional_t<(sizeof(Key) > sizeof(void*)),
-            std::reference_wrapper<const Key>, Key>;
-
-public:
-    using size_type = uint32_t;
-
-    /**
-     * Creates an LRU cache with the specified capacity.
-     *
-     * @param name Name to identify the backing Arena allocator.
-     * @param capacity The maximum number of elements the cache can hold.
-     */
-    LruCache(const char* UTILS_NONNULL name, size_type capacity)
-            : mCapacity(capacity),
-              // HACK: FreeList cannot handle a capacity of 0.
-              mArena(name, (capacity ? capacity : 1) * sizeof(Node)),
-              mHead(nullptr),
-              mTail(nullptr) {
-        mMap.reserve(capacity);
-    }
-
-    ~LruCache() {
-        clear();
-    }
-
-    LruCache(const LruCache&) = delete;
-    LruCache& operator=(const LruCache&) = delete;
-    LruCache(LruCache&&) = default;
-    LruCache& operator=(LruCache&&) = default;
-
-    /**
-     * Returns a reference to the value associated with the key.
-     *
-     * Additionally moves the key/value pair to the front of the most recently used (MRU) list.
-     *
-     * @param key The key to look up.
-     * @param hash The precomputed hash of the key.
-     * @return Pointer to the value, if exists; otherwise, nullptr.
-     */
-    T* UTILS_NULLABLE get(Key const& key, size_t hash) {
-        auto it = mMap.find(key, hash);
-        if (UTILS_UNLIKELY(it == mMap.end())) {
-            return nullptr;
-        }
-
-        Node* node = it->second;
-        assert(node != nullptr);
-
-        moveToFront(node);
-
-        return &node->value;
-    }
-
-    inline T* UTILS_NULLABLE get(Key const& key) {
-        return get(key, Hash{}(key));
-    }
-
-    /**
-     * Moves the value out of the cache, if it exists.
-     *
-     * Because this moves the actual value, any previous references to this object returned by a
-     * call to get() or put() is invalidated by this call.
-     *
-     * @param key The key to look up.
-     * @param hash The precomputed hash of the key.
-     */
-    std::optional<T> pop(Key const& key, size_t hash) {
-        auto it = mMap.find(key, hash);
-        if (UTILS_UNLIKELY(it == mMap.end())) {
-            return std::nullopt;
-        }
-
-        Node* node = it->second;
-        assert(node != nullptr);
-
-        if (node == mHead && node == mTail) {
-            mHead = nullptr;
-            mTail = nullptr;
-        } else if (node == mHead) {
-            assert(node->next != nullptr);
-            mHead = node->next;
-            mHead->prev = nullptr;
-        } else if (node == mTail) {
-            assert(node->prev != nullptr);
-            mTail = node->prev;
-            mTail->next = nullptr;
-        } else {
-            assert(node->prev != nullptr);
-            assert(node->next != nullptr);
-            node->prev->next = node->next;
-            node->next->prev = node->prev;
-        }
-
-        T r = std::move(node->value);
-        mMap.erase(it);
-        mArena.destroy(node);
-        return r;
-    }
-
-    inline std::optional<T> pop(Key const& key) {
-        return pop(key, Hash{}(key));
-    }
-
-    /**
-     * Inserts a new entry or updates an existing one.
-     *
-     * Prepends it to the front of the most recently used (MRU) list. Potentially evicts the
-     * least-recently used (LRU) key/value pair by calling the releaser function.
-     *
-     * Due to the evicting nature of this function, any pointers to any objects within the LRU cache
-     * that had been previously returned by a call to get() or another call to put() can be
-     * considered INVALID and any access to them will result in undefined behavior.
-     *
-     * An evicted node is removed from the map before the releaser callback is called. Calls to
-     * get() or pop() in the body of the releaser for the item in question will fail.
-     *
-     * @tparam F Callable type accepting T&&.
-     * @param key The key to insert or update.
-     * @param t The value to insert or update.
-     * @param hash The precomputed hash of the key.
-     * @param releaser Function called with the evicted value (T&&) if eviction occurs.
-     * @return Reference to the inserted or updated value.
-     */
