wip: shader compilation benchmark

2024-01-16 17:03:41 -08:00
380 changed files with 4644 additions and 11690 deletions
@@ -10,7 +10,7 @@ on:
 jobs:
  build-android:
    name: build-android
-    runs-on: macos-14
+    runs-on: macos-latest

    steps:
      - uses: actions/checkout@v3.3.0
@@ -13,7 +13,7 @@ on:
 jobs:
  cocoapods-deploy:
    name: cocoapods-deploy
-    runs-on: macos-14
+    runs-on: macos-latest
    steps:
      - name: Check out iOS/CocoaPods directory
        uses: Bhacaz/checkout-files@49fc3050859046bf4f4873678d46099985640e89
@@ -10,7 +10,7 @@ on:
 jobs:
  build-ios:
    name: build-ios
-    runs-on: macos-14
+    runs-on: macos-latest

    steps:
      - uses: actions/checkout@v3.3.0
@@ -10,7 +10,7 @@ on:
 jobs:
  build-mac:
    name: build-mac
-    runs-on: macos-14
+    runs-on: macos-latest

    steps:
      - uses: actions/checkout@v3.3.0
@@ -11,7 +11,7 @@ on:
 jobs:
  npm-deploy:
    name: npm-deploy
-    runs-on: macos-14
+    runs-on: macos-latest
    steps:
      - uses: actions/checkout@v3.3.0
        with:
@@ -15,7 +15,7 @@ jobs:

    strategy:
      matrix:
-        os: [macos-14, ubuntu-22.04]
+        os: [macos-latest, ubuntu-22.04]

    steps:
      - uses: actions/checkout@v3.3.0
@@ -40,7 +40,7 @@ jobs:

  build-android:
    name: build-android
-    runs-on: macos-14
+    runs-on: macos-latest

    steps:
      - uses: actions/checkout@v3.3.0
@@ -54,7 +54,7 @@ jobs:

  build-ios:
    name: build-iOS
-    runs-on: macos-14
+    runs-on: macos-latest

    steps:
      - uses: actions/checkout@v3.3.0
@@ -67,7 +67,7 @@ jobs:

  build-web:
    name: build-web
-    runs-on: macos-14
+    runs-on: macos-latest

    steps:
      - uses: actions/checkout@v3.3.0
@@ -31,7 +31,7 @@ jobs:

    strategy:
      matrix:
-        os: [macos-14, ubuntu-22.04]
+        os: [macos-latest, ubuntu-22.04]

    steps:
      - name: Decide Git ref
@@ -65,7 +65,7 @@ jobs:

  build-web:
    name: build-web
-    runs-on: macos-14
+    runs-on: macos-latest
    if: github.event_name == 'release' || github.event.inputs.platform == 'web'

    steps:
@@ -98,7 +98,7 @@ jobs:

  build-android:
    name: build-android
-    runs-on: macos-14
+    runs-on: macos-latest
    if: github.event_name == 'release' || github.event.inputs.platform == 'android'

    steps:
@@ -129,7 +129,7 @@ jobs:
      - name: Sign sample-gltf-viewer
        run: |
          echo "${APK_KEYSTORE_BASE64}" > filament.jks.base64
-          base64 --decode -i filament.jks.base64 > filament.jks
+          base64 --decode filament.jks.base64 > filament.jks
          BUILD_TOOLS_VERSION=$(ls ${ANDROID_HOME}/build-tools | sort -V | tail -n 1)
          APKSIGNER=${ANDROID_HOME}/build-tools/${BUILD_TOOLS_VERSION}/apksigner
          IN_FILE="out/sample-gltf-viewer-release.apk"
@@ -152,7 +152,7 @@ jobs:

  build-ios:
    name: build-ios
-    runs-on: macos-14
+    runs-on: macos-latest
    if: github.event_name == 'release' || github.event.inputs.platform == 'ios'

    steps:
@@ -205,7 +205,7 @@ jobs:
          TAG: ${{ steps.git_ref.outputs.tag }}
        run: |
          build\windows\build-github.bat release
-          move out\filament-windows.tgz out\filament-%TAG%-windows.tgz
+          move out\filament-windows.tgz out\filament-$Env:TAG-windows.tgz
        shell: cmd
      - uses: actions/github-script@v6
        env:
@@ -10,7 +10,7 @@ on:
 jobs:
  build-web:
    name: build-web
-    runs-on: macos-14
+    runs-on: macos-latest

    steps:
      - uses: actions/checkout@v3.3.0
@@ -45,8 +45,6 @@ option(FILAMENT_ENABLE_TSAN "Enable Thread Sanitizer" OFF)

 option(FILAMENT_ENABLE_FEATURE_LEVEL_0 "Enable Feature Level 0" ON)

-option(FILAMENT_ENABLE_MULTIVIEW "Enable multiview for Filament" OFF)
-
 set(FILAMENT_NDK_VERSION "" CACHE STRING
    "Android NDK version or version prefix to be used when building for Android."
 )
@@ -533,21 +531,6 @@ else()
    option(FILAMENT_DISABLE_MATOPT "Disable material optimizations" ON)
 endif()

-# This only affects the prebuilt shader files in gltfio and samples, not filament library.
-# The value can be either "instanced" or "multiview".
-set(FILAMENT_SAMPLES_STEREO_TYPE "instanced" CACHE STRING
-    "Stereoscopic type that shader files in gltfio and samples are built for."
-)
-string(TOLOWER "${FILAMENT_SAMPLES_STEREO_TYPE}" FILAMENT_SAMPLES_STEREO_TYPE)
-if (NOT FILAMENT_SAMPLES_STEREO_TYPE STREQUAL "instanced" AND NOT FILAMENT_SAMPLES_STEREO_TYPE STREQUAL "multiview")
-    message(FATAL_ERROR "Invalid stereo type: \"${FILAMENT_SAMPLES_STEREO_TYPE}\" choose either \"instanced\" or \"multiview\" ")
-endif ()
-
-# Compiling samples for multiview implies enabling multiview feature as well.
-if (FILAMENT_SAMPLES_STEREO_TYPE STREQUAL "multiview")
-    set(FILAMENT_ENABLE_MULTIVIEW ON)
-endif ()
-
 # ==================================================================================================
 # Material compilation flags
 # ==================================================================================================
@@ -651,9 +634,9 @@ function(combine_static_libs TARGET OUTPUT DEPS)
    # Loop through the dependent libraries and query their location on disk.
    set(DEPS_FILES )
    foreach(DEPENDENCY ${DEPS})
-        if (TARGET ${DEPENDENCY})
+        if(TARGET ${DEPENDENCY})
            get_property(dep_type TARGET ${DEPENDENCY} PROPERTY TYPE)
-            if (dep_type STREQUAL "STATIC_LIBRARY")
+            if(dep_type STREQUAL "STATIC_LIBRARY")
                list(APPEND DEPS_FILES "$<TARGET_FILE:${DEPENDENCY}>")
            endif()
        endif()
@@ -723,7 +706,7 @@ function(get_resgen_vars ARCHIVE_DIR ARCHIVE_NAME)
        set(RESGEN_OUTPUTS "${OUTPUTS}" PARENT_SCOPE)
        set(RESGEN_FLAGS -qx ${ARCHIVE_DIR} -p ${ARCHIVE_NAME} PARENT_SCOPE)
        set(RESGEN_SOURCE "${ARCHIVE_DIR}/${ARCHIVE_NAME}${ASM_SUFFIX}.S" PARENT_SCOPE)
-        set(RESGEN_SOURCE_FLAGS "-I'${ARCHIVE_DIR}' ${ASM_ARCH_FLAG}" PARENT_SCOPE)
+        set(RESGEN_SOURCE_FLAGS "-I${ARCHIVE_DIR} ${ASM_ARCH_FLAG}" PARENT_SCOPE)
    endif()
 endfunction()

@@ -738,6 +721,7 @@ add_subdirectory(${LIBRARIES}/filabridge)
 add_subdirectory(${LIBRARIES}/filaflat)
 add_subdirectory(${LIBRARIES}/filagui)
 add_subdirectory(${LIBRARIES}/filameshio)
+add_subdirectory(${LIBRARIES}/geometry)
 add_subdirectory(${LIBRARIES}/gltfio)
 add_subdirectory(${LIBRARIES}/ibl)
 add_subdirectory(${LIBRARIES}/iblprefilter)
@@ -765,9 +749,6 @@ add_subdirectory(${EXTERNAL}/jsmn/tnt)
 add_subdirectory(${EXTERNAL}/stb/tnt)
 add_subdirectory(${EXTERNAL}/getopt)

-# Note that this has to be placed after mikktspace in order for combine_static_libs to work.
-add_subdirectory(${LIBRARIES}/geometry)
-
 if (FILAMENT_BUILD_FILAMAT OR IS_HOST_PLATFORM)
    # spirv-tools must come before filamat, as filamat relies on the presence of the
    # spirv-tools_SOURCE_DIR variable.
@@ -7,3 +7,5 @@ for next branch cut* header.
 appropriate header in [RELEASE_NOTES.md](./RELEASE_NOTES.md).

 ## Release notes for next branch cut
+
+- Metal: fix some shader artifacts by disabling fast math optimizations.
@@ -31,7 +31,7 @@ repositories {
 }

 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.51.3'
+    implementation 'com.google.android.filament:filament-android:1.49.3'
 }
 ```

@@ -51,7 +51,7 @@ Here are all the libraries available in the group `com.google.android.filament`:
 iOS projects can use CocoaPods to install the latest release:

