wip: color space stuff

2023-11-13 14:32:31 -08:00
421 changed files with 4105 additions and 7215 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:
@@ -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
@@ -41,10 +41,6 @@ option(FILAMENT_LINUX_IS_MOBILE "Treat Linux as Mobile" OFF)

 option(FILAMENT_ENABLE_ASAN_UBSAN "Enable Address and Undefined Behavior Sanitizers" OFF)

-option(FILAMENT_ENABLE_TSAN "Enable Thread Sanitizer" OFF)
-
-option(FILAMENT_ENABLE_FEATURE_LEVEL_0 "Enable Feature Level 0" ON)
-
 set(FILAMENT_NDK_VERSION "" CACHE STRING
    "Android NDK version or version prefix to be used when building for Android."
 )
@@ -69,9 +65,6 @@ set(FILAMENT_METAL_HANDLE_ARENA_SIZE_IN_MB "8" CACHE STRING
    "Size of the Metal handle arena, default 8."
 )

-# Enable exceptions by default in spirv-cross.
-set(SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS OFF)
-
 # ==================================================================================================
 # CMake policies
 # ==================================================================================================
@@ -232,21 +225,6 @@ if (WIN32)
            # we don't need them on CI.
            string(REPLACE "/INCREMENTAL" "/INCREMENTAL:NO" ${LinkerFlag} ${${LinkerFlag}})
        endforeach()
-
-        # We turn off compile-time optimizations for CI, as options that speed up the compile-time
-        # (e.g. /MP) might increase memory usage, leading to instabilities on limited CI machines.
-        option(FILAMENT_SHORTEN_MSVC_COMPILATION "Shorten compile-time in Visual Studio" OFF)
-    else()
-        option(FILAMENT_SHORTEN_MSVC_COMPILATION "Shorten compile-time in Visual Studio" ON)
-    endif()
-
-    if (MSVC)
-        if (FILAMENT_SHORTEN_MSVC_COMPILATION)
-            # enable multi-processor compilation
-            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MP")
-            # disable run-time STL checks to improve tools (e.g. matc) performance
-            set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /D_ITERATOR_DEBUG_LEVEL=0")
-        endif()
    endif()
 endif()

@@ -359,7 +337,6 @@ endif()

 if (CYGWIN)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-exceptions -fno-rtti")
-    set(SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS ON)
 endif()

 if (MSVC)
@@ -396,7 +373,6 @@ endif()
 # saved by -fno-exception and 10 KiB saved by -fno-rtti).
 if (ANDROID OR IOS OR WEBGL)
    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fno-exceptions -fno-rtti")
-    set(SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS ON)

    if (ANDROID OR WEBGL)
        # Omitting unwind info prevents the generation of readable stack traces in crash reports on iOS
@@ -408,7 +384,6 @@ endif()
 # std::visit, which is not supported on iOS 11.0 when exceptions are enabled.
 if (IOS)
    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fno-exceptions")
-    set(SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS ON)
 endif()

 # With WebGL, we disable RTTI even for debug builds because we pass emscripten::val back and forth
@@ -428,13 +403,7 @@ endif()
 if (FILAMENT_ENABLE_ASAN_UBSAN)
   set(EXTRA_SANITIZE_OPTIONS "-fsanitize=address -fsanitize=undefined")
 endif()
-if (FILAMENT_ENABLE_TSAN)
-   set(EXTRA_SANITIZE_OPTIONS "-fsanitize=thread")
-endif()
-if (ANDROID)
-    # keep STL debug infos (mimics what the NDK does)
-    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fno-limit-debug-info")
-endif()
+
 if (NOT MSVC AND NOT WEBGL)
    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fstack-protector")
 endif()
@@ -554,11 +523,6 @@ if (FILAMENT_SUPPORTS_METAL)
    set(MATC_API_FLAGS ${MATC_API_FLAGS} -a metal)
 endif()

-# Disable ESSL 1.0 code generation.
-if (NOT FILAMENT_ENABLE_FEATURE_LEVEL_0)
-    set(MATC_API_FLAGS ${MATC_API_FLAGS} -1)
-endif()
-
 # Enable debug info (preserves names in SPIR-V)
 if (FILAMENT_ENABLE_MATDBG)
    set(MATC_OPT_FLAGS ${MATC_OPT_FLAGS} -d)
@@ -634,9 +598,9 @@ 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:${DEPENDENCY}>")
            endif()
        endif()
@@ -706,7 +670,7 @@ function(get_resgen_vars ARCHIVE_DIR ARCHIVE_NAME)
        set(RESGEN_OUTPUTS "${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()

@@ -721,6 +685,7 @@ 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)
@@ -748,9 +713,6 @@ 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(${LIBRARIES}/geometry)
-
 if (FILAMENT_BUILD_FILAMAT OR IS_HOST_PLATFORM)
    # spirv-tools must come before filamat, as filamat relies on the presence of the
    # spirv-tools_SOURCE_DIR variable.
@@ -768,8 +730,6 @@ endif()
 if (FILAMENT_SUPPORTS_VULKAN)
    add_subdirectory(${LIBRARIES}/bluevk)
    add_subdirectory(${EXTERNAL}/vkmemalloc/tnt)
-    set(SPIRV_HEADERS_SKIP_EXAMPLES ON)
-    add_subdirectory(${EXTERNAL}/spirv-headers)
 endif()

 set(FILAMENT_SAMPLES_BINARY_DIR ${PROJECT_BINARY_DIR}/samples)
@@ -7,3 +7,12 @@ for next branch cut* header.
 appropriate header in [RELEASE_NOTES.md](./RELEASE_NOTES.md).

 ## Release notes for next branch cut
+
+- engine: Allow instantiating Engine at a given feature level via `Engine::Builder::featureLevel`
+- matc: Enable `GL_OES_standard_derivatives` extension in ESSL 1.0 shaders
+- matc: Fix code generation of double sided and masked materials in ESSL 1.0 shaders
+- filagui: Add support for feature level 0
+- matc: Add support for post-process materials in feature level 0
+- engine: Add `Material::getFeatureLevel()`
+- engine: Add missing `Material::getReflectionMode()` method in Java
+- engine: Support basic usage of post-processing materials on feature level 0
@@ -31,7 +31,7 @@ repositories {
 }

 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.50.3'
+    implementation 'com.google.android.filament:filament-android:1.45.0'
 }
 ```

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

