wip: shader compilation benchmark

2024-01-16 17:03:41 -08:00
188 changed files with 1771 additions and 3083 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:
@@ -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
@@ -634,9 +634,9 @@ function(combine_static_libs TARGET OUTPUT DEPS)
    # Loop through the dependent libraries and query their location on disk.
    set(DEPS_FILES )
    foreach(DEPENDENCY ${DEPS})
-        if (TARGET ${DEPENDENCY})
+        if(TARGET ${DEPENDENCY})
            get_property(dep_type TARGET ${DEPENDENCY} PROPERTY TYPE)
-            if (dep_type STREQUAL "STATIC_LIBRARY")
+            if(dep_type STREQUAL "STATIC_LIBRARY")
                list(APPEND DEPS_FILES "$<TARGET_FILE:${DEPENDENCY}>")
            endif()
        endif()
@@ -706,7 +706,7 @@ function(get_resgen_vars ARCHIVE_DIR ARCHIVE_NAME)
        set(RESGEN_OUTPUTS "${OUTPUTS}" PARENT_SCOPE)
        set(RESGEN_FLAGS -qx ${ARCHIVE_DIR} -p ${ARCHIVE_NAME} PARENT_SCOPE)
        set(RESGEN_SOURCE "${ARCHIVE_DIR}/${ARCHIVE_NAME}${ASM_SUFFIX}.S" PARENT_SCOPE)
-        set(RESGEN_SOURCE_FLAGS "-I'${ARCHIVE_DIR}' ${ASM_ARCH_FLAG}" PARENT_SCOPE)
+        set(RESGEN_SOURCE_FLAGS "-I${ARCHIVE_DIR} ${ASM_ARCH_FLAG}" PARENT_SCOPE)
    endif()
 endfunction()

@@ -721,6 +721,7 @@ add_subdirectory(${LIBRARIES}/filabridge)
 add_subdirectory(${LIBRARIES}/filaflat)
 add_subdirectory(${LIBRARIES}/filagui)
 add_subdirectory(${LIBRARIES}/filameshio)
+add_subdirectory(${LIBRARIES}/geometry)
 add_subdirectory(${LIBRARIES}/gltfio)
 add_subdirectory(${LIBRARIES}/ibl)
 add_subdirectory(${LIBRARIES}/iblprefilter)
@@ -748,9 +749,6 @@ add_subdirectory(${EXTERNAL}/jsmn/tnt)
 add_subdirectory(${EXTERNAL}/stb/tnt)
 add_subdirectory(${EXTERNAL}/getopt)

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

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

 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.50.2'
+    implementation 'com.google.android.filament:filament-android:1.49.3'
 }
 ```

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

 ```shell
-pod 'Filament', '~> 1.50.2'
+pod 'Filament', '~> 1.49.3'
 ```

 ### Snapshots
@@ -7,18 +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**]

@@ -484,8 +484,7 @@ extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBu
 extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBuilderConfig(JNIEnv*,
        jclass, jlong nativeBuilder, jlong commandBufferSizeMB, jlong perRenderPassArenaSizeMB,
        jlong driverHandleArenaSizeMB, jlong minCommandBufferSizeMB, jlong perFrameCommandsSizeMB,
-        jlong jobSystemThreadCount, jlong stereoscopicEyeCount,
-        jlong resourceAllocatorCacheSizeMB, jlong resourceAllocatorCacheMaxAge) {
+        jlong jobSystemThreadCount, jlong stereoscopicEyeCount) {
    Engine::Builder* builder = (Engine::Builder*) nativeBuilder;
    Engine::Config config = {
            .commandBufferSizeMB = (uint32_t) commandBufferSizeMB,
@@ -495,8 +494,6 @@ extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBu
            .perFrameCommandsSizeMB = (uint32_t) perFrameCommandsSizeMB,
            .jobSystemThreadCount = (uint32_t) jobSystemThreadCount,
            .stereoscopicEyeCount = (uint8_t) stereoscopicEyeCount,
-            .resourceAllocatorCacheSizeMB = (uint32_t) resourceAllocatorCacheSizeMB,
-            .resourceAllocatorCacheMaxAge = (uint8_t) resourceAllocatorCacheMaxAge,
    };
    builder->config(&config);
 }
@@ -211,8 +211,7 @@ public class Engine {
            nSetBuilderConfig(mNativeBuilder, config.commandBufferSizeMB,
                    config.perRenderPassArenaSizeMB, config.driverHandleArenaSizeMB,
                    config.minCommandBufferSizeMB, config.perFrameCommandsSizeMB,
-                    config.jobSystemThreadCount, config.stereoscopicEyeCount,
-                    config.resourceAllocatorCacheSizeMB, config.resourceAllocatorCacheMaxAge);
+                    config.jobSystemThreadCount, config.stereoscopicEyeCount);
            return this;
        }

@@ -357,18 +356,6 @@ public class Engine {
         * @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) {
@@ -1240,8 +1227,7 @@ public class Engine {
    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 stereoscopicEyeCount);
    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);
@@ -1882,7 +1882,7 @@ public class View {
     * Options for Temporal Anti-aliasing (TAA)
     * Most TAA parameters are extremely costly to change, as they will trigger the TAA post-process
     * shaders to be recompiled. These options should be changed or set during initialization.
-     * `filterWidth`, `feedback` and `jitterPattern`, however, can be changed at any time.
+     * `filterWidth`, `feedback` and `jitterPattern`, however, could be changed at any time.
     *
     * `feedback` of 0.1 effectively accumulates a maximum of 19 samples in steady state.
     * see "A Survey of Temporal Antialiasing Techniques" by Lei Yang and all for more information.
@@ -1925,7 +1925,6 @@ public class View {
            UNIFORM_HELIX_X4,
            HALTON_23_X8,
            HALTON_23_X16,
-            HALTON_23_X32,
        }

        /**
@@ -1940,10 +1939,6 @@ public class View {
         * texturing lod bias (typically -1 or -2)
         */
        public float lodBias = -1.0f;
-        /**
-         * post-TAA sharpen, especially useful when upscaling is true.
-         */
-        public float sharpness = 0.0f;
        /**
         * enables or disables temporal anti-aliasing
         */
@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.50.2
+VERSION_NAME=1.49.3

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

@@ -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.
@@ -238,7 +238,6 @@ set(MATERIAL_SRCS
        src/materials/resolveDepth.mat
        src/materials/separableGaussianBlur.mat
        src/materials/skybox.mat
-        src/materials/shadowmap.mat
        src/materials/ssao/bilateralBlur.mat
        src/materials/ssao/bilateralBlurBentNormals.mat
        src/materials/ssao/mipmapDepth.mat
@@ -23,6 +23,7 @@
 #include <utils/ostream.h>

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

 namespace filament::backend {

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

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

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

 #include <math/vec4.h>

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

 #include <stddef.h>
 #include <stdint.h>
@@ -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>
@@ -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 +17,7 @@
 #ifndef TNT_FILAMENT_BACKEND_PRIVATE_CIRCULARBUFFER_H
 #define TNT_FILAMENT_BACKEND_PRIVATE_CIRCULARBUFFER_H

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

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

    ~CircularBuffer() noexcept;

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

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

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

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

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

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

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

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

 #include "private/backend/CircularBuffer.h"

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

 #include <vector>

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

 /*
 * A producer-consumer command queue that uses a CircularBuffer as main storage
 */
 class CommandBufferQueue {
-    struct Range {
+    struct Slice {
        void* begin;
        void* end;
    };
@@ -46,7 +44,7 @@ class CommandBufferQueue {

    mutable utils::Mutex mLock;
    mutable utils::Condition mCondition;
-    mutable std::vector<Range> mCommandBuffersToExecute;
+    mutable std::vector<Slice> mCommandBuffersToExecute;
    size_t mFreeSpace = 0;
    size_t mHighWatermark = 0;
    uint32_t mExitRequested = 0;
@@ -58,20 +56,17 @@ public:
    CommandBufferQueue(size_t requiredSize, size_t bufferSize);
    ~CommandBufferQueue();

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

    size_t getHighWatermark() const noexcept { return mHighWatermark; }

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

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

    // all commands buffers (Slices) written to this point are returned by waitForCommand(). This
    // call blocks until the CircularBuffer has at least mRequiredSize bytes available.