 ```shell
-pod 'Filament', '~> 1.51.3'
+pod 'Filament', '~> 1.49.3'
 ```

 ### Snapshots
@@ -7,48 +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).

-## v1.51.3
-
-
-## v1.51.2
-
- engine: Add experimental APIs `Engine::builder::paused()` and `Engine::setPaused()`
-
-## v1.51.1
-
-
-## v1.51.0
-
- materials: add support for post-lighting mix factor (b/328498606) [⚠️ **New Material Version**]
-
-## v1.50.6
-
- Add new API `SwapChain::getFrameScheduledCallback`
- vulkan: fixed validation error VUID-vkAcquireNextImageKHR-semaphore-01779
- opengl: Add support for protected content swapchains and contexts
-
-## v1.50.5
-
- android: NDK 26.1.10909125 is used by default
- android: Minimum API level on Android is now API 21 instead of API 19. This allows the use of OpenGL ES 3.1
- rendering: New PBR Neutral tone mapper, designed to preserve materials color appearance
- android: Change default frameRateOptions.interval to 1.0
-
-## v1.50.4
-
-
-## v1.50.3
-
-
-## v1.50.2
-
-
-## v1.50.1
-
- Metal: fix some shader artifacts by disabling fast math optimizations.
- backend: remove `atan2` overload which had a typo and wasn't useful. Fixes b/320856413.
- utils: remove usages of `SpinLock`. Fixes b/321101014.
-
 ## v1.50.0
 - engine: TAA now supports 4x upscaling [BETA] [⚠️ **New Material Version**]

@@ -80,7 +80,7 @@ buildscript {

    ext.versions = [
        'jdk': 17,
-        'minSdk': 21,
+        'minSdk': 19,
        'targetSdk': 34,
        'compileSdk': 34,
        'kotlin': '1.9.21',
@@ -391,13 +391,6 @@ Java_com_google_android_filament_Engine_nFlush(JNIEnv*, jclass,
    engine->flush();
 }

-extern "C" JNIEXPORT void JNICALL
-Java_com_google_android_filament_Engine_nSetPaused(JNIEnv*, jclass,
-        jlong nativeEngine, jboolean paused) {
-    Engine* engine = (Engine*) nativeEngine;
-    engine->setPaused(paused);
-}
-
 // Managers...

 extern "C" JNIEXPORT jlong JNICALL
@@ -491,11 +484,7 @@ extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBu
 extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBuilderConfig(JNIEnv*,
        jclass, jlong nativeBuilder, jlong commandBufferSizeMB, jlong perRenderPassArenaSizeMB,
        jlong driverHandleArenaSizeMB, jlong minCommandBufferSizeMB, jlong perFrameCommandsSizeMB,
-        jlong jobSystemThreadCount,
-        jlong textureUseAfterFreePoolSize, jboolean disableParallelShaderCompile,
-        jint stereoscopicType, jlong stereoscopicEyeCount,
-        jlong resourceAllocatorCacheSizeMB, jlong resourceAllocatorCacheMaxAge,
-        jboolean disableHandleUseAfterFreeCheck) {
+        jlong jobSystemThreadCount, jlong stereoscopicEyeCount) {
    Engine::Builder* builder = (Engine::Builder*) nativeBuilder;
    Engine::Config config = {
            .commandBufferSizeMB = (uint32_t) commandBufferSizeMB,
@@ -504,13 +493,7 @@ extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBu
            .minCommandBufferSizeMB = (uint32_t) minCommandBufferSizeMB,
            .perFrameCommandsSizeMB = (uint32_t) perFrameCommandsSizeMB,
            .jobSystemThreadCount = (uint32_t) jobSystemThreadCount,
-            .textureUseAfterFreePoolSize = (uint32_t) textureUseAfterFreePoolSize,
-            .disableParallelShaderCompile = (bool) disableParallelShaderCompile,
-            .stereoscopicType = (Engine::StereoscopicType) stereoscopicType,
            .stereoscopicEyeCount = (uint8_t) stereoscopicEyeCount,
-            .resourceAllocatorCacheSizeMB = (uint32_t) resourceAllocatorCacheSizeMB,
-            .resourceAllocatorCacheMaxAge = (uint8_t) resourceAllocatorCacheMaxAge,
-            .disableHandleUseAfterFreeCheck = (bool) disableHandleUseAfterFreeCheck,
    };
    builder->config(&config);
 }
@@ -527,12 +510,6 @@ extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBu
    builder->sharedContext((void*) sharedContext);
 }

-extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBuilderPaused(
-        JNIEnv*, jclass, jlong nativeBuilder, jboolean paused) {
-    Engine::Builder* builder = (Engine::Builder*) nativeBuilder;
-    builder->paused((bool) paused);
-}
-
 extern "C" JNIEXPORT jlong JNICALL
 Java_com_google_android_filament_Engine_nBuilderBuild(JNIEnv*, jclass, jlong nativeBuilder) {
    Engine::Builder* builder = (Engine::Builder*) nativeBuilder;
@@ -104,14 +104,6 @@ Java_com_google_android_filament_RenderableManager_nBuilderGeometry__JIIJJIIII(J
            (size_t) count);
 }

-extern "C"
-JNIEXPORT void JNICALL
-Java_com_google_android_filament_RenderableManager_nBuilderGeometryType(JNIEnv*, jclass,
-        jlong nativeBuilder, int type) {
-    RenderableManager::Builder *builder = (RenderableManager::Builder *) nativeBuilder;
-    builder->geometryType((RenderableManager::Builder::GeometryType)type);
-}
-
 extern "C"
 JNIEXPORT void JNICALL
 Java_com_google_android_filament_RenderableManager_nBuilderMaterial(JNIEnv*, jclass,
@@ -34,15 +34,7 @@ Java_com_google_android_filament_SwapChain_nSetFrameCompletedCallback(JNIEnv* en
 }

 extern "C" JNIEXPORT jboolean JNICALL
-Java_com_google_android_filament_SwapChain_nIsSRGBSwapChainSupported(
-        JNIEnv *, jclass, jlong nativeEngine) {
+Java_com_google_android_filament_SwapChain_nIsSRGBSwapChainSupported(JNIEnv *, jclass, jlong nativeEngine) {
    Engine* engine = (Engine*) nativeEngine;
-    return (jboolean)SwapChain::isSRGBSwapChainSupported(*engine);
-}
-
-extern "C" JNIEXPORT jboolean JNICALL
-Java_com_google_android_filament_SwapChain_nIsProtectedContentSupported(
-        JNIEnv *, jclass, jlong nativeEngine) {
-    Engine* engine = (Engine*) nativeEngine;
-    return (jboolean)SwapChain::isProtectedContentSupported(*engine);
+    return (bool)SwapChain::isSRGBSwapChainSupported(*engine);
 }
@@ -47,11 +47,6 @@ Java_com_google_android_filament_ToneMapper_nCreateFilmicToneMapper(JNIEnv*, jcl
    return (jlong) new FilmicToneMapper();
 }

-extern "C" JNIEXPORT jlong JNICALL
-Java_com_google_android_filament_ToneMapper_nCreatePBRNeutralToneMapper(JNIEnv*, jclass) {
-    return (jlong) new PBRNeutralToneMapper();
-}
-
 extern "C" JNIEXPORT jlong JNICALL
 Java_com_google_android_filament_ToneMapper_nCreateAgxToneMapper(JNIEnv*, jclass, jint look) {
    return (jlong) new AgxToneMapper(AgxToneMapper::AgxLook(look));
@@ -158,16 +158,6 @@ public class Engine {
        FEATURE_LEVEL_3,
    };

-    /**
-     * The type of technique for stereoscopic rendering
-     */
-    public enum StereoscopicType {
-        /** Stereoscopic rendering is performed using instanced rendering technique. */
-        INSTANCED,
-        /** Stereoscopic rendering is performed using the multiview feature from the graphics backend. */
-        MULTIVIEW,
-    };
-
    /**
     * Constructs <code>Engine</code> objects using a builder pattern.
     */
@@ -221,11 +211,7 @@ public class Engine {
            nSetBuilderConfig(mNativeBuilder, config.commandBufferSizeMB,
                    config.perRenderPassArenaSizeMB, config.driverHandleArenaSizeMB,
                    config.minCommandBufferSizeMB, config.perFrameCommandsSizeMB,
-                    config.jobSystemThreadCount,
-                    config.textureUseAfterFreePoolSize, config.disableParallelShaderCompile,
-                    config.stereoscopicType.ordinal(), config.stereoscopicEyeCount,
-                    config.resourceAllocatorCacheSizeMB, config.resourceAllocatorCacheMaxAge,
-                    config.disableHandleUseAfterFreeCheck);
+                    config.jobSystemThreadCount, config.stereoscopicEyeCount);
            return this;
        }

@@ -240,20 +226,6 @@ public class Engine {
            return this;
        }

-        /**
-         * Sets the initial paused state of the rendering thread.
-         *
-         * <p>Warning: This is an experimental API. See {@link Engine#setPaused(boolean)} for
-         * caveats.
-         *
-         * @param paused Whether to start the rendering thread paused.
-         * @return A reference to this Builder for chaining calls.
-         */
-        public Builder paused(boolean paused) {
-            nSetBuilderPaused(mNativeBuilder, paused);
-            return this;
-        }
-
        /**
         * Creates an instance of Engine
         *
@@ -376,35 +348,6 @@ public class Engine {
         */
        public long jobSystemThreadCount = 0;

-        /**
-         * Number of most-recently destroyed textures to track for use-after-free.
-         *
-         * This will cause the backend to throw an exception when a texture is freed but still bound
-         * to a SamplerGroup and used in a draw call. 0 disables completely.
-         *
-         * Currently only respected by the Metal backend.
-         */
-        public long textureUseAfterFreePoolSize = 0;
-
-        /**
-         * Set to `true` to forcibly disable parallel shader compilation in the backend.
-         * Currently only honored by the GL backend.
-         */
-        public boolean disableParallelShaderCompile = false;
-
-        /**
-         * The type of technique for stereoscopic rendering.
-         *
-         * This setting determines the algorithm used when stereoscopic rendering is enabled. This
-         * decision applies to the entire Engine for the lifetime of the Engine. E.g., multiple
-         * Views created from the Engine must use the same stereoscopic type.
-         *
-         * Each view can enable stereoscopic rendering via the StereoscopicOptions::enable flag.
-         *
-         * @see View#setStereoscopicOptions
-         */
-        public StereoscopicType stereoscopicType = StereoscopicType.INSTANCED;
-
        /**
         * The number of eyes to render when stereoscopic rendering is enabled. Supported values are
         * between 1 and Engine#getMaxStereoscopicEyes() (inclusive).
@@ -413,21 +356,6 @@ public class Engine {
         * @see Engine#getMaxStereoscopicEyes
         */
        public long stereoscopicEyeCount = 2;
-
-        /*
-         * @Deprecated This value is no longer used.
-         */
-        public long resourceAllocatorCacheSizeMB = 64;
-
-        /*
-         * This value determines for how many frames are texture entries kept in the cache.
-         */
-        public long resourceAllocatorCacheMaxAge = 2;
-
-        /*
-         * Disable backend handles use-after-free checks.
-         */
-        public boolean disableHandleUseAfterFreeCheck = false;
    }

    private Engine(long nativeEngine, Config config) {
@@ -674,7 +602,7 @@ public class Engine {
     */
    @NonNull
    public SwapChain createSwapChain(@NonNull Object surface) {
-        return createSwapChain(surface, SwapChainFlags.CONFIG_DEFAULT);
+        return createSwapChain(surface, SwapChain.CONFIG_DEFAULT);
    }

    /**
@@ -682,15 +610,15 @@ public class Engine {
     *
     * @param surface on Android, <b>must be</b> an instance of {@link android.view.Surface}
     *
-     * @param flags configuration flags, see {@link SwapChainFlags}
+     * @param flags configuration flags, see {@link SwapChain}
     *
     * @return a newly created {@link SwapChain} object
     *
     * @exception IllegalStateException can be thrown if the SwapChain couldn't be created
     *
-     * @see SwapChainFlags#CONFIG_DEFAULT
-     * @see SwapChainFlags#CONFIG_TRANSPARENT
-     * @see SwapChainFlags#CONFIG_READABLE
+     * @see SwapChain#CONFIG_DEFAULT
+     * @see SwapChain#CONFIG_TRANSPARENT
+     * @see SwapChain#CONFIG_READABLE
     *
     */
    @NonNull
@@ -708,22 +636,21 @@ public class Engine {
     *
     * @param width  width of the rendering buffer
     * @param height height of the rendering buffer
-     * @param flags  configuration flags, see {@link SwapChainFlags}
+     * @param flags  configuration flags, see {@link SwapChain}
     *
     * @return a newly created {@link SwapChain} object
     *
     * @exception IllegalStateException can be thrown if the SwapChain couldn't be created
     *
-     * @see SwapChainFlags#CONFIG_DEFAULT
-     * @see SwapChainFlags#CONFIG_TRANSPARENT
-     * @see SwapChainFlags#CONFIG_READABLE
+     * @see SwapChain#CONFIG_DEFAULT
+     * @see SwapChain#CONFIG_TRANSPARENT
+     * @see SwapChain#CONFIG_READABLE
     *
     */
    @NonNull
    public SwapChain createSwapChain(int width, int height, long flags) {
        if (width >= 0 && height >= 0) {
-            long nativeSwapChain =
-                nCreateSwapChainHeadless(getNativeObject(), width, height, flags);
+            long nativeSwapChain = nCreateSwapChainHeadless(getNativeObject(), width, height, flags);
            if (nativeSwapChain == 0) throw new IllegalStateException("Couldn't create SwapChain");
            return new SwapChain(nativeSwapChain, null);
        }
@@ -735,12 +662,11 @@ public class Engine {
     *
     * @param surface a properly initialized {@link NativeSurface}
     *
-     * @param flags configuration flags, see {@link SwapChainFlags}
+     * @param flags configuration flags, see {@link SwapChain}
     *
     * @return a newly created {@link SwapChain} object
     *
-     * @exception IllegalStateException can be thrown if the {@link SwapChainFlags} couldn't be
-     *            created
+     * @exception IllegalStateException can be thrown if the {@link SwapChain} couldn't be created
     */
    @NonNull
    public SwapChain createSwapChainFromNativeSurface(@NonNull NativeSurface surface, long flags) {
@@ -1209,22 +1135,6 @@ public class Engine {
        nFlush(getNativeObject());
    }

-    /**
-     * Pause or resume the rendering thread.
-     *
-     * <p>Warning: This is an experimental API. In particular, note the following caveats.
-     *
-     * <ul><li>
-     * Buffer callbacks will never be called as long as the rendering thread is paused.
-     * Do not rely on a buffer callback to unpause the thread.
-     * </li><li>
-     * While the rendering thread is paused, rendering commands will continue to be queued until the
-     * buffer limit is reached. When the limit is reached, the program will abort.
-     * </li></ul>
-     */
-    public void setPaused(boolean paused) {
-        nSetPaused(getNativeObject(), paused);
-    }

    @UsedByReflection("TextureHelper.java")
    public long getNativeObject() {
@@ -1299,7 +1209,6 @@ public class Engine {
    private static native void nDestroyEntity(long nativeEngine, int entity);
    private static native void nFlushAndWait(long nativeEngine);
    private static native void nFlush(long nativeEngine);
-    private static native void nSetPaused(long nativeEngine, boolean paused);
    private static native long nGetTransformManager(long nativeEngine);
    private static native long nGetLightManager(long nativeEngine);
    private static native long nGetRenderableManager(long nativeEngine);
@@ -1318,12 +1227,8 @@ public class Engine {
    private static native void nSetBuilderConfig(long nativeBuilder, long commandBufferSizeMB,
            long perRenderPassArenaSizeMB, long driverHandleArenaSizeMB,
            long minCommandBufferSizeMB, long perFrameCommandsSizeMB, long jobSystemThreadCount,
-            long textureUseAfterFreePoolSize, boolean disableParallelShaderCompile,
-            int stereoscopicType, long stereoscopicEyeCount,
-            long resourceAllocatorCacheSizeMB, long resourceAllocatorCacheMaxAge,
-            boolean disableHandleUseAfterFreeCheck);
+            long stereoscopicEyeCount);
    private static native void nSetBuilderFeatureLevel(long nativeBuilder, int ordinal);
    private static native void nSetBuilderSharedContext(long nativeBuilder, long sharedContext);
-    private static native void nSetBuilderPaused(long nativeBuilder, boolean paused);
    private static native long nBuilderBuild(long nativeBuilder);
 }
@@ -175,32 +175,6 @@ public class RenderableManager {
            return this;
        }

-        /**
-         * Type of geometry for a Renderable
-         */
-        public enum GeometryType {
-            /** dynamic gemoetry has no restriction */
-            DYNAMIC,
-            /** bounds and world space transform are immutable */
-            STATIC_BOUNDS,
-            /** skinning/morphing not allowed and Vertex/IndexBuffer immutables */
-            STATIC
-        }
-
-        /**
-         * Specify whether this renderable has static bounds. In this context his means that
-         * the renderable's bounding box cannot change and that the renderable's transform is
-         * assumed immutable. Changing the renderable's transform via the TransformManager
-         * can lead to corrupted graphics. Note that skinning and morphing are not forbidden.
-         * Disabled by default.
-         * @param enable whether this renderable has static bounds. false by default.
-         */
-        @NonNull
-        public Builder geometryType(GeometryType type) {
-            nBuilderGeometryType(mNativeBuilder, type.ordinal());
-            return this;
-        }
-
        /**
         * Binds a material instance to the specified primitive.
         *
@@ -990,7 +964,6 @@ public class RenderableManager {
    private static native void nBuilderGeometry(long nativeBuilder, int index, int value, long nativeVertexBuffer, long nativeIndexBuffer);
    private static native void nBuilderGeometry(long nativeBuilder, int index, int value, long nativeVertexBuffer, long nativeIndexBuffer, int offset, int count);
    private static native void nBuilderGeometry(long nativeBuilder, int index, int value, long nativeVertexBuffer, long nativeIndexBuffer, int offset, int minIndex, int maxIndex, int count);
-    private static native void nBuilderGeometryType(long nativeBuilder, int type);
    private static native void nBuilderMaterial(long nativeBuilder, int index, long nativeMaterialInstance);
    private static native void nBuilderBlendOrder(long nativeBuilder, int index, int blendOrder);
    private static native void nBuilderGlobalBlendOrderEnabled(long nativeBuilder, int index, boolean enabled);
@@ -101,7 +101,7 @@ public class Renderer {
        /**
         * Desired frame interval in unit of 1 / DisplayInfo.refreshRate.
         */
-        public float interval = 1.0f;
+        public float interval = 1.0f / 60.0f;

        /**
         * Additional headroom for the GPU as a ratio of the targetFrameTime.
@@ -68,30 +68,77 @@ public class SwapChain {
    private final Object mSurface;
    private long mNativeObject;

+    public static final long CONFIG_DEFAULT = 0x0;
+
+    /**
+     * This flag indicates that the <code>SwapChain</code> must be allocated with an
+     * alpha-channel.
+     */
+    public static final long CONFIG_TRANSPARENT = 0x1;
+
+    /**
+     * This flag indicates that the <code>SwapChain</code> may be used as a source surface
+     * for reading back render results.  This config must be set when creating
+     * any <code>SwapChain</code>  that will be used as the source for a blit operation.
+     *
+     * @see Renderer#copyFrame
+     */
+    public static final long CONFIG_READABLE = 0x2;
+
+    /**
+     * Indicates that the native X11 window is an XCB window rather than an XLIB window.
+     * This is ignored on non-Linux platforms and in builds that support only one X11 API.
+     */
+    public static final long CONFIG_ENABLE_XCB = 0x4;
+
+    /**
+     * Indicates that the SwapChain must automatically perform linear to sRGB encoding.
+     *
+     * This flag is ignored if isSRGBSwapChainSupported() is false.
+     *
+     * When using this flag, post-processing should be disabled.
+     *
+     * @see SwapChain#isSRGBSwapChainSupported
+     * @see View#setPostProcessingEnabled
+     */
+    public static final long CONFIG_SRGB_COLORSPACE = 0x10;
+
+    /**
+     * Indicates that this SwapChain should allocate a stencil buffer in addition to a depth buffer.
+     *
+     * This flag is necessary when using View::setStencilBufferEnabled and rendering directly into
+     * the SwapChain (when post-processing is disabled).
+     *
+     * The specific format of the stencil buffer depends on platform support. The following pixel
+     * formats are tried, in order of preference:
+     *
+     * Depth only (without CONFIG_HAS_STENCIL_BUFFER):
+     * - DEPTH32F
+     * - DEPTH24
+     *
+     * Depth + stencil (with CONFIG_HAS_STENCIL_BUFFER):
+     * - DEPTH32F_STENCIL8
+     * - DEPTH24F_STENCIL8
+     *
+     * Note that enabling the stencil buffer may hinder depth precision and should only be used if
+     * necessary.
+     *
+     * @see View#setStencilBufferEnabled
+     * @see View#setPostProcessingEnabled
+     */
+    public static final long CONFIG_HAS_STENCIL_BUFFER = 0x20;
+
    SwapChain(long nativeSwapChain, Object surface) {
        mNativeObject = nativeSwapChain;
        mSurface = surface;
    }

    /**
-     * Return whether createSwapChain supports the CONFIG_PROTECTED_CONTENT flag.
+     * Return whether createSwapChain supports the SWAP_CHAIN_CONFIG_SRGB_COLORSPACE flag.
     * The default implementation returns false.
     *
     * @param engine A reference to the filament Engine
-     * @return true if CONFIG_PROTECTED_CONTENT is supported, false otherwise.
-     * @see SwapChainFlags#CONFIG_PROTECTED_CONTENT
-     */
-    public static boolean isProtectedContentSupported(@NonNull Engine engine) {
-        return nIsProtectedContentSupported(engine.getNativeObject());
-    }
-
-    /**
-     * Return whether createSwapChain supports the CONFIG_SRGB_COLORSPACE flag.
-     * The default implementation returns false.
-     *
-     * @param engine A reference to the filament Engine
-     * @return true if CONFIG_SRGB_COLORSPACE is supported, false otherwise.
-     * @see SwapChainFlags#CONFIG_SRGB_COLORSPACE
+     * @return true if SWAP_CHAIN_CONFIG_SRGB_COLORSPACE is supported, false otherwise.
     */
    public static boolean isSRGBSwapChainSupported(@NonNull Engine engine) {
        return nIsSRGBSwapChainSupported(engine.getNativeObject());
@@ -139,5 +186,4 @@ public class SwapChain {

    private static native void nSetFrameCompletedCallback(long nativeSwapChain, Object handler, Runnable callback);
    private static native boolean nIsSRGBSwapChainSupported(long nativeEngine);
-    private static native boolean nIsProtectedContentSupported(long nativeEngine);
 }
@@ -1,97 +0,0 @@
-/*
- * Copyright (C) 2024 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.
- */
-
-package com.google.android.filament;
-
-// Note: SwapChainFlags is kept separate from SwapChain so that UiHelper does not need to depend
-// on SwapChain. This allows clients to use UiHelper without requiring all of Filament's Java
-// classes.
-
-/**
- * Flags that a <code>SwapChain</code> can be created with to control behavior.
- *
- * @see Engine#createSwapChain
- * @see Engine#createSwapChainFromNativeSurface
- */
-public final class SwapChainFlags {
-
-    public static final long CONFIG_DEFAULT = 0x0;
-
-    /**
-     * This flag indicates that the <code>SwapChain</code> must be allocated with an
-     * alpha-channel.
-     */
-    public static final long CONFIG_TRANSPARENT = 0x1;
-
-    /**
-     * This flag indicates that the <code>SwapChain</code> may be used as a source surface
-     * for reading back render results.  This config must be set when creating
-     * any <code>SwapChain</code>  that will be used as the source for a blit operation.
-     *
-     * @see Renderer#copyFrame
-     */
-    public static final long CONFIG_READABLE = 0x2;
-
-    /**
-     * Indicates that the native X11 window is an XCB window rather than an XLIB window.
-     * This is ignored on non-Linux platforms and in builds that support only one X11 API.
-     */
-    public static final long CONFIG_ENABLE_XCB = 0x4;
-
-    /**
-     * Indicates that the SwapChain must automatically perform linear to sRGB encoding.
-     *
-     * This flag is ignored if isSRGBSwapChainSupported() is false.
-     *
-     * When using this flag, post-processing should be disabled.
-     *
-     * @see SwapChain#isSRGBSwapChainSupported
-     * @see View#setPostProcessingEnabled
-     */
-    public static final long CONFIG_SRGB_COLORSPACE = 0x10;
-
-    /**
-     * Indicates that this SwapChain should allocate a stencil buffer in addition to a depth buffer.
-     *
-     * This flag is necessary when using View::setStencilBufferEnabled and rendering directly into
-     * the SwapChain (when post-processing is disabled).
-     *
-     * The specific format of the stencil buffer depends on platform support. The following pixel
-     * formats are tried, in order of preference:
-     *
-     * Depth only (without CONFIG_HAS_STENCIL_BUFFER):
-     * - DEPTH32F
-     * - DEPTH24
-     *
-     * Depth + stencil (with CONFIG_HAS_STENCIL_BUFFER):
-     * - DEPTH32F_STENCIL8
-     * - DEPTH24F_STENCIL8
-     *
-     * Note that enabling the stencil buffer may hinder depth precision and should only be used if
-     * necessary.
-     *
-     * @see View#setStencilBufferEnabled
-     * @see View#setPostProcessingEnabled
-     */
-    public static final long CONFIG_HAS_STENCIL_BUFFER = 0x20;
-
-    /**
-     * The SwapChain contains protected content. Only supported when isProtectedContentSupported()
-     * is true.
-     */
-    public static final long CONFIG_PROTECTED_CONTENT   = 0x40;
-}
-
@@ -16,14 +16,12 @@ package com.google.android.filament;
 * <li>Configurable tone mapping operators</li>
 * <ul>
 *   <li>GenericToneMapper</li>
- *   <li>AgXToneMapper</li>
 * </ul>
 * <li>Fixed-aesthetic tone mapping operators</li>
 * <ul>
 *   <li>ACESToneMapper</li>
 *   <li>ACESLegacyToneMapper</li>
 *   <li>FilmicToneMapper</li>
- *   <li>PBRNeutralToneMapper</li>
 * </ul>
 * <li>Debug/validation tone mapping operators</li>
 * <ul>
@@ -102,21 +100,11 @@ public class ToneMapper {
        }
    }

-    /**
-     * Khronos PBR Neutral tone mapping operator. This tone mapper was designed
-     * to preserve the appearance of materials across lighting conditions while
-     * avoiding artifacts in the highlights in high dynamic range conditions.
-     */
-    public static class PBRNeutralToneMapper extends ToneMapper {
-        public PBRNeutralToneMapper() {
-            super(nCreatePBRNeutralToneMapper());
-        }
-    }
-
    /**
     * AgX tone mapping operator.
     */
    public static class Agx extends ToneMapper {
+
        public enum AgxLook {
            /**
             * Base contrast with no look applied
@@ -245,7 +233,6 @@ public class ToneMapper {
    private static native long nCreateACESToneMapper();
    private static native long nCreateACESLegacyToneMapper();
    private static native long nCreateFilmicToneMapper();
-    private static native long nCreatePBRNeutralToneMapper();
    private static native long nCreateAgxToneMapper(int look);
    private static native long nCreateGenericToneMapper(
            float contrast, float midGrayIn, float midGrayOut, float hdrMax);
@@ -1882,7 +1882,7 @@ public class View {
     * Options for Temporal Anti-aliasing (TAA)
     * Most TAA parameters are extremely costly to change, as they will trigger the TAA post-process
     * shaders to be recompiled. These options should be changed or set during initialization.
-     * `filterWidth`, `feedback` and `jitterPattern`, however, can be changed at any time.
+     * `filterWidth`, `feedback` and `jitterPattern`, however, could be changed at any time.
     *
     * `feedback` of 0.1 effectively accumulates a maximum of 19 samples in steady state.
     * see "A Survey of Temporal Antialiasing Techniques" by Lei Yang and all for more information.
@@ -1925,7 +1925,6 @@ public class View {
            UNIFORM_HELIX_X4,
            HALTON_23_X8,
            HALTON_23_X16,
-            HALTON_23_X32,
        }

        /**
@@ -1940,10 +1939,6 @@ public class View {
         * texturing lod bias (typically -1 or -2)
         */
        public float lodBias = -1.0f;
-        /**
-         * post-TAA sharpen, especially useful when upscaling is true.
-         */
-        public float sharpness = 0.0f;
        /**
         * enables or disables temporal anti-aliasing
         */
@@ -27,7 +27,7 @@ import android.view.SurfaceHolder;
 import android.view.SurfaceView;
 import android.view.TextureView;

-import com.google.android.filament.SwapChainFlags;
+import com.google.android.filament.SwapChain;

 /**
 * UiHelper is a simple class that can manage either a SurfaceView, TextureView, or a SurfaceHolder
@@ -538,7 +538,7 @@ public class UiHelper {
     * the options set on this UiHelper.
     */
    public long getSwapChainFlags() {
-        return isOpaque() ? SwapChainFlags.CONFIG_DEFAULT : SwapChainFlags.CONFIG_TRANSPARENT;
+        return isOpaque() ? SwapChain.CONFIG_DEFAULT : SwapChain.CONFIG_TRANSPARENT;
    }

    /**
@@ -81,17 +81,9 @@ set(GLTFIO_SRCS
        ${GLTFIO_DIR}/src/TangentsJob.cpp
        ${GLTFIO_DIR}/src/TangentsJob.h
        ${GLTFIO_DIR}/src/UbershaderProvider.cpp
-        ${GLTFIO_DIR}/src/Utility.cpp
-        ${GLTFIO_DIR}/src/Utility.h
        ${GLTFIO_DIR}/src/Wireframe.cpp
        ${GLTFIO_DIR}/src/Wireframe.h
        ${GLTFIO_DIR}/src/downcast.h
-        ${GLTFIO_DIR}/src/extended/AssetLoaderExtended.h
-        ${GLTFIO_DIR}/src/extended/TangentsJobExtended.cpp
-        ${GLTFIO_DIR}/src/extended/TangentsJobExtended.h
-        ${GLTFIO_DIR}/src/extended/TangentSpaceMeshWrapper.cpp
-        ${GLTFIO_DIR}/src/extended/TangentSpaceMeshWrapper.h
-

        src/main/cpp/Animator.cpp
        src/main/cpp/AssetLoader.cpp
@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.51.3
+VERSION_NAME=1.49.3

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

@@ -30,7 +30,7 @@ android {
    compileSdkVersion versions.compileSdk
    defaultConfig {
        applicationId "com.google.android.filament.gltf"
-        minSdkVersion versions.minSdk
+        minSdkVersion 19
        targetSdkVersion versions.targetSdk
    }

@@ -516,7 +516,7 @@ function build_android {
            if [[ "${BUILD_ANDROID_SAMPLES}" == "true" ]]; then
                for sample in ${ANDROID_SAMPLES}; do
                    echo "Installing out/${sample}-debug.apk"
-                    cp samples/${sample}/build/outputs/apk/debug/${sample}-debug.apk \
+                    cp samples/${sample}/build/outputs/apk/debug/${sample}-debug-unsigned.apk \
                        ../out/${sample}-debug.apk
                done
            fi
@@ -1 +1 @@
-26.1.10909125
+25.1.8937393
@@ -125,13 +125,7 @@ if [[ "${has_universal}" == "true" ]]; then

        arch_output="${OUTPUT_PATH%.a}_${arch}.a"
        arch_outputs+=("$arch_output")
-
-        archives=()
-        while IFS=  read -r -d $'\0'; do
-            archives+=("$REPLY")
-        done < <(find "$(pwd)/${archs_temp_dir}/${arch}" -iname '*.a' -print0)
-
-        combine_static_libs "$arch_output" "$archives"
+        combine_static_libs "$arch_output" $(find "$(pwd)/${archs_temp_dir}/${arch}" -iname '*.a')
    done

    # Finally, combine the single-architecture archives into a universal binary.
@@ -21,7 +21,7 @@ set(CMAKE_SYSTEM_NAME Linux)
 set(CMAKE_SYSTEM_VERSION 1)

 # android
-set(API_LEVEL 21)
+set(API_LEVEL 19)

 # architecture
 set(ARCH armv7a-linux-androideabi)
@@ -21,7 +21,7 @@ set(CMAKE_SYSTEM_NAME Linux)
 set(CMAKE_SYSTEM_VERSION 1)

 # android
-set(API_LEVEL 21)
+set(API_LEVEL 19)

 # architecture
 set(ARCH i686-linux-android)
@@ -1397,7 +1397,7 @@ Type
 :    `string`

 Value
-:     Any of `opaque`, `transparent`, `fade`, `add`, `masked`, `multiply`, `screen`, `custom`. Defaults to `opaque`.
+:     Any of `opaque`, `transparent`, `fade`, `add`, `masked`, `multiply`, `screen`. Defaults to `opaque`.

 Description
 :     Defines how/if the rendered object is blended with the content of the render target.
@@ -1420,7 +1420,6 @@ Description
      of the material's output defines whether a fragment is discarded or not. Additionally,
      ALPHA_TO_COVERAGE is enabled for non-translucent views. See the maskThreshold section for more
      information.
-    - **Custom**: blending is enabled. But the blending function is user specified. See `blendFunction`.

 !!! Note
    When `blending` is set to `masked`, alpha to coverage is automatically enabled for the material.
@@ -1433,36 +1432,6 @@ material {
 }
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

-### Blending and transparency: blendFunction
-
-Type
-:    `object`
-
-Fields
-:     `srcRGB`, `srcA`, `dstRGB`, `dstA`
-
-Description
-:       - *srcRGB*: source function applied to the RGB channels
-        - *srcA*: source function applied to the alpha channel
-        - *srcRGB*: destination function applied to the RGB channels
-        - *srcRGB*: destination function applied to the alpha channel
-        The values possible for each functions are one of `zero`, `one`, `srcColor`, `oneMinusSrcColor`,
-               `dstColor`, `oneMinusDstColor`, `srcAlpha`, `oneMinusSrcAlpha`, `dstAlpha`,
-               `oneMinusDstAlpha`, `srcAlphaSaturate`
-
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ JSON
-material {
-    blending : custom,
-    blendFunction :
-    {
-        srcRGB: one,
-        srcA: one,
-        dstRGB: oneMinusSrcColor,
-        dstA: oneMinusSrcAlpha
-    }
- }
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
 ### Blending and transparency: postLightingBlending

 Type
@@ -66,7 +66,6 @@ set(SRCS
        src/Froxelizer.cpp
        src/Frustum.cpp
        src/HwRenderPrimitiveFactory.cpp
-        src/HwVertexBufferInfoFactory.cpp
        src/IndexBuffer.cpp
        src/IndirectLight.cpp
        src/InstanceBuffer.cpp
@@ -138,7 +137,6 @@ set(SRCS

 set(PRIVATE_HDRS
        src/Allocators.h
-        src/Bimap.h
        src/BufferPoolAllocator.h
        src/ColorSpaceUtils.h
        src/Culler.h
@@ -149,7 +147,6 @@ set(PRIVATE_HDRS
        src/FrameSkipper.h
        src/Froxelizer.h
        src/HwRenderPrimitiveFactory.h
-        src/HwVertexBufferInfoFactory.h
        src/Intersections.h
        src/MaterialParser.h
        src/PerViewUniforms.h
@@ -215,7 +212,6 @@ set(MATERIAL_SRCS
        src/materials/antiAliasing/fxaa.mat
        src/materials/antiAliasing/taa.mat
        src/materials/blitLow.mat
-        src/materials/blitArray.mat
        src/materials/bloom/bloomDownsample.mat
        src/materials/bloom/bloomDownsample2x.mat
        src/materials/bloom/bloomDownsample9.mat
@@ -242,7 +238,6 @@ set(MATERIAL_SRCS
        src/materials/resolveDepth.mat
        src/materials/separableGaussianBlur.mat
        src/materials/skybox.mat
-        src/materials/shadowmap.mat
        src/materials/ssao/bilateralBlur.mat
        src/materials/ssao/bilateralBlurBentNormals.mat
        src/materials/ssao/mipmapDepth.mat
@@ -252,13 +247,8 @@ set(MATERIAL_SRCS
 )

 set(MATERIAL_FL0_SRCS
-        src/materials/defaultMaterial.mat
-        src/materials/skybox.mat
-)
-
-set(MATERIAL_MULTIVIEW_SRCS
-        src/materials/defaultMaterial.mat
-        src/materials/skybox.mat
+        src/materials/defaultMaterial0.mat
+        src/materials/skybox0.mat
 )

 # Embed the binary resource blob for materials.
@@ -290,11 +280,6 @@ if (FILAMENT_ENABLE_FEATURE_LEVEL_0)
    add_definitions(-DFILAMENT_ENABLE_FEATURE_LEVEL_0)
 endif()

-# Whether to include MULTIVIEW materials.
-if (FILAMENT_ENABLE_MULTIVIEW)
-    add_definitions(-DFILAMENT_ENABLE_MULTIVIEW)
-endif()
-
 # ==================================================================================================
 # Definitions
 # ==================================================================================================
@@ -325,43 +310,34 @@ foreach (mat_src ${MATERIAL_SRCS})
    get_filename_component(localname "${mat_src}" NAME_WE)
    get_filename_component(fullname "${mat_src}" ABSOLUTE)
    set(output_path "${MATERIAL_DIR}/${localname}.filamat")
+
    add_custom_command(
            OUTPUT ${output_path}
            COMMAND matc ${MATC_BASE_FLAGS} -o ${output_path} ${fullname}
            MAIN_DEPENDENCY ${fullname}
            DEPENDS matc
-            COMMENT "Compiling material ${fullname} to ${output_path}"
+            COMMENT "Compiling material ${mat_src} to ${output_path}"
    )
    list(APPEND MATERIAL_BINS ${output_path})
-
-    list(FIND MATERIAL_FL0_SRCS ${mat_src} index)
-    if (${index} GREATER -1 AND FILAMENT_ENABLE_FEATURE_LEVEL_0)
-        string(REGEX REPLACE "[.]filamat$" "_fl0.filamat" output_path_fl0 ${output_path})
-        add_custom_command(
-                OUTPUT ${output_path_fl0}
-                COMMAND matc ${MATC_BASE_FLAGS} -PfeatureLevel=0 -o ${output_path_fl0} ${fullname}
-                MAIN_DEPENDENCY ${fullname}
-                DEPENDS matc
-                COMMENT "Compiling material ${fullname} to ${output_path_fl0}"
-        )
-        list(APPEND MATERIAL_BINS ${output_path_fl0})
-    endif ()
-
-    list(FIND MATERIAL_MULTIVIEW_SRCS ${mat_src} index)
-    if (${index} GREATER -1 AND FILAMENT_ENABLE_MULTIVIEW)
-        string(REGEX REPLACE "[.]filamat$" "_multiview.filamat" output_path_multiview ${output_path})
-        add_custom_command(
-                OUTPUT ${output_path_multiview}
-                COMMAND matc ${MATC_BASE_FLAGS} -PstereoscopicType=multiview -o ${output_path_multiview} ${fullname}
-                MAIN_DEPENDENCY ${fullname}
-                DEPENDS matc
-                COMMENT "Compiling material ${fullname} to ${output_path_multiview}"
-        )
-        list(APPEND MATERIAL_BINS ${output_path_multiview})
-    endif ()
-
 endforeach()

+if (FILAMENT_ENABLE_FEATURE_LEVEL_0)
+    foreach (mat_src ${MATERIAL_FL0_SRCS})
+        get_filename_component(localname "${mat_src}" NAME_WE)
+        get_filename_component(fullname "${mat_src}" ABSOLUTE)
+        set(output_path "${MATERIAL_DIR}/${localname}.filamat")
+
+        add_custom_command(
+                OUTPUT ${output_path}
+                COMMAND matc ${MATC_BASE_FLAGS} -o ${output_path} ${fullname}
+                MAIN_DEPENDENCY ${fullname}
+                DEPENDS matc
+                COMMENT "Compiling material ${mat_src} to ${output_path}"
+        )
+        list(APPEND MATERIAL_BINS ${output_path})
+    endforeach ()
+endif ()
+
 # Additional dependencies on included files for materials

 add_custom_command(
@@ -166,8 +166,6 @@ endif()
 if (FILAMENT_SUPPORTS_VULKAN)
    list(APPEND SRCS
            include/backend/platforms/VulkanPlatform.h
-            src/vulkan/caching/VulkanDescriptorSet.cpp
-            src/vulkan/caching/VulkanDescriptorSet.h
            src/vulkan/platform/VulkanPlatform.cpp
            src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp
            src/vulkan/platform/VulkanPlatformSwapChainImpl.h
@@ -23,6 +23,7 @@
 #include <utils/ostream.h>

 #include <stddef.h>
+#include <stdint.h>

 namespace filament::backend {

@@ -17,6 +17,8 @@
 #ifndef TNT_FILAMENT_BACKEND_CALLBACKHANDLER_H
 #define TNT_FILAMENT_BACKEND_CALLBACKHANDLER_H

+#include <utils/compiler.h>
+
 namespace filament::backend {

 /**
@@ -28,8 +28,7 @@

 #include <math/vec4.h>

-#include <array>        // FIXME: STL headers are not allowed in public headers
-#include <type_traits>  // FIXME: STL headers are not allowed in public headers
+#include <array>    // FIXME: STL headers are not allowed in public headers

 #include <stddef.h>
 #include <stdint.h>
@@ -81,14 +80,7 @@ static constexpr uint64_t SWAP_CHAIN_CONFIG_SRGB_COLORSPACE     = 0x10;
 /**
 * Indicates that the SwapChain should also contain a stencil component.
 */
-static constexpr uint64_t SWAP_CHAIN_CONFIG_HAS_STENCIL_BUFFER  = 0x20;
-static constexpr uint64_t SWAP_CHAIN_HAS_STENCIL_BUFFER         = SWAP_CHAIN_CONFIG_HAS_STENCIL_BUFFER;
-
-/**
- * The SwapChain contains protected content. Currently only supported by OpenGLPlatform and
- * only when OpenGLPlatform::isProtectedContextSupported() is true.
- */
-static constexpr uint64_t SWAP_CHAIN_CONFIG_PROTECTED_CONTENT   = 0x40;
+static constexpr uint64_t SWAP_CHAIN_HAS_STENCIL_BUFFER         = 0x20;


 static constexpr size_t MAX_VERTEX_ATTRIBUTE_COUNT  = 16;   // This is guaranteed by OpenGL ES.
@@ -136,12 +128,6 @@ enum class Backend : uint8_t {
    NOOP = 4,     //!< Selects the no-op driver for testing purposes.
 };

-enum class TimerQueryResult : int8_t {
-    ERROR = -1,     // an error occurred, result won't be available
-    NOT_READY = 0,  // result to ready yet
-    AVAILABLE = 1,  // result is available
-};
-
 static constexpr const char* backendToString(Backend backend) {
    switch (backend) {
        case Backend::NOOP:
@@ -672,17 +658,16 @@ enum class TextureFormat : uint16_t {