 ```shell
-pod 'Filament', '~> 1.50.3'
+pod 'Filament', '~> 1.45.0'
 ```

 ### Snapshots
@@ -7,59 +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).

-## v1.50.3
-
-
-## 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**]
-
-## v1.49.3
-
- matc: Generate stereo variants for FL0 materials [⚠️ **Recompile materials**]
-
-## v1.49.2
-
-
-## v1.49.1
-
-
-## v1.49.0
-
- matc: Fix ESSL 1.0 codegen when using external samplers [⚠️ **Recompile materials**]
-
-## v1.48.0
-
- matc: New option `-1` to disable generation of ESSL 1.0 code in Feature Level 0 materials
- matc: Support optimizations for ESSL 1.0 code [⚠️ **Recompile materials**]
-
-## v1.47.0
-
- engine: Support up to 4 side-by-side stereoscopic eyes, configurable at Engine creation time. See
-  `Engine::Config::stereoscopicEyeCount`. [⚠️ **Recompile Materials**]
-
-## v1.46.0
-
- engine: Allow instantiating Engine at a given feature level via `Engine::Builder::featureLevel`
- matc: Enable `GL_OES_standard_derivatives` extension in ESSL 1.0 shaders
- matc: Fix code generation of double sided and masked materials in ESSL 1.0 shaders
- filagui: Add support for feature level 0
- matc: Add support for post-process materials in feature level 0
- engine: Add `Material::getFeatureLevel()`
- engine: Add missing `Material::getReflectionMode()` method in Java
- engine: Support basic usage of post-processing materials on feature level 0
- engine: Fix critical GLES 2.0 bugs
- engine: Add `FILAMENT_ENABLE_FEATURE_LEVEL_0` build-time option optionally allow building Filament
-  without FL0 support.
-
 ## v1.45.1

 - engine: Added parameter for configuring JobSystem thread count
@@ -81,14 +81,14 @@ buildscript {
    ext.versions = [
        'jdk': 17,
        'minSdk': 19,
-        'targetSdk': 34,
-        'compileSdk': 34,
-        'kotlin': '1.9.21',
-        'kotlin_coroutines': '1.7.3',
+        'targetSdk': 33,
+        'compileSdk': 33,
+        'kotlin': '1.9.0',
+        'kotlin_coroutines': '1.7.2',
        'buildTools': '34.0.0',
-        'ndk': '26.1.10909125',
-        'androidx_core': '1.12.0',
-        'androidx_annotations': '1.7.0'
+        'ndk': '25.1.8937393',
+        'androidx_core': '1.10.1',
+        'androidx_annotations': '1.6.0'
    ]

    ext.deps = [
@@ -104,7 +104,7 @@ buildscript {
    ]

    dependencies {
-        classpath 'com.android.tools.build:gradle:8.2.0'
+        classpath 'com.android.tools.build:gradle:8.1.0'
        classpath "org.jetbrains.kotlin:kotlin-gradle-plugin:${versions.kotlin}"
    }

@@ -384,13 +384,6 @@ Java_com_google_android_filament_Engine_nFlushAndWait(JNIEnv*, jclass,
    engine->flushAndWait();
 }

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

 extern "C" JNIEXPORT jlong JNICALL
@@ -435,13 +428,6 @@ Java_com_google_android_filament_Engine_nIsAutomaticInstancingEnabled(JNIEnv*, j
    return (jboolean)engine->isAutomaticInstancingEnabled();
 }

-extern "C" JNIEXPORT jlong JNICALL
-Java_com_google_android_filament_Engine_nGetMaxStereoscopicEyes(JNIEnv*, jclass, jlong nativeEngine) {
-    Engine* engine = (Engine*) nativeEngine;
-    return (jlong) engine->getMaxStereoscopicEyes();
-}
-
-
 extern "C" JNIEXPORT jint JNICALL
 Java_com_google_android_filament_Engine_nGetSupportedFeatureLevel(JNIEnv *, jclass,
        jlong nativeEngine) {
@@ -484,8 +470,7 @@ 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 stereoscopicEyeCount,
-        jlong resourceAllocatorCacheSizeMB, jlong resourceAllocatorCacheMaxAge) {
+        jlong jobSystemThreadCount) {
    Engine::Builder* builder = (Engine::Builder*) nativeBuilder;
    Engine::Config config = {
            .commandBufferSizeMB = (uint32_t) commandBufferSizeMB,
@@ -494,9 +479,6 @@ extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBu
            .minCommandBufferSizeMB = (uint32_t) minCommandBufferSizeMB,
            .perFrameCommandsSizeMB = (uint32_t) perFrameCommandsSizeMB,
            .jobSystemThreadCount = (uint32_t) jobSystemThreadCount,
-            .stereoscopicEyeCount = (uint8_t) stereoscopicEyeCount,
-            .resourceAllocatorCacheSizeMB = (uint32_t) resourceAllocatorCacheSizeMB,
-            .resourceAllocatorCacheMaxAge = (uint8_t) resourceAllocatorCacheMaxAge,
    };
    builder->config(&config);
 }
@@ -38,7 +38,7 @@ Java_com_google_android_filament_EntityManager_nCreateArray(JNIEnv* env, jclass,
    // (which it is), but still.
    em->create((size_t) n, reinterpret_cast<Entity *>(entities));

-    env->ReleaseIntArrayElements(entities_, entities, JNI_ABORT);
+    env->ReleaseIntArrayElements(entities_, entities, 0);
 }

 extern "C" JNIEXPORT jint JNICALL
@@ -19,7 +19,6 @@
 #include <filament/Material.h>

 #include "common/NioUtils.h"
-#include "common/CallbackUtils.h"

 using namespace filament;

@@ -272,17 +271,3 @@ Java_com_google_android_filament_Material_nHasParameter(JNIEnv* env, jclass,
    env->ReleaseStringUTFChars(name_, name);
    return (jboolean) hasParameter;
 }
-
-extern "C"
-JNIEXPORT void JNICALL
-Java_com_google_android_filament_Material_nCompile(JNIEnv *env, jclass clazz,
-        jlong nativeMaterial, jint priority, jint variants, jobject handler, jobject runnable) {
-    Material* material = (Material*) nativeMaterial;
-    JniCallback* jniCallback = JniCallback::make(env, handler, runnable);
-    material->compile(
-            (Material::CompilerPriorityQueue) priority,
-            (UserVariantFilterBit) variants,
-            jniCallback->getHandler(), [jniCallback](Material*){
-                JniCallback::postToJavaAndDestroy(jniCallback);
-            });
-}
@@ -205,7 +205,7 @@ Java_com_google_android_filament_MaterialInstance_nSetIntParameterArray(JNIEnv *
            break;
    }

-    env->ReleaseIntArrayElements(v_, v, JNI_ABORT);
+    env->ReleaseIntArrayElements(v_, v, 0);

    env->ReleaseStringUTFChars(name_, name);
 }
@@ -98,22 +98,3 @@ Java_com_google_android_filament_Scene_nHasEntity(JNIEnv *env, jclass type, jlon
    Entity entity = Entity::import(entityId);
    return (jboolean) scene->hasEntity(entity);
 }
-
-extern "C"
-JNIEXPORT jboolean JNICALL
-Java_com_google_android_filament_Scene_nGetEntities(JNIEnv *env, jclass ,
-        jlong nativeScene, jintArray outArray, jint length) {
-    Scene const* const scene = (Scene*) nativeScene;
-    if (length < scene->getEntityCount()) {
-        // should not happen because we already checked on the java side
-        return JNI_FALSE;
-    }
-    jint *out = (jint *) env->GetIntArrayElements(outArray, nullptr);
-    scene->forEach([out, length, i = 0](Entity entity)mutable {
-        if (i < length) { // this is just paranoia here
-            out[i++] = (jint) entity.getId();
-        }
-    });
-    env->ReleaseIntArrayElements(outArray, (jint*) out, 0);
-    return JNI_TRUE;
-}
@@ -480,17 +480,6 @@ Java_com_google_android_filament_View_nIsStencilBufferEnabled(JNIEnv *, jclass,
    return view->isStencilBufferEnabled();
 }

-extern "C"
-JNIEXPORT void JNICALL
-Java_com_google_android_filament_View_nSetStereoscopicOptions(JNIEnv *, jclass, jlong nativeView,
-        jboolean enabled) {
-    View* view = (View*) nativeView;
-    View::StereoscopicOptions options {
-        .enabled = (bool) enabled
-    };
-    view->setStereoscopicOptions(options);
-}
-
 extern "C"
 JNIEXPORT void JNICALL
 Java_com_google_android_filament_View_nSetGuardBandOptions(JNIEnv *, jclass,
@@ -111,8 +111,6 @@ public class Engine {

    private long mNativeObject;

-    private Config mConfig;
-
    @NonNull private final TransformManager mTransformManager;
    @NonNull private final LightManager mLightManager;
    @NonNull private final RenderableManager mRenderableManager;
@@ -152,10 +150,8 @@ public class Engine {
        FEATURE_LEVEL_0,
        /** OpenGL ES 3.0 features (default) */
        FEATURE_LEVEL_1,
-        /** OpenGL ES 3.1 features + 16 textures units + cubemap arrays */
-        FEATURE_LEVEL_2,
        /** OpenGL ES 3.1 features + 31 textures units + cubemap arrays */
-        FEATURE_LEVEL_3,
+        FEATURE_LEVEL_2
    };

    /**
@@ -165,7 +161,6 @@ public class Engine {
        @SuppressWarnings({"FieldCanBeLocal", "UnusedDeclaration"})
        private final BuilderFinalizer mFinalizer;
        private final long mNativeBuilder;
-        private Config mConfig;

        public Builder() {
            mNativeBuilder = nCreateBuilder();
@@ -207,12 +202,10 @@ public class Engine {
         * @return A reference to this Builder for chaining calls.
         */
        public Builder config(Config config) {
-            mConfig = config;
            nSetBuilderConfig(mNativeBuilder, config.commandBufferSizeMB,
                    config.perRenderPassArenaSizeMB, config.driverHandleArenaSizeMB,
                    config.minCommandBufferSizeMB, config.perFrameCommandsSizeMB,
-                    config.jobSystemThreadCount, config.stereoscopicEyeCount,
-                    config.resourceAllocatorCacheSizeMB, config.resourceAllocatorCacheMaxAge);
+                    config.jobSystemThreadCount);
            return this;
        }

@@ -240,7 +233,7 @@ public class Engine {
        public Engine build() {
            long nativeEngine = nBuilderBuild(mNativeBuilder);
            if (nativeEngine == 0) throw new IllegalStateException("Couldn't create Engine");
-            return new Engine(nativeEngine, mConfig);
+            return new Engine(nativeEngine);
        }

        private static class BuilderFinalizer {
@@ -348,36 +341,14 @@ public class Engine {
         * the number of threads to use.
         */
        public long jobSystemThreadCount = 0;
-
-        /**
-         * The number of eyes to render when stereoscopic rendering is enabled. Supported values are
-         * between 1 and Engine#getMaxStereoscopicEyes() (inclusive).
-         *
-         * @see View#setStereoscopicOptions
-         * @see Engine#getMaxStereoscopicEyes
-         */
-        public long stereoscopicEyeCount = 2;
-
-        /*
-         * Size in MiB of the frame graph texture cache. This should be adjusted based on the
-         * size of used render targets (typically the screen).
-         */
-        public long resourceAllocatorCacheSizeMB = 64;
-
-        /*
-         * This value determines for how many frames are texture entries kept in the cache.
-         * The default value of 30 corresponds to about half a second at 60 fps.
-         */
-        public long resourceAllocatorCacheMaxAge = 30;
    }

-    private Engine(long nativeEngine, Config config) {
+    private Engine(long nativeEngine) {
        mNativeObject = nativeEngine;
        mTransformManager = new TransformManager(nGetTransformManager(nativeEngine));
        mLightManager = new LightManager(nGetLightManager(nativeEngine));
        mRenderableManager = new RenderableManager(nGetRenderableManager(nativeEngine));
        mEntityManager = new EntityManager(nGetEntityManager(nativeEngine));
-        mConfig = config;
    }

    /**
@@ -570,37 +541,6 @@ public class Engine {
        return nIsAutomaticInstancingEnabled(getNativeObject());
    }

-    /**
-     * Retrieves the configuration settings of this {@link Engine}.
-     *
-     * This method returns the configuration object that was supplied to the Engine's {@link
-     * Builder#config} method during the creation of this Engine. If the {@link Builder::config}
-     * method was not explicitly called (or called with null), this method returns the default
-     * configuration settings.
-     *
-     * @return a {@link Config} object with this Engine's configuration
-     * @see Builder#config
-     */
-    @NonNull
-    public Config getConfig() {
-        if (mConfig == null) {
-            mConfig = new Config();
-        }
-        return mConfig;
-    }
-
-    /**
-     * Returns the maximum number of stereoscopic eyes supported by Filament. The actual number of
-     * eyes rendered is set at Engine creation time with the {@link
-     * Engine#Config#stereoscopicEyeCount} setting.
-     *
-     * @return the max number of stereoscopic eyes supported
-     * @see Engine#Config#stereoscopicEyeCount
-     */
-    public long getMaxStereoscopicEyes() {
-        return nGetMaxStereoscopicEyes(getNativeObject());
-    }
-

    // SwapChain

@@ -1137,18 +1077,6 @@ public class Engine {
        nFlushAndWait(getNativeObject());
    }

-    /**
-     * Kicks the hardware thread (e.g. the OpenGL, Vulkan or Metal thread) but does not wait
-     * for commands to be either executed or the hardware finished.
-     *
-     * <p>This is typically used after creating a lot of objects to start draining the command
-     * queue which has a limited size.</p>
-     */
-    public void flush() {
-        nFlush(getNativeObject());
-    }
-
-
    @UsedByReflection("TextureHelper.java")
    public long getNativeObject() {
        if (mNativeObject == 0) {
@@ -1221,7 +1149,6 @@ public class Engine {
    private static native boolean nIsValidSwapChain(long nativeEngine, long nativeSwapChain);
    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 long nGetTransformManager(long nativeEngine);
    private static native long nGetLightManager(long nativeEngine);
    private static native long nGetRenderableManager(long nativeEngine);
@@ -1229,7 +1156,6 @@ public class Engine {
    private static native long nGetEntityManager(long nativeEngine);
    private static native void nSetAutomaticInstancingEnabled(long nativeEngine, boolean enable);
    private static native boolean nIsAutomaticInstancingEnabled(long nativeEngine);
-    private static native long nGetMaxStereoscopicEyes(long nativeEngine);
    private static native int nGetSupportedFeatureLevel(long nativeEngine);
    private static native int nSetActiveFeatureLevel(long nativeEngine, int ordinal);
    private static native int nGetActiveFeatureLevel(long nativeEngine);
@@ -1239,9 +1165,7 @@ public class Engine {
    private static native void nSetBuilderBackend(long nativeBuilder, long backend);
    private static native void nSetBuilderConfig(long nativeBuilder, long commandBufferSizeMB,
            long perRenderPassArenaSizeMB, long driverHandleArenaSizeMB,
-            long minCommandBufferSizeMB, long perFrameCommandsSizeMB, long jobSystemThreadCount,
-            long stereoscopicEyeCount,
-            long resourceAllocatorCacheSizeMB, long resourceAllocatorCacheMaxAge);
+            long minCommandBufferSizeMB, long perFrameCommandsSizeMB, long jobSystemThreadCount);
    private static native void nSetBuilderFeatureLevel(long nativeBuilder, int ordinal);
    private static native void nSetBuilderSharedContext(long nativeBuilder, long sharedContext);
    private static native long nBuilderBuild(long nativeBuilder);
@@ -18,7 +18,6 @@ package com.google.android.filament;

 import androidx.annotation.IntRange;
 import androidx.annotation.NonNull;
-import androidx.annotation.Nullable;
 import androidx.annotation.Size;

 import com.google.android.filament.proguard.UsedByNative;
@@ -239,31 +238,6 @@ public class Material {
        FRONT_AND_BACK
    }

-    public enum CompilerPriorityQueue {
-        HIGH,
-        LOW
-    }
-
-    public static class UserVariantFilterBit {
-        /** Directional lighting */
-        public static int DIRECTIONAL_LIGHTING = 0x01;
-        /** Dynamic lighting */
-        public static int DYNAMIC_LIGHTING = 0x02;
-        /** Shadow receiver */
-        public static int SHADOW_RECEIVER = 0x04;
-        /** Skinning */
-        public static int SKINNING = 0x08;
-        /** Fog */
-        public static int FOG = 0x10;
-        /** Variance shadow maps */
-        public static int VSM = 0x20;
-        /** Screen-space reflections */
-        public static int SSR = 0x40;
-        /** Instanced stereo rendering */
-        public static int STE = 0x80;
-        public static int ALL = 0xFF;
-    }
-
    @UsedByNative("Material.cpp")
    public static class Parameter {
        private static final Type[] sTypeValues = Type.values();
@@ -378,55 +352,6 @@ public class Material {
        }
    }

-
-    /**
-     * Asynchronously ensures that a subset of this Material's variants are compiled. After issuing
-     * several compile() calls in a row, it is recommended to call {@link Engine#flush}
-     * such that the backend can start the compilation work as soon as possible.
-     * The provided callback is guaranteed to be called on the main thread after all specified
-     * variants of the material are compiled. This can take hundreds of milliseconds.
-     *<p>
-     * If all the material's variants are already compiled, the callback will be scheduled as
-     * soon as possible, but this might take a few dozen millisecond, corresponding to how
-     * many previous frames are enqueued in the backend. This also varies by backend. Therefore,
-     * it is recommended to only call this method once per material shortly after creation.
-     *</p>
-     *<p>
-     * If the same variant is scheduled for compilation multiple times, the first scheduling
-     * takes precedence; later scheduling are ignored.
-     *</p>
-     *<p>
-     * caveat: A consequence is that if a variant is scheduled on the low priority queue and later
-     * scheduled again on the high priority queue, the later scheduling is ignored.
-     * Therefore, the second callback could be called before the variant is compiled.
-     * However, the first callback, if specified, will trigger as expected.
-     *</p>
-     *<p>
-     * The callback is guaranteed to be called. If the engine is destroyed while some material
-     * variants are still compiling or in the queue, these will be discarded and the corresponding
-     * callback will be called. In that case however the Material pointer passed to the callback
-     * is guaranteed to be invalid (either because it's been destroyed by the user already, or,
-     * because it's been cleaned-up by the Engine).
-     *</p>
-     *<p>
-     * {@link UserVariantFilterBit#ALL} should be used with caution. Only variants that an application
-     * needs should be included in the variants argument. For example, the STE variant is only used
-     * for stereoscopic rendering. If an application is not planning to render in stereo, this bit
-     * should be turned off to avoid unnecessary material compilations.
-     *</p>
-     * @param priority      Which priority queue to use, LOW or HIGH.
-     * @param variants      Variants to include to the compile command.
-     * @param handler       An {@link java.util.concurrent.Executor Executor}. On Android this can also be a {@link android.os.Handler Handler}.
-     * @param callback      callback called on the main thread when the compilation is done on
-     *                      by backend.
-     */
-    public void compile(@NonNull CompilerPriorityQueue priority,
-                        int variants,
-                        @Nullable Object handler,
-                        @Nullable Runnable callback) {
-        nCompile(getNativeObject(), priority.ordinal(), variants, handler, callback);
-    }
-
    /**
     * Creates a new instance of this material. Material instances should be freed using
     * {@link Engine#destroyMaterialInstance(MaterialInstance)}.
@@ -1028,7 +953,6 @@ public class Material {
    private static native long nCreateInstanceWithName(long nativeMaterial, @NonNull String name);
    private static native long nGetDefaultInstance(long nativeMaterial);

-    private static native void nCompile(long nativeMaterial, int priority, int variants, Object handler, Runnable runnable);
    private static native String nGetName(long nativeMaterial);
    private static native int nGetShading(long nativeMaterial);
    private static native int nGetInterpolation(long nativeMaterial);
@@ -16,7 +16,6 @@

 package com.google.android.filament;

-import androidx.annotation.NonNull;
 import androidx.annotation.Nullable;

 /**
@@ -191,52 +190,6 @@ public class Scene {
        return mNativeObject;
    }

-    /**
-     * Returns the list of all entities in the Scene. If outArray is provided and large enough,
-     * it is used to store the list and returned, otherwise a new array is allocated and returned.
-     * @param outArray an array to store the list of entities in the scene.
-     * @return outArray if it was used or a newly allocated array.
-     * @see #getEntityCount
-     */
-    public int[] getEntities(@Nullable int[] outArray) {
-        int c = getEntityCount();
-        if (outArray == null || outArray.length < c) {
-            outArray = new int[c];
-        }
-        boolean success = nGetEntities(getNativeObject(), outArray, outArray.length);
-        if (!success) {
-            throw new IllegalStateException("Error retriving Scene's entities");
-        }
-        return outArray;
-    }
-
-    /**
-     * Returns the list of all entities in the Scene in a newly allocated array.
-     * @return an array containing the list of all entities in the scene.
-     * @see #getEntityCount
-     */
-    public int[] getEntities() {
-        return getEntities(null);
-    }
-
-    public interface EntityProcessor {
-        void process(@Entity int entity);
-    }
-
-    /**
-     * Invokes user functor on each entity in the scene.
-     *
-     * It is not allowed to add or remove an entity from the scene within the functor.
-     *
-     * @param entityProcessor User provided functor called for each entity in the scene
-     */
-    public void forEach(@NonNull EntityProcessor entityProcessor) {
-        int[] entities = getEntities(null);
-        for (int entity : entities) {
-            entityProcessor.process(entity);
-        }
-    }
-
    void clearNativeObject() {
        mNativeObject = 0;
    }
@@ -251,5 +204,4 @@ public class Scene {
    private static native int nGetRenderableCount(long nativeScene);
    private static native int nGetLightCount(long nativeScene);
    private static native boolean nHasEntity(long nativeScene, int entity);
-    private static native boolean nGetEntities(long nativeScene, int[] outArray, int length);
 }
@@ -849,10 +849,6 @@ public class Texture {
        public static final int SAMPLEABLE = 0x10;
        /** Texture can be used as a subpass input */
        public static final int SUBPASS_INPUT = 0x20;
-        /** Texture can be used the source of a blit() */
-        public static final int BLIT_SRC = 0x40;
-        /** Texture can be used the destination of a blit() */
-        public static final int BLIT_DST = 0x80;
        /** by default textures are <code>UPLOADABLE</code> and <code>SAMPLEABLE</code>*/
        public static final int DEFAULT = UPLOADABLE | SAMPLEABLE;
    }
@@ -75,7 +75,6 @@ public class View {
    private AmbientOcclusionOptions mAmbientOcclusionOptions;
    private BloomOptions mBloomOptions;
    private FogOptions mFogOptions;
-    private StereoscopicOptions mStereoscopicOptions;
    private RenderTarget mRenderTarget;
    private BlendMode mBlendMode;
    private DepthOfFieldOptions mDepthOfFieldOptions;
@@ -1056,51 +1055,6 @@ public class View {
        return nIsStencilBufferEnabled(getNativeObject());
    }

-    /**
-     * Sets the stereoscopic rendering options for this view.
-     *
-     * <p>
-     * Currently, only one type of stereoscopic rendering is supported: side-by-side.
-     * Side-by-side stereo rendering splits the viewport into two halves: a left and right half.
-     * Eye 0 will render to the left half, while Eye 1 will render into the right half.
-     * </p>
-     *
-     * <p>
-     * Currently, the following features are not supported with stereoscopic rendering:
-     * - post-processing
-     * - shadowing
-     * - punctual lights
-     * </p>
-     *
-     * <p>
-     * Stereo rendering depends on device and platform support. To check if stereo rendering is
-     * supported, use {@link Engine#isStereoSupported()}. If stereo rendering is not supported, then
-     * the stereoscopic options have no effect.
-     * </p>
-     *
-     * @param options The stereoscopic options to use on this view
-     * @see #getStereoscopicOptions
-     */
-    public void setStereoscopicOptions(@NonNull StereoscopicOptions options) {
-        mStereoscopicOptions = options;
-        nSetStereoscopicOptions(getNativeObject(), options.enabled);
-    }
-
-    /**
-     * Gets the stereoscopic options.
-     *
-     * @return options Stereoscopic options currently set.
-     * @see #setStereoscopicOptions
-     */
-    @NonNull
-    public StereoscopicOptions getStereoscopicOptions() {
-        if (mStereoscopicOptions == null) {
-            mStereoscopicOptions = new StereoscopicOptions();
-        }
-        return mStereoscopicOptions;
-    }
-
-
    /**
     * A class containing the result of a picking query
     */
@@ -1266,7 +1220,6 @@ public class View {
    private static native void nSetBloomOptions(long nativeView, long dirtNativeObject, float dirtStrength, float strength, int resolution, int levels, int blendMode, boolean threshold, boolean enabled, float highlight,
            boolean lensFlare, boolean starburst, float chromaticAberration, int ghostCount, float ghostSpacing, float ghostThreshold, float haloThickness, float haloRadius, float haloThreshold);
    private static native void nSetFogOptions(long nativeView, float distance, float maximumOpacity, float height, float heightFalloff, float cutOffDistance, float v, float v1, float v2, float density, float inScatteringStart, float inScatteringSize, boolean fogColorFromIbl, long skyColorNativeObject, boolean enabled);
-    private static native void nSetStereoscopicOptions(long nativeView, boolean enabled);
    private static native void nSetBlendMode(long nativeView, int blendMode);
    private static native void nSetDepthOfFieldOptions(long nativeView, float cocScale, float maxApertureDiameter, boolean enabled, int filter,
            boolean nativeResolution, int foregroundRingCount, int backgroundRingCount, int fastGatherRingCount, int maxForegroundCOC, int maxBackgroundCOC);
@@ -1637,10 +1590,6 @@ public class View {
         * circle of confusion scale factor (amount of blur)
         */
        public float cocScale = 1.0f;
-        /**
-         * width/height aspect ratio of the circle of confusion (simulate anamorphic lenses)
-         */
-        public float cocAspectRatio = 1.0f;
        /**
         * maximum aperture diameter in meters (zero to disable rotation)
         */
@@ -1856,7 +1805,7 @@ public class View {
    }

    /**
-     * Options for Multi-Sample Anti-aliasing (MSAA)
+     * Options for Temporal Multi-Sample Anti-aliasing (MSAA)
     * @see setMultiSampleAntiAliasingOptions()
     */
    public static class MultiSampleAntiAliasingOptions {
@@ -1880,111 +1829,21 @@ 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.
-     *
-     * `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.
-     *
     * @see setTemporalAntiAliasingOptions()
     */
    public static class TemporalAntiAliasingOptions {
-        public enum BoxType {
-            /**
-             * use an AABB neighborhood
-             */
-            AABB,
-            /**
-             * use the variance of the neighborhood (not recommended)
-             */
-            VARIANCE,
-            /**
-             * use both AABB and variance
-             */
-            AABB_VARIANCE,
-        }
-
-        public enum BoxClipping {
-            /**
-             * Accurate box clipping
-             */
-            ACCURATE,
-            /**
-             * clamping
-             */
-            CLAMP,
-            /**
-             * no rejections (use for debugging)
-             */
-            NONE,
-        }
-
-        public enum JitterPattern {
-            RGSS_X4,
-            UNIFORM_HELIX_X4,
-            HALTON_23_X8,
-            HALTON_23_X16,
-            HALTON_23_X32,
-        }
-
        /**
-         * reconstruction filter width typically between 0.2 (sharper, aliased) and 1.5 (smoother)
+         * reconstruction filter width typically between 0 (sharper, aliased) and 1 (smoother)
         */
        public float filterWidth = 1.0f;
        /**
         * history feedback, between 0 (maximum temporal AA) and 1 (no temporal AA).
         */
-        public float feedback = 0.12f;
-        /**
-         * 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;
+        public float feedback = 0.04f;
        /**
         * enables or disables temporal anti-aliasing
         */
        public boolean enabled = false;
-        /**
-         * 4x TAA upscaling. Disables Dynamic Resolution. [BETA]
-         */
-        public boolean upscaling = false;
-        /**
-         * whether to filter the history buffer
-         */
-        public boolean filterHistory = true;
-        /**
-         * whether to apply the reconstruction filter to the input
-         */
-        public boolean filterInput = true;
-        /**
-         * whether to use the YcoCg color-space for history rejection
-         */
-        public boolean useYCoCg = false;
-        /**
-         * type of color gamut box
-         */
-        @NonNull
-        public TemporalAntiAliasingOptions.BoxType boxType = TemporalAntiAliasingOptions.BoxType.AABB;
-        /**
-         * clipping algorithm
-         */
-        @NonNull
-        public TemporalAntiAliasingOptions.BoxClipping boxClipping = TemporalAntiAliasingOptions.BoxClipping.ACCURATE;
-        @NonNull
-        public TemporalAntiAliasingOptions.JitterPattern jitterPattern = TemporalAntiAliasingOptions.JitterPattern.HALTON_23_X16;
-        public float varianceGamma = 1.0f;
-        /**
-         * adjust the feedback dynamically to reduce flickering
-         */
-        public boolean preventFlickering = false;
-        /**
-         * whether to apply history reprojection (debug option)
-         */
-        public boolean historyReprojection = true;
    }

    /**
@@ -182,85 +182,28 @@ public class UiHelper {
        void detach();
    }

-    private class SurfaceViewHandler implements RenderSurface, SurfaceHolder.Callback {
-        @NonNull private final SurfaceView mSurfaceView;
+    private static class SurfaceViewHandler implements RenderSurface {
+        private final SurfaceView mSurfaceView;

-        SurfaceViewHandler(@NonNull SurfaceView surfaceView) {
-            mSurfaceView = surfaceView;
-
-            @NonNull SurfaceHolder holder = surfaceView.getHolder();
-            holder.addCallback(this);
-
-            if (mDesiredWidth > 0 && mDesiredHeight > 0) {
-                holder.setFixedSize(mDesiredWidth, mDesiredHeight);
-            }
-
-            // in case the SurfaceView's surface already existed
-            final Surface surface = holder.getSurface();
-            if (surface != null && surface.isValid()) {
-                surfaceCreated(holder);
-                // there is no way to retrieve the actual PixelFormat, since it is not used
-                // in the callback, we can use whatever we want.
-                surfaceChanged(holder, PixelFormat.RGBA_8888,
-                    holder.getSurfaceFrame().width(), holder.getSurfaceFrame().height());
-            }
+        SurfaceViewHandler(SurfaceView surface) {
+            mSurfaceView = surface;
        }

        @Override
        public void resize(int width, int height) {
-            @NonNull SurfaceHolder holder = mSurfaceView.getHolder();
-            holder.setFixedSize(width, height);
+            mSurfaceView.getHolder().setFixedSize(width, height);