@@ -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>

 // ------------------------------------------------------------------------------------------------
@@ -213,8 +211,6 @@ public:
    CommandStream(CommandStream const& rhs) noexcept = delete;
    CommandStream& operator=(CommandStream const& rhs) noexcept = delete;

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

 #include <functional>

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

 // Command debugging off. debugging virtuals are not called.
@@ -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 = 0) 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>

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

 #include "private/backend/CircularBuffer.h"

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

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

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

 namespace filament::backend {

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

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

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


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

 } // namespace filament::backend
@@ -15,25 +15,14 @@
 */

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

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

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

 using namespace utils;

@@ -76,53 +65,50 @@ void CommandBufferQueue::flush() noexcept {
    // always guaranteed to have enough space for the NoopCommand
    new(circularBuffer.allocate(sizeof(NoopCommand))) NoopCommand(nullptr);

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

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

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

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

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

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

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

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

+    mCondition.notify_one();
+    if (UTILS_LIKELY(mFreeSpace < requiredSize)) {
        SYSTRACE_NAME("waiting: CircularBuffer::flush()");
        mCondition.wait(lock, [this, requiredSize]() -> bool {
-            // TODO: on macOS, we need to call pumpEvents from time to time
            return mFreeSpace >= requiredSize;
        });
    }
 }

-std::vector<CommandBufferQueue::Range> CommandBufferQueue::waitForCommands() const {
+std::vector<CommandBufferQueue::Slice> CommandBufferQueue::waitForCommands() const {
    if (!UTILS_HAS_THREADING) {
        return std::move(mCommandBuffersToExecute);
    }
@@ -137,7 +123,7 @@ std::vector<CommandBufferQueue::Range> CommandBufferQueue::waitForCommands() con
    return std::move(mCommandBuffersToExecute);
 }

-void CommandBufferQueue::releaseBuffer(CommandBufferQueue::Range const& buffer) {
+void CommandBufferQueue::releaseBuffer(CommandBufferQueue::Slice const& buffer) {
    std::lock_guard<utils::Mutex> const lock(mLock);
    mFreeSpace += uintptr_t(buffer.end) - uintptr_t(buffer.begin);
    mCondition.notify_one();
@@ -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
    });
 }
@@ -20,7 +20,6 @@
 #include "MetalUtils.h"

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

 namespace filament::backend {

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

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

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

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

 #define NSERROR_CHECK(message)                                                                     \
@@ -245,53 +245,30 @@ void MetalSwapChain::present() {
    }
 }

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

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

 void MetalSwapChain::scheduleFrameScheduledCallback() {
@@ -300,16 +277,21 @@ void MetalSwapChain::scheduleFrameScheduledCallback() {
    }

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

@@ -18,8 +18,6 @@

 #include "MetalEnums.h"

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

@@ -427,6 +427,7 @@ void OpenGLContext::initBugs(Bugs* bugs, Extensions const& exts,
            if (strstr(renderer, "Mali-T")) {
                bugs->disable_glFlush = true;
                bugs->disable_shared_context_draws = true;
+                bugs->texture_external_needs_rebind = true;
                // We have not verified that timer queries work on Mali-T, so we disable to be safe.
                bugs->dont_use_timer_query = true;
            }
@@ -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;
 }

@@ -171,8 +171,7 @@ 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);

@@ -400,9 +400,6 @@ 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;
@@ -151,12 +151,6 @@ bool ShaderCompilerService::isParallelShaderCompileSupported() const noexcept {
 }

 void ShaderCompilerService::init() noexcept {
-    if (UTILS_UNLIKELY(mDriver.getDriverConfig().disableParallelShaderCompile)) {
-        // user disabled parallel shader compile
-        mMode = Mode::SYNCHRONOUS;
-        return;
-    }
-
    // Here we decide which mode we'll be using. We always prefer our own thread-pool if
    // that mode is available because, we have no control on how the compilation queues are
    // handled if done by the driver (so at the very least we'd need to decode this per-driver).
@@ -109,7 +109,7 @@ private:
    };

    enum class Mode {
-        UNDEFINED,      // init() has not been called yet.
+        UNDEFINED,      // init() has not beed 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
@@ -15,7 +15,6 @@
 */

 #include <backend/platforms/VulkanPlatform.h>
-#include <backend/DriverEnums.h>

 #include "vulkan/VulkanConstants.h"
 #include "vulkan/VulkanDriverFactory.h"
@@ -24,11 +23,6 @@

 #include <bluevk/BlueVK.h>

-#include <tuple>
-
-#include <stdint.h>
-#include <stddef.h>
-
 #if defined(__linux__) || defined(__FreeBSD__)
 #define LINUX_OR_FREEBSD 1
 #endif
@@ -21,14 +21,10 @@

 #include <utils/compiler.h>

-#include <math/mat3.h>
+#include <limits>
+
 #include <math/mat4.h>
 #include <math/vec3.h>
-#include <math/vec4.h>
-
-#include <float.h>
-
-#include <stddef.h>

 namespace filament {

@@ -74,7 +70,7 @@ public:
     * @return This bounding box
     */
    Box& set(const math::float3& min, const math::float3& max) noexcept {
-        // float3 ctor needed for Visual Studio
+        // float3 ctor needed for visual studio
        center     = (max + min) * math::float3(0.5f);
        halfExtent = (max - min) * math::float3(0.5f);
        return *this;
@@ -134,17 +130,17 @@ public:
 struct UTILS_PUBLIC Aabb {

    /** min coordinates */
-    math::float3 min = FLT_MAX;
+    math::float3 min = std::numeric_limits<float>::max();

    /** max coordinates */
-    math::float3 max = -FLT_MAX;
+    math::float3 max = std::numeric_limits<float>::lowest();

    /**
     * Computes the center of the box.
     * @return (max + min)/2
     */
    math::float3 center() const noexcept {
-        // float3 ctor needed for Visual Studio
+        // float3 ctor needed for visual studio
        return (max + min) * math::float3(0.5f);
    }

@@ -153,7 +149,7 @@ struct UTILS_PUBLIC Aabb {
     * @return (max - min)/2
     */
    math::float3 extent() const noexcept {
-        // float3 ctor needed for Visual Studio
+        // float3 ctor needed for visual studio
        return (max - min) * math::float3(0.5f);
    }

@@ -202,14 +198,12 @@ struct UTILS_PUBLIC Aabb {
     * @return the maximum signed distance to the box. Negative if p is in the box
     */
    float contains(math::float3 p) const noexcept {
-        // we don't use std::max to avoid a dependency on <algorithm>
-        auto const maximum = [](auto a, auto b) { return a > b ? a : b; };
        float d = min.x - p.x;
-        d = maximum(d, min.y - p.y);
-        d = maximum(d, min.z - p.z);
-        d = maximum(d, p.x - max.x);
-        d = maximum(d, p.y - max.y);
-        d = maximum(d, p.z - max.z);
+        d = std::max(d, min.y - p.y);
+        d = std::max(d, min.z - p.z);
+        d = std::max(d, p.x - max.x);
+        d = std::max(d, p.y - max.y);
+        d = std::max(d, p.z - max.z);
        return d;
    }

@@ -22,13 +22,11 @@
 #include <filament/FilamentAPI.h>

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

 #include <utils/compiler.h>

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

 class FBufferObject;
@@ -84,7 +82,8 @@ public:
         *
         * @param engine Reference to the filament::Engine to associate this BufferObject with.