 };

 //! Bitmask describing the intended Texture Usage
-enum class TextureUsage : uint16_t {
-    NONE                = 0x0000,
-    COLOR_ATTACHMENT    = 0x0001,            //!< Texture can be used as a color attachment
-    DEPTH_ATTACHMENT    = 0x0002,            //!< Texture can be used as a depth attachment
-    STENCIL_ATTACHMENT  = 0x0004,            //!< Texture can be used as a stencil attachment
-    UPLOADABLE          = 0x0008,            //!< Data can be uploaded into this texture (default)
-    SAMPLEABLE          = 0x0010,            //!< Texture can be sampled (default)
-    SUBPASS_INPUT       = 0x0020,            //!< Texture can be used as a subpass input
-    BLIT_SRC            = 0x0040,            //!< Texture can be used the source of a blit()
-    BLIT_DST            = 0x0080,            //!< Texture can be used the destination of a blit()
-    PROTECTED           = 0x0100,            //!< Texture can be used the destination of a blit()
+enum class TextureUsage : uint8_t {
+    NONE                = 0x00,
+    COLOR_ATTACHMENT    = 0x01,                     //!< Texture can be used as a color attachment
+    DEPTH_ATTACHMENT    = 0x02,                     //!< Texture can be used as a depth attachment
+    STENCIL_ATTACHMENT  = 0x04,                     //!< Texture can be used as a stencil attachment
+    UPLOADABLE          = 0x08,                     //!< Data can be uploaded into this texture (default)
+    SAMPLEABLE          = 0x10,                     //!< Texture can be sampled (default)
+    SUBPASS_INPUT       = 0x20,                     //!< Texture can be used as a subpass input
+    BLIT_SRC            = 0x40,                     //!< Texture can be used the source of a blit()
+    BLIT_DST            = 0x80,                     //!< Texture can be used the destination of a blit()
    DEFAULT             = UPLOADABLE | SAMPLEABLE   //!< Default texture usage
 };

@@ -1186,27 +1171,11 @@ struct StencilState {

    //! Stencil operations for front-facing polygons
    StencilOperations front = {
-            .stencilFunc = StencilFunction::A,
-            .stencilOpStencilFail = StencilOperation::KEEP,
-            .padding0 = 0,
-            .stencilOpDepthFail = StencilOperation::KEEP,
-            .stencilOpDepthStencilPass = StencilOperation::KEEP,
-            .padding1 = 0,
-            .ref = 0,
-            .readMask = 0xff,
-            .writeMask = 0xff };
+            .stencilFunc = StencilFunction::A, .ref = 0, .readMask = 0xff, .writeMask = 0xff };

    //! Stencil operations for back-facing polygons
    StencilOperations back  = {
-            .stencilFunc = StencilFunction::A,
-            .stencilOpStencilFail = StencilOperation::KEEP,
-            .padding0 = 0,
-            .stencilOpDepthFail = StencilOperation::KEEP,
-            .stencilOpDepthStencilPass = StencilOperation::KEEP,
-            .padding1 = 0,
-            .ref = 0,
-            .readMask = 0xff,
-            .writeMask = 0xff };
+            .stencilFunc = StencilFunction::A, .ref = 0, .readMask = 0xff, .writeMask = 0xff };

    //! Whether stencil-buffer writes are enabled
    bool stencilWrite = false;
@@ -1242,14 +1211,6 @@ enum class Workaround : uint16_t {
    DISABLE_THREAD_AFFINITY
 };

-//! The type of technique for stereoscopic rendering
-enum class StereoscopicType : uint8_t {
-    // Stereoscopic rendering is performed using instanced rendering technique.
-    INSTANCED,
-    // Stereoscopic rendering is performed using the multiview feature from the graphics backend.
-    MULTIVIEW,
-};
-
 } // namespace filament::backend

 template<> struct utils::EnableBitMaskOperators<filament::backend::ShaderStageFlags>
@@ -17,15 +17,17 @@
 #ifndef TNT_FILAMENT_BACKEND_HANDLE_H
 #define TNT_FILAMENT_BACKEND_HANDLE_H

+#include <utils/compiler.h>
 #if !defined(NDEBUG)
-#include <utils/ostream.h>
+#include <utils/Log.h>
 #endif
 #include <utils/debug.h>

-#include <type_traits> // FIXME: STL headers are not allowed in public headers
-
 #include <stdint.h>

+#include <limits>
+#include <type_traits>
+
 namespace filament::backend {

 struct HwBufferObject;
@@ -39,7 +41,6 @@ struct HwStream;
 struct HwSwapChain;
 struct HwTexture;
 struct HwTimerQuery;
-struct HwVertexBufferInfo;
 struct HwVertexBuffer;

 /*
@@ -53,7 +54,7 @@ struct HwVertexBuffer;
 class HandleBase {
 public:
    using HandleId = uint32_t;
-    static constexpr const HandleId nullid = HandleId{ UINT32_MAX };
+    static constexpr const HandleId nullid = HandleId{ std::numeric_limits<HandleId>::max() };

    constexpr HandleBase() noexcept: object(nullid) {}

@@ -63,6 +64,14 @@ public:
    // clear the handle, this doesn't free associated resources
    void clear() noexcept { object = nullid; }

+    // compare handles
+    bool operator==(const HandleBase& rhs) const noexcept { return object == rhs.object; }
+    bool operator!=(const HandleBase& rhs) const noexcept { return object != rhs.object; }
+    bool operator<(const HandleBase& rhs) const noexcept { return object < rhs.object; }
+    bool operator<=(const HandleBase& rhs) const noexcept { return object <= rhs.object; }
+    bool operator>(const HandleBase& rhs) const noexcept { return object > rhs.object; }
+    bool operator>=(const HandleBase& rhs) const noexcept { return object >= rhs.object; }
+
    // get this handle's handleId
    HandleId getId() const noexcept { return object; }

@@ -94,14 +103,6 @@ struct Handle : public HandleBase {

    explicit Handle(HandleId id) noexcept : HandleBase(id) { }

-    // compare handles of the same type
-    bool operator==(const Handle& rhs) const noexcept { return getId() == rhs.getId(); }
-    bool operator!=(const Handle& rhs) const noexcept { return getId() != rhs.getId(); }
-    bool operator<(const Handle& rhs) const noexcept { return getId() < rhs.getId(); }
-    bool operator<=(const Handle& rhs) const noexcept { return getId() <= rhs.getId(); }
-    bool operator>(const Handle& rhs) const noexcept { return getId() > rhs.getId(); }
-    bool operator>=(const Handle& rhs) const noexcept { return getId() >= rhs.getId(); }
-
    // type-safe Handle cast
    template<typename B, typename = std::enable_if_t<std::is_base_of<T, B>::value> >
    Handle(Handle<B> const& base) noexcept : HandleBase(base) { } // NOLINT(hicpp-explicit-conversions,google-explicit-constructor)
@@ -115,19 +116,18 @@ private:

 // Types used by the command stream
 // (we use this renaming because the macro-system doesn't deal well with "<" and ">")
-using BufferObjectHandle     = Handle<HwBufferObject>;
-using FenceHandle            = Handle<HwFence>;
-using IndexBufferHandle      = Handle<HwIndexBuffer>;
-using ProgramHandle          = Handle<HwProgram>;
-using RenderPrimitiveHandle  = Handle<HwRenderPrimitive>;
-using RenderTargetHandle     = Handle<HwRenderTarget>;
-using SamplerGroupHandle     = Handle<HwSamplerGroup>;
-using StreamHandle           = Handle<HwStream>;
-using SwapChainHandle        = Handle<HwSwapChain>;
-using TextureHandle          = Handle<HwTexture>;
-using TimerQueryHandle       = Handle<HwTimerQuery>;
-using VertexBufferHandle     = Handle<HwVertexBuffer>;
-using VertexBufferInfoHandle = Handle<HwVertexBufferInfo>;
+using BufferObjectHandle    = Handle<HwBufferObject>;
+using FenceHandle           = Handle<HwFence>;
+using IndexBufferHandle     = Handle<HwIndexBuffer>;
+using ProgramHandle         = Handle<HwProgram>;
+using RenderPrimitiveHandle = Handle<HwRenderPrimitive>;
+using RenderTargetHandle    = Handle<HwRenderTarget>;
+using SamplerGroupHandle    = Handle<HwSamplerGroup>;
+using StreamHandle          = Handle<HwStream>;
+using SwapChainHandle       = Handle<HwSwapChain>;
+using TextureHandle         = Handle<HwTexture>;
+using TimerQueryHandle      = Handle<HwTimerQuery>;
+using VertexBufferHandle    = Handle<HwVertexBuffer>;

 } // namespace filament::backend

@@ -20,7 +20,7 @@
 #include <backend/DriverEnums.h>
 #include <backend/Handle.h>

-#include <utils/ostream.h>
+#include <limits>

 #include <stdint.h>

@@ -29,13 +29,14 @@ namespace filament::backend {
 //! \privatesection

 struct PipelineState {
-    Handle<HwProgram> program;                                              //  4
-    Handle<HwVertexBufferInfo> vertexBufferInfo;                            //  4
-    RasterState rasterState;                                                //  4
-    StencilState stencilState;                                              // 12
-    PolygonOffset polygonOffset;                                            //  8
-    PrimitiveType primitiveType = PrimitiveType::TRIANGLES;                 //  1
-    uint8_t padding[3] = {};                                                //  3
+    Handle<HwProgram> program;
+    RasterState rasterState;
+    StencilState stencilState;
+    PolygonOffset polygonOffset;
+    Viewport scissor{ 0, 0,
+                      (uint32_t)std::numeric_limits<int32_t>::max(),
+                      (uint32_t)std::numeric_limits<int32_t>::max()
+    };
 };

 } // namespace filament::backend
@@ -24,7 +24,6 @@

 #include <utils/compiler.h>
 #include <utils/debug.h>
-#include <utils/ostream.h>

 #include <stddef.h>
 #include <stdint.h>
@@ -19,12 +19,11 @@
 #ifndef TNT_FILAMENT_BACKEND_PLATFORM_H
 #define TNT_FILAMENT_BACKEND_PLATFORM_H

+#include <backend/DriverEnums.h>
+
 #include <utils/compiler.h>
 #include <utils/Invocable.h>

-#include <stddef.h>
-#include <stdint.h>
-
 namespace filament::backend {

 class Driver;
@@ -42,29 +41,18 @@ public:
    struct Stream {};

    struct DriverConfig {
-        /**
-         * Size of handle arena in bytes. Setting to 0 indicates default value is to be used.
+        /*
+         * size of handle arena in bytes. Setting to 0 indicates default value is to be used.
         * Driver clamps to valid values.
         */
        size_t handleArenaSize = 0;

-        /**
-         * This number of most-recently destroyed textures will be tracked for use-after-free.
+        /*
+         * this number of most-recently destroyed textures will be tracked for use-after-free.
         * Throws an exception when a texture is freed but still bound to a SamplerGroup and used in
         * a draw call. 0 disables completely. Currently only respected by the Metal backend.
         */
        size_t textureUseAfterFreePoolSize = 0;
-
-        /**
-         * Set to `true` to forcibly disable parallel shader compilation in the backend.
-         * Currently only honored by the GL and Metal backends.
-         */
-        bool disableParallelShaderCompile = false;
-
-        /**
-         * Disable backend handles use-after-free checks.
-         */
-        bool disableHandleUseAfterFreeCheck = false;
    };

    Platform() noexcept;
@@ -90,7 +78,7 @@ public:
     *
     * @return nullptr on failure, or a pointer to the newly created driver.
     */
-    virtual backend::Driver* UTILS_NULLABLE createDriver(void* UTILS_NULLABLE sharedContext,
+    virtual backend::Driver* createDriver(void* sharedContext,
            const DriverConfig& driverConfig) noexcept = 0;

    /**
@@ -108,8 +96,7 @@ public:
     * cache.
     */
    using InsertBlobFunc = utils::Invocable<
-            void(const void* UTILS_NONNULL key, size_t keySize,
-                    const void* UTILS_NONNULL value, size_t valueSize)>;
+            void(const void* key, size_t keySize, const void* value, size_t valueSize)>;

    /*
     * RetrieveBlobFunc is an Invocable to an application-provided function that a
@@ -117,8 +104,7 @@ public:
     * cache.
     */
    using RetrieveBlobFunc = utils::Invocable<
-            size_t(const void* UTILS_NONNULL key, size_t keySize,
-                    void* UTILS_NONNULL value, size_t valueSize)>;
+            size_t(const void* key, size_t keySize, void* value, size_t valueSize)>;

    /**
     * Sets the callback functions that the backend can use to interact with caching functionality
@@ -171,8 +157,7 @@ public:
     * @param value         pointer to the beginning of the value data that is to be inserted
     * @param valueSize     specifies the size in byte of the data pointed to by <value>
     */
-    void insertBlob(const void* UTILS_NONNULL key, size_t keySize,
-            const void* UTILS_NONNULL value, size_t valueSize);
+    void insertBlob(const void* key, size_t keySize, const void* value, size_t valueSize);

    /**
     * To retrieve the binary value associated with a given key from the cache, a
@@ -191,43 +176,11 @@ public:
     * @return             If the cache contains a value associated with the given key then the
     *                     size of that binary value in bytes is returned. Otherwise 0 is returned.
     */
-    size_t retrieveBlob(const void* UTILS_NONNULL key, size_t keySize,
-            void* UTILS_NONNULL value, size_t valueSize);
-
-    using DebugUpdateStatFunc = utils::Invocable<void(const char* UTILS_NONNULL key, uint64_t value)>;
-
-    /**
-     * Sets the callback function that the backend can use to update backend-specific statistics
-     * to aid with debugging. This callback is guaranteed to be called on the Filament driver
-     * thread.
-     *
-     * @param debugUpdateStat   an Invocable that updates debug statistics
-     */
-    void setDebugUpdateStatFunc(DebugUpdateStatFunc&& debugUpdateStat) noexcept;
-
-    /**
-     * @return true if debugUpdateStat is valid.
-     */
-    bool hasDebugUpdateStatFunc() const noexcept;
-
-    /**
-     * To track backend-specific statistics, the backend implementation can call the
-     * application-provided callback function debugUpdateStatFunc to associate or update a value
-     * with a given key. It is possible for this function to be called multiple times with the
-     * same key, in which case newer values should overwrite older values.
-     *
-     * This function is guaranteed to be called only on a single thread, the Filament driver
-     * thread.
-     *
-     * @param key          a null-terminated C-string with the key of the debug statistic
-     * @param value        the updated value of key
-     */
-    void debugUpdateStat(const char* UTILS_NONNULL key, uint64_t value);
+    size_t retrieveBlob(const void* key, size_t keySize, void* value, size_t valueSize);

 private:
    InsertBlobFunc mInsertBlob;
    RetrieveBlobFunc mRetrieveBlob;
-    DebugUpdateStatFunc mDebugUpdateStat;
 };

 } // namespace filament
@@ -21,7 +21,8 @@

 #include <utils/compiler.h>

-namespace filament::backend {
+namespace filament {
+namespace backend {

 /**
 * A PresentCallable is a callable object that, when called, schedules a frame for presentation on
@@ -97,6 +98,7 @@ private:
 */
 using FrameFinishedCallback UTILS_DEPRECATED = void(*)(PresentCallable callable, void* user);

-} // namespace filament::backend
+} // namespace backend
+} // namespace filament

 #endif // TNT_FILAMENT_BACKEND_PRESENTCALLABLE
@@ -17,19 +17,17 @@
 #ifndef TNT_FILAMENT_BACKEND_PRIVATE_PROGRAM_H
 #define TNT_FILAMENT_BACKEND_PRIVATE_PROGRAM_H

+#include <utils/compiler.h>
 #include <utils/CString.h>
 #include <utils/FixedCapacityVector.h>
 #include <utils/Invocable.h>
+#include <utils/Log.h>
 #include <utils/ostream.h>

 #include <backend/DriverEnums.h>

-#include <array>    // FIXME: STL headers are not allowed in public headers
-#include <utility>  // FIXME: STL headers are not allowed in public headers
-#include <variant>  // FIXME: STL headers are not allowed in public headers
-
-#include <stddef.h>
-#include <stdint.h>
+#include <array>
+#include <variant>

 namespace filament::backend {

@@ -116,8 +114,6 @@ public:

    Program& cacheId(uint64_t cacheId) noexcept;

-    Program& multiview(bool multiview) noexcept;
-
    ShaderSource const& getShadersSource() const noexcept { return mShadersSource; }
    ShaderSource& getShadersSource() noexcept { return mShadersSource; }

@@ -145,8 +141,6 @@ public:

    uint64_t getCacheId() const noexcept { return mCacheId; }

-    bool isMultiview() const noexcept { return mMultiview; }
-
    CompilerPriorityQueue getPriorityQueue() const noexcept { return mPriorityQueue; }

 private:
@@ -162,11 +156,6 @@ private:
    utils::FixedCapacityVector<std::pair<utils::CString, uint8_t>> mAttributes;
    std::array<UniformInfo, Program::UNIFORM_BINDING_COUNT> mBindingUniformInfo;
    CompilerPriorityQueue mPriorityQueue = CompilerPriorityQueue::HIGH;
-    // Indicates the current engine was initialized with multiview stereo, and the variant for this
-    // program contains STE flag. This will be referred later for the OpenGL shader compiler to
-    // determine whether shader code replacement for the num_views should be performed.
-    // This variable could be promoted as a more generic variable later if other similar needs occur.
-    bool mMultiview = false;
 };

 } // namespace filament::backend
@@ -24,6 +24,9 @@

 #include <utils/compiler.h>

+#include <stddef.h>
+#include <stdint.h>
+
 namespace filament::backend {

 struct UTILS_PUBLIC SamplerDescriptor {
@@ -17,11 +17,9 @@
 #ifndef TNT_FILAMENT_BACKEND_TARGETBUFFERINFO_H
 #define TNT_FILAMENT_BACKEND_TARGETBUFFERINFO_H

+#include <backend/DriverEnums.h>
 #include <backend/Handle.h>

-#include <utils/ostream.h>
-
-#include <stddef.h>
 #include <stdint.h>

 namespace filament::backend {
@@ -32,10 +30,6 @@ struct TargetBufferInfo {
    // texture to be used as render target
    Handle<HwTexture> handle;

-    // starting layer index for multiview. This value is only used when the `layerCount` for the
-    // render target is greater than 1.
-    uint8_t baseViewIndex = 0;
-
    // level to be used
    uint8_t level = 0;

@@ -84,7 +78,7 @@ public:

    // this is here for backward compatibility
    MRT(Handle<HwTexture> handle, uint8_t level, uint16_t layer) noexcept
-            : mInfos{{ handle, 0, level, layer }} {
+            : mInfos{{ handle, level, layer }} {
    }
 };

@@ -18,15 +18,8 @@
 #define TNT_FILAMENT_BACKEND_PRIVATE_OPENGLPLATFORM_H

 #include <backend/AcquiredImage.h>
-#include <backend/DriverEnums.h>
 #include <backend/Platform.h>

-#include <utils/compiler.h>
-#include <utils/Invocable.h>
-
-#include <stddef.h>
-#include <stdint.h>
-
 namespace filament::backend {

 class Driver;
@@ -45,8 +38,8 @@ protected:
     * Derived classes can use this to instantiate the default OpenGLDriver backend.
     * This is typically called from your implementation of createDriver()
     */
-    static Driver* UTILS_NULLABLE createDefaultDriver(OpenGLPlatform* UTILS_NONNULL platform,
-            void* UTILS_NULLABLE sharedContext, const DriverConfig& driverConfig);
+    static Driver* createDefaultDriver(OpenGLPlatform* platform,
+            void* sharedContext, const DriverConfig& driverConfig);

    ~OpenGLPlatform() noexcept override;

@@ -64,22 +57,6 @@ public:
     */
    virtual void terminate() noexcept = 0;

-    /**
-     * Return whether createSwapChain supports the SWAP_CHAIN_CONFIG_SRGB_COLORSPACE flag.
-     * The default implementation returns false.
-     *
-     * @return true if SWAP_CHAIN_CONFIG_SRGB_COLORSPACE is supported, false otherwise.
-     */
-    virtual bool isSRGBSwapChainSupported() const noexcept;
-
-    /**
-     * Return whether protected contexts are supported by this backend.
-     * If protected context are supported, the SWAP_CHAIN_CONFIG_PROTECTED_CONTENT flag can be
-     * used when creating a SwapChain.
-     * The default implementation returns false.
-     */
-    virtual bool isProtectedContextSupported() const noexcept;
-
    /**
     * Called by the driver to create a SwapChain for this driver.
     *
@@ -89,8 +66,15 @@ public:
     * @return              The driver's SwapChain object.
     *
     */
-    virtual SwapChain* UTILS_NULLABLE createSwapChain(
-            void* UTILS_NULLABLE nativeWindow, uint64_t flags) noexcept = 0;
+    virtual SwapChain* createSwapChain(void* nativeWindow, uint64_t flags) noexcept = 0;
+
+    /**
+     * Return whether createSwapChain supports the SWAP_CHAIN_CONFIG_SRGB_COLORSPACE flag.
+     * The default implementation returns false.
+     *
+     * @return true if SWAP_CHAIN_CONFIG_SRGB_COLORSPACE is supported, false otherwise.
+     */
+    virtual bool isSRGBSwapChainSupported() const noexcept;

    /**
     * Called by the driver create a headless SwapChain.
@@ -103,14 +87,13 @@ public:
     * TODO: we need a more generic way of passing construction parameters
     *       A void* might be enough.
     */
-    virtual SwapChain* UTILS_NULLABLE createSwapChain(
-            uint32_t width, uint32_t height, uint64_t flags) noexcept = 0;
+    virtual SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept = 0;

    /**
     * Called by the driver to destroys the SwapChain
     * @param swapChain SwapChain to be destroyed.
     */
-    virtual void destroySwapChain(SwapChain* UTILS_NONNULL swapChain) noexcept = 0;
+    virtual void destroySwapChain(SwapChain* swapChain) noexcept = 0;

    /**
     * Returns the set of buffers that must be preserved up to the call to commit().
@@ -123,80 +106,28 @@ public:
     * @return buffer that must be preserved
     * @see commit()
     */
-    virtual TargetBufferFlags getPreservedFlags(SwapChain* UTILS_NONNULL swapChain) noexcept;
-
-    /**
-     * Returns true if the swapchain is protected
-     */
-    virtual bool isSwapChainProtected(Platform::SwapChain* UTILS_NONNULL swapChain) noexcept;
+    virtual TargetBufferFlags getPreservedFlags(SwapChain* swapChain) noexcept;

    /**
     * Called by the driver to establish the default FBO. The default implementation returns 0.
-     *
-     * This method can be called either on the regular or protected OpenGL contexts and can return
-     * a different or identical name, since these names exist in different namespaces.
-     *
-     * @return a GLuint casted to a uint32_t that is an OpenGL framebuffer object.
+      * @return a GLuint casted to a uint32_t that is an OpenGL framebuffer object.
     */
-    virtual uint32_t getDefaultFramebufferObject() noexcept;
-
-
-    /**
-     * Type of contexts available
-     */
-    enum class ContextType {
-        NONE,           //!< No current context
-        UNPROTECTED,    //!< current context is unprotected
-        PROTECTED       //!< current context supports protected content
-    };
-
-    /**
-     * Returns the type of the context currently in use. This value is updated by makeCurrent()
-     * and therefore can be cached between calls. ContextType::PROTECTED can only be returned
-     * if isProtectedContextSupported() is true.
-     * @return ContextType
-     */
-    virtual ContextType getCurrentContextType() const noexcept;
-
-    /**
-     * Binds the requested context to the current thread and drawSwapChain to the default FBO
-     * returned by getDefaultFramebufferObject().
-     *
-     * @param type type of context to bind to the current thread.
-     * @param drawSwapChain SwapChain to draw to. It must be bound to the default FBO.
-     * @param readSwapChain SwapChain to read from (for operation like `glBlitFramebuffer`)
-     * @return true on success, false on error.
-     */
-    virtual bool makeCurrent(ContextType type,
-            SwapChain* UTILS_NONNULL drawSwapChain,
-            SwapChain* UTILS_NONNULL readSwapChain) noexcept = 0;
+    virtual uint32_t createDefaultRenderTarget() noexcept;

    /**
     * Called by the driver to make the OpenGL context active on the calling thread and bind
-     * the drawSwapChain to the default FBO returned by getDefaultFramebufferObject().
-     * The context used is either the default context or the protected context. When a context
-     * change is necessary, the preContextChange and postContextChange callbacks are called,
-     * before and after the context change respectively. postContextChange is given the index
-     * of the new context (0 for default and 1 for protected).
-     * The default implementation just calls makeCurrent(getCurrentContextType(), SwapChain*, SwapChain*).
-     *
+     * the drawSwapChain to the default render target (FBO) created with createDefaultRenderTarget.
     * @param drawSwapChain SwapChain to draw to. It must be bound to the default FBO.
     * @param readSwapChain SwapChain to read from (for operation like `glBlitFramebuffer`)
-     * @param preContextChange called before the context changes
-     * @param postContextChange called after the context changes
     */
-    virtual void makeCurrent(
-            SwapChain* UTILS_NONNULL drawSwapChain,
-            SwapChain* UTILS_NONNULL readSwapChain,
-            utils::Invocable<void()> preContextChange,
-            utils::Invocable<void(size_t index)> postContextChange) noexcept;
+    virtual void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept = 0;

    /**
     * Called by the driver once the current frame finishes drawing. Typically, this should present
     * the drawSwapChain. This is for example where `eglMakeCurrent()` would be called.
     * @param swapChain the SwapChain to present.
     */
-    virtual void commit(SwapChain* UTILS_NONNULL swapChain) noexcept = 0;
+    virtual void commit(SwapChain* swapChain) noexcept = 0;

    /**
     * Set the time the next committed buffer should be presented to the user at.
@@ -221,14 +152,14 @@ public:
     *
     * @return A Fence object. The default implementation returns nullptr.
     */
-    virtual Fence* UTILS_NULLABLE createFence() noexcept;
+    virtual Fence* createFence() noexcept;

    /**
     * Destroys a Fence object. The default implementation does nothing.
     *
     * @param fence Fence to destroy.
     */
-    virtual void destroyFence(Fence* UTILS_NONNULL fence) noexcept;
+    virtual void destroyFence(Fence* fence) noexcept;

    /**
     * Waits on a Fence.
@@ -238,7 +169,7 @@ public:
     * @return Whether the fence signaled or timed out. See backend::FenceStatus.
     *         The default implementation always return backend::FenceStatus::ERROR.
     */
-    virtual backend::FenceStatus waitFence(Fence* UTILS_NONNULL fence, uint64_t timeout) noexcept;
+    virtual backend::FenceStatus waitFence(Fence* fence, uint64_t timeout) noexcept;


    // --------------------------------------------------------------------------------------------
@@ -252,13 +183,13 @@ public:
     * @param nativeStream The native stream, this parameter depends on the concrete implementation.
     * @return A new Stream object.
     */
-    virtual Stream* UTILS_NULLABLE createStream(void* UTILS_NULLABLE nativeStream) noexcept;
+    virtual Stream* createStream(void* nativeStream) noexcept;

    /**
     * Destroys a Stream.
     * @param stream Stream to destroy.
     */
-    virtual void destroyStream(Stream* UTILS_NONNULL stream) noexcept;
+    virtual void destroyStream(Stream* stream) noexcept;

    /**
     * The specified stream takes ownership of the texture (tname) object
@@ -268,21 +199,20 @@ public:
     * @param stream Stream to take ownership of the texture
     * @param tname  GL texture id to "bind" to the Stream.
     */
-    virtual void attach(Stream* UTILS_NONNULL stream, intptr_t tname) noexcept;
+    virtual void attach(Stream* stream, intptr_t tname) noexcept;

    /**
     * Destroys the texture associated to the stream
     * @param stream Stream to detach from its texture
     */
-    virtual void detach(Stream* UTILS_NONNULL stream) noexcept;
+    virtual void detach(Stream* stream) noexcept;

    /**
     * Updates the content of the texture attached to the stream.
     * @param stream Stream to update
     * @param timestamp Output parameter: Timestamp of the image bound to the texture.
     */
-    virtual void updateTexImage(Stream* UTILS_NONNULL stream,
-            int64_t* UTILS_NONNULL timestamp) noexcept;
+    virtual void updateTexImage(Stream* stream, int64_t* timestamp) noexcept;


    // --------------------------------------------------------------------------------------------
@@ -295,13 +225,13 @@ public:
     *         implementation could just return { 0, GL_TEXTURE_2D } at this point. The actual
     *         values can be delayed until setExternalImage.
     */
-    virtual ExternalTexture* UTILS_NULLABLE createExternalImageTexture() noexcept;
+    virtual ExternalTexture *createExternalImageTexture() noexcept;

    /**
     * Destroys an external texture handle and associated data.
     * @param texture a pointer to the handle to destroy.
     */
-    virtual void destroyExternalImage(ExternalTexture* UTILS_NONNULL texture) noexcept;
+    virtual void destroyExternalImage(ExternalTexture* texture) noexcept;

    // called on the application thread to allow Filament to take ownership of the image

@@ -314,7 +244,7 @@ public:
     * @param externalImage A token representing the platform's external image.
     * @see destroyExternalImage
     */
-    virtual void retainExternalImage(void* UTILS_NONNULL externalImage) noexcept;
+    virtual void retainExternalImage(void* externalImage) noexcept;

    /**
     * Called to bind the platform-specific externalImage to an ExternalTexture.
@@ -328,8 +258,7 @@ public:
     * @param texture an in/out pointer to ExternalTexture, id and target can be updated if necessary.
     * @return true on success, false on error.
     */
-    virtual bool setExternalImage(void* UTILS_NONNULL externalImage,
-            ExternalTexture* UTILS_NONNULL texture) noexcept;
+    virtual bool setExternalImage(void* externalImage, ExternalTexture* texture) noexcept;

    /**
     * The method allows platforms to convert a user-supplied external image object into a new type
@@ -17,10 +17,11 @@
 #ifndef TNT_FILAMENT_BACKEND_OPENGL_OPENGL_PLATFORM_COCOA_GL_H
 #define TNT_FILAMENT_BACKEND_OPENGL_OPENGL_PLATFORM_COCOA_GL_H

-#include <backend/DriverEnums.h>
+#include <stdint.h>
+
 #include <backend/platforms/OpenGLPlatform.h>

-#include <stdint.h>
+#include <backend/DriverEnums.h>

 namespace filament::backend {

@@ -57,7 +58,7 @@ protected:
    SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
    SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
    void destroySwapChain(SwapChain* swapChain) noexcept override;
-    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
+    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
    void commit(SwapChain* swapChain) noexcept override;
    OpenGLPlatform::ExternalTexture* createExternalImageTexture() noexcept override;
    void destroyExternalImage(ExternalTexture* texture) noexcept override;
@@ -30,7 +30,7 @@ struct PlatformCocoaTouchGLImpl;
 class PlatformCocoaTouchGL : public OpenGLPlatform {
 public:
    PlatformCocoaTouchGL();
-    ~PlatformCocoaTouchGL() noexcept override;
+    ~PlatformCocoaTouchGL() noexcept;

    // --------------------------------------------------------------------------------------------
    // Platform Interface
@@ -45,7 +45,7 @@ public:

    void terminate() noexcept override;

-    uint32_t getDefaultFramebufferObject() noexcept override;
+    uint32_t createDefaultRenderTarget() noexcept override;

    bool isExtraContextSupported() const noexcept override;
    void createContext(bool shared) override;
@@ -53,7 +53,7 @@ public:
    SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
    SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
    void destroySwapChain(SwapChain* swapChain) noexcept override;
-    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
+    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
    void commit(SwapChain* swapChain) noexcept override;

    OpenGLPlatform::ExternalTexture* createExternalImageTexture() noexcept override;
@@ -17,23 +17,18 @@
 #ifndef TNT_FILAMENT_BACKEND_OPENGL_OPENGL_PLATFORM_EGL_H
 #define TNT_FILAMENT_BACKEND_OPENGL_OPENGL_PLATFORM_EGL_H

-#include <backend/DriverEnums.h>
-#include <backend/Platform.h>
-#include <backend/platforms/OpenGLPlatform.h>
+#include <stdint.h>

 #include <EGL/egl.h>
 #include <EGL/eglext.h>
-#include <EGL/eglplatform.h>

-#include <utils/Invocable.h>
+#include <backend/platforms/OpenGLPlatform.h>
+
+#include <backend/DriverEnums.h>

-#include <initializer_list>
 #include <utility>
 #include <vector>

-#include <stddef.h>
-#include <stdint.h>
-
 namespace filament::backend {

 /**
@@ -43,11 +38,15 @@ class PlatformEGL : public OpenGLPlatform {
 public:

    PlatformEGL() noexcept;
+    bool isExtraContextSupported() const noexcept override;
+    void createContext(bool shared) override;
+    void releaseContext() noexcept override;

    // Return true if we're on an OpenGL platform (as opposed to OpenGL ES). false by default.
    virtual bool isOpenGL() const noexcept;

 protected:
+
    // --------------------------------------------------------------------------------------------
    // Helper for EGL configs and attributes parameters

@@ -87,30 +86,13 @@ protected:
    // --------------------------------------------------------------------------------------------
    // OpenGLPlatform Interface

-    bool isExtraContextSupported() const noexcept override;
-    void createContext(bool shared) override;
-    void releaseContext() noexcept override;
-
    void terminate() noexcept override;

-    bool isProtectedContextSupported() const noexcept override;
-
    bool isSRGBSwapChainSupported() const noexcept override;
    SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
    SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
    void destroySwapChain(SwapChain* swapChain) noexcept override;
-    bool isSwapChainProtected(SwapChain* swapChain) noexcept override;
-
-    ContextType getCurrentContextType() const noexcept override;
-
-    bool makeCurrent(ContextType type,
-            SwapChain* drawSwapChain,
-            SwapChain* readSwapChain) noexcept override;
-
-    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain,
-            utils::Invocable<void()> preContextChange,
-            utils::Invocable<void(size_t index)> postContextChange) noexcept override;
-
+    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
    void commit(SwapChain* swapChain) noexcept override;

    bool canCreateFence() noexcept override;
@@ -137,27 +119,16 @@ protected:
    static void clearGlError() noexcept;

    /**
-     * Always use this instead of eglMakeCurrent(), as it tracks some state.
+     * Always use this instead of eglMakeCurrent().
     */
-
-    EGLContext getContextForType(ContextType type) const noexcept;
-
-    // makes the draw and read surface current without changing the current context
-    EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) noexcept {
-        return egl.makeCurrent(drawSurface, readSurface);
-    }
-
-    // makes context current and set draw and read surfaces to EGL_NO_SURFACE
-    EGLBoolean makeCurrent(EGLContext context) noexcept {
-        return egl.makeCurrent(context, mEGLDummySurface, mEGLDummySurface);
-    }
+    EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) noexcept;

    // TODO: this should probably use getters instead.
    EGLDisplay mEGLDisplay = EGL_NO_DISPLAY;
    EGLContext mEGLContext = EGL_NO_CONTEXT;
-    EGLContext mEGLContextProtected = EGL_NO_CONTEXT;
+    EGLSurface mCurrentDrawSurface = EGL_NO_SURFACE;
+    EGLSurface mCurrentReadSurface = EGL_NO_SURFACE;
    EGLSurface mEGLDummySurface = EGL_NO_SURFACE;
-    ContextType mCurrentContextType = ContextType::NONE;
    // mEGLConfig is valid only if ext.egl.KHR_no_config_context is false
    EGLConfig mEGLConfig = EGL_NO_CONFIG_KHR;
    Config mContextAttribs;
@@ -174,38 +145,13 @@ protected:
            bool KHR_gl_colorspace = false;
            bool KHR_no_config_context = false;
            bool KHR_surfaceless_context = false;
-            bool EXT_protected_content = false;
        } egl;
    } ext;

-    struct SwapChainEGL : public Platform::SwapChain {
-        EGLSurface sur = EGL_NO_SURFACE;
-        Config attribs{};
-        EGLNativeWindowType nativeWindow{};
-        EGLConfig config{};
-        uint64_t flags{};
-    };
-
    void initializeGlExtensions() noexcept;

 protected:
    EGLConfig findSwapChainConfig(uint64_t flags, bool window, bool pbuffer) const;
-
-private:
-    class EGL {
-        EGLDisplay& mEGLDisplay;
-        EGLSurface mCurrentDrawSurface = EGL_NO_SURFACE;
-        EGLSurface mCurrentReadSurface = EGL_NO_SURFACE;
-        EGLContext mCurrentContext = EGL_NO_CONTEXT;
-    public:
-        explicit EGL(EGLDisplay& dpy) : mEGLDisplay(dpy) {}
-        EGLBoolean makeCurrent(EGLContext context,
-                EGLSurface drawSurface, EGLSurface readSurface) noexcept;
-
-        EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) noexcept {
-            return makeCurrent(mCurrentContext, drawSurface, readSurface);
-        }
-    } egl{ mEGLDisplay };
 };

 } // namespace filament::backend
@@ -17,14 +17,8 @@
 #ifndef TNT_FILAMENT_BACKEND_OPENGL_OPENGL_PLATFORM_EGL_ANDROID_H
 #define TNT_FILAMENT_BACKEND_OPENGL_OPENGL_PLATFORM_EGL_ANDROID_H

-#include <backend/AcquiredImage.h>
-#include <backend/Platform.h>
-#include <backend/platforms/OpenGLPlatform.h>
 #include <backend/platforms/PlatformEGL.h>

-#include <stddef.h>
-#include <stdint.h>
-
 namespace filament::backend {

 class ExternalStreamManagerAndroid;
@@ -51,7 +51,7 @@ protected:
    SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
    SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
    void destroySwapChain(SwapChain* swapChain) noexcept override;
-    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
+    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
    void commit(SwapChain* swapChain) noexcept override;

 private:
@@ -53,7 +53,7 @@ protected:
    SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
    SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
    void destroySwapChain(SwapChain* swapChain) noexcept override;
-    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
+    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
    void commit(SwapChain* swapChain) noexcept override;

 protected:
@@ -46,7 +46,7 @@ protected:
    SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
    SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
    void destroySwapChain(SwapChain* swapChain) noexcept override;
-    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
+    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
    void commit(SwapChain* swapChain) noexcept override;
 };

@@ -20,18 +20,13 @@
 #include <backend/Platform.h>

 #include <bluevk/BlueVK.h>
-
 #include <utils/CString.h>
 #include <utils/FixedCapacityVector.h>
 #include <utils/PrivateImplementation.h>

-#include <string_view>
 #include <tuple>
 #include <unordered_set>

-#include <stddef.h>
-#include <stdint.h>
-
 namespace filament::backend {

 using SwapChain = Platform::SwapChain;
@@ -17,7 +17,7 @@
 #ifndef TNT_FILAMENT_BACKEND_PRIVATE_CIRCULARBUFFER_H
 #define TNT_FILAMENT_BACKEND_PRIVATE_CIRCULARBUFFER_H

-#include <utils/debug.h>
+#include <utils/compiler.h>

 #include <stddef.h>
 #include <stdint.h>
@@ -40,36 +40,28 @@ public:

    ~CircularBuffer() noexcept;

-    static size_t getBlockSize() noexcept { return sPageSize; }
-
-    // Total size of circular buffer. This is a constant.
-    size_t size() const noexcept { return mSize; }
-
-    // Allocates `s` bytes in the circular buffer and returns a pointer to the memory. All
-    // allocations must not exceed size() bytes.
-    inline void* allocate(size_t s) noexcept {
-        // We can never allocate more that size().
-        assert_invariant(getUsed() + s <= size());
+    // allocates 'size' bytes in the circular buffer and returns a pointer to the memory
+    // return the current head and moves it forward by size bytes
+    inline void* allocate(size_t size) noexcept {
        char* const cur = static_cast<char*>(mHead);
-        mHead = cur + s;
+        mHead = cur + size;
        return cur;
    }

-    // Returns true if the buffer is empty, i.e.: no allocations were made since
-    // calling getBuffer();
+    // Total size of circular buffer
+    size_t size() const noexcept { return mSize; }
+
+    // returns true if the buffer is empty (e.g. after calling flush)
    bool empty() const noexcept { return mTail == mHead; }

-    // Returns the size used since the last call to getBuffer()
-    size_t getUsed() const noexcept { return intptr_t(mHead) - intptr_t(mTail); }
+    void* getHead() const noexcept { return mHead; }

-    // Retrieves the current allocated range and frees it. It is the responsibility of the caller