-    template<typename F>
-    T& put(Key key, T value, size_t hash, F releaser) {
-        // Assert that we have capacity to store at least one item
-        assert(mCapacity > 0);
-
-        auto it = mMap.find(key, hash);
-        if (it != mMap.end()) {
-            Node* node = it->second;
-            assert(node != nullptr);
-
-            moveToFront(node);
-
-            node->value = std::move(value);
-            return node->value;
-        }
-
-        if (UTILS_LIKELY(mMap.size() >= mCapacity)) {
-            evict(releaser);
-        }
-
-        // Create new node at front of list.
-        Node* node = mArena.template make<Node>(std::move(key), std::move(value), mHead);
-        assert(node != nullptr);
-        attach(node);
-
-        mMap.emplace(node->key, node);
-
-        return node->value;
-    }
-
-    template<typename F>
-    inline T& put(Key key, T value, F releaser) {
-        return put(std::move(key), std::move(value), Hash{}(key), std::move(releaser));
-    }
-
-    /**
-     * Clear the cache, calling releaser on each item evicted.
-     *
-     * @param releaser Function called with the evicted value (T&&)
-     */
-    template<typename F>
-    void clear(F releaser) {
-        // Clear map first, just in case it tries to dereference any keys.
-        mMap.clear();
-        // Destroy everything in the arena.
-        Node* node = mHead;
-        while (UTILS_LIKELY(node)) {
-            Node* next = node->next;
-            releaser(std::move(node->value));
-            mArena.destroy(node);
-            node = next;
-        }
-        mHead = nullptr;
-        mTail = nullptr;
-    }
-
-    void clear() {
-        clear([](T&&){});
-    }
-
-    /** Returns the number of elements in the cache. */
-    size_type size() const noexcept { return mMap.size(); }
-
-    /** Returns the capacity of the cache. */
-    size_type capacity() const noexcept { return mCapacity; }
-
-private:
-    struct Node {
-        Key key;
-        T value;
-        Node* UTILS_NULLABLE prev;
-        Node* UTILS_NULLABLE next;
-
-        Node(Key&& k, T&& v, Node* UTILS_NULLABLE next)
-                : key(std::move(k)),
-                  value(std::move(v)),
-                  prev(nullptr),
-                  next(next) {}
-    };
-
-    void moveToFront(Node* UTILS_NONNULL node) {
-        if (node == mHead) {
-            return;
-        }
-
-        // First, detach from list.
-        if (UTILS_UNLIKELY(node == mTail)) {
-            assert(node->prev != nullptr);
-            mTail = node->prev;
-            mTail->next = nullptr;
-        } else {
-            assert(node->prev != nullptr);
-            assert(node->next != nullptr);
-            node->prev->next = node->next;
-            node->next->prev = node->prev;
-        }
-
-        // Then attach to head.
-        node->prev = nullptr;
-        node->next = mHead;
-        attach(node);
-    }
-
-    // Attach node to front.
-    void attach(Node* UTILS_NONNULL node) {
-        if (UTILS_LIKELY(mHead != nullptr)) {
-            mHead->prev = node;
-        } else {
-            assert(mTail == nullptr);
-            mTail = node;
-        }
-        mHead = node;
-    }
-
-    // Evict the least recently used node.
-    template<typename F>
-    void evict(F releaser) {
-        Node* node = mTail;
-        assert(node != nullptr);
-
-        if (UTILS_UNLIKELY(node == mHead)) {
-            mTail = nullptr;
-            mHead = nullptr;
-        } else {
-            // Move new tail node to mTail.
-            assert(node->prev != nullptr);
-            mTail = node->prev;
-            mTail->next = nullptr;
-        }
-
-        // Finally free the node.
-        mMap.erase(node->key);
-        releaser(std::move(node->value));
-        mArena.destroy(node);
-    }
-
-    size_type mCapacity;
-    utils::Arena<utils::ObjectPoolAllocator<Node>, utils::LockingPolicy::NoLock> mArena;
-    tsl::robin_map<MapKey, Node* UTILS_NONNULL, Hash, KeyEqual> mMap;
-    Node* UTILS_NULLABLE mHead;
-    Node* UTILS_NULLABLE mTail;
-};
-
-} // namespace utils
-
-#endif // TNT_UTILS_LRUCACHE_H
@@ -19,7 +19,6 @@
 #include <utils/Panic.h>
 #include <utils/compiler.h>
 #include <utils/debug.h>
-#include <utils/LruCache.h>