        }

        @Override
        public void detach() {
-            @NonNull SurfaceHolder holder = mSurfaceView.getHolder();
-            holder.removeCallback(this);
-        }
-
-        @Override
-        public void surfaceCreated(@NonNull SurfaceHolder holder) {
-            if (LOGGING) Log.d(LOG_TAG, "surfaceCreated()");
-            createSwapChain(holder.getSurface());
-        }
-
-        @Override
-        public void surfaceChanged(
-            @NonNull SurfaceHolder holder, int format, int width, int height) {
-            // Note: this is always called at least once after surfaceCreated()
-            if (LOGGING) Log.d(LOG_TAG, "surfaceChanged(" + width + ", " + height + ")");
-            if (mRenderCallback != null) {
-                mRenderCallback.onResized(width, height);
-            }
-        }
-
-        @Override
-        public void surfaceDestroyed(@NonNull SurfaceHolder holder) {
-            if (LOGGING) Log.d(LOG_TAG, "surfaceDestroyed()");
-            destroySwapChain();
        }
    }

-    private class SurfaceHolderHandler implements RenderSurface, SurfaceHolder.Callback {
+    private static class SurfaceHolderHandler implements RenderSurface {
        private final SurfaceHolder mSurfaceHolder;

-        SurfaceHolderHandler(@NonNull SurfaceHolder holder) {
-            mSurfaceHolder = holder;
-            holder.addCallback(this);
-
-            if (mDesiredWidth > 0 && mDesiredHeight > 0) {
-                holder.setFixedSize(mDesiredWidth, mDesiredHeight);
-            }
-
-            // in case the SurfaceHolder's surface already existed
-            final Surface surface = holder.getSurface();
-            if (surface != null && surface.isValid()) {
-                surfaceCreated(holder);
-                // there is no way to retrieve the actual PixelFormat, since it is not used
-                // in the callback, we can use whatever we want.
-                surfaceChanged(holder, PixelFormat.RGBA_8888,
-                    holder.getSurfaceFrame().width(), holder.getSurfaceFrame().height());
-            }
+        SurfaceHolderHandler(SurfaceHolder surface) {
+            mSurfaceHolder = surface;
        }

        @Override
@@ -270,55 +213,19 @@ public class UiHelper {

        @Override
        public void detach() {
-            mSurfaceHolder.removeCallback(this);
-        }
-
-        @Override
-        public void surfaceCreated(@NonNull SurfaceHolder holder) {
-            if (LOGGING) Log.d(LOG_TAG, "surfaceCreated()");
-            createSwapChain(holder.getSurface());
-        }
-
-        @Override
-        public void surfaceChanged(@NonNull SurfaceHolder holder, int format, int width, int height) {
-            // Note: this is always called at least once after surfaceCreated()
-            if (LOGGING) Log.d(LOG_TAG, "surfaceChanged(" + width + ", " + height + ")");
-            if (mRenderCallback != null) {
-                mRenderCallback.onResized(width, height);
-            }
-        }
-
-        @Override
-        public void surfaceDestroyed(@NonNull SurfaceHolder surfaceHolder) {
-            if (LOGGING) Log.d(LOG_TAG, "surfaceDestroyed()");
-            destroySwapChain();
        }
    }

-    private class TextureViewHandler implements RenderSurface, TextureView.SurfaceTextureListener {
+    private class TextureViewHandler implements RenderSurface {
        private final TextureView mTextureView;
        private Surface mSurface;

-        TextureViewHandler(@NonNull TextureView view) {
-            mTextureView = view;
-            mTextureView.setSurfaceTextureListener(this);
-            // in case the View's SurfaceTexture already existed
-            if (view.isAvailable()) {
-                SurfaceTexture surfaceTexture = view.getSurfaceTexture();
-                if (surfaceTexture != null) {
-                    this.onSurfaceTextureAvailable(surfaceTexture,
-                        mDesiredWidth, mDesiredHeight);
-                }
-            }
-        }
+        TextureViewHandler(TextureView surface) { mTextureView = surface; }

        @Override
        public void resize(int width, int height) {
            if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.ICE_CREAM_SANDWICH_MR1) {
-                final SurfaceTexture surfaceTexture = mTextureView.getSurfaceTexture();
-                if (surfaceTexture != null) {
-                    surfaceTexture.setDefaultBufferSize(width, height);
-                }
+                mTextureView.getSurfaceTexture().setDefaultBufferSize(width, height);
            }
            if (mRenderCallback != null) {
                // the call above won't cause TextureView.onSurfaceTextureSizeChanged()
@@ -328,69 +235,10 @@ public class UiHelper {

        @Override
        public void detach() {
-            mTextureView.setSurfaceTextureListener(null);
-        }
-
-
-        @Override
-        public void onSurfaceTextureAvailable(
-            @NonNull SurfaceTexture surfaceTexture, int width, int height) {
-            if (LOGGING) Log.d(LOG_TAG, "onSurfaceTextureAvailable()");
-
-            if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.ICE_CREAM_SANDWICH_MR1) {
-                if (mDesiredWidth > 0 && mDesiredHeight > 0) {
-                    surfaceTexture.setDefaultBufferSize(mDesiredWidth, mDesiredHeight);
-                }
-            }
-
-            final Surface surface = new Surface(surfaceTexture);
-            setSurface(surface);
-            createSwapChain(surface);
-
-            if (mRenderCallback != null) {
-                // Call this the first time because onSurfaceTextureSizeChanged()
-                // isn't called at initialization time
-                mRenderCallback.onResized(width, height);
-            }
-        }
-
-        @Override
-        public void onSurfaceTextureSizeChanged(
-            @NonNull SurfaceTexture surfaceTexture, int width, int height) {
-            if (LOGGING) Log.d(LOG_TAG, "onSurfaceTextureSizeChanged()");
-            if (mRenderCallback != null) {
-                if (mDesiredWidth > 0 && mDesiredHeight > 0) {
-                    surfaceTexture.setDefaultBufferSize(mDesiredWidth, mDesiredHeight);
-                    mRenderCallback.onResized(mDesiredWidth, mDesiredHeight);
-                } else {
-                    mRenderCallback.onResized(width, height);
-                }
-                // We must recreate the SwapChain to guarantee that it sees the new size.
-                // More precisely, for an EGL client, the EGLSurface must be recreated. For
-                // a Vulkan client, the SwapChain must be recreated. Calling
-                // onNativeWindowChanged() will accomplish that.
-                // This requirement comes from SurfaceTexture.setDefaultBufferSize()
-                // documentation.
-                final Surface surface = getSurface();
-                if (surface != null) {
-                    mRenderCallback.onNativeWindowChanged(surface);
-                }
-            }
-        }
-
-        @Override
-        public boolean onSurfaceTextureDestroyed(@NonNull SurfaceTexture surfaceTexture) {
-            if (LOGGING) Log.d(LOG_TAG, "onSurfaceTextureDestroyed()");
            setSurface(null);
-            destroySwapChain();
-            return true;
        }

-        @Override
-        public void onSurfaceTextureUpdated(@NonNull SurfaceTexture surface) { }
-
-
-        private void setSurface(@Nullable Surface surface) {
+        void setSurface(Surface surface) {
            if (surface == null) {
                if (mSurface != null) {
                    mSurface.release();
@@ -399,7 +247,7 @@ public class UiHelper {
            mSurface = surface;
        }

-        private Surface getSurface() {
+        public Surface getSurface() {
            return mSurface;
        }
    }
@@ -445,9 +293,6 @@ public class UiHelper {
     * {@link #attachTo(TextureView)}, or {@link #attachTo(SurfaceHolder)}.
     */
    public void detach() {
-        if (mRenderSurface != null) {
-            mRenderSurface.detach();
-        }
        destroySwapChain();
        mNativeWindow = null;
        mRenderSurface = null;
@@ -557,8 +402,48 @@ public class UiHelper {
                view.setZOrderOnTop(translucent);
            }

-            view.getHolder().setFormat(isOpaque() ? PixelFormat.OPAQUE : PixelFormat.TRANSLUCENT);
+            int format = isOpaque() ? PixelFormat.OPAQUE : PixelFormat.TRANSLUCENT;
+            view.getHolder().setFormat(format);
+
            mRenderSurface = new SurfaceViewHandler(view);
+
+            final SurfaceHolder.Callback callback = new SurfaceHolder.Callback() {
+                @Override
+                public void surfaceCreated(@NonNull SurfaceHolder holder) {
+                    if (LOGGING) Log.d(LOG_TAG, "surfaceCreated()");
+                    createSwapChain(holder.getSurface());
+                }
+
+                @Override
+                public void surfaceChanged(
+                    @NonNull SurfaceHolder holder, int format, int width, int height) {
+                    // Note: this is always called at least once after surfaceCreated()
+                    if (LOGGING) Log.d(LOG_TAG, "surfaceChanged(" + width + ", " + height + ")");
+                    if (mRenderCallback != null) {
+                        mRenderCallback.onResized(width, height);
+                    }
+                }
+
+                @Override
+                public void surfaceDestroyed(@NonNull SurfaceHolder holder) {
+                    if (LOGGING) Log.d(LOG_TAG, "surfaceDestroyed()");
+                    destroySwapChain();
+                }
+            };
+
+            SurfaceHolder holder = view.getHolder();
+            holder.addCallback(callback);
+            if (mDesiredWidth > 0 && mDesiredHeight > 0) {
+                holder.setFixedSize(mDesiredWidth, mDesiredHeight);
+            }
+
+            // in case the SurfaceView's surface already existed
+            final Surface surface = holder.getSurface();
+            if (surface != null && surface.isValid()) {
+                callback.surfaceCreated(holder);
+                callback.surfaceChanged(holder, format,
+                        holder.getSurfaceFrame().width(), holder.getSurfaceFrame().height());
+            }
        }
    }

@@ -570,7 +455,76 @@ public class UiHelper {
    public void attachTo(@NonNull TextureView view) {
        if (attach(view)) {
            view.setOpaque(isOpaque());
+
            mRenderSurface = new TextureViewHandler(view);
+
+            TextureView.SurfaceTextureListener listener = new TextureView.SurfaceTextureListener() {
+                @Override
+                public void onSurfaceTextureAvailable(
+                    @NonNull SurfaceTexture surfaceTexture, int width, int height) {
+                    if (LOGGING) Log.d(LOG_TAG, "onSurfaceTextureAvailable()");
+
+                    if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.ICE_CREAM_SANDWICH_MR1) {
+                        if (mDesiredWidth > 0 && mDesiredHeight > 0) {
+                            surfaceTexture.setDefaultBufferSize(mDesiredWidth, mDesiredHeight);
+                        }
+                    }
+
+                    Surface surface = new Surface(surfaceTexture);
+                    TextureViewHandler textureViewHandler = (TextureViewHandler) mRenderSurface;
+                    textureViewHandler.setSurface(surface);
+
+                    createSwapChain(surface);
+
+                    if (mRenderCallback != null) {
+                        // Call this the first time because onSurfaceTextureSizeChanged()
+                        // isn't called at initialization time
+                        mRenderCallback.onResized(width, height);
+                    }
+                }
+
+                @Override
+                public void onSurfaceTextureSizeChanged(
+                    @NonNull SurfaceTexture surfaceTexture, int width, int height) {
+                    if (LOGGING) Log.d(LOG_TAG, "onSurfaceTextureSizeChanged()");
+                    if (mRenderCallback != null) {
+                        if (mDesiredWidth > 0 && mDesiredHeight > 0) {
+                            surfaceTexture.setDefaultBufferSize(mDesiredWidth, mDesiredHeight);
+                            mRenderCallback.onResized(mDesiredWidth, mDesiredHeight);
+                        } else {
+                            mRenderCallback.onResized(width, height);
+                        }
+                        // We must recreate the SwapChain to guarantee that it sees the new size.
+                        // More precisely, for an EGL client, the EGLSurface must be recreated. For
+                        // a Vulkan client, the SwapChain must be recreated. Calling
+                        // onNativeWindowChanged() will accomplish that.
+                        // This requirement comes from SurfaceTexture.setDefaultBufferSize()
+                        // documentation.
+                        TextureViewHandler textureViewHandler = (TextureViewHandler) mRenderSurface;
+                        mRenderCallback.onNativeWindowChanged(textureViewHandler.getSurface());
+                    }
+                }
+
+                @Override
+                public boolean onSurfaceTextureDestroyed(@NonNull SurfaceTexture surfaceTexture) {
+                    if (LOGGING) Log.d(LOG_TAG, "onSurfaceTextureDestroyed()");
+                    destroySwapChain();
+                    return true;
+                }
+
+                @Override
+                public void onSurfaceTextureUpdated(@NonNull SurfaceTexture surface) { }
+            };
+
+            view.setSurfaceTextureListener(listener);
+
+            // in case the View's SurfaceTexture already existed
+            if (view.isAvailable()) {
+                SurfaceTexture surfaceTexture = view.getSurfaceTexture();
+                if (surfaceTexture != null) {
+                    listener.onSurfaceTextureAvailable(surfaceTexture, mDesiredWidth, mDesiredHeight);
+                }
+            }
        }
    }

@@ -581,8 +535,46 @@ public class UiHelper {
     */
    public void attachTo(@NonNull SurfaceHolder holder) {
        if (attach(holder)) {
-            holder.setFormat(isOpaque() ? PixelFormat.OPAQUE : PixelFormat.TRANSLUCENT);
+            int format = isOpaque() ? PixelFormat.OPAQUE : PixelFormat.TRANSLUCENT;
+            holder.setFormat(format);
+
            mRenderSurface = new SurfaceHolderHandler(holder);
+
+            final SurfaceHolder.Callback callback = new SurfaceHolder.Callback() {
+                @Override
+                public void surfaceCreated(@NonNull SurfaceHolder surfaceHolder) {
+                    if (LOGGING) Log.d(LOG_TAG, "surfaceCreated()");
+                    createSwapChain(holder.getSurface());
+                }
+
+                @Override
+                public void surfaceChanged(@NonNull SurfaceHolder holder, int format, int width, int height) {
+                    // Note: this is always called at least once after surfaceCreated()
+                    if (LOGGING) Log.d(LOG_TAG, "surfaceChanged(" + width + ", " + height + ")");
+                    if (mRenderCallback != null) {
+                        mRenderCallback.onResized(width, height);
+                    }
+                }
+
+                @Override
+                public void surfaceDestroyed(@NonNull SurfaceHolder surfaceHolder) {
+                    if (LOGGING) Log.d(LOG_TAG, "surfaceDestroyed()");
+                    destroySwapChain();
+                }
+            };
+
+            holder.addCallback(callback);
+            if (mDesiredWidth > 0 && mDesiredHeight > 0) {
+                holder.setFixedSize(mDesiredWidth, mDesiredHeight);
+            }
+
+            // in case the SurfaceHolder's surface already existed
+            final Surface surface = holder.getSurface();
+            if (surface != null && surface.isValid()) {
+                callback.surfaceCreated(holder);
+                callback.surfaceChanged(holder, format,
+                        holder.getSurfaceFrame().width(), holder.getSurfaceFrame().height());
+            }
        }
    }

@@ -593,10 +585,6 @@ public class UiHelper {
                // nothing to do
                return false;
            }
-            if (mRenderSurface != null) {
-                mRenderSurface.detach();
-                mRenderSurface = null;
-            }
            destroySwapChain();
        }
        mNativeWindow = nativeWindow;
@@ -611,6 +599,9 @@ public class UiHelper {
    }

    private void destroySwapChain() {
+        if (mRenderSurface != null) {
+            mRenderSurface.detach();
+        }
        if (mRenderCallback != null) {
            mRenderCallback.onDetachedFromSurface();
        }
@@ -144,7 +144,7 @@ public class FilamentInstance {
     *
     * Ignored if variantIndex is out of bounds.
     */
-    public void applyMaterialVariant(@IntRange(from = 0) int variantIndex) {
+    void applyMaterialVariant(@IntRange(from = 0) int variantIndex) {
        nApplyMaterialVariant(mNativeObject, variantIndex);
    }

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.50.3
+VERSION_NAME=1.45.0

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

@@ -1,6 +1,6 @@
 #Wed Nov 17 10:40:18 PST 2021
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-8.2-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-8.0-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists
@@ -26,7 +26,6 @@ import android.widget.TextView
 import android.widget.Toast
 import com.google.android.filament.Fence
 import com.google.android.filament.IndirectLight
-import com.google.android.filament.Material
 import com.google.android.filament.Skybox
 import com.google.android.filament.View
 import com.google.android.filament.View.OnPickCallback
@@ -393,36 +392,6 @@ class MainActivity : Activity() {
                    Log.i(TAG, "The Filament backend took $total ms to load the model geometry.")
                    modelViewer.engine.destroyFence(it)
                    loadStartFence = null
-
-                    val materials = mutableSetOf<Material>()
-                    val rcm = modelViewer.engine.renderableManager
-                    modelViewer.scene.forEach {
-                        val entity = it
-                        if (rcm.hasComponent(entity)) {
-                            val ri = rcm.getInstance(entity)
-                            val c = rcm.getPrimitiveCount(ri)
-                            for (i in 0 until c) {
-                                val mi = rcm.getMaterialInstanceAt(ri, i)
-                                val ma = mi.material
-                                materials.add(ma)
-                            }
-                        }
-                    }
-                    materials.forEach {
-                        it.compile(
-                            Material.CompilerPriorityQueue.HIGH,
-                            Material.UserVariantFilterBit.DIRECTIONAL_LIGHTING or
-                            Material.UserVariantFilterBit.DYNAMIC_LIGHTING or
-                            Material.UserVariantFilterBit.SHADOW_RECEIVER,
-                            null, null)
-                        it.compile(
-                            Material.CompilerPriorityQueue.LOW,
-                            Material.UserVariantFilterBit.FOG or
-                            Material.UserVariantFilterBit.SKINNING or
-                            Material.UserVariantFilterBit.SSR or
-                            Material.UserVariantFilterBit.VSM,
-                            null, null)
-                    }
                }
            }

@@ -20,8 +20,6 @@ import android.animation.ValueAnimator
 import android.app.Activity
 import android.opengl.Matrix
 import android.os.Bundle
-import android.os.Handler
-import android.os.Looper
 import android.view.Choreographer
 import android.view.Surface
 import android.view.SurfaceView
@@ -159,14 +157,6 @@ class MainActivity : Activity() {
    private fun loadMaterial() {
        readUncompressedAsset("materials/baked_color.filamat").let {
            material = Material.Builder().payload(it, it.remaining()).build(engine)
-            material.compile(
-                Material.CompilerPriorityQueue.HIGH,
-                Material.UserVariantFilterBit.ALL,
-                Handler(Looper.getMainLooper())) {
-                        android.util.Log.i("hellotriangle",
-                            "Material " + material.name + " compiled.")
-            }
-            engine.flush()
        }
    }

@@ -50,7 +50,7 @@ function replace {
    FIND_STR="${1//\{\{VERSION\}\}/${VERSION_REGEX}}"
    REPLACE_STR="${1//\{\{VERSION\}\}/${NEW_VERSION}}"
    local FILE_NAME="$2"
-    if [ $IS_DARWIN == 1 ]; then
+    if [ IS_DARWIN ]; then
        sed -i '' -E "s/${FIND_STR}/${REPLACE_STR}/" "${FILE_NAME}"
    else
        sed -i -E "s/${FIND_STR}/${REPLACE_STR}/" "${FILE_NAME}"
@@ -125,13 +125,7 @@ if [[ "${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.
@@ -587,7 +587,6 @@
 <li><em>overrides</em> Dictionary with one or more of the following properties: mapSize, shadowCascades, constantBias, normalBias, shadowFar, shadowNearHint, shadowFarHint, stable, polygonOffsetConstant, polygonOffsetSlope, \</li>
 </ul>
 </li>
-<li><strong>engine.setStereoscopicOptions()</strong></li>
 <li><strong>engine.setTemporalAntiAliasingOptions()</strong></li>
 <li><strong>engine.setVignetteOptions()</strong></li>
 </ul>
@@ -1463,8 +1462,6 @@ This defines the following functions:</p>
 <li>STENCIL_ATTACHMENT</li>
 <li>UPLOADABLE</li>
 <li>SAMPLEABLE</li>
-<li>BLIT_SRC</li>
-<li>BLIT_DST</li>
 <li>SUBPASS_INPUT</li>
 </ul>
 </div>
@@ -209,49 +209,44 @@ set(PRIVATE_HDRS
 )

 set(MATERIAL_SRCS
-        src/materials/antiAliasing/fxaa.mat
-        src/materials/antiAliasing/taa.mat
-        src/materials/blitLow.mat
-        src/materials/bloom/bloomDownsample.mat
-        src/materials/bloom/bloomDownsample2x.mat
-        src/materials/bloom/bloomDownsample9.mat
-        src/materials/bloom/bloomUpsample.mat
+        src/materials/fsr/fsr_easu.mat
+        src/materials/fsr/fsr_easu_mobile.mat
+        src/materials/fsr/fsr_easu_mobileF.mat
+        src/materials/fsr/fsr_rcas.mat
        src/materials/colorGrading/colorGrading.mat
        src/materials/colorGrading/colorGradingAsSubpass.mat
        src/materials/colorGrading/customResolveAsSubpass.mat
        src/materials/debugShadowCascades.mat
        src/materials/defaultMaterial.mat
+        src/materials/defaultMaterial0.mat
        src/materials/dof/dof.mat
        src/materials/dof/dofCoc.mat
-        src/materials/dof/dofCombine.mat
-        src/materials/dof/dofDilate.mat
        src/materials/dof/dofDownsample.mat
-        src/materials/dof/dofMedian.mat
-        src/materials/dof/dofMipmap.mat
+        src/materials/dof/dofCombine.mat
        src/materials/dof/dofTiles.mat
        src/materials/dof/dofTilesSwizzle.mat
+        src/materials/dof/dofDilate.mat
+        src/materials/dof/dofMipmap.mat
+        src/materials/dof/dofMedian.mat
        src/materials/flare/flare.mat
-        src/materials/fsr/fsr_easu.mat
-        src/materials/fsr/fsr_easu_mobile.mat
-        src/materials/fsr/fsr_easu_mobileF.mat
-        src/materials/fsr/fsr_rcas.mat
-        src/materials/resolveDepth.mat
-        src/materials/separableGaussianBlur.mat
-        src/materials/skybox.mat
-        src/materials/shadowmap.mat
+        src/materials/blitLow.mat
+        src/materials/bloom/bloomDownsample.mat
+        src/materials/bloom/bloomDownsample2x.mat
+        src/materials/bloom/bloomDownsample9.mat
+        src/materials/bloom/bloomUpsample.mat
        src/materials/ssao/bilateralBlur.mat
        src/materials/ssao/bilateralBlurBentNormals.mat
        src/materials/ssao/mipmapDepth.mat
+        src/materials/skybox.mat
+        src/materials/skybox0.mat
        src/materials/ssao/sao.mat
        src/materials/ssao/saoBentNormals.mat
+        src/materials/separableGaussianBlur.mat
+        src/materials/antiAliasing/fxaa.mat
+        src/materials/antiAliasing/taa.mat
        src/materials/vsmMipmap.mat
 )

-set(MATERIAL_FL0_SRCS
-        src/materials/defaultMaterial0.mat
-        src/materials/skybox0.mat
-)
-
 # Embed the binary resource blob for materials.
 get_resgen_vars(${RESOURCE_DIR} materials)
 list(APPEND PRIVATE_HDRS ${RESGEN_HEADER})
@@ -276,11 +271,6 @@ if (NOT DFG_LUT_SIZE)
 endif()
 message(STATUS "DFG LUT size set to ${DFG_LUT_SIZE}x${DFG_LUT_SIZE}")

-# Whether to include FL0 materials.
-if (FILAMENT_ENABLE_FEATURE_LEVEL_0)
-    add_definitions(-DFILAMENT_ENABLE_FEATURE_LEVEL_0)
-endif()
-
 # ==================================================================================================
 # Definitions
 # ==================================================================================================
@@ -322,23 +312,6 @@ foreach (mat_src ${MATERIAL_SRCS})
    list(APPEND MATERIAL_BINS ${output_path})
 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(
@@ -169,8 +169,6 @@ if (FILAMENT_SUPPORTS_VULKAN)
            src/vulkan/platform/VulkanPlatform.cpp
            src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp
            src/vulkan/platform/VulkanPlatformSwapChainImpl.h
-            src/vulkan/spirv/VulkanSpirvUtils.cpp
-            src/vulkan/spirv/VulkanSpirvUtils.h
            src/vulkan/VulkanBlitter.cpp
            src/vulkan/VulkanBlitter.h
            src/vulkan/VulkanBuffer.cpp
@@ -315,7 +313,6 @@ endif()

 if (FILAMENT_SUPPORTS_VULKAN)
    target_link_libraries(${TARGET} PUBLIC bluevk vkmemalloc vkshaders smol-v)
-    target_link_libraries(${TARGET} PRIVATE SPIRV-Headers)
 endif()

 if (FILAMENT_SUPPORTS_METAL)
@@ -400,8 +397,8 @@ install(DIRECTORY ${PUBLIC_HDR_DIR}/backend DESTINATION include)
 # ==================================================================================================
 option(INSTALL_BACKEND_TEST "Install the backend test library so it can be consumed on iOS" OFF)

-if (APPLE OR LINUX)
-    set(BACKEND_TEST_SRC
+if (APPLE)
+    add_library(backend_test STATIC
        test/BackendTest.cpp
        test/ShaderGenerator.cpp
        test/TrianglePrimitive.cpp
@@ -413,26 +410,19 @@ if (APPLE OR LINUX)
        test/test_BufferUpdates.cpp
        test/test_MRT.cpp
        test/test_LoadImage.cpp
+        test/test_RenderExternalImage.cpp
        test/test_StencilBuffer.cpp
        test/test_Scissor.cpp
        test/test_MipLevels.cpp
-    )
-    set(BACKEND_TEST_LIBS
+        )
+
+    target_link_libraries(backend_test PRIVATE
        backend
        getopt
        gtest
-        imageio
        filamat
        SPIRV
        spirv-cross-glsl)
-endif()
-
-if (APPLE)
-    # TODO: we should expand this test to Linux and other platforms.
-    list(APPEND BACKEND_TEST_SRC
-         test/test_RenderExternalImage.cpp)
-    add_library(backend_test STATIC ${BACKEND_TEST_SRC})
-    target_link_libraries(backend_test PRIVATE ${BACKEND_TEST_LIBS})

    set(BACKEND_TEST_DEPS
            OGLCompiler
@@ -446,11 +436,12 @@ if (APPLE)
            glslang
            spirv-cross-core
            spirv-cross-glsl
-            spirv-cross-msl)
+            spirv-cross-msl
+            )

    if (NOT IOS)
        target_link_libraries(backend_test PRIVATE image imageio)
-        list(APPEND BACKEND_TEST_DEPS image)
+        list(APPEND BACKEND_TEST_DEPS image imageio)
    endif()

    set(BACKEND_TEST_COMBINED_OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/libbackendtest_combined.a")
@@ -464,21 +455,15 @@ if (APPLE)
    endif()

    set_target_properties(backend_test PROPERTIES FOLDER Tests)
-
-    if (APPLE AND NOT IOS)
-       add_executable(backend_test_mac test/mac_runner.mm)
-       target_link_libraries(backend_test_mac PRIVATE "-framework Metal -framework AppKit -framework QuartzCore")
-       # Because each test case is a separate file, the -force_load flag is necessary to prevent the
-       # linker from removing "unused" symbols.
-       target_link_libraries(backend_test_mac PRIVATE -force_load backend_test)
-       set_target_properties(backend_test_mac PROPERTIES FOLDER Tests)
-    endif()
 endif()

-if (LINUX)
-    add_executable(backend_test_linux test/linux_runner.cpp ${BACKEND_TEST_SRC})
-    target_link_libraries(backend_test_linux PRIVATE ${BACKEND_TEST_LIBS})
-    set_target_properties(backend_test_linux PROPERTIES FOLDER Tests)
+if (APPLE AND NOT IOS)
+    add_executable(backend_test_mac test/mac_runner.mm)
+    target_link_libraries(backend_test_mac PRIVATE "-framework Metal -framework AppKit -framework QuartzCore")
+    # Because each test case is a separate file, the -force_load flag is necessary to prevent the
+    # linker from removing "unused" symbols.
+    target_link_libraries(backend_test_mac PRIVATE -force_load backend_test)
+    set_target_properties(backend_test_mac PROPERTIES FOLDER Tests)
 endif()

 # ==================================================================================================
@@ -23,6 +23,7 @@
 #include <utils/ostream.h>

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

 namespace filament::backend {

@@ -112,7 +113,7 @@ public:
    /**
     * Helper to create a BufferDescriptor that uses a KNOWN method pointer w/ object passed
     * by pointer as the callback. e.g.:
-     *     auto bd = BufferDescriptor::make<Foo, &Foo::method>(buffer, size, foo);
+     *     auto bd = BufferDescriptor::make(buffer, size, &Foo::method, foo);
     *
     * @param buffer    Memory address of the CPU buffer to reference
     * @param size      Size of the CPU buffer in bytes
@@ -120,12 +121,12 @@ public:
     * @return          a new BufferDescriptor
     */
    template<typename T, void(T::*method)(void const*, size_t)>
-    static BufferDescriptor make(void const* buffer, size_t size, T* data,
-            CallbackHandler* handler = nullptr) noexcept {
+    static BufferDescriptor make(
+            void const* buffer, size_t size, T* data, CallbackHandler* handler = nullptr) noexcept {
        return {
                buffer, size,
                handler, [](void* b, size_t s, void* u) {
-                    (static_cast<T*>(u)->*method)(b, s);
+                    (*static_cast<T**>(u)->*method)(b, s);
                }, data
        };
    }
@@ -144,14 +145,14 @@ public:
     * @return          a new BufferDescriptor
     */
    template<typename T>
-    static BufferDescriptor make(void const* buffer, size_t size, T&& functor,
-            CallbackHandler* handler = nullptr) noexcept {
+    static BufferDescriptor make(
+            void const* buffer, size_t size, T&& functor, CallbackHandler* handler = nullptr) noexcept {
        return {
                buffer, size,
                handler, [](void* b, size_t s, void* u) {
-                    T* const that = static_cast<T*>(u);
-                    that->operator()(b, s);
-                    delete that;
+                    T& that = *static_cast<T*>(u);
+                    that(b, s);
+                    delete &that;
                },
                new T(std::forward<T>(functor))
        };
@@ -200,7 +201,7 @@ public:
        return mUser;
    }

-    //! CPU memory-buffer virtual address
+    //! CPU mempry-buffer virtual address
    void* buffer = nullptr;