-    // to make sure the returned range is no longer in use by the time allocate() allocates
-    // (size() - getUsed()) bytes.
-    struct Range {
-        void* tail;
-        void* head;
-    };
-    Range getBuffer() noexcept;
+    void* getTail() const noexcept { return mTail; }
+
+    // call at least once every getRequiredSize() bytes allocated from the buffer
+    void circularize() noexcept;
+
+    static size_t getBlockSize() noexcept { return sPageSize; }

 private:
    void* alloc(size_t size) noexcept;
@@ -77,10 +69,10 @@ private:

    // pointer to the beginning of the circular buffer (constant)
    void* mData = nullptr;
-    int mAshmemFd = -1;
+    int mUsesAshmem = -1;

    // size of the circular buffer (constant)
-    size_t const mSize;
+    size_t mSize = 0;

    // pointer to the beginning of recorded data
    void* mTail = nullptr;
@@ -19,21 +19,19 @@

 #include "private/backend/CircularBuffer.h"

+#include <utils/compiler.h>
 #include <utils/Condition.h>
 #include <utils/Mutex.h>

 #include <vector>

-#include <stddef.h>
-#include <stdint.h>
-
 namespace filament::backend {

 /*
 * A producer-consumer command queue that uses a CircularBuffer as main storage
 */
 class CommandBufferQueue {
-    struct Range {
+    struct Slice {
        void* begin;
        void* end;
    };
@@ -46,33 +44,29 @@ class CommandBufferQueue {

    mutable utils::Mutex mLock;
    mutable utils::Condition mCondition;
-    mutable std::vector<Range> mCommandBuffersToExecute;
+    mutable std::vector<Slice> mCommandBuffersToExecute;
    size_t mFreeSpace = 0;
    size_t mHighWatermark = 0;
    uint32_t mExitRequested = 0;
-    bool mPaused = false;

    static constexpr uint32_t EXIT_REQUESTED = 0x31415926;

 public:
    // requiredSize: guaranteed available space after flush()
-    CommandBufferQueue(size_t requiredSize, size_t bufferSize, bool paused);
+    CommandBufferQueue(size_t requiredSize, size_t bufferSize);
    ~CommandBufferQueue();

-    CircularBuffer& getCircularBuffer() noexcept { return mCircularBuffer; }
-    CircularBuffer const& getCircularBuffer() const noexcept { return mCircularBuffer; }
-
-    size_t getCapacity() const noexcept { return mRequiredSize; }
+    CircularBuffer& getCircularBuffer() { return mCircularBuffer; }

    size_t getHighWatermark() const noexcept { return mHighWatermark; }

    // wait for commands to be available and returns an array containing these commands
-    std::vector<Range> waitForCommands() const;
+    std::vector<Slice> waitForCommands() const;

    // return the memory used by this command buffer to the circular buffer
    // WARNING: releaseBuffer() must be called in sequence of the Slices returned by
    // waitForCommands()
-    void releaseBuffer(Range const& buffer);
+    void releaseBuffer(Slice const& buffer);

    // all commands buffers (Slices) written to this point are returned by waitForCommand(). This
    // call blocks until the CircularBuffer has at least mRequiredSize bytes available.
@@ -81,9 +75,6 @@ public:
    // returns from waitForCommands() immediately.
    void requestExit();

-    // suspend or unsuspend the queue.
-    void setPaused(bool paused);
-
    bool isExitRequested() const;
 };

@@ -32,13 +32,11 @@
 #include <backend/TargetBufferInfo.h>

 #include <utils/compiler.h>
-#include <utils/debug.h>
 #include <utils/ThreadUtils.h>

 #include <cstddef>
 #include <functional>
 #include <tuple>
-#include <type_traits>
 #include <utility>

 #ifndef NDEBUG
@@ -134,7 +132,7 @@ struct CommandType<void (Driver::*)(ARGS...)> {

    public:
        template<typename M, typename D>
-        static inline void execute(M&& method, D&& driver, CommandBase* base, intptr_t* next) {
+        static inline void execute(M&& method, D&& driver, CommandBase* base, intptr_t* next) noexcept {
            Command* self = static_cast<Command*>(base);
            *next = align(sizeof(Command));
 #if DEBUG_COMMAND_STREAM
@@ -154,21 +152,21 @@ struct CommandType<void (Driver::*)(ARGS...)> {
        }

        // placement new declared as "throw" to avoid the compiler's null-check
-        inline void* operator new(std::size_t, void* ptr) {
+        inline void* operator new(std::size_t size, void* ptr) {
            assert_invariant(ptr);
            return ptr;
        }
    };
 };

-// convert a method of "class Driver" into a Command<> type
+// convert an method of "class Driver" into a Command<> type
 #define COMMAND_TYPE(method) CommandType<decltype(&Driver::method)>::Command<&Driver::method>

 // ------------------------------------------------------------------------------------------------

 class CustomCommand : public CommandBase {
    std::function<void()> mCommand;
-    static void execute(Driver&, CommandBase* base, intptr_t* next);
+    static void execute(Driver&, CommandBase* base, intptr_t* next) noexcept;
 public:
    inline CustomCommand(CustomCommand&& rhs) = default;
    inline explicit CustomCommand(std::function<void()> cmd)
@@ -213,8 +211,6 @@ public:
    CommandStream(CommandStream const& rhs) noexcept = delete;
    CommandStream& operator=(CommandStream const& rhs) noexcept = delete;

-    CircularBuffer const& getCircularBuffer() const noexcept { return mCurrentBuffer; }
-
 public:
 #define DECL_DRIVER_API(methodName, paramsDecl, params)                                         \
    inline void methodName(paramsDecl) {                                                        \
@@ -29,7 +29,6 @@

 #include <functional>

-#include <stddef.h>
 #include <stdint.h>

 // Command debugging off. debugging virtuals are not called.
@@ -76,7 +75,7 @@ public:
    // the fn function will execute a batch of driver commands
    // this gives the driver a chance to wrap their execution in a meaningful manner
    // the default implementation simply calls fn
-    virtual void execute(std::function<void(void)> const& fn);
+    virtual void execute(std::function<void(void)> const& fn) noexcept;

    // This is called on debug build, or when enabled manually on the backend thread side.
    virtual void debugCommandBegin(CommandStream* cmds,
@@ -167,14 +167,11 @@ DECL_DRIVER_API_0(resetState)
 * -----------------------
 */

-DECL_DRIVER_API_R_N(backend::VertexBufferInfoHandle, createVertexBufferInfo,
+DECL_DRIVER_API_R_N(backend::VertexBufferHandle, createVertexBuffer,
        uint8_t, bufferCount,
        uint8_t, attributeCount,
-        backend::AttributeArray, attributes)
-
-DECL_DRIVER_API_R_N(backend::VertexBufferHandle, createVertexBuffer,
        uint32_t, vertexCount,
-        backend::VertexBufferInfoHandle, vbih)
+        backend::AttributeArray, attributes)

 DECL_DRIVER_API_R_N(backend::IndexBufferHandle, createIndexBuffer,
        backend::ElementType, elementType,
@@ -227,7 +224,11 @@ DECL_DRIVER_API_R_N(backend::SamplerGroupHandle, createSamplerGroup,
 DECL_DRIVER_API_R_N(backend::RenderPrimitiveHandle, createRenderPrimitive,
        backend::VertexBufferHandle, vbh,
        backend::IndexBufferHandle, ibh,
-        backend::PrimitiveType, pt)
+        backend::PrimitiveType, pt,
+        uint32_t, offset,
+        uint32_t, minIndex,
+        uint32_t, maxIndex,
+        uint32_t, count)

 DECL_DRIVER_API_R_N(backend::ProgramHandle, createProgram,
        backend::Program&&, program)
@@ -239,7 +240,6 @@ DECL_DRIVER_API_R_N(backend::RenderTargetHandle, createRenderTarget,
        uint32_t, width,
        uint32_t, height,
        uint8_t, samples,
-        uint8_t, layerCount,
        backend::MRT, color,
        backend::TargetBufferInfo, depth,
        backend::TargetBufferInfo, stencil)
@@ -264,7 +264,6 @@ DECL_DRIVER_API_R_0(backend::TimerQueryHandle, createTimerQuery)
 */

 DECL_DRIVER_API_N(destroyVertexBuffer,    backend::VertexBufferHandle, vbh)
-DECL_DRIVER_API_N(destroyVertexBufferInfo,backend::VertexBufferInfoHandle, vbih)
 DECL_DRIVER_API_N(destroyIndexBuffer,     backend::IndexBufferHandle, ibh)
 DECL_DRIVER_API_N(destroyBufferObject,    backend::BufferObjectHandle, ibh)
 DECL_DRIVER_API_N(destroyRenderPrimitive, backend::RenderPrimitiveHandle, rph)
@@ -299,16 +298,14 @@ DECL_DRIVER_API_SYNCHRONOUS_0(bool, isFrameBufferFetchMultiSampleSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isFrameTimeSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isAutoDepthResolveSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isSRGBSwapChainSupported)
-DECL_DRIVER_API_SYNCHRONOUS_0(bool, isProtectedContentSupported)
-DECL_DRIVER_API_SYNCHRONOUS_N(bool, isStereoSupported, backend::StereoscopicType, stereoscopicType)
+DECL_DRIVER_API_SYNCHRONOUS_0(bool, isStereoSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isParallelShaderCompileSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isDepthStencilResolveSupported)
-DECL_DRIVER_API_SYNCHRONOUS_0(bool, isProtectedTexturesSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(uint8_t, getMaxDrawBuffers)
 DECL_DRIVER_API_SYNCHRONOUS_0(size_t, getMaxUniformBufferSize)
 DECL_DRIVER_API_SYNCHRONOUS_0(math::float2, getClipSpaceParams)
 DECL_DRIVER_API_SYNCHRONOUS_N(void, setupExternalImage, void*, image)
-DECL_DRIVER_API_SYNCHRONOUS_N(backend::TimerQueryResult, getTimerQueryValue, backend::TimerQueryHandle, query, uint64_t*, elapsedTime)
+DECL_DRIVER_API_SYNCHRONOUS_N(bool, getTimerQueryValue, backend::TimerQueryHandle, query, uint64_t*, elapsedTime)
 DECL_DRIVER_API_SYNCHRONOUS_N(bool, isWorkaroundNeeded, backend::Workaround, workaround)
 DECL_DRIVER_API_SYNCHRONOUS_0(backend::FeatureLevel, getFeatureLevel)

@@ -494,32 +491,15 @@ DECL_DRIVER_API_N(blit,
        math::uint2, srcOrigin,
        math::uint2, size)

-DECL_DRIVER_API_N(bindPipeline,
-        backend::PipelineState, state)
-
-DECL_DRIVER_API_N(bindRenderPrimitive,
-        backend::RenderPrimitiveHandle, rph)
-
-DECL_DRIVER_API_N(draw2,
-        uint32_t, indexOffset,
-        uint32_t, indexCount,
-        uint32_t, instanceCount)
-
 DECL_DRIVER_API_N(draw,
        backend::PipelineState, state,
        backend::RenderPrimitiveHandle, rph,
-        uint32_t, indexOffset,
-        uint32_t, indexCount,
        uint32_t, instanceCount)

-
 DECL_DRIVER_API_N(dispatchCompute,
        backend::ProgramHandle, program,
        math::uint3, workGroupCount)

-DECL_DRIVER_API_N(scissor,
-        Viewport, scissor)
-

 #pragma clang diagnostic pop

@@ -22,34 +22,29 @@
 #include <utils/Allocator.h>
 #include <utils/Log.h>
 #include <utils/compiler.h>
-#include <utils/debug.h>
-#include <utils/ostream.h>
-#include <utils/Panic.h>
-
 #include <tsl/robin_map.h>
-
-#include <cstddef>
-#include <exception>
-#include <type_traits>
 #include <unordered_map>
-#include <utility>

-#include <stddef.h>
-#include <stdint.h>
+#if !defined(NDEBUG) && UTILS_HAS_RTTI
+#   define HANDLE_TYPE_SAFETY 1
+#else
+#   define HANDLE_TYPE_SAFETY 0
+#endif

-#define HandleAllocatorGL  HandleAllocator<32,  64, 136>    // ~4520 / pool / MiB
-#define HandleAllocatorVK  HandleAllocator<64, 160, 312>    // ~1820 / pool / MiB
-#define HandleAllocatorMTL HandleAllocator<32,  48, 552>    // ~1660 / pool / MiB
+#define HandleAllocatorGL  HandleAllocator<16, 64, 208>
+#define HandleAllocatorVK  HandleAllocator<16, 64, 880>
+#define HandleAllocatorMTL HandleAllocator<16, 64, 584>

 namespace filament::backend {

 /*
 * A utility class to efficiently allocate and manage Handle<>
 */
-template<size_t P0, size_t P1, size_t P2>
+template <size_t P0, size_t P1, size_t P2>
 class HandleAllocator {
 public:
-    HandleAllocator(const char* name, size_t size, bool disableUseAfterFreeCheck) noexcept;
+
+    HandleAllocator(const char* name, size_t size) noexcept;
    HandleAllocator(HandleAllocator const& rhs) = delete;
    HandleAllocator& operator=(HandleAllocator const& rhs) = delete;
    ~HandleAllocator();
@@ -65,10 +60,15 @@ public:
     *
     */
    template<typename D, typename ... ARGS>
-    Handle<D> allocateAndConstruct(ARGS&& ... args) {
-        Handle<D> h{ allocateHandle<D>() };
+    Handle<D> allocateAndConstruct(ARGS&& ... args) noexcept {
+        Handle<D> h{ allocateHandle<sizeof(D)>() };
        D* addr = handle_cast<D*>(h);
        new(addr) D(std::forward<ARGS>(args)...);
+#if HANDLE_TYPE_SAFETY
+        mLock.lock();
+        mHandleTypeId[addr] = typeid(D).name();
+        mLock.unlock();
+#endif
        return h;
    }

@@ -84,7 +84,13 @@ public:
     */
    template<typename D>
    Handle<D> allocate() noexcept {
-        Handle<D> h{ allocateHandle<D>() };
+        Handle<D> h{ allocateHandle<sizeof(D)>() };
+#if HANDLE_TYPE_SAFETY
+        D* addr = handle_cast<D*>(h);
+        mLock.lock();
+        mHandleTypeId[addr] = typeid(D).name();
+        mLock.unlock();
+#endif
        return h;
    }

@@ -97,14 +103,21 @@ public:
     */
    template<typename D, typename B, typename ... ARGS>
    typename std::enable_if_t<std::is_base_of_v<B, D>, D>*
-    destroyAndConstruct(Handle<B> const& handle, ARGS&& ... args) {
+    destroyAndConstruct(Handle<B> const& handle, ARGS&& ... args) noexcept {
        assert_invariant(handle);
        D* addr = handle_cast<D*>(const_cast<Handle<B>&>(handle));
        assert_invariant(addr);
+
        // currently we implement construct<> with dtor+ctor, we could use operator= also
        // but all our dtors are trivial, ~D() is actually a noop.
        addr->~D();
        new(addr) D(std::forward<ARGS>(args)...);
+
+#if HANDLE_TYPE_SAFETY
+        mLock.lock();
+        mHandleTypeId[addr] = typeid(D).name();
+        mLock.unlock();
+#endif
        return addr;
    }

@@ -121,6 +134,12 @@ public:
        D* addr = handle_cast<D*>(const_cast<Handle<B>&>(handle));
        assert_invariant(addr);
        new(addr) D(std::forward<ARGS>(args)...);
+
+#if HANDLE_TYPE_SAFETY
+        mLock.lock();
+        mHandleTypeId[addr] = typeid(D).name();
+        mLock.unlock();
+#endif
        return addr;
    }

@@ -136,8 +155,19 @@ public:
    void deallocate(Handle<B>& handle, D const* p) noexcept {
        // allow to destroy the nullptr, similarly to operator delete
        if (p) {
+#if HANDLE_TYPE_SAFETY
+            mLock.lock();
+            auto typeId = mHandleTypeId[p];
+            mHandleTypeId.erase(p);
+            mLock.unlock();
+            if (UTILS_UNLIKELY(typeId != typeid(D).name())) {
+                utils::slog.e << "Destroying handle " << handle.getId() << ", type " << typeid(D).name()
+                       << ", but handle's actual type is " << typeId << utils::io::endl;
+                std::terminate();
+            }
+#endif
            p->~D();
-            deallocateHandle<D>(handle.getId());
+            deallocateHandle<sizeof(D)>(handle.getId());
        }
    }

@@ -163,21 +193,9 @@ public:
    inline typename std::enable_if_t<
            std::is_pointer_v<Dp> &&
            std::is_base_of_v<B, typename std::remove_pointer_t<Dp>>, Dp>
-    handle_cast(Handle<B>& handle) {
+    handle_cast(Handle<B>& handle) noexcept {
        assert_invariant(handle);
-        auto [p, tag] = handleToPointer(handle.getId());
-
-        if (isPoolHandle(handle.getId())) {
-            // check for use after free
-            if (UTILS_UNLIKELY(!mUseAfterFreeCheckDisabled)) {
-                uint8_t const age = (tag & HANDLE_AGE_MASK) >> HANDLE_AGE_SHIFT;
-                auto const pNode = static_cast<typename Allocator::Node*>(p);
-                uint8_t const expectedAge = pNode[-1].age;
-                ASSERT_POSTCONDITION(expectedAge == age,
-                        "use-after-free of Handle with id=%d", handle.getId());
-            }
-        }
-
+        void* const p = handleToPointer(handle.getId());
        return static_cast<Dp>(p);
    }

@@ -185,66 +203,36 @@ public:
    inline typename std::enable_if_t<
            std::is_pointer_v<Dp> &&
            std::is_base_of_v<B, typename std::remove_pointer_t<Dp>>, Dp>
-    handle_cast(Handle<B> const& handle) {
+    handle_cast(Handle<B> const& handle) noexcept {
        return handle_cast<Dp>(const_cast<Handle<B>&>(handle));
    }

+
 private:

-    template<typename D>
-    static constexpr size_t getBucketSize() noexcept {
-        if constexpr (sizeof(D) <= P0) { return P0; }
-        if constexpr (sizeof(D) <= P1) { return P1; }
-        static_assert(sizeof(D) <= P2);
-        return P2;
-    }
-
+    // template <int P0, int P1, int P2>
    class Allocator {
        friend class HandleAllocator;
-        static constexpr size_t MIN_ALIGNMENT = alignof(std::max_align_t);
-        struct Node { uint8_t age; };
-        // Note: using the `extra` parameter of PoolAllocator<>, even with a 1-byte structure,
-        // generally increases all pool allocations by 8-bytes because of alignment restrictions.
-        template<size_t SIZE>
-        using Pool = utils::PoolAllocator<SIZE, MIN_ALIGNMENT, sizeof(Node)>;
-        Pool<P0> mPool0;
-        Pool<P1> mPool1;
-        Pool<P2> mPool2;
+        utils::PoolAllocator<P0, 16>   mPool0;
+        utils::PoolAllocator<P1, 16>   mPool1;
+        utils::PoolAllocator<P2, 16>   mPool2;
        UTILS_UNUSED_IN_RELEASE const utils::AreaPolicy::HeapArea& mArea;
-        bool mUseAfterFreeCheckDisabled;
    public:
-        explicit Allocator(const utils::AreaPolicy::HeapArea& area, bool disableUseAfterFreeCheck);
-
-        static constexpr size_t getAlignment() noexcept { return MIN_ALIGNMENT; }
+        static constexpr size_t MIN_ALIGNMENT_SHIFT = 4;
+        explicit Allocator(const utils::AreaPolicy::HeapArea& area);

        // this is in fact always called with a constexpr size argument
-        [[nodiscard]] inline void* alloc(size_t size, size_t, size_t, uint8_t* outAge) noexcept {
+        [[nodiscard]] inline void* alloc(size_t size, size_t alignment, size_t extra) noexcept {
            void* p = nullptr;
-            if      (size <= mPool0.getSize()) p = mPool0.alloc(size);
-            else if (size <= mPool1.getSize()) p = mPool1.alloc(size);
-            else if (size <= mPool2.getSize()) p = mPool2.alloc(size);
-            if (UTILS_LIKELY(p)) {
-                Node const* const pNode = static_cast<Node const*>(p);
-                // we are guaranteed to have at least sizeof<Node> bytes of extra storage before
-                // the allocation address.
-                *outAge = pNode[-1].age;
-            }
+                 if (size <= mPool0.getSize()) p = mPool0.alloc(size, 16, extra);
+            else if (size <= mPool1.getSize()) p = mPool1.alloc(size, 16, extra);
+            else if (size <= mPool2.getSize()) p = mPool2.alloc(size, 16, extra);
            return p;
        }

        // this is in fact always called with a constexpr size argument
-        inline void free(void* p, size_t size, uint8_t age) noexcept {
+        inline void free(void* p, size_t size) noexcept {
            assert_invariant(p >= mArea.begin() && (char*)p + size <= (char*)mArea.end());
-
-            // check for double-free
-            Node* const pNode = static_cast<Node*>(p);
-            uint8_t& expectedAge = pNode[-1].age;
-            if (UTILS_UNLIKELY(!mUseAfterFreeCheckDisabled)) {
-                ASSERT_POSTCONDITION(expectedAge == age,
-                        "double-free of Handle of size %d at %p", size, p);
-            }
-            expectedAge = (expectedAge + 1) & 0xF; // fixme
-
            if (size <= mPool0.getSize()) { mPool0.free(p); return; }
            if (size <= mPool1.getSize()) { mPool1.free(p); return; }
            if (size <= mPool2.getSize()) { mPool2.free(p); return; }
@@ -266,16 +254,24 @@ private:
    // allocateHandle()/deallocateHandle() selects the pool to use at compile-time based on the
    // allocation size this is always inlined, because all these do is to call
    // allocateHandleInPool()/deallocateHandleFromPool() with the right pool size.
-    template<typename D>
+    template<size_t SIZE>
    HandleBase::HandleId allocateHandle() noexcept {
-        constexpr size_t BUCKET_SIZE = getBucketSize<D>();
-        return allocateHandleInPool<BUCKET_SIZE>();
+        if constexpr (SIZE <= P0) { return allocateHandleInPool<P0>(); }
+        if constexpr (SIZE <= P1) { return allocateHandleInPool<P1>(); }
+        static_assert(SIZE <= P2);
+        return allocateHandleInPool<P2>();
    }

-    template<typename D>
+    template<size_t SIZE>
    void deallocateHandle(HandleBase::HandleId id) noexcept {
-        constexpr size_t BUCKET_SIZE = getBucketSize<D>();
-        deallocateHandleFromPool<BUCKET_SIZE>(id);
+        if constexpr (SIZE <= P0) {
+            deallocateHandleFromPool<P0>(id);
+        } else if constexpr (SIZE <= P1) {
+            deallocateHandleFromPool<P1>(id);
+        } else {
+            static_assert(SIZE <= P2);
+            deallocateHandleFromPool<P2>(id);
+        }
    }

    // allocateHandleInPool()/deallocateHandleFromPool() is NOT inlined, which will cause three
@@ -284,11 +280,9 @@ private:
    template<size_t SIZE>
    UTILS_NOINLINE
    HandleBase::HandleId allocateHandleInPool() noexcept {
-        uint8_t age;
-        void* p = mHandleArena.alloc(SIZE, alignof(std::max_align_t), 0, &age);
+        void* p = mHandleArena.alloc(SIZE);
        if (UTILS_LIKELY(p)) {
-            uint32_t const tag = (uint32_t(age) << HANDLE_AGE_SHIFT) & HANDLE_AGE_MASK;
-            return arenaPointerToHandle(p, tag);
+            return pointerToHandle(p);
        } else {
            return allocateHandleSlow(SIZE);
        }
@@ -298,51 +292,42 @@ private:
    UTILS_NOINLINE
    void deallocateHandleFromPool(HandleBase::HandleId id) noexcept {
        if (UTILS_LIKELY(isPoolHandle(id))) {
-            auto [p, tag] = handleToPointer(id);
-            uint8_t const age = (tag & HANDLE_AGE_MASK) >> HANDLE_AGE_SHIFT;
-            mHandleArena.free(p, SIZE, age);
+            void* p = handleToPointer(id);
+            mHandleArena.free(p, SIZE);
        } else {
            deallocateHandleSlow(id, SIZE);
        }
    }

-    // we handle a 4 bits age per address
-    static constexpr uint32_t HANDLE_HEAP_FLAG      = 0x80000000u;      // pool vs heap handle
-    static constexpr uint32_t HANDLE_AGE_MASK       = 0x78000000u;      // handle's age
-    static constexpr uint32_t HANDLE_INDEX_MASK     = 0x07FFFFFFu;      // handle index
-    static constexpr uint32_t HANDLE_TAG_MASK       = HANDLE_AGE_MASK;
-    static constexpr uint32_t HANDLE_AGE_SHIFT      = 27;
+    static constexpr uint32_t HEAP_HANDLE_FLAG = 0x80000000u;

    static bool isPoolHandle(HandleBase::HandleId id) noexcept {
-        return (id & HANDLE_HEAP_FLAG) == 0u;
+        return (id & HEAP_HANDLE_FLAG) == 0u;
    }

-    HandleBase::HandleId allocateHandleSlow(size_t size);
+    HandleBase::HandleId allocateHandleSlow(size_t size) noexcept;
    void deallocateHandleSlow(HandleBase::HandleId id, size_t size) noexcept;

    // We inline this because it's just 4 instructions in the fast case
-    inline std::pair<void*, uint32_t> handleToPointer(HandleBase::HandleId id) const noexcept {
+    inline void* handleToPointer(HandleBase::HandleId id) const noexcept {
        // note: the null handle will end-up returning nullptr b/c it'll be handled as
        // a non-pool handle.
        if (UTILS_LIKELY(isPoolHandle(id))) {
            char* const base = (char*)mHandleArena.getArea().begin();
-            uint32_t const tag = id & HANDLE_TAG_MASK;
-            size_t const offset = (id & HANDLE_INDEX_MASK) * Allocator::getAlignment();
-            return { static_cast<void*>(base + offset), tag };
+            size_t offset = id << Allocator::MIN_ALIGNMENT_SHIFT;
+            return static_cast<void*>(base + offset);
        }
-        return { handleToPointerSlow(id), 0 };
+        return handleToPointerSlow(id);
    }

    void* handleToPointerSlow(HandleBase::HandleId id) const noexcept;

    // We inline this because it's just 3 instructions
-    inline HandleBase::HandleId arenaPointerToHandle(void* p, uint32_t tag) const noexcept {
+    inline HandleBase::HandleId pointerToHandle(void* p) const noexcept {
        char* const base = (char*)mHandleArena.getArea().begin();
-        size_t const offset = (char*)p - base;
-        assert_invariant((offset % Allocator::getAlignment()) == 0);
-        auto id = HandleBase::HandleId(offset / Allocator::getAlignment());
-        id |= tag & HANDLE_TAG_MASK;
-        assert_invariant((id & HANDLE_HEAP_FLAG) == 0);
+        size_t offset = (char*)p - base;
+        auto id = HandleBase::HandleId(offset >> Allocator::MIN_ALIGNMENT_SHIFT);
+        assert_invariant((id & HEAP_HANDLE_FLAG) == 0);
        return id;
    }

@@ -352,7 +337,9 @@ private:
    mutable utils::Mutex mLock;
    tsl::robin_map<HandleBase::HandleId, void*> mOverflowMap;
    HandleBase::HandleId mId = 0;
-    bool mUseAfterFreeCheckDisabled = false;
+#if HANDLE_TYPE_SAFETY
+    mutable std::unordered_map<const void*, const char*> mHandleTypeId;
+#endif
 };

 } // namespace filament::backend
@@ -19,9 +19,9 @@
 #ifndef TNT_FILAMENT_BACKEND_PLATFORM_FACTORY_H
 #define TNT_FILAMENT_BACKEND_PLATFORM_FACTORY_H

-#include <backend/DriverEnums.h>
+#include "backend/DriverEnums.h"

-#include <utils/compiler.h>
+#include "utils/compiler.h"

 namespace filament::backend {

@@ -17,12 +17,14 @@
 #ifndef TNT_FILAMENT_BACKEND_PRIVATE_SAMPLERGROUP_H
 #define TNT_FILAMENT_BACKEND_PRIVATE_SAMPLERGROUP_H

-#include <backend/DriverApiForward.h>
-#include <backend/DriverEnums.h>
-#include <backend/SamplerDescriptor.h>
+#include "backend/DriverApiForward.h"

+#include <utils/compiler.h>
 #include <utils/FixedCapacityVector.h>
-#include <utils/ostream.h>
+
+#include <backend/DriverEnums.h>
+#include <backend/Handle.h>
+#include <backend/SamplerDescriptor.h>

 #include <stddef.h>

@@ -16,14 +16,6 @@

 #include "private/backend/CircularBuffer.h"

-#include <utils/Log.h>
-#include <utils/Panic.h>
-#include <utils/architecture.h>
-#include <utils/ashmem.h>
-#include <utils/compiler.h>
-#include <utils/debug.h>
-#include <utils/ostream.h>
-
 #if !defined(WIN32) && !defined(__EMSCRIPTEN__) && !defined(IOS)
 #    include <sys/mman.h>
 #    include <unistd.h>
@@ -32,20 +24,23 @@
 #    define HAS_MMAP 0
 #endif

-#include <stdint.h>
-#include <stddef.h>
-#include <stdlib.h>
 #include <stdio.h>

+#include <utils/architecture.h>
+#include <utils/ashmem.h>
+#include <utils/debug.h>
+#include <utils/Log.h>
+#include <utils/Panic.h>
+
 using namespace utils;

 namespace filament::backend {

 size_t CircularBuffer::sPageSize = arch::getPageSize();

-CircularBuffer::CircularBuffer(size_t size)
-    : mSize(size) {
+CircularBuffer::CircularBuffer(size_t size) {
    mData = alloc(size);
+    mSize = size;
    mTail = mData;
    mHead = mData;
 }
@@ -90,7 +85,7 @@ void* CircularBuffer::alloc(size_t size) noexcept {
                            MAP_PRIVATE, fd, (off_t)size);
                    if (vaddr_guard != MAP_FAILED && (vaddr_guard == (char*)vaddr_shadow + size)) {
                        // woo-hoo success!
-                        mAshmemFd = fd;
+                        mUsesAshmem = fd;
                        data = vaddr;
                    }
                }
@@ -98,7 +93,7 @@ void* CircularBuffer::alloc(size_t size) noexcept {
        }
    }

-    if (UTILS_UNLIKELY(mAshmemFd < 0)) {
+    if (UTILS_UNLIKELY(mUsesAshmem < 0)) {
        // ashmem failed
        if (vaddr_guard != MAP_FAILED) {
            munmap(vaddr_guard, size);
@@ -142,9 +137,9 @@ void CircularBuffer::dealloc() noexcept {
    if (mData) {
        size_t const BLOCK_SIZE = getBlockSize();
        munmap(mData, mSize * 2 + BLOCK_SIZE);
-        if (mAshmemFd >= 0) {
-            close(mAshmemFd);
-            mAshmemFd = -1;
+        if (mUsesAshmem >= 0) {
+            close(mUsesAshmem);
+            mUsesAshmem = -1;
        }
    }
 #else
@@ -154,37 +149,23 @@ void CircularBuffer::dealloc() noexcept {
 }


-CircularBuffer::Range CircularBuffer::getBuffer() noexcept {
-    Range const range{ .tail = mTail, .head = mHead };
-
-    char* const pData = static_cast<char*>(mData);
-    char const* const pEnd = pData + mSize;
-    char const* const pHead = static_cast<char const*>(mHead);
-    if (UTILS_UNLIKELY(pHead >= pEnd)) {
-        size_t const overflow = pHead - pEnd;
-        if (UTILS_LIKELY(mAshmemFd > 0)) {
-            assert_invariant(overflow <= mSize);
-            mHead = static_cast<void*>(pData + overflow);
-            // Data         Tail  End   Head              [virtual]
-            //  v             v    v     v
-            //  +-------------:----+-----:--------------+
-            //  |             :    |     :              |
-            //  +-----:------------+--------------------+
-            //       Head          |<------ copy ------>| [physical]
-        } else {
-            // Data         Tail  End   Head
-            //  v             v    v     v
-            //  +-------------:----+-----:--------------+
-            //  |             :    |     :              |
-            //  +-----|------------+-----|--------------+
-            //        |<---------------->|
-            //           sliding window
+void CircularBuffer::circularize() noexcept {
+    if (mUsesAshmem > 0) {
+        intptr_t const overflow = intptr_t(mHead) - (intptr_t(mData) + ssize_t(mSize));
+        if (overflow >= 0) {
+            assert_invariant(size_t(overflow) <= mSize);
+            mHead = (void *) (intptr_t(mData) + overflow);
+            #ifndef NDEBUG
+            memset(mData, 0xA5, size_t(overflow));
+            #endif
+        }
+    } else {
+        // Only circularize if mHead if in the second buffer.
+        if (intptr_t(mHead) - intptr_t(mData) > ssize_t(mSize)) {
            mHead = mData;
        }
    }
    mTail = mHead;
-
-    return range;
 }

 } // namespace filament::backend
@@ -15,35 +15,23 @@
 */

 #include "private/backend/CommandBufferQueue.h"
-#include "private/backend/CircularBuffer.h"
-#include "private/backend/CommandStream.h"

-#include <utils/compiler.h>
 #include <utils/Log.h>
-#include <utils/Mutex.h>
-#include <utils/ostream.h>
-#include <utils/Panic.h>
 #include <utils/Systrace.h>
+#include <utils/Panic.h>
 #include <utils/debug.h>

-#include <algorithm>
-#include <mutex>
-#include <iterator>
-#include <utility>
-#include <vector>
-
-#include <stddef.h>
-#include <stdint.h>
+#include "private/backend/BackendUtils.h"
+#include "private/backend/CommandStream.h"

 using namespace utils;

 namespace filament::backend {

-CommandBufferQueue::CommandBufferQueue(size_t requiredSize, size_t bufferSize, bool paused)
+CommandBufferQueue::CommandBufferQueue(size_t requiredSize, size_t bufferSize)
        : mRequiredSize((requiredSize + (CircularBuffer::getBlockSize() - 1u)) & ~(CircularBuffer::getBlockSize() -1u)),
          mCircularBuffer(bufferSize),
-          mFreeSpace(mCircularBuffer.size()),
-          mPaused(paused) {
+          mFreeSpace(mCircularBuffer.size()) {
    assert_invariant(mCircularBuffer.size() > requiredSize);
 }

@@ -57,16 +45,6 @@ void CommandBufferQueue::requestExit() {
    mCondition.notify_one();
 }

-void CommandBufferQueue::setPaused(bool paused) {
-    std::lock_guard<utils::Mutex> const lock(mLock);
-    if (paused) {
-        mPaused = true;
-    } else {
-        mPaused = false;
-        mCondition.notify_one();
-    }
-}
-
 bool CommandBufferQueue::isExitRequested() const {
    std::lock_guard<utils::Mutex> const lock(mLock);
    ASSERT_PRECONDITION( mExitRequested == 0 || mExitRequested == EXIT_REQUESTED,
@@ -87,61 +65,55 @@ void CommandBufferQueue::flush() noexcept {
    // always guaranteed to have enough space for the NoopCommand
    new(circularBuffer.allocate(sizeof(NoopCommand))) NoopCommand(nullptr);

-    const size_t requiredSize = mRequiredSize;
+    // end of this slice
+    void* const head = circularBuffer.getHead();

-    // get the current buffer
-    auto const [begin, end] = circularBuffer.getBuffer();
+    // beginning of this slice
+    void* const tail = circularBuffer.getTail();

-    assert_invariant(circularBuffer.empty());
+    // size of this slice
+    uint32_t const used = uint32_t(intptr_t(head) - intptr_t(tail));

-    // size of the current buffer
-    size_t const used = std::distance(
-            static_cast<char const*>(begin), static_cast<char const*>(end));
+    circularBuffer.circularize();

    std::unique_lock<utils::Mutex> lock(mLock);
-    mCommandBuffersToExecute.push_back({ begin, end });
-    mCondition.notify_one();
+    mCommandBuffersToExecute.push_back({ tail, head });

    // circular buffer is too small, we corrupted the stream
    ASSERT_POSTCONDITION(used <= mFreeSpace,
            "Backend CommandStream overflow. Commands are corrupted and unrecoverable.\n"
            "Please increase minCommandBufferSizeMB inside the Config passed to Engine::create.\n"
-            "Space used at this time: %u bytes, overflow: %u bytes",
-            (unsigned)used, unsigned(used - mFreeSpace));
+            "Space used at this time: %u bytes",
+            (unsigned)used);

    // wait until there is enough space in the buffer
    mFreeSpace -= used;
-    if (UTILS_UNLIKELY(mFreeSpace < requiredSize)) {
-
+    const size_t requiredSize = mRequiredSize;

 #ifndef NDEBUG
-        size_t const totalUsed = circularBuffer.size() - mFreeSpace;
-        slog.d << "CommandStream used too much space (will block): "
-                << "needed space " << requiredSize << " out of " << mFreeSpace
-                << ", totalUsed=" << totalUsed << ", current=" << used
-                << ", queue size=" << mCommandBuffersToExecute.size() << " buffers"
-                << io::endl;
-
-        mHighWatermark = std::max(mHighWatermark, totalUsed);
+    size_t totalUsed = circularBuffer.size() - mFreeSpace;
+    mHighWatermark = std::max(mHighWatermark, totalUsed);
+    if (UTILS_UNLIKELY(totalUsed > requiredSize)) {
+        slog.d << "CommandStream used too much space: " << totalUsed
+            << ", out of " << requiredSize << " (will block)" << io::endl;
+    }
 #endif

+    mCondition.notify_one();
+    if (UTILS_LIKELY(mFreeSpace < requiredSize)) {
        SYSTRACE_NAME("waiting: CircularBuffer::flush()");
-        ASSERT_POSTCONDITION(!mPaused,
-                "CommandStream is full, but since the rendering thread is paused, "
-                "the buffer cannot flush and we will deadlock. Instead, abort.");
        mCondition.wait(lock, [this, requiredSize]() -> bool {
-            // TODO: on macOS, we need to call pumpEvents from time to time
            return mFreeSpace >= requiredSize;
        });
    }
 }

-std::vector<CommandBufferQueue::Range> CommandBufferQueue::waitForCommands() const {
+std::vector<CommandBufferQueue::Slice> CommandBufferQueue::waitForCommands() const {
    if (!UTILS_HAS_THREADING) {
        return std::move(mCommandBuffersToExecute);
    }
    std::unique_lock<utils::Mutex> lock(mLock);
-    while ((mCommandBuffersToExecute.empty() || mPaused) && !mExitRequested) {
+    while (mCommandBuffersToExecute.empty() && !mExitRequested) {
        mCondition.wait(lock);
    }

@@ -151,7 +123,7 @@ std::vector<CommandBufferQueue::Range> CommandBufferQueue::waitForCommands() con
    return std::move(mCommandBuffersToExecute);
 }

-void CommandBufferQueue::releaseBuffer(CommandBufferQueue::Range const& buffer) {
+void CommandBufferQueue::releaseBuffer(CommandBufferQueue::Slice const& buffer) {
    std::lock_guard<utils::Mutex> const lock(mLock);
    mFreeSpace += uintptr_t(buffer.end) - uintptr_t(buffer.begin);
    mCondition.notify_one();
@@ -149,7 +149,7 @@ void CommandType<void (Driver::*)(ARGS...)>::Command<METHOD>::log() noexcept  {

 // ------------------------------------------------------------------------------------------------

-void CustomCommand::execute(Driver&, CommandBase* base, intptr_t* next) {
+void CustomCommand::execute(Driver&, CommandBase* base, intptr_t* next) noexcept {
    *next = CustomCommand::align(sizeof(CustomCommand));
    static_cast<CustomCommand*>(base)->mCommand();
    static_cast<CustomCommand*>(base)->~CustomCommand();
@@ -19,7 +19,6 @@

 #include <backend/PixelBufferDescriptor.h>

-#include <cstring>
 #include <stddef.h>
 #include <stdint.h>

@@ -21,12 +21,7 @@

 #include <backend/AcquiredImage.h>
 #include <backend/BufferDescriptor.h>
-#include <backend/DriverEnums.h>

-#include <utils/compiler.h>
-#include <utils/debug.h>
-#include <utils/Log.h>
-#include <utils/ostream.h>
 #include <utils/Systrace.h>

 #include <math/half.h>
@@ -34,13 +29,6 @@
 #include <math/vec3.h>
 #include <math/vec4.h>

-#include <functional>
-#include <mutex>
-#include <utility>
-
-#include <stddef.h>
-#include <stdint.h>
-
 using namespace utils;
 using namespace filament::math;

@@ -131,8 +119,7 @@ void DriverBase::purge() noexcept {
 // ------------------------------------------------------------------------------------------------

 void DriverBase::scheduleDestroySlow(BufferDescriptor&& buffer) noexcept {
-    auto const handler = buffer.getHandler();
-    scheduleCallback(handler, [buffer = std::move(buffer)]() {
+    scheduleCallback(buffer.getHandler(), [buffer = std::move(buffer)]() {
        // user callback is called when BufferDescriptor gets destroyed
    });
 }
@@ -214,7 +201,7 @@ size_t Driver::getElementTypeSize(ElementType type) noexcept {

 Driver::~Driver() noexcept = default;

-void Driver::execute(std::function<void(void)> const& fn) {
+void Driver::execute(std::function<void(void)> const& fn) noexcept {
    fn();
 }

@@ -49,25 +49,21 @@ struct AcquiredImage;
 struct HwBase {
 };

-
-struct HwVertexBufferInfo : public HwBase {
+struct HwVertexBuffer : public HwBase {
+    AttributeArray attributes{};          // 8 * MAX_VERTEX_ATTRIBUTE_COUNT
+    uint32_t vertexCount{};               //   4
    uint8_t bufferCount{};                //   1
    uint8_t attributeCount{};             //   1
-    bool padding[2]{};                    //   2
-    HwVertexBufferInfo() noexcept = default;
-    HwVertexBufferInfo(uint8_t bufferCount, uint8_t attributeCount) noexcept
-            : bufferCount(bufferCount),
-              attributeCount(attributeCount) {
-    }
-};
+    bool padding{};                       //   1
+    uint8_t bufferObjectsVersion{};       //   1 -> total struct is 136 bytes

-struct HwVertexBuffer : public HwBase {
-    uint32_t vertexCount{};               //   4
-    uint8_t bufferObjectsVersion{0xff};   //   1
-    bool padding[3]{};                    //   2
    HwVertexBuffer() noexcept = default;
-    explicit HwVertexBuffer(uint32_t vertextCount) noexcept
-            : vertexCount(vertextCount) {
+    HwVertexBuffer(uint8_t bufferCount, uint8_t attributeCount, uint32_t elementCount,
+            AttributeArray const& attributes) noexcept
+            : attributes(attributes),
+              vertexCount(elementCount),
+              bufferCount(bufferCount),
+              attributeCount(attributeCount) {
    }
 };

@@ -92,6 +88,11 @@ struct HwIndexBuffer : public HwBase {
 };

 struct HwRenderPrimitive : public HwBase {
+    uint32_t offset{};
+    uint32_t minIndex{};
+    uint32_t maxIndex{};
+    uint32_t count{};
+    uint32_t maxVertexCount{};
    PrimitiveType type = PrimitiveType::TRIANGLES;
 };

@@ -113,9 +114,7 @@ struct HwTexture : public HwBase {
    uint8_t levels : 4;  // This allows up to 15 levels (max texture size of 32768 x 32768)
    uint8_t samples : 4; // Sample count per pixel (should always be a power of 2)
    TextureFormat format{};
-    uint8_t reserved0 = 0;
    TextureUsage usage{};
-    uint16_t reserved1 = 0;
    HwStream* hwStream = nullptr;

    HwTexture() noexcept : levels{}, samples{} {}
@@ -16,22 +16,9 @@

 #include "private/backend/HandleAllocator.h"

-#include <backend/Handle.h>
-
-#include <utils/Allocator.h>
-#include <utils/Log.h>
 #include <utils/Panic.h>
-#include <utils/compiler.h>
-#include <utils/debug.h>
-#include <utils/ostream.h>
-
-#include <algorithm>
-#include <exception>
-#include <limits>
-#include <mutex>

 #include <stdlib.h>
-#include <string.h>

 namespace filament::backend {

@@ -39,47 +26,23 @@ using namespace utils;

 template <size_t P0, size_t P1, size_t P2>
 UTILS_NOINLINE
-HandleAllocator<P0, P1, P2>::Allocator::Allocator(AreaPolicy::HeapArea const& area,
-        bool disableUseAfterFreeCheck)
-        : mArea(area),
-          mUseAfterFreeCheckDisabled(disableUseAfterFreeCheck) {
-
-    // The largest handle this allocator can generate currently depends on the architecture's
-    // min alignment, typically 8 or 16 bytes.