 #include <tsl/robin_map.h>

@@ -76,6 +75,8 @@ class RefCountedMap {
        T value;
    };

+    using Map = tsl::robin_map<Key, Entry, Hash>;
+
    static constexpr TValue& deref(T& a) {
        if constexpr (refcountedmap::IsPointer<T>) {
            return *a;
@@ -98,9 +99,6 @@ class RefCountedMap {
            "Attempted to get missing value";

 public:
-    explicit RefCountedMap(const char* UTILS_NONNULL name = "RefCountedMap", size_t lruCapacity = 0)
-            : mLruCache(name, lruCapacity) {}
-
    /** Acquire and return a value by key, initializing it with F if it doesn't exist.
     *
     * If F returns NullValue{}(), this indicates a failure to create the object. If T is a value
@@ -113,16 +111,12 @@ public:
            it.value().referenceCount++;
            return &deref(it.value().value);
        }
-
-        if (std::optional<T> lruValue = mLruCache.pop(key, hash)) {
-            return &insert(key, std::move(*lruValue));
-        }
-
        T r = factory();
        if (r == NullValue{}()) {
            return nullptr;
        }
-        return &insert(key, std::move(r));
+        // TODO: how to use above computed hash here?
+        return &deref(mMap.insert({ key, Entry{ 1, std::move(r) } }).first.value().value);
    }

    template<typename F>
@@ -144,11 +138,6 @@ public:
            it.value().referenceCount++;
            return &deref(it.value().value);
        }
-
-        if (std::optional<T> lruValue = mLruCache.pop(key, hash)) {
-            return &insert(key, std::move(*lruValue));
-        }
-
        // TODO: how to use above computed hash here?
        mMap.insert({ key, Entry{ 1, NullValue{}() } });
        return nullptr;
@@ -167,14 +156,8 @@ public:
        auto it = mMap.find(key, hash);
        FILAMENT_CHECK_PRECONDITION(it != mMap.end()) << MISSING_ENTRY_ERROR_STRING;
        if (--it.value().referenceCount == 0) {
-            if (it.value().value != NullValue{}()) {
-                if (mLruCache.capacity() > 0) {
-                    mLruCache.put(key, std::move(it.value().value), hash, [&releaser](T&& v) {
-                        releaser(deref(v));
-                    });
-                } else {
-                    releaser(deref(it.value().value));
-                }
+            if (it.value().value != NullValue{}()){
+                releaser(deref(it.value().value));
            }
            // TODO: change to erase_fast
            mMap.erase(it);
@@ -194,9 +177,6 @@ public:
        auto it = mMap.find(key, hash);
        FILAMENT_CHECK_PRECONDITION(it != mMap.end()) << MISSING_ENTRY_ERROR_STRING;
        if (--it.value().referenceCount == 0) {
-            if (mLruCache.capacity() > 0) {
-                mLruCache.put(key, std::move(it.value().value), hash, [](T&&){});
-            }
            // TODO: change to erase_fast
            mMap.erase(it);
        }
@@ -237,7 +217,7 @@ public:
     * Panics if no entry found in map.
     */
    TValue& get(KeyRef key, size_t hash) {
-        auto it = mMap.find(key, hash);
+        auto it = mMap.find(key);
        FILAMENT_CHECK_PRECONDITION(it != mMap.end()) << MISSING_ENTRY_ERROR_STRING;
        FILAMENT_CHECK_PRECONDITION(it.value().value != NullValue{}())
                << MISSING_VALUE_ERROR_STRING;
@@ -256,33 +236,13 @@ public:

    inline TValue const& get(KeyRef key) const noexcept { return get(key, Hash{}(key)); }