         *
-         * @return pointer to the newly created object
+         * @return pointer to the newly created object or nullptr if exceptions are disabled and
+         *         an error occurred.
         *
         * @exception utils::PostConditionPanic if a runtime error occurred, such as running out of
         *            memory or other resources.
@@ -92,7 +91,7 @@ public:
         *
         * @see IndexBuffer::setBuffer
         */
-        BufferObject* UTILS_NONNULL build(Engine& engine);
+        BufferObject* build(Engine& engine);
    private:
        friend class FBufferObject;
    };
@@ -28,11 +28,6 @@
 #include <math/vec4.h>
 #include <math/mat4.h>

-#include <math.h>
-
-#include <stdint.h>
-#include <stddef.h>
-
 namespace utils {
 class Entity;
 } // namespace utils
@@ -189,7 +184,7 @@ public:
     * @see Fov.
     */
    static math::mat4 projection(Fov direction, double fovInDegrees,
-            double aspect, double near, double far = INFINITY);
+            double aspect, double near, double far = std::numeric_limits<double>::infinity());

    /** Returns the projection matrix from the focal length.
     *
@@ -199,7 +194,7 @@ public:
     * @param far         distance in world units from the camera to the far plane. \p far > \p near.
     */
    static math::mat4 projection(double focalLengthInMillimeters,
-            double aspect, double near, double far = INFINITY);
+            double aspect, double near, double far = std::numeric_limits<double>::infinity());


    /** Sets the projection matrix from a frustum defined by six planes.
@@ -306,7 +301,7 @@ public:
     * @see setCustomProjection
     * @see Engine::Config::stereoscopicEyeCount
     */
-    void setCustomEyeProjection(math::mat4 const* UTILS_NONNULL projection, size_t count,
+    void setCustomEyeProjection(math::mat4 const* projection, size_t count,
            math::mat4 const& projectionForCulling, double near, double far);

    /** Sets an additional matrix that scales the projection matrix.
@@ -24,9 +24,6 @@
 #include <math/vec3.h>
 #include <math/vec4.h>

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

 //! RGB color in linear space
@@ -26,9 +26,6 @@

 #include <math/mathfwd.h>

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

 class Engine;
@@ -203,7 +200,7 @@ public:
         *
         * @return This Builder, for chaining calls
         */
-        Builder& toneMapper(ToneMapper const* UTILS_NULLABLE toneMapper) noexcept;
+        Builder& toneMapper(const ToneMapper* toneMapper) noexcept;

        /**
         * Selects the tone mapping operator to apply to the HDR color buffer as the last
@@ -473,9 +470,10 @@ public:
         *
         * @param engine Reference to the filament::Engine to associate this ColorGrading with.
         *
-         * @return pointer to the newly created object.
+         * @return pointer to the newly created object or nullptr if exceptions are disabled and
+         *         an error occurred.
         */
-        ColorGrading* UTILS_NONNULL build(Engine& engine);
+        ColorGrading* build(Engine& engine);

    private:
        friend class FColorGrading;
@@ -18,6 +18,7 @@
 #define TNT_FILAMENT_COLOR_SPACE_H

 #include <math/vec2.h>
+#include <math/vec3.h>

 namespace filament::color {

@@ -25,7 +25,7 @@

 #include <math/mathfwd.h>

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

 namespace filament {

@@ -39,12 +39,27 @@ namespace filament {
 class UTILS_PUBLIC DebugRegistry : public FilamentAPI {
 public:

+    /**
+     * Type of a property
+     */
+    enum Type {
+        BOOL, INT, FLOAT, FLOAT2, FLOAT3, FLOAT4
+    };
+
+    /**
+     * Information about a property
+     */
+    struct Property {
+        const char* name;   //!< property name
+        Type type;          //!< property type
+    };
+
    /**
     * Queries whether a property exists
     * @param name The name of the property to query
     * @return true if the property exists, false otherwise
     */
-    bool hasProperty(const char* UTILS_NONNULL name) const noexcept;
+    bool hasProperty(const char* name) const noexcept;

    /**
     * Queries the address of a property's data from its name
@@ -52,30 +67,28 @@ public:
     * @return Address of the data of the \p name property
     * @{
     */
-    void* UTILS_NULLABLE getPropertyAddress(const char* UTILS_NONNULL name);
+    void* getPropertyAddress(const char* name);

-    void const* UTILS_NULLABLE getPropertyAddress(const char* UTILS_NONNULL name) const noexcept;
+    void const* getPropertyAddress(const char* name) const noexcept;

    template<typename T>
-    inline T* UTILS_NULLABLE getPropertyAddress(const char* UTILS_NONNULL name) {
+    inline T* getPropertyAddress(const char* name) {
        return static_cast<T*>(getPropertyAddress(name));
    }

    template<typename T>
-    inline T const* UTILS_NULLABLE getPropertyAddress(const char* UTILS_NONNULL name) const noexcept {
+    inline T const* getPropertyAddress(const char* name) const noexcept {
        return static_cast<T*>(getPropertyAddress(name));
    }

    template<typename T>
-    inline bool getPropertyAddress(const char* UTILS_NONNULL name,
-            T* UTILS_NULLABLE* UTILS_NONNULL p) {
+    inline bool getPropertyAddress(const char* name, T** p) {
        *p = getPropertyAddress<T>(name);
        return *p != nullptr;
    }

    template<typename T>
-    inline bool getPropertyAddress(const char* UTILS_NONNULL name,
-            T* const UTILS_NULLABLE* UTILS_NONNULL p) const noexcept {
+    inline bool getPropertyAddress(const char* name, T* const* p) const noexcept {
        *p = getPropertyAddress<T>(name);
        return *p != nullptr;
    }
@@ -88,12 +101,12 @@ public:
     * @return true if the operation was successful, false otherwise.
     * @{
     */
-    bool setProperty(const char* UTILS_NONNULL name, bool v) noexcept;
-    bool setProperty(const char* UTILS_NONNULL name, int v) noexcept;
-    bool setProperty(const char* UTILS_NONNULL name, float v) noexcept;
-    bool setProperty(const char* UTILS_NONNULL name, math::float2 v) noexcept;
-    bool setProperty(const char* UTILS_NONNULL name, math::float3 v) noexcept;
-    bool setProperty(const char* UTILS_NONNULL name, math::float4 v) noexcept;
+    bool setProperty(const char* name, bool v) noexcept;
+    bool setProperty(const char* name, int v) noexcept;
+    bool setProperty(const char* name, float v) noexcept;
+    bool setProperty(const char* name, math::float2 v) noexcept;
+    bool setProperty(const char* name, math::float3 v) noexcept;
+    bool setProperty(const char* name, math::float4 v) noexcept;
    /** @}*/

    /**
@@ -103,20 +116,20 @@ public:
     * @return true if the call was successful and \p v was updated
     * @{
     */
-    bool getProperty(const char* UTILS_NONNULL name, bool* UTILS_NONNULL v) const noexcept;
-    bool getProperty(const char* UTILS_NONNULL name, int* UTILS_NONNULL v) const noexcept;
-    bool getProperty(const char* UTILS_NONNULL name, float* UTILS_NONNULL v) const noexcept;
-    bool getProperty(const char* UTILS_NONNULL name, math::float2* UTILS_NONNULL v) const noexcept;
-    bool getProperty(const char* UTILS_NONNULL name, math::float3* UTILS_NONNULL v) const noexcept;
-    bool getProperty(const char* UTILS_NONNULL name, math::float4* UTILS_NONNULL v) const noexcept;
+    bool getProperty(const char* name, bool* v) const noexcept;
+    bool getProperty(const char* name, int* v) const noexcept;
+    bool getProperty(const char* name, float* v) const noexcept;
+    bool getProperty(const char* name, math::float2* v) const noexcept;
+    bool getProperty(const char* name, math::float3* v) const noexcept;
+    bool getProperty(const char* name, math::float4* v) const noexcept;
    /** @}*/

    struct DataSource {
-        void const* UTILS_NULLABLE data;
+        void const* data;
        size_t count;
    };

-    DataSource getDataSource(const char* UTILS_NONNULL name) const noexcept;
+    DataSource getDataSource(const char* name) const noexcept;

    struct FrameHistory {
        using duration_ms = float;
@@ -19,14 +19,9 @@

 #include <filament/FilamentAPI.h>

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

 #include <utils/compiler.h>
-#include <utils/Invocable.h>
-
-#include <stdint.h>
-#include <stddef.h>

 namespace utils {
 class Entity;
@@ -305,23 +300,11 @@ public:
         * @see Engine::getMaxStereoscopicEyes
         */
        uint8_t 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).