-    // e.g. On Android armv8, the alignment is 16 bytes, so for a 1 MiB heap, the largest handle
-    //      index will be 65536. Note that this is not the same as the number of handles (which
-    //      will always be less).
-    // Because our maximum representable handle currently is 0x07FFFFFF, the maximum no-nonsensical
-    // heap size is 2 GiB, which amounts to 7.6 millions handles per pool (in the GL case).
-    size_t const maxHeapSize = std::min(area.size(), HANDLE_INDEX_MASK * getAlignment());
-
-    if (UTILS_UNLIKELY(maxHeapSize != area.size())) {
-        slog.w << "HandleAllocator heap size reduced to "
-               << maxHeapSize << " from " << area.size() << io::endl;
-    }
-
-    // make sure we start with a clean arena. This is needed to ensure that all blocks start
-    // with an age of 0.
-    memset(area.data(), 0, maxHeapSize);
-
-    // size the different pools so that they can all contain the same number of handles
-    size_t const count = maxHeapSize / (P0 + P1 + P2);
-    char* const p0 = static_cast<char*>(area.begin());
-    char* const p1 = p0 + count * P0;
-    char* const p2 = p1 + count * P1;
-
-    mPool0 = Pool<P0>(p0, count * P0);
-    mPool1 = Pool<P1>(p1, count * P1);
-    mPool2 = Pool<P2>(p2, count * P2);
+HandleAllocator<P0, P1, P2>::Allocator::Allocator(AreaPolicy::HeapArea const& area)
+        : mArea(area) {
+    // TODO: we probably need a better way to set the size of these pools
+    const size_t unit = area.size() / 32;
+    const size_t offsetPool1 =      unit;
+    const size_t offsetPool2 = 16 * unit;
+    char* const p = (char*)area.begin();
+    mPool0 = PoolAllocator< P0, 16>(p, p + offsetPool1);
+    mPool1 = PoolAllocator< P1, 16>(p + offsetPool1, p + offsetPool2);
+    mPool2 = PoolAllocator< P2, 16>(p + offsetPool2, area.end());
 }

 // ------------------------------------------------------------------------------------------------

 template <size_t P0, size_t P1, size_t P2>
-HandleAllocator<P0, P1, P2>::HandleAllocator(const char* name, size_t size,
-        bool disableUseAfterFreeCheck) noexcept
-    : mHandleArena(name, size, disableUseAfterFreeCheck),
-      mUseAfterFreeCheckDisabled(disableUseAfterFreeCheck) {
+HandleAllocator<P0, P1, P2>::HandleAllocator(const char* name, size_t size) noexcept
+    : mHandleArena(name, size) {
 }

 template <size_t P0, size_t P1, size_t P2>
@@ -107,20 +70,14 @@ void* HandleAllocator<P0, P1, P2>::handleToPointerSlow(HandleBase::HandleId id)
 }

 template <size_t P0, size_t P1, size_t P2>
-HandleBase::HandleId HandleAllocator<P0, P1, P2>::allocateHandleSlow(size_t size) {
+HandleBase::HandleId HandleAllocator<P0, P1, P2>::allocateHandleSlow(size_t size) noexcept {
    void* p = ::malloc(size);
    std::unique_lock lock(mLock);
-
-    HandleBase::HandleId id = (++mId) | HANDLE_HEAP_FLAG;
-
-    ASSERT_POSTCONDITION(mId < HANDLE_HEAP_FLAG,
-            "No more Handle ids available! This can happen if HandleAllocator arena has been full"
-            " for a while. Please increase FILAMENT_OPENGL_HANDLE_ARENA_SIZE_IN_MB");
-
+    HandleBase::HandleId id = (++mId) | HEAP_HANDLE_FLAG;
    mOverflowMap.emplace(id, p);
    lock.unlock();

-    if (UTILS_UNLIKELY(id == (HANDLE_HEAP_FLAG | 1u))) { // meaning id was zero
+    if (UTILS_UNLIKELY(id == (HEAP_HANDLE_FLAG|1u))) { // meaning id was zero
        PANIC_LOG("HandleAllocator arena is full, using slower system heap. Please increase "
                  "the appropriate constant (e.g. FILAMENT_OPENGL_HANDLE_ARENA_SIZE_IN_MB).");
    }
@@ -129,7 +86,7 @@ HandleBase::HandleId HandleAllocator<P0, P1, P2>::allocateHandleSlow(size_t size

 template <size_t P0, size_t P1, size_t P2>
 void HandleAllocator<P0, P1, P2>::deallocateHandleSlow(HandleBase::HandleId id, size_t) noexcept {
-    assert_invariant(id & HANDLE_HEAP_FLAG);
+    assert_invariant(id & HEAP_HANDLE_FLAG);
    void* p = nullptr;
    auto& overflowMap = mOverflowMap;

@@ -53,18 +53,4 @@ size_t Platform::retrieveBlob(void const* key, size_t keySize, void* value, size
    return 0;
 }

-void Platform::setDebugUpdateStatFunc(DebugUpdateStatFunc&& debugUpdateStat) noexcept {
-    mDebugUpdateStat = std::move(debugUpdateStat);
-}
-
-bool Platform::hasDebugUpdateStatFunc() const noexcept {
-    return bool(mDebugUpdateStat);
-}
-
-void Platform::debugUpdateStat(const char* key, uint64_t value) {
-    if (mDebugUpdateStat) {
-        mDebugUpdateStat(key, value);
-    }
-}
-
 } // namespace filament::backend
@@ -91,11 +91,6 @@ Program& Program::cacheId(uint64_t cacheId) noexcept {
    return *this;
 }

-Program& Program::multiview(bool multiview) noexcept {
-    mMultiview = multiview;
-    return *this;
-}
-
 io::ostream& operator<<(io::ostream& out, const Program& builder) {
    out << "Program{";
    builder.mLogger(out);
@@ -20,7 +20,6 @@
 #include "MetalUtils.h"

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

 namespace filament::backend {

@@ -18,6 +18,7 @@
 #define TNT_FILAMENT_DRIVER_METALBUFFER_H

 #include "MetalContext.h"
+#include "MetalBufferPool.h"

 #include <backend/DriverEnums.h>

@@ -27,50 +28,9 @@

 #include <utility>
 #include <memory>
-#include <atomic>

 namespace filament::backend {

-class TrackedMetalBuffer {
-public:
-    TrackedMetalBuffer() noexcept : mBuffer(nil) {}
-    TrackedMetalBuffer(id<MTLBuffer> buffer) noexcept : mBuffer(buffer) {
-        if (buffer) {
-            aliveBuffers++;
-        }
-    }
-    ~TrackedMetalBuffer() {
-        if (mBuffer) {
-            aliveBuffers--;
-        }
-    }
-
-    TrackedMetalBuffer(TrackedMetalBuffer&&) = delete;
-    TrackedMetalBuffer(TrackedMetalBuffer const&) = delete;
-    TrackedMetalBuffer& operator=(TrackedMetalBuffer const&) = delete;
-
-    TrackedMetalBuffer& operator=(TrackedMetalBuffer&& rhs) noexcept {
-        swap(rhs);
-        return *this;
-    }
-
-    id<MTLBuffer> get() const noexcept { return mBuffer; }
-    operator bool() const noexcept { return bool(mBuffer); }
-
-    static uint64_t getAliveBuffers() { return aliveBuffers; }
-
-private:
-    void swap(TrackedMetalBuffer& other) noexcept {
-        id<MTLBuffer> temp = mBuffer;
-        mBuffer = other.mBuffer;
-        other.mBuffer = temp;
-    }
-
-    id<MTLBuffer> mBuffer;
-
-    static std::atomic<uint64_t> aliveBuffers;
-};
-
 class MetalBuffer {
 public:

@@ -122,7 +82,7 @@ public:

 private:

-    TrackedMetalBuffer mBuffer;
+    id<MTLBuffer> mBuffer = nil;
    size_t mBufferSize = 0;
    void* mCpuBuffer = nullptr;
    MetalContext& mContext;
@@ -191,7 +151,7 @@ public:
            // finishes executing.
            mAuxBuffer = [mDevice newBufferWithLength:mSlotSizeBytes options:mBufferOptions];
            assert_invariant(mAuxBuffer);
-            return {mAuxBuffer.get(), 0};
+            return {mAuxBuffer, 0};
        }
        mCurrentSlot = (mCurrentSlot + 1) % mSlotCount;
        mOccupiedSlots->fetch_add(1, std::memory_order_relaxed);
@@ -220,9 +180,9 @@ public:
     */
    std::pair<id<MTLBuffer>, NSUInteger> getCurrentAllocation() const {
        if (UTILS_UNLIKELY(mAuxBuffer)) {
-            return { mAuxBuffer.get(), 0 };
+            return { mAuxBuffer, 0 };
        }
-        return { mBuffer.get(), mCurrentSlot * mSlotSizeBytes };
+        return { mBuffer, mCurrentSlot * mSlotSizeBytes };
    }

    bool canAccomodateLayout(MTLSizeAndAlign layout) const {
@@ -231,8 +191,8 @@ public:

 private:
    id<MTLDevice> mDevice;
-    TrackedMetalBuffer mBuffer;
-    TrackedMetalBuffer mAuxBuffer;
+    id<MTLBuffer> mBuffer;
+    id<MTLBuffer> mAuxBuffer;

    MTLResourceOptions mBufferOptions;

@@ -15,15 +15,12 @@
 */

 #include "MetalBuffer.h"
-#include "MetalBufferPool.h"

 #include "MetalContext.h"

 namespace filament {
 namespace backend {

-std::atomic<uint64_t> TrackedMetalBuffer::aliveBuffers = 0;
-
 MetalBuffer::MetalBuffer(MetalContext& context, BufferObjectBinding bindingType, BufferUsage usage,
        size_t size, bool forceGpuBuffer) : mBufferSize(size), mContext(context) {
    // If the buffer is less than 4K in size and is updated frequently, we don't use an explicit
@@ -64,7 +61,7 @@ void MetalBuffer::copyIntoBuffer(void* src, size_t size, size_t byteOffset) {
    // Acquire a staging buffer to hold the contents of this update.
    MetalBufferPool* bufferPool = mContext.bufferPool;
    const MetalBufferPoolEntry* const staging = bufferPool->acquireBuffer(size);
-    memcpy(staging->buffer.get().contents, src, size);
+    memcpy(staging->buffer.contents, src, size);

    // The blit below requires that byteOffset be a multiple of 4.
    ASSERT_PRECONDITION(!(byteOffset & 0x3u), "byteOffset must be a multiple of 4");
@@ -73,9 +70,9 @@ void MetalBuffer::copyIntoBuffer(void* src, size_t size, size_t byteOffset) {
    id<MTLCommandBuffer> cmdBuffer = getPendingCommandBuffer(&mContext);
    id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
    blitEncoder.label = @"Buffer upload blit";
-    [blitEncoder copyFromBuffer:staging->buffer.get()
+    [blitEncoder copyFromBuffer:staging->buffer
                   sourceOffset:0
-                       toBuffer:mBuffer.get()
+                       toBuffer:mBuffer
              destinationOffset:byteOffset
                           size:size];
    [blitEncoder endEncoding];
@@ -96,7 +93,7 @@ id<MTLBuffer> MetalBuffer::getGpuBufferForDraw(id<MTLCommandBuffer> cmdBuffer) n
        return nil;
    }
    assert_invariant(mBuffer);
-    return mBuffer.get();
+    return mBuffer;
 }

 void MetalBuffer::bindBuffers(id<MTLCommandBuffer> cmdBuffer, id<MTLCommandEncoder> encoder,
@@ -19,8 +19,6 @@

 #include <Metal/Metal.h>

-#include "MetalBuffer.h"
-
 #include <map>
 #include <mutex>
 #include <unordered_set>
@@ -32,7 +30,7 @@ struct MetalContext;

 // Immutable POD representing a shared CPU-GPU buffer.
 struct MetalBufferPoolEntry {
-    TrackedMetalBuffer buffer;
+    id<MTLBuffer> buffer;
    size_t capacity;
    mutable uint64_t lastAccessed;
    mutable uint32_t referenceCount;
@@ -19,7 +19,6 @@
 #include "MetalContext.h"

 #include <utils/Panic.h>
-#include <utils/Log.h>
 #include <utils/trap.h>

 #include <thread>
@@ -45,12 +44,12 @@ MetalBufferPoolEntry const* MetalBufferPool::acquireBuffer(size_t numBytes) {
    id<MTLBuffer> buffer = [mContext.device newBufferWithLength:numBytes
                                                        options:MTLResourceStorageModeShared];
    ASSERT_POSTCONDITION(buffer, "Could not allocate Metal staging buffer of size %zu.", numBytes);
-    MetalBufferPoolEntry* stage = new MetalBufferPoolEntry {
+    MetalBufferPoolEntry* stage = new MetalBufferPoolEntry({
        .buffer = buffer,
        .capacity = numBytes,
        .lastAccessed = mCurrentFrame,
        .referenceCount = 1
-    };
+    });
    mUsedStages.insert(stage);

    return stage;
@@ -99,7 +99,6 @@ struct MetalContext {
    std::array<BufferState, MAX_SSBO_COUNT> ssboState;
    CullModeStateTracker cullModeState;
    WindingStateTracker windingState;
-    Handle<HwRenderPrimitive> currentRenderPrimitive;

    // State caches.
    DepthStencilStateCache depthStencilStateCache;
@@ -133,9 +133,12 @@ private:
        mHandleAllocator.deallocate(handle, p);
    }

-    inline void setRenderPrimitiveBuffer(Handle<HwRenderPrimitive> rph, PrimitiveType pt,
+    inline void setRenderPrimitiveBuffer(Handle<HwRenderPrimitive> rph,
            Handle<HwVertexBuffer> vbh, Handle<HwIndexBuffer> ibh);

+    inline void setRenderPrimitiveRange(Handle<HwRenderPrimitive> rph, PrimitiveType pt,
+            uint32_t offset, uint32_t minIndex, uint32_t maxIndex, uint32_t count);
+
    void finalizeSamplerGroup(MetalSamplerGroup* sg);
    void enumerateBoundBuffers(BufferObjectBinding bindingType,
            const std::function<void(const BufferState&, MetalBuffer*, uint32_t)>& f);
@@ -20,7 +20,6 @@
 #include "metal/MetalDriver.h"

 #include "MetalBlitter.h"
-#include "MetalBufferPool.h"
 #include "MetalContext.h"
 #include "MetalDriverFactory.h"
 #include "MetalEnums.h"
@@ -37,7 +36,6 @@

 #include <utils/Log.h>
 #include <utils/Panic.h>
-#include <utils/sstream.h>

 #include <algorithm>

@@ -45,42 +43,6 @@ namespace filament {
 namespace backend {

 Driver* MetalDriverFactory::create(MetalPlatform* const platform, const Platform::DriverConfig& driverConfig) {
-#if 0
-    // this is useful for development, but too verbose even for debug builds
-    // For reference on a 64-bits machine in Release mode:
-    //    MetalTimerQuery              :  16       few
-    //    HwStream                     :  24       few
-    //    MetalRenderPrimitive         :  24       many
-    //    MetalVertexBuffer            :  32       moderate
-    // -- less than or equal 32 bytes
-    //    MetalIndexBuffer             :  40       moderate
-    //    MetalFence                   :  48       few
-    //    MetalBufferObject            :  48       many
-    // -- less than or equal 48 bytes
-    //    MetalSamplerGroup            : 112       few
-    //    MetalProgram                 : 152       moderate
-    //    MetalTexture                 : 152       moderate
-    //    MetalSwapChain               : 184       few
-    //    MetalRenderTarget            : 272       few
-    //    MetalVertexBufferInfo        : 552       moderate
-    // -- less than or equal to 552 bytes
-
-    utils::slog.d
-           << "\nMetalSwapChain: " << sizeof(MetalSwapChain)
-           << "\nMetalBufferObject: " << sizeof(MetalBufferObject)
-           << "\nMetalVertexBuffer: " << sizeof(MetalVertexBuffer)
-           << "\nMetalVertexBufferInfo: " << sizeof(MetalVertexBufferInfo)
-           << "\nMetalIndexBuffer: " << sizeof(MetalIndexBuffer)
-           << "\nMetalSamplerGroup: " << sizeof(MetalSamplerGroup)
-           << "\nMetalRenderPrimitive: " << sizeof(MetalRenderPrimitive)
-           << "\nMetalTexture: " << sizeof(MetalTexture)
-           << "\nMetalTimerQuery: " << sizeof(MetalTimerQuery)
-           << "\nHwStream: " << sizeof(HwStream)
-           << "\nMetalRenderTarget: " << sizeof(MetalRenderTarget)
-           << "\nMetalFence: " << sizeof(MetalFence)
-           << "\nMetalProgram: " << sizeof(MetalProgram)
-           << utils::io::endl;
-#endif
    return MetalDriver::create(platform, driverConfig);
 }

@@ -100,9 +62,7 @@ Dispatcher MetalDriver::getDispatcher() const noexcept {
 MetalDriver::MetalDriver(MetalPlatform* platform, const Platform::DriverConfig& driverConfig) noexcept
        : mPlatform(*platform),
          mContext(new MetalContext(driverConfig.textureUseAfterFreePoolSize)),
-          mHandleAllocator("Handles",
-                  driverConfig.handleArenaSize,
-                  driverConfig.disableHandleUseAfterFreeCheck) {
+          mHandleAllocator("Handles", driverConfig.handleArenaSize) {
    mContext->driver = this;

    mContext->device = mPlatform.createDevice();
@@ -184,10 +144,7 @@ MetalDriver::MetalDriver(MetalPlatform* platform, const Platform::DriverConfig&
        mContext->eventListener = [[MTLSharedEventListener alloc] initWithDispatchQueue:queue];
    }

-    const MetalShaderCompiler::Mode compilerMode = driverConfig.disableParallelShaderCompile
-            ? MetalShaderCompiler::Mode::SYNCHRONOUS
-            : MetalShaderCompiler::Mode::ASYNCHRONOUS;
-    mContext->shaderCompiler = new MetalShaderCompiler(mContext->device, *this, compilerMode);
+    mContext->shaderCompiler = new MetalShaderCompiler(mContext->device, *this);
    mContext->shaderCompiler->init();

 #if defined(FILAMENT_METAL_PROFILING)
@@ -216,9 +173,6 @@ void MetalDriver::beginFrame(int64_t monotonic_clock_ns, uint32_t frameId) {
 #if defined(FILAMENT_METAL_PROFILING)
    os_signpost_interval_begin(mContext->log, mContext->signpostId, "Frame encoding", "%{public}d", frameId);
 #endif
-    if (mPlatform.hasDebugUpdateStatFunc()) {
-        mPlatform.debugUpdateStat("filament.metal.alive_buffers", TrackedMetalBuffer::getAliveBuffers());
-    }
 }

 void MetalDriver::setFrameScheduledCallback(Handle<HwSwapChain> sch,
@@ -291,16 +245,10 @@ void MetalDriver::finish(int) {
    [oneOffBuffer waitUntilCompleted];
 }

-void MetalDriver::createVertexBufferInfoR(Handle<HwVertexBufferInfo> vbih, uint8_t bufferCount,
-        uint8_t attributeCount, AttributeArray attributes) {
-    construct_handle<MetalVertexBufferInfo>(vbih, *mContext,
-            bufferCount, attributeCount, attributes);
-}
-
-void MetalDriver::createVertexBufferR(Handle<HwVertexBuffer> vbh,
-        uint32_t vertexCount, Handle<HwVertexBufferInfo> vbih) {
-    MetalVertexBufferInfo const* const vbi = handle_cast<const MetalVertexBufferInfo>(vbih);
-    construct_handle<MetalVertexBuffer>(vbh, *mContext, vertexCount, vbi->bufferCount, vbih);
+void MetalDriver::createVertexBufferR(Handle<HwVertexBuffer> vbh, uint8_t bufferCount,
+        uint8_t attributeCount, uint32_t vertexCount, AttributeArray attributes) {
+    construct_handle<MetalVertexBuffer>(vbh, *mContext, bufferCount,
+            attributeCount, vertexCount, attributes);
 }

 void MetalDriver::createIndexBufferR(Handle<HwIndexBuffer> ibh, ElementType elementType,
@@ -368,9 +316,11 @@ void MetalDriver::createSamplerGroupR(

 void MetalDriver::createRenderPrimitiveR(Handle<HwRenderPrimitive> rph,
        Handle<HwVertexBuffer> vbh, Handle<HwIndexBuffer> ibh,
-        PrimitiveType pt) {
+        PrimitiveType pt, uint32_t offset,
+        uint32_t minIndex, uint32_t maxIndex, uint32_t count) {
    construct_handle<MetalRenderPrimitive>(rph);
-    MetalDriver::setRenderPrimitiveBuffer(rph, pt, vbh, ibh);
+    MetalDriver::setRenderPrimitiveBuffer(rph, vbh, ibh);
+    MetalDriver::setRenderPrimitiveRange(rph, pt, offset, minIndex, maxIndex, count);
 }

 void MetalDriver::createProgramR(Handle<HwProgram> rph, Program&& program) {
@@ -383,7 +333,7 @@ void MetalDriver::createDefaultRenderTargetR(Handle<HwRenderTarget> rth, int dum

 void MetalDriver::createRenderTargetR(Handle<HwRenderTarget> rth,
        TargetBufferFlags targetBufferFlags, uint32_t width, uint32_t height,
-        uint8_t samples, uint8_t layerCount, MRT color,
+        uint8_t samples, MRT color,
        TargetBufferInfo depth, TargetBufferInfo stencil) {
    ASSERT_PRECONDITION(!isInRenderPass(mContext),
            "createRenderTarget must be called outside of a render pass.");
@@ -456,10 +406,6 @@ void MetalDriver::createTimerQueryR(Handle<HwTimerQuery> tqh, int) {
    // nothing to do, timer query was constructed in createTimerQueryS
 }

-Handle<HwVertexBufferInfo> MetalDriver::createVertexBufferInfoS() noexcept {
-    return alloc_handle<MetalVertexBufferInfo>();
-}
-
 Handle<HwVertexBuffer> MetalDriver::createVertexBufferS() noexcept {
    return alloc_handle<MetalVertexBuffer>();
 }
@@ -524,12 +470,6 @@ Handle<HwTimerQuery> MetalDriver::createTimerQueryS() noexcept {
    return alloc_and_construct_handle<MetalTimerQuery, HwTimerQuery>();
 }

-void MetalDriver::destroyVertexBufferInfo(Handle<HwVertexBufferInfo> vbih) {
-    if (vbih) {
-        destruct_handle<MetalVertexBufferInfo>(vbih);
-    }
-}
-
 void MetalDriver::destroyVertexBuffer(Handle<HwVertexBuffer> vbh) {
    if (vbh) {
        destruct_handle<MetalVertexBuffer>(vbh);
@@ -782,33 +722,18 @@ bool MetalDriver::isSRGBSwapChainSupported() {
    return false;
 }

-bool MetalDriver::isProtectedContentSupported() {
-    // the SWAP_CHAIN_CONFIG_PROTECTED_CONTENT flag is not supported
-    return false;
-}
-
-bool MetalDriver::isStereoSupported(backend::StereoscopicType stereoscopicType) {
-    switch (stereoscopicType) {
-    case backend::StereoscopicType::INSTANCED:
-        return true;
-    case backend::StereoscopicType::MULTIVIEW:
-        // TODO: implement multiview feature in Metal.
-        return false;
-    }
+bool MetalDriver::isStereoSupported() {
+    return true;
 }

 bool MetalDriver::isParallelShaderCompileSupported() {
-    return mContext->shaderCompiler->isParallelShaderCompileSupported();
+    return true;
 }

 bool MetalDriver::isDepthStencilResolveSupported() {
    return false;
 }

-bool MetalDriver::isProtectedTexturesSupported() {
-    return false;
-}
-
 bool MetalDriver::isWorkaroundNeeded(Workaround workaround) {
    switch (workaround) {
        case Workaround::SPLIT_EASU:
@@ -935,10 +860,9 @@ void MetalDriver::setExternalImagePlane(Handle<HwTexture> th, void* image, uint3
 void MetalDriver::setExternalStream(Handle<HwTexture> th, Handle<HwStream> sh) {
 }

-TimerQueryResult MetalDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint64_t* elapsedTime) {
+bool MetalDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint64_t* elapsedTime) {
    auto* tq = handle_cast<MetalTimerQuery>(tqh);
-    return mContext->timerQueryImpl->getQueryResult(tq, elapsedTime) ?
-           TimerQueryResult::AVAILABLE : TimerQueryResult::NOT_READY;
+    return mContext->timerQueryImpl->getQueryResult(tq, elapsedTime);
 }

 void MetalDriver::generateMipmaps(Handle<HwTexture> th) {
@@ -1105,14 +1029,23 @@ void MetalDriver::endRenderPass(int dummy) {
    mContext->currentRenderPassEncoder = nil;
 }

-void MetalDriver::setRenderPrimitiveBuffer(Handle<HwRenderPrimitive> rph, PrimitiveType pt,
+void MetalDriver::setRenderPrimitiveBuffer(Handle<HwRenderPrimitive> rph,
        Handle<HwVertexBuffer> vbh, Handle<HwIndexBuffer> ibh) {
    auto primitive = handle_cast<MetalRenderPrimitive>(rph);
    auto vertexBuffer = handle_cast<MetalVertexBuffer>(vbh);
    auto indexBuffer = handle_cast<MetalIndexBuffer>(ibh);
-    MetalVertexBufferInfo const* const vbi = handle_cast<MetalVertexBufferInfo>(vertexBuffer->vbih);
-    primitive->setBuffers(vbi, vertexBuffer, indexBuffer);
+    primitive->setBuffers(vertexBuffer, indexBuffer);
+}
+
+void MetalDriver::setRenderPrimitiveRange(Handle<HwRenderPrimitive> rph,
+        PrimitiveType pt, uint32_t offset, uint32_t minIndex, uint32_t maxIndex,
+        uint32_t count) {
+    auto primitive = handle_cast<MetalRenderPrimitive>(rph);
    primitive->type = pt;
+    primitive->offset = offset * primitive->indexBuffer->elementSize;
+    primitive->count = count;
+    primitive->minIndex = minIndex;
+    primitive->maxIndex = maxIndex > minIndex ? maxIndex : primitive->maxVertexCount - 1;
 }

 void MetalDriver::makeCurrent(Handle<HwSwapChain> schDraw, Handle<HwSwapChain> schRead) {
@@ -1614,13 +1547,10 @@ void MetalDriver::finalizeSamplerGroup(MetalSamplerGroup* samplerGroup) {
    }
 }

-void MetalDriver::bindPipeline(PipelineState ps) {
+void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph, uint32_t instanceCount) {
    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder != nullptr,
-            "bindPipeline() without a valid command encoder.");
-
-    MetalVertexBufferInfo const* const vbi =
-            handle_cast<MetalVertexBufferInfo>(ps.vertexBufferInfo);
-
+            "Attempted to draw without a valid command encoder.");
+    auto primitive = handle_cast<MetalRenderPrimitive>(rph);
    auto program = handle_cast<MetalProgram>(ps.program);
    const auto& rs = ps.rasterState;

@@ -1634,7 +1564,7 @@ void MetalDriver::bindPipeline(PipelineState ps) {
        return;
    }

-    functions.validate();
+    ASSERT_PRECONDITION(bool(functions), "Attempting to draw with an invalid Metal program.");

    auto [fragment, vertex] = functions.getRasterFunctions();

@@ -1661,7 +1591,7 @@ void MetalDriver::bindPipeline(PipelineState ps) {
    MetalPipelineState const pipelineState {
        .vertexFunction = vertex,
        .fragmentFunction = fragment,
-        .vertexDescription = vbi->vertexDescription,
+        .vertexDescription = primitive->vertexDescription,
        .colorAttachmentPixelFormat = {
            colorPixelFormat[0],
            colorPixelFormat[1],
@@ -1755,6 +1685,50 @@ void MetalDriver::bindPipeline(PipelineState ps) {
        mContext->currentPolygonOffset = ps.polygonOffset;
    }

+    // Set scissor-rectangle.
+    // In order to do this, we compute the intersection between:
+    //  1. the scissor rectangle
+    //  2. the render target attachment dimensions (important, as the scissor can't be set larger)
+    // fmax/min are used below to guard against NaN and because the MTLViewport/MTLRegion
+    // coordinates are doubles.
+    MTLRegion scissor = mContext->currentRenderTarget->getRegionFromClientRect(ps.scissor);
+    const float sleft = scissor.origin.x, sright = scissor.origin.x + scissor.size.width;
+    const float stop = scissor.origin.y, sbottom = scissor.origin.y + scissor.size.height;
+
+    // Attachment extent
+    const auto attachmentSize = mContext->currentRenderTarget->getAttachmentSize();
+    const float aleft = 0.0f, atop = 0.0f;
+    const float aright = static_cast<float>(attachmentSize.x);
+    const float abottom = static_cast<float>(attachmentSize.y);
+
+    const auto left   = std::fmax(sleft, aleft);
+    const auto right  = std::fmin(sright, aright);
+    const auto top    = std::fmax(stop, atop);
+    const auto bottom = std::fmin(sbottom, abottom);
+
+    MTLScissorRect scissorRect = {
+        .x      = static_cast<NSUInteger>(left),
+        .y      = static_cast<NSUInteger>(top),
+        .width  = static_cast<NSUInteger>(right  - left),
+        .height = static_cast<NSUInteger>(bottom - top)
+    };
+
+    [mContext->currentRenderPassEncoder setScissorRect:scissorRect];
+
+    // Bind uniform buffers.
+    MetalBuffer* uniformsToBind[Program::UNIFORM_BINDING_COUNT] = { nil };
+    NSUInteger offsets[Program::UNIFORM_BINDING_COUNT] = { 0 };
+
+    enumerateBoundBuffers(BufferObjectBinding::UNIFORM,
+            [&uniformsToBind, &offsets](const BufferState& state, MetalBuffer* buffer,
+                    uint32_t index) {
+        uniformsToBind[index] = buffer;
+        offsets[index] = state.offset;
+    });
+    MetalBuffer::bindBuffers(getPendingCommandBuffer(mContext), mContext->currentRenderPassEncoder,
+            UNIFORM_BUFFER_BINDING_START, MetalBuffer::Stage::VERTEX | MetalBuffer::Stage::FRAGMENT,
+            uniformsToBind, offsets, Program::UNIFORM_BINDING_COUNT);
+
    // Bind sampler groups (argument buffers).
    for (size_t s = 0; s < Program::SAMPLER_BINDING_COUNT; s++) {
        MetalSamplerGroup* const samplerGroup = mContext->samplerBindings[s];
@@ -1785,29 +1759,19 @@ void MetalDriver::bindPipeline(PipelineState ps) {
                                                          atIndex:(SAMPLER_GROUP_BINDING_START + s)];
        }
    }
-}
-
-void MetalDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
-    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder != nullptr,
-            "bindRenderPrimitive() without a valid command encoder.");

    // Bind the user vertex buffers.
+
    MetalBuffer* vertexBuffers[MAX_VERTEX_BUFFER_COUNT] = {};
    size_t vertexBufferOffsets[MAX_VERTEX_BUFFER_COUNT] = {};
    size_t maxBufferIndex = 0;

-    MetalRenderPrimitive const* const primitive = handle_cast<MetalRenderPrimitive>(rph);
-    MetalVertexBufferInfo const* const vbi =
-            handle_cast<MetalVertexBufferInfo>(primitive->vertexBuffer->vbih);
-
-    mContext->currentRenderPrimitive = rph;
-
    auto vb = primitive->vertexBuffer;
-    for (auto m : vbi->bufferMapping) {
+    for (auto m : primitive->bufferMapping) {
        assert_invariant(
                m.bufferArgumentIndex >= USER_VERTEX_BUFFER_BINDING_START &&
                m.bufferArgumentIndex < USER_VERTEX_BUFFER_BINDING_START + MAX_VERTEX_BUFFER_COUNT);
-        size_t const vertexBufferIndex = m.bufferArgumentIndex - USER_VERTEX_BUFFER_BINDING_START;
+        size_t vertexBufferIndex = m.bufferArgumentIndex - USER_VERTEX_BUFFER_BINDING_START;
        vertexBuffers[vertexBufferIndex] = vb->buffers[m.sourceBufferIndex];
        maxBufferIndex = std::max(maxBufferIndex, vertexBufferIndex);
    }
@@ -1822,50 +1786,19 @@ void MetalDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
    [mContext->currentRenderPassEncoder setVertexBytes:bytes
                                                length:16
                                               atIndex:ZERO_VERTEX_BUFFER_BINDING];
-}
-
-void MetalDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
-    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder != nullptr,
-            "draw() without a valid command encoder.");
-
-    // Bind uniform buffers.
-    MetalBuffer* uniformsToBind[Program::UNIFORM_BINDING_COUNT] = { nil };
-    NSUInteger offsets[Program::UNIFORM_BINDING_COUNT] = { 0 };
-
-    enumerateBoundBuffers(BufferObjectBinding::UNIFORM,
-            [&uniformsToBind, &offsets](const BufferState& state, MetalBuffer* buffer,
-                    uint32_t index) {
-                uniformsToBind[index] = buffer;
-                offsets[index] = state.offset;
-            });
-    MetalBuffer::bindBuffers(getPendingCommandBuffer(mContext), mContext->currentRenderPassEncoder,
-            UNIFORM_BUFFER_BINDING_START, MetalBuffer::Stage::VERTEX | MetalBuffer::Stage::FRAGMENT,
-            uniformsToBind, offsets, Program::UNIFORM_BINDING_COUNT);
-
-    auto primitive = handle_cast<MetalRenderPrimitive>(mContext->currentRenderPrimitive);

    MetalIndexBuffer* indexBuffer = primitive->indexBuffer;

    id<MTLCommandBuffer> cmdBuffer = getPendingCommandBuffer(mContext);
    id<MTLBuffer> metalIndexBuffer = indexBuffer->buffer.getGpuBufferForDraw(cmdBuffer);
    [mContext->currentRenderPassEncoder drawIndexedPrimitives:getMetalPrimitiveType(primitive->type)
-                                                   indexCount:indexCount
+                                                   indexCount:primitive->count
                                                    indexType:getIndexType(indexBuffer->elementSize)
                                                  indexBuffer:metalIndexBuffer
-                                            indexBufferOffset:indexOffset * primitive->indexBuffer->elementSize
+                                            indexBufferOffset:primitive->offset
                                                instanceCount:instanceCount];
 }

-void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph,
-        uint32_t const indexOffset, uint32_t const indexCount, uint32_t const instanceCount) {
-    MetalRenderPrimitive const* const rp = handle_cast<MetalRenderPrimitive>(rph);
-    ps.primitiveType = rp->type;
-    ps.vertexBufferInfo = rp->vertexBuffer->vbih;
-    bindPipeline(ps);
-    bindRenderPrimitive(rph);
-    draw2(indexOffset, indexCount, instanceCount);
-}
-
 void MetalDriver::dispatchCompute(Handle<HwProgram> program, math::uint3 workGroupCount) {
    ASSERT_PRECONDITION(!isInRenderPass(mContext),
            "dispatchCompute must be called outside of a render pass.");
@@ -1935,38 +1868,6 @@ void MetalDriver::dispatchCompute(Handle<HwProgram> program, math::uint3 workGro
    [computeEncoder endEncoding];
 }

-void MetalDriver::scissor(Viewport scissorBox) {
-    // Set scissor-rectangle.
-    // In order to do this, we compute the intersection between:
-    //  1. the scissor rectangle
-    //  2. the render target attachment dimensions (important, as the scissor can't be set larger)
-    // fmax/min are used below to guard against NaN and because the MTLViewport/MTLRegion
-    // coordinates are doubles.
-    MTLRegion scissor = mContext->currentRenderTarget->getRegionFromClientRect(scissorBox);
-    const float sleft = scissor.origin.x, sright = scissor.origin.x + scissor.size.width;
-    const float stop = scissor.origin.y, sbottom = scissor.origin.y + scissor.size.height;
-
-    // Attachment extent
-    const auto attachmentSize = mContext->currentRenderTarget->getAttachmentSize();
-    const float aleft = 0.0f, atop = 0.0f;
-    const float aright = static_cast<float>(attachmentSize.x);
-    const float abottom = static_cast<float>(attachmentSize.y);
-
-    const auto left   = std::fmax(sleft, aleft);
-    const auto right  = std::fmin(sright, aright);
-    const auto top    = std::fmax(stop, atop);
-    const auto bottom = std::fmin(sbottom, abottom);
-
-    MTLScissorRect scissorRect = {
-            .x      = static_cast<NSUInteger>(left),
-            .y      = static_cast<NSUInteger>(top),
-            .width  = static_cast<NSUInteger>(right  - left),
-            .height = static_cast<NSUInteger>(bottom - top)
-    };
-
-    [mContext->currentRenderPassEncoder setScissorRect:scissorRect];
-}
-
 void MetalDriver::beginTimerQuery(Handle<HwTimerQuery> tqh) {
    ASSERT_PRECONDITION(!isInRenderPass(mContext),
            "beginTimerQuery must be called outside of a render pass.");
@@ -21,7 +21,6 @@
 #include "MetalUtils.h"

 #include <utils/Panic.h>
-#include <utils/Log.h>
 #include <utils/trap.h>

 #define NSERROR_CHECK(message)                                                                     \
@@ -145,9 +145,28 @@ private:
    MetalBuffer buffer;
 };

-struct MetalVertexBufferInfo : public HwVertexBufferInfo {
-    MetalVertexBufferInfo(MetalContext& context,
-            uint8_t bufferCount, uint8_t attributeCount, AttributeArray const& attributes);
+struct MetalVertexBuffer : public HwVertexBuffer {
+    MetalVertexBuffer(MetalContext& context, uint8_t bufferCount, uint8_t attributeCount,
+            uint32_t vertexCount, AttributeArray const& attributes);
+
+    utils::FixedCapacityVector<MetalBuffer*> buffers;
+};
+
+struct MetalIndexBuffer : public HwIndexBuffer {
+    MetalIndexBuffer(MetalContext& context, BufferUsage usage, uint8_t elementSize,
+            uint32_t indexCount);
+
+    MetalBuffer buffer;
+};
+
+struct MetalRenderPrimitive : public HwRenderPrimitive {
+    MetalRenderPrimitive();
+    void setBuffers(MetalVertexBuffer* vertexBuffer, MetalIndexBuffer* indexBuffer);
+    // The pointers to MetalVertexBuffer and MetalIndexBuffer are "weak".
+    // The MetalVertexBuffer and MetalIndexBuffer must outlive the MetalRenderPrimitive.
+
+    MetalVertexBuffer* vertexBuffer = nullptr;
+    MetalIndexBuffer* indexBuffer = nullptr;

    // This struct is used to create the pipeline description to describe vertex assembly.
    VertexDescription vertexDescription = {};
@@ -166,32 +185,6 @@ struct MetalVertexBufferInfo : public HwVertexBufferInfo {
    utils::FixedCapacityVector<Entry> bufferMapping;
 };

-struct MetalVertexBuffer : public HwVertexBuffer {
-    MetalVertexBuffer(MetalContext& context,
-            uint32_t vertexCount, uint32_t bufferCount, Handle<HwVertexBufferInfo> vbih);
-
-    Handle<HwVertexBufferInfo> vbih;
-    utils::FixedCapacityVector<MetalBuffer*> buffers;
-};
-
-struct MetalIndexBuffer : public HwIndexBuffer {
-    MetalIndexBuffer(MetalContext& context, BufferUsage usage, uint8_t elementSize,
-            uint32_t indexCount);
-
-    MetalBuffer buffer;
-};
-
-struct MetalRenderPrimitive : public HwRenderPrimitive {
-    MetalRenderPrimitive();
-    void setBuffers(MetalVertexBufferInfo const* const vbi,
-            MetalVertexBuffer* vertexBuffer, MetalIndexBuffer* indexBuffer);
-    // The pointers to MetalVertexBuffer and MetalIndexBuffer are "weak".
-    // The MetalVertexBuffer and MetalIndexBuffer must outlive the MetalRenderPrimitive.
-
-    MetalVertexBuffer* vertexBuffer = nullptr;
-    MetalIndexBuffer* indexBuffer = nullptr;
-};
-
 class MetalProgram : public HwProgram {
 public:
    MetalProgram(MetalContext& context, Program&& program) noexcept;
@@ -19,7 +19,6 @@
 #include "MetalBlitter.h"
 #include "MetalEnums.h"
 #include "MetalUtils.h"
-#include "MetalBufferPool.h"

 #include <filament/SwapChain.h>

@@ -246,53 +245,30 @@ void MetalSwapChain::present() {
    }
 }

-#ifndef FILAMENT_RELEASE_PRESENT_DRAWABLE_MAIN_THREAD
-#define FILAMENT_RELEASE_PRESENT_DRAWABLE_MAIN_THREAD 1
-#endif
-
-class PresentDrawableData {
-public:
-    PresentDrawableData() = delete;
-    PresentDrawableData(const PresentDrawableData&) = delete;
-    PresentDrawableData& operator=(const PresentDrawableData&) = delete;
-
-    static PresentDrawableData* create(id<CAMetalDrawable> drawable, MetalDriver* driver) {
-        assert_invariant(driver);
-        return new PresentDrawableData(drawable, driver);
-    }
-
-    static void maybePresentAndDestroyAsync(PresentDrawableData* that, bool shouldPresent) {
-        if (shouldPresent) {
-           [that->mDrawable present];
-        }
-
-#if FILAMENT_RELEASE_PRESENT_DRAWABLE_MAIN_THREAD == 1
-        // mDrawable is acquired on the driver thread. Typically, we would release this object on
-        // the same thread, but after receiving consistent crash reports from within
-        // [CAMetalDrawable dealloc], we suspect this object requires releasing on the main thread.
-        dispatch_async(dispatch_get_main_queue(), ^{ cleanupAndDestroy(that); });
-#else
-        that->mDriver->runAtNextTick([that]() { cleanupAndDestroy(that); });
-#endif
-    }
-
-private:
-    PresentDrawableData(id<CAMetalDrawable> drawable, MetalDriver* driver)
-        : mDrawable(drawable), mDriver(driver) {}
-
-    static void cleanupAndDestroy(PresentDrawableData *that) {
-        that->mDrawable = nil;
-        that->mDriver = nullptr;
-        delete that;
-    }
-
-    id<CAMetalDrawable> mDrawable;
-    MetalDriver* mDriver = nullptr;
+struct PresentDrawableData {
+    void* drawable = nullptr;
+    MetalDriver* driver = nullptr;
 };

 void presentDrawable(bool presentFrame, void* user) {
    auto* presentDrawableData = static_cast<PresentDrawableData*>(user);
-    PresentDrawableData::maybePresentAndDestroyAsync(presentDrawableData, presentFrame);
+
+    // CFBridgingRelease here is used to balance the CFBridgingRetain inside acquireDrawable.
+    id<CAMetalDrawable> drawable =
+            (id<CAMetalDrawable>)CFBridgingRelease(presentDrawableData->drawable);
+    if (presentFrame) {
+        [drawable present];
+    }
+
+    // Schedule the drawable destruction on the driver thread.
+    void* voidDrawable = (void*) CFBridgingRetain(drawable);
+    MetalDriver* driver = presentDrawableData->driver;
+    driver->runAtNextTick([voidDrawable]() {
+        // The drawable is released here.
+        CFBridgingRelease(voidDrawable);
+    });
+
+    delete presentDrawableData;
 }

 void MetalSwapChain::scheduleFrameScheduledCallback() {
@@ -301,16 +277,21 @@ void MetalSwapChain::scheduleFrameScheduledCallback() {
    }

    assert_invariant(drawable);
-
-    // Destroy this by calling maybePresentAndDestroyAsync() later.
-    auto* presentData = PresentDrawableData::create(drawable, context.driver);
-
-    FrameScheduledCallback userCallback = frameScheduledCallback;
+    FrameScheduledCallback callback = frameScheduledCallback;
+    // This block strongly captures drawable to keep it alive until the handler executes.
+    // We cannot simply reference this->drawable inside the block because the block would then only
+    // capture the _this_ pointer (MetalSwapChain*) instead of the drawable.
+    id<CAMetalDrawable> d = drawable;
    void* userData = frameScheduledUserData;
-
+    MetalDriver* driver = context.driver;
    [getPendingCommandBuffer(&context) addScheduledHandler:^(id<MTLCommandBuffer> cb) {
-        PresentCallable callable(presentDrawable, static_cast<void*>(presentData));
-        userCallback(callable, userData);
+        // CFBridgingRetain is used here to give the drawable a +1 retain count before
+        // casting it to a void*.
+        auto* presentDrawableData = new PresentDrawableData;
+        presentDrawableData->drawable = (void*) CFBridgingRetain(d);
+        presentDrawableData->driver = driver;
+        PresentCallable callable(presentDrawable, (void*) presentDrawableData);
+        callback(callable, userData);
    }];
 }

@@ -341,12 +322,23 @@ void MetalBufferObject::updateBufferUnsynchronized(void* data, size_t size, uint
    buffer.copyIntoBufferUnsynchronized(data, size, byteOffset);
 }

-MetalVertexBufferInfo::MetalVertexBufferInfo(MetalContext& context, uint8_t bufferCount,
-        uint8_t attributeCount, AttributeArray const& attributes)
-        : HwVertexBufferInfo(bufferCount, attributeCount),
-          bufferMapping(utils::FixedCapacityVector<Entry>::with_capacity(MAX_VERTEX_BUFFER_COUNT)) {
+MetalVertexBuffer::MetalVertexBuffer(MetalContext& context, uint8_t bufferCount,
+            uint8_t attributeCount, uint32_t vertexCount, AttributeArray const& attributes)
+    : HwVertexBuffer(bufferCount, attributeCount, vertexCount, attributes), buffers(bufferCount, nullptr) {}

-    const size_t maxAttributeCount = attributes.size();
+MetalIndexBuffer::MetalIndexBuffer(MetalContext& context, BufferUsage usage, uint8_t elementSize,