-    /** Clear the LRU cache, calling releaser on each item evicted. */
-    template<typename F>
-    void clearLruCache(F releaser) {
-        mLruCache.clear([&releaser](T&& v) {
-            releaser(deref(v));
-        });
-    }
-
-    /** Clear the LRU cache without calling any releasers. */
-    void clearLruCache() {
-        mLruCache.clear();
-    }
-
    /** Returns true if the map is empty. */
    inline bool empty() const noexcept { return mMap.empty(); }

 private:
-    tsl::robin_map<Key, Entry, Hash> mMap;
-    utils::LruCache<Key, T, Hash> mLruCache;
-
-    TValue& insert(KeyRef key, T value) {
-        // TODO: how to use computed hash here?
-        auto it = mMap.insert({ key, Entry{ 1, std::move(value) } });
-        return deref(it.first.value().value);
-    }
+    Map mMap;
 };

-} // utils
+}

 #endif  // TNT_UTILS_REFCOUNTEDMAP_H
@@ -87,7 +87,7 @@ FreeList::Node* FreeList::init(void* begin, void* end,
    void* const p = pointermath::align(begin, alignment, extra);
    void* const n = pointermath::align(pointermath::add(p, elementSize), alignment, extra);
    assert_invariant(p >= begin && p < end);
-    assert_invariant(n >= begin && n <= end && n > p);
+    assert_invariant(n >= begin && n < end && n > p);

    const size_t d = uintptr_t(n) - uintptr_t(p);
    const size_t num = (uintptr_t(end) - uintptr_t(p)) / d;
@@ -128,7 +128,7 @@ AtomicFreeList::AtomicFreeList(void* begin, void* end,
    void* const p = pointermath::align(begin, alignment, extra);
    void* const n = pointermath::align(pointermath::add(p, elementSize), alignment, extra);
    assert_invariant(p >= begin && p < end);
-    assert_invariant(n >= begin && n <= end && n > p);
+    assert_invariant(n >= begin && n < end && n > p);