-         */
-        uint32_t 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.
-         */
-        uint32_t resourceAllocatorCacheMaxAge = 30;
    };


 #if UTILS_HAS_THREADING
-    using CreateCallback = void(void* UTILS_NULLABLE user, void* UTILS_NONNULL token);
+    using CreateCallback = void(void* user, void* token);
 #endif

    /**
@@ -360,7 +343,7 @@ public:
         *
         * @return A reference to this Builder for chaining calls.
         */
-        Builder& platform(Platform* UTILS_NULLABLE platform) noexcept;
+        Builder& platform(Platform* platform) noexcept;

        /**
         * @param config    A pointer to optional parameters to specify memory size
@@ -368,7 +351,7 @@ public:
         *
         * @return A reference to this Builder for chaining calls.
         */
-        Builder& config(const Config* UTILS_NULLABLE config) noexcept;
+        Builder& config(const Config* config) noexcept;

        /**
         * @param sharedContext A platform-dependant context used as a shared context
@@ -376,7 +359,7 @@ public:
         *
         * @return A reference to this Builder for chaining calls.
         */
-        Builder& sharedContext(void* UTILS_NULLABLE sharedContext) noexcept;
+        Builder& sharedContext(void* sharedContext) noexcept;

        /**
         * @param featureLevel The feature level at which initialize Filament.
@@ -391,7 +374,7 @@ public:
         * @param callback  Callback called once the engine is initialized and it is safe to
         *                  call Engine::getEngine().
         */
-        void build(utils::Invocable<void(void* UTILS_NONNULL token)>&& callback) const;
+        void build(utils::Invocable<void(void* token)>&& callback) const;
 #endif

        /**
@@ -406,17 +389,16 @@ public:
         *              allocate the command buffer. If exceptions are disabled, this condition if
         *              fatal and this function will abort.
         */
-        Engine* UTILS_NULLABLE build() const;
+        Engine* build() const;
    };

    /**
     * Backward compatibility helper to create an Engine.
     * @see Builder
     */
-    static inline Engine* UTILS_NULLABLE create(Backend backend = Backend::DEFAULT,
-            Platform* UTILS_NULLABLE platform = nullptr,
-            void* UTILS_NULLABLE sharedContext = nullptr,
-            const Config* UTILS_NULLABLE config = nullptr) {
+    static inline Engine* create(Backend backend = Backend::DEFAULT,
+            Platform* platform = nullptr, void* sharedContext = nullptr,
+            const Config* config = nullptr) {
        return Engine::Builder()
                .backend(backend)
                .platform(platform)
@@ -431,18 +413,16 @@ public:
     * Backward compatibility helper to create an Engine asynchronously.
     * @see Builder
     */
-    static inline void createAsync(CreateCallback callback,
-            void* UTILS_NULLABLE user,
+    static inline void createAsync(CreateCallback callback, void* user,
            Backend backend = Backend::DEFAULT,
-            Platform* UTILS_NULLABLE platform = nullptr,
-            void* UTILS_NULLABLE sharedContext = nullptr,
-            const Config* UTILS_NULLABLE config = nullptr) {
+            Platform* platform = nullptr, void* sharedContext = nullptr,
+            const Config* config = nullptr) {
        Engine::Builder()
                .backend(backend)
                .platform(platform)
                .sharedContext(sharedContext)
                .config(config)
-                .build([callback, user](void* UTILS_NONNULL token) {
+                .build([callback, user](void* token) {
                    callback(user, token);
                });
    }
@@ -459,7 +439,7 @@ public:
     * allocate the command buffer. If exceptions are disabled, this condition if fatal and
     * this function will abort.
     */
-    static Engine* UTILS_NULLABLE getEngine(void* UTILS_NONNULL token);
+    static Engine* getEngine(void* token);
 #endif


@@ -489,7 +469,7 @@ public:
     * \remark
     * This method is thread-safe.
     */
-    static void destroy(Engine* UTILS_NULLABLE* UTILS_NULLABLE engine);
+    static void destroy(Engine** engine);

    /**
     * Destroy the Engine instance and all associated resources.
@@ -516,7 +496,7 @@ public:
     * \remark
     * This method is thread-safe.
     */
-    static void destroy(Engine* UTILS_NULLABLE engine);
+    static void destroy(Engine* engine);

    /**
     * Query the feature level supported by the selected backend.
@@ -664,11 +644,11 @@ public:
     *                     `ANativeWindow*`.
     * @param flags One or more configuration flags as defined in `SwapChain`.
     *
-     * @return A pointer to the newly created SwapChain.
+     * @return A pointer to the newly created SwapChain or nullptr if it couldn't be created.
     *
     * @see Renderer.beginFrame()
     */
-    SwapChain* UTILS_NONNULL createSwapChain(void* UTILS_NULLABLE nativeWindow, uint64_t flags = 0) noexcept;
+    SwapChain* createSwapChain(void* nativeWindow, uint64_t flags = 0) noexcept;


    /**
@@ -678,42 +658,42 @@ public:
      * @param height   Height of the drawing buffer in pixels.
     * @param flags     One or more configuration flags as defined in `SwapChain`.
     *
-     * @return A pointer to the newly created SwapChain.
+     * @return A pointer to the newly created SwapChain or nullptr if it couldn't be created.
     *
     * @see Renderer.beginFrame()
     */
-    SwapChain* UTILS_NONNULL createSwapChain(uint32_t width, uint32_t height, uint64_t flags = 0) noexcept;
+    SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags = 0) noexcept;

    /**
     * Creates a renderer associated to this engine.
     *
     * A Renderer is intended to map to a *window* on screen.
     *
-     * @return A pointer to the newly created Renderer.
+     * @return A pointer to the newly created Renderer or nullptr if it couldn't be created.
     */
-    Renderer* UTILS_NONNULL createRenderer() noexcept;
+    Renderer* createRenderer() noexcept;

    /**
     * Creates a View.
     *
-     * @return A pointer to the newly created View.
+     * @return A pointer to the newly created View or nullptr if it couldn't be created.
     */
-    View* UTILS_NONNULL createView() noexcept;
+    View* createView() noexcept;

    /**
     * Creates a Scene.
     *
-     * @return A pointer to the newly created Scene.
+     * @return A pointer to the newly created Scene or nullptr if it couldn't be created.
     */
-    Scene* UTILS_NONNULL createScene() noexcept;
+    Scene* createScene() noexcept;

    /**
     * Creates a Camera component.
     *
     * @param entity Entity to add the camera component to.
-     * @return A pointer to the newly created Camera.
+     * @return A pointer to the newly created Camera or nullptr if it couldn't be created.