+        uint32_t indexCount) : HwIndexBuffer(elementSize, indexCount),
+        buffer(context, BufferObjectBinding::VERTEX, usage, elementSize * indexCount, true) { }
+
+MetalRenderPrimitive::MetalRenderPrimitive()
+    : bufferMapping(utils::FixedCapacityVector<Entry>::with_capacity(MAX_VERTEX_BUFFER_COUNT)) {}
+
+void MetalRenderPrimitive::setBuffers(MetalVertexBuffer* vertexBuffer, MetalIndexBuffer*
+        indexBuffer) {
+    this->vertexBuffer = vertexBuffer;
+    this->indexBuffer = indexBuffer;
+
+    const size_t attributeCount = vertexBuffer->attributes.size();

    auto& mapping = bufferMapping;
    mapping.clear();
@@ -386,8 +378,8 @@ MetalVertexBufferInfo::MetalVertexBufferInfo(MetalContext& context, uint8_t buff
        }
    };

-    for (uint32_t attributeIndex = 0; attributeIndex < maxAttributeCount; attributeIndex++) {
-        const auto& attribute = attributes[attributeIndex];
+    for (uint32_t attributeIndex = 0; attributeIndex < attributeCount; attributeIndex++) {
+        const auto& attribute = vertexBuffer->attributes[attributeIndex];

        // If the attribute is unused, bind it to the zero buffer. It's a Metal error for a shader
        // to read from missing vertex attributes.
@@ -414,24 +406,6 @@ MetalVertexBufferInfo::MetalVertexBufferInfo(MetalContext& context, uint8_t buff
    }
 }

-MetalVertexBuffer::MetalVertexBuffer(MetalContext& context,
-        uint32_t vertexCount, uint32_t bufferCount, Handle<HwVertexBufferInfo> vbih)
-    : HwVertexBuffer(vertexCount), vbih(vbih), buffers(bufferCount, nullptr) {
-}
-
-MetalIndexBuffer::MetalIndexBuffer(MetalContext& context, BufferUsage usage, uint8_t elementSize,
-        uint32_t indexCount) : HwIndexBuffer(elementSize, indexCount),
-        buffer(context, BufferObjectBinding::VERTEX, usage, elementSize * indexCount, true) { }
-
-MetalRenderPrimitive::MetalRenderPrimitive() {
-}
-
-void MetalRenderPrimitive::setBuffers(MetalVertexBufferInfo const* const vbi,
-        MetalVertexBuffer* vertexBuffer, MetalIndexBuffer* indexBuffer) {
-    this->vertexBuffer = vertexBuffer;
-    this->indexBuffer = indexBuffer;
-}
-
 MetalProgram::MetalProgram(MetalContext& context, Program&& program) noexcept
    : HwProgram(program.getName()), mContext(context) {

@@ -771,13 +745,13 @@ void MetalTexture::loadWithCopyBuffer(uint32_t level, uint32_t slice, MTLRegion
        PixelBufferDescriptor const& data, const PixelBufferShape& shape) {
    const size_t stagingBufferSize = shape.totalBytes;
    auto entry = context.bufferPool->acquireBuffer(stagingBufferSize);
-    memcpy(entry->buffer.get().contents,
+    memcpy(entry->buffer.contents,
            static_cast<uint8_t*>(data.buffer) + shape.sourceOffset,
            stagingBufferSize);
    id<MTLCommandBuffer> blitCommandBuffer = getPendingCommandBuffer(&context);
    id<MTLBlitCommandEncoder> blitCommandEncoder = [blitCommandBuffer blitCommandEncoder];
    blitCommandEncoder.label = @"Texture upload buffer blit";
-    [blitCommandEncoder copyFromBuffer:entry->buffer.get()
+    [blitCommandEncoder copyFromBuffer:entry->buffer
                          sourceOffset:0
                     sourceBytesPerRow:shape.bytesPerRow
                   sourceBytesPerImage:shape.bytesPerSlice
@@ -30,8 +30,6 @@

 #include <array>
 #include <memory>
-#include <tuple>
-#include <variant>

 namespace filament::backend {

@@ -41,81 +39,43 @@ class MetalShaderCompiler {
    struct MetalProgramToken;

 public:
-    enum class Mode {
-        SYNCHRONOUS,    // synchronous shader compilation
-        ASYNCHRONOUS    // asynchronous shader compilation
-    };
-
    class MetalFunctionBundle {
    public:
-        using Raster = std::tuple<id<MTLFunction>, id<MTLFunction>>;
-        using Compute = id<MTLFunction>;
-        using Error = std::tuple<NSString*, NSString*>; // error message, Program name
-        struct None {};
-
-        MetalFunctionBundle() : mPrograms{None{}} {}
-
-        explicit operator bool() const { return isValid(); }
-
-        bool isValid() const noexcept {
-            return std::holds_alternative<Raster>(mPrograms) ||
-                std::holds_alternative<Compute>(mPrograms);
-        }
-
-        /**
-         * Throws an NSException if this MetalFunctionBundle either contains an error or is empty.
-         *
-         * If this MetalFunctionBundle contains an error, will throw a MetalCompilationFailure
-         * NSException with the error string and program name passed to
-         * MetalFunctionBundle::error(NSString*, NSString*).
-         *
-         * If this MetalFunctionBundle is empty, will throw a MetalEmptyFunctionBundle NSException.
-         */
-        void validate() const;
-
-        Raster getRasterFunctions() const {
-            assert_invariant(std::holds_alternative<Raster>(mPrograms));
-            return std::get<Raster>(mPrograms);
-        }
-
-        Compute getComputeFunction() const {
-            assert_invariant(std::holds_alternative<Compute>(mPrograms));
-            return std::get<Compute>(mPrograms);
-        }
-
-        static MetalFunctionBundle none() {
-            return MetalFunctionBundle(None{});
-        }
-
-        static MetalFunctionBundle raster(id<MTLFunction> fragment, id<MTLFunction> vertex) {
+        MetalFunctionBundle() = default;
+        MetalFunctionBundle(id<MTLFunction> fragment, id<MTLFunction> vertex)
+            : functions{fragment, vertex} {
            assert_invariant(fragment && vertex);
            assert_invariant(fragment.functionType == MTLFunctionTypeFragment);
            assert_invariant(vertex.functionType == MTLFunctionTypeVertex);
-            return MetalFunctionBundle(Raster{fragment, vertex});
        }
-
-        static MetalFunctionBundle compute(id<MTLFunction> compute) {
+        explicit MetalFunctionBundle(id<MTLFunction> compute) : functions{compute, nil} {
            assert_invariant(compute);
            assert_invariant(compute.functionType == MTLFunctionTypeKernel);
-            return MetalFunctionBundle(Compute{compute});
        }

-        static MetalFunctionBundle error(NSString* errorMessage, NSString* programName) {
-            return MetalFunctionBundle(Error{errorMessage, programName});
+        std::pair<id<MTLFunction>, id<MTLFunction>> getRasterFunctions() const noexcept {
+            assert_invariant(functions[0].functionType == MTLFunctionTypeFragment);
+            assert_invariant(functions[1].functionType == MTLFunctionTypeVertex);
+            return {functions[0], functions[1]};
        }

+        id<MTLFunction> getComputeFunction() const noexcept {
+            assert_invariant(functions[0].functionType == MTLFunctionTypeKernel);
+            return functions[0];
+        }
+
+        explicit operator bool() const { return functions[0] != nil; }
+
    private:
-        MetalFunctionBundle(None&& t) : mPrograms(std::move(t)) {}
-        MetalFunctionBundle(Raster&& t) : mPrograms(std::move(t)) {}
-        MetalFunctionBundle(Compute&& t) : mPrograms(std::move(t)) {}
-        MetalFunctionBundle(Error&& t) : mPrograms(std::move(t)) {}
-
-        std::variant<Raster, Compute, None, Error> mPrograms;
+        // Can hold two functions, either:
+        // - fragment and vertex (for rasterization pipelines)
+        // - compute (for compute pipelines)
+        id<MTLFunction> functions[2] = {nil, nil};
    };

    using program_token_t = std::shared_ptr<MetalProgramToken>;

-    explicit MetalShaderCompiler(id<MTLDevice> device, MetalDriver& driver, Mode mode);
+    explicit MetalShaderCompiler(id<MTLDevice> device, MetalDriver& driver);

    MetalShaderCompiler(MetalShaderCompiler const& rhs) = delete;
    MetalShaderCompiler(MetalShaderCompiler&& rhs) = delete;
@@ -125,15 +85,15 @@ public:
    void init() noexcept;
    void terminate() noexcept;

-    bool isParallelShaderCompileSupported() const noexcept;
-
-    // Creates a program, either synchronously or asynchronously, depending on the Mode
-    // MetalShaderCompiler was constructed with.
+    // Creates a program asynchronously
    program_token_t createProgram(utils::CString const& name, Program&& program);

    // Returns the functions, blocking if necessary. The Token is destroyed and becomes invalid.
    MetalFunctionBundle getProgram(program_token_t& token);

+    // Destroys a valid token and all associated resources. Used to "cancel" a program compilation.
+    static void terminate(program_token_t& token);
+
    void notifyWhenAllProgramsAreReady(
            CallbackHandler* handler, CallbackHandler::Callback callback, void* user);

@@ -143,7 +103,6 @@ private:
    CompilerThreadPool mCompilerThreadPool;
    id<MTLDevice> mDevice;
    CallbackManager mCallbackManager;
-    Mode mMode;
 };

 } // namespace filament::backend
@@ -70,31 +70,22 @@ struct MetalShaderCompiler::MetalProgramToken : ProgramToken {

 MetalShaderCompiler::MetalProgramToken::~MetalProgramToken() = default;

-MetalShaderCompiler::MetalShaderCompiler(id<MTLDevice> device, MetalDriver& driver, Mode mode)
+MetalShaderCompiler::MetalShaderCompiler(id<MTLDevice> device, MetalDriver& driver)
        : mDevice(device),
-          mCallbackManager(driver),
-          mMode(mode) {
+          mCallbackManager(driver) {

 }

 void MetalShaderCompiler::init() noexcept {
-    const uint32_t poolSize = 1;
-    if (mMode == Mode::ASYNCHRONOUS) {
-        mCompilerThreadPool.init(poolSize, []() {}, []() {});
-    }
+    const uint32_t poolSize = 2;
+    mCompilerThreadPool.init(poolSize, []() {}, []() {});
 }

 void MetalShaderCompiler::terminate() noexcept {
-    if (mMode == Mode::ASYNCHRONOUS) {
-        mCompilerThreadPool.terminate();
-    }
+    mCompilerThreadPool.terminate();
    mCallbackManager.terminate();
 }

-bool MetalShaderCompiler::isParallelShaderCompileSupported() const noexcept {
-    return mMode == Mode::ASYNCHRONOUS;
-}
-
 /* static */ MetalShaderCompiler::MetalFunctionBundle MetalShaderCompiler::compileProgram(
        const Program& program, id<MTLDevice> device) {
    std::array<id<MTLFunction>, Program::SHADER_TYPE_COUNT> functions = { nil };
@@ -127,16 +118,13 @@ bool MetalShaderCompiler::isParallelShaderCompileSupported() const noexcept {
                                                      options:options
                                                        error:&error];
        if (library == nil) {
-            NSString* errorMessage = @"unknown error";
            if (error) {
                auto description =
                        [error.localizedDescription cStringUsingEncoding:NSUTF8StringEncoding];
                utils::slog.w << description << utils::io::endl;
-                errorMessage = error.localizedDescription;
            }
            PANIC_LOG("Failed to compile Metal program.");
-            NSString* programName = [NSString stringWithFormat:@"%s", program.getName().c_str_safe()];
-            return MetalFunctionBundle::error(errorMessage, programName);
+            return {};
        }

        MTLFunctionConstantValues* constants = [MTLFunctionConstantValues new];
@@ -172,15 +160,14 @@ bool MetalShaderCompiler::isParallelShaderCompileSupported() const noexcept {
    assert_invariant(isRasterizationProgram != isComputeProgram);

    if (isRasterizationProgram) {
-        return MetalFunctionBundle::raster(fragmentFunction, vertexFunction);
+        return {fragmentFunction, vertexFunction};
    }

    if (isComputeProgram) {
-        return MetalFunctionBundle::compute(computeFunction);
+        return MetalFunctionBundle{computeFunction};
    }

-    // Should never reach here.
-    return MetalFunctionBundle::none();
+    return {};
 }

 MetalShaderCompiler::program_token_t MetalShaderCompiler::createProgram(
@@ -189,26 +176,14 @@ MetalShaderCompiler::program_token_t MetalShaderCompiler::createProgram(

    token->handle = mCallbackManager.get();

-    switch (mMode) {
-        case Mode::ASYNCHRONOUS: {
-            CompilerPriorityQueue const priorityQueue = program.getPriorityQueue();
-            mCompilerThreadPool.queue(priorityQueue, token,
-                    [this, name, device = mDevice, program = std::move(program), token]() {
-                        MetalFunctionBundle compiledProgram = compileProgram(program, device);
-                        token->set(compiledProgram);
-                        mCallbackManager.put(token->handle);
-                    });
+    CompilerPriorityQueue const priorityQueue = program.getPriorityQueue();
+    mCompilerThreadPool.queue(priorityQueue, token,
+            [this, name, device = mDevice, program = std::move(program), token]() {
+                MetalFunctionBundle compiledProgram = compileProgram(program, device);

-            break;
-        }
-
-        case Mode::SYNCHRONOUS: {
-            MetalFunctionBundle compiledProgram = compileProgram(program, mDevice);
-            token->set(compiledProgram);
-            mCallbackManager.put(token->handle);
-            break;
-        }
-    }
+                token->set(compiledProgram);
+                mCallbackManager.put(token->handle);
+            });

    return token;
 }
@@ -216,46 +191,38 @@ MetalShaderCompiler::program_token_t MetalShaderCompiler::createProgram(
 MetalShaderCompiler::MetalFunctionBundle MetalShaderCompiler::getProgram(program_token_t& token) {
    assert_invariant(token);

-    if (mMode == Mode::ASYNCHRONOUS) {
-        if (!token->isReady()) {
-            auto job = mCompilerThreadPool.dequeue(token);
-            if (job) {
-                job();
-            }
+    if (!token->isReady()) {
+        auto job = mCompilerThreadPool.dequeue(token);
+        if (job) {
+            job();
        }
    }

-    // The job isn't guaranteed to have finished yet. We may have failed to dequeue it above,
-    // which means it's currently running. In that case get() will block until it finishes.
-
    MetalShaderCompiler::MetalFunctionBundle program = token->get();
+
    token = nullptr;
+
    return program;
 }

+/* static */ void MetalShaderCompiler::terminate(program_token_t& token) {
+    assert_invariant(token);
+
+    auto job = token->compiler.mCompilerThreadPool.dequeue(token);
+    if (!job) {
+        // The job is being executed right now (or has already executed).
+        token->wait();
+    } else {
+        // The job has not executed yet.
+        token->compiler.mCallbackManager.put(token->handle);
+    }
+
+    token.reset();
+}
+
 void MetalShaderCompiler::notifyWhenAllProgramsAreReady(
        CallbackHandler* handler, CallbackHandler::Callback callback, void* user) {
    mCallbackManager.setCallback(handler, callback, user);
 }

-UTILS_NOINLINE
-void MetalShaderCompiler::MetalFunctionBundle::validate() const {
-    if (UTILS_UNLIKELY(std::holds_alternative<Error>(mPrograms))) {
-        auto [errorMessage, programName] = std::get<Error>(mPrograms);
-        NSString* reason =
-                [NSString stringWithFormat:
-                        @"Attempting to draw with an id<MTLFunction> that failed to compile.\n"
-                        @"Program: %@\n"
-                        @"%@", programName, errorMessage];
-        [[NSException exceptionWithName:@"MetalCompilationFailure"
-                                reason:reason
-                              userInfo:nil] raise];
-    } else if (UTILS_UNLIKELY(std::holds_alternative<None>(mPrograms))) {
-        NSString* reason = @"Attempting to draw with an empty id<MTLFunction>.";
-        [[NSException exceptionWithName:@"MetalEmptyFunctionBundle"
-                                reason:reason
-                              userInfo:nil] raise];
-    }
-}
-
 } // namespace filament::backend
@@ -18,8 +18,6 @@

 #include "MetalEnums.h"

-#include <utils/Log.h>
-
 namespace filament {
 namespace backend {

@@ -77,9 +77,6 @@ void NoopDriver::finish(int) {
 void NoopDriver::destroyRenderPrimitive(Handle<HwRenderPrimitive> rph) {
 }

-void NoopDriver::destroyVertexBufferInfo(Handle<HwVertexBufferInfo> vbih) {
-}
-
 void NoopDriver::destroyVertexBuffer(Handle<HwVertexBuffer> vbh) {
 }

@@ -177,11 +174,7 @@ bool NoopDriver::isSRGBSwapChainSupported() {
    return false;
 }

-bool NoopDriver::isProtectedContentSupported() {
-    return false;
-}
-
-bool NoopDriver::isStereoSupported(backend::StereoscopicType) {
+bool NoopDriver::isStereoSupported() {
    return false;
 }

@@ -193,10 +186,6 @@ bool NoopDriver::isDepthStencilResolveSupported() {
    return true;
 }

-bool NoopDriver::isProtectedTexturesSupported() {
-    return true;
-}
-
 bool NoopDriver::isWorkaroundNeeded(Workaround) {
    return false;
 }
@@ -252,8 +241,8 @@ void NoopDriver::update3DImage(Handle<HwTexture> th,
 void NoopDriver::setupExternalImage(void* image) {
 }

-TimerQueryResult NoopDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint64_t* elapsedTime) {
-    return TimerQueryResult::ERROR;
+bool NoopDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint64_t* elapsedTime) {
+    return false;
 }

 void NoopDriver::setExternalImage(Handle<HwTexture> th, void* image) {
@@ -350,26 +339,13 @@ void NoopDriver::blit(
        math::uint2 size) {
 }

-void NoopDriver::bindPipeline(PipelineState pipelineState) {
-}
-
-void NoopDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
-}
-
-void NoopDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
-}
-
 void NoopDriver::draw(PipelineState pipelineState, Handle<HwRenderPrimitive> rph,
-        uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
+        uint32_t instanceCount) {
 }

 void NoopDriver::dispatchCompute(Handle<HwProgram> program, math::uint3 workGroupCount) {
 }

-void NoopDriver::scissor(
-        Viewport scissor) {
-}
-
 void NoopDriver::beginTimerQuery(Handle<HwTimerQuery> tqh) {
 }

@@ -16,25 +16,10 @@

 #include "OpenGLContext.h"

-#include "GLUtils.h"
-#include "OpenGLTimerQuery.h"
-
 #include <backend/platforms/OpenGLPlatform.h>
-#include <backend/DriverEnums.h>

-#include <utils/compiler.h>
-#include <utils/debug.h>
-#include <utils/Log.h>
-#include <utils/ostream.h>
-
-#include <functional>
-#include <string_view>
 #include <utility>

-#include <stddef.h>
-#include <stdio.h>
-#include <string.h>
-
 // change to true to display all GL extensions in the console on start-up
 #define DEBUG_PRINT_EXTENSIONS false

@@ -63,8 +48,7 @@ bool OpenGLContext::queryOpenGLVersion(GLint* major, GLint* minor) noexcept {
 #endif
 }

-OpenGLContext::OpenGLContext(OpenGLPlatform& platform) noexcept
-        : mPlatform(platform) {
+OpenGLContext::OpenGLContext() noexcept {

    state.vao.p = &mDefaultVAO;

@@ -247,49 +231,6 @@ OpenGLContext::OpenGLContext(OpenGLPlatform& platform) noexcept
        glDebugMessageCallback(cb, nullptr);
    }
 #endif
-
-    mTimerQueryFactory = TimerQueryFactory::init(platform, *this);
-}
-
-OpenGLContext::~OpenGLContext() noexcept {
-    delete mTimerQueryFactory;
-}
-
-void OpenGLContext::destroyWithContext(
-        size_t index, std::function<void(OpenGLContext&)> const& closure) noexcept {
-    if (index == 0) {
-        // Note: we only need to delay the destruction of objects on the unprotected context
-        // (index 0) because the protected context is always immediately destroyed and all its
-        // active objects and bindings are then automatically destroyed.
-        // TODO: this is only guaranteed for EGLPlatform, but that's the only one we care about.
-        mDestroyWithNormalContext.push_back(closure);
-    }
-}
-
-void OpenGLContext::unbindEverything() noexcept {
-    // TODO:  we're supposed to unbind everything here so that resources don't get
-    //        stuck in this context (contextIndex) when destroyed in the other context.
-    //        However, because EGLPlatform always immediately destroys the protected context (1),
-    //        the bindings will automatically be severed when we switch back to the default context.
-    //        Since bindings now only exist in one context, we don't have a ref-counting issue to
-    //        worry about.
-}
-
-void OpenGLContext::synchronizeStateAndCache(size_t index) noexcept {
-
-    // if we're just switching back to context 0, run all the pending destructors
-    if (index == 0) {
-        auto list = std::move(mDestroyWithNormalContext);
-        for (auto&& fn: list) {
-            fn(*this);
-        }
-    }
-
-    // the default FBO could be invalid
-    mDefaultFbo[index].reset();
-
-    contextIndex = index;
-    resetState();
 }

 void OpenGLContext::setDefaultState() noexcept {
@@ -486,6 +427,7 @@ void OpenGLContext::initBugs(Bugs* bugs, Extensions const& exts,
            if (strstr(renderer, "Mali-T")) {
                bugs->disable_glFlush = true;
                bugs->disable_shared_context_draws = true;
+                bugs->texture_external_needs_rebind = true;
                // We have not verified that timer queries work on Mali-T, so we disable to be safe.
                bugs->dont_use_timer_query = true;
            }
@@ -658,7 +600,6 @@ void OpenGLContext::initExtensionsGLES(Extensions* ext, GLint major, GLint minor
    ext->EXT_disjoint_timer_query = exts.has("GL_EXT_disjoint_timer_query"sv);
    ext->EXT_multisampled_render_to_texture = exts.has("GL_EXT_multisampled_render_to_texture"sv);
    ext->EXT_multisampled_render_to_texture2 = exts.has("GL_EXT_multisampled_render_to_texture2"sv);
-    ext->EXT_protected_textures = exts.has("GL_EXT_protected_textures"sv);
 #endif
    ext->EXT_shader_framebuffer_fetch = exts.has("GL_EXT_shader_framebuffer_fetch"sv);
 #ifndef __EMSCRIPTEN__
@@ -682,7 +623,6 @@ void OpenGLContext::initExtensionsGLES(Extensions* ext, GLint major, GLint minor
    ext->OES_standard_derivatives = exts.has("GL_OES_standard_derivatives"sv);
    ext->OES_texture_npot = exts.has("GL_OES_texture_npot"sv);
    ext->OES_vertex_array_object = exts.has("GL_OES_vertex_array_object"sv);
-    ext->OVR_multiview2 = exts.has("GL_OVR_multiview2"sv);
    ext->WEBGL_compressed_texture_etc = exts.has("WEBGL_compressed_texture_etc"sv);
    ext->WEBGL_compressed_texture_s3tc = exts.has("WEBGL_compressed_texture_s3tc"sv);
    ext->WEBGL_compressed_texture_s3tc_srgb = exts.has("WEBGL_compressed_texture_s3tc_srgb"sv);
@@ -747,7 +687,6 @@ void OpenGLContext::initExtensionsGL(Extensions* ext, GLint major, GLint minor)
    ext->OES_standard_derivatives = true;
    ext->OES_texture_npot = true;
    ext->OES_vertex_array_object = true;
-    ext->OVR_multiview2 = exts.has("GL_OVR_multiview2"sv);
    ext->WEBGL_compressed_texture_etc = false;
    ext->WEBGL_compressed_texture_s3tc = false;
    ext->WEBGL_compressed_texture_s3tc_srgb = false;
@@ -769,51 +708,6 @@ void OpenGLContext::initExtensionsGL(Extensions* ext, GLint major, GLint minor)

 #endif // BACKEND_OPENGL_VERSION_GL

-
-GLuint OpenGLContext::bindFramebuffer(GLenum target, GLuint buffer) noexcept {
-    if (UTILS_UNLIKELY(buffer == 0)) {
-        // we're binding the default frame buffer, resolve its actual name
-        auto& defaultFboForThisContext = mDefaultFbo[contextIndex];
-        if (UTILS_UNLIKELY(!defaultFboForThisContext.has_value())) {
-            defaultFboForThisContext = GLuint(mPlatform.getDefaultFramebufferObject());
-        }
-        buffer = defaultFboForThisContext.value();
-    }
-    bindFramebufferResolved(target, buffer);
-    return buffer;
-}
-
-void OpenGLContext::unbindFramebuffer(GLenum target) noexcept {
-    bindFramebufferResolved(target, 0);
-}
-
-void OpenGLContext::bindFramebufferResolved(GLenum target, GLuint buffer) noexcept {
-    switch (target) {
-        case GL_FRAMEBUFFER:
-            if (state.draw_fbo != buffer || state.read_fbo != buffer) {
-                state.draw_fbo = state.read_fbo = buffer;
-                glBindFramebuffer(target, buffer);
-            }
-            break;
-#ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
-        case GL_DRAW_FRAMEBUFFER:
-            if (state.draw_fbo != buffer) {
-                state.draw_fbo = buffer;
-                glBindFramebuffer(target, buffer);
-            }
-            break;
-        case GL_READ_FRAMEBUFFER:
-            if (state.read_fbo != buffer) {
-                state.read_fbo = buffer;
-                glBindFramebuffer(target, buffer);
-            }
-            break;
-#endif
-        default:
-            break;
-    }
-}
-
 void OpenGLContext::bindBuffer(GLenum target, GLuint buffer) noexcept {
    if (target == GL_ELEMENT_ARRAY_BUFFER) {
        constexpr size_t targetIndex = getIndexForBufferTarget(GL_ELEMENT_ARRAY_BUFFER);
@@ -924,22 +818,19 @@ void OpenGLContext::deleteBuffers(GLsizei n, const GLuint* buffers, GLenum targe
 #endif
 }

-void OpenGLContext::deleteVertexArray(GLuint vao) noexcept {
-    if (UTILS_LIKELY(vao)) {
-        procs.deleteVertexArrays(1, &vao);
-        // if the destroyed VAO is bound, clear the binding.
-        if (state.vao.p->vao[contextIndex] == vao) {
+void OpenGLContext::deleteVertexArrays(GLsizei n, const GLuint* arrays) noexcept {
+    procs.deleteVertexArrays(n, arrays);
+    // if one of the destroyed VAO is bound, clear the binding.
+    for (GLsizei i = 0; i < n; ++i) {
+        if (state.vao.p->vao == arrays[i]) {
            bindVertexArray(nullptr);
+            break;
        }
    }
 }

 void OpenGLContext::resetState() noexcept {
    // Force GL state to match the Filament state
-
-    // increase the state version so other parts of the state know to reset
-    state.age++;
-
    if (state.major > 2) {
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
        glBindFramebuffer(GL_DRAW_FRAMEBUFFER, state.draw_fbo);
@@ -956,8 +847,11 @@ void OpenGLContext::resetState() noexcept {
    glUseProgram(state.program.use);

    // state.vao
-    state.vao.p = nullptr;
-    bindVertexArray(nullptr);
+    if (state.vao.p) {
+        procs.bindVertexArray(state.vao.p->vao);
+    } else {
+        bindVertexArray(nullptr);
+    }

    // state.raster
    glFrontFace(state.raster.frontFace);
@@ -1113,22 +1007,7 @@ void OpenGLContext::resetState() noexcept {
        state.window.viewport.w
    );
    glDepthRangef(state.window.depthRange.x, state.window.depthRange.y);
-}
-
-void OpenGLContext::createTimerQuery(GLTimerQuery* query) {
-    mTimerQueryFactory->createTimerQuery(query);
-}
-
-void OpenGLContext::destroyTimerQuery(GLTimerQuery* query) {
-    mTimerQueryFactory->destroyTimerQuery(query);
-}
-
-void OpenGLContext::beginTimeElapsedQuery(GLTimerQuery* query) {
-    mTimerQueryFactory->beginTimeElapsedQuery(query);
-}
-
-void OpenGLContext::endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQuery* query) {
-    mTimerQueryFactory->endTimeElapsedQuery(driver, query);
+    
 }

 } // namesapce filament
@@ -17,37 +17,25 @@
 #ifndef TNT_FILAMENT_BACKEND_OPENGLCONTEXT_H
 #define TNT_FILAMENT_BACKEND_OPENGLCONTEXT_H

-
-#include "OpenGLTimerQuery.h"
-
-#include <backend/platforms/OpenGLPlatform.h>
-
-#include <backend/DriverEnums.h>
-#include <backend/Handle.h>
-
-#include "gl_headers.h"
-
-#include <utils/compiler.h>
-#include <utils/bitset.h>
-#include <utils/debug.h>
-
-#include <math/vec2.h>
 #include <math/vec4.h>

-#include <array>
-#include <functional>
-#include <optional>
-#include <tuple>
-#include <vector>
+#include <utils/CString.h>
+#include <utils/debug.h>

-#include <stddef.h>
-#include <stdint.h>
+#include <backend/Handle.h>
+
+#include "GLUtils.h"
+
+#include <array>
+#include <set>
+#include <tuple>
+#include <utility>

 namespace filament::backend {

 class OpenGLPlatform;

-class OpenGLContext final : public TimerQueryFactoryInterface {
+class OpenGLContext {
 public:
    static constexpr const size_t MAX_TEXTURE_UNIT_COUNT = MAX_SAMPLER_COUNT;
    static constexpr const size_t DUMMY_TEXTURE_BINDING = 7; // highest binding guaranteed to work with ES2
@@ -58,29 +46,19 @@ public:
    struct RenderPrimitive {
        static_assert(MAX_VERTEX_ATTRIBUTE_COUNT <= 16);

-        GLuint vao[2] = {};                                     // 4
+        GLuint vao = 0;                                         // 4
        GLuint elementArray = 0;                                // 4
-        mutable utils::bitset<uint16_t> vertexAttribArray;      // 2
+        utils::bitset<uint16_t> vertexAttribArray;              // 2

-        // if this differs from vertexBufferWithObjects->bufferObjectsVersion, this VAO needs to
-        // be updated (see OpenGLDriver::updateVertexArrayObject())
+        // If this version number does not match vertexBufferWithObjects->bufferObjectsVersion,
+        // then the VAO needs to be updated.
        uint8_t vertexBufferVersion = 0;                        // 1
-
-        // if this differs from OpenGLContext::state.age, this VAO needs to
-        // be updated (see OpenGLDriver::updateVertexArrayObject())
-        uint8_t stateVersion = 0;                               // 1
-
-        // If this differs from OpenGLContext::state.age, this VAO's name needs to be updated.
-        // See OpenGLContext::bindVertexArray()
-        uint8_t nameVersion = 0;                                // 1
-
-        // Size in bytes of indices in the index buffer
        uint8_t indicesSize = 0;                                // 1

        // The optional 32-bit handle to a GLVertexBuffer is necessary only if the referenced
        // VertexBuffer supports buffer objects. If this is zero, then the VBO handles array is
        // immutable.
-        Handle<HwVertexBuffer> vertexBufferWithObjects;         // 4
+        Handle<HwVertexBuffer> vertexBufferWithObjects = {};    // 4

        GLenum getIndicesType() const noexcept {
            return indicesSize == 4 ? GL_UNSIGNED_INT : GL_UNSIGNED_SHORT;
@@ -89,18 +67,7 @@ public:

    static bool queryOpenGLVersion(GLint* major, GLint* minor) noexcept;

-    explicit OpenGLContext(OpenGLPlatform& platform) noexcept;
-    ~OpenGLContext() noexcept final;
-
-    // TimerQueryInterface ------------------------------------------------------------------------
-
-    // note: OpenGLContext being final ensures (clang) these are not called through the vtable
-    void createTimerQuery(GLTimerQuery* query) override;
-    void destroyTimerQuery(GLTimerQuery* query) override;
-    void beginTimeElapsedQuery(GLTimerQuery* query) override;
-    void endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQuery* query) override;
-
-    // --------------------------------------------------------------------------------------------
+    OpenGLContext() noexcept;

    template<int MAJOR, int MINOR>
    inline bool isAtLeastGL() const noexcept {
@@ -156,11 +123,10 @@ public:
    inline void bindBufferRange(GLenum target, GLuint index, GLuint buffer,
            GLintptr offset, GLsizeiptr size) noexcept;

-    GLuint bindFramebuffer(GLenum target, GLuint buffer) noexcept;
-    void unbindFramebuffer(GLenum target) noexcept;
+    inline void bindFramebuffer(GLenum target, GLuint buffer) noexcept;

-    inline void enableVertexAttribArray(RenderPrimitive const* rp, GLuint index) noexcept;
-    inline void disableVertexAttribArray(RenderPrimitive const* rp, GLuint index) noexcept;
+    inline void enableVertexAttribArray(GLuint index) noexcept;
+    inline void disableVertexAttribArray(GLuint index) noexcept;
    inline void enable(GLenum cap) noexcept;
    inline void disable(GLenum cap) noexcept;
    inline void frontFace(GLenum mode) noexcept;
@@ -182,9 +148,7 @@ public:
    inline void depthRange(GLclampf near, GLclampf far) noexcept;

    void deleteBuffers(GLsizei n, const GLuint* buffers, GLenum target) noexcept;
-    void deleteVertexArray(GLuint vao) noexcept;
-
-    void destroyWithContext(size_t index, std::function<void(OpenGLContext&)> const& closure) noexcept;
+    void deleteVertexArrays(GLsizei n, const GLuint* arrays) noexcept;

    // glGet*() values
    struct Gets {
@@ -219,7 +183,6 @@ public:
        bool EXT_discard_framebuffer;
        bool EXT_multisampled_render_to_texture2;
        bool EXT_multisampled_render_to_texture;
-        bool EXT_protected_textures;
        bool EXT_shader_framebuffer_fetch;
        bool EXT_texture_compression_bptc;
        bool EXT_texture_compression_etc2;
@@ -242,7 +205,6 @@ public:
        bool OES_standard_derivatives;
        bool OES_texture_npot;
        bool OES_vertex_array_object;
-        bool OVR_multiview2;
        bool WEBGL_compressed_texture_etc;
        bool WEBGL_compressed_texture_s3tc;
        bool WEBGL_compressed_texture_s3tc_srgb;
@@ -260,6 +222,10 @@ public:
        // Some drivers have gl state issues when drawing from shared contexts
        bool disable_shared_context_draws;

+        // Some drivers require the GL_TEXTURE_EXTERNAL_OES target to be bound when
+        // the texture image changes, even if it's already bound to that texture
+        bool texture_external_needs_rebind;
+
        // Some web browsers seem to immediately clear the default framebuffer when calling
        // glInvalidateFramebuffer with WebGL 2.0
        bool disable_invalidate_framebuffer;
@@ -328,19 +294,8 @@ public:

    FeatureLevel getFeatureLevel() const noexcept { return mFeatureLevel; }

-    // This is the index of the context in use. Must be 0 or 1. This is used to manange the
-    // OpenGL name of ContainerObjects within each context.
-    uint32_t contextIndex = 0;
-
    // Try to keep the State structure sorted by data-access patterns
    struct State {
-        State() noexcept = default;
-        // make sure we don't copy this state by accident
-        State(State const& rhs) = delete;
-        State(State&& rhs) noexcept = delete;
-        State& operator=(State const& rhs) = delete;
-        State& operator=(State&& rhs) noexcept = delete;
-
        GLint major = 0;
        GLint minor = 0;

@@ -445,7 +400,6 @@ public:
            vec4gli viewport { 0 };
            vec2glf depthRange { 0.0f, 1.0f };
        } window;
-        uint8_t age = 0;
    } state;

    struct Procs {
@@ -465,19 +419,9 @@ public:
        void (* maxShaderCompilerThreadsKHR)(GLuint count);
    } procs{};

-    void unbindEverything() noexcept;
-    void synchronizeStateAndCache(size_t index) noexcept;
-
 private:
-    OpenGLPlatform& mPlatform;
    ShaderModel mShaderModel = ShaderModel::MOBILE;
    FeatureLevel mFeatureLevel = FeatureLevel::FEATURE_LEVEL_1;
-    TimerQueryFactoryInterface* mTimerQueryFactory = nullptr;
-    std::vector<std::function<void(OpenGLContext&)>> mDestroyWithNormalContext;
-    RenderPrimitive mDefaultVAO;
-    std::optional<GLuint> mDefaultFbo[2];
-
-    void bindFramebufferResolved(GLenum target, GLuint buffer) noexcept;

    const std::array<std::tuple<bool const&, char const*, char const*>, sizeof(bugs)> mBugDatabase{{
            {   bugs.disable_glFlush,
@@ -489,6 +433,9 @@ private:
            {   bugs.disable_shared_context_draws,
                    "disable_shared_context_draws",
                    ""},
+            {   bugs.texture_external_needs_rebind,
+                    "texture_external_needs_rebind",
+                    ""},
            {   bugs.disable_invalidate_framebuffer,
                    "disable_invalidate_framebuffer",
                    ""},
@@ -530,6 +477,8 @@ private:
                    ""},
    }};

+    RenderPrimitive mDefaultVAO;
+
    // this is chosen to minimize code size
 #if defined(BACKEND_OPENGL_VERSION_GLES)
    static void initExtensionsGLES(Extensions* ext, GLint major, GLint minor) noexcept;
@@ -687,26 +636,11 @@ void OpenGLContext::depthRange(GLclampf near, GLclampf far) noexcept {
 void OpenGLContext::bindVertexArray(RenderPrimitive const* p) noexcept {
    RenderPrimitive* vao = p ? const_cast<RenderPrimitive *>(p) : &mDefaultVAO;
    update_state(state.vao.p, vao, [&]() {
-
-        // See if we need to create a name for this VAO on the fly, this would happen if:
-        // - we're not the default VAO, because its name is always 0
-        // - our name is 0, this could happen if this VAO was created in the "other" context
-        // - the nameVersion is out of date *and* we're on the protected context, in this case:
-        //      - the name must be stale from a previous use of this context because we always
-        //        destroy the protected context when we're done with it.
-        bool const recreateVaoName = p != &mDefaultVAO &&
-                ((vao->vao[contextIndex] == 0) ||
-                        (vao->nameVersion != state.age && contextIndex == 1));
-        if (UTILS_UNLIKELY(recreateVaoName)) {
-            vao->nameVersion = state.age;
-            procs.genVertexArrays(1, &vao->vao[contextIndex]);
-        }
-
-        procs.bindVertexArray(vao->vao[contextIndex]);
+        procs.bindVertexArray(vao->vao);
        // update GL_ELEMENT_ARRAY_BUFFER, which is updated by glBindVertexArray
        size_t const targetIndex = getIndexForBufferTarget(GL_ELEMENT_ARRAY_BUFFER);
        state.buffers.genericBinding[targetIndex] = vao->elementArray;
-        if (UTILS_UNLIKELY(bugs.vao_doesnt_store_element_array_buffer_binding || recreateVaoName)) {
+        if (UTILS_UNLIKELY(bugs.vao_doesnt_store_element_array_buffer_binding)) {
            // This shouldn't be needed, but it looks like some drivers don't do the implicit
            // glBindBuffer().
            glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, vao->elementArray);
@@ -744,13 +678,40 @@ void OpenGLContext::bindBufferRange(GLenum target, GLuint index, GLuint buffer,
 #endif
 }

+void OpenGLContext::bindFramebuffer(GLenum target, GLuint buffer) noexcept {
+    switch (target) {
+        case GL_FRAMEBUFFER:
+            if (state.draw_fbo != buffer || state.read_fbo != buffer) {
+                state.draw_fbo = state.read_fbo = buffer;
+                glBindFramebuffer(target, buffer);
+            }
+            break;
+#ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
+        case GL_DRAW_FRAMEBUFFER:
+            if (state.draw_fbo != buffer) {
+                state.draw_fbo = buffer;
+                glBindFramebuffer(target, buffer);
+            }
+            break;
+        case GL_READ_FRAMEBUFFER:
+            if (state.read_fbo != buffer) {
+                state.read_fbo = buffer;
+                glBindFramebuffer(target, buffer);
+            }
+            break;
+#endif
+        default:
+            break;
+    }
+}
+
 void OpenGLContext::bindTexture(GLuint unit, GLuint target, GLuint texId, size_t targetIndex) noexcept {
    assert_invariant(targetIndex == getIndexForTextureTarget(target));
    assert_invariant(targetIndex < TEXTURE_TARGET_COUNT);
    update_state(state.textures.units[unit].targets[targetIndex].texture_id, texId, [&]() {
        activeTexture(unit);
        glBindTexture(target, texId);
-    }, target == GL_TEXTURE_EXTERNAL_OES);
+    }, (target == GL_TEXTURE_EXTERNAL_OES) && bugs.texture_external_needs_rebind);
 }

 void OpenGLContext::bindTexture(GLuint unit, GLuint target, GLuint texId) noexcept {
@@ -763,22 +724,20 @@ void OpenGLContext::useProgram(GLuint program) noexcept {
    });
 }

-void OpenGLContext::enableVertexAttribArray(RenderPrimitive const* rp, GLuint index) noexcept {
-    assert_invariant(rp);
-    assert_invariant(index < rp->vertexAttribArray.size());
-    bool const force = rp->stateVersion != state.age;
-    if (UTILS_UNLIKELY(force || !rp->vertexAttribArray[index])) {
-        rp->vertexAttribArray.set(index);
+void OpenGLContext::enableVertexAttribArray(GLuint index) noexcept {
+    assert_invariant(state.vao.p);
+    assert_invariant(index < state.vao.p->vertexAttribArray.size());
+    if (UTILS_UNLIKELY(!state.vao.p->vertexAttribArray[index])) {
+        state.vao.p->vertexAttribArray.set(index);
        glEnableVertexAttribArray(index);
    }
 }

-void OpenGLContext::disableVertexAttribArray(RenderPrimitive const* rp, GLuint index) noexcept {
-    assert_invariant(rp);
-    assert_invariant(index < rp->vertexAttribArray.size());
-    bool const force = rp->stateVersion != state.age;
-    if (UTILS_UNLIKELY(force || rp->vertexAttribArray[index])) {
-        rp->vertexAttribArray.unset(index);
+void OpenGLContext::disableVertexAttribArray(GLuint index) noexcept {
+    assert_invariant(state.vao.p);
+    assert_invariant(index < state.vao.p->vertexAttribArray.size());
+    if (UTILS_UNLIKELY(state.vao.p->vertexAttribArray[index])) {
+        state.vao.p->vertexAttribArray.unset(index);
        glDisableVertexAttribArray(index);
    }
 }
@@ -16,54 +16,21 @@

 #include "OpenGLDriver.h"

+#include "private/backend/DriverApi.h"
+
 #include "CommandStreamDispatcher.h"