    //! CPU memory-buffer size in bytes
@@ -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>
@@ -228,7 +227,6 @@ struct Viewport {
    int32_t top() const noexcept { return bottom + int32_t(height); }
 };

-
 /**
 * Specifies the mapping of the near and far clipping plane to window coordinates.
 */
@@ -358,6 +356,13 @@ enum class SamplerFormat : uint8_t {
    SHADOW = 3      //!< shadow sampler (PCF)
 };

+//! Texture color space
+enum class SamplerTransferFunction : uint8_t {
+    UNDEFINED = 0,      //!< don't care
+    IDENTITY = 1,       //!< sampled as-is
+    SRGB_TO_LINEAR = 2, //!< sampled as sRGB and converted to linear
+};
+
 /**
 * Supported element types
 */
@@ -660,15 +665,13 @@ enum class TextureFormat : uint16_t {

 //! Bitmask describing the intended Texture Usage
 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)
+    NONE                = 0x0,
+    COLOR_ATTACHMENT    = 0x1,                      //!< Texture can be used as a color attachment
+    DEPTH_ATTACHMENT    = 0x2,                      //!< Texture can be used as a depth attachment
+    STENCIL_ATTACHMENT  = 0x4,                      //!< Texture can be used as a stencil attachment
+    UPLOADABLE          = 0x8,                      //!< 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
 };

@@ -696,17 +699,6 @@ static constexpr bool isDepthFormat(TextureFormat format) noexcept {
    }
 }

-static constexpr bool isStencilFormat(TextureFormat format) noexcept {
-    switch (format) {
-        case TextureFormat::STENCIL8:
-        case TextureFormat::DEPTH24_STENCIL8:
-        case TextureFormat::DEPTH32F_STENCIL8:
-            return true;
-        default:
-            return false;
-    }
-}
-
 static constexpr bool isUnsignedIntFormat(TextureFormat format) {
    switch (format) {
        case TextureFormat::R8UI:
@@ -1195,8 +1187,8 @@ using FrameScheduledCallback = void(*)(PresentCallable callable, void* user);
 enum class Workaround : uint16_t {
    // The EASU pass must split because shader compiler flattens early-exit branch
    SPLIT_EASU,
-    // Backend allows feedback loop with ancillary buffers (depth/stencil) as long as they're
-    // read-only for the whole render pass.
+    // Backend allows feedback loop with ancillary buffers (depth/stencil) as long as they're read-only for
+    // the whole render pass.
    ALLOW_READ_ONLY_ANCILLARY_FEEDBACK_LOOP,
    // for some uniform arrays, it's needed to do an initialization to avoid crash on adreno gpu
    ADRENO_UNIFORM_ARRAY_CRASH,
@@ -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;
@@ -52,7 +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) {}

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

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

 #include <stdint.h>

@@ -33,7 +33,10 @@ struct PipelineState {
    RasterState rasterState;
    StencilState stencilState;
    PolygonOffset polygonOffset;
-    Viewport scissor{ 0, 0, (uint32_t)INT32_MAX, (uint32_t)INT32_MAX };
+    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>
@@ -142,7 +141,7 @@ public:
            CallbackHandler* handler = nullptr) noexcept {
        return { buffer, size, format, type, alignment, left, top, stride,
                handler, [](void* b, size_t s, void* u) {
-                    (static_cast<T*>(u)->*method)(b, s); }, data };
+                    (*static_cast<T**>(u)->*method)(b, s); }, data };
    }

    template<typename T, void(T::*method)(void const*, size_t)>
@@ -150,7 +149,7 @@ public:
            PixelDataFormat format, PixelDataType type, T* data,
            CallbackHandler* handler = nullptr) noexcept {
        return { buffer, size, format, type, handler, [](void* b, size_t s, void* u) {
-                    (static_cast<T*>(u)->*method)(b, s); }, data };
+                    (*static_cast<T**>(u)->*method)(b, s); }, data };
    }

    template<typename T, void(T::*method)(void const*, size_t)>
@@ -158,7 +157,7 @@ public:
            backend::CompressedPixelDataType format, uint32_t imageSize, T* data,
            CallbackHandler* handler = nullptr) noexcept {
        return { buffer, size, format, imageSize, handler, [](void* b, size_t s, void* u) {
-                    (static_cast<T*>(u)->*method)(b, s); }, data
+                    (*static_cast<T**>(u)->*method)(b, s); }, data
        };
    }

@@ -169,9 +168,9 @@ public:
            CallbackHandler* handler = nullptr) noexcept {
        return { buffer, size, format, type, alignment, left, top, stride,
                handler, [](void* b, size_t s, void* u) {
-                    T* const that = static_cast<T*>(u);
-                    that->operator()(b, s);
-                    delete that;
+                    T& that = *static_cast<T*>(u);
+                    that(b, s);
+                    delete &that;
                }, new T(std::forward<T>(functor))
        };
    }
@@ -182,9 +181,9 @@ public:
            CallbackHandler* handler = nullptr) noexcept {
        return { buffer, size, format, type,
                 handler, [](void* b, size_t s, void* u) {
-                    T* const that = static_cast<T*>(u);
-                    that->operator()(b, s);
-                    delete that;
+                    T& that = *static_cast<T*>(u);
+                    that(b, s);
+                    delete &that;
                }, new T(std::forward<T>(functor))
        };
    }
@@ -195,9 +194,9 @@ public:
            CallbackHandler* handler = nullptr) noexcept {
        return { buffer, size, format, imageSize,
                 handler, [](void* b, size_t s, void* u) {
-                    T* const that = static_cast<T*>(u);
-                    that->operator()(b, s);
-                    delete that;
+                    T& that = *static_cast<T*>(u);
+                    that(b, s);
+                    delete &that;
                }, new T(std::forward<T>(functor))
        };
    }
@@ -19,11 +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>
-
 namespace filament::backend {

 class Driver;
@@ -41,24 +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 backend.
-         */
-        bool disableParallelShaderCompile = false;
    };

    Platform() noexcept;
@@ -84,7 +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;

    /**
@@ -102,8 +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
@@ -111,8 +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
@@ -165,8 +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
@@ -185,8 +176,7 @@ 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);
+    size_t retrieveBlob(const void* key, size_t keySize, void* value, size_t valueSize);

 private:
    InsertBlobFunc mInsertBlob;
@@ -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 {

@@ -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 {
@@ -18,13 +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 <stdint.h>
-
 namespace filament::backend {

 class Driver;
@@ -43,8 +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;

@@ -71,8 +66,7 @@ public:
     * @return              The driver's SwapChain object.
     *
     */
-    virtual SwapChain* UTILS_NONNULL 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.
@@ -93,14 +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().
@@ -113,7 +106,7 @@ public:
     * @return buffer that must be preserved
     * @see commit()
     */
-    virtual TargetBufferFlags getPreservedFlags(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.
@@ -127,16 +120,14 @@ public:
     * @param drawSwapChain SwapChain to draw to. It must be bound to the default FBO.
     * @param readSwapChain SwapChain to read from (for operation like `glBlitFramebuffer`)
     */
-    virtual void makeCurrent(
-            SwapChain* UTILS_NONNULL drawSwapChain,
-            SwapChain* UTILS_NONNULL readSwapChain) noexcept = 0;
+    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.
@@ -161,14 +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.
@@ -178,7 +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;


    // --------------------------------------------------------------------------------------------
@@ -192,13 +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
@@ -208,21 +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;


    // --------------------------------------------------------------------------------------------
@@ -235,13 +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

@@ -254,7 +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.
@@ -268,8 +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 {

@@ -30,7 +30,7 @@ struct PlatformCocoaTouchGLImpl;
 class PlatformCocoaTouchGL : public OpenGLPlatform {
 public:
    PlatformCocoaTouchGL();
-    ~PlatformCocoaTouchGL() noexcept override;
+    ~PlatformCocoaTouchGL() noexcept;

    // --------------------------------------------------------------------------------------------
    // Platform Interface
@@ -17,21 +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 <initializer_list>
+#include <backend/platforms/OpenGLPlatform.h>
+
+#include <backend/DriverEnums.h>
+
 #include <utility>
 #include <vector>

-#include <stddef.h>
-#include <stdint.h>
-
 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;
@@ -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 +17,10 @@
 #ifndef TNT_FILAMENT_BACKEND_PRIVATE_CIRCULARBUFFER_H
 #define TNT_FILAMENT_BACKEND_PRIVATE_CIRCULARBUFFER_H

+#include <utils/compiler.h>
+
 #include <stddef.h>
+#include <stdint.h>

 namespace filament::backend {

@@ -19,14 +19,12 @@

 #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 {

 /*
@@ -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
@@ -154,14 +152,14 @@ 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>

 // ------------------------------------------------------------------------------------------------
@@ -29,7 +29,6 @@

 #include <functional>

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

 // Command debugging off. debugging virtuals are not called.
@@ -300,10 +300,10 @@ DECL_DRIVER_API_SYNCHRONOUS_0(bool, isAutoDepthResolveSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isSRGBSwapChainSupported)
 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(uint8_t, getMaxDrawBuffers)
 DECL_DRIVER_API_SYNCHRONOUS_0(size_t, getMaxUniformBufferSize)
 DECL_DRIVER_API_SYNCHRONOUS_0(math::float2, getClipSpaceParams)
+DECL_DRIVER_API_SYNCHRONOUS_0(bool, canGenerateMipmaps)
 DECL_DRIVER_API_SYNCHRONOUS_N(void, setupExternalImage, void*, image)
 DECL_DRIVER_API_SYNCHRONOUS_N(bool, getTimerQueryValue, backend::TimerQueryHandle, query, uint64_t*, elapsedTime)
 DECL_DRIVER_API_SYNCHRONOUS_N(bool, isWorkaroundNeeded, backend::Workaround, workaround)
@@ -468,7 +468,7 @@ DECL_DRIVER_API_N(readBufferSubData,
 * --------------------
 */

-DECL_DRIVER_API_N(blitDEPRECATED,
+DECL_DRIVER_API_N(blit,
        backend::TargetBufferFlags, buffers,
        backend::RenderTargetHandle, dst,
        backend::Viewport, dstRect,
@@ -476,21 +476,6 @@ DECL_DRIVER_API_N(blitDEPRECATED,
        backend::Viewport, srcRect,
        backend::SamplerMagFilter, filter)

-DECL_DRIVER_API_N(resolve,
-        backend::TextureHandle, dst,
-        uint8_t, dstLevel, uint8_t, dstLayer,
-        backend::TextureHandle, src,
-        uint8_t, srcLevel, uint8_t, srcLayer)
-
-DECL_DRIVER_API_N(blit,
-        backend::TextureHandle, dst,
-        uint8_t, dstLevel, uint8_t, dstLayer,
-        math::uint2, dstOrigin,
-        backend::TextureHandle, src,
-        uint8_t, srcLevel, uint8_t, srcLayer,
-        math::uint2, srcOrigin,
-        math::uint2, size)
-
 DECL_DRIVER_API_N(draw,
        backend::PipelineState, state,
        backend::RenderPrimitiveHandle, rph,
@@ -22,18 +22,9 @@
 #include <utils/Allocator.h>
 #include <utils/Log.h>
 #include <utils/compiler.h>
-#include <utils/debug.h>
-#include <utils/ostream.h>
-
 #include <tsl/robin_map.h>
-
-#include <exception>
-#include <type_traits>
 #include <unordered_map>

-#include <stddef.h>
-#include <stdint.h>
-
 #if !defined(NDEBUG) && UTILS_HAS_RTTI
 #   define HANDLE_TYPE_SAFETY 1
 #else
@@ -231,7 +222,7 @@ private:
        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 extra) 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, 16, extra);
            else if (size <= mPool1.getSize()) p = mPool1.alloc(size, 16, extra);
@@ -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>

@@ -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
    });
 }
@@ -24,7 +24,8 @@
 #include <tsl/robin_map.h>
 #include <utils/Hash.h>

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

 struct MetalContext;

@@ -36,7 +37,8 @@ public:

    struct BlitArgs {
        struct Attachment {
-            id<MTLTexture> texture = nil;
+            id<MTLTexture> color = nil;
+            id<MTLTexture> depth = nil;
            MTLRegion region = {};
            uint8_t level = 0;
            uint32_t slice = 0;      // must be 0 on source attachment
@@ -44,13 +46,25 @@ public:

        // Valid source formats:       2D, 2DArray, 2DMultisample, 3D
        // Valid destination formats:  2D, 2DArray, 3D, Cube
-        Attachment source;
-        Attachment destination;
+        Attachment source, destination;
        SamplerMagFilter filter;

-        bool destinationIsFullAttachment() const {
-            return destination.texture.width == destination.region.size.width &&
-                   destination.texture.height == destination.region.size.height;
+        bool blitColor() const {
+            return source.color != nil && destination.color != nil;
+        }
+
+        bool blitDepth() const {
+            return source.depth != nil && destination.depth != nil;
+        }
+
+        bool colorDestinationIsFullAttachment() const {
+            return destination.color.width == destination.region.size.width &&
+                   destination.color.height == destination.region.size.height;
+        }
+
+        bool depthDestinationIsFullAttachment() const {
+            return destination.depth.width == destination.region.size.width &&
+                   destination.depth.height == destination.region.size.height;
        }
    };

@@ -64,36 +78,48 @@ public:

 private:

-    static void setupAttachment(MTLRenderPassAttachmentDescriptor* descriptor,
-            const BlitArgs& args, uint32_t depthPlane);
+    static void setupColorAttachment(const BlitArgs& args, MTLRenderPassDescriptor* descriptor,
+            uint32_t depthPlane);
+    static void setupDepthAttachment(const BlitArgs& args, MTLRenderPassDescriptor* descriptor,
+            uint32_t depthPlane);

    struct BlitFunctionKey {
-        bool msaaColorSource{};
-        bool sources3D{};
-        char padding[2]{};
+        bool blitColor;
+        bool blitDepth;
+        bool msaaColorSource;
+        bool msaaDepthSource;
+        bool sources3D;
+
+        char padding[3];

        bool isValid() const noexcept {
            // MSAA 3D textures do not exist.
-            bool const hasMsaa = msaaColorSource;
+            bool hasMsaa = msaaColorSource || msaaDepthSource;
            return !(hasMsaa && sources3D);
        }

        bool operator==(const BlitFunctionKey& rhs) const noexcept {
-            return msaaColorSource == rhs.msaaColorSource &&
+            return blitColor == rhs.blitColor &&
+                   blitDepth == rhs.blitDepth &&
+                   msaaColorSource == rhs.msaaColorSource &&
+                   msaaDepthSource == rhs.msaaDepthSource &&
                   sources3D == rhs.sources3D;
        }
+
+        BlitFunctionKey() {
+            std::memset(this, 0, sizeof(BlitFunctionKey));
+        }
    };

-    static bool blitFastPath(id<MTLCommandBuffer> cmdBuffer,
+    void blitFastPath(id<MTLCommandBuffer> cmdBuffer, bool& blitColor, bool& blitDepth,
            const BlitArgs& args, const char* label);
-
-    void blitSlowPath(id<MTLCommandBuffer> cmdBuffer,
+    void blitSlowPath(id<MTLCommandBuffer> cmdBuffer, bool& blitColor, bool& blitDepth,
            const BlitArgs& args, const char* label);
-
-    void blitDepthPlane(id <MTLCommandBuffer> cmdBuffer, const BlitArgs& args,
-            uint32_t depthPlaneSource, uint32_t depthPlaneDest, const char* label);
-
-    id<MTLFunction> compileFragmentFunction(BlitFunctionKey key) const;
+    void blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, bool blitColor, bool blitDepth,
+            const BlitArgs& args, uint32_t depthPlaneSource, uint32_t depthPlaneDest,
+            const char* label);
+    id<MTLTexture> createIntermediateTexture(id<MTLTexture> t, MTLSize size);
+    id<MTLFunction> compileFragmentFunction(BlitFunctionKey key);
    id<MTLFunction> getBlitVertexFunction();
    id<MTLFunction> getBlitFragmentFunction(BlitFunctionKey key);

@@ -104,9 +130,10 @@ private:
    tsl::robin_map<BlitFunctionKey, Function, HashFn> mBlitFunctions;

    id<MTLFunction> mVertexFunction = nil;
+
 };

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

 #endif //TNT_METALBLITTER_H
@@ -20,9 +20,9 @@
 #include "MetalUtils.h"

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

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

 static const char* functionLibrary = R"(
 #include <metal_stdlib>
@@ -37,7 +37,13 @@ struct VertexOut

 struct FragmentOut
 {
+#ifdef BLIT_COLOR
    float4 color [[color(0)]];
+#endif
+
+#ifdef BLIT_DEPTH
+    float depth [[depth(any)]];
+#endif
 };

 vertex VertexOut
@@ -61,6 +67,7 @@ fragment FragmentOut
 blitterFrag(VertexOut in [[stage_in]],
            sampler sourceSampler [[sampler(0)]],

+#ifdef BLIT_COLOR
 #ifdef MSAA_COLOR_SOURCE
            texture2d_ms<float, access::read> sourceColor [[texture(0)]],
 #elif SOURCES_3D
@@ -68,18 +75,32 @@ blitterFrag(VertexOut in [[stage_in]],
 #else
            texture2d<float, access::sample> sourceColor [[texture(0)]],
 #endif  // MSAA_COLOR_SOURCE
+#endif  // BLIT_COLOR
+
+#ifdef BLIT_DEPTH
+#ifdef MSAA_DEPTH_SOURCE
+            texture2d_ms<float, access::read> sourceDepth [[texture(1)]],
+#elif SOURCES_3D
+            texture3d<float, access::sample> sourceDepth [[texture(1)]],
+#else
+            texture2d<float, access::sample> sourceDepth [[texture(1)]],
+#endif  // MSAA_DEPTH_SOURCE
+#endif  // BLIT_DEPTH

            constant FragmentArgs* args [[buffer(0)]])
 {
    FragmentOut out = {};

+#if defined(BLIT_COLOR) || defined(BLIT_DEPTH)
    // These coordinates match the Vulkan vkCmdBlitImage spec:
    // https://www.khronos.org/registry/vulkan/specs/1.2-extensions/man/html/vkCmdBlitImage.html
    float2 uvbase = in.position.xy; // unnormalized coordinates at center of texel: (1.5, 2.5, etc)
    float2 uvoffset = uvbase - args->dstOffset;
    float2 uvscaled = uvoffset * args->scale;
    float2 uv = uvscaled + args->srcOffset;
+#endif

+#ifdef BLIT_COLOR
 #ifdef MSAA_COLOR_SOURCE
    out.color = float4(0.0);
    for (uint s = 0; s < sourceColor.get_num_samples(); s++) {
@@ -95,106 +116,201 @@ blitterFrag(VertexOut in [[stage_in]],
    float2 uvnorm = uv / float2(sourceColor.get_width(args->lod), sourceColor.get_height(args->lod));
    out.color += sourceColor.sample(sourceSampler, uvnorm, level(args->lod));
 #endif  // MSAA_COLOR_SOURCE
+#endif  // BLIT_COLOR

+#ifdef BLIT_DEPTH
+#ifdef MSAA_DEPTH_SOURCE
+    out.depth = 0.0;
+    for (uint s = 0; s < sourceDepth.get_num_samples(); s++) {
+        out.depth += sourceDepth.read(static_cast<uint2>(uv), s).r;
+    }
+    out.depth /= sourceDepth.get_num_samples();
+#elif SOURCES_3D
+    float2 uvnormd = uv / float2(sourceDepth.get_width(args->lod), sourceDepth.get_height(args->lod));
+    float3 coords = float3(uvnormd, (static_cast<float>(args->depthPlane) + 0.5) /
+            sourceDepth.get_depth(args->lod));
+    out.depth = sourceDepth.sample(sourceSampler, coords, level(args->lod)).r;
+#else
+    float2 uvnormd = uv / float2(sourceDepth.get_width(args->lod), sourceDepth.get_height(args->lod));
+    out.depth = sourceDepth.sample(sourceSampler, uvnormd, level(args->lod)).r;
+#endif  // MSAA_DEPTH_SOURCE
+#endif  // BLIT_DEPTH
    return out;
 }
 )";

-
-template<typename T>
-inline bool MTLSizeEqual(T a, T b) noexcept {
-    return (a.width == b.width && a.height == b.height && a.depth == b.depth);
-}
-
 MetalBlitter::MetalBlitter(MetalContext& context) noexcept : mContext(context) { }

+#define MTLSizeEqual(a, b) (a.width == b.width && a.height == b.height && a.depth == b.depth)
+
 void MetalBlitter::blit(id<MTLCommandBuffer> cmdBuffer, const BlitArgs& args, const char* label) {
+    bool blitColor = args.blitColor();
+    bool blitDepth = args.blitDepth();

    ASSERT_PRECONDITION(args.source.region.size.depth == args.destination.region.size.depth,
            "Blitting requires the source and destination regions to have the same depth.");

+    if (args.source.color && args.source.depth) {
+        MTLTextureType colorType = args.source.color.textureType;
+        MTLTextureType depthType = args.source.depth.textureType;
+
+        if (colorType == MTLTextureType2DMultisample) colorType = MTLTextureType2D;
+        if (depthType == MTLTextureType2DMultisample) depthType = MTLTextureType2D;
+
+        ASSERT_PRECONDITION(colorType == depthType,
+                "Blitting requires color and depth sources to be the same texture type.");
+    }
+
    // Determine if the blit for color or depth are eligible to use a MTLBlitCommandEncoder.
-    // blitFastPath returns true upon success.
-    // blitFastPath fails if either the format or sampleCount don't match (i.e. resolve)
-    if (blitFastPath(cmdBuffer, args, label)) {
+    // blitColor and / or blitDepth are set to false upon success, to indicate that no more work is
+    // necessary for that attachment.
+    blitFastPath(cmdBuffer, blitColor, blitDepth, args, label);
+
+    if (!blitColor && !blitDepth) {
        return;
    }

-    // If we end-up here, it means that either:
-    // - we're resolving a color texture
-    // - src/dest formats didn't match, or we're scaling -- this can only happen with the legacy
-    //   blit() path and implies that the format is not depth.
-    // note: in the future we will support a fast-path resolve
+    // If the destination is MSAA and we weren't able to use the fast path, report an error, as
+    // blitting to a MSAA texture isn't supported through the "slow path" yet.
+    const bool colorDestinationIsMultisample =
+            blitColor && args.destination.color.textureType == MTLTextureType2DMultisample;
+    const bool depthDestinationIsMultisample =
+            blitDepth && args.destination.depth.textureType == MTLTextureType2DMultisample;
+    ASSERT_PRECONDITION(!colorDestinationIsMultisample && !depthDestinationIsMultisample,
+        "Blitting between MSAA render targets with differing pixel formats and/or regions is not supported.");

-    UTILS_UNUSED_IN_RELEASE
-    const bool destinationIsMultisample =
-            args.destination.texture.textureType == MTLTextureType2DMultisample;
+    // If the destination texture doesn't have the MTLTextureUsageRenderTarget flag, we have to blit
+    // to an intermediate texture first to perform the format conversion. Then, we can perform a
+    // "fast blit" to the final destination texture.

-    assert_invariant(!destinationIsMultisample);
-    assert_invariant((args.destination.texture.usage & MTLTextureUsageRenderTarget));
-    blitSlowPath(cmdBuffer, args, label);
-}
+    id<MTLTexture> intermediateColor = nil;
+    id<MTLTexture> intermediateDepth = nil;
+    BlitArgs slowBlit = args;
+    BlitArgs finalBlit = args;

-bool MetalBlitter::blitFastPath(id<MTLCommandBuffer> cmdBuffer,
-        const BlitArgs& args, const char* label) {
+    MTLRegion sourceRegionNoOffset = MTLRegionMake3D(0, 0, 0,
+            args.source.region.size.width, args.source.region.size.height,
+            args.source.region.size.depth);

-    if (args.source.texture.sampleCount == args.destination.texture.sampleCount &&
-        args.source.texture.pixelFormat == args.destination.texture.pixelFormat &&
-        MTLSizeEqual(args.source.region.size, args.destination.region.size)) {
-
-        id<MTLBlitCommandEncoder> const blitEncoder = [cmdBuffer blitCommandEncoder];
-        blitEncoder.label = @(label);
-        [blitEncoder copyFromTexture:args.source.texture
-                         sourceSlice:args.source.slice
-                         sourceLevel:args.source.level
-                        sourceOrigin:args.source.region.origin
-                          sourceSize:args.source.region.size
-                           toTexture:args.destination.texture
-                    destinationSlice:args.destination.slice
-                    destinationLevel:args.destination.level
-                   destinationOrigin:args.destination.region.origin];
-        [blitEncoder endEncoding];
-
-        return true;
+    if (blitColor && !(args.destination.color.usage & MTLTextureUsageRenderTarget)) {
+        intermediateColor = createIntermediateTexture(args.destination.color, args.source.region.size);
+        slowBlit.destination.color = finalBlit.source.color = intermediateColor;
+        slowBlit.destination.level = finalBlit.source.level = 0;
+        slowBlit.destination.slice = finalBlit.source.slice = 0;
+        slowBlit.destination.region = finalBlit.source.region = sourceRegionNoOffset;
    }
-    return false;
+    if (blitDepth && !(args.destination.depth.usage & MTLTextureUsageRenderTarget)) {
+        intermediateDepth = createIntermediateTexture(args.destination.depth, args.source.region.size);
+        slowBlit.destination.depth = finalBlit.source.depth = intermediateDepth;
+        slowBlit.destination.level = finalBlit.source.level = 0;
+        slowBlit.destination.slice = finalBlit.source.slice = 0;
+        slowBlit.destination.region = finalBlit.source.region = sourceRegionNoOffset;
+    }
+
+    blitSlowPath(cmdBuffer, blitColor, blitDepth, slowBlit, label);
+
+    bool finalBlitColor = intermediateColor != nil;
+    bool finalBlitDepth = intermediateDepth != nil;
+    blitFastPath(cmdBuffer, finalBlitColor, finalBlitDepth, finalBlit, label);
 }

-void MetalBlitter::blitSlowPath(id<MTLCommandBuffer> cmdBuffer,
+void MetalBlitter::blitFastPath(id<MTLCommandBuffer> cmdBuffer, bool& blitColor, bool& blitDepth,
        const BlitArgs& args, const char* label) {
-    // scaling in any dimension is not allowed
-    assert_invariant(args.source.region.size.depth == args.destination.region.size.depth);
-    // we're always blitting a single plane
+    if (blitColor) {
+        if (args.source.color.sampleCount == args.destination.color.sampleCount &&
+            args.source.color.pixelFormat == args.destination.color.pixelFormat &&
+            MTLSizeEqual(args.source.region.size, args.destination.region.size)) {
+
+            id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
+            blitEncoder.label = @(label);
+            [blitEncoder copyFromTexture:args.source.color
+                             sourceSlice:args.source.slice
+                             sourceLevel:args.source.level
+                            sourceOrigin:args.source.region.origin
+                              sourceSize:args.source.region.size
+                               toTexture:args.destination.color
+                        destinationSlice:args.destination.slice
+                        destinationLevel:args.destination.level
+                       destinationOrigin:args.destination.region.origin];
+            [blitEncoder endEncoding];
+
+            blitColor = false;
+        }
+    }
+
+    if (blitDepth) {
+        if (args.source.depth.sampleCount == args.destination.depth.sampleCount &&
+            args.source.depth.pixelFormat == args.destination.depth.pixelFormat &&
+            MTLSizeEqual(args.source.region.size, args.destination.region.size)) {
+
+            id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
+            blitEncoder.label = @(label);
+            [blitEncoder copyFromTexture:args.source.depth
+                             sourceSlice:args.source.slice
+                             sourceLevel:args.source.level
+                            sourceOrigin:args.source.region.origin
+                              sourceSize:args.source.region.size
+                               toTexture:args.destination.depth
+                        destinationSlice:args.destination.slice
+                        destinationLevel:args.destination.level
+                       destinationOrigin:args.destination.region.origin];
+            [blitEncoder endEncoding];
+
+            blitDepth = false;
+        }
+    }
+}
+
+void MetalBlitter::blitSlowPath(id<MTLCommandBuffer> cmdBuffer, bool& blitColor, bool& blitDepth,
+        const BlitArgs& args, const char* label) {
+
    uint32_t depthPlaneSource = args.source.region.origin.z;
    uint32_t depthPlaneDest = args.destination.region.origin.z;
-    for (NSUInteger d = 0; d < args.source.region.size.depth; d++) {
-        blitDepthPlane(cmdBuffer, args,
-                depthPlaneSource++, depthPlaneDest++, label);
+
+    assert_invariant(args.source.region.size.depth == args.destination.region.size.depth);
+    uint32_t depthPlaneCount = args.source.region.size.depth;
+    for (NSUInteger d = 0; d < depthPlaneCount; d++) {
+        blitDepthPlane(cmdBuffer, blitColor, blitDepth, args, depthPlaneSource++, depthPlaneDest++,
+                label);
    }
+
+    blitColor = false;
+    blitDepth = false;
 }

-void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, const BlitArgs& args,
-        uint32_t depthPlaneSource, uint32_t depthPlaneDest, const char* label) {
+void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, bool blitColor, bool blitDepth,
+        const BlitArgs& args, uint32_t depthPlaneSource, uint32_t depthPlaneDest,
+        const char* label) {
+    MTLRenderPassDescriptor* descriptor = [MTLRenderPassDescriptor renderPassDescriptor];

-    MTLRenderPassDescriptor* const descriptor = [MTLRenderPassDescriptor renderPassDescriptor];
+    if (blitColor) {
+        setupColorAttachment(args, descriptor, depthPlaneDest);
+    }

-    setupAttachment(descriptor.colorAttachments[0], args, depthPlaneDest);
+    if (blitDepth) {
+        setupDepthAttachment(args, descriptor, depthPlaneDest);
+    }

-    id<MTLRenderCommandEncoder> const encoder =
-            [cmdBuffer renderCommandEncoderWithDescriptor:descriptor];
+    id<MTLRenderCommandEncoder> encoder = [cmdBuffer renderCommandEncoderWithDescriptor:descriptor];
    encoder.label = @(label);

    BlitFunctionKey key;
-    key.msaaColorSource = args.source.texture.textureType == MTLTextureType2DMultisample;
-    key.sources3D       = args.source.texture.textureType == MTLTextureType3D;
-    id<MTLFunction> const fragmentFunction = getBlitFragmentFunction(key);
+    key.blitColor = blitColor;
+    key.blitDepth = blitDepth;
+    key.msaaColorSource = blitColor && args.source.color.textureType == MTLTextureType2DMultisample;
+    key.msaaDepthSource = blitDepth && args.source.depth.textureType == MTLTextureType2DMultisample;
+    key.sources3D       = blitColor && args.source.color.textureType == MTLTextureType3D;
+    if (key.sources3D && blitDepth) {
+        assert_invariant(args.source.depth.textureType == MTLTextureType3D);
+    }
+    id<MTLFunction> fragmentFunction = getBlitFragmentFunction(key);

-    MetalPipelineState const pipelineState {
+    MetalPipelineState pipelineState {
        .vertexFunction = getBlitVertexFunction(),