     */
-    Camera* UTILS_NONNULL createCamera(utils::Entity entity) noexcept;
+    Camera* createCamera(utils::Entity entity) noexcept;

    /**
     * Returns the Camera component of the given entity.
@@ -723,7 +703,7 @@ public:
     *         have a Camera component. The pointer is valid until destroyCameraComponent()
     *         is called or the entity itself is destroyed.
     */
-    Camera* UTILS_NULLABLE getCameraComponent(utils::Entity entity) noexcept;
+    Camera* getCameraComponent(utils::Entity entity) noexcept;

    /**
     * Destroys the Camera component associated with the given entity.
@@ -735,17 +715,17 @@ public:
    /**
     * Creates a Fence.
     *
-     * @return A pointer to the newly created Fence.
+     * @return A pointer to the newly created Fence or nullptr if it couldn't be created.
     */
-    Fence* UTILS_NONNULL createFence() noexcept;
+    Fence* createFence() noexcept;

-    bool destroy(const BufferObject* UTILS_NULLABLE p);         //!< Destroys a BufferObject object.
-    bool destroy(const VertexBuffer* UTILS_NULLABLE p);         //!< Destroys an VertexBuffer object.
-    bool destroy(const Fence* UTILS_NULLABLE p);                //!< Destroys a Fence object.
-    bool destroy(const IndexBuffer* UTILS_NULLABLE p);          //!< Destroys an IndexBuffer object.
-    bool destroy(const SkinningBuffer* UTILS_NULLABLE p);       //!< Destroys a SkinningBuffer object.
-    bool destroy(const MorphTargetBuffer* UTILS_NULLABLE p);    //!< Destroys a MorphTargetBuffer object.
-    bool destroy(const IndirectLight* UTILS_NULLABLE p);        //!< Destroys an IndirectLight object.
+    bool destroy(const BufferObject* p);        //!< Destroys a BufferObject object.
+    bool destroy(const VertexBuffer* p);        //!< Destroys an VertexBuffer object.
+    bool destroy(const Fence* p);               //!< Destroys a Fence object.
+    bool destroy(const IndexBuffer* p);         //!< Destroys an IndexBuffer object.
+    bool destroy(const SkinningBuffer* p);      //!< Destroys a SkinningBuffer object.
+    bool destroy(const MorphTargetBuffer* p);   //!< Destroys a MorphTargetBuffer object.
+    bool destroy(const IndirectLight* p);       //!< Destroys an IndirectLight object.

    /**
     * Destroys a Material object
@@ -755,38 +735,38 @@ public:
     * @exception utils::PreConditionPanic is thrown if some MaterialInstances remain.
     * no-op if exceptions are disabled and some MaterialInstances remain.
     */
-    bool destroy(const Material* UTILS_NULLABLE p);
-    bool destroy(const MaterialInstance* UTILS_NULLABLE p); //!< Destroys a MaterialInstance object.
-    bool destroy(const Renderer* UTILS_NULLABLE p);         //!< Destroys a Renderer object.
-    bool destroy(const Scene* UTILS_NULLABLE p);            //!< Destroys a Scene object.
-    bool destroy(const Skybox* UTILS_NULLABLE p);           //!< Destroys a SkyBox object.
-    bool destroy(const ColorGrading* UTILS_NULLABLE p);     //!< Destroys a ColorGrading object.
-    bool destroy(const SwapChain* UTILS_NULLABLE p);        //!< Destroys a SwapChain object.
-    bool destroy(const Stream* UTILS_NULLABLE p);           //!< Destroys a Stream object.
-    bool destroy(const Texture* UTILS_NULLABLE p);          //!< Destroys a Texture object.
-    bool destroy(const RenderTarget* UTILS_NULLABLE p);     //!< Destroys a RenderTarget object.
-    bool destroy(const View* UTILS_NULLABLE p);             //!< Destroys a View object.
-    bool destroy(const InstanceBuffer* UTILS_NULLABLE p);   //!< Destroys an InstanceBuffer object.
-    void destroy(utils::Entity e);    //!< Destroys all filament-known components from this entity
+    bool destroy(const Material* p);
+    bool destroy(const MaterialInstance* p);    //!< Destroys a MaterialInstance object.
+    bool destroy(const Renderer* p);            //!< Destroys a Renderer object.
+    bool destroy(const Scene* p);               //!< Destroys a Scene object.
+    bool destroy(const Skybox* p);              //!< Destroys a SkyBox object.
+    bool destroy(const ColorGrading* p);        //!< Destroys a ColorGrading object.
+    bool destroy(const SwapChain* p);           //!< Destroys a SwapChain object.
+    bool destroy(const Stream* p);              //!< Destroys a Stream object.
+    bool destroy(const Texture* p);             //!< Destroys a Texture object.
+    bool destroy(const RenderTarget* p);        //!< Destroys a RenderTarget object.
+    bool destroy(const View* p);                //!< Destroys a View object.
+    bool destroy(const InstanceBuffer* p);      //!< Destroys an InstanceBuffer object.
+    void destroy(utils::Entity e);              //!< Destroys all filament-known components from this entity

-    bool isValid(const BufferObject* UTILS_NULLABLE p);        //!< Tells whether a BufferObject object is valid
-    bool isValid(const VertexBuffer* UTILS_NULLABLE p);        //!< Tells whether an VertexBuffer object is valid
-    bool isValid(const Fence* UTILS_NULLABLE p);               //!< Tells whether a Fence object is valid
-    bool isValid(const IndexBuffer* UTILS_NULLABLE p);         //!< Tells whether an IndexBuffer object is valid
-    bool isValid(const SkinningBuffer* UTILS_NULLABLE p);      //!< Tells whether a SkinningBuffer object is valid
-    bool isValid(const MorphTargetBuffer* UTILS_NULLABLE p);   //!< Tells whether a MorphTargetBuffer object is valid
-    bool isValid(const IndirectLight* UTILS_NULLABLE p);       //!< Tells whether an IndirectLight object is valid
-    bool isValid(const Material* UTILS_NULLABLE p);            //!< Tells whether an IndirectLight object is valid
-    bool isValid(const Renderer* UTILS_NULLABLE p);            //!< Tells whether a Renderer object is valid
-    bool isValid(const Scene* UTILS_NULLABLE p);               //!< Tells whether a Scene object is valid
-    bool isValid(const Skybox* UTILS_NULLABLE p);              //!< Tells whether a SkyBox object is valid
-    bool isValid(const ColorGrading* UTILS_NULLABLE p);        //!< Tells whether a ColorGrading object is valid
-    bool isValid(const SwapChain* UTILS_NULLABLE p);           //!< Tells whether a SwapChain object is valid
-    bool isValid(const Stream* UTILS_NULLABLE p);              //!< Tells whether a Stream object is valid
-    bool isValid(const Texture* UTILS_NULLABLE p);             //!< Tells whether a Texture object is valid
-    bool isValid(const RenderTarget* UTILS_NULLABLE p);        //!< Tells whether a RenderTarget object is valid
-    bool isValid(const View* UTILS_NULLABLE p);                //!< Tells whether a View object is valid
-    bool isValid(const InstanceBuffer* UTILS_NULLABLE p);      //!< Tells whether an InstanceBuffer object is valid
+    bool isValid(const BufferObject* p);        //!< Tells whether a BufferObject object is valid
+    bool isValid(const VertexBuffer* p);        //!< Tells whether an VertexBuffer object is valid
+    bool isValid(const Fence* p);               //!< Tells whether a Fence object is valid
+    bool isValid(const IndexBuffer* p);         //!< Tells whether an IndexBuffer object is valid
+    bool isValid(const SkinningBuffer* p);      //!< Tells whether a SkinningBuffer object is valid
+    bool isValid(const MorphTargetBuffer* p);   //!< Tells whether a MorphTargetBuffer object is valid
+    bool isValid(const IndirectLight* p);       //!< Tells whether an IndirectLight object is valid
+    bool isValid(const Material* p);            //!< Tells whether an IndirectLight object is valid
+    bool isValid(const Renderer* p);            //!< Tells whether a Renderer object is valid
+    bool isValid(const Scene* p);               //!< Tells whether a Scene object is valid
+    bool isValid(const Skybox* p);              //!< Tells whether a SkyBox object is valid
+    bool isValid(const ColorGrading* p);        //!< Tells whether a ColorGrading object is valid
+    bool isValid(const SwapChain* p);           //!< Tells whether a SwapChain object is valid
+    bool isValid(const Stream* p);              //!< Tells whether a Stream object is valid
+    bool isValid(const Texture* p);             //!< Tells whether a Texture object is valid
+    bool isValid(const RenderTarget* p);        //!< Tells whether a RenderTarget object is valid
+    bool isValid(const View* p);                //!< Tells whether a View object is valid
+    bool isValid(const InstanceBuffer* p);      //!< Tells whether an InstanceBuffer object is valid

    /**
     * Kicks the hardware thread (e.g. the OpenGL, Vulkan or Metal thread) and blocks until
@@ -826,7 +806,7 @@ public:
     *
     * @return A pointer to the default Material instance (a singleton).