-#include "GLUtils.h"
 #include "OpenGLContext.h"
 #include "OpenGLDriverFactory.h"
 #include "OpenGLProgram.h"
 #include "OpenGLTimerQuery.h"
-#include "gl_headers.h"

 #include <backend/platforms/OpenGLPlatform.h>
-
-#include <backend/BufferDescriptor.h>
-#include <backend/CallbackHandler.h>
-#include <backend/DriverApiForward.h>
-#include <backend/DriverEnums.h>
-#include <backend/Handle.h>
-#include <backend/PipelineState.h>
-#include <backend/Platform.h>
-#include <backend/Program.h>
 #include <backend/SamplerDescriptor.h>
-#include <backend/TargetBufferInfo.h>

-#include "private/backend/Dispatcher.h"
-#include "private/backend/DriverApi.h"
-
-#include <utils/BitmaskEnum.h>
-#include <utils/CString.h>
+#include <utils/compiler.h>
 #include <utils/Log.h>
 #include <utils/Panic.h>
 #include <utils/Systrace.h>
-#include <utils/compiler.h>
-#include <utils/debug.h>
-#include <utils/ostream.h>
-
-#include <math/vec2.h>
-#include <math/vec3.h>
-
-#include <algorithm>
-#include <chrono>
-#include <functional>
-#include <limits>
-#include <memory>
-#include <mutex>
-#include <new>
-#include <utility>
-
-#include <stdint.h>
-#include <stdlib.h>
-#include <string.h>

 #if defined(__EMSCRIPTEN__)
 #include <emscripten.h>
@@ -123,28 +90,26 @@ Driver* OpenGLDriver::create(OpenGLPlatform* const platform,
 #if 0
    // this is useful for development, but too verbose even for debug builds
    // For reference on a 64-bits machine in Release mode:
+    //    GLFence                   :   8       few
    //    GLIndexBuffer             :   8       moderate
-    //    GLSamplerGroup            :  16       few
-    //    GLSwapChain               :  16       few
-    //    GLTimerQuery              :  16       few
-    //    GLFence                   :  24       few
-    //    GLRenderPrimitive         :  32       many
-    //    GLBufferObject            :  32       many
-    // -- less than or equal 32 bytes
-    //    OpenGLProgram             :  56       moderate
-    //    GLTexture                 :  64       moderate
+    //    GLSamplerGroup            :   8       few
+    // -- less than or equal 16 bytes
+    //    GLBufferObject            :  24       many
+    //    GLSync                    :  24       few
+    //    GLTimerQuery              :  32       few
+    //    OpenGLProgram             :  32       moderate
+    //    GLRenderPrimitive         :  48       many
    // -- less than or equal 64 bytes
-    //    GLVertexBuffer            :  76       moderate
-    //    GLStream                  : 104       few
+    //    GLTexture                 :  72       moderate
    //    GLRenderTarget            : 112       few
-    //    GLVertexBufferInfo        : 132       moderate
-    // -- less than or equal to 136 bytes
+    //    GLStream                  : 184       few
+    //    GLVertexBuffer            : 200       moderate
+    // -- less than or equal to 208 bytes

    slog.d
-           << "\nGLSwapChain: " << sizeof(GLSwapChain)
+           << "HwFence: " << sizeof(HwFence)
           << "\nGLBufferObject: " << sizeof(GLBufferObject)
           << "\nGLVertexBuffer: " << sizeof(GLVertexBuffer)
-           << "\nGLVertexBufferInfo: " << sizeof(GLVertexBufferInfo)
           << "\nGLIndexBuffer: " << sizeof(GLIndexBuffer)
           << "\nGLSamplerGroup: " << sizeof(GLSamplerGroup)
           << "\nGLRenderPrimitive: " << sizeof(GLRenderPrimitive)
@@ -152,7 +117,7 @@ Driver* OpenGLDriver::create(OpenGLPlatform* const platform,
           << "\nGLTimerQuery: " << sizeof(GLTimerQuery)
           << "\nGLStream: " << sizeof(GLStream)
           << "\nGLRenderTarget: " << sizeof(GLRenderTarget)
-           << "\nGLFence: " << sizeof(GLFence)
+           << "\nGLSync: " << sizeof(GLSync)
           << "\nOpenGLProgram: " << sizeof(OpenGLProgram)
           << io::endl;
 #endif
@@ -182,9 +147,9 @@ Driver* OpenGLDriver::create(OpenGLPlatform* const platform,
 #endif

    size_t const defaultSize = FILAMENT_OPENGL_HANDLE_ARENA_SIZE_IN_MB * 1024U * 1024U;
-    Platform::DriverConfig validConfig{ driverConfig };
+    Platform::DriverConfig validConfig {driverConfig};
    validConfig.handleArenaSize = std::max(driverConfig.handleArenaSize, defaultSize);
-    OpenGLDriver* const driver = new(std::nothrow) OpenGLDriver(ec, validConfig);
+    OpenGLDriver* const driver = new OpenGLDriver(ec, validConfig);
    return driver;
 }

@@ -203,13 +168,10 @@ OpenGLDriver::DebugMarker::~DebugMarker() noexcept {

 OpenGLDriver::OpenGLDriver(OpenGLPlatform* platform, const Platform::DriverConfig& driverConfig) noexcept
        : mPlatform(*platform),
-          mContext(mPlatform),
+          mContext(),
          mShaderCompilerService(*this),
-          mHandleAllocator("Handles",
-                  driverConfig.handleArenaSize,
-                  driverConfig.disableHandleUseAfterFreeCheck),
-          mSamplerMap(32),
-          mDriverConfig(driverConfig) {
+          mHandleAllocator("Handles", driverConfig.handleArenaSize),
+          mSamplerMap(32) {

    std::fill(mSamplerBindings.begin(), mSamplerBindings.end(), nullptr);

@@ -229,6 +191,8 @@ OpenGLDriver::OpenGLDriver(OpenGLPlatform* platform, const Platform::DriverConfi
    assert_invariant(mContext.ext.EXT_disjoint_timer_query);
 #endif

+    mTimerQueryImpl = OpenGLTimerQueryFactory::init(mPlatform, *this);
+
    mShaderCompilerService.init();
 }

@@ -273,6 +237,8 @@ void OpenGLDriver::terminate() {
    }
 #endif

+    delete mTimerQueryImpl;
+
    mPlatform.terminate();
 }

@@ -419,10 +385,6 @@ void OpenGLDriver::setStencilState(StencilState ss) noexcept {
 // Creating driver objects
 // ------------------------------------------------------------------------------------------------

-Handle<HwVertexBufferInfo> OpenGLDriver::createVertexBufferInfoS() noexcept {
-    return initHandle<GLVertexBufferInfo>();
-}
-
 Handle<HwVertexBuffer> OpenGLDriver::createVertexBufferS() noexcept {
    return initHandle<GLVertexBuffer>();
 }
@@ -483,21 +445,14 @@ Handle<HwTimerQuery> OpenGLDriver::createTimerQueryS() noexcept {
    return initHandle<GLTimerQuery>();
 }

-void OpenGLDriver::createVertexBufferInfoR(
-        Handle<HwVertexBufferInfo> vbih,
-        uint8_t bufferCount,
-        uint8_t attributeCount,
-        AttributeArray attributes) {
-    DEBUG_MARKER()
-    construct<GLVertexBufferInfo>(vbih, bufferCount, attributeCount, attributes);
-}
-
 void OpenGLDriver::createVertexBufferR(
        Handle<HwVertexBuffer> vbh,
-        uint32_t vertexCount,
-        Handle<HwVertexBufferInfo> vbih) {
+        uint8_t bufferCount,
+        uint8_t attributeCount,
+        uint32_t elementCount,
+        AttributeArray attributes) {
    DEBUG_MARKER()
-    construct<GLVertexBuffer>(vbh, vertexCount, vbih);
+    construct<GLVertexBuffer>(vbh, bufferCount, attributeCount, elementCount, attributes);
 }

 void OpenGLDriver::createIndexBufferR(
@@ -545,33 +500,31 @@ void OpenGLDriver::createBufferObjectR(Handle<HwBufferObject> boh,

 void OpenGLDriver::createRenderPrimitiveR(Handle<HwRenderPrimitive> rph,
        Handle<HwVertexBuffer> vbh, Handle<HwIndexBuffer> ibh,
-        PrimitiveType pt) {
+        PrimitiveType pt, uint32_t offset,
+        uint32_t minIndex, uint32_t maxIndex, uint32_t count) {
    DEBUG_MARKER()

    auto& gl = mContext;

+    GLVertexBuffer const* const eb = handle_cast<const GLVertexBuffer*>(vbh);
    GLIndexBuffer const* const ib = handle_cast<const GLIndexBuffer*>(ibh);
    assert_invariant(ib->elementSize == 2 || ib->elementSize == 4);

-    GLVertexBuffer* const vb = handle_cast<GLVertexBuffer*>(vbh);
-    GLRenderPrimitive* const rp = handle_cast<GLRenderPrimitive*>(rph);
+    GLRenderPrimitive* rp = handle_cast<GLRenderPrimitive*>(rph);
    rp->gl.indicesSize = (ib->elementSize == 4u) ? 4u : 2u;
    rp->gl.vertexBufferWithObjects = vbh;
    rp->type = pt;
-    rp->vbih = vb->vbih;
+    rp->offset = offset * rp->gl.indicesSize;
+    rp->count = count;
+    rp->minIndex = minIndex;
+    rp->maxIndex = maxIndex > minIndex ? maxIndex : rp->maxVertexCount - 1; // sanitize max index

-    // create a name for this VAO in the current context
-    gl.procs.genVertexArrays(1, &rp->gl.vao[gl.contextIndex]);
+    gl.procs.genVertexArrays(1, &rp->gl.vao);

-    // this implies our name is up-to-date
-    rp->gl.nameVersion = gl.state.age;
-
-    // binding the VAO will actually create it
    gl.bindVertexArray(&rp->gl);

-    // Note: we don't update the vertex buffer bindings in the VAO just yet, we will do that
-    // later in draw() or bindRenderPrimitive(). At this point, the HwVertexBuffer might not
-    // have all its buffers set.
+    // update the VBO bindings in the VAO
+    updateVertexArrayObject(rp, eb);

    // this records the index buffer into the currently bound VAO
    gl.bindBuffer(GL_ELEMENT_ARRAY_BUFFER, ib->gl.buffer);
@@ -618,22 +571,13 @@ void OpenGLDriver::createSamplerGroupR(Handle<HwSamplerGroup> sbh, uint32_t size

 UTILS_NOINLINE
 void OpenGLDriver::textureStorage(OpenGLDriver::GLTexture* t,
-        uint32_t width, uint32_t height, uint32_t depth, bool useProtectedMemory) noexcept {
+        uint32_t width, uint32_t height, uint32_t depth) noexcept {

    auto& gl = mContext;

    bindTexture(OpenGLContext::DUMMY_TEXTURE_BINDING, t);
    gl.activeTexture(OpenGLContext::DUMMY_TEXTURE_BINDING);

-#ifdef GL_EXT_protected_textures
-#ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
-    if (UTILS_UNLIKELY(useProtectedMemory)) {
-        assert_invariant(gl.ext.EXT_protected_textures);
-        glTexParameteri(t->gl.target, GL_TEXTURE_PROTECTED_EXT, 1);
-    }
-#endif
-#endif
-
    switch (t->gl.target) {
        case GL_TEXTURE_2D:
        case GL_TEXTURE_CUBE_MAP:
@@ -724,21 +668,6 @@ void OpenGLDriver::createTextureR(Handle<HwTexture> th, SamplerType target, uint
    GLenum internalFormat = getInternalFormat(format);
    assert_invariant(internalFormat);

-    if (any(usage & TextureUsage::PROTECTED)) {
-        // renderbuffers don't have a protected mode, so we need to use a texture
-        // because protected textures are only supported on GLES 3.2, MSAA will be available.
-        usage |= TextureUsage::SAMPLEABLE;
-    } else if (any(usage & TextureUsage::UPLOADABLE)) {
-        // if we have the uploadable flag, we also need to use a texture
-        usage |= TextureUsage::SAMPLEABLE;
-    } else if (target != SamplerType::SAMPLER_2D) {
-        // renderbuffers can only be 2D
-        usage |= TextureUsage::SAMPLEABLE;
-    } else if (levels > 1) {
-        // renderbuffers can't have mip levels
-        usage |= TextureUsage::SAMPLEABLE;
-    }
-
    auto& gl = mContext;
    samples = std::clamp(samples, uint8_t(1u), uint8_t(gl.gets.max_samples));
    GLTexture* t = construct<GLTexture>(th, target, levels, samples, w, h, depth, format, usage);
@@ -774,23 +703,23 @@ void OpenGLDriver::createTextureR(Handle<HwTexture> th, SamplerType target, uint
                    // we can't be here -- doesn't matter what we do
                case SamplerType::SAMPLER_2D:
                    t->gl.target = GL_TEXTURE_2D;
-                    t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_2D);
+                    t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_2D);
                    break;
                case SamplerType::SAMPLER_3D:
                    t->gl.target = GL_TEXTURE_3D;
-                    t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_3D);
+                    t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_3D);
                    break;
                case SamplerType::SAMPLER_2D_ARRAY:
                    t->gl.target = GL_TEXTURE_2D_ARRAY;
-                    t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_2D_ARRAY);
+                    t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_2D_ARRAY);
                    break;
                case SamplerType::SAMPLER_CUBEMAP:
                    t->gl.target = GL_TEXTURE_CUBE_MAP;
-                    t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_CUBE_MAP);
+                    t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_CUBE_MAP);
                    break;
                case SamplerType::SAMPLER_CUBEMAP_ARRAY:
                    t->gl.target = GL_TEXTURE_CUBE_MAP_ARRAY;
-                    t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_CUBE_MAP_ARRAY);
+                    t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_CUBE_MAP_ARRAY);
                    break;
            }

@@ -802,16 +731,21 @@ void OpenGLDriver::createTextureR(Handle<HwTexture> th, SamplerType target, uint
                    // multi-sample texture on GL 3.2 / GLES 3.1 and above
                    t->gl.target = GL_TEXTURE_2D_MULTISAMPLE;
                    t->gl.targetIndex = (uint8_t)
-                            OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_2D_MULTISAMPLE);
+                            gl.getIndexForTextureTarget(GL_TEXTURE_2D_MULTISAMPLE);
                } else {
                    // Turn off multi-sampling for that texture. It's just not supported.
                }
 #endif
            }
-
-            textureStorage(t, w, h, depth, bool(usage & TextureUsage::PROTECTED));
+            textureStorage(t, w, h, depth);
        }
    } else {
+        assert_invariant(any(usage & (
+                TextureUsage::COLOR_ATTACHMENT |
+                TextureUsage::DEPTH_ATTACHMENT |
+                TextureUsage::STENCIL_ATTACHMENT)));
+        assert_invariant(levels == 1);
+        assert_invariant(target == SamplerType::SAMPLER_2D);
        t->gl.internalFormat = internalFormat;
        t->gl.target = GL_RENDERBUFFER;
        glGenRenderbuffers(1, &t->gl.id);
@@ -866,27 +800,27 @@ void OpenGLDriver::importTextureR(Handle<HwTexture> th, intptr_t id,
    switch (target) {
        case SamplerType::SAMPLER_EXTERNAL:
            t->gl.target = GL_TEXTURE_EXTERNAL_OES;
-            t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_EXTERNAL_OES);
+            t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_EXTERNAL_OES);
            break;
        case SamplerType::SAMPLER_2D:
            t->gl.target = GL_TEXTURE_2D;
-            t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_2D);
+            t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_2D);
            break;
        case SamplerType::SAMPLER_3D:
            t->gl.target = GL_TEXTURE_3D;
-            t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_3D);
+            t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_3D);
            break;
        case SamplerType::SAMPLER_2D_ARRAY:
            t->gl.target = GL_TEXTURE_2D_ARRAY;
-            t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_2D_ARRAY);
+            t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_2D_ARRAY);
            break;
        case SamplerType::SAMPLER_CUBEMAP:
            t->gl.target = GL_TEXTURE_CUBE_MAP;
-            t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_CUBE_MAP);
+            t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_CUBE_MAP);
            break;
        case SamplerType::SAMPLER_CUBEMAP_ARRAY:
            t->gl.target = GL_TEXTURE_CUBE_MAP_ARRAY;
-            t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_CUBE_MAP_ARRAY);
+            t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_CUBE_MAP_ARRAY);
            break;
    }

@@ -897,7 +831,7 @@ void OpenGLDriver::importTextureR(Handle<HwTexture> th, intptr_t id,
        if (gl.features.multisample_texture) {
            // multi-sample texture on GL 3.2 / GLES 3.1 and above
            t->gl.target = GL_TEXTURE_2D_MULTISAMPLE;
-            t->gl.targetIndex = OpenGLContext::getIndexForTextureTarget(GL_TEXTURE_2D_MULTISAMPLE);
+            t->gl.targetIndex = (uint8_t)gl.getIndexForTextureTarget(GL_TEXTURE_2D_MULTISAMPLE);
        } else {
            // Turn off multi-sampling for that texture. It's just not supported.
        }
@@ -908,65 +842,61 @@ void OpenGLDriver::importTextureR(Handle<HwTexture> th, intptr_t id,
 }

 void OpenGLDriver::updateVertexArrayObject(GLRenderPrimitive* rp, GLVertexBuffer const* vb) {
-    // NOTE: this is called from draw() and must be as efficient as possible.

    auto& gl = mContext;

+    // NOTE: this is called from draw() and must be as efficient as possible.
+
+
    if (UTILS_LIKELY(gl.ext.OES_vertex_array_object)) {
        // The VAO for the given render primitive must already be bound.
+#ifndef NDEBUG
        GLint vaoBinding;
        glGetIntegerv(GL_VERTEX_ARRAY_BINDING, &vaoBinding);
-        assert_invariant(vaoBinding == (GLint)rp->gl.vao[gl.contextIndex]);
+        assert_invariant(vaoBinding == (GLint)rp->gl.vao);
+#endif
+        rp->gl.vertexBufferVersion = vb->bufferObjectsVersion;
+    } else {
+        // if we don't have OES_vertex_array_object, we never update the buffer version so
+        // that it's always reset in draw
    }

-    if (UTILS_LIKELY(rp->gl.vertexBufferVersion == vb->bufferObjectsVersion &&
-                     rp->gl.stateVersion == gl.state.age)) {
-        return;
-    }
-
-    GLVertexBufferInfo const* const vbi = handle_cast<const GLVertexBufferInfo*>(vb->vbih);
-
-    for (size_t i = 0, n = vbi->attributes.size(); i < n; i++) {
-        const auto& attribute = vbi->attributes[i];
+    rp->maxVertexCount = vb->vertexCount;
+    for (size_t i = 0, n = vb->attributes.size(); i < n; i++) {
+        const auto& attribute = vb->attributes[i];
        const uint8_t bi = attribute.buffer;
-        if (bi != Attribute::BUFFER_UNUSED) {
-            // if a buffer is defined it must not be invalid.
-            assert_invariant(vb->gl.buffers[bi]);

-            // if w're on ES2, the user shouldn't use FLAG_INTEGER_TARGET
+        // Invoking glVertexAttribPointer without a bound VBO is an invalid operation, so we must
+        // take care to avoid it. This can occur when VertexBuffer is only partially populated with
+        // BufferObject items.
+        if (bi != Attribute::BUFFER_UNUSED && UTILS_LIKELY(vb->gl.buffers[bi] != 0)) {
+
            assert_invariant(!(gl.isES2() && (attribute.flags & Attribute::FLAG_INTEGER_TARGET)));

            gl.bindBuffer(GL_ARRAY_BUFFER, vb->gl.buffers[bi]);
-            GLuint const index = i;
-            GLint const size = (GLint)getComponentCount(attribute.type);
-            GLenum const type = getComponentType(attribute.type);
-            GLboolean const normalized = getNormalization(attribute.flags & Attribute::FLAG_NORMALIZED);
-            GLsizei const stride = attribute.stride;
-            void const* pointer = reinterpret_cast<void const *>(attribute.offset);
-
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
            if (UTILS_UNLIKELY(attribute.flags & Attribute::FLAG_INTEGER_TARGET)) {
-                // integer attributes can't be floats
-                assert_invariant(type == GL_BYTE || type == GL_UNSIGNED_BYTE || type == GL_SHORT ||
-                        type == GL_UNSIGNED_SHORT || type == GL_INT || type == GL_UNSIGNED_INT);
-                glVertexAttribIPointer(index, size, type, stride, pointer);
+                glVertexAttribIPointer(GLuint(i),
+                        (GLint)getComponentCount(attribute.type),
+                        getComponentType(attribute.type),
+                        attribute.stride,
+                        (void*) uintptr_t(attribute.offset));
            } else
 #endif
            {
-                glVertexAttribPointer(index, size, type, normalized, stride, pointer);
+                glVertexAttribPointer(GLuint(i),
+                        (GLint)getComponentCount(attribute.type),
+                        getComponentType(attribute.type),
+                        getNormalization(attribute.flags & Attribute::FLAG_NORMALIZED),
+                        attribute.stride,
+                        (void*) uintptr_t(attribute.offset));
            }

-            gl.enableVertexAttribArray(&rp->gl, GLuint(i));
+            gl.enableVertexAttribArray(GLuint(i));
        } else {
-            // In some OpenGL implementations, we must supply a properly-typed placeholder for
-            // every integer input that is declared in the vertex shader.
-            // Note that the corresponding doesn't have to be enabled and in fact won't be. If it
-            // was enabled, it would indicate a user-error (providing the wrong type of array).
-            // With a disabled array, the vertex shader gets the attribute from glVertexAttrib,
-            // and must have the proper intergerness.
-            // But at this point, we don't know what the shader requirements are, and so we must
-            // rely on the attribute.

+            // In some OpenGL implementations, we must supply a properly-typed placeholder for
+            // every integer input that is declared in the vertex shader, even if disabled.
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
            if (UTILS_UNLIKELY(attribute.flags & Attribute::FLAG_INTEGER_TARGET)) {
                if (!gl.isES2()) {
@@ -979,21 +909,13 @@ void OpenGLDriver::updateVertexArrayObject(GLRenderPrimitive* rp, GLVertexBuffer
                glVertexAttrib4f(GLuint(i), 0, 0, 0, 0);
            }

-            gl.disableVertexAttribArray(&rp->gl, GLuint(i));
+            gl.disableVertexAttribArray(GLuint(i));
        }
    }
-
-    rp->gl.stateVersion = gl.state.age;
-    if (UTILS_LIKELY(gl.ext.OES_vertex_array_object)) {
-        rp->gl.vertexBufferVersion = vb->bufferObjectsVersion;
-    } else {
-        // if we don't have OES_vertex_array_object, we never update the buffer version so
-        // that it's always reset in draw
-    }
 }

 void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
-        GLRenderTarget const* rt, GLenum attachment, uint8_t layerCount) noexcept {
+        GLRenderTarget const* rt, GLenum attachment) noexcept {

 #if !defined(NDEBUG)
    // Only used by assert_invariant() checks below
@@ -1131,20 +1053,9 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
            case GL_TEXTURE_2D_ARRAY:
            case GL_TEXTURE_CUBE_MAP_ARRAY:
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
-
-                // TODO: support multiview for iOS and WebGL
-#if !defined(__EMSCRIPTEN__) && !defined(IOS)
-                if (layerCount > 1) {
-                    // if layerCount > 1, it means we use the multiview extension.
-                    glFramebufferTextureMultiviewOVR(GL_FRAMEBUFFER, attachment,
-                        t->gl.id, 0, binfo.baseViewIndex, layerCount);
-                } else
-#endif // !defined(__EMSCRIPTEN__) && !defined(IOS)
-                {
-                    // GL_TEXTURE_2D_MULTISAMPLE_ARRAY is not supported in GLES
-                    glFramebufferTextureLayer(GL_FRAMEBUFFER, attachment,
+                // GL_TEXTURE_2D_MULTISAMPLE_ARRAY is not supported in GLES
+                glFramebufferTextureLayer(GL_FRAMEBUFFER, attachment,
                        t->gl.id, binfo.level, binfo.layer);
-                }
 #endif
                break;
            default:
@@ -1305,9 +1216,11 @@ void OpenGLDriver::createDefaultRenderTargetR(

    construct<GLRenderTarget>(rth, 0, 0);  // FIXME: we don't know the width/height

+    uint32_t const framebuffer = mPlatform.createDefaultRenderTarget();
+
    GLRenderTarget* rt = handle_cast<GLRenderTarget*>(rth);
    rt->gl.isDefault = true;
-    rt->gl.fbo = 0; // the actual id is resolved at binding time
+    rt->gl.fbo = framebuffer;
    rt->gl.samples = 1;
    // FIXME: these flags should reflect the actual attachments present
    rt->targets = TargetBufferFlags::COLOR0 | TargetBufferFlags::DEPTH;
@@ -1318,7 +1231,6 @@ void OpenGLDriver::createRenderTargetR(Handle<HwRenderTarget> rth,
        uint32_t width,
        uint32_t height,
        uint8_t samples,
-        uint8_t layerCount,
        MRT color,
        TargetBufferInfo depth,
        TargetBufferInfo stencil) {
@@ -1385,7 +1297,7 @@ void OpenGLDriver::createRenderTargetR(Handle<HwRenderTarget> rth,
            if (any(targets & getTargetBufferFlagsAt(i))) {
                assert_invariant(color[i].handle);
                rt->gl.color[i] = handle_cast<GLTexture*>(color[i].handle);
-                framebufferTexture(color[i], rt, GL_COLOR_ATTACHMENT0 + i, layerCount);
+                framebufferTexture(color[i], rt, GL_COLOR_ATTACHMENT0 + i);
                bufs[i] = GL_COLOR_ATTACHMENT0 + i;
                checkDimensions(rt->gl.color[i], color[i].level);
            }
@@ -1407,7 +1319,7 @@ void OpenGLDriver::createRenderTargetR(Handle<HwRenderTarget> rth,
        // either we supplied only the depth handle or both depth/stencil are identical and not null
        if (depth.handle && (stencil.handle == depth.handle || !stencil.handle)) {
            rt->gl.depth = handle_cast<GLTexture*>(depth.handle);
-            framebufferTexture(depth, rt, GL_DEPTH_STENCIL_ATTACHMENT, layerCount);
+            framebufferTexture(depth, rt, GL_DEPTH_STENCIL_ATTACHMENT);
            specialCased = true;
            checkDimensions(rt->gl.depth, depth.level);
        }
@@ -1418,13 +1330,13 @@ void OpenGLDriver::createRenderTargetR(Handle<HwRenderTarget> rth,
        if (any(targets & TargetBufferFlags::DEPTH)) {
            assert_invariant(depth.handle);
            rt->gl.depth = handle_cast<GLTexture*>(depth.handle);
-            framebufferTexture(depth, rt, GL_DEPTH_ATTACHMENT, layerCount);
+            framebufferTexture(depth, rt, GL_DEPTH_ATTACHMENT);
            checkDimensions(rt->gl.depth, depth.level);
        }
        if (any(targets & TargetBufferFlags::STENCIL)) {
            assert_invariant(stencil.handle);
            rt->gl.stencil = handle_cast<GLTexture*>(stencil.handle);
-            framebufferTexture(stencil, rt, GL_STENCIL_ATTACHMENT, layerCount);
+            framebufferTexture(stencil, rt, GL_STENCIL_ATTACHMENT);
            checkDimensions(rt->gl.stencil, stencil.level);
        }
    }
@@ -1466,13 +1378,6 @@ void OpenGLDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow,
    GLSwapChain* sc = handle_cast<GLSwapChain*>(sch);
    sc->swapChain = mPlatform.createSwapChain(nativeWindow, flags);

-#if !defined(__EMSCRIPTEN__)
-    // note: in practice this should never happen on Android
-    ASSERT_POSTCONDITION(sc->swapChain,
-            "createSwapChain(%p, 0x%lx) failed. See logs for details.",
-            nativeWindow, flags);
-#endif
-
    // See if we need the emulated rec709 output conversion
    if (UTILS_UNLIKELY(mContext.isES2())) {
        sc->rec709 = (flags & SWAP_CHAIN_CONFIG_SRGB_COLORSPACE &&
@@ -1487,13 +1392,6 @@ void OpenGLDriver::createSwapChainHeadlessR(Handle<HwSwapChain> sch,
    GLSwapChain* sc = handle_cast<GLSwapChain*>(sch);
    sc->swapChain = mPlatform.createSwapChain(width, height, flags);

-#if !defined(__EMSCRIPTEN__)
-    // note: in practice this should never happen on Android
-    ASSERT_POSTCONDITION(sc->swapChain,
-            "createSwapChainHeadless(%u, %u, 0x%lx) failed. See logs for details.",
-            width, height, flags);
-#endif
-
    // See if we need the emulated rec709 output conversion
    if (UTILS_UNLIKELY(mContext.isES2())) {
        sc->rec709 = (flags & SWAP_CHAIN_CONFIG_SRGB_COLORSPACE &&
@@ -1504,21 +1402,13 @@ void OpenGLDriver::createSwapChainHeadlessR(Handle<HwSwapChain> sch,
 void OpenGLDriver::createTimerQueryR(Handle<HwTimerQuery> tqh, int) {
    DEBUG_MARKER()
    GLTimerQuery* tq = handle_cast<GLTimerQuery*>(tqh);
-    mContext.createTimerQuery(tq);
+    mTimerQueryImpl->createTimerQuery(tq);
 }

 // ------------------------------------------------------------------------------------------------
 // Destroying driver objects
 // ------------------------------------------------------------------------------------------------

-void OpenGLDriver::destroyVertexBufferInfo(Handle<HwVertexBufferInfo> vbih) {
-    DEBUG_MARKER()
-    if (vbih) {
-        GLVertexBufferInfo const* vbi = handle_cast<const GLVertexBufferInfo*>(vbih);
-        destruct(vbih, vbi);
-    }
-}
-
 void OpenGLDriver::destroyVertexBuffer(Handle<HwVertexBuffer> vbh) {
    DEBUG_MARKER()
    if (vbh) {
@@ -1558,20 +1448,7 @@ void OpenGLDriver::destroyRenderPrimitive(Handle<HwRenderPrimitive> rph) {
    if (rph) {
        auto& gl = mContext;
        GLRenderPrimitive const* rp = handle_cast<const GLRenderPrimitive*>(rph);
-        gl.deleteVertexArray(rp->gl.vao[gl.contextIndex]);
-
-        // If we have a name in the "other" context, we need to schedule the destroy for
-        // later, because it can't be done here. VAOs are "container objects" and are not
-        // shared between contexts.
-        size_t const otherContextIndex = 1 - gl.contextIndex;
-        GLuint const nameInOtherContext = rp->gl.vao[otherContextIndex];
-        if (UTILS_UNLIKELY(nameInOtherContext)) {
-            gl.destroyWithContext(otherContextIndex,
-                    [name = nameInOtherContext](OpenGLContext& gl) {
-                gl.deleteVertexArray(name);
-            });
-        }
-
+        gl.deleteVertexArrays(1, &rp->gl.vao);
        destruct(rph, rp);
    }
 }
@@ -1636,11 +1513,11 @@ void OpenGLDriver::destroyRenderTarget(Handle<HwRenderTarget> rth) {
        GLRenderTarget* rt = handle_cast<GLRenderTarget*>(rth);
        if (rt->gl.fbo) {
            // first unbind this framebuffer if needed
-            gl.unbindFramebuffer(GL_FRAMEBUFFER);
+            gl.bindFramebuffer(GL_FRAMEBUFFER, 0);
        }
        if (rt->gl.fbo_read) {
            // first unbind this framebuffer if needed
-            gl.unbindFramebuffer(GL_FRAMEBUFFER);
+            gl.bindFramebuffer(GL_FRAMEBUFFER, 0);
        }

 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
@@ -1712,7 +1589,7 @@ void OpenGLDriver::destroyTimerQuery(Handle<HwTimerQuery> tqh) {

    if (tqh) {
        GLTimerQuery* tq = handle_cast<GLTimerQuery*>(tqh);
-        mContext.destroyTimerQuery(tq);
+        mTimerQueryImpl->destroyTimerQuery(tq);
        destruct(tqh, tq);
    }
 }
@@ -2003,7 +1880,7 @@ bool OpenGLDriver::isFrameBufferFetchMultiSampleSupported() {
 }

 bool OpenGLDriver::isFrameTimeSupported() {
-    return TimerQueryFactory::isGpuTimeSupported();
+    return OpenGLTimerQueryFactory::isGpuTimeSupported();
 }

 bool OpenGLDriver::isAutoDepthResolveSupported() {
@@ -2021,24 +1898,12 @@ bool OpenGLDriver::isSRGBSwapChainSupported() {
    return mPlatform.isSRGBSwapChainSupported();
 }

-bool OpenGLDriver::isProtectedContentSupported() {
-    return mPlatform.isProtectedContextSupported();
-}
-
-bool OpenGLDriver::isStereoSupported(backend::StereoscopicType stereoscopicType) {
-    // Instanced-stereo requires instancing and EXT_clip_cull_distance.
-    // Multiview-stereo requires ES 3.0 and OVR_multiview2.
+bool OpenGLDriver::isStereoSupported() {
+    // Stereo requires instancing and EXT_clip_cull_distance.
    if (UTILS_UNLIKELY(mContext.isES2())) {
        return false;
    }
-    switch (stereoscopicType) {
-    case backend::StereoscopicType::INSTANCED:
-        return mContext.ext.EXT_clip_cull_distance;
-    case backend::StereoscopicType::MULTIVIEW:
-        return mContext.ext.OVR_multiview2;
-    default:
-        return false;
-    }
+    return mContext.ext.EXT_clip_cull_distance;
 }

 bool OpenGLDriver::isParallelShaderCompileSupported() {
@@ -2049,10 +1914,6 @@ bool OpenGLDriver::isDepthStencilResolveSupported() {
    return true;
 }

-bool OpenGLDriver::isProtectedTexturesSupported() {
-    return getContext().ext.EXT_protected_textures;
-}
-
 bool OpenGLDriver::isWorkaroundNeeded(Workaround workaround) {
    switch (workaround) {
        case Workaround::SPLIT_EASU:
@@ -2120,18 +1981,7 @@ void OpenGLDriver::makeCurrent(Handle<HwSwapChain> schDraw, Handle<HwSwapChain>

    GLSwapChain* scDraw = handle_cast<GLSwapChain*>(schDraw);
    GLSwapChain* scRead = handle_cast<GLSwapChain*>(schRead);
-
-    mPlatform.makeCurrent(scDraw->swapChain, scRead->swapChain,
-            [this]() {
-                // OpenGL context is about to change, unbind everything
-                mContext.unbindEverything();
-            },
-            [this](size_t index) {
-                // OpenGL context has changed, resynchronize the state with the cache
-                mContext.synchronizeStateAndCache(index);
-                slog.d << "*** OpenGL context change : " << (index ? "protected" : "default") << io::endl;
-            });
-
+    mPlatform.makeCurrent(scDraw->swapChain, scRead->swapChain);
    mCurrentDrawSwapChain = scDraw;

    // From the GL spec for glViewport and glScissor:
@@ -2474,9 +2324,7 @@ void OpenGLDriver::setTextureData(GLTexture* t, uint32_t level,
    size_t const stride = p.stride ? p.stride : width;
    size_t const bpp = PBD::computeDataSize(p.format, p.type, 1, 1, 1);
    size_t const bpr = PBD::computeDataSize(p.format, p.type, stride, 1, p.alignment);
-    size_t const bpl = bpr * height; // TODO: PBD should have a "layer stride"
-    void const* const buffer = static_cast<char const*>(p.buffer)
-            + bpp* p.left + bpr * p.top + bpl * 0; // TODO: PBD should have a p.depth
+    void const* const buffer = static_cast<char const*>(p.buffer) + p.left * bpp + bpr * p.top;

    switch (t->target) {
        case SamplerType::SAMPLER_EXTERNAL:
@@ -2541,7 +2389,7 @@ void OpenGLDriver::setTextureData(GLTexture* t, uint32_t level,

 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
    if (!gl.isES2()) {
-        // Update the base/max LOD, so we don't access undefined LOD. this allows the app to
+        // update the base/max LOD, so we don't access undefined LOD. this allows the app to
        // specify levels as they become available.
        if (int8_t(level) < t->gl.baseLevel) {
            t->gl.baseLevel = int8_t(level);
@@ -2642,7 +2490,7 @@ void OpenGLDriver::setCompressedTextureData(GLTexture* t, uint32_t level,

 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
    if (!gl.isES2()) {
-        // Update the base/max LOD, so we don't access undefined LOD. this allows the app to
+        // update the base/max LOD, so we don't access undefined LOD. this allows the app to
        // specify levels as they become available.
        if (int8_t(level) < t->gl.baseLevel) {
            t->gl.baseLevel = int8_t(level);
@@ -2781,21 +2629,21 @@ void OpenGLDriver::replaceStream(GLTexture* texture, GLStream* newStream) noexce
 void OpenGLDriver::beginTimerQuery(Handle<HwTimerQuery> tqh) {
    DEBUG_MARKER()
    GLTimerQuery* tq = handle_cast<GLTimerQuery*>(tqh);
-    mContext.beginTimeElapsedQuery(tq);
+    mTimerQueryImpl->beginTimeElapsedQuery(tq);
 }

 void OpenGLDriver::endTimerQuery(Handle<HwTimerQuery> tqh) {
    DEBUG_MARKER()
    GLTimerQuery* tq = handle_cast<GLTimerQuery*>(tqh);
-    mContext.endTimeElapsedQuery(*this, tq);
+    mTimerQueryImpl->endTimeElapsedQuery(tq);
 }

-TimerQueryResult OpenGLDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint64_t* elapsedTime) {
+bool OpenGLDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint64_t* elapsedTime) {
    GLTimerQuery* tq = handle_cast<GLTimerQuery*>(tqh);
-    return TimerQueryFactoryInterface::getTimerQueryValue(tq, elapsedTime);
+    return OpenGLTimerQueryInterface::getTimerQueryValue(tq, elapsedTime);
 }

-void OpenGLDriver::compilePrograms(CompilerPriorityQueue,
+void OpenGLDriver::compilePrograms(CompilerPriorityQueue priority,
        CallbackHandler* handler, CallbackHandler::Callback callback, void* user) {
    if (callback) {
        getShaderCompilerService().notifyWhenAllProgramsAreReady(handler, callback, user);
@@ -2823,13 +2671,13 @@ void OpenGLDriver::beginRenderPass(Handle<HwRenderTarget> rth,
    const TargetBufferFlags clearFlags = params.flags.clear & rt->targets;
    TargetBufferFlags discardFlags = params.flags.discardStart & rt->targets;

-    GLuint const fbo = gl.bindFramebuffer(GL_FRAMEBUFFER, rt->gl.fbo);
+    gl.bindFramebuffer(GL_FRAMEBUFFER, rt->gl.fbo);
    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_FRAMEBUFFER)

    if (gl.ext.EXT_discard_framebuffer
            && !gl.bugs.disable_invalidate_framebuffer) {
        AttachmentArray attachments; // NOLINT
-        GLsizei const attachmentCount = getAttachments(attachments, discardFlags, !fbo);
+        GLsizei const attachmentCount = getAttachments(attachments, rt, discardFlags);
        if (attachmentCount) {
            gl.procs.invalidateFramebuffer(GL_FRAMEBUFFER, attachmentCount, attachments.data());
        }
@@ -2917,9 +2765,9 @@ void OpenGLDriver::endRenderPass(int) {
        }
        if (!gl.bugs.disable_invalidate_framebuffer) {
            // we wouldn't have to bind the framebuffer if we had glInvalidateNamedFramebuffer()
-            GLuint const fbo = gl.bindFramebuffer(GL_FRAMEBUFFER, rt->gl.fbo);
+            gl.bindFramebuffer(GL_FRAMEBUFFER, rt->gl.fbo);
            AttachmentArray attachments; // NOLINT
-            GLsizei const attachmentCount = getAttachments(attachments, effectiveDiscardFlags, !fbo);
+            GLsizei const attachmentCount = getAttachments(attachments, rt, effectiveDiscardFlags);
            if (attachmentCount) {
                gl.procs.invalidateFramebuffer(GL_FRAMEBUFFER, attachmentCount, attachments.data());
            }
@@ -2981,48 +2829,50 @@ void OpenGLDriver::resolvePass(ResolveAction action, GLRenderTarget const* rt,
 }

 GLsizei OpenGLDriver::getAttachments(AttachmentArray& attachments,
-        TargetBufferFlags buffers, bool isDefaultFramebuffer) noexcept {
+        GLRenderTarget const* rt, TargetBufferFlags buffers) noexcept {
+    assert_invariant(buffers <= rt->targets);
+
    GLsizei attachmentCount = 0;
    // the default framebuffer uses different constants!!!
-
+    const bool defaultFramebuffer = (rt->gl.fbo == 0);
    if (any(buffers & TargetBufferFlags::COLOR0)) {
-        attachments[attachmentCount++] = isDefaultFramebuffer ? GL_COLOR : GL_COLOR_ATTACHMENT0;
+        attachments[attachmentCount++] = defaultFramebuffer ? GL_COLOR : GL_COLOR_ATTACHMENT0;
    }
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
    if (any(buffers & TargetBufferFlags::COLOR1)) {
-        assert_invariant(!isDefaultFramebuffer);
+        assert_invariant(!defaultFramebuffer);
        attachments[attachmentCount++] = GL_COLOR_ATTACHMENT1;
    }
    if (any(buffers & TargetBufferFlags::COLOR2)) {
-        assert_invariant(!isDefaultFramebuffer);
+        assert_invariant(!defaultFramebuffer);
        attachments[attachmentCount++] = GL_COLOR_ATTACHMENT2;
    }
    if (any(buffers & TargetBufferFlags::COLOR3)) {
-        assert_invariant(!isDefaultFramebuffer);
+        assert_invariant(!defaultFramebuffer);
        attachments[attachmentCount++] = GL_COLOR_ATTACHMENT3;
    }
    if (any(buffers & TargetBufferFlags::COLOR4)) {
-        assert_invariant(!isDefaultFramebuffer);
+        assert_invariant(!defaultFramebuffer);
        attachments[attachmentCount++] = GL_COLOR_ATTACHMENT4;
    }
    if (any(buffers & TargetBufferFlags::COLOR5)) {
-        assert_invariant(!isDefaultFramebuffer);
+        assert_invariant(!defaultFramebuffer);
        attachments[attachmentCount++] = GL_COLOR_ATTACHMENT5;
    }
    if (any(buffers & TargetBufferFlags::COLOR6)) {
-        assert_invariant(!isDefaultFramebuffer);
+        assert_invariant(!defaultFramebuffer);
        attachments[attachmentCount++] = GL_COLOR_ATTACHMENT6;
    }
    if (any(buffers & TargetBufferFlags::COLOR7)) {
-        assert_invariant(!isDefaultFramebuffer);
+        assert_invariant(!defaultFramebuffer);
        attachments[attachmentCount++] = GL_COLOR_ATTACHMENT7;
    }
 #endif
    if (any(buffers & TargetBufferFlags::DEPTH)) {
-        attachments[attachmentCount++] = isDefaultFramebuffer ? GL_DEPTH : GL_DEPTH_ATTACHMENT;
+        attachments[attachmentCount++] = defaultFramebuffer ? GL_DEPTH : GL_DEPTH_ATTACHMENT;
    }
    if (any(buffers & TargetBufferFlags::STENCIL)) {
-        attachments[attachmentCount++] = isDefaultFramebuffer ? GL_STENCIL : GL_STENCIL_ATTACHMENT;
+        attachments[attachmentCount++] = defaultFramebuffer ? GL_STENCIL : GL_STENCIL_ATTACHMENT;
    }
    return attachmentCount;
 }
@@ -3742,8 +3592,8 @@ void OpenGLDriver::blit(
            mask, GL_NEAREST);
    CHECK_GL_ERROR(utils::slog.e)

-    gl.unbindFramebuffer(GL_DRAW_FRAMEBUFFER);
-    gl.unbindFramebuffer(GL_READ_FRAMEBUFFER);
+    gl.bindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