    const size_t d = uintptr_t(n) - uintptr_t(p);
    const size_t num = (uintptr_t(end) - uintptr_t(p)) / d;
@@ -1,268 +0,0 @@
-/*
- * Copyright (C) 2026 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 <utils/LruCache.h>
-
-#include <gtest/gtest.h>
-
-#include <string>
-#include <vector>
-
-using namespace utils;
-
-TEST(LruCacheTest, BasicPutGet) {
-    LruCache<int, std::string> cache("LruCacheTest", 3);
-
-    cache.put(1, "one", [](std::string&&) {});
-    cache.put(2, "two", [](std::string&&) {});
-
-    auto* v1 = cache.get(1);
-    ASSERT_NE(v1, nullptr);
-    EXPECT_EQ(*v1, "one");
-
-    auto* v2 = cache.get(2);
-    ASSERT_NE(v2, nullptr);
-    EXPECT_EQ(*v2, "two");
-
-    EXPECT_EQ(cache.size(), 2);
-}
-
-TEST(LruCacheTest, MissingKey) {
-    LruCache<int, int> cache("LruCacheTest", 3);
-    EXPECT_EQ(cache.get(999), nullptr);
-}
-
-TEST(LruCacheTest, Eviction) {
-    LruCache<int, int> cache("LruCacheTest", 3);
-    std::vector<int> evicted;
-    auto releaser = [&](int&& v) { evicted.push_back(v); };
-
-    cache.put(1, 10, releaser);
-    cache.put(2, 20, releaser);
-    cache.put(3, 30, releaser);
-
-    // Cache: 3(MRU), 2, 1(LRU)
-    EXPECT_EQ(cache.size(), 3);
-    EXPECT_TRUE(evicted.empty());
-
-    cache.put(4, 40, releaser);
-    // Evicted 1. Cache: 4(MRU), 3, 2
-    EXPECT_EQ(cache.size(), 3);
-    ASSERT_EQ(evicted.size(), 1);
-    EXPECT_EQ(evicted[0], 10);
-
-    // Verify contents
-    EXPECT_NE(cache.get(2), nullptr);
-    EXPECT_NE(cache.get(3), nullptr);
-    EXPECT_NE(cache.get(4), nullptr);
-    EXPECT_EQ(cache.get(1), nullptr);
-
-    // Accessing 2 makes it MRU. Cache: 2(MRU), 4, 3(LRU)
-    cache.get(2);
-
-    cache.put(5, 50, releaser);
-    // Should evict 3.
-    ASSERT_EQ(evicted.size(), 2);
-    EXPECT_EQ(evicted[1], 30);
-}
-
-TEST(LruCacheTest, UpdatePromotesToMRU) {
-    LruCache<int, int> cache("LruCacheTest", 3);
-    auto releaser = [](int&&) {};
-
-    cache.put(1, 10, releaser);
-    cache.put(2, 20, releaser);
-    cache.put(3, 30, releaser);
-    // Order: 3, 2, 1
-
-    cache.put(1, 11, releaser);
-    // Order: 1, 3, 2
-
-    cache.put(4, 40, releaser);
-    // Order: 4, 1, 3, 2 (evicted)
-
-    // Check if 1 exists
-    auto* v1 = cache.get(1);
-    ASSERT_NE(v1, nullptr);
-    EXPECT_EQ(*v1, 11);
-}
-
-TEST(LruCacheTest, GetPromotesToMRU) {
-    LruCache<int, int> cache("LruCacheTest", 3);
-    std::vector<int> evicted;
-    auto releaser = [&](int&& v) { evicted.push_back(v); };
-
-    cache.put(1, 10, releaser);
-    cache.put(2, 20, releaser);
-    cache.put(3, 30, releaser);
-    // 3, 2, 1
-
-    cache.get(1);
-    // 1, 3, 2
-
-    cache.put(4, 40, releaser);
-    // Evicts 2.
-    ASSERT_EQ(evicted.size(), 1);
-    EXPECT_EQ(evicted[0], 20);
-}
-
-TEST(LruCacheTest, MoveToFrontFromTail) {
-    // This specific test targets the potential bug where moving from tail doesn't update tail.
-    LruCache<int, int> cache("LruCacheTest", 3);
-    std::vector<int> evicted;
-    auto releaser = [&](int&& v) { evicted.push_back(v); };
-
-    cache.put(1, 10, releaser);
-    cache.put(2, 20, releaser);
-    cache.put(3, 30, releaser);
-    // State: 3 (MRU) -> 2 -> 1 (LRU, Tail)
-
-    // Move 1 to front.
-    cache.get(1);
-    // Expected State: 1 (MRU) -> 3 -> 2 (LRU, Tail)
-
-    // Add 4. Should evict 2.
-    cache.put(4, 40, releaser);
-
-    ASSERT_EQ(evicted.size(), 1);
-    EXPECT_EQ(evicted[0], 20);