     */
-    Material const* UTILS_NONNULL getDefaultMaterial() const noexcept;
+    const Material* getDefaultMaterial() const noexcept;

    /**
     * Returns the resolved backend.
@@ -857,7 +837,7 @@ public:
     * @return A pointer to the Platform object that was provided to Engine::create, or the
     * Filament-created one.
     */
-    Platform* UTILS_NULLABLE getPlatform() const noexcept;
+    Platform* getPlatform() const noexcept;

    /**
     * Allocate a small amount of memory directly in the command stream. The allocated memory is
@@ -870,7 +850,7 @@ public:
     * @note there is no need to destroy this buffer, it will be freed automatically when
     *       the current command buffer is executed.
     */
-    void* UTILS_NULLABLE streamAlloc(size_t size, size_t alignment = alignof(double)) noexcept;
+    void* streamAlloc(size_t size, size_t alignment = alignof(double)) noexcept;

    /**
      * Invokes one iteration of the render loop, used only on single-threaded platforms.
@@ -25,8 +25,6 @@

 #include <utils/compiler.h>

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

 /**
@@ -76,7 +74,7 @@ public:
     * @return  FenceStatus::CONDITION_SATISFIED on success,
     *          FenceStatus::ERROR otherwise.
     */
-    static FenceStatus waitAndDestroy(Fence* UTILS_NONNULL fence, Mode mode = Mode::FLUSH);
+    static FenceStatus waitAndDestroy(Fence* fence, Mode mode = Mode::FLUSH);

 protected:
    // prevent heap allocation
@@ -23,12 +23,9 @@

 #include <math/mat4.h>
 #include <math/vec3.h>
-#include <math/vec4.h>

 #include <utils/unwindows.h> // Because we define NEAR and FAR in the Plane enum.

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

 class Box;
@@ -79,14 +76,14 @@ public:
     * @param planes six plane equations encoded as in getNormalizedPlane() in
     *              left, right, bottom, top, far, near order
     */
-    void getNormalizedPlanes(math::float4 planes[UTILS_NONNULL 6]) const noexcept;
+    void getNormalizedPlanes(math::float4 planes[6]) const noexcept;

    /**
     * Returns all six frustum planes in left, right, bottom, top, far, near order
     * @return six plane equations encoded as in getNormalizedPlane() in
     *              left, right, bottom, top, far, near order
     */
-    math::float4 const* UTILS_NONNULL getNormalizedPlanes() const noexcept { return mPlanes; }
+    math::float4 const* getNormalizedPlanes() const noexcept { return mPlanes; }

    /**
     * Returns whether a box intersects the frustum (i.e. is visible)
@@ -27,7 +27,6 @@

 #include <utils/compiler.h>

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

 namespace filament {
@@ -89,7 +88,8 @@ public:
         *
         * @param engine Reference to the filament::Engine to associate this IndexBuffer with.
         *
-         * @return pointer to the newly created object.
+         * @return pointer to the newly created object or nullptr if exceptions are disabled and
+         *         an error occurred.
         *
         * @exception utils::PostConditionPanic if a runtime error occurred, such as running out of
         *            memory or other resources.
@@ -97,7 +97,7 @@ public:
         *
         * @see IndexBuffer::setBuffer
         */
-        IndexBuffer* UTILS_NONNULL build(Engine& engine);
+        IndexBuffer* build(Engine& engine);
    private:
        friend class FIndexBuffer;
    };
@@ -25,8 +25,6 @@

 #include <math/mathfwd.h>

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

 class Engine;
@@ -116,7 +114,7 @@ public:
         * @return This Builder, for chaining calls.
         *
         */
-        Builder& reflections(Texture const* UTILS_NULLABLE cubemap) noexcept;
+        Builder& reflections(Texture const* cubemap) noexcept;

        /**
         * Sets the irradiance as Spherical Harmonics.
@@ -162,7 +160,7 @@ public:
         * Because the coefficients are pre-scaled, `sh[0]` is the environment's
         * average irradiance.
         */
-        Builder& irradiance(uint8_t bands, math::float3 const* UTILS_NONNULL sh) noexcept;
+        Builder& irradiance(uint8_t bands, math::float3 const* sh) noexcept;

        /**
         * Sets the irradiance from the radiance expressed as Spherical Harmonics.
@@ -194,7 +192,7 @@ public:
         *
         * @return This Builder, for chaining calls.
         */
-        Builder& radiance(uint8_t bands, math::float3 const* UTILS_NONNULL sh) noexcept;
+        Builder& radiance(uint8_t bands, math::float3 const* sh) noexcept;

        /**
         * Sets the irradiance as a cubemap.
@@ -213,7 +211,7 @@ public:
         *
         * @see irradiance(uint8_t bands, math::float3 const* sh)
         */
-        Builder& irradiance(Texture const* UTILS_NULLABLE cubemap) noexcept;
+        Builder& irradiance(Texture const* cubemap) noexcept;

        /**
         * (optional) Environment intensity.
@@ -249,7 +247,7 @@ public:
         *            memory or other resources.
         * @exception utils::PreConditionPanic if a parameter to a builder function was invalid.
         */
-        IndirectLight* UTILS_NONNULL build(Engine& engine);
+        IndirectLight* build(Engine& engine);

    private:
        friend class FIndirectLight;
@@ -286,12 +284,12 @@ public:
    /**
     * Returns the associated reflection map, or null if it does not exist.
     */
-    Texture const* UTILS_NULLABLE getReflectionsTexture() const noexcept;
+    Texture const* getReflectionsTexture() const noexcept;

    /**
     * Returns the associated irradiance map, or null if it does not exist.