+    gl.bindFramebuffer(GL_READ_FRAMEBUFFER, 0);
    glDeleteFramebuffers(2, fbo);

    if (any(d->usage & TextureUsage::SAMPLEABLE)) {
@@ -3847,52 +3697,51 @@ void OpenGLDriver::updateTextureLodRange(GLTexture* texture, int8_t targetLevel)
 #endif
 }

-void OpenGLDriver::bindPipeline(PipelineState state) {
+void OpenGLDriver::draw(PipelineState state, Handle<HwRenderPrimitive> rph, uint32_t instanceCount) {
    DEBUG_MARKER()
    auto& gl = mContext;
-    setRasterState(state.rasterState);
-    setStencilState(state.stencilState);
-    gl.polygonOffset(state.polygonOffset.slope, state.polygonOffset.constant);
-    OpenGLProgram* const p = handle_cast<OpenGLProgram*>(state.program);
-    mValidProgram = useProgram(p);
-}

-void OpenGLDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
-    DEBUG_MARKER()
-    auto& gl = mContext;
+    OpenGLProgram* const p = handle_cast<OpenGLProgram*>(state.program);
+
+    bool const success = useProgram(p);
+    if (UTILS_UNLIKELY(!success)) {
+        // Avoid fatal (or cascading) errors that can occur during the draw call when the program
+        // is invalid. The shader compile error has already been dumped to the console at this
+        // point, so it's fine to simply return early.
+        return;
+    }

    GLRenderPrimitive* const rp = handle_cast<GLRenderPrimitive*>(rph);

    // Gracefully do nothing if the render primitive has not been set up.
    VertexBufferHandle vb = rp->gl.vertexBufferWithObjects;
    if (UTILS_UNLIKELY(!vb)) {
-        mBoundRenderPrimitive = nullptr;
        return;
    }

+    gl.bindVertexArray(&rp->gl);
+
    // If necessary, mutate the bindings in the VAO.
-    gl.bindVertexArray(&rp->gl);
    GLVertexBuffer const* const glvb = handle_cast<GLVertexBuffer*>(vb);
-    updateVertexArrayObject(rp, glvb);
-
-    mBoundRenderPrimitive = rp;
-}
-
-void OpenGLDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
-    GLRenderPrimitive const* const rp = mBoundRenderPrimitive;
-    if (UTILS_UNLIKELY(!rp || !mValidProgram)) {
-        return;
+    if (UTILS_UNLIKELY(rp->gl.vertexBufferVersion != glvb->bufferObjectsVersion)) {
+        updateVertexArrayObject(rp, glvb);
    }

+    setRasterState(state.rasterState);
+    setStencilState(state.stencilState);
+
+    gl.polygonOffset(state.polygonOffset.slope, state.polygonOffset.constant);
+
+    setScissor(state.scissor);
+
    if (UTILS_LIKELY(instanceCount <= 1)) {
-        glDrawElements(GLenum(rp->type), (GLsizei)indexCount, rp->gl.getIndicesType(),
-                reinterpret_cast<const void*>(indexOffset * rp->gl.indicesSize));
+        glDrawElements(GLenum(rp->type), (GLsizei)rp->count, rp->gl.getIndicesType(),
+                reinterpret_cast<const void*>(rp->offset));
    } else {
        assert_invariant(!mContext.isES2());
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
-        glDrawElementsInstanced(GLenum(rp->type), (GLsizei)indexCount,
-                rp->gl.getIndicesType(),
-                reinterpret_cast<const void*>(indexOffset * rp->gl.indicesSize),
+        glDrawElementsInstanced(GLenum(rp->type), (GLsizei)rp->count,
+                rp->gl.getIndicesType(), reinterpret_cast<const void*>(rp->offset),
                (GLsizei)instanceCount);
 #endif
    }
@@ -3904,21 +3753,6 @@ void OpenGLDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t ins
 #endif
 }

-void OpenGLDriver::scissor(Viewport scissor) {
-    setScissor(scissor);
-}
-
-void OpenGLDriver::draw(PipelineState state, Handle<HwRenderPrimitive> rph,
-        uint32_t const indexOffset, uint32_t const indexCount, uint32_t const instanceCount) {
-    DEBUG_MARKER()
-    GLRenderPrimitive* const rp = handle_cast<GLRenderPrimitive*>(rph);
-    state.primitiveType = rp->type;
-    state.vertexBufferInfo = rp->vbih;
-    bindPipeline(state);
-    bindRenderPrimitive(rph);
-    draw2(indexOffset, indexCount, instanceCount);
-}
-
 void OpenGLDriver::dispatchCompute(Handle<HwProgram> program, math::uint3 workGroupCount) {
    getShaderCompilerService().tick();

@@ -18,42 +18,27 @@
 #define TNT_FILAMENT_BACKEND_OPENGL_OPENGLDRIVER_H

 #include "DriverBase.h"
+#include "GLUtils.h"
 #include "OpenGLContext.h"
-#include "OpenGLTimerQuery.h"
 #include "ShaderCompilerService.h"

-#include <backend/platforms/OpenGLPlatform.h>
-
-#include <backend/AcquiredImage.h>
-#include <backend/DriverEnums.h>
-#include <backend/Handle.h>
-#include <backend/Platform.h>
-#include <backend/Program.h>
-#include <backend/TargetBufferInfo.h>
-
 #include "private/backend/Driver.h"
 #include "private/backend/HandleAllocator.h"

-#include <utils/FixedCapacityVector.h>
+#include <backend/platforms/OpenGLPlatform.h>
+
+#include <backend/AcquiredImage.h>
+#include <backend/Program.h>
+#include <backend/TargetBufferInfo.h>
+
 #include <utils/compiler.h>
-#include <utils/debug.h>
+#include <utils/Allocator.h>

 #include <math/vec4.h>

 #include <tsl/robin_map.h>

-#include <array>
-#include <condition_variable>
-#include <functional>
-#include <memory>
-#include <mutex>
-#include <tuple>
-#include <type_traits>
-#include <utility>
-#include <vector>
-
-#include <stddef.h>
-#include <stdint.h>
+#include <set>

 #ifndef FILAMENT_OPENGL_HANDLE_ARENA_SIZE_IN_MB
 #    define FILAMENT_OPENGL_HANDLE_ARENA_SIZE_IN_MB 4
@@ -66,17 +51,15 @@ class PixelBufferDescriptor;
 struct TargetBufferInfo;

 class OpenGLProgram;
-class TimerQueryFactoryInterface;
+class OpenGLTimerQueryInterface;

 class OpenGLDriver final : public DriverBase {
-    inline explicit OpenGLDriver(OpenGLPlatform* platform,
-            const Platform::DriverConfig& driverConfig) noexcept;
+    inline explicit OpenGLDriver(OpenGLPlatform* platform, const Platform::DriverConfig& driverConfig) noexcept;
    ~OpenGLDriver() noexcept final;
    Dispatcher getDispatcher() const noexcept final;

 public:
-    static Driver* create(OpenGLPlatform* platform, void* sharedGLContext,
-            const Platform::DriverConfig& driverConfig) noexcept;
+    static Driver* create(OpenGLPlatform* platform, void* sharedGLContext, const Platform::DriverConfig& driverConfig) noexcept;

    class DebugMarker {
        OpenGLDriver& driver;
@@ -111,22 +94,8 @@ public:
        uint16_t age = 0;
    };

-    struct GLVertexBufferInfo : public HwVertexBufferInfo {
-        GLVertexBufferInfo() noexcept = default;
-        GLVertexBufferInfo(uint8_t bufferCount, uint8_t attributeCount,
-                AttributeArray const& attributes)
-                : HwVertexBufferInfo(bufferCount, attributeCount),
-                  attributes(attributes) {
-        }
-        AttributeArray attributes;
-    };
-
    struct GLVertexBuffer : public HwVertexBuffer {
-        GLVertexBuffer() noexcept = default;
-        GLVertexBuffer(uint32_t vertexCount, Handle<HwVertexBufferInfo> vbih)
-                : HwVertexBuffer(vertexCount), vbih(vbih) {
-        }
-        Handle<HwVertexBufferInfo> vbih;
+        using HwVertexBuffer::HwVertexBuffer;
        struct {
            // 4 * MAX_VERTEX_ATTRIBUTE_COUNT bytes
            std::array<GLuint, MAX_VERTEX_ATTRIBUTE_COUNT> buffers{};
@@ -154,7 +123,6 @@ public:
    struct GLRenderPrimitive : public HwRenderPrimitive {
        using HwRenderPrimitive::HwRenderPrimitive;
        OpenGLContext::RenderPrimitive gl;
-        Handle<HwVertexBufferInfo> vbih;
    };

    struct GLTexture : public HwTexture {
@@ -179,7 +147,15 @@ public:
        OpenGLPlatform::ExternalTexture* externalTexture = nullptr;
    };

-    using GLTimerQuery = filament::backend::GLTimerQuery;
+    struct GLTimerQuery : public HwTimerQuery {
+        struct State {
+            struct {
+                GLuint query;
+            } gl;
+            std::atomic<int64_t> elapsed{};
+        };
+        std::shared_ptr<State> state;
+    };

    struct GLStream : public HwStream {
        using HwStream::HwStream;
@@ -233,8 +209,8 @@ private:
    OpenGLContext mContext;
    ShaderCompilerService mShaderCompilerService;

-    friend class TimerQueryFactory;
-    friend class TimerQueryNativeFactory;
+    friend class OpenGLTimerQueryFactory;
+    friend class TimerQueryNative;
    OpenGLContext& getContext() noexcept { return mContext; }

    ShaderCompilerService& getShaderCompilerService() noexcept {
@@ -268,13 +244,13 @@ private:
    HandleAllocatorGL mHandleAllocator;

    template<typename D, typename ... ARGS>
-    Handle<D> initHandle(ARGS&& ... args) {
+    Handle<D> initHandle(ARGS&& ... args) noexcept {
        return mHandleAllocator.allocateAndConstruct<D>(std::forward<ARGS>(args) ...);
    }

    template<typename D, typename B, typename ... ARGS>
    typename std::enable_if<std::is_base_of<B, D>::value, D>::type*
-    construct(Handle<B> const& handle, ARGS&& ... args) {
+    construct(Handle<B> const& handle, ARGS&& ... args) noexcept {
        return mHandleAllocator.destroyAndConstruct<D, B>(handle, std::forward<ARGS>(args) ...);
    }

@@ -288,7 +264,7 @@ private:
    typename std::enable_if_t<
            std::is_pointer_v<Dp> &&
            std::is_base_of_v<B, typename std::remove_pointer_t<Dp>>, Dp>
-    handle_cast(Handle<B>& handle) {
+    handle_cast(Handle<B>& handle) noexcept {
        return mHandleAllocator.handle_cast<Dp, B>(handle);
    }

@@ -296,7 +272,7 @@ private:
    inline typename std::enable_if_t<
            std::is_pointer_v<Dp> &&
            std::is_base_of_v<B, typename std::remove_pointer_t<Dp>>, Dp>
-    handle_cast(Handle<B> const& handle) {
+    handle_cast(Handle<B> const& handle) noexcept {
        return mHandleAllocator.handle_cast<Dp, B>(handle);
    }

@@ -314,7 +290,7 @@ private:
    void updateVertexArrayObject(GLRenderPrimitive* rp, GLVertexBuffer const* vb);

    void framebufferTexture(TargetBufferInfo const& binfo,
-            GLRenderTarget const* rt, GLenum attachment, uint8_t layerCount) noexcept;
+            GLRenderTarget const* rt, GLenum attachment) noexcept;

    void setRasterState(RasterState rs) noexcept;

@@ -335,8 +311,8 @@ private:
    void renderBufferStorage(GLuint rbo, GLenum internalformat, uint32_t width,
            uint32_t height, uint8_t samples) const noexcept;

-    void textureStorage(OpenGLDriver::GLTexture* t, uint32_t width, uint32_t height,
-            uint32_t depth, bool useProtectedMemory) noexcept;
+    void textureStorage(GLTexture* t,
+            uint32_t width, uint32_t height, uint32_t depth) noexcept;

    /* State tracking GL wrappers... */

@@ -369,8 +345,8 @@ private:
    }

    using AttachmentArray = std::array<GLenum, MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT + 2>;
-    static GLsizei getAttachments(AttachmentArray& attachments, TargetBufferFlags buffers,
-            bool isDefaultFramebuffer) noexcept;
+    static GLsizei getAttachments(AttachmentArray& attachments,
+            GLRenderTarget const* rt, TargetBufferFlags buffers) noexcept;

    // state required to represent the current render pass
    Handle<HwRenderTarget> mRenderPassTarget;
@@ -379,10 +355,6 @@ private:
    GLboolean mRenderPassDepthWrite{};
    GLboolean mRenderPassStencilWrite{};

-    GLRenderPrimitive const* mBoundRenderPrimitive = nullptr;
-    bool mValidProgram = false;
-
-
    void clearWithRasterPipe(TargetBufferFlags clearFlags,
            math::float4 const& linearColor, GLfloat depth, GLint stencil) noexcept;

@@ -425,8 +397,8 @@ private:
    void executeEveryNowAndThenOps() noexcept;
    std::vector<std::function<bool()>> mEveryNowAndThenOps;

-    const Platform::DriverConfig mDriverConfig;
-    Platform::DriverConfig const& getDriverConfig() const noexcept { return mDriverConfig; }
+    // timer query implementation
+    OpenGLTimerQueryInterface* mTimerQueryImpl = nullptr;

    // for ES2 sRGB support
    GLSwapChain* mCurrentDrawSwapChain = nullptr;
@@ -18,17 +18,6 @@

 #include "OpenGLDriverFactory.h"

-#include <backend/AcquiredImage.h>
-#include <backend/DriverEnums.h>
-#include <backend/Platform.h>
-
-#include <utils/compiler.h>
-
-#include <utils/Invocable.h>
-#include <stddef.h>
-#include <stdint.h>
-#include <stdlib.h>
-
 namespace filament::backend {

 Driver* OpenGLPlatform::createDefaultDriver(OpenGLPlatform* platform,
@@ -38,27 +27,14 @@ Driver* OpenGLPlatform::createDefaultDriver(OpenGLPlatform* platform,

 OpenGLPlatform::~OpenGLPlatform() noexcept = default;

-void OpenGLPlatform::makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain,
-        utils::Invocable<void()>, utils::Invocable<void(size_t)>) noexcept {
-    makeCurrent(getCurrentContextType(), drawSwapChain, readSwapChain);
-}
-
-bool OpenGLPlatform::isProtectedContextSupported() const noexcept {
-    return false;
-}
-
 bool OpenGLPlatform::isSRGBSwapChainSupported() const noexcept {
    return false;
 }

-uint32_t OpenGLPlatform::getDefaultFramebufferObject() noexcept {
+uint32_t OpenGLPlatform::createDefaultRenderTarget() noexcept {
    return 0;
 }

-OpenGLPlatform::ContextType OpenGLPlatform::getCurrentContextType() const noexcept {
-    return ContextType::UNPROTECTED;
-}
-
 void OpenGLPlatform::setPresentationTime(
        UTILS_UNUSED int64_t presentationTimeInNanosecond) noexcept {
 }
@@ -129,14 +105,10 @@ AcquiredImage OpenGLPlatform::transformAcquiredImage(AcquiredImage source) noexc
    return source;
 }

-TargetBufferFlags OpenGLPlatform::getPreservedFlags(UTILS_UNUSED SwapChain*) noexcept {
+TargetBufferFlags OpenGLPlatform::getPreservedFlags(UTILS_UNUSED SwapChain* swapChain) noexcept {
    return TargetBufferFlags::NONE;
 }

-bool OpenGLPlatform::isSwapChainProtected(UTILS_UNUSED SwapChain*) noexcept {
-    return false;
-}
-
 bool OpenGLPlatform::isExtraContextSupported() const noexcept {
    return false;
 }
@@ -16,25 +16,16 @@

 #include "OpenGLProgram.h"

-#include "GLUtils.h"
+#include "BlobCacheKey.h"
 #include "OpenGLDriver.h"
 #include "ShaderCompilerService.h"

-#include <backend/Program.h>
-
-#include <private/backend/BackendUtils.h>
-
 #include <utils/debug.h>
 #include <utils/compiler.h>
 #include <utils/Log.h>
 #include <utils/Systrace.h>

-#include <array>
-#include <string_view>
-#include <utility>
-#include <new>
-
-#include <stddef.h>
+#include <private/backend/BackendUtils.h>

 namespace filament::backend {

@@ -28,9 +28,6 @@
 #include <utils/compiler.h>
 #include <utils/FixedCapacityVector.h>

-#include <array>
-#include <limits>
-
 #include <stddef.h>
 #include <stdint.h>

@@ -16,132 +16,110 @@

 #include "OpenGLTimerQuery.h"

-#include "GLUtils.h"
-#include "OpenGLDriver.h"
-
-#include <backend/Platform.h>
 #include <backend/platforms/OpenGLPlatform.h>
-#include <backend/DriverEnums.h>

 #include <utils/compiler.h>
-#include <utils/debug.h>
 #include <utils/JobSystem.h>
 #include <utils/Log.h>
-#include <utils/Mutex.h>
 #include <utils/Systrace.h>
-
-#include <atomic>
-#include <memory>
-#include <mutex>
-#include <new>
-#include <utility>
-
-#include <stdint.h>
+#include <utils/debug.h>

 namespace filament::backend {

 using namespace backend;
 using namespace GLUtils;

-class OpenGLDriver;
-
 // ------------------------------------------------------------------------------------------------

-bool TimerQueryFactory::mGpuTimeSupported = false;
+bool OpenGLTimerQueryFactory::mGpuTimeSupported = false;

-TimerQueryFactoryInterface* TimerQueryFactory::init(
-        OpenGLPlatform& platform, OpenGLContext& context) noexcept {
-    (void)context;
+OpenGLTimerQueryInterface* OpenGLTimerQueryFactory::init(
+        OpenGLPlatform& platform, OpenGLDriver& driver) noexcept {
+    (void)driver;

-    TimerQueryFactoryInterface* impl = nullptr;
+    OpenGLTimerQueryInterface* impl;

 #if defined(BACKEND_OPENGL_VERSION_GL) || defined(GL_EXT_disjoint_timer_query)
+    auto& context = driver.getContext();
    if (context.ext.EXT_disjoint_timer_query) {
        // timer queries are available
        if (context.bugs.dont_use_timer_query && platform.canCreateFence()) {
            // however, they don't work well, revert to using fences if we can.
-            impl = new(std::nothrow) TimerQueryFenceFactory(platform);
+            impl = new(std::nothrow) OpenGLTimerQueryFence(platform);
        } else {
-            impl = new(std::nothrow) TimerQueryNativeFactory(context);
+            impl = new(std::nothrow) TimerQueryNative(driver);
        }
        mGpuTimeSupported = true;
    } else
 #endif
    if (platform.canCreateFence()) {
        // no timer queries, but we can use fences
-        impl = new(std::nothrow) TimerQueryFenceFactory(platform);
+        impl = new(std::nothrow) OpenGLTimerQueryFence(platform);
        mGpuTimeSupported = true;
    } else {
        // no queries, no fences -- that's a problem
-        impl = new(std::nothrow) TimerQueryFallbackFactory();
+        impl = new(std::nothrow) TimerQueryFallback();
        mGpuTimeSupported = false;
    }
-    assert_invariant(impl);
    return impl;
 }

 // ------------------------------------------------------------------------------------------------

-TimerQueryFactoryInterface::~TimerQueryFactoryInterface() = default;
+OpenGLTimerQueryInterface::~OpenGLTimerQueryInterface() = default;

 // This is a backend synchronous call
-TimerQueryResult TimerQueryFactoryInterface::getTimerQueryValue(
-        GLTimerQuery* tq, uint64_t* elapsedTime) noexcept {
+bool OpenGLTimerQueryInterface::getTimerQueryValue(GLTimerQuery* tq, uint64_t* elapsedTime) noexcept {
    if (UTILS_LIKELY(tq->state)) {
        int64_t const elapsed = tq->state->elapsed.load(std::memory_order_relaxed);
-        if (elapsed > 0) {
+        bool const available = elapsed > 0;
+        if (available) {
            *elapsedTime = elapsed;
-            return TimerQueryResult::AVAILABLE;
        }
-        return TimerQueryResult(elapsed);
+        return available;
    }
-    return TimerQueryResult::ERROR;
+    return false;
 }

 // ------------------------------------------------------------------------------------------------

 #if defined(BACKEND_OPENGL_VERSION_GL) || defined(GL_EXT_disjoint_timer_query)

-TimerQueryNativeFactory::TimerQueryNativeFactory(OpenGLContext& context)
-        : mContext(context) {
+TimerQueryNative::TimerQueryNative(OpenGLDriver& driver)
+        : mDriver(driver) {
 }

-TimerQueryNativeFactory::~TimerQueryNativeFactory() = default;
+TimerQueryNative::~TimerQueryNative() = default;

-void TimerQueryNativeFactory::createTimerQuery(GLTimerQuery* tq) {
-    assert_invariant(!tq->state);
-
-    tq->state = std::make_shared<GLTimerQuery::State>();
-    mContext.procs.genQueries(1u, &tq->state->gl.query);
+void TimerQueryNative::createTimerQuery(GLTimerQuery* tq) {
+    if (UTILS_UNLIKELY(!tq->state)) {
+        tq->state = std::make_shared<GLTimerQuery::State>();
+    }
+    mDriver.getContext().procs.genQueries(1u, &tq->state->gl.query);
    CHECK_GL_ERROR(utils::slog.e)
 }

-void TimerQueryNativeFactory::destroyTimerQuery(GLTimerQuery* tq) {
+void TimerQueryNative::destroyTimerQuery(GLTimerQuery* tq) {
    assert_invariant(tq->state);
-
-    mContext.procs.deleteQueries(1u, &tq->state->gl.query);
-    CHECK_GL_ERROR(utils::slog.e)
-
-    tq->state.reset();
-}
-
-void TimerQueryNativeFactory::beginTimeElapsedQuery(GLTimerQuery* tq) {
-    assert_invariant(tq->state);
-
-    tq->state->elapsed.store(int64_t(TimerQueryResult::NOT_READY), std::memory_order_relaxed);
-    mContext.procs.beginQuery(GL_TIME_ELAPSED, tq->state->gl.query);
+    mDriver.getContext().procs.deleteQueries(1u, &tq->state->gl.query);
    CHECK_GL_ERROR(utils::slog.e)
 }

-void TimerQueryNativeFactory::endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQuery* tq) {
+void TimerQueryNative::beginTimeElapsedQuery(GLTimerQuery* tq) {
    assert_invariant(tq->state);
+    tq->state->elapsed.store(0);
+    mDriver.getContext().procs.beginQuery(GL_TIME_ELAPSED, tq->state->gl.query);
+    CHECK_GL_ERROR(utils::slog.e)
+}

-    mContext.procs.endQuery(GL_TIME_ELAPSED);
+void TimerQueryNative::endTimeElapsedQuery(GLTimerQuery* tq) {
+    assert_invariant(tq->state);
+    mDriver.getContext().procs.endQuery(GL_TIME_ELAPSED);
    CHECK_GL_ERROR(utils::slog.e)

    std::weak_ptr<GLTimerQuery::State> const weak = tq->state;

-    driver.runEveryNowAndThen([&context = mContext, weak]() -> bool {
+    mDriver.runEveryNowAndThen([context = mDriver.getContext(), weak]() -> bool {
        auto state = weak.lock();
        if (state) {
            GLuint available = 0;
@@ -155,8 +133,6 @@ void TimerQueryNativeFactory::endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQ
            // we won't end-up here if we're on ES and don't have GL_EXT_disjoint_timer_query
            context.procs.getQueryObjectui64v(state->gl.query, GL_QUERY_RESULT, &elapsedTime);
            state->elapsed.store((int64_t)elapsedTime, std::memory_order_relaxed);
-        } else {
-            state->elapsed.store(int64_t(TimerQueryResult::ERROR), std::memory_order_relaxed);
        }
        return true;
    });
@@ -166,7 +142,7 @@ void TimerQueryNativeFactory::endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQ

 // ------------------------------------------------------------------------------------------------

-TimerQueryFenceFactory::TimerQueryFenceFactory(OpenGLPlatform& platform)
+OpenGLTimerQueryFence::OpenGLTimerQueryFence(OpenGLPlatform& platform)
        : mPlatform(platform) {
    mQueue.reserve(2);
    mThread = std::thread([this]() {
@@ -190,8 +166,7 @@ TimerQueryFenceFactory::TimerQueryFenceFactory(OpenGLPlatform& platform)
    });
 }

-TimerQueryFenceFactory::~TimerQueryFenceFactory() {
-    assert_invariant(mQueue.empty());
+OpenGLTimerQueryFence::~OpenGLTimerQueryFence() {
    if (mThread.joinable()) {
        std::unique_lock<utils::Mutex> lock(mLock);
        mExitRequested = true;
@@ -203,26 +178,27 @@ TimerQueryFenceFactory::~TimerQueryFenceFactory() {
    }
 }

-void TimerQueryFenceFactory::push(TimerQueryFenceFactory::Job&& job) {
+void OpenGLTimerQueryFence::enqueue(OpenGLTimerQueryFence::Job&& job) {
    std::unique_lock<utils::Mutex> const lock(mLock);
-    mQueue.push_back(std::move(job));
+    mQueue.push_back(std::forward<Job>(job));
    mCondition.notify_one();
 }

-void TimerQueryFenceFactory::createTimerQuery(GLTimerQuery* tq) {
-    assert_invariant(!tq->state);
-    tq->state = std::make_shared<GLTimerQuery::State>();
+void OpenGLTimerQueryFence::createTimerQuery(GLTimerQuery* tq) {
+    if (UTILS_UNLIKELY(!tq->state)) {
+        tq->state = std::make_shared<GLTimerQuery::State>();
+    }
 }

-void TimerQueryFenceFactory::destroyTimerQuery(GLTimerQuery* tq) {
+void OpenGLTimerQueryFence::destroyTimerQuery(GLTimerQuery* tq) {
    assert_invariant(tq->state);
-    tq->state.reset();
 }

-void TimerQueryFenceFactory::beginTimeElapsedQuery(GLTimerQuery* tq) {
+void OpenGLTimerQueryFence::beginTimeElapsedQuery(GLTimerQuery* tq) {
    assert_invariant(tq->state);
-    tq->state->elapsed.store(int64_t(TimerQueryResult::NOT_READY), std::memory_order_relaxed);
+    tq->state->elapsed.store(0);

+    Platform::Fence* fence = mPlatform.createFence();
    std::weak_ptr<GLTimerQuery::State> const weak = tq->state;

    // FIXME: this implementation of beginTimeElapsedQuery is usually wrong; it ends up
@@ -231,11 +207,12 @@ void TimerQueryFenceFactory::beginTimeElapsedQuery(GLTimerQuery* tq) {
    //    on a dummy target for instance, or somehow latch the begin time at the next renderpass
    //    start.

-    push([&platform = mPlatform, fence = mPlatform.createFence(), weak]() {
+    push([&platform = mPlatform, fence, weak]() {
        auto state = weak.lock();
        if (state) {
            platform.waitFence(fence, FENCE_WAIT_FOR_EVER);
-            state->then = clock::now().time_since_epoch().count();
+            int64_t const then = clock::now().time_since_epoch().count();
+            state->elapsed.store(-then, std::memory_order_relaxed);
            SYSTRACE_CONTEXT();
            SYSTRACE_ASYNC_BEGIN("OpenGLTimerQueryFence", intptr_t(state.get()));
        }
@@ -243,16 +220,19 @@ void TimerQueryFenceFactory::beginTimeElapsedQuery(GLTimerQuery* tq) {
    });
 }

-void TimerQueryFenceFactory::endTimeElapsedQuery(OpenGLDriver&, GLTimerQuery* tq) {
+void OpenGLTimerQueryFence::endTimeElapsedQuery(GLTimerQuery* tq) {
    assert_invariant(tq->state);
+    Platform::Fence* fence = mPlatform.createFence();
    std::weak_ptr<GLTimerQuery::State> const weak = tq->state;

-    push([&platform = mPlatform, fence = mPlatform.createFence(), weak]() {
+    push([&platform = mPlatform, fence, weak]() {
        auto state = weak.lock();
        if (state) {
            platform.waitFence(fence, FENCE_WAIT_FOR_EVER);
            int64_t const now = clock::now().time_since_epoch().count();
-            state->elapsed.store(now - state->then, std::memory_order_relaxed);
+            int64_t const then = state->elapsed.load(std::memory_order_relaxed);
+            assert_invariant(then < 0);
+            state->elapsed.store(now + then, std::memory_order_relaxed);
            SYSTRACE_CONTEXT();
            SYSTRACE_ASYNC_END("OpenGLTimerQueryFence", intptr_t(state.get()));
        }
@@ -262,32 +242,34 @@ void TimerQueryFenceFactory::endTimeElapsedQuery(OpenGLDriver&, GLTimerQuery* tq

 // ------------------------------------------------------------------------------------------------

-TimerQueryFallbackFactory::TimerQueryFallbackFactory() = default;
+TimerQueryFallback::TimerQueryFallback() = default;

-TimerQueryFallbackFactory::~TimerQueryFallbackFactory() = default;
+TimerQueryFallback::~TimerQueryFallback() = default;

-void TimerQueryFallbackFactory::createTimerQuery(GLTimerQuery* tq) {
-    assert_invariant(!tq->state);
-    tq->state = std::make_shared<GLTimerQuery::State>();
+void TimerQueryFallback::createTimerQuery(GLTimerQuery* tq) {
+    if (UTILS_UNLIKELY(!tq->state)) {
+        tq->state = std::make_shared<GLTimerQuery::State>();
+    }
 }

-void TimerQueryFallbackFactory::destroyTimerQuery(GLTimerQuery* tq) {
+void TimerQueryFallback::destroyTimerQuery(GLTimerQuery* tq) {
    assert_invariant(tq->state);
-    tq->state.reset();
 }

-void TimerQueryFallbackFactory::beginTimeElapsedQuery(GLTimerQuery* tq) {
+void TimerQueryFallback::beginTimeElapsedQuery(OpenGLTimerQueryInterface::GLTimerQuery* tq) {
    assert_invariant(tq->state);
    // this implementation measures the CPU time, but we have no h/w support
-    tq->state->then = clock::now().time_since_epoch().count();
-    tq->state->elapsed.store(int64_t(TimerQueryResult::NOT_READY), std::memory_order_relaxed);
+    int64_t const then = clock::now().time_since_epoch().count();
+    tq->state->elapsed.store(-then, std::memory_order_relaxed);
 }

-void TimerQueryFallbackFactory::endTimeElapsedQuery(OpenGLDriver&, GLTimerQuery* tq) {
+void TimerQueryFallback::endTimeElapsedQuery(OpenGLTimerQueryInterface::GLTimerQuery* tq) {
    assert_invariant(tq->state);
    // this implementation measures the CPU time, but we have no h/w support
    int64_t const now = clock::now().time_since_epoch().count();
-    tq->state->elapsed.store(now - tq->state->then, std::memory_order_relaxed);
+    int64_t const then = tq->state->elapsed.load(std::memory_order_relaxed);
+    assert_invariant(then < 0);
+    tq->state->elapsed.store(now + then, std::memory_order_relaxed);
 }

 } // namespace filament::backend
@@ -17,41 +17,18 @@
 #ifndef TNT_FILAMENT_BACKEND_OPENGL_TIMERQUERY_H
 #define TNT_FILAMENT_BACKEND_OPENGL_TIMERQUERY_H

-#include <backend/DriverEnums.h>
-
-#include "DriverBase.h"
+#include "OpenGLDriver.h"

 #include <utils/Condition.h>
 #include <utils/Mutex.h>

-#include "gl_headers.h"
-
-#include <atomic>
-#include <chrono>
-#include <functional>
-#include <memory>
 #include <thread>
 #include <vector>

-#include <stdint.h>
-
 namespace filament::backend {

 class OpenGLPlatform;
-class OpenGLContext;
-class OpenGLDriver;
-class TimerQueryFactoryInterface;
-
-struct GLTimerQuery : public HwTimerQuery {
-    struct State {
-        struct {
-            GLuint query;
-        } gl;
-        int64_t then{};
-        std::atomic<int64_t> elapsed{};
-    };
-    std::shared_ptr<State> state;
-};
+class OpenGLTimerQueryInterface;

 /*
 * We need two implementation of timer queries (only elapsed time), because
@@ -61,80 +38,83 @@ struct GLTimerQuery : public HwTimerQuery {
 * These classes implement the various strategies...
 */

-class TimerQueryFactory {
+
+class OpenGLTimerQueryFactory {
    static bool mGpuTimeSupported;
 public:
-    static TimerQueryFactoryInterface* init(
-            OpenGLPlatform& platform, OpenGLContext& context) noexcept;
+    static OpenGLTimerQueryInterface* init(
+            OpenGLPlatform& platform, OpenGLDriver& driver) noexcept;

    static bool isGpuTimeSupported() noexcept {
        return mGpuTimeSupported;
    }
 };

-class TimerQueryFactoryInterface {
+class OpenGLTimerQueryInterface {
 protected:
-    using GLTimerQuery = filament::backend::GLTimerQuery;
+    using GLTimerQuery = OpenGLDriver::GLTimerQuery;
    using clock = std::chrono::steady_clock;

 public:
-    virtual ~TimerQueryFactoryInterface();
+    virtual ~OpenGLTimerQueryInterface();
    virtual void createTimerQuery(GLTimerQuery* query) = 0;
    virtual void destroyTimerQuery(GLTimerQuery* query) = 0;
    virtual void beginTimeElapsedQuery(GLTimerQuery* query) = 0;
-    virtual void endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQuery* query) = 0;
+    virtual void endTimeElapsedQuery(GLTimerQuery* query) = 0;

-    static TimerQueryResult getTimerQueryValue(GLTimerQuery* tq, uint64_t* elapsedTime) noexcept;
+    static bool getTimerQueryValue(GLTimerQuery* tq, uint64_t* elapsedTime) noexcept;
 };

 #if defined(BACKEND_OPENGL_VERSION_GL) || defined(GL_EXT_disjoint_timer_query)

-class TimerQueryNativeFactory final : public TimerQueryFactoryInterface {
+class TimerQueryNative : public OpenGLTimerQueryInterface {
 public:
-    explicit TimerQueryNativeFactory(OpenGLContext& context);
-    ~TimerQueryNativeFactory() override;
+    explicit TimerQueryNative(OpenGLDriver& driver);
+    ~TimerQueryNative() override;
 private:
    void createTimerQuery(GLTimerQuery* query) override;
    void destroyTimerQuery(GLTimerQuery* query) override;
    void beginTimeElapsedQuery(GLTimerQuery* query) override;
-    void endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQuery* query) override;
-    OpenGLContext& mContext;
+    void endTimeElapsedQuery(GLTimerQuery* query) override;
+    OpenGLDriver& mDriver;
 };

 #endif

-class TimerQueryFenceFactory final : public TimerQueryFactoryInterface {
+class OpenGLTimerQueryFence : public OpenGLTimerQueryInterface {
 public:
-    explicit TimerQueryFenceFactory(OpenGLPlatform& platform);
-    ~TimerQueryFenceFactory() override;
+    explicit OpenGLTimerQueryFence(OpenGLPlatform& platform);
+    ~OpenGLTimerQueryFence() override;
 private:
    using Job = std::function<void()>;
-    using Container = std::vector<Job>;
-
    void createTimerQuery(GLTimerQuery* query) override;
    void destroyTimerQuery(GLTimerQuery* query) override;
    void beginTimeElapsedQuery(GLTimerQuery* tq) override;
-    void endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQuery* tq) override;
+    void endTimeElapsedQuery(GLTimerQuery* tq) override;

-    void push(Job&& job);
+    void enqueue(Job&& job);
+    template<typename CALLABLE, typename ... ARGS>
+    void push(CALLABLE&& func, ARGS&& ... args) {
+        enqueue(Job(std::bind(std::forward<CALLABLE>(func), std::forward<ARGS>(args)...)));
+    }

    OpenGLPlatform& mPlatform;
    std::thread mThread;
    mutable utils::Mutex mLock;
    mutable utils::Condition mCondition;
-    Container mQueue;
+    std::vector<Job> mQueue;
    bool mExitRequested = false;
 };

-class TimerQueryFallbackFactory final : public TimerQueryFactoryInterface {
+class TimerQueryFallback : public OpenGLTimerQueryInterface {
 public:
-    explicit TimerQueryFallbackFactory();
-    ~TimerQueryFallbackFactory() override;
+    explicit TimerQueryFallback();
+    ~TimerQueryFallback() override;
 private:
    void createTimerQuery(GLTimerQuery* query) override;
    void destroyTimerQuery(GLTimerQuery* query) override;
    void beginTimeElapsedQuery(GLTimerQuery* query) override;
-    void endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQuery* query) override;
+    void endTimeElapsedQuery(GLTimerQuery* query) override;
 };

 } // namespace filament::backend
@@ -30,7 +30,6 @@
 #include <utils/Log.h>
 #include <utils/Systrace.h>

-#include <cctype>
 #include <chrono>
 #include <string>
 #include <string_view>
@@ -152,12 +151,6 @@ bool ShaderCompilerService::isParallelShaderCompileSupported() const noexcept {
 }

 void ShaderCompilerService::init() noexcept {
-    if (UTILS_UNLIKELY(mDriver.getDriverConfig().disableParallelShaderCompile)) {
-        // user disabled parallel shader compile
-        mMode = Mode::SYNCHRONOUS;
-        return;
-    }
-
    // Here we decide which mode we'll be using. We always prefer our own thread-pool if
    // that mode is available because, we have no control on how the compilation queues are
    // handled if done by the driver (so at the very least we'd need to decode this per-driver).
@@ -262,12 +255,12 @@ ShaderCompilerService::program_token_t ShaderCompilerService::createProgram(
                [this, &gl, program = std::move(program), token]() mutable {
                    // compile the shaders
                    std::array<GLuint, Program::SHADER_TYPE_COUNT> shaders{};
+                    std::array<utils::CString, Program::SHADER_TYPE_COUNT> shaderSourceCode;
                    compileShaders(gl,
                            std::move(program.getShadersSource()),
                            program.getSpecializationConstants(),
-                            program.isMultiview(),
                            shaders,
-                            token->shaderSourceCode);
+                            shaderSourceCode);

                    // link the program
                    GLuint const glProgram = linkProgram(gl, shaders, token->attributes);
@@ -301,7 +294,6 @@ ShaderCompilerService::program_token_t ShaderCompilerService::createProgram(
        compileShaders(gl,
                std::move(program.getShadersSource()),
                program.getSpecializationConstants(),
-                program.isMultiview(),
                token->gl.shaders,
                token->shaderSourceCode);

@@ -504,7 +496,6 @@ GLuint ShaderCompilerService::initialize(program_token_t& token) noexcept {
 void ShaderCompilerService::compileShaders(OpenGLContext& context,
        Program::ShaderSource shadersSource,
        utils::FixedCapacityVector<Program::SpecializationConstant> const& specializationConstants,
-        bool multiview,
        std::array<GLuint, Program::SHADER_TYPE_COUNT>& outShaders,
        UTILS_UNUSED_IN_RELEASE std::array<CString, Program::SHADER_TYPE_COUNT>& outShaderSourceCode) noexcept {

@@ -518,16 +509,8 @@ void ShaderCompilerService::compileShaders(OpenGLContext& context,
    };

    std::string specializationConstantString;
-    int32_t numViews = 2;
    for (auto const& sc : specializationConstants) {
        appendSpecConstantString(specializationConstantString, sc);
-        if (sc.id == 8) {
-            // This constant must match
-            // ReservedSpecializationConstants::CONFIG_STEREO_EYE_COUNT
-            // which we can't use here because it's defined in EngineEnums.h.
-            // (we're breaking layering here, but it's for the good cause).
-            numViews = std::get<int32_t>(sc.value);
-        }
    }
    if (!specializationConstantString.empty()) {
        specializationConstantString += '\n';
@@ -556,23 +539,17 @@ void ShaderCompilerService::compileShaders(OpenGLContext& context,

        if (UTILS_LIKELY(!shadersSource[i].empty())) {
            Program::ShaderBlob& shader = shadersSource[i];
-            char* shader_src = reinterpret_cast<char*>(shader.data());
-            size_t shader_len = shader.size();

            // remove GOOGLE_cpp_style_line_directive
-            process_GOOGLE_cpp_style_line_directive(context, shader_src, shader_len);
-
-            // replace the value of layout(num_views = X) for multiview extension
-            if (multiview && stage == ShaderStage::VERTEX) {
-                process_OVR_multiview2(context, numViews, shader_src, shader_len);
-            }
+            std::string_view const source = process_GOOGLE_cpp_style_line_directive(context,
+                    reinterpret_cast<char*>(shader.data()), shader.size());

            // add support for ARB_shading_language_packing if needed
            auto const packingFunctions = process_ARB_shading_language_packing(context);

            // split shader source, so we can insert the specialization constants and the packing
            // functions
-            auto const [prolog, body] = splitShaderSource({ shader_src, shader_len });
+            auto const [prolog, body] = splitShaderSource(source);

            const std::array<const char*, 4> sources = {
                    prolog.data(),
@@ -595,7 +572,7 @@ void ShaderCompilerService::compileShaders(OpenGLContext& context,
 #ifndef NDEBUG
            // for debugging we return the original shader source (without the modifications we
            // made here), otherwise the line numbers wouldn't match.
-            outShaderSourceCode[i] = { shader_src, shader_len };
+            outShaderSourceCode[i] = { source.data(), source.length() };
 #endif

            outShaders[i] = shaderId;
@@ -604,59 +581,15 @@ void ShaderCompilerService::compileShaders(OpenGLContext& context,
 }

 // If usages of the Google-style line directive are present, remove them, as some
-// drivers don't allow the quotation marks. This source modification happens in-place.
-void ShaderCompilerService::process_GOOGLE_cpp_style_line_directive(OpenGLContext& context,
+// drivers don't allow the quotation marks. This happens in-place.
+std::string_view ShaderCompilerService::process_GOOGLE_cpp_style_line_directive(OpenGLContext& context,
        char* source, size_t len) noexcept {
    if (!context.ext.GOOGLE_cpp_style_line_directive) {
        if (UTILS_UNLIKELY(requestsGoogleLineDirectivesExtension({ source, len }))) {
            removeGoogleLineDirectives(source, len); // length is unaffected
        }
    }
-}
-
-// Look up the `source` to replace the number of eyes for multiview with the given number. This is
-// necessary for OpenGL because OpenGL relies on the number specified in shader files to determine
-// the number of views, which is assumed as a single digit, for multiview.
-// This source modification happens in-place.
-void ShaderCompilerService::process_OVR_multiview2(OpenGLContext& context,
-        int32_t eyeCount, char* source, size_t len) noexcept {
-    // We don't use regular expression in favor of performance.
-    if (context.ext.OVR_multiview2) {
-        const std::string_view shader{ source, len };
-        const std::string_view layout = "layout";
-        const std::string_view num_views = "num_views";
-        size_t found = 0;
-        while (true) {
-            found = shader.find(layout, found);
-            if (found == std::string_view::npos) {
-                break;
-            }
-            found = shader.find_first_not_of(' ', found + layout.size());
-            if (found == std::string_view::npos || shader[found] != '(') {
-                continue;
-            }
-            found = shader.find_first_not_of(' ', found + 1);
-            if (found == std::string_view::npos) {
-                continue;
-            }
-            if (shader.compare(found, num_views.size(), num_views) != 0) {
-                continue;
-            }
-            found = shader.find_first_not_of(' ', found + num_views.size());
-            if (found == std::string_view::npos || shader[found] != '=') {
-                continue;
-            }
-            found = shader.find_first_not_of(' ', found + 1);
-            if (found == std::string_view::npos) {
-                continue;
-            }
-            // We assume the value should be one-digit number.
-            assert_invariant(eyeCount < 10);
-            assert_invariant(!::isdigit(source[found + 1]));
-            source[found] = '0' + eyeCount;
-            break;
-        }
-    }
+    return { source, len };
 }

 // Tragically, OpenGL 4.1 doesn't support unpackHalf2x16 (appeared in 4.2) and
--- a/Show More
+++ b/Show More