        .fragmentFunction = fragmentFunction,
        .vertexDescription = {},
        .colorAttachmentPixelFormat = {
-            args.destination.texture.pixelFormat,
+            blitColor ? args.destination.color.pixelFormat : MTLPixelFormatInvalid,
            MTLPixelFormatInvalid,
            MTLPixelFormatInvalid,
            MTLPixelFormatInvalid,
@@ -203,21 +319,31 @@ void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, const BlitArgs
            MTLPixelFormatInvalid,
            MTLPixelFormatInvalid
        },
-        .depthAttachmentPixelFormat = MTLPixelFormatInvalid,
+        .depthAttachmentPixelFormat =
+                blitDepth ? args.destination.depth.pixelFormat : MTLPixelFormatInvalid,
        .sampleCount = 1,
        .blendState = {}
    };
-    id<MTLRenderPipelineState> const pipeline =
-            mContext.pipelineStateCache.getOrCreateState(pipelineState);
+    id<MTLRenderPipelineState> pipeline = mContext.pipelineStateCache.getOrCreateState(pipelineState);
    [encoder setRenderPipelineState:pipeline];

    // For texture arrays, create a view of the texture at the given slice (layer).
-
-    id<MTLTexture> srcTextureColor = args.source.texture;
+    id<MTLTexture> srcTextureColor = args.source.color;
    if (srcTextureColor && srcTextureColor.textureType == MTLTextureType2DArray) {
        srcTextureColor = createTextureViewWithSingleSlice(srcTextureColor, args.source.slice);
    }
-    [encoder setFragmentTexture:srcTextureColor atIndex:0];
+    id<MTLTexture> srcTextureDepth = args.source.depth;
+    if (srcTextureDepth && srcTextureDepth.textureType == MTLTextureType2DArray) {
+        srcTextureDepth = createTextureViewWithSingleSlice(srcTextureDepth, args.source.slice);
+    }
+
+    if (blitColor) {
+        [encoder setFragmentTexture:srcTextureColor atIndex:0];
+    }
+
+    if (blitDepth) {
+        [encoder setFragmentTexture:srcTextureDepth atIndex:1];
+    }

    SamplerMinFilter filterMin = SamplerMinFilter::NEAREST_MIPMAP_NEAREST;
    if (args.filter == SamplerMagFilter::NEAREST) {
@@ -226,13 +352,13 @@ void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, const BlitArgs
        filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST;
    }

-    SamplerState const s {
+    SamplerState s {
        .samplerParams = {
            .filterMag = args.filter,
            .filterMin = filterMin
        }
    };
-    id<MTLSamplerState> const sampler = mContext.samplerStateCache.getOrCreateState(s);
+    id<MTLSamplerState> sampler = mContext.samplerStateCache.getOrCreateState(s);
    [encoder setFragmentSamplerState:sampler atIndex:0];

    MTLViewport viewport;
@@ -244,11 +370,11 @@ void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, const BlitArgs
    viewport.zfar = 1.0;
    [encoder setViewport:viewport];

-    DepthStencilState const depthStencilState {
+    DepthStencilState depthStencilState {
        .depthCompare = MTLCompareFunctionAlways,
-        .depthWriteEnabled = false
+        .depthWriteEnabled = blitDepth
    };
-    id<MTLDepthStencilState> const depthStencil =
+    id<MTLDepthStencilState> depthStencil =
            mContext.depthStencilStateCache.getOrCreateState(depthStencilState);
    [encoder setDepthStencilState:depthStencil];

@@ -284,42 +410,80 @@ void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, const BlitArgs
    [encoder endEncoding];
 }

+id<MTLTexture> MetalBlitter::createIntermediateTexture(id<MTLTexture> t, MTLSize size) {
+    MTLTextureDescriptor* descriptor = [MTLTextureDescriptor new];
+    descriptor.textureType = size.depth == 1 ? MTLTextureType2D : MTLTextureType3D;
+    descriptor.pixelFormat = t.pixelFormat;
+    descriptor.width = size.width;
+    descriptor.height = size.height;
+    descriptor.depth = size.depth;
+    descriptor.usage = t.usage & MTLTextureUsageRenderTarget;
+    return [mContext.device newTextureWithDescriptor:descriptor];
+}
+
 void MetalBlitter::shutdown() noexcept {
    mBlitFunctions.clear();
    mVertexFunction = nil;
 }

-void MetalBlitter::setupAttachment(MTLRenderPassAttachmentDescriptor* descriptor,
-        const BlitArgs& args, uint32_t depthPlane) {
-    descriptor.texture = args.destination.texture;
-    descriptor.level = args.destination.level;
-    descriptor.slice = args.destination.slice;
-    descriptor.depthPlane = depthPlane;
-    descriptor.loadAction = MTLLoadActionLoad;
+void MetalBlitter::setupColorAttachment(const BlitArgs& args,
+        MTLRenderPassDescriptor* descriptor, uint32_t depthPlane) {
+    descriptor.colorAttachments[0].texture = args.destination.color;
+    descriptor.colorAttachments[0].level = args.destination.level;
+    descriptor.colorAttachments[0].slice = args.destination.slice;
+    descriptor.colorAttachments[0].depthPlane = depthPlane;
+
+    descriptor.colorAttachments[0].loadAction = MTLLoadActionLoad;
    // We don't need to load the contents of the attachment if we're blitting over all of it.
-    if (args.destinationIsFullAttachment()) {
-        descriptor.loadAction = MTLLoadActionDontCare;
+    if (args.colorDestinationIsFullAttachment()) {
+        descriptor.colorAttachments[0].loadAction = MTLLoadActionDontCare;
    }
-    descriptor.storeAction = MTLStoreActionStore;
+
+    descriptor.colorAttachments[0].storeAction = MTLStoreActionStore;
 }

-id<MTLFunction> MetalBlitter::compileFragmentFunction(BlitFunctionKey key) const {
-    MTLCompileOptions* const options = [MTLCompileOptions new];
-    NSMutableDictionary* const macros = [NSMutableDictionary dictionary];
+void MetalBlitter::setupDepthAttachment(const BlitArgs& args, MTLRenderPassDescriptor* descriptor,
+        uint32_t depthPlane) {
+    descriptor.depthAttachment.texture = args.destination.depth;
+    descriptor.depthAttachment.level = args.destination.level;
+    descriptor.depthAttachment.slice = args.destination.slice;
+    descriptor.depthAttachment.depthPlane = depthPlane;
+
+    descriptor.depthAttachment.loadAction = MTLLoadActionLoad;
+    // We don't need to load the contents of the attachment if we're blitting over all of it.
+    if (args.depthDestinationIsFullAttachment()) {
+        descriptor.depthAttachment.loadAction = MTLLoadActionDontCare;
+    }
+
+    descriptor.depthAttachment.storeAction = MTLStoreActionStore;
+}
+
+id<MTLFunction> MetalBlitter::compileFragmentFunction(BlitFunctionKey key) {
+    MTLCompileOptions* options = [MTLCompileOptions new];
+    NSMutableDictionary* macros = [NSMutableDictionary dictionary];
+    if (key.blitColor) {
+        macros[@"BLIT_COLOR"] = @"1";
+    }
+    if (key.blitDepth) {
+        macros[@"BLIT_DEPTH"] = @"1";
+    }
    if (key.msaaColorSource) {
        macros[@"MSAA_COLOR_SOURCE"] = @"1";
    }
+    if (key.msaaDepthSource) {
+        macros[@"MSAA_DEPTH_SOURCE"] = @"1";
+    }
    if (key.sources3D) {
        macros[@"SOURCES_3D"] = @"1";
    }
    options.preprocessorMacros = macros;
-    NSString* const objcSource = [NSString stringWithCString:functionLibrary
-                                                    encoding:NSUTF8StringEncoding];
+    NSString* objcSource = [NSString stringWithCString:functionLibrary
+                                              encoding:NSUTF8StringEncoding];
    NSError* error = nil;
-    id <MTLLibrary> const library = [mContext.device newLibraryWithSource:objcSource
-                                                                  options:options
-                                                                    error:&error];
-    id <MTLFunction> const function = [library newFunctionWithName:@"blitterFrag"];
+    id<MTLLibrary> library = [mContext.device newLibraryWithSource:objcSource
+                                                            options:options
+                                                              error:&error];
+    id<MTLFunction> function = [library newFunctionWithName:@"blitterFrag"];

    if (!library || !function) {
        if (error) {
@@ -337,14 +501,13 @@ id<MTLFunction> MetalBlitter::getBlitVertexFunction() {
        return mVertexFunction;
    }

-    NSString* const objcSource = [NSString stringWithCString:functionLibrary
-                                                    encoding:NSUTF8StringEncoding];
+    NSString* objcSource = [NSString stringWithCString:functionLibrary
+                                              encoding:NSUTF8StringEncoding];
    NSError* error = nil;
-    id <MTLLibrary> const library = [mContext.device newLibraryWithSource:objcSource
-                                                                  options:nil
-                                                                    error:&error];
-
-    id<MTLFunction> const function = [library newFunctionWithName:@"blitterVertex"];
+    id<MTLLibrary> library = [mContext.device newLibraryWithSource:objcSource
+                                                           options:nil
+                                                             error:&error];
+    id<MTLFunction> function = [library newFunctionWithName:@"blitterVertex"];

    if (!library || !function) {
        if (error) {
@@ -372,5 +535,5 @@ id<MTLFunction> MetalBlitter::getBlitFragmentFunction(BlitFunctionKey key) {
    return function;
 }

-} // namespace filament::backend
-
+} // namespace backend
+} // namespace filament
@@ -73,10 +73,7 @@ private:
    std::unordered_set<MetalBufferPoolEntry const*> mUsedStages;

    // Store the current "time" (really just a frame count) and LRU eviction parameters.
-    // An atomic is necessary as mCurrentFrame is incremented in gc() (called on
-    // the driver thread) and read from acquireBuffer() and releaseBuffer(),
-    // which may be called on non-driver threads.
-    std::atomic<uint64_t> mCurrentFrame = 0;
+    uint64_t mCurrentFrame = 0;
    static constexpr uint32_t TIME_BEFORE_EVICTION = 10;
 };

@@ -19,7 +19,6 @@
 #include "MetalContext.h"

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

 #include <thread>
@@ -142,10 +142,7 @@ struct MetalContext {
    // Fences, only supported on macOS 10.14 and iOS 12 and above.
    API_AVAILABLE(macos(10.14), ios(12.0))
    MTLSharedEventListener* eventListener = nil;
-    // signalId is incremented in the MetalFence constructor, which is called on
-    // both the driver (MetalTimerQueryFence::beginTimeElapsedQuery) and main
-    // threads (in createFenceS), so an atomic is necessary.
-    std::atomic<uint64_t> signalId = 1;
+    uint64_t signalId = 1;

    MetalTimerQueryInterface* timerQueryImpl;

@@ -28,10 +28,6 @@
 #include <utils/Log.h>
 #include <utils/debug.h>

-#include <functional>
-#include <mutex>
-#include <vector>
-
 namespace filament {
 namespace backend {

@@ -56,13 +52,13 @@ class MetalDriver final : public DriverBase {

 public:
    static Driver* create(MetalPlatform* platform, const Platform::DriverConfig& driverConfig);
-    void runAtNextTick(const std::function<void()>& fn) noexcept;

 private:

    friend class MetalSwapChain;

    MetalPlatform& mPlatform;
+
    MetalContext* mContext;

    ShaderModel getShaderModel() const noexcept final;
@@ -70,13 +66,6 @@ private:
    // Overrides the default implementation by wrapping the call to fn in an @autoreleasepool block.
    void execute(std::function<void(void)> const& fn) noexcept final;

-    /*
-     * Tasks run regularly on the driver thread.
-     */
-    void executeTickOps() noexcept;
-    std::vector<std::function<void()>> mTickOps;
-    std::mutex mTickOpsLock;
-
    /*
     * Driver interface
     */
@@ -166,7 +166,6 @@ MetalDriver::~MetalDriver() noexcept {
 }

 void MetalDriver::tick(int) {
-    executeTickOps();
 }

 void MetalDriver::beginFrame(int64_t monotonic_clock_ns, uint32_t frameId) {
@@ -584,8 +583,6 @@ void MetalDriver::terminate() {
    // This must be done before calling bufferPool->reset() to ensure no buffers are in flight.
    finish();

-    executeTickOps();
-
    mContext->bufferPool->reset();
    mContext->commandQueue = nil;

@@ -730,10 +727,6 @@ bool MetalDriver::isParallelShaderCompileSupported() {
    return true;
 }

-bool MetalDriver::isDepthStencilResolveSupported() {
-    return false;
-}
-
 bool MetalDriver::isWorkaroundNeeded(Workaround workaround) {
    switch (workaround) {
        case Workaround::SPLIT_EASU:
@@ -872,6 +865,10 @@ void MetalDriver::generateMipmaps(Handle<HwTexture> th) {
    tex->generateMipmaps();
 }

+bool MetalDriver::canGenerateMipmaps() {
+    return true;
+}
+
 void MetalDriver::updateSamplerGroup(Handle<HwSamplerGroup> sbh, BufferDescriptor&& data) {
    ASSERT_PRECONDITION(!isInRenderPass(mContext),
            "updateSamplerGroup must be called outside of a render pass.");
@@ -1259,14 +1256,14 @@ void MetalDriver::readPixels(Handle<HwRenderTarget> src, uint32_t x, uint32_t y,
    textureDescriptor.usage = MTLTextureUsageShaderWrite | MTLTextureUsageRenderTarget;
    id<MTLTexture> readPixelsTexture = [mContext->device newTextureWithDescriptor:textureDescriptor];

-    MetalBlitter::BlitArgs args{};
+    MetalBlitter::BlitArgs args;
    args.filter = SamplerMagFilter::NEAREST;
    args.source.level = miplevel;
    args.source.region = MTLRegionMake2D(0, 0, srcTexture.width >> miplevel, srcTexture.height >> miplevel);
-    args.source.texture = srcTexture;
    args.destination.level = 0;
    args.destination.region = MTLRegionMake2D(0, 0, readPixelsTexture.width, readPixelsTexture.height);
-    args.destination.texture = readPixelsTexture;
+    args.source.color = srcTexture;
+    args.destination.color = readPixelsTexture;

    mContext->blitter->blit(getPendingCommandBuffer(mContext), args, "readPixels blit");

@@ -1302,123 +1299,24 @@ void MetalDriver::readBufferSubData(backend::BufferObjectHandle boh,
    scheduleDestroy(std::move(p));
 }

-void MetalDriver::resolve(
-        Handle<HwTexture> dst, uint8_t srcLevel, uint8_t srcLayer,
-        Handle<HwTexture> src, uint8_t dstLevel, uint8_t dstLayer) {
-    auto* const srcTexture = handle_cast<MetalTexture>(src);
-    auto* const dstTexture = handle_cast<MetalTexture>(dst);
-    assert_invariant(srcTexture);
-    assert_invariant(dstTexture);
-
-    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder == nil,
-            "resolve() cannot be invoked inside a render pass.");
-
-    ASSERT_PRECONDITION(
-            dstTexture->width == srcTexture->width && dstTexture->height == srcTexture->height,
-            "invalid resolve: src and dst sizes don't match");
-
-    ASSERT_PRECONDITION(srcTexture->samples > 1 && dstTexture->samples == 1,
-            "invalid resolve: src.samples=%u, dst.samples=%u",
-            +srcTexture->samples, +dstTexture->samples);
-
-    ASSERT_PRECONDITION(srcTexture->format == dstTexture->format,
-            "src and dst texture format don't match");
-
-    ASSERT_PRECONDITION(!isDepthFormat(srcTexture->format),
-            "can't resolve depth formats");
-
-    ASSERT_PRECONDITION(!isStencilFormat(srcTexture->format),
-            "can't resolve stencil formats");
-
-    ASSERT_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST),
-            "texture doesn't have BLIT_DST");
-
-    ASSERT_PRECONDITION(any(srcTexture->usage & TextureUsage::BLIT_SRC),
-            "texture doesn't have BLIT_SRC");
-
-    // FIXME: on metal the blit() call below always take the slow path (using a shader)
-
-    blit(   dst, dstLevel, dstLayer, {},
-            src, srcLevel, srcLayer, {},
-            { dstTexture->width, dstTexture->height });
-}
-
-void MetalDriver::blit(
-        Handle<HwTexture> dst, uint8_t srcLevel, uint8_t srcLayer, math::uint2 dstOrigin,
-        Handle<HwTexture> src, uint8_t dstLevel, uint8_t dstLayer, math::uint2 srcOrigin,
-        math::uint2 size) {
-
-
-    auto isBlitableTextureType = [](MTLTextureType t) -> bool {
-        return t == MTLTextureType2D || t == MTLTextureType2DMultisample ||
-               t == MTLTextureType2DArray;
-    };
-
-    auto* const srcTexture = handle_cast<MetalTexture>(src);
-    auto* const dstTexture = handle_cast<MetalTexture>(dst);
-    assert_invariant(srcTexture);
-    assert_invariant(dstTexture);
-
-    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder == nil,
-            "blit() cannot be invoked inside a render pass.");
-
-    ASSERT_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST),
-            "texture doesn't have BLIT_DST");
-
-    ASSERT_PRECONDITION(any(srcTexture->usage & TextureUsage::BLIT_SRC),
-            "texture doesn't have BLIT_SRC");
-
-    ASSERT_PRECONDITION(srcTexture->format == dstTexture->format,
-            "src and dst texture format don't match");
-
-    ASSERT_PRECONDITION(isBlitableTextureType(srcTexture->getMtlTextureForRead().textureType) &&
-                        isBlitableTextureType(dstTexture->getMtlTextureForWrite().textureType),
-            "Metal does not support blitting to/from non-2D textures.");
-
-    MetalBlitter::BlitArgs args{};
-    args.filter = SamplerMagFilter::NEAREST;
-    args.source.region = MTLRegionMake2D(
-            (NSUInteger)srcOrigin.x,
-            std::max(srcTexture->height - (int64_t)srcOrigin.y - size.y, (int64_t)0),
-            size.x,  size.y);
-    args.destination.region = MTLRegionMake2D(
-            (NSUInteger)dstOrigin.x,
-            std::max(dstTexture->height - (int64_t)dstOrigin.y - size.y, (int64_t)0),
-            size.x,  size.y);
-
-    // FIXME: we shouldn't need to know the type here. This is an artifact of using the old API.
-
-    args.source.texture = srcTexture->getMtlTextureForRead();
-    args.destination.texture = dstTexture->getMtlTextureForWrite();
-    args.source.level = srcLevel;
-    args.source.slice = srcLayer;
-    args.destination.level = dstLevel;
-    args.destination.slice = dstLayer;
-
-    // TODO: The blit() call below always take the fast path.
-
-    mContext->blitter->blit(getPendingCommandBuffer(mContext), args, "blit/resolve");
-
-    dstTexture->extendLodRangeTo(dstLevel);
-}
-
-void MetalDriver::blitDEPRECATED(TargetBufferFlags buffers,
+void MetalDriver::blit(TargetBufferFlags buffers,
        Handle<HwRenderTarget> dst, Viewport dstRect,
        Handle<HwRenderTarget> src, Viewport srcRect,
        SamplerMagFilter filter) {
-
-    // Note: blitDEPRECATED is only used by Renderer::copyFrame
-    // It is called between beginFrame and endFrame, but should never be called in the middle of
-    // a render pass.
-
-    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder == nil,
-            "blitDEPRECATED() cannot be invoked inside a render pass.");
+    // If we're the in middle of a render pass, finish it.
+    // This condition should only occur during copyFrame. It's okay to end the render pass because
+    // we don't issue any other rendering commands.
+    if (mContext->currentRenderPassEncoder) {
+        [mContext->currentRenderPassEncoder endEncoding];
+        mContext->currentRenderPassEncoder = nil;
+    }

    auto srcTarget = handle_cast<MetalRenderTarget>(src);
    auto dstTarget = handle_cast<MetalRenderTarget>(dst);

-    ASSERT_PRECONDITION(buffers == TargetBufferFlags::COLOR0,
-            "blitDEPRECATED only supports COLOR0");
+    ASSERT_PRECONDITION(
+            !(buffers & (TargetBufferFlags::COLOR_ALL & ~TargetBufferFlags::COLOR0)),
+            "Blitting only supports COLOR0");

    ASSERT_PRECONDITION(srcRect.left >= 0 && srcRect.bottom >= 0 &&
                        dstRect.left >= 0 && dstRect.bottom >= 0,
@@ -1429,28 +1327,54 @@ void MetalDriver::blitDEPRECATED(TargetBufferFlags buffers,
               t == MTLTextureType2DArray;
    };

-    // We always blit from/to the COLOR0 attachment.
-    MetalRenderTarget::Attachment const srcColorAttachment = srcTarget->getReadColorAttachment(0);
-    MetalRenderTarget::Attachment const dstColorAttachment = dstTarget->getDrawColorAttachment(0);
+    // MetalBlitter supports blitting color and depth simultaneously, but for simplicitly we'll blit
+    // them separately. In practice, Filament only ever blits a single buffer at a time anyway.

-    if (srcColorAttachment && dstColorAttachment) {
-        ASSERT_PRECONDITION(isBlitableTextureType(srcColorAttachment.getTexture().textureType) &&
-                            isBlitableTextureType(dstColorAttachment.getTexture().textureType),
-                           "Metal does not support blitting to/from non-2D textures.");
+    if (any(buffers & TargetBufferFlags::COLOR_ALL)) {
+        // We always blit from/to the COLOR0 attachment.
+        MetalRenderTarget::Attachment srcColorAttachment = srcTarget->getReadColorAttachment(0);
+        MetalRenderTarget::Attachment dstColorAttachment = dstTarget->getDrawColorAttachment(0);

-        MetalBlitter::BlitArgs args{};
-        args.filter = filter;
-        args.source.region = srcTarget->getRegionFromClientRect(srcRect);
-        args.source.texture = srcColorAttachment.getTexture();
-        args.source.level = srcColorAttachment.level;
-        args.source.slice = srcColorAttachment.layer;
+        if (srcColorAttachment && dstColorAttachment) {
+            ASSERT_PRECONDITION(isBlitableTextureType(srcColorAttachment.getTexture().textureType) &&
+                                isBlitableTextureType(dstColorAttachment.getTexture().textureType),
+                               "Metal does not support blitting to/from non-2D textures.");

-        args.destination.region = dstTarget->getRegionFromClientRect(dstRect);
-        args.destination.texture = dstColorAttachment.getTexture();
-        args.destination.level = dstColorAttachment.level;
-        args.destination.slice = dstColorAttachment.layer;
+            MetalBlitter::BlitArgs args;
+            args.filter = filter;
+            args.source.region = srcTarget->getRegionFromClientRect(srcRect);
+            args.destination.region = dstTarget->getRegionFromClientRect(dstRect);
+            args.source.color = srcColorAttachment.getTexture();
+            args.destination.color = dstColorAttachment.getTexture();
+            args.source.level = srcColorAttachment.level;
+            args.destination.level = dstColorAttachment.level;
+            args.source.slice = srcColorAttachment.layer;
+            args.destination.slice = dstColorAttachment.layer;
+            mContext->blitter->blit(getPendingCommandBuffer(mContext), args, "Color blit");
+        }
+    }

-        mContext->blitter->blit(getPendingCommandBuffer(mContext), args, "blitDEPRECATED");
+    if (any(buffers & TargetBufferFlags::DEPTH)) {
+        MetalRenderTarget::Attachment srcDepthAttachment = srcTarget->getDepthAttachment();
+        MetalRenderTarget::Attachment dstDepthAttachment = dstTarget->getDepthAttachment();
+
+        if (srcDepthAttachment && dstDepthAttachment) {
+            ASSERT_PRECONDITION(isBlitableTextureType(srcDepthAttachment.getTexture().textureType) &&
+                                isBlitableTextureType(dstDepthAttachment.getTexture().textureType),
+                               "Metal does not support blitting to/from non-2D textures.");
+
+            MetalBlitter::BlitArgs args;
+            args.filter = filter;
+            args.source.region = srcTarget->getRegionFromClientRect(srcRect);
+            args.destination.region = dstTarget->getRegionFromClientRect(dstRect);
+            args.source.depth = srcDepthAttachment.getTexture();
+            args.destination.depth = dstDepthAttachment.getTexture();
+            args.source.level = srcDepthAttachment.level;
+            args.destination.level = dstDepthAttachment.level;
+            args.source.slice = srcDepthAttachment.layer;
+            args.destination.slice = dstDepthAttachment.layer;
+            mContext->blitter->blit(getPendingCommandBuffer(mContext), args, "Depth blit");
+        }
    }
 }

@@ -1588,7 +1512,7 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph, uint32_t
        stencilPixelFormat = stencilAttachment.getPixelFormat();
        assert_invariant(isMetalFormatStencil(stencilPixelFormat));
    }
-    MetalPipelineState const pipelineState {
+    MetalPipelineState pipelineState {
        .vertexFunction = vertex,
        .fragmentFunction = fragment,
        .vertexDescription = primitive->vertexDescription,
@@ -1606,13 +1530,13 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph, uint32_t
        .stencilAttachmentPixelFormat = stencilPixelFormat,
        .sampleCount = mContext->currentRenderTarget->getSamples(),
        .blendState = BlendState {
-            .alphaBlendOperation = getMetalBlendOperation(rs.blendEquationAlpha),
-            .rgbBlendOperation = getMetalBlendOperation(rs.blendEquationRGB),
-            .destinationAlphaBlendFactor = getMetalBlendFactor(rs.blendFunctionDstAlpha),
-            .destinationRGBBlendFactor = getMetalBlendFactor(rs.blendFunctionDstRGB),
-            .sourceAlphaBlendFactor = getMetalBlendFactor(rs.blendFunctionSrcAlpha),
-            .sourceRGBBlendFactor = getMetalBlendFactor(rs.blendFunctionSrcRGB),
            .blendingEnabled = rs.hasBlending(),
+            .rgbBlendOperation = getMetalBlendOperation(rs.blendEquationRGB),
+            .alphaBlendOperation = getMetalBlendOperation(rs.blendEquationAlpha),
+            .sourceRGBBlendFactor = getMetalBlendFactor(rs.blendFunctionSrcRGB),
+            .sourceAlphaBlendFactor = getMetalBlendFactor(rs.blendFunctionSrcAlpha),
+            .destinationRGBBlendFactor = getMetalBlendFactor(rs.blendFunctionDstRGB),
+            .destinationAlphaBlendFactor = getMetalBlendFactor(rs.blendFunctionDstAlpha)
        },
        .colorWrite = rs.colorWrite
    };
@@ -1911,21 +1835,6 @@ void MetalDriver::enumerateBoundBuffers(BufferObjectBinding bindingType,
 void MetalDriver::resetState(int) {
 }

-void MetalDriver::runAtNextTick(const std::function<void()>& fn) noexcept {
-    std::lock_guard<std::mutex> const lock(mTickOpsLock);
-    mTickOps.push_back(fn);
-}
-
-void MetalDriver::executeTickOps() noexcept {
-    std::vector<std::function<void()>> ops;
-    mTickOpsLock.lock();
-    std::swap(ops, mTickOps);
-    mTickOpsLock.unlock();
-    for (const auto& f : ops) {
-        f();
-    }
-}
-
 // explicit instantiation of the Dispatcher
 template class ConcreteDispatcher<MetalDriver>;

@@ -21,7 +21,6 @@
 #include "MetalUtils.h"

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

 #define NSERROR_CHECK(message)                                                                     \
@@ -39,15 +39,7 @@ namespace filament {
 namespace backend {

 static inline MTLTextureUsage getMetalTextureUsage(TextureUsage usage) {
-    NSUInteger u = MTLTextureUsageUnknown;
-
-    if (any(usage & TextureUsage::SAMPLEABLE)) {
-        u |= MTLTextureUsageShaderRead;
-    }
-    if (any(usage & TextureUsage::UPLOADABLE)) {
-        // This is only needed because of the slowpath is MetalBlitter
-        u |= MTLTextureUsageRenderTarget;
-    }
+    NSUInteger u = 0;
    if (any(usage & TextureUsage::COLOR_ATTACHMENT)) {
        u |= MTLTextureUsageRenderTarget;
    }
@@ -57,13 +49,9 @@ static inline MTLTextureUsage getMetalTextureUsage(TextureUsage usage) {
    if (any(usage & TextureUsage::STENCIL_ATTACHMENT)) {
        u |= MTLTextureUsageRenderTarget;
    }
-    if (any(usage & TextureUsage::BLIT_DST)) {
-        // This is only needed because of the slowpath is MetalBlitter
-        u |= MTLTextureUsageRenderTarget;
-    }
-    if (any(usage & TextureUsage::BLIT_SRC)) {
-        u |= MTLTextureUsageShaderRead;
-    }
+
+    // All textures can be blitted from, so they must have the UsageShaderRead flag.
+    u |= MTLTextureUsageShaderRead;

    return MTLTextureUsage(u);
 }
@@ -245,53 +233,13 @@ 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;
-};
-
 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 of acquireDrawable.
+    id<CAMetalDrawable> drawable = (id<CAMetalDrawable>) CFBridgingRelease(user);
+    if (presentFrame) {
+        [drawable present];
+    }
+    // The drawable will be released here when the "drawable" variable goes out of scope.
 }

 void MetalSwapChain::scheduleFrameScheduledCallback() {
@@ -300,16 +248,19 @@ 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;
-
    [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*.
+        PresentCallable callable(presentDrawable, (void*) CFBridgingRetain(d));
+        dispatch_async(dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT, 0), ^{
+            callback(callable, userData);
+        });
    }];
 }

@@ -805,7 +756,6 @@ void MetalTexture::loadWithBlit(uint32_t level, uint32_t slice, MTLRegion region
 #endif

    id<MTLTexture> stagingTexture = [context.device newTextureWithDescriptor:descriptor];
-    // FIXME? Why is this not just `MTLRegion sourceRegion = region;` ?
    MTLRegion sourceRegion = MTLRegionMake3D(0, 0, 0,
            region.size.width, region.size.height, region.size.depth);
    [stagingTexture replaceRegion:sourceRegion
@@ -832,16 +782,16 @@ void MetalTexture::loadWithBlit(uint32_t level, uint32_t slice, MTLRegion region
                                                             slices:NSMakeRange(0, slices)];
    }

-    MetalBlitter::BlitArgs args{};
+    MetalBlitter::BlitArgs args;
    args.filter = SamplerMagFilter::NEAREST;
    args.source.level = 0;
    args.source.slice = 0;
    args.source.region = sourceRegion;
-    args.source.texture = stagingTexture;
    args.destination.level = level;
    args.destination.slice = slice;
    args.destination.region = region;
-    args.destination.texture = destinationTexture;
+    args.source.color = stagingTexture;
+    args.destination.color = destinationTexture;
    context.blitter->blit(getPendingCommandBuffer(&context), args, "Texture upload blit");
 }

@@ -103,19 +103,10 @@ void MetalShaderCompiler::terminate() noexcept {
        NSString* objcSource = [[NSString alloc] initWithBytes:source.data()
                                                        length:source.size() - 1
                                                      encoding:NSUTF8StringEncoding];
-
-        // By default, Metal uses the most recent language version.
-        MTLCompileOptions* options = [MTLCompileOptions new];
-
-        // Disable Fast Math optimizations.
-        // This ensures that operations adhere to IEEE standards for floating-point arithmetic,
-        // which is crucial for half precision floats in scenarios where fast math optimizations
-        // lead to inaccuracies, such as in handling special values like NaN or Infinity.
-        options.fastMathEnabled = NO;
-
        NSError* error = nil;
+        // When options is nil, Metal uses the most recent language version available.
        id<MTLLibrary> library = [device newLibraryWithSource:objcSource
-                                                      options:options
+                                                      options:nil
                                                        error:&error];
        if (library == nil) {
            if (error) {
@@ -179,6 +170,9 @@ MetalShaderCompiler::program_token_t MetalShaderCompiler::createProgram(
    CompilerPriorityQueue const priorityQueue = program.getPriorityQueue();
    mCompilerThreadPool.queue(priorityQueue, token,
            [this, name, device = mDevice, program = std::move(program), token]() {
+                int sleepTime = atoi(name.c_str());
+                sleep(sleepTime);
+
                MetalFunctionBundle compiledProgram = compileProgram(program, device);

                token->set(compiledProgram);
@@ -18,8 +18,6 @@