-}
-
-TEST(LruCacheTest, MoveToFrontFromMiddle) {
-    LruCache<int, int> cache("LruCacheTest", 3);
-    std::vector<int> evicted;
-    auto releaser = [&](int&& v) { evicted.push_back(v); };
-
-    cache.put(1, 10, releaser);
-    cache.put(2, 20, releaser);
-    cache.put(3, 30, releaser);
-    // 3, 2, 1
-
-    cache.get(2);
-    // 2, 3, 1
-
-    cache.put(4, 40, releaser);
-    // Evicts 1.
-    ASSERT_EQ(evicted.size(), 1);
-    EXPECT_EQ(evicted[0], 10);
-}
-
-TEST(LruCacheTest, CapacityOne) {
-    LruCache<int, int> cache("LruCacheTest", 1);
-    std::vector<int> evicted;
-    auto releaser = [&](int&& v) { evicted.push_back(v); };
-
-    cache.put(1, 10, releaser);
-    EXPECT_NE(cache.get(1), nullptr);
-    EXPECT_EQ(*cache.get(1), 10);
-
-    cache.put(2, 20, releaser);
-    ASSERT_EQ(evicted.size(), 1);
-    EXPECT_EQ(evicted[0], 10);
-    EXPECT_NE(cache.get(2), nullptr);
-    EXPECT_EQ(*cache.get(2), 20);
-}
-
-#ifdef GTEST_HAS_DEATH_TEST
-TEST(LruCacheTest, CapacityZero) {
-    LruCache<int, int> cache("LruCacheTest", 0);
-    EXPECT_EQ(cache.capacity(), 0);
-    EXPECT_EQ(cache.size(), 0);
-    EXPECT_DEATH(cache.put(1, 1, [](int&&){}), "");
-}
-#endif
-
-TEST(LruCacheTest, Clear) {
-    LruCache<int, int> cache("LruCacheTest", 3);
-    std::vector<int> evicted;
-    auto releaser = [&](int&& v) { evicted.push_back(v); };
-
-    cache.put(1, 10, releaser);
-    cache.put(2, 20, releaser);
-
-    EXPECT_EQ(cache.size(), 2);
-
-    cache.clear(releaser);
-
-    EXPECT_EQ(cache.size(), 0);
-    ASSERT_EQ(evicted.size(), 2);
-    // Implementation traverses mHead -> mTail (MRU to LRU).
-    EXPECT_EQ(evicted[0], 20); // MRU
-    EXPECT_EQ(evicted[1], 10); // LRU
-
-    // We should now be able to refill the cache.
-    cache.put(3, 30, releaser);
-    cache.put(4, 40, releaser);
-
-    EXPECT_EQ(cache.size(), 2);
-    EXPECT_NE(cache.get(3), nullptr);
-    EXPECT_NE(cache.get(4), nullptr);
-}
-
-TEST(LruCacheTest, Pop) {
-    LruCache<int, int> cache("LruCacheTest", 3);
-    auto releaser = [](int&&) {};
-
-    cache.put(1, 10, releaser);
-    cache.put(2, 20, releaser);
-
-    std::optional<int> val1 = cache.pop(1);
-    EXPECT_TRUE(val1.has_value());
-    EXPECT_EQ(*val1, 10);
-    EXPECT_EQ(cache.size(), 1);
-    EXPECT_EQ(cache.get(1), nullptr);
-    EXPECT_NE(cache.get(2), nullptr);
-
-    std::optional<int> val2 = cache.pop(2);
-    EXPECT_TRUE(val2.has_value());
-    EXPECT_EQ(*val2, 20);
-    EXPECT_EQ(cache.size(), 0);
-}
-
-/* Large key tests. */
-
-struct LargeKey {
-    long long a, b;
-    bool operator==(LargeKey const& other) const {
-        return a == other.a && b == other.b;
-    }
-};
-
-struct LargeKeyHash {
-    size_t operator()(LargeKey const& k) const {
-        return std::hash<long long>{}(k.a) ^ std::hash<long long>{}(k.b);
-    }
-};
-
-TEST(LruCacheTest, LargeKey) {
-    static_assert(sizeof(LargeKey) > sizeof(void*), "LargeKey must be larger than pointer");
-    LruCache<LargeKey, int, LargeKeyHash> cache("LruCacheTest", 2);
-
-    LargeKey k1{1, 1};
-    LargeKey k2{2, 2};
-    LargeKey k3{3, 3};
-
-    cache.put(k1, 100, [](int&&){});
-    cache.put(k2, 200, [](int&&){});
-
-    EXPECT_NE(cache.get(k1), nullptr);
-    EXPECT_EQ(*cache.get(k1), 100);
-
-    // Evict k2 (LRU because k1 was accessed)
-    cache.put(k3, 300, [](int&&){});
-
-    EXPECT_NE(cache.get(k3), nullptr); // k3 present
-}
--- a/Show More
+++ b/Show More