     */
-    Texture const* UTILS_NULLABLE getIrradianceTexture() const noexcept;
+    Texture const* getIrradianceTexture() const noexcept;

    /**
     * Helper to estimate the direction of the dominant light in the environment represented by
@@ -314,7 +312,7 @@ public:
     * @see LightManager::Builder::direction()
     * @see getColorEstimate()
     */
-    static math::float3 getDirectionEstimate(const math::float3 sh[UTILS_NONNULL 9]) noexcept;
+    static math::float3 getDirectionEstimate(const math::float3 sh[9]) noexcept;

    /**
     * Helper to estimate the color and relative intensity of the environment represented by
@@ -334,8 +332,7 @@ public:
     * @see LightManager::Builder::intensity()
     * @see getDirectionEstimate, getIntensity, setIntensity
     */
-    static math::float4 getColorEstimate(const math::float3 sh[UTILS_NONNULL 9],
-            math::float3 direction) noexcept;
+    static math::float4 getColorEstimate(const math::float3 sh[9], math::float3 direction) noexcept;


    /** @deprecated use static versions instead */
@@ -18,14 +18,11 @@
 #define TNT_FILAMENT_INSTANCEBUFFER_H

 #include <filament/FilamentAPI.h>
+
 #include <filament/Engine.h>

-#include <utils/compiler.h>
-
 #include <math/mathfwd.h>

-#include <stddef.h>
-
 namespace filament {

 /**
@@ -48,7 +45,7 @@ public:
         *                      >= 1 and <= \c Engine::getMaxAutomaticInstances()
         * @see Engine::getMaxAutomaticInstances
         */
-        explicit Builder(size_t instanceCount) noexcept;
+        Builder(size_t instanceCount) noexcept;

        Builder(Builder const& rhs) noexcept;
        Builder(Builder&& rhs) noexcept;
@@ -68,12 +65,12 @@ public:
         * @param localTransforms an array of math::mat4f with length instanceCount, must remain
         *                        valid until after build() is called
         */
-        Builder& localTransforms(math::mat4f const* UTILS_NULLABLE localTransforms) noexcept;
+        Builder& localTransforms(math::mat4f const* localTransforms) noexcept;

        /**
         * Creates the InstanceBuffer object and returns a pointer to it.
         */
-        InstanceBuffer* UTILS_NONNULL build(Engine& engine);
+        InstanceBuffer* build(Engine& engine);

    private:
        friend class FInstanceBuffer;
@@ -93,8 +90,7 @@ public:
     * @param count the number of local transforms
     * @param offset index of the first instance to set local transforms
     */
-    void setLocalTransforms(math::mat4f const* UTILS_NONNULL localTransforms,
-            size_t count, size_t offset = 0);
+    void setLocalTransforms(math::mat4f const* localTransforms, size_t count, size_t offset = 0);

 protected:
    // prevent heap allocation
@@ -27,9 +27,6 @@
 #include <math/mathfwd.h>
 #include <math/quat.h>

-#include <stdint.h>
-#include <stddef.h>
-
 namespace utils {
    class Entity;
 } // namespace utils
@@ -176,7 +173,7 @@ public:
     * Retrieve the Entities of all the components of this manager.
     * @return A list, in no particular order, of all the entities managed by this manager.
     */
-    utils::Entity const* UTILS_NONNULL getEntities() const noexcept;
+    utils::Entity const* getEntities() const noexcept;

    /**
     * Gets an Instance representing the Light component associated with the given Entity.
@@ -394,7 +391,7 @@ public:
         *                          positions into
         * @param cascades          the number of shadow cascades, at most 4
         */
-        static void computeUniformSplits(float* UTILS_NONNULL splitPositions, uint8_t cascades);
+        static void computeUniformSplits(float* splitPositions, uint8_t cascades);

        /**
         * Utility method to compute ShadowOptions::cascadeSplitPositions according to a logarithmic
@@ -406,7 +403,7 @@ public:
         * @param near              the camera near plane
         * @param far               the camera far plane
         */
-        static void computeLogSplits(float* UTILS_NONNULL splitPositions, uint8_t cascades,
+        static void computeLogSplits(float* splitPositions, uint8_t cascades,
                float near, float far);

        /**
@@ -426,7 +423,7 @@ public:
         * @param lambda            a float in the range [0, 1] that interpolates between log and
         *                          uniform split schemes
         */
-        static void computePracticalSplits(float* UTILS_NONNULL splitPositions, uint8_t cascades,
+        static void computePracticalSplits(float* splitPositions, uint8_t cascades,
                float near, float far, float lambda);
    };

@@ -30,12 +30,7 @@

 #include <math/mathfwd.h>

-#include <type_traits>
-#include <utility>
-
-#include <stddef.h>
 #include <stdint.h>
-#include <string.h>

 namespace utils {
    class CString;
@@ -74,7 +69,7 @@ public:
     */
    struct ParameterInfo {
        //! Name of the parameter.
-        const char* UTILS_NONNULL name;
+        const char* name;
        //! Whether the parameter is a sampler (texture).
        bool isSampler;
        //! Whether the parameter is a subpass type.
@@ -110,7 +105,7 @@ public:
         * @param payload Pointer to the material data, must stay valid until build() is called.
         * @param size Size of the material data pointed to by "payload" in bytes.
         */
-        Builder& package(const void* UTILS_NONNULL payload, size_t size);
+        Builder& package(const void* payload, size_t size);

        template<typename T>
        using is_supported_constant_parameter_t = typename std::enable_if<
@@ -132,11 +127,11 @@ public:
         *              in the material definition.
         */
        template<typename T, typename = is_supported_constant_parameter_t<T>>
-        Builder& constant(const char* UTILS_NONNULL name, size_t nameLength, T value);
+        Builder& constant(const char* name, size_t nameLength, T value);

        /** inline helper to provide the constant name as a null-terminated C string */
        template<typename T, typename = is_supported_constant_parameter_t<T>>
-        inline Builder& constant(const char* UTILS_NONNULL name, T value) {
+        inline Builder& constant(const char* name, T value) {
            return constant(name, strlen(name), value);
        }

@@ -152,7 +147,7 @@ public:
         *            memory or other resources.
         * @exception utils::PreConditionPanic if a parameter to a builder function was invalid.
         */
-        Material* UTILS_NULLABLE build(Engine& engine);
+        Material* build(Engine& engine);
    private:
        friend class FMaterial;
    };
@@ -198,22 +193,22 @@ public:
     */
    void compile(CompilerPriorityQueue priority,
            UserVariantFilterMask variants,
-            backend::CallbackHandler* UTILS_NULLABLE handler = nullptr,
-            utils::Invocable<void(Material* UTILS_NONNULL)>&& callback = {}) noexcept;
+            backend::CallbackHandler* handler = nullptr,
+            utils::Invocable<void(Material*)>&& callback = {}) noexcept;

    inline void compile(CompilerPriorityQueue priority,
            UserVariantFilterBit variants,
-            backend::CallbackHandler* UTILS_NULLABLE handler = nullptr,
-            utils::Invocable<void(Material* UTILS_NONNULL)>&& callback = {}) noexcept {
+            backend::CallbackHandler* handler = nullptr,
+            utils::Invocable<void(Material*)>&& callback = {}) noexcept {
        compile(priority, UserVariantFilterMask(variants), handler,
-                std::forward<utils::Invocable<void(Material* UTILS_NONNULL)>>(callback));
+                std::forward<utils::Invocable<void(Material*)>>(callback));
    }

    inline void compile(CompilerPriorityQueue priority,
-            backend::CallbackHandler* UTILS_NULLABLE handler = nullptr,
-            utils::Invocable<void(Material* UTILS_NONNULL)>&& callback = {}) noexcept {
+            backend::CallbackHandler* handler = nullptr,
+            utils::Invocable<void(Material*)>&& callback = {}) noexcept {
        compile(priority, UserVariantFilterBit::ALL, handler,
-                std::forward<utils::Invocable<void(Material* UTILS_NONNULL)>>(callback));
+                std::forward<utils::Invocable<void(Material*)>>(callback));
    }

    /**
@@ -225,13 +220,13 @@ public:
     *
     * @return A pointer to the new instance.
     */
-    MaterialInstance* UTILS_NONNULL createInstance(const char* UTILS_NULLABLE name = nullptr) const noexcept;
+    MaterialInstance* createInstance(const char* name = nullptr) const noexcept;

    //! Returns the name of this material as a null-terminated string.
-    const char* UTILS_NONNULL getName() const noexcept;
+    const char* getName() const noexcept;

    //! Returns the shading model of this material.