 #include "MetalEnums.h"

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

@@ -182,10 +182,6 @@ bool NoopDriver::isParallelShaderCompileSupported() {
    return false;
 }

-bool NoopDriver::isDepthStencilResolveSupported() {
-    return true;
-}
-
 bool NoopDriver::isWorkaroundNeeded(Workaround) {
    return false;
 }
@@ -256,6 +252,10 @@ void NoopDriver::setExternalStream(Handle<HwTexture> th, Handle<HwStream> sh) {

 void NoopDriver::generateMipmaps(Handle<HwTexture> th) { }

+bool NoopDriver::canGenerateMipmaps() {
+    return true;
+}
+
 void NoopDriver::updateSamplerGroup(Handle<HwSamplerGroup> sbh,
        BufferDescriptor&& data) {
    scheduleDestroy(std::move(data));
@@ -322,23 +322,12 @@ void NoopDriver::readBufferSubData(backend::BufferObjectHandle boh,
    scheduleDestroy(std::move(p));
 }

-void NoopDriver::blitDEPRECATED(TargetBufferFlags buffers,
+void NoopDriver::blit(TargetBufferFlags buffers,
        Handle<HwRenderTarget> dst, Viewport dstRect,
        Handle<HwRenderTarget> src, Viewport srcRect,
        SamplerMagFilter filter) {
 }

-void NoopDriver::resolve(
-        Handle<HwTexture> dst, uint8_t srcLevel, uint8_t srcLayer,
-        Handle<HwTexture> src, uint8_t dstLevel, uint8_t dstLayer) {
-}
-
-void NoopDriver::blit(
-        Handle<HwTexture> dst, uint8_t srcLevel, uint8_t srcLayer, math::uint2 dstOrigin,
-        Handle<HwTexture> src, uint8_t dstLevel, uint8_t dstLayer, math::uint2 srcOrigin,
-        math::uint2 size) {
-}
-
 void NoopDriver::draw(PipelineState pipelineState, Handle<HwRenderPrimitive> rph,
        uint32_t instanceCount) {
 }
@@ -274,7 +274,6 @@ void OpenGLContext::setDefaultState() noexcept {
    glHint(GL_FRAGMENT_SHADER_DERIVATIVE_HINT, GL_NICEST);
 #endif

-#if !defined(__EMSCRIPTEN__)
    if (ext.EXT_clip_control) {
 #if defined(BACKEND_OPENGL_VERSION_GL)
        glClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE);
@@ -282,11 +281,9 @@ void OpenGLContext::setDefaultState() noexcept {
        glClipControlEXT(GL_LOWER_LEFT_EXT, GL_ZERO_TO_ONE_EXT);
 #endif
    }
-#endif

    if (ext.EXT_clip_cull_distance) {
        glEnable(GL_CLIP_DISTANCE0);
-        glEnable(GL_CLIP_DISTANCE1);
    }
 }

@@ -311,9 +308,7 @@ void OpenGLContext::initProcs(Procs* procs,
 #   ifdef BACKEND_OPENGL_VERSION_GL
    procs->getQueryObjectui64v = glGetQueryObjectui64v; // only core in GL
 #   elif defined(GL_EXT_disjoint_timer_query)
-#       ifndef __EMSCRIPTEN__
-            procs->getQueryObjectui64v = glGetQueryObjectui64vEXT;
-#       endif
+    procs->getQueryObjectui64v = glGetQueryObjectui64vEXT;
 #   endif // BACKEND_OPENGL_VERSION_GL

    // core in ES 3.0 and GL 4.3
@@ -325,7 +320,6 @@ void OpenGLContext::initProcs(Procs* procs,

 #ifdef BACKEND_OPENGL_VERSION_GLES
 #   ifndef IOS // IOS is guaranteed to have ES3.x
-#       ifndef __EMSCRIPTEN__
    if (UTILS_UNLIKELY(major == 2)) {
        // Runtime OpenGL version is ES 2.x
        if (UTILS_LIKELY(ext.OES_vertex_array_object)) {
@@ -353,7 +347,6 @@ void OpenGLContext::initProcs(Procs* procs,

        procs->maxShaderCompilerThreadsKHR = glMaxShaderCompilerThreadsKHR;
    }
-#       endif // __EMSCRIPTEN__
 #   endif // IOS
 #else
    procs->maxShaderCompilerThreadsKHR = glMaxShaderCompilerThreadsARB;
@@ -427,6 +420,7 @@ 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;
            }
@@ -481,6 +475,7 @@ void OpenGLContext::initBugs(Bugs* bugs, Extensions const& exts,
            // (that should be regardless of ANGLE, but we should double-check)
            bugs->split_easu = true;
        }
+        // TODO: see if we could use `bugs.allow_read_only_ancillary_feedback_loop = true`
    }

 #ifdef BACKEND_OPENGL_VERSION_GLES
@@ -585,23 +580,17 @@ void OpenGLContext::initExtensionsGLES(Extensions* ext, GLint major, GLint minor
    // figure out and initialize the extensions we need
    using namespace std::literals;
    ext->APPLE_color_buffer_packed_float = exts.has("GL_APPLE_color_buffer_packed_float"sv);
-#ifndef __EMSCRIPTEN__
    ext->EXT_clip_control = exts.has("GL_EXT_clip_control"sv);
-#endif
    ext->EXT_clip_cull_distance = exts.has("GL_EXT_clip_cull_distance"sv);
    ext->EXT_color_buffer_float = exts.has("GL_EXT_color_buffer_float"sv);
    ext->EXT_color_buffer_half_float = exts.has("GL_EXT_color_buffer_half_float"sv);
-#ifndef __EMSCRIPTEN__
    ext->EXT_debug_marker = exts.has("GL_EXT_debug_marker"sv);
-#endif
    ext->EXT_discard_framebuffer = exts.has("GL_EXT_discard_framebuffer"sv);
-#ifndef __EMSCRIPTEN__
    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);
-#endif
    ext->EXT_shader_framebuffer_fetch = exts.has("GL_EXT_shader_framebuffer_fetch"sv);
-#ifndef __EMSCRIPTEN__
+#if !defined(__EMSCRIPTEN__)
    ext->EXT_texture_compression_etc2 = true;
 #endif
    ext->EXT_texture_compression_s3tc = exts.has("GL_EXT_texture_compression_s3tc"sv);
@@ -222,6 +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;
@@ -429,6 +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",
                    ""},
@@ -704,7 +711,7 @@ void OpenGLContext::bindTexture(GLuint unit, GLuint target, GLuint texId, size_t
    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 {
@@ -147,9 +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;
 }

@@ -171,9 +171,8 @@ OpenGLDriver::OpenGLDriver(OpenGLPlatform* platform, const Platform::DriverConfi
          mContext(),
          mShaderCompilerService(*this),
          mHandleAllocator("Handles", driverConfig.handleArenaSize),
-          mSamplerMap(32),
-          mDriverConfig(driverConfig) {
-
+          mSamplerMap(32) {
+  
    std::fill(mSamplerBindings.begin(), mSamplerBindings.end(), nullptr);

    // set a reasonable default value for our stream array
@@ -263,27 +262,21 @@ void OpenGLDriver::bindTexture(GLuint unit, GLTexture const* t) noexcept {
    mContext.bindTexture(unit, t->gl.target, t->gl.id, t->gl.targetIndex);
 }

-bool OpenGLDriver::useProgram(OpenGLProgram* p) noexcept {
-    if (UTILS_UNLIKELY(!p->isValid())) {
-        // If the program is not valid, we can't call use().
-        return false;
-    }
-
+void OpenGLDriver::useProgram(OpenGLProgram* p) noexcept {
    // set-up textures and samplers in the proper TMUs (as specified in setSamplers)
    p->use(this, mContext);

    if (UTILS_UNLIKELY(mContext.isES2())) {
        for (uint32_t i = 0; i < Program::UNIFORM_BINDING_COUNT; i++) {
-            auto [id, buffer, age] = mUniformBindings[i];
+            auto [buffer, age] = mUniformBindings[i];
            if (buffer) {
-                p->updateUniforms(i, id, buffer, age);
+                p->updateUniforms(i, buffer, age);
            }
        }
        // Set the output colorspace for this program (linear or rec709). This in only relevant
        // when mPlatform.isSRGBSwapChainSupported() is false (no need to check though).
        p->setRec709ColorSpace(mRec709OutputColorspace);
    }
-    return true;
 }


@@ -486,7 +479,6 @@ void OpenGLDriver::createBufferObjectR(Handle<HwBufferObject> boh,

    GLBufferObject* bo = construct<GLBufferObject>(boh, byteCount, bindingType, usage);
    if (UTILS_UNLIKELY(bindingType == BufferObjectBinding::UNIFORM && gl.isES2())) {
-        bo->gl.id = ++mLastAssignedEmulatedUboId;
        bo->gl.buffer = malloc(byteCount);
        memset(bo->gl.buffer, 0, byteCount);
    } else {
@@ -538,7 +530,7 @@ void OpenGLDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {


    if (UTILS_UNLIKELY(mContext.isES2())) {
-        // Here we patch the specialization constants to enable or not the rec709 output
+        // Here we patch the specification constants to enable or not the rec709 output
        // color space emulation in this program. Obviously, the backend shouldn't know about
        // specific spec-constants, so we need to handle failures gracefully. This cannot be
        // done at Material creation time because only the backend has access to
@@ -603,9 +595,8 @@ void OpenGLDriver::textureStorage(OpenGLDriver::GLTexture* t,
                        }
                    } else {
                        glTexImage2D(t->gl.target, level, GLint(t->gl.internalFormat),
-                                std::max(GLsizei(1), GLsizei(width >> level)),
-                                std::max(GLsizei(1), GLsizei(height >> level)),
-                                0, format, type, nullptr);
+                                GLsizei(width), GLsizei(height), 0,
+                                format, type, nullptr);
                    }
                }
            }
@@ -1065,7 +1056,6 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
        }
        CHECK_GL_ERROR(utils::slog.e)
    } else
-#ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_multisampled_render_to_texture
        // EXT_multisampled_render_to_texture only support GL_COLOR_ATTACHMENT0
    if (!attachmentTypeNotSupportedByMSRTT && (t->depth <= 1)
@@ -1078,7 +1068,7 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
        // This extension only exists on OpenGL ES.
        gl.bindFramebuffer(GL_FRAMEBUFFER, rt->gl.fbo);
        if (any(t->usage & TextureUsage::SAMPLEABLE)) {
-            glFramebufferTexture2DMultisampleEXT(GL_FRAMEBUFFER,
+            glext::glFramebufferTexture2DMultisampleEXT(GL_FRAMEBUFFER,
                    attachment, target, t->gl.id, binfo.level, rt->gl.samples);
        } else {
            glFramebufferRenderbuffer(GL_FRAMEBUFFER, attachment,
@@ -1086,8 +1076,7 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
        }
        CHECK_GL_ERROR(utils::slog.e)
    } else
-#endif // GL_EXT_multisampled_render_to_texture
-#endif // __EMSCRIPTEN__
+#endif
    if (!any(t->usage & TextureUsage::SAMPLEABLE) && t->samples > 1) {
        assert_invariant(rt->gl.samples > 1);
        assert_invariant(glIsRenderbuffer(t->gl.id));
@@ -1186,7 +1175,6 @@ void OpenGLDriver::renderBufferStorage(GLuint rbo, GLenum internalformat, uint32
        uint32_t height, uint8_t samples) const noexcept {
    glBindRenderbuffer(GL_RENDERBUFFER, rbo);
    if (samples > 1) {
-#ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_multisampled_render_to_texture
        auto& gl = mContext;
        if (gl.ext.EXT_multisampled_render_to_texture ||
@@ -1194,8 +1182,7 @@ void OpenGLDriver::renderBufferStorage(GLuint rbo, GLenum internalformat, uint32
            glext::glRenderbufferStorageMultisampleEXT(GL_RENDERBUFFER,
                    samples, internalformat, width, height);
        } else
-#endif // GL_EXT_multisampled_render_to_texture
-#endif // __EMSCRIPTEN__
+#endif
        {
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
            glRenderbufferStorageMultisample(GL_RENDERBUFFER,
@@ -1911,10 +1898,6 @@ bool OpenGLDriver::isParallelShaderCompileSupported() {
    return mShaderCompilerService.isParallelShaderCompileSupported();
 }

-bool OpenGLDriver::isDepthStencilResolveSupported() {
-    return true;
-}
-
 bool OpenGLDriver::isWorkaroundNeeded(Workaround workaround) {
    switch (workaround) {
        case Workaround::SPLIT_EASU:
@@ -1984,13 +1967,6 @@ void OpenGLDriver::makeCurrent(Handle<HwSwapChain> schDraw, Handle<HwSwapChain>
    GLSwapChain* scRead = handle_cast<GLSwapChain*>(schRead);
    mPlatform.makeCurrent(scDraw->swapChain, scRead->swapChain);
    mCurrentDrawSwapChain = scDraw;
-
-    // From the GL spec for glViewport and glScissor:
-    // When a GL context is first attached to a window, width and height are set to the
-    // dimensions of that window.
-    // So basically, our viewport/scissor can be reset to "something" here.
-    mContext.state.window.viewport = {};
-    mContext.state.window.scissor = {};
 }

 // ------------------------------------------------------------------------------------------------
@@ -2236,7 +2212,8 @@ void OpenGLDriver::setMinMaxLevels(Handle<HwTexture> th, uint32_t minLevel, uint
        t->gl.baseLevel = (int8_t)minLevel;
        glTexParameteri(t->gl.target, GL_TEXTURE_BASE_LEVEL, t->gl.baseLevel);

-        t->gl.maxLevel = (int8_t)maxLevel; // NOTE: according to the GL spec, the default value of this 1000
+        t->gl
+         .maxLevel = (int8_t)maxLevel; // NOTE: according to the GL spec, the default value of this 1000
        glTexParameteri(t->gl.target, GL_TEXTURE_MAX_LEVEL, t->gl.maxLevel);
    }
 #endif
@@ -2286,13 +2263,16 @@ void OpenGLDriver::generateMipmaps(Handle<HwTexture> th) {
    CHECK_GL_ERROR(utils::slog.e)
 }

+bool OpenGLDriver::canGenerateMipmaps() {
+    return true;
+}
+
 void OpenGLDriver::setTextureData(GLTexture* t, uint32_t level,
        uint32_t xoffset, uint32_t yoffset, uint32_t zoffset,
        uint32_t width, uint32_t height, uint32_t depth,
        PixelBufferDescriptor&& p) {
    auto& gl = mContext;

-    assert_invariant(t != nullptr);
    assert_invariant(xoffset + width <= std::max(1u, t->width >> level));
    assert_invariant(yoffset + height <= std::max(1u, t->height >> level));
    assert_invariant(t->samples <= 1);
@@ -2921,11 +2901,7 @@ void OpenGLDriver::bindBufferRange(BufferObjectBinding bindingType, uint32_t ind
    assert_invariant(offset + size <= ub->byteCount);

    if (UTILS_UNLIKELY(ub->bindingType == BufferObjectBinding::UNIFORM && gl.isES2())) {
-        mUniformBindings[index] = {
-                ub->gl.id,
-                static_cast<uint8_t const*>(ub->gl.buffer) + offset,
-                ub->age,
-        };
+        mUniformBindings[index] = { static_cast<uint8_t const*>(ub->gl.buffer) + offset, ub->age };
    } else {
        GLenum const target = GLUtils::getBufferBindingType(bindingType);

@@ -2991,18 +2967,16 @@ GLuint OpenGLDriver::getSamplerSlow(SamplerParams params) const noexcept {
 #endif

 void OpenGLDriver::insertEventMarker(char const* string, uint32_t len) {
-#ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_debug_marker
    auto& gl = mContext;
    if (gl.ext.EXT_debug_marker) {
        glInsertEventMarkerEXT(GLsizei(len ? len : strlen(string)), string);
    }
 #endif
-#endif
 }

 void OpenGLDriver::pushGroupMarker(char const* string, uint32_t len) {
-#ifndef __EMSCRIPTEN__
+
 #ifdef GL_EXT_debug_marker
 #if DEBUG_MARKER_LEVEL & DEBUG_MARKER_OPENGL
    if (UTILS_LIKELY(mContext.ext.EXT_debug_marker)) {
@@ -3015,11 +2989,9 @@ void OpenGLDriver::pushGroupMarker(char const* string, uint32_t len) {
    SYSTRACE_CONTEXT();
    SYSTRACE_NAME_BEGIN(string);
 #endif
-#endif
 }

 void OpenGLDriver::popGroupMarker(int) {
-#ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_debug_marker
 #if DEBUG_MARKER_LEVEL & DEBUG_MARKER_OPENGL
    if (UTILS_LIKELY(mContext.ext.EXT_debug_marker)) {
@@ -3032,7 +3004,6 @@ void OpenGLDriver::popGroupMarker(int) {
    SYSTRACE_CONTEXT();
    SYSTRACE_NAME_END();
 #endif
-#endif
 }

 void OpenGLDriver::startCapture(int) {
@@ -3448,230 +3419,75 @@ void OpenGLDriver::clearWithRasterPipe(TargetBufferFlags clearFlags,
    CHECK_GL_ERROR(utils::slog.e)
 }

-void OpenGLDriver::resolve(
-        Handle<HwTexture> dst, uint8_t srcLevel, uint8_t srcLayer,
-        Handle<HwTexture> src, uint8_t dstLevel, uint8_t dstLayer) {
-    DEBUG_MARKER()
-    GLTexture const* const s = handle_cast<GLTexture*>(src);
-    GLTexture const* const d = handle_cast<GLTexture*>(dst);
-    assert_invariant(s);
-    assert_invariant(d);
-
-    ASSERT_PRECONDITION(
-            d->width == s->width && d->height == s->height,
-            "invalid resolve: src and dst sizes don't match");
-
-    ASSERT_PRECONDITION(s->samples > 1 && d->samples == 1,
-            "invalid resolve: src.samples=%u, dst.samples=%u",
-            +s->samples, +d->samples);
-
-    blit(   dst, dstLevel, dstLayer, {},
-            src, srcLevel, srcLayer, {},
-            { d->width, d->height });
-}
-
-void OpenGLDriver::blit(
-        Handle<HwTexture> dst, uint8_t srcLevel, uint8_t srcLayer, uint2 dstOrigin,
-        Handle<HwTexture> src, uint8_t dstLevel, uint8_t dstLayer, uint2 srcOrigin,
-        uint2 size) {
-    DEBUG_MARKER()
-    UTILS_UNUSED_IN_RELEASE auto& gl = mContext;
-    assert_invariant(!gl.isES2());
-
-#ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
-
-    GLTexture* d = handle_cast<GLTexture*>(dst);
-    GLTexture* s = handle_cast<GLTexture*>(src);
-    assert_invariant(d);
-    assert_invariant(s);
-
-    ASSERT_PRECONDITION_NON_FATAL(any(d->usage & TextureUsage::BLIT_DST),
-            "texture doesn't have BLIT_DST");
-
-    ASSERT_PRECONDITION_NON_FATAL(any(s->usage & TextureUsage::BLIT_SRC),
-            "texture doesn't have BLIT_SRC");
-
-    ASSERT_PRECONDITION_NON_FATAL(s->format == d->format,
-            "src and dst texture format don't match");
-
-    enum class AttachmentType : GLenum {
-        COLOR = GL_COLOR_ATTACHMENT0,
-        DEPTH = GL_DEPTH_ATTACHMENT,
-        STENCIL = GL_STENCIL_ATTACHMENT,
-        DEPTH_STENCIL = GL_DEPTH_STENCIL_ATTACHMENT,
-    };
-
-    auto getFormatType = [](TextureFormat format) -> AttachmentType {
-        bool const depth = isDepthFormat(format);
-        bool const stencil = isStencilFormat(format);
-        if (depth && stencil) return AttachmentType::DEPTH_STENCIL;
-        if (depth) return AttachmentType::DEPTH;
-        if (stencil) return AttachmentType::STENCIL;
-        return AttachmentType::COLOR;
-    };
-
-    AttachmentType const type = getFormatType(d->format);
-    assert_invariant(type == getFormatType(s->format));
-
-    // GL_INVALID_OPERATION is generated if mask contains any of the GL_DEPTH_BUFFER_BIT or
-    // GL_STENCIL_BUFFER_BIT and filter is not GL_NEAREST.
-    GLbitfield mask = {};
-    switch (type) {
-        case AttachmentType::COLOR:
-            mask = GL_COLOR_BUFFER_BIT;
-            break;
-        case AttachmentType::DEPTH:
-            mask = GL_DEPTH_BUFFER_BIT;
-            break;
-        case AttachmentType::STENCIL:
-            mask = GL_STENCIL_BUFFER_BIT;
-            break;
-        case AttachmentType::DEPTH_STENCIL:
-            mask = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT;
-            break;
-    };
-
-    GLuint fbo[2] = {};
-    glGenFramebuffers(2, fbo);
-
-    gl.bindFramebuffer(GL_DRAW_FRAMEBUFFER, fbo[0]);
-    switch (d->target) {
-        case SamplerType::SAMPLER_2D:
-            if (any(d->usage & TextureUsage::SAMPLEABLE)) {
-                glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GLenum(type),
-                        GL_TEXTURE_2D, d->gl.id, dstLevel);
-            } else {
-                glFramebufferRenderbuffer(GL_DRAW_FRAMEBUFFER, GLenum(type),
-                        GL_RENDERBUFFER, d->gl.id);
-            }
-            break;
-        case SamplerType::SAMPLER_CUBEMAP:
-            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GLenum(type),
-                    GL_TEXTURE_CUBE_MAP_POSITIVE_X + dstLayer, d->gl.id, dstLevel);
-            break;
-        case SamplerType::SAMPLER_2D_ARRAY:
-        case SamplerType::SAMPLER_CUBEMAP_ARRAY:
-        case SamplerType::SAMPLER_3D:
-            glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GLenum(type),
-                    d->gl.id, dstLevel, dstLayer);
-            break;
-        case SamplerType::SAMPLER_EXTERNAL:
-            break;
-    }
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_READ_FRAMEBUFFER)
-
-    gl.bindFramebuffer(GL_READ_FRAMEBUFFER, fbo[1]);
-    switch (s->target) {
-        case SamplerType::SAMPLER_2D:
-            if (any(s->usage & TextureUsage::SAMPLEABLE)) {
-                glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GLenum(type),
-                        GL_TEXTURE_2D, s->gl.id, srcLevel);
-            } else {
-                glFramebufferRenderbuffer(GL_READ_FRAMEBUFFER, GLenum(type),
-                        GL_RENDERBUFFER, s->gl.id);
-            }
-            break;
-        case SamplerType::SAMPLER_CUBEMAP:
-            glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GLenum(type),
-                    GL_TEXTURE_CUBE_MAP_POSITIVE_X + srcLayer, s->gl.id, srcLevel);
-            break;
-        case SamplerType::SAMPLER_2D_ARRAY:
-        case SamplerType::SAMPLER_CUBEMAP_ARRAY:
-        case SamplerType::SAMPLER_3D:
-            glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GLenum(type),
-                    s->gl.id, srcLevel, srcLayer);
-            break;
-        case SamplerType::SAMPLER_EXTERNAL:
-            break;
-    }
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_DRAW_FRAMEBUFFER)
-
-    gl.disable(GL_SCISSOR_TEST);
-    glBlitFramebuffer(
-            srcOrigin.x, srcOrigin.y, srcOrigin.x + size.x, srcOrigin.y + size.y,
-            dstOrigin.x, dstOrigin.y, dstOrigin.x + size.x, dstOrigin.y + size.y,
-            mask, GL_NEAREST);
-    CHECK_GL_ERROR(utils::slog.e)
-
-    gl.bindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
-    gl.bindFramebuffer(GL_READ_FRAMEBUFFER, 0);
-    glDeleteFramebuffers(2, fbo);
-
-    if (any(d->usage & TextureUsage::SAMPLEABLE)) {
-        // In a sense, blitting to a texture level is similar to calling setTextureData on it; in
-        // both cases, we update the base/max LOD to give shaders access to levels as they become
-        // available.  Note that this can only expand the LOD range (never shrink it), and that
-        // users can override this range by calling setMinMaxLevels().
-        updateTextureLodRange(d, int8_t(dstLevel));
-    }
-
-#endif
-}
-
-void OpenGLDriver::blitDEPRECATED(TargetBufferFlags buffers,
+void OpenGLDriver::blit(TargetBufferFlags buffers,
        Handle<HwRenderTarget> dst, Viewport dstRect,
        Handle<HwRenderTarget> src, Viewport srcRect,
        SamplerMagFilter filter) {
-
-    // Note: blitDEPRECATED is only used by Renderer::copyFrame
-
    DEBUG_MARKER()
    UTILS_UNUSED_IN_RELEASE auto& gl = mContext;
    assert_invariant(!gl.isES2());

-    ASSERT_PRECONDITION(buffers == TargetBufferFlags::COLOR0,
-            "blitDEPRECATED only supports COLOR0");
-
-    ASSERT_PRECONDITION(srcRect.left >= 0 && srcRect.bottom >= 0 &&
-                        dstRect.left >= 0 && dstRect.bottom >= 0,
-            "Source and destination rects must be positive.");
-
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
+    GLbitfield const mask = getAttachmentBitfield(buffers);
+    if (mask) {
+        GLenum glFilterMode = (filter == SamplerMagFilter::NEAREST) ? GL_NEAREST : GL_LINEAR;
+        if (mask & (GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT)) {
+            // GL_INVALID_OPERATION is generated if mask contains any of the GL_DEPTH_BUFFER_BIT or
+            // GL_STENCIL_BUFFER_BIT and filter is not GL_NEAREST.
+            glFilterMode = GL_NEAREST;
+        }

-    GLenum const glFilterMode = (filter == SamplerMagFilter::NEAREST) ? GL_NEAREST : GL_LINEAR;
+        // note: for msaa RenderTargets with non-msaa attachments, we copy from the msaa sidecar
+        // buffer -- this should produce the same output that if we copied from the resolved
+        // texture. EXT_multisampled_render_to_texture seems to allow both behaviours, and this
+        // is an emulation of that.  We cannot use the resolved texture easily because it's not
+        // actually attached to the RenderTarget. Another implementation would be to do a
+        // reverse-resolve, but that wouldn't buy us anything.
+        GLRenderTarget const* s = handle_cast<GLRenderTarget const*>(src);
+        GLRenderTarget const* d = handle_cast<GLRenderTarget const*>(dst);

-    // note: for msaa RenderTargets with non-msaa attachments, we copy from the msaa sidecar
-    // buffer -- this should produce the same output that if we copied from the resolved
-    // texture. EXT_multisampled_render_to_texture seems to allow both behaviours, and this
-    // is an emulation of that.  We cannot use the resolved texture easily because it's not
-    // actually attached to the RenderTarget. Another implementation would be to do a
-    // reverse-resolve, but that wouldn't buy us anything.
-    GLRenderTarget const* s = handle_cast<GLRenderTarget const*>(src);
-    GLRenderTarget const* d = handle_cast<GLRenderTarget const*>(dst);
+        // blit operations are only supported from RenderTargets that have only COLOR0 (or
+        // no color buffer at all)
+        assert_invariant(
+                !(s->targets & (TargetBufferFlags::COLOR_ALL & ~TargetBufferFlags::COLOR0)));

-    // With GLES 3.x, GL_INVALID_OPERATION is generated if the value of GL_SAMPLE_BUFFERS
-    // for the draw buffer is greater than zero. This works with OpenGL, so we want to
-    // make sure to catch this scenario.
-    assert_invariant(d->gl.samples <= 1);
+        assert_invariant(
+                !(d->targets & (TargetBufferFlags::COLOR_ALL & ~TargetBufferFlags::COLOR0)));

-    // GL_INVALID_OPERATION is generated if GL_SAMPLE_BUFFERS for the read buffer is greater
-    // than zero and the formats of draw and read buffers are not identical.
-    // However, it's not well-defined in the spec what "format" means. So it's difficult
-    // to have an assert here -- especially when dealing with the default framebuffer
+        // With GLES 3.x, GL_INVALID_OPERATION is generated if the value of GL_SAMPLE_BUFFERS
+        // for the draw buffer is greater than zero. This works with OpenGL, so we want to
+        // make sure to catch this scenario.
+        assert_invariant(d->gl.samples <= 1);

-    // GL_INVALID_OPERATION is generated if GL_SAMPLE_BUFFERS for the read buffer is greater
-    // than zero and (...) the source and destination rectangles are not defined with the
-    // same (X0, Y0) and (X1, Y1) bounds.
+        // GL_INVALID_OPERATION is generated if GL_SAMPLE_BUFFERS for the read buffer is greater
+        // than zero and the formats of draw and read buffers are not identical.
+        // However, it's not well-defined in the spec what "format" means. So it's difficult
+        // to have an assert here -- especially when dealing with the default framebuffer

-    // Additionally, the EXT_multisampled_render_to_texture extension doesn't specify what
-    // happens when blitting from an "implicit" resolve render target (does it work?), so
-    // to ere on the safe side, we don't allow it.
-    if (s->gl.samples > 1) {
-        assert_invariant(!memcmp(&dstRect, &srcRect, sizeof(srcRect)));
+        // GL_INVALID_OPERATION is generated if GL_SAMPLE_BUFFERS for the read buffer is greater
+        // than zero and (...) the source and destination rectangles are not defined with the
+        // same (X0, Y0) and (X1, Y1) bounds.
+
+        // Additionally, the EXT_multisampled_render_to_texture extension doesn't specify what
+        // happens when blitting from an "implicit" resolve render target (does it work?), so
+        // to ere on the safe side, we don't allow it.
+        if (s->gl.samples > 1) {
+            assert_invariant(!memcmp(&dstRect, &srcRect, sizeof(srcRect)));
+        }
+
+        gl.bindFramebuffer(GL_READ_FRAMEBUFFER, s->gl.fbo);
+        gl.bindFramebuffer(GL_DRAW_FRAMEBUFFER, d->gl.fbo);
+
+        CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_READ_FRAMEBUFFER)
+        CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_DRAW_FRAMEBUFFER)
+
+        gl.disable(GL_SCISSOR_TEST);
+        glBlitFramebuffer(
+                srcRect.left, srcRect.bottom, srcRect.right(), srcRect.top(),
+                dstRect.left, dstRect.bottom, dstRect.right(), dstRect.top(),
+                mask, glFilterMode);
+        CHECK_GL_ERROR(utils::slog.e)
    }
-
-    gl.bindFramebuffer(GL_READ_FRAMEBUFFER, s->gl.fbo);
-    gl.bindFramebuffer(GL_DRAW_FRAMEBUFFER, d->gl.fbo);
-
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_READ_FRAMEBUFFER)
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_DRAW_FRAMEBUFFER)
-
-    gl.disable(GL_SCISSOR_TEST);
-    glBlitFramebuffer(
-            srcRect.left, srcRect.bottom, srcRect.right(), srcRect.top(),
-            dstRect.left, dstRect.bottom, dstRect.right(), dstRect.top(),
-            GL_COLOR_BUFFER_BIT, glFilterMode);
-    CHECK_GL_ERROR(utils::slog.e)
 #endif
 }

@@ -3702,17 +3518,18 @@ void OpenGLDriver::draw(PipelineState state, Handle<HwRenderPrimitive> rph, uint
    DEBUG_MARKER()
    auto& gl = mContext;

-    OpenGLProgram* const p = handle_cast<OpenGLProgram*>(state.program);
+    OpenGLProgram* 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.
+    // If the material debugger is enabled, avoid fatal (or cascading) errors and 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.
+    if (FILAMENT_ENABLE_MATDBG && UTILS_UNLIKELY(!p->isValid())) {
        return;
    }

-    GLRenderPrimitive* const rp = handle_cast<GLRenderPrimitive*>(rph);
+    useProgram(p);
+
+    GLRenderPrimitive* rp = handle_cast<GLRenderPrimitive *>(rph);

    // Gracefully do nothing if the render primitive has not been set up.
    VertexBufferHandle vb = rp->gl.vertexBufferWithObjects;
@@ -3723,7 +3540,7 @@ void OpenGLDriver::draw(PipelineState state, Handle<HwRenderPrimitive> rph, uint
    gl.bindVertexArray(&rp->gl);

    // If necessary, mutate the bindings in the VAO.
-    GLVertexBuffer const* const glvb = handle_cast<GLVertexBuffer*>(vb);
+    const GLVertexBuffer* glvb = handle_cast<GLVertexBuffer*>(vb);
    if (UTILS_UNLIKELY(rp->gl.vertexBufferVersion != glvb->bufferObjectsVersion)) {
        updateVertexArrayObject(rp, glvb);
    }
@@ -3757,16 +3574,17 @@ void OpenGLDriver::draw(PipelineState state, Handle<HwRenderPrimitive> rph, uint
 void OpenGLDriver::dispatchCompute(Handle<HwProgram> program, math::uint3 workGroupCount) {
    getShaderCompilerService().tick();

-    OpenGLProgram* const p = handle_cast<OpenGLProgram*>(program);
+    OpenGLProgram* p = handle_cast<OpenGLProgram*>(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.
+    // If the material debugger is enabled, avoid fatal (or cascading) errors and 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.
+    if (FILAMENT_ENABLE_MATDBG && UTILS_UNLIKELY(!p->isValid())) {
        return;
    }

+    useProgram(p);
+
 #if defined(BACKEND_OPENGL_LEVEL_GLES31)

 #if defined(__ANDROID__)
@@ -83,9 +83,11 @@ public:
        }

        struct {
-            GLuint id;
            union {
-                GLenum binding;
+                struct {
+                    GLuint id;
+                    GLenum binding;
+                };
                void* buffer;
            };
        } gl;
@@ -318,7 +320,7 @@ private:

           void bindTexture(GLuint unit, GLTexture const* t) noexcept;
           void bindSampler(GLuint unit, GLuint sampler) noexcept;
-    inline bool useProgram(OpenGLProgram* p) noexcept;
+    inline void useProgram(OpenGLProgram* p) noexcept;

    enum class ResolveAction { LOAD, STORE };
    void resolvePass(ResolveAction action, GLRenderTarget const* rt,
@@ -361,8 +363,7 @@ private:
    void setScissor(Viewport const& scissor) noexcept;

    // ES2 only. Uniform buffer emulation binding points
-    GLuint mLastAssignedEmulatedUboId = 0;
-    std::array<std::tuple<GLuint, void const*, uint16_t>, Program::UNIFORM_BINDING_COUNT> mUniformBindings = {};
+    std::array<std::pair<void const*, uint16_t>, Program::UNIFORM_BINDING_COUNT> mUniformBindings = {};

    // sampler buffer binding points (nullptr if not used)
    std::array<GLSamplerGroup*, Program::SAMPLER_BINDING_COUNT> mSamplerBindings = {};   // 4 pointers
@@ -400,9 +401,6 @@ private:
    // timer query implementation
    OpenGLTimerQueryInterface* mTimerQueryImpl = nullptr;

-    const Platform::DriverConfig mDriverConfig;
-    Platform::DriverConfig const& getDriverConfig() const noexcept { return mDriverConfig; }
-
    // for ES2 sRGB support
    GLSwapChain* mCurrentDrawSwapChain = nullptr;
    bool mRec709OutputColorspace = false;
@@ -129,7 +129,7 @@ void OpenGLProgram::initializeProgramState(OpenGLContext& context, GLuint progra
 #endif
    {
        // ES2 initialization of (fake) UBOs
-        UniformsRecord* const uniformsRecords = new(std::nothrow) UniformsRecord[Program::UNIFORM_BINDING_COUNT];
+        UniformsRecord* const uniformsRecords = new UniformsRecord[Program::UNIFORM_BINDING_COUNT];
        UTILS_NOUNROLL
        for (GLuint binding = 0, n = Program::UNIFORM_BINDING_COUNT; binding < n; binding++) {
            Program::UniformInfo& uniforms = lazyInitializationData.bindingUniformInfo[binding];
@@ -229,16 +229,15 @@ void OpenGLProgram::updateSamplers(OpenGLDriver* const gld) const noexcept {
    CHECK_GL_ERROR(utils::slog.e)
 }

-void OpenGLProgram::updateUniforms(uint32_t index, GLuint id, void const* buffer, uint16_t age) noexcept {
+void OpenGLProgram::updateUniforms(uint32_t index, void const* buffer, uint16_t age) noexcept {
    assert_invariant(mUniformsRecords);
    assert_invariant(buffer);

    // only update the uniforms if the UBO has changed since last time we updated
    UniformsRecord const& records = mUniformsRecords[index];
-    if (records.id == id && records.age == age) {
+    if (records.age == age) {
        return;
    }
-    records.id = id;
    records.age = age;

    assert_invariant(records.uniforms.size() == records.locations.size());
@@ -59,7 +59,7 @@ public:
            // - the content of any bound sampler buffer has changed
            // ... since last time we used this program

-            // Turns out the former might be relatively cheap to check, the latter requires
+            // turns out the former might be relatively cheap to check, the later requires
            // a bit less. Compared to what updateSamplers() actually does, which is
            // pretty little, I'm not sure if we'll get ahead.

@@ -72,7 +72,7 @@ public:
    } gl; // 12 bytes

    // For ES2 only
-    void updateUniforms(uint32_t index, GLuint id, void const* buffer, uint16_t age) noexcept;
+    void updateUniforms(uint32_t index, void const* buffer, uint16_t age) noexcept;
    void setRec709ColorSpace(bool rec709) const noexcept;

 private:
@@ -98,7 +98,6 @@ private:
    struct UniformsRecord {
        Program::UniformInfo uniforms;
        LocationInfo locations;
-        mutable GLuint id = 0;
        mutable uint16_t age = std::numeric_limits<uint16_t>::max();
    };
    UniformsRecord const* mUniformsRecords = nullptr;
@@ -22,6 +22,7 @@

 #include <private/backend/BackendUtils.h>

+#include <backend/platforms/OpenGLPlatform.h>
 #include <backend/Program.h>

 #include <utils/compiler.h>
@@ -140,43 +141,20 @@ void* ShaderCompilerService::getUserData(const program_token_t& token) noexcept
 ShaderCompilerService::ShaderCompilerService(OpenGLDriver& driver)
        : mDriver(driver),
          mBlobCache(driver.getContext()),
-          mCallbackManager(driver) {
+          mCallbackManager(driver),
+          KHR_parallel_shader_compile(driver.getContext().ext.KHR_parallel_shader_compile) {
 }

 ShaderCompilerService::~ShaderCompilerService() noexcept = default;

 bool ShaderCompilerService::isParallelShaderCompileSupported() const noexcept {
-    assert_invariant(mMode != Mode::UNDEFINED);
-    return mMode != Mode::SYNCHRONOUS;
+    return KHR_parallel_shader_compile || mShaderCompilerThreadCount;
 }

 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).
-    // In theory, using Mode::ASYNCHRONOUS (a.k.a. KHR_parallel_shader_compile) could be more
-    // efficient, since it doesn't require shared contexts.
-    // In practice, we already know that with ANGLE, Mode::ASYNCHRONOUS can cause very long
-    // pauses at glDraw() time in the situation where glLinkProgram() has been emitted, but has
-    // other programs ahead of it in ANGLE's queue.
-    if (mDriver.mPlatform.isExtraContextSupported()) {
-        // our thread-pool if possible
-        mMode = Mode::THREAD_POOL;
-    } else if (mDriver.getContext().ext.KHR_parallel_shader_compile) {
-        // if not, async shader compilation and link if the driver supports it
-        mMode = Mode::ASYNCHRONOUS;
-    } else {
-        // fallback to synchronous shader compilation
-        mMode = Mode::SYNCHRONOUS;
-    }
-
-    if (mMode == Mode::THREAD_POOL) {
+    // If we have KHR_parallel_shader_compile, we always use it, it should be more resource
+    // friendly.
+    if (!KHR_parallel_shader_compile) {
        // - on Adreno there is a single compiler object. We can't use a pool > 1
        //   also glProgramBinary blocks if other threads are compiling.
        // - on Mali shader compilation can be multi-threaded, but program linking happens on
@@ -186,42 +164,37 @@ void ShaderCompilerService::init() noexcept {
        // - on macOS (M1 MacBook Pro/Ventura) there is global lock around all GL APIs when using
        //   a shared context, so parallel shader compilation yields no benefit.
        // - on windows/linux we could use more threads, tbd.
+        if (mDriver.mPlatform.isExtraContextSupported()) {
+            // By default, we use one thread at the same priority as the gl thread. This is the
+            // safest choice that avoids priority inversions.
+            uint32_t poolSize = 1;
+            JobSystem::Priority priority = JobSystem::Priority::DISPLAY;

-        // By default, we use one thread at the same priority as the gl thread. This is the
-        // safest choice that avoids priority inversions.
-        uint32_t poolSize = 1;
-        JobSystem::Priority priority = JobSystem::Priority::DISPLAY;
+            auto const& renderer = mDriver.getContext().state.renderer;
+            if (UTILS_UNLIKELY(strstr(renderer, "PowerVR"))) {
+                // The PowerVR driver support parallel shader compilation well, so we use 2
+                // threads, we can use lower priority threads here because urgent compilations
+                // will most likely happen on the main gl thread. Using too many thread can
+                // increase memory pressure significantly.
+                poolSize = 2;
+                priority = JobSystem::Priority::BACKGROUND;
+            }

-        auto const& renderer = mDriver.getContext().state.renderer;
-        // Some drivers support parallel shader compilation well, so we use N
-        // threads, we can use lower priority threads here because urgent compilations
-        // will most likely happen on the main gl thread. Using too many thread can
-        // increase memory pressure significantly.
-        if (UTILS_UNLIKELY(strstr(renderer, "PowerVR"))) {
-            // For PowerVR, it's unclear what the cost of extra shared contexts is, so we only
-            // use 2 to be safe.
-            poolSize = 2;
-            priority = JobSystem::Priority::BACKGROUND;
-        } else if (UTILS_UNLIKELY(strstr(renderer, "ANGLE"))) {
-            // Angle shared contexts are not expensive once we have two.
-            poolSize = (std::thread::hardware_concurrency() + 1) / 2;
-            priority = JobSystem::Priority::BACKGROUND;
+            mShaderCompilerThreadCount = poolSize;
+            mCompilerThreadPool.init(mShaderCompilerThreadCount,
+                    [&platform = mDriver.mPlatform, priority]() {
+                        // give the thread a name
+                        JobSystem::setThreadName("CompilerThreadPool");
+                        // run at a slightly lower priority than other filament threads
+                        JobSystem::setThreadPriority(priority);
+                        // create a gl context current to this thread
+                        platform.createContext(true);
+                    },
+                    [&platform = mDriver.mPlatform]() {
+                        // release context and thread state
+                        platform.releaseContext();
+                    });
        }
-
-        mShaderCompilerThreadCount = poolSize;
-        mCompilerThreadPool.init(mShaderCompilerThreadCount,
-                [&platform = mDriver.mPlatform, priority]() {
-                    // give the thread a name
-                    JobSystem::setThreadName("CompilerThreadPool");
-                    // run at a slightly lower priority than other filament threads
-                    JobSystem::setThreadPriority(priority);
-                    // create a gl context current to this thread
-                    platform.createContext(true);
-                },
-                [&platform = mDriver.mPlatform]() {
-                    // release context and thread state
-                    platform.releaseContext();
-                });
    }
 }

@@ -255,7 +228,7 @@ ShaderCompilerService::program_token_t ShaderCompilerService::createProgram(
    token->handle = mCallbackManager.get();

    CompilerPriorityQueue const priorityQueue = program.getPriorityQueue();
-    if (mMode == Mode::THREAD_POOL) {
+    if (mShaderCompilerThreadCount) {
        // queue a compile job
        mCompilerThreadPool.queue(priorityQueue, token,
                [this, &gl, program = std::move(program), token]() mutable {
@@ -281,6 +254,8 @@ ShaderCompilerService::program_token_t ShaderCompilerService::createProgram(
                    glGetProgramiv(glProgram, GL_LINK_STATUS, &status);
                    programData.program = glProgram;

+                    token->gl.program = programData.program;
+
                    // we don't need to check for success here, it'll be done on the
                    // main thread side.
                    token->set(programData);
@@ -304,8 +279,7 @@ ShaderCompilerService::program_token_t ShaderCompilerService::createProgram(
                token->shaderSourceCode);

        runAtNextTick(priorityQueue, token, [this, token](Job const&) {
-            assert_invariant(mMode != Mode::THREAD_POOL);
-            if (mMode == Mode::ASYNCHRONOUS) {
+            if (KHR_parallel_shader_compile) {
                // don't attempt to link this program if all shaders are not done compiling
                GLint status;
                if (token->gl.program) {
@@ -329,7 +303,7 @@ ShaderCompilerService::program_token_t ShaderCompilerService::createProgram(
                // link the program, this also cannot fail because status is checked later.
                token->gl.program = linkProgram(mDriver.getContext(),
                        token->gl.shaders, token->attributes);
-                if (mMode == Mode::ASYNCHRONOUS) {
+                if (KHR_parallel_shader_compile) {
                    // wait until the link finishes...
                    return false;
                }
@@ -363,14 +337,14 @@ GLuint ShaderCompilerService::getProgram(ShaderCompilerService::program_token_t&
    return program;
 }

-void ShaderCompilerService::terminate(program_token_t& token) {
+/* static*/ void ShaderCompilerService::terminate(program_token_t& token) {
    assert_invariant(token);

    token->canceled = true;

-    bool const canceled = token->compiler.cancelTickOp(token);
+    bool canceled = token->compiler.cancelTickOp(token);

-    if (token->compiler.mMode == Mode::THREAD_POOL) {
+    if (token->compiler.mShaderCompilerThreadCount) {
        auto job = token->compiler.mCompilerThreadPool.dequeue(token);
        if (!job) {
            // The job is being executed right now. We need to wait for it to finish to avoid a
@@ -404,7 +378,7 @@ void ShaderCompilerService::terminate(program_token_t& token) {

 void ShaderCompilerService::tick() {
    // we don't need to run executeTickOps() if we're using the thread-pool
-    if (UTILS_UNLIKELY(mMode != Mode::THREAD_POOL)) {
+    if (UTILS_UNLIKELY(!mShaderCompilerThreadCount)) {
        executeTickOps();
    }
 }
@@ -429,7 +403,7 @@ void ShaderCompilerService::getProgramFromCompilerPool(program_token_t& token) n
 GLuint ShaderCompilerService::initialize(program_token_t& token) noexcept {
    SYSTRACE_CALL();
    if (!token->gl.program) {
-        if (mMode == Mode::THREAD_POOL) {
+        if (mShaderCompilerThreadCount) {
            // we need this program right now, remove it from the queue
            auto job = mCompilerThreadPool.dequeue(token);
            if (job) {
@@ -445,7 +419,7 @@ GLuint ShaderCompilerService::initialize(program_token_t& token) noexcept {
            // Block until we get the program from the pool. Generally this wouldn't block
            // because we just compiled the program above, when executing job.
            ShaderCompilerService::getProgramFromCompilerPool(token);
-        } else if (mMode == Mode::ASYNCHRONOUS) {
+        } else if (KHR_parallel_shader_compile) {
            // we force the program link -- which might stall, either here or below in
            // checkProgramStatus(), but we don't have a choice, we need to use the program now.
            token->compiler.cancelTickOp(token);
@@ -553,8 +527,7 @@ void ShaderCompilerService::compileShaders(OpenGLContext& context,
            // 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
+            // split shader source, so we can insert the specification constants and the packing functions
            auto const [prolog, body] = splitShaderSource(source);

            const std::array<const char*, 4> sources = {
@@ -95,42 +95,19 @@ public:
            CallbackHandler* handler, CallbackHandler::Callback callback, void* user);

 private:
-    struct Job {
-        template<typename FUNC>
-        Job(FUNC&& fn) : fn(std::forward<FUNC>(fn)) {}
-        Job(std::function<bool(Job const& job)> fn,
-                CallbackHandler* handler, void* user, CallbackHandler::Callback callback)
-                : fn(std::move(fn)), handler(handler), user(user), callback(callback) {
-        }
-        std::function<bool(Job const& job)> fn;
-        CallbackHandler* handler = nullptr;
-        void* user = nullptr;
-        CallbackHandler::Callback callback{};
-    };
-
-    enum class Mode {
-        UNDEFINED,      // init() has not been called yet.
-        SYNCHRONOUS,    // synchronous shader compilation
-        THREAD_POOL,    // asynchronous shader compilation using a thread-pool (most common)
-        ASYNCHRONOUS    // asynchronous shader compilation using KHR_parallel_shader_compile
-    };
-
    OpenGLDriver& mDriver;
    OpenGLBlobCache mBlobCache;
    CallbackManager mCallbackManager;
    CompilerThreadPool mCompilerThreadPool;

+    const bool KHR_parallel_shader_compile;
    uint32_t mShaderCompilerThreadCount = 0u;
-    Mode mMode = Mode::UNDEFINED; // valid after init() is called
-
-    using ContainerType = std::tuple<CompilerPriorityQueue, program_token_t, Job>;
-    std::vector<ContainerType> mRunAtNextTickOps;

    GLuint initialize(ShaderCompilerService::program_token_t& token) noexcept;

    static void getProgramFromCompilerPool(program_token_t& token) noexcept;

-    static void compileShaders(
+        static void compileShaders(
            OpenGLContext& context,
            Program::ShaderSource shadersSource,
            utils::FixedCapacityVector<Program::SpecializationConstant> const& specializationConstants,
@@ -150,11 +127,27 @@ private:

    static bool checkProgramStatus(program_token_t const& token) noexcept;

+    struct Job {
+        template<typename FUNC>
+        Job(FUNC&& fn) : fn(std::forward<FUNC>(fn)) {}
+        Job(std::function<bool(Job const& job)> fn,
+                CallbackHandler* handler, void* user, CallbackHandler::Callback callback)
+                : fn(std::move(fn)), handler(handler), user(user), callback(callback) {
+        }
+        std::function<bool(Job const& job)> fn;
+        CallbackHandler* handler = nullptr;
+        void* user = nullptr;
+        CallbackHandler::Callback callback{};
+    };
+
    void runAtNextTick(CompilerPriorityQueue priority,
            const program_token_t& token, Job job) noexcept;
    void executeTickOps() noexcept;
    bool cancelTickOp(program_token_t token) noexcept;
    // order of insertion is important
+
+    using ContainerType = std::tuple<CompilerPriorityQueue, program_token_t, Job>;
+    std::vector<ContainerType> mRunAtNextTickOps;
 };

 } // namespace filament::backend
@@ -28,7 +28,6 @@ static void getProcAddress(T& pfn, const char* name) noexcept {
 }

 namespace glext {
-#ifndef __EMSCRIPTEN__
 #ifdef GL_OES_EGL_image
 PFNGLEGLIMAGETARGETTEXTURE2DOESPROC glEGLImageTargetTexture2DOES;
 #endif
@@ -73,11 +72,10 @@ PFNGLMAXSHADERCOMPILERTHREADSKHRPROC glMaxShaderCompilerThreadsKHR;
 // use getProcAddress for ES3.1 and above entry points.
 PFNGLDISPATCHCOMPUTEPROC glDispatchCompute;
 #endif
+
 static std::once_flag sGlExtInitialized;
-#endif // __EMSCRIPTEN__

 void importGLESExtensionsEntryPoints() {
-#ifndef __EMSCRIPTEN__
    std::call_once(sGlExtInitialized, +[]() {
 #ifdef GL_OES_EGL_image
    getProcAddress(glEGLImageTargetTexture2DOES, "glEGLImageTargetTexture2DOES");
@@ -121,7 +119,6 @@ void importGLESExtensionsEntryPoints() {
        getProcAddress(glDispatchCompute, "glDispatchCompute");
 #endif
    });
-#endif // __EMSCRIPTEN__
 }

 } // namespace glext
@@ -112,7 +112,6 @@ namespace glext {
 // it is currently called from PlatformEGL.
 void importGLESExtensionsEntryPoints();

-#ifndef __EMSCRIPTEN__
 #ifdef GL_OES_EGL_image
 extern PFNGLEGLIMAGETARGETTEXTURE2DOESPROC glEGLImageTargetTexture2DOES;
 #endif
@@ -154,7 +153,6 @@ extern PFNGLMAXSHADERCOMPILERTHREADSKHRPROC glMaxShaderCompilerThreadsKHR;
 #if defined(__ANDROID__) && !defined(FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2)
 extern PFNGLDISPATCHCOMPUTEPROC glDispatchCompute;
 #endif
-#endif // __EMSCRIPTEN__
 } // namespace glext

 using namespace glext;
@@ -192,10 +190,8 @@ using namespace glext;

 #if defined(GL_EXT_clip_cull_distance)
 #   define GL_CLIP_DISTANCE0                        GL_CLIP_DISTANCE0_EXT
-#   define GL_CLIP_DISTANCE1                        GL_CLIP_DISTANCE1_EXT
 #else
 #   define GL_CLIP_DISTANCE0                        0x3000
-#   define GL_CLIP_DISTANCE1                        0x3001
 #endif

 #if defined(GL_KHR_debug)
@@ -293,6 +293,20 @@ io::ostream& operator<<(io::ostream& out, TextureFormat format) {
    return out;
 }

+io::ostream& operator<<(io::ostream& out, TextureUsage usage) {
+    switch (usage) {
+        CASE(TextureUsage, NONE)
+        CASE(TextureUsage, DEFAULT)
+        CASE(TextureUsage, COLOR_ATTACHMENT)
+        CASE(TextureUsage, DEPTH_ATTACHMENT)
+        CASE(TextureUsage, STENCIL_ATTACHMENT)
+        CASE(TextureUsage, UPLOADABLE)
+        CASE(TextureUsage, SAMPLEABLE)
+        CASE(TextureUsage, SUBPASS_INPUT)
+    }
+    return out;
+}
+
 io::ostream& operator<<(io::ostream& out, TextureCubemapFace face) {
    switch (face) {
        CASE(TextureCubemapFace, NEGATIVE_X)
@@ -455,7 +469,7 @@ io::ostream& operator<<(io::ostream& out, filament::backend::Viewport const& vie

 io::ostream& operator<<(io::ostream& out, TargetBufferFlags flags) {
    // TODO: implement decoding of enum
-    out << uint32_t(flags);
+    out << uint8_t(flags);
    return out;
 }

@@ -38,68 +38,22 @@ using ImgUtil = VulkanImageUtility;
 namespace {

 inline void blitFast(const VkCommandBuffer cmdbuffer, VkImageAspectFlags aspect, VkFilter filter,
-        VulkanAttachment src, VulkanAttachment dst,
+        const VkExtent2D srcExtent, VulkanAttachment src, VulkanAttachment dst,
        const VkOffset3D srcRect[2], const VkOffset3D dstRect[2]) {
-
-    if constexpr (FVK_ENABLED(FVK_DEBUG_BLITTER)) {
-        utils::slog.d << "Fast blit from=" << src.texture->getVkImage() << ",level=" << (int) src.level
-                      << " layout=" << src.getLayout()
-                      << " to=" << dst.texture->getVkImage() << ",level=" << (int) dst.level
-                      << " layout=" << dst.getLayout() << utils::io::endl;
-    }
-
-    const VkImageSubresourceRange srcRange = {
-            .aspectMask = aspect,
-            .baseMipLevel = src.level,
-            .levelCount = 1,
-            .baseArrayLayer = src.layer,
-            .layerCount = 1,
-    };
-
-    const VkImageSubresourceRange dstRange = {
-            .aspectMask = aspect,
-            .baseMipLevel = dst.level,
-            .levelCount = 1,
-            .baseArrayLayer = dst.layer,
-            .layerCount = 1,
-    };
-
-    VulkanLayout oldSrcLayout = src.getLayout();
-    VulkanLayout oldDstLayout = dst.getLayout();
-
-    src.texture->transitionLayout(cmdbuffer, srcRange, VulkanLayout::TRANSFER_SRC);
-    dst.texture->transitionLayout(cmdbuffer, dstRange, VulkanLayout::TRANSFER_DST);
-
    const VkImageBlit blitRegions[1] = {{
-            .srcSubresource = { aspect, src.level, src.layer, 1 },
-            .srcOffsets = { srcRect[0], srcRect[1] },
-            .dstSubresource = { aspect, dst.level, dst.layer, 1 },
-            .dstOffsets = { dstRect[0], dstRect[1] },
-    }};
-    vkCmdBlitImage(cmdbuffer,
-            src.getImage(), ImgUtil::getVkLayout(VulkanLayout::TRANSFER_SRC),
-            dst.getImage(), ImgUtil::getVkLayout(VulkanLayout::TRANSFER_DST),
-            1, blitRegions, filter);
+            .srcSubresource = {aspect, src.level, src.layer, 1},
+            .srcOffsets = {srcRect[0], srcRect[1]},
+            .dstSubresource = {aspect, dst.level, dst.layer, 1},
+            .dstOffsets = {dstRect[0], dstRect[1]},
+        }};

-    if (oldSrcLayout == VulkanLayout::UNDEFINED) {
-        oldSrcLayout = ImgUtil::getDefaultLayout(src.texture->usage);
-    }
-    if (oldDstLayout == VulkanLayout::UNDEFINED) {
-        oldDstLayout = ImgUtil::getDefaultLayout(dst.texture->usage);
-    }
-    src.texture->transitionLayout(cmdbuffer, srcRange, oldSrcLayout);
-    dst.texture->transitionLayout(cmdbuffer, dstRange, oldDstLayout);
-}
-
-inline void resolveFast(const VkCommandBuffer cmdbuffer, VkImageAspectFlags aspect,
-        VulkanAttachment src, VulkanAttachment dst) {
-
-    if constexpr (FVK_ENABLED(FVK_DEBUG_BLITTER)) {
-        utils::slog.d << "Fast blit from=" << src.texture->getVkImage() << ",level=" << (int) src.level
-                      << " layout=" << src.getLayout()
-                      << " to=" << dst.texture->getVkImage() << ",level=" << (int) dst.level
-                      << " layout=" << dst.getLayout() << utils::io::endl;
-    }
+    const VkImageResolve resolveRegions[1] = {{
+            .srcSubresource = {aspect, src.level, src.layer, 1},
+            .srcOffset = srcRect[0],
+            .dstSubresource = {aspect, dst.level, dst.layer, 1},
+            .dstOffset = dstRect[0],
+            .extent = {srcExtent.width, srcExtent.height, 1},
+        }};

    const VkImageSubresourceRange srcRange = {
            .aspectMask = aspect,
@@ -117,25 +71,32 @@ inline void resolveFast(const VkCommandBuffer cmdbuffer, VkImageAspectFlags aspe
            .layerCount = 1,
    };

+    if constexpr (FVK_ENABLED(FVK_DEBUG_BLITTER)) {
+        utils::slog.d << "Fast blit from=" << src.texture->getVkImage() << ",level=" << (int) src.level
+                      << " layout=" << src.getLayout()
+                      << " to=" << dst.texture->getVkImage() << ",level=" << (int) dst.level
+                      << " layout=" << dst.getLayout() << utils::io::endl;
+    }
+
    VulkanLayout oldSrcLayout = src.getLayout();
    VulkanLayout oldDstLayout = dst.getLayout();

    src.texture->transitionLayout(cmdbuffer, srcRange, VulkanLayout::TRANSFER_SRC);
    dst.texture->transitionLayout(cmdbuffer, dstRange, VulkanLayout::TRANSFER_DST);

-    assert_invariant(
-            aspect != VK_IMAGE_ASPECT_DEPTH_BIT && "Resolve with depth is not yet supported.");
-    const VkImageResolve resolveRegions[1] = {{
-            .srcSubresource = { aspect, src.level, src.layer, 1 },
-            .srcOffset = { 0, 0 },
-            .dstSubresource = { aspect, dst.level, dst.layer, 1 },
-            .dstOffset = { 0, 0 },
-            .extent = { src.getExtent2D().width, src.getExtent2D().height, 1 },
-    }};
-    vkCmdResolveImage(cmdbuffer,
-            src.getImage(), ImgUtil::getVkLayout(VulkanLayout::TRANSFER_SRC),
-            dst.getImage(), ImgUtil::getVkLayout(VulkanLayout::TRANSFER_DST),
-            1, resolveRegions);
+    if (src.texture->samples > 1 && dst.texture->samples == 1) {
+        assert_invariant(
+                aspect != VK_IMAGE_ASPECT_DEPTH_BIT && "Resolve with depth is not yet supported.");
+        vkCmdResolveImage(cmdbuffer,
+                src.getImage(), ImgUtil::getVkLayout(VulkanLayout::TRANSFER_SRC),
+                dst.getImage(), ImgUtil::getVkLayout(VulkanLayout::TRANSFER_DST),
+                1, resolveRegions);
+    } else {
+        vkCmdBlitImage(cmdbuffer,
+                src.getImage(), ImgUtil::getVkLayout(VulkanLayout::TRANSFER_SRC),
+                dst.getImage(), ImgUtil::getVkLayout(VulkanLayout::TRANSFER_DST),
+                1, blitRegions, filter);
+    }

    if (oldSrcLayout == VulkanLayout::UNDEFINED) {
        oldSrcLayout = ImgUtil::getDefaultLayout(src.texture->usage);
@@ -154,22 +115,52 @@ struct BlitterUniforms {

 }// anonymous namespace

-VulkanBlitter::VulkanBlitter() noexcept = default;
+VulkanBlitter::VulkanBlitter(VulkanStagePool& stagePool, VulkanPipelineCache& pipelineCache,
+        VulkanFboCache& fboCache, VulkanSamplerCache& samplerCache) noexcept
+    : mStagePool(stagePool), mPipelineCache(pipelineCache), mFramebufferCache(fboCache),
+      mSamplerCache(samplerCache) {}

 void VulkanBlitter::initialize(VkPhysicalDevice physicalDevice, VkDevice device,
-        VmaAllocator allocator, VulkanCommands* commands) noexcept {
+        VmaAllocator allocator, VulkanCommands* commands, VulkanTexture* emptyTexture) noexcept {
    mPhysicalDevice = physicalDevice;
    mDevice = device;
    mAllocator = allocator;
    mCommands = commands;
+    mEmptyTexture = emptyTexture;
 }

-void VulkanBlitter::resolve(VulkanAttachment dst, VulkanAttachment src) {
+void VulkanBlitter::blitColor(BlitArgs args) {
+    const VulkanAttachment src = args.srcTarget->getColor(args.targetIndex);
+    const VulkanAttachment dst = args.dstTarget->getColor(0);
+    const VkImageAspectFlags aspect = VK_IMAGE_ASPECT_COLOR_BIT;

-    // src and dst should have the same aspect here
-    VkImageAspectFlags const aspect = src.texture->getImageAspect();
+#if FVK_ENABLED(FVK_DEBUG_BLIT_FORMAT)
+    VkPhysicalDevice const gpu = mPhysicalDevice;
+    VkFormatProperties info;
+    vkGetPhysicalDeviceFormatProperties(gpu, src.getFormat(), &info);
+    if (!ASSERT_POSTCONDITION_NON_FATAL(info.optimalTilingFeatures & VK_FORMAT_FEATURE_BLIT_SRC_BIT,
+                "Source format is not blittable %d", src.getFormat())) {
+        return;
+    }
+    vkGetPhysicalDeviceFormatProperties(gpu, dst.getFormat(), &info);
+    if (!ASSERT_POSTCONDITION_NON_FATAL(info.optimalTilingFeatures & VK_FORMAT_FEATURE_BLIT_DST_BIT,
+                "Destination format is not blittable %d", dst.getFormat())) {
+        return;
+    }
+#endif
+    VulkanCommandBuffer& commands = mCommands->get();
+    VkCommandBuffer const cmdbuffer = commands.buffer();
+    commands.acquire(src.texture);
+    commands.acquire(dst.texture);

-    assert_invariant(!(aspect & VK_IMAGE_ASPECT_DEPTH_BIT));
+    blitFast(cmdbuffer, aspect, args.filter, args.srcTarget->getExtent(), src, dst,
+            args.srcRectPair, args.dstRectPair);
+}
+
+void VulkanBlitter::blitDepth(BlitArgs args) {
+    const VulkanAttachment src = args.srcTarget->getDepth();
+    const VulkanAttachment dst = args.dstTarget->getDepth();
+    const VkImageAspectFlags aspect = VK_IMAGE_ASPECT_DEPTH_BIT;

 #if FVK_ENABLED(FVK_DEBUG_BLIT_FORMAT)
    VkPhysicalDevice const gpu = mPhysicalDevice;
@@ -186,40 +177,286 @@ void VulkanBlitter::resolve(VulkanAttachment dst, VulkanAttachment src) {
    }
 #endif

-    VulkanCommandBuffer& commands = mCommands->get();
-    VkCommandBuffer const cmdbuffer = commands.buffer();
-    commands.acquire(src.texture);
-    commands.acquire(dst.texture);
-    resolveFast(cmdbuffer, aspect, src, dst);
-}
+    assert_invariant(src.texture && dst.texture);

-void VulkanBlitter::blit(VkFilter filter,
-        VulkanAttachment dst, const VkOffset3D* dstRectPair,
-        VulkanAttachment src, const VkOffset3D* srcRectPair) {
-#if FVK_ENABLED(FVK_DEBUG_BLIT_FORMAT)
-    VkPhysicalDevice const gpu = mPhysicalDevice;
-    VkFormatProperties info;
-    vkGetPhysicalDeviceFormatProperties(gpu, src.getFormat(), &info);
-    if (!ASSERT_POSTCONDITION_NON_FATAL(info.optimalTilingFeatures & VK_FORMAT_FEATURE_BLIT_SRC_BIT,
-                "Depth src format is not blittable %d", src.getFormat())) {
+    if (src.texture->samples > 1 && dst.texture->samples == 1) {
+        blitSlowDepth(args.filter, args.srcTarget->getExtent(), src, dst, args.srcRectPair,
+                args.dstRectPair);
        return;
    }
-    vkGetPhysicalDeviceFormatProperties(gpu, dst.getFormat(), &info);
-    if (!ASSERT_POSTCONDITION_NON_FATAL(info.optimalTilingFeatures & VK_FORMAT_FEATURE_BLIT_DST_BIT,
-                "Depth dst format is not blittable %d", dst.getFormat())) {
-        return;
-    }
-#endif
-    // src and dst should have the same aspect here
-    VkImageAspectFlags const aspect = src.texture->getImageAspect();
+
    VulkanCommandBuffer& commands = mCommands->get();
    VkCommandBuffer const cmdbuffer = commands.buffer();
    commands.acquire(src.texture);
    commands.acquire(dst.texture);
-    blitFast(cmdbuffer, aspect, filter, src, dst, srcRectPair, dstRectPair);
+    blitFast(cmdbuffer, aspect, args.filter, args.srcTarget->getExtent(), src, dst, args.srcRectPair,
+            args.dstRectPair);
 }

 void VulkanBlitter::terminate() noexcept {
+    if (mDepthResolveProgram) {
+        delete mDepthResolveProgram;
+        mDepthResolveProgram = nullptr;
+    }
+
+    if (mTriangleBuffer) {
+        delete mTriangleBuffer;
+        mTriangleBuffer = nullptr;
+    }
+
+    if (mParamsBuffer) {
+        delete mParamsBuffer;
+        mParamsBuffer = nullptr;
+    }
+}
+
+// If we created these shader modules in the constructor, the device might not be ready yet.
+// It is easier to do lazy initialization, which can also improve load time.
+void VulkanBlitter::lazyInit() noexcept {
+    if (mDepthResolveProgram != nullptr) {
+        return;
+    }
+    assert_invariant(mDevice);
+
+    auto decode = [device = mDevice](const uint8_t* compressed, int compressedSize) {
+        const size_t spirvSize = smolv::GetDecodedBufferSize(compressed, compressedSize);
+        FixedCapacityVector<uint8_t> spirv(spirvSize);
+        smolv::Decode(compressed, compressedSize, spirv.data(), spirvSize);
+
+        VkShaderModuleCreateInfo moduleInfo = {};
+        moduleInfo.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
+        moduleInfo.codeSize = spirvSize;
+        moduleInfo.pCode = (uint32_t*) spirv.data();
+
+        VkShaderModule result = VK_NULL_HANDLE;
+        vkCreateShaderModule(device, &moduleInfo, VKALLOC, &result);
+        assert_invariant(result);
+
+        return result;
+    };
+
+    VkShaderModule vertexShader = decode(VKSHADERS_BLITDEPTHVS_DATA, VKSHADERS_BLITDEPTHVS_SIZE);
+    VkShaderModule fragmentShader = decode(VKSHADERS_BLITDEPTHFS_DATA, VKSHADERS_BLITDEPTHFS_SIZE);
+
+    // Allocate one anonymous sampler at slot 0.
+    VulkanProgram::CustomSamplerInfoList samplers = {
+        {0, 0, ShaderStageFlags::FRAGMENT},
+    };
+    mDepthResolveProgram = new VulkanProgram(mDevice, vertexShader, fragmentShader, samplers);
+
+#if FVK_ENABLED(FVK_DEBUG_BLITTER)
+    utils::slog.d << "Created Shader Module for VulkanBlitter "
+                  << "shaders = (" << vertexShader << ", " << fragmentShader << ")"
+                  << utils::io::endl;
+#endif
+
+    static const float kTriangleVertices[] = {
+        -1.0f, -1.0f,
+        +1.0f, -1.0f,
+        -1.0f, +1.0f,
+        +1.0f, +1.0f,
+    };
+
+    VulkanCommandBuffer& commands = mCommands->get();
+    VkCommandBuffer const cmdbuffer = commands.buffer();
+
+    mTriangleBuffer = new VulkanBuffer(mAllocator, mStagePool, VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
+            sizeof(kTriangleVertices));
+
+    mTriangleBuffer->loadFromCpu(cmdbuffer, kTriangleVertices, 0, sizeof(kTriangleVertices));
+
+    mParamsBuffer = new VulkanBuffer(mAllocator, mStagePool, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT,
+            sizeof(BlitterUniforms));
+}
+
+// At a high level, the procedure for resolving depth looks like this:
+// 1. Load uniforms
+// 2. Begin render pass
+// 3. Draw a big triangle
+// 4. End render pass.
+void VulkanBlitter::blitSlowDepth(VkFilter filter, const VkExtent2D srcExtent, VulkanAttachment src,
+        VulkanAttachment dst, const VkOffset3D srcRect[2], const VkOffset3D dstRect[2]) {
+    lazyInit();
+
+    VulkanCommandBuffer* commands = &mCommands->get();
+    VkCommandBuffer const cmdbuffer = commands->buffer();
+    commands->acquire(src.texture);
+    commands->acquire(dst.texture);
+
+    BlitterUniforms const uniforms = {
+            .sampleCount = src.texture->samples,
+            .inverseSampleCount = 1.0f / float(src.texture->samples),
+    };
+    mParamsBuffer->loadFromCpu(cmdbuffer, &uniforms, 0, sizeof(uniforms));
+
+    VkImageAspectFlags const aspect = VK_IMAGE_ASPECT_DEPTH_BIT;
+
+    // We will transition the src into sampler layout and also keep it in sampler layout for
+    // consistency.
+    VulkanLayout const samplerLayout = VulkanLayout::DEPTH_SAMPLER;
+
+    // BEGIN RENDER PASS
+    // -----------------
+
+    const VulkanFboCache::RenderPassKey rpkey = {
+        .initialColorLayoutMask = 0,
+        .initialDepthLayout = VulkanLayout::UNDEFINED,
+        .renderPassDepthLayout = samplerLayout,
+        .finalDepthLayout = samplerLayout,
+        .depthFormat = dst.getFormat(),
+        .clear = {},
+        .discardStart = TargetBufferFlags::DEPTH,
+        .samples = 1,
+    };
+
+    const VkRenderPass renderPass = mFramebufferCache.getRenderPass(rpkey);
+    mPipelineCache.bindRenderPass(renderPass, 0);
+
+    const VulkanFboCache::FboKey fbkey {
+        .renderPass = renderPass,
+        .width = (uint16_t) (dst.texture->width >> dst.level),
+        .height = (uint16_t) (dst.texture->height >> dst.level),
+        .layers = 1,
+        .samples = rpkey.samples,
+        .color = {},
+        .resolve = {},
+        .depth = dst.getImageView(aspect),
+    };
+    const VkFramebuffer vkfb = mFramebufferCache.getFramebuffer(fbkey);
+
+    VkRenderPassBeginInfo renderPassInfo {
+        .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
+        .renderPass = renderPass,
+        .framebuffer = vkfb,
+    };
+    renderPassInfo.renderArea.offset.x = dstRect[0].x;
+    renderPassInfo.renderArea.offset.y = dstRect[0].y;
+    renderPassInfo.renderArea.extent.width = dstRect[1].x - dstRect[0].x;
+    renderPassInfo.renderArea.extent.height = dstRect[1].y - dstRect[0].y;
+
+    // We need to transition the source into a sampler since it'll be sampled in the shader.
+    const VkImageSubresourceRange srcRange = {
+            .aspectMask = aspect,
+            .baseMipLevel = src.level,
+            .levelCount = 1,
+            .baseArrayLayer = src.layer,
+            .layerCount = 1,
+    };
+    src.texture->transitionLayout(cmdbuffer, srcRange, samplerLayout);
+
+    vkCmdBeginRenderPass(cmdbuffer, &renderPassInfo, VK_SUBPASS_CONTENTS_INLINE);
+
+    VkViewport viewport = {
+        .x = (float) dstRect[0].x,
+        .y = (float) dstRect[0].y,
+        .width = (float) renderPassInfo.renderArea.extent.width,
+        .height = (float) renderPassInfo.renderArea.extent.height,
+        .minDepth = 0.0f,
+        .maxDepth = 1.0f
+    };
+
+    vkCmdSetViewport(cmdbuffer, 0, 1, &viewport);
+
+    // DRAW THE TRIANGLE
+    // -----------------
+
+    mPipelineCache.bindProgram(mDepthResolveProgram);
+    mPipelineCache.bindPrimitiveTopology(VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP);
+
+    auto vkraster = mPipelineCache.getCurrentRasterState();
+    vkraster.depthWriteEnable = true;
+    vkraster.depthCompareOp = SamplerCompareFunc::A;
+    vkraster.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT,
+    vkraster.alphaToCoverageEnable = false,
+    vkraster.blendEnable = VK_FALSE,
+    vkraster.srcColorBlendFactor = VK_BLEND_FACTOR_ONE,
+    vkraster.dstColorBlendFactor = VK_BLEND_FACTOR_ZERO,
+    vkraster.colorBlendOp = BlendEquation::ADD,
+    vkraster.srcAlphaBlendFactor = VK_BLEND_FACTOR_ONE,
+    vkraster.dstAlphaBlendFactor = VK_BLEND_FACTOR_ZERO,
+    vkraster.alphaBlendOp = BlendEquation::ADD,
+    vkraster.colorWriteMask = (VkColorComponentFlags) 0,
+    vkraster.cullMode = VK_CULL_MODE_NONE;
+    vkraster.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE;
+    vkraster.depthBiasEnable = VK_FALSE;
+    vkraster.colorTargetCount = 0;
+    mPipelineCache.bindRasterState(vkraster);
+
+    VkBuffer buffers[1] = {};
+    VkDeviceSize offsets[1] = {};
+    buffers[0] = mTriangleBuffer->getGpuBuffer();
+    VkVertexInputAttributeDescription attribDesc = {
+            .location = 0,
+            .binding = 0,
+            .format = VK_FORMAT_R32G32_SFLOAT,
+    };
+    VkVertexInputBindingDescription bufferDesc = {
+            .binding = 0,
+            .stride = sizeof(float) * 2,
+    };
+    mPipelineCache.bindVertexArray(&attribDesc, &bufferDesc, 1);
+
+    // Select nearest filtering and clamp_to_edge.
+    VkSampler vksampler = mSamplerCache.getSampler({});
+
+    VkDescriptorImageInfo samplers[VulkanPipelineCache::SAMPLER_BINDING_COUNT];
+    VulkanTexture* textures[VulkanPipelineCache::SAMPLER_BINDING_COUNT] = {nullptr};
+    for (auto& sampler : samplers) {
+        sampler = {
+            .sampler = vksampler,
+            .imageView = mEmptyTexture->getPrimaryImageView(),
+            .imageLayout = ImgUtil::getVkLayout(VulkanLayout::READ_WRITE),
+        };
+    }
+
+    samplers[0] = {
+        .sampler = vksampler,
+        .imageView = src.getImageView(VK_IMAGE_ASPECT_DEPTH_BIT),
+        .imageLayout = ImgUtil::getVkLayout(samplerLayout),
+    };
+    textures[0] = src.texture;
+
+    mPipelineCache.bindSamplers(samplers, textures,
+            VulkanPipelineCache::getUsageFlags(0, ShaderStageFlags::FRAGMENT));
+
+    auto previousUbo = mPipelineCache.getUniformBufferBinding(0);
+    mPipelineCache.bindUniformBuffer(0, mParamsBuffer->getGpuBuffer());
+
+    if (!mPipelineCache.bindDescriptors(cmdbuffer)) {
+        assert_invariant(false);
+    }
+
+    const VkRect2D scissor = {
+        .offset = renderPassInfo.renderArea.offset,
+        .extent = renderPassInfo.renderArea.extent,
+    };
+
+    mPipelineCache.bindScissor(cmdbuffer, scissor);
+
+    if (!mPipelineCache.bindPipeline(commands)) {
+        assert_invariant(false);
+    }
+
+    vkCmdBindVertexBuffers(cmdbuffer, 0, 1, buffers, offsets);
+    vkCmdDraw(cmdbuffer, 4, 1, 0, 0);
+
+    // END RENDER PASS
+    // ---------------
+
+    vkCmdEndRenderPass(cmdbuffer);
+
+    VkMemoryBarrier barrier {
+        .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+        .srcAccessMask = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
+        .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+    };
+    VkPipelineStageFlags srcStageMask = VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
+    vkCmdPipelineBarrier(cmdbuffer, srcStageMask,
+            VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT | // <== For Mali
+            VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
+            0, 1, &barrier, 0, nullptr, 0, nullptr);
+
+    mPipelineCache.bindUniformBuffer(0, previousUbo.buffer, previousUbo.offset, previousUbo.size);
 }

 } // namespace filament::backend
@@ -32,24 +32,46 @@ struct VulkanProgram;

 class VulkanBlitter {
 public:
-    VulkanBlitter() noexcept;
+    VulkanBlitter(VulkanStagePool& stagePool, VulkanPipelineCache& pipelineCache,
+            VulkanFboCache& fboCache, VulkanSamplerCache& samplerCache) noexcept;

-    void initialize(VkPhysicalDevice physicalDevice, VkDevice device,
-            VmaAllocator allocator, VulkanCommands* commands) noexcept;
+    void initialize(VkPhysicalDevice physicalDevice, VkDevice device, VmaAllocator allocator,
+            VulkanCommands* commands, VulkanTexture* emptyTexture) noexcept;

-    void blit(VkFilter filter,
-            VulkanAttachment dst, const VkOffset3D* dstRectPair,
-            VulkanAttachment src, const VkOffset3D* srcRectPair);
+    struct BlitArgs {
+        const VulkanRenderTarget* dstTarget;
+        const VkOffset3D* dstRectPair;
+        const VulkanRenderTarget* srcTarget;
+        const VkOffset3D* srcRectPair;
+        VkFilter filter = VK_FILTER_NEAREST;
+        int targetIndex = 0;
+    };

-    void resolve(VulkanAttachment dst, VulkanAttachment src);
+    void blitColor(BlitArgs args);
+    void blitDepth(BlitArgs args);

    void terminate() noexcept;

 private:
+    void lazyInit() noexcept;
+
+    void blitSlowDepth(VkFilter filter, const VkExtent2D srcExtent, VulkanAttachment src,
+            VulkanAttachment dst, const VkOffset3D srcRect[2], const VkOffset3D dstRect[2]);
+
+    VulkanBuffer* mTriangleBuffer = nullptr;
+    VulkanBuffer* mParamsBuffer = nullptr;
+    VulkanProgram* mDepthResolveProgram = nullptr;
+
    UTILS_UNUSED VkPhysicalDevice mPhysicalDevice;
    VkDevice mDevice;
    VmaAllocator mAllocator;
    VulkanCommands* mCommands;
+    VulkanTexture* mEmptyTexture;
+
+    VulkanStagePool& mStagePool;
+    VulkanPipelineCache& mPipelineCache;
+    VulkanFboCache& mFramebufferCache;
+    VulkanSamplerCache& mSamplerCache;
 };

 } // namespace filament::backend
@@ -268,18 +268,13 @@ bool VulkanCommands::flush() {
            VK_NULL_HANDLE,
            VK_NULL_HANDLE,
    };
-    uint32_t waitSemaphoreCount = 0;
-    if (mSubmissionSignal) {
-        signals[waitSemaphoreCount++] = mSubmissionSignal;
-    }
-    if (mInjectedSignal) {
-        signals[waitSemaphoreCount++] = mInjectedSignal;
-    }
+
    VkCommandBuffer const cmdbuffer = currentbuf->buffer();
+
    VkSubmitInfo submitInfo{
            .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
-            .waitSemaphoreCount = waitSemaphoreCount,
-            .pWaitSemaphores = waitSemaphoreCount > 0 ? signals : nullptr,
+            .waitSemaphoreCount = 0,
+            .pWaitSemaphores = signals,
            .pWaitDstStageMask = waitDestStageMasks,
            .commandBufferCount = 1,
            .pCommandBuffers = &cmdbuffer,
@@ -287,6 +282,19 @@ bool VulkanCommands::flush() {
            .pSignalSemaphores = &renderingFinished,
    };

+    if (mSubmissionSignal) {
+        signals[submitInfo.waitSemaphoreCount++] = mSubmissionSignal;
+    }
+
+    if (mInjectedSignal) {
+        signals[submitInfo.waitSemaphoreCount++] = mInjectedSignal;
+    }
+
+    // To address a validation warning.
+    if (submitInfo.waitSemaphoreCount == 0) {
+        submitInfo.pWaitSemaphores = VK_NULL_HANDLE;
+    }
+
 #if FVK_ENABLED(FVK_DEBUG_COMMAND_BUFFER)
    slog.i << "Submitting cmdbuffer=" << cmdbuffer
           << " wait=(" << signals[0] << ", " << signals[1] << ") "
@@ -54,8 +54,7 @@ VmaAllocator createAllocator(VkInstance instance, VkPhysicalDevice physicalDevic
    VmaAllocator allocator;
    VmaVulkanFunctions const funcs {
 #if VMA_DYNAMIC_VULKAN_FUNCTIONS
-        .vkGetInstanceProcAddr = vkGetInstanceProcAddr,
-        .vkGetDeviceProcAddr = vkGetDeviceProcAddr,
+        .vkGetInstanceProcAddr = vkGetInstanceProcAddr, .vkGetDeviceProcAddr = vkGetDeviceProcAddr,
 #else
        .vkGetPhysicalDeviceProperties = vkGetPhysicalDeviceProperties,
        .vkGetPhysicalDeviceMemoryProperties = vkGetPhysicalDeviceMemoryProperties,
@@ -156,6 +155,7 @@ VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext const& contex
      mResourceManager(&mResourceAllocator),
      mThreadSafeResourceManager(&mResourceAllocator),
      mPipelineCache(&mResourceAllocator),
+      mBlitter(mStagePool, mPipelineCache, mFramebufferCache, mSamplerCache),
      mReadPixels(mPlatform->getDevice()),
      mIsSRGBSwapChainSupported(mPlatform->getCustomization().isSRGBSwapChainSupported) {

@@ -205,7 +205,7 @@ VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext const& contex

    mPipelineCache.setDummyTexture(mEmptyTexture->getPrimaryImageView());
    mBlitter.initialize(mPlatform->getPhysicalDevice(), mPlatform->getDevice(), mAllocator,
-            mCommands.get());
+            mCommands.get(), mEmptyTexture.get());
 }

 VulkanDriver::~VulkanDriver() noexcept = default;
@@ -426,8 +426,7 @@ void VulkanDriver::destroyTexture(Handle<HwTexture> th) {
 }

 void VulkanDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {
-    auto vkprogram
-            = mResourceAllocator.construct<VulkanProgram>(ph, mPlatform->getDevice(), program);
+    auto vkprogram = mResourceAllocator.construct<VulkanProgram>(ph, mPlatform->getDevice(), program);
    mResourceManager.acquire(vkprogram);
 }

@@ -756,7 +755,6 @@ bool VulkanDriver::isFrameTimeSupported() {
 }

 bool VulkanDriver::isAutoDepthResolveSupported() {
-    // TODO: this could be supported with vk 1.2 or VK_KHR_depth_stencil_resolve
    return false;
 }

@@ -772,11 +770,6 @@ bool VulkanDriver::isParallelShaderCompileSupported() {
    return false;
 }

-bool VulkanDriver::isDepthStencilResolveSupported() {
-    // TODO: apparently it could be supported in core 1.2 and/or with VK_KHR_depth_stencil_resolve
-    return false;
-}
-
 bool VulkanDriver::isWorkaroundNeeded(Workaround workaround) {
    switch (workaround) {
        case Workaround::SPLIT_EASU: {
@@ -960,43 +953,10 @@ void VulkanDriver::setExternalImagePlane(Handle<HwTexture> th, void* image, uint
 void VulkanDriver::setExternalStream(Handle<HwTexture> th, Handle<HwStream> sh) {
 }

-void VulkanDriver::generateMipmaps(Handle<HwTexture> th) {
-    auto* const t = mResourceAllocator.handle_cast<VulkanTexture*>(th);
-    assert_invariant(t);
+void VulkanDriver::generateMipmaps(Handle<HwTexture> th) { }

-    int32_t layerCount = int32_t(t->depth);
-    if (t->target == SamplerType::SAMPLER_CUBEMAP_ARRAY ||
-        t->target == SamplerType::SAMPLER_CUBEMAP) {
-        layerCount *= 6;
-    }
-
-    // FIXME: the loop below can perform many layout transitions and back. We should be
-    //        able to optimize that.
-
-    uint8_t level = 0;
-    int32_t srcw = int32_t(t->width);
-    int32_t srch = int32_t(t->height);
-    do {
-        int32_t const dstw = std::max(srcw >> 1, 1);
-        int32_t const dsth = std::max(srch >> 1, 1);
-        const VkOffset3D srcOffsets[2] = {{ 0, 0, 0 }, { srcw, srch, 1 }};
-        const VkOffset3D dstOffsets[2] = {{ 0, 0, 0 }, { dstw, dsth, 1 }};
-
-        // TODO: there should be a way to do this using layerCount in vkBlitImage
-        // TODO: vkBlitImage should be able to handle 3D textures too
-        for (int32_t layer = 0; layer < layerCount; layer++) {
-            mBlitter.blit(VK_FILTER_LINEAR,
-                    { .texture = t, .level = uint8_t(level + 1), .layer = (uint16_t)layer },
-                    dstOffsets,
-                    { .texture = t, .level = uint8_t(level    ), .layer = (uint16_t)layer },
-                    srcOffsets);
-        }
-
-        level++;
-        srcw = dstw;
-        srch = dsth;
-    } while ((srcw > 1 || srch > 1) && level < t->levels);
-    t->setPrimaryRange(0, t->levels - 1);
+bool VulkanDriver::canGenerateMipmaps() {
+    return false;
 }

 void VulkanDriver::updateSamplerGroup(Handle<HwSamplerGroup> sbh,
@@ -1477,138 +1437,50 @@ void VulkanDriver::readBufferSubData(backend::BufferObjectHandle boh,
    scheduleDestroy(std::move(p));
 }

-void VulkanDriver::resolve(
-        Handle<HwTexture> dst, uint8_t srcLevel, uint8_t srcLayer,
-        Handle<HwTexture> src, uint8_t dstLevel, uint8_t dstLayer) {
-    FVK_SYSTRACE_CONTEXT();
-    FVK_SYSTRACE_START("resolve");
-
-    ASSERT_PRECONDITION(mCurrentRenderPass.renderPass == VK_NULL_HANDLE,
-            "resolve() cannot be invoked inside a render pass.");
-
-    auto* const srcTexture = mResourceAllocator.handle_cast<VulkanTexture*>(src);
-    auto* const dstTexture = mResourceAllocator.handle_cast<VulkanTexture*>(dst);
-    assert_invariant(srcTexture);
-    assert_invariant(dstTexture);
-
-    ASSERT_PRECONDITION(
-            dstTexture->width == srcTexture->width && dstTexture->height == srcTexture->height,
-            "invalid resolve: src and dst sizes don't match");
-
-    ASSERT_PRECONDITION(srcTexture->samples > 1 && dstTexture->samples == 1,
-            "invalid resolve: src.samples=%u, dst.samples=%u",
-            +srcTexture->samples, +dstTexture->samples);
-
-    ASSERT_PRECONDITION(srcTexture->format == dstTexture->format,
-            "src and dst texture format don't match");
-
-    ASSERT_PRECONDITION(!isDepthFormat(srcTexture->format),
-            "can't resolve depth formats");
-
-    ASSERT_PRECONDITION(!isStencilFormat(srcTexture->format),
-            "can't resolve stencil formats");
-
-    ASSERT_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST),
-            "texture doesn't have BLIT_DST");
-
-    ASSERT_PRECONDITION(any(srcTexture->usage & TextureUsage::BLIT_SRC),
-            "texture doesn't have BLIT_SRC");
-
-    mBlitter.resolve(
-            { .texture = dstTexture, .level = dstLevel, .layer = dstLayer },
-            { .texture = srcTexture, .level = srcLevel, .layer = srcLayer });
-
-    FVK_SYSTRACE_END();
-}
-
-void VulkanDriver::blit(
-        Handle<HwTexture> dst, uint8_t srcLevel, uint8_t srcLayer, math::uint2 dstOrigin,
-        Handle<HwTexture> src, uint8_t dstLevel, uint8_t dstLayer, math::uint2 srcOrigin,
-        math::uint2 size) {
+void VulkanDriver::blit(TargetBufferFlags buffers, Handle<HwRenderTarget> dst, Viewport dstRect,
+        Handle<HwRenderTarget> src, Viewport srcRect, SamplerMagFilter filter) {
    FVK_SYSTRACE_CONTEXT();
    FVK_SYSTRACE_START("blit");

-    ASSERT_PRECONDITION(mCurrentRenderPass.renderPass == VK_NULL_HANDLE,
-            "blit() cannot be invoked inside a render pass.");
+    assert_invariant(mCurrentRenderPass.renderPass == VK_NULL_HANDLE);

-    auto* const srcTexture = mResourceAllocator.handle_cast<VulkanTexture*>(src);
-    auto* const dstTexture = mResourceAllocator.handle_cast<VulkanTexture*>(dst);
+    // blit operation only support COLOR0 color buffer
+    assert_invariant(
+            !(buffers & (TargetBufferFlags::COLOR_ALL & ~TargetBufferFlags::COLOR0)));

-    ASSERT_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST),
-            "texture doesn't have BLIT_DST");
-
-    ASSERT_PRECONDITION(any(srcTexture->usage & TextureUsage::BLIT_SRC),
-            "texture doesn't have BLIT_SRC");
-
-    ASSERT_PRECONDITION(srcTexture->format == dstTexture->format,
-            "src and dst texture format don't match");
-
-    // The Y inversion below makes it so that Vk matches GL and Metal.
-
-    auto const srcLeft   = int32_t(srcOrigin.x);
-    auto const dstLeft   = int32_t(dstOrigin.x);
-    auto const srcTop    = int32_t(srcTexture->height - (srcOrigin.y + size.y));
-    auto const dstTop    = int32_t(dstTexture->height - (dstOrigin.y + size.y));
-    auto const srcRight  = int32_t(srcOrigin.x + size.x);
-    auto const dstRight  = int32_t(dstOrigin.x + size.x);
-    auto const srcBottom = int32_t(srcTop + size.y);
-    auto const dstBottom = int32_t(dstTop + size.y);
-    VkOffset3D const srcOffsets[2] = { { srcLeft, srcTop, 0 }, { srcRight, srcBottom, 1 }};
-    VkOffset3D const dstOffsets[2] = { { dstLeft, dstTop, 0 }, { dstRight, dstBottom, 1 }};
-
-    // no scaling guaranteed
-    mBlitter.blit(VK_FILTER_NEAREST,
-            { .texture = dstTexture, .level = dstLevel, .layer = dstLayer }, dstOffsets,
-            { .texture = srcTexture, .level = srcLevel, .layer = srcLayer }, srcOffsets);
-
-    FVK_SYSTRACE_END();
-}
-
-void VulkanDriver::blitDEPRECATED(TargetBufferFlags buffers,
-        Handle<HwRenderTarget> dst, Viewport dstRect,
-        Handle<HwRenderTarget> src, Viewport srcRect,
-        SamplerMagFilter filter) {
-    FVK_SYSTRACE_CONTEXT();
-    FVK_SYSTRACE_START("blitDEPRECATED");
-
-    // Note: blitDEPRECATED is only used for Renderer::copyFrame()
-
-    ASSERT_PRECONDITION(mCurrentRenderPass.renderPass == VK_NULL_HANDLE,
-            "blitDEPRECATED() cannot be invoked inside a render pass.");
-
-    ASSERT_PRECONDITION(buffers == TargetBufferFlags::COLOR0,
-            "blitDEPRECATED only supports COLOR0");
-
-    ASSERT_PRECONDITION(srcRect.left >= 0 && srcRect.bottom >= 0 &&
-                        dstRect.left >= 0 && dstRect.bottom >= 0,
-            "Source and destination rects must be positive.");
+    if (UTILS_UNLIKELY(mCurrentRenderPass.renderPass)) {
+        utils::slog.e << "Blits cannot be invoked inside a render pass." << utils::io::endl;
+        return;
+    }

    VulkanRenderTarget* dstTarget = mResourceAllocator.handle_cast<VulkanRenderTarget*>(dst);
    VulkanRenderTarget* srcTarget = mResourceAllocator.handle_cast<VulkanRenderTarget*>(src);

-    VkFilter const vkfilter = (filter == SamplerMagFilter::NEAREST) ?
-            VK_FILTER_NEAREST : VK_FILTER_LINEAR;
+    VkFilter vkfilter = filter == SamplerMagFilter::NEAREST ? VK_FILTER_NEAREST : VK_FILTER_LINEAR;

    // The Y inversion below makes it so that Vk matches GL and Metal.

-    VkExtent2D const srcExtent = srcTarget->getExtent();
-    VkExtent2D const dstExtent = dstTarget->getExtent();
+    const VkExtent2D srcExtent = srcTarget->getExtent();
+    const int32_t srcLeft = srcRect.left;
+    const int32_t srcTop = srcExtent.height - srcRect.bottom - srcRect.height;
+    const int32_t srcRight = srcRect.left + srcRect.width;
+    const int32_t srcBottom = srcTop + srcRect.height;
+    const VkOffset3D srcOffsets[2] = { { srcLeft, srcTop, 0 }, { srcRight, srcBottom, 1 }};

-    auto const dstLeft   = int32_t(dstRect.left);
-    auto const srcLeft   = int32_t(srcRect.left);
-    auto const dstTop    = int32_t(dstExtent.height - (dstRect.bottom + dstRect.height));
-    auto const srcTop    = int32_t(srcExtent.height - (srcRect.bottom + srcRect.height));
-    auto const dstRight  = int32_t(dstRect.left + dstRect.width);
-    auto const srcRight  = int32_t(srcRect.left + srcRect.width);
-    auto const dstBottom = int32_t(dstTop + dstRect.height);
-    auto const srcBottom = int32_t(srcTop + srcRect.height);
-    VkOffset3D const srcOffsets[2] = { { srcLeft, srcTop, 0 }, { srcRight, srcBottom, 1 }};
-    VkOffset3D const dstOffsets[2] = { { dstLeft, dstTop, 0 }, { dstRight, dstBottom, 1 }};
+    const VkExtent2D dstExtent = dstTarget->getExtent();
+    const int32_t dstLeft = dstRect.left;
+    const int32_t dstTop = dstExtent.height - dstRect.bottom - dstRect.height;
+    const int32_t dstRight = dstRect.left + dstRect.width;
+    const int32_t dstBottom = dstTop + dstRect.height;
+    const VkOffset3D dstOffsets[2] = { { dstLeft, dstTop, 0 }, { dstRight, dstBottom, 1 }};

-    mBlitter.blit(vkfilter,
-            dstTarget->getColor(0), dstOffsets,
-            srcTarget->getColor(0), srcOffsets);
+    if (any(buffers & TargetBufferFlags::DEPTH) && srcTarget->hasDepth() && dstTarget->hasDepth()) {
+        mBlitter.blitDepth({dstTarget, dstOffsets, srcTarget, srcOffsets});
+    }

+    if (any(buffers & TargetBufferFlags::COLOR0)) {
+        mBlitter.blitColor({ dstTarget, dstOffsets, srcTarget, srcOffsets, vkfilter, int(0) });
+    }
    FVK_SYSTRACE_END();
 }

@@ -18,11 +18,10 @@

 #include "VulkanConstants.h"
 #include "VulkanMemory.h"
-#include "spirv/VulkanSpirvUtils.h"

 #include <backend/platforms/VulkanPlatform.h>

-#include <utils/Panic.h>    // ASSERT_POSTCONDITION
+#include <utils/Panic.h>

 using namespace bluevk;

@@ -47,39 +46,61 @@ static void clampToFramebuffer(VkRect2D* rect, uint32_t fbWidth, uint32_t fbHeig
    rect->extent.height = std::max(top - y, 0);
 }

-VulkanProgram::VulkanProgram(VkDevice device, Program const& builder) noexcept
+VulkanProgram::VulkanProgram(VkDevice device, const Program& builder) noexcept
    : HwProgram(builder.getName()),
      VulkanResource(VulkanResourceType::PROGRAM),
-      mInfo(new PipelineInfo()),
+      mInfo(new PipelineInfo(builder.getSpecializationConstants().size())),
      mDevice(device) {
-    Program::ShaderSource const& blobs = builder.getShadersSource();
+    auto& blobs = builder.getShadersSource();
    auto& modules = mInfo->shaders;
-
-    auto const& specializationConstants = builder.getSpecializationConstants();
-
-    std::vector<uint32_t> shader;
    for (size_t i = 0; i < MAX_SHADER_MODULES; i++) {
-        Program::ShaderBlob const& blob = blobs[i];
-
-        uint32_t* data = (uint32_t*) blob.data();
-        size_t dataSize = blob.size();
-
-        if (!specializationConstants.empty()) {
-            workaroundSpecConstant(blob, specializationConstants, shader);
-            data = (uint32_t*) shader.data();
-            dataSize = shader.size() * 4;
-        }
-
+        const auto& blob = blobs[i];
+        uint32_t* data = (uint32_t*)blob.data();
        VkShaderModule& module = modules[i];
        VkShaderModuleCreateInfo moduleInfo = {
            .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
-            .codeSize = dataSize,
+            .codeSize = blob.size(),
            .pCode = data,
        };
        VkResult result = vkCreateShaderModule(mDevice, &moduleInfo, VKALLOC, &module);
        ASSERT_POSTCONDITION(result == VK_SUCCESS, "Unable to create shader module.");
    }

+    // Note that bools are 4-bytes in Vulkan
+    // https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/VkBool32.html
+    constexpr uint32_t const CONSTANT_SIZE = 4;
+
+    // populate the specialization constants requirements right now
+    auto const& specializationConstants = builder.getSpecializationConstants();
+    uint32_t const specConstCount = static_cast<uint32_t>(specializationConstants.size());
+    char* specData = mInfo->specConstData.get();
+    if (specConstCount > 0) {
+        mInfo->specializationInfo = {
+            .mapEntryCount = specConstCount,
+            .pMapEntries = mInfo->specConsts.data(),
+            .dataSize = specConstCount * CONSTANT_SIZE,
+            .pData = specData,
+        };
+    }
+    for (uint32_t i = 0; i < specConstCount; ++i) {
+        uint32_t const offset = i * CONSTANT_SIZE;
+        mInfo->specConsts[i] = {
+            .constantID = specializationConstants[i].id,
+            .offset = offset,
+            .size = CONSTANT_SIZE,
+        };
+        using SpecConstant = Program::SpecializationConstant::Type;
+        char const* addr = (char*)specData + offset;
+        SpecConstant const& arg = specializationConstants[i].value;
+        if (std::holds_alternative<bool>(arg)) {
+            *((VkBool32*)addr) = std::get<bool>(arg) ? VK_TRUE : VK_FALSE;
+        } else if (std::holds_alternative<float>(arg)) {
+            *((float*)addr) = std::get<float>(arg);
+        } else {
+            *((int32_t*)addr) = std::get<int32_t>(arg);
+        }
+    }
+
    auto& groupInfo = builder.getSamplerGroupInfo();
    auto& bindingToSamplerIndex = mInfo->bindingToSamplerIndex;
    auto& usage = mInfo->usage;
@@ -102,7 +123,7 @@ VulkanProgram::VulkanProgram(VkDevice device, Program const& builder) noexcept
 VulkanProgram::VulkanProgram(VkDevice device, VkShaderModule vs, VkShaderModule fs,
        CustomSamplerInfoList const& samplerInfo) noexcept
    : VulkanResource(VulkanResourceType::PROGRAM),
-      mInfo(new PipelineInfo()),
+      mInfo(new PipelineInfo(0)),
      mDevice(device) {
    mInfo->shaders[0] = vs;
    mInfo->shaders[1] = fs;
@@ -25,6 +25,7 @@
 #include "VulkanResources.h"
 #include "VulkanSwapChain.h"
 #include "VulkanTexture.h"
+#include "VulkanUtility.h"

 #include "private/backend/SamplerGroup.h"

@@ -37,7 +38,7 @@ class VulkanTimestamps;

 struct VulkanProgram : public HwProgram, VulkanResource {

-    VulkanProgram(VkDevice device, Program const& builder) noexcept;
+    VulkanProgram(VkDevice device, const Program& builder) noexcept;

    struct CustomSamplerInfo {
        uint8_t groupIndex;
@@ -64,14 +65,21 @@ struct VulkanProgram : public HwProgram, VulkanResource {
        return mInfo->bindingToSamplerIndex;
    }

+    inline VkSpecializationInfo const& getSpecConstInfo() const {
+        return mInfo->specializationInfo;
+    }
+
 private:
    // TODO: handle compute shaders.
    // The expected order of shaders - from frontend to backend - is vertex, fragment, compute.
-    static constexpr uint8_t const MAX_SHADER_MODULES = 2;
+    static constexpr uint8_t MAX_SHADER_MODULES = 2;

    struct PipelineInfo {
-        PipelineInfo()
-            : bindingToSamplerIndex(MAX_SAMPLER_COUNT, 0xffff) {}
+        PipelineInfo(size_t specConstsCount) :
+            bindingToSamplerIndex(MAX_SAMPLER_COUNT, 0xffff),
+            specConsts(specConstsCount, VkSpecializationMapEntry{}),
+            specConstData(new char[specConstsCount * 4])
+        {}

        // This bitset maps to each of the sampler in the sampler groups associated with this
        // program, and whether each sampler is used in which shader (i.e. vert, frag, compute).
@@ -79,7 +87,10 @@ private:

        // We store the samplerGroupIndex as the top 8-bit and the index within each group as the lower 8-bit.
        utils::FixedCapacityVector<uint16_t> bindingToSamplerIndex;
-        VkShaderModule shaders[MAX_SHADER_MODULES] = { VK_NULL_HANDLE };
+        VkShaderModule shaders[MAX_SHADER_MODULES] = {VK_NULL_HANDLE};
+        VkSpecializationInfo specializationInfo = {};
+        utils::FixedCapacityVector<VkSpecializationMapEntry> specConsts;
+        std::unique_ptr<char[]> specConstData;
    };

    PipelineInfo* mInfo;
@@ -360,7 +360,9 @@ VulkanPipelineCache::PipelineCacheEntry* VulkanPipelineCache::createPipeline() n

    // If we reach this point, we need to create and stash a brand new pipeline object.
    shaderStages[0].module = mPipelineRequirements.shaders[0];
+    shaderStages[0].pSpecializationInfo = mSpecializationRequirements;
    shaderStages[1].module = mPipelineRequirements.shaders[1];
+    shaderStages[1].pSpecializationInfo = mSpecializationRequirements;

    // Expand our size-optimized structs into the proper Vk structs.
    uint32_t numVertexAttribs = 0;
@@ -566,6 +568,7 @@ VulkanPipelineCache::PipelineLayoutCacheEntry* VulkanPipelineCache::getOrCreateP
 void VulkanPipelineCache::bindProgram(VulkanProgram* program) noexcept {
    mPipelineRequirements.shaders[0] = program->getVertexShader();
    mPipelineRequirements.shaders[1] = program->getFragmentShader();
+    mSpecializationRequirements = &program->getSpecConstInfo();
 }

 void VulkanPipelineCache::bindRasterState(const RasterState& rasterState) noexcept {
--- a/Show More
+++ b/Show More