-    Shading getShading() const noexcept;
+    Shading getShading()  const noexcept;

    //! Returns the interpolation mode of this material. This affects how variables are interpolated.
    Interpolation getInterpolation() const noexcept;
@@ -317,13 +312,13 @@ public:
     *
     * @return The number of parameters written to the parameters pointer.
     */
-    size_t getParameters(ParameterInfo* UTILS_NONNULL parameters, size_t count) const noexcept;
+    size_t getParameters(ParameterInfo* parameters, size_t count) const noexcept;

    //! Indicates whether a parameter of the given name exists on this material.
-    bool hasParameter(const char* UTILS_NONNULL name) const noexcept;
+    bool hasParameter(const char* name) const noexcept;

    //! Indicates whether an existing parameter is a sampler or not.
-    bool isSampler(const char* UTILS_NONNULL name) const noexcept;
+    bool isSampler(const char* name) const noexcept;

    /**
     * Sets the value of the given parameter on this material's default instance.
@@ -334,7 +329,7 @@ public:
     * @see getDefaultInstance()
     */
    template <typename T>
-    void setDefaultParameter(const char* UTILS_NONNULL name, T value) noexcept {
+    void setDefaultParameter(const char* name, T value) noexcept {
        getDefaultInstance()->setParameter(name, value);
    }

@@ -347,8 +342,8 @@ public:
     *
     * @see getDefaultInstance()
     */
-    void setDefaultParameter(const char* UTILS_NONNULL name,
-            Texture const* UTILS_NULLABLE texture, TextureSampler const& sampler) noexcept {
+    void setDefaultParameter(const char* name, Texture const* texture,
+            TextureSampler const& sampler) noexcept {
        getDefaultInstance()->setParameter(name, texture, sampler);
    }

@@ -361,7 +356,7 @@ public:
     *
     * @see getDefaultInstance()
     */
-    void setDefaultParameter(const char* UTILS_NONNULL name, RgbType type, math::float3 color) noexcept {
+    void setDefaultParameter(const char* name, RgbType type, math::float3 color) noexcept {
        getDefaultInstance()->setParameter(name, type, color);
    }

@@ -374,15 +369,15 @@ public:
     *
     * @see getDefaultInstance()
     */
-    void setDefaultParameter(const char* UTILS_NONNULL name, RgbaType type, math::float4 color) noexcept {
+    void setDefaultParameter(const char* name, RgbaType type, math::float4 color) noexcept {
        getDefaultInstance()->setParameter(name, type, color);
    }

    //! Returns this material's default instance.
-    MaterialInstance* UTILS_NONNULL getDefaultInstance() noexcept;
+    MaterialInstance* getDefaultInstance() noexcept;

    //! Returns this material's default instance.
-    MaterialInstance const* UTILS_NONNULL getDefaultInstance() const noexcept;
+    MaterialInstance const* getDefaultInstance() const noexcept;

 protected:
    // prevent heap allocation
@@ -28,12 +28,6 @@

 #include <math/mathfwd.h>

-#include <type_traits>
-
-#include <stddef.h>
-#include <stdint.h>
-#include <string.h>
-
 namespace filament {

 class Material;
@@ -47,7 +41,7 @@ class UTILS_PUBLIC MaterialInstance : public FilamentAPI {
    using StringLiteralHelper = const char[N];

    struct StringLiteral {
-        const char* UTILS_NONNULL data;
+        const char* data;
        size_t size;
        template<size_t N>
        StringLiteral(StringLiteralHelper<N> const& s) noexcept // NOLINT(google-explicit-constructor)
@@ -95,18 +89,17 @@ public:
     * @param name  A name for the new MaterialInstance or nullptr to use the template's name
     * @return      A new MaterialInstance
     */
-    static MaterialInstance* UTILS_NONNULL duplicate(MaterialInstance const* UTILS_NONNULL other,
-            const char* UTILS_NULLABLE name = nullptr) noexcept;
+    static MaterialInstance* duplicate(MaterialInstance const* other, const char* name = nullptr) noexcept;

    /**
     * @return the Material associated with this instance
     */
-    Material const* UTILS_NONNULL getMaterial() const noexcept;
+    Material const* getMaterial() const noexcept;

    /**
     * @return the name associated with this instance
     */
-    const char* UTILS_NONNULL getName() const noexcept;
+    const char* getName() const noexcept;

    /**
     * Set a uniform by name
@@ -117,7 +110,7 @@ public:
     * @throws utils::PreConditionPanic if name doesn't exist or no-op if exceptions are disabled.
     */
    template<typename T, typename = is_supported_parameter_t<T>>
-    void setParameter(const char* UTILS_NONNULL name, size_t nameLength, T const& value);
+    void setParameter(const char* name, size_t nameLength, T const& value);

    /** inline helper to provide the name as a null-terminated string literal */
    template<typename T, typename = is_supported_parameter_t<T>>
@@ -127,7 +120,7 @@ public:

    /** inline helper to provide the name as a null-terminated C string */
    template<typename T, typename = is_supported_parameter_t<T>>
-    inline void setParameter(const char* UTILS_NONNULL name, T const& value) {
+    inline void setParameter(const char* name, T const& value) {
        setParameter<T>(name, strlen(name), value);
    }

@@ -142,19 +135,17 @@ public:
     * @throws utils::PreConditionPanic if name doesn't exist or no-op if exceptions are disabled.
     */
    template<typename T, typename = is_supported_parameter_t<T>>
-    void setParameter(const char* UTILS_NONNULL name, size_t nameLength,
-            const T* UTILS_NONNULL values, size_t count);
+    void setParameter(const char* name, size_t nameLength, const T* values, size_t count);

    /** inline helper to provide the name as a null-terminated string literal */
    template<typename T, typename = is_supported_parameter_t<T>>
-    inline void setParameter(StringLiteral name, const T* UTILS_NONNULL values, size_t count) {
+    inline void setParameter(StringLiteral name, const T* values, size_t count) {
        setParameter<T>(name.data, name.size, values, count);
    }

    /** inline helper to provide the name as a null-terminated C string */
    template<typename T, typename = is_supported_parameter_t<T>>
-    inline void setParameter(const char* UTILS_NONNULL name,
-            const T* UTILS_NONNULL values, size_t count) {
+    inline void setParameter(const char* name, const T* values, size_t count) {
        setParameter<T>(name, strlen(name), values, count);
    }

@@ -171,18 +162,18 @@ public:
     * @param sampler       Sampler parameters.
     * @throws utils::PreConditionPanic if name doesn't exist or no-op if exceptions are disabled.
     */
-    void setParameter(const char* UTILS_NONNULL name, size_t nameLength,
-            Texture const* UTILS_NULLABLE texture, TextureSampler const& sampler);
+    void setParameter(const char* name, size_t nameLength,
+            Texture const* texture, TextureSampler const& sampler);

    /** inline helper to provide the name as a null-terminated string literal */
    inline void setParameter(StringLiteral name,
-            Texture const* UTILS_NULLABLE texture, TextureSampler const& sampler) {
+            Texture const* texture, TextureSampler const& sampler) {
        setParameter(name.data, name.size, texture, sampler);
    }

    /** inline helper to provide the name as a null-terminated C string */
-    inline void setParameter(const char* UTILS_NONNULL name,
-            Texture const* UTILS_NULLABLE texture, TextureSampler const& sampler) {
+    inline void setParameter(const char* name,
+            Texture const* texture, TextureSampler const& sampler) {
        setParameter(name, strlen(name), texture, sampler);
    }

@@ -197,8 +188,7 @@ public:
     * @param color         Array of read, green, blue channels values.
     * @throws utils::PreConditionPanic if name doesn't exist or no-op if exceptions are disabled.
     */
-    void setParameter(const char* UTILS_NONNULL name, size_t nameLength,
-            RgbType type, math::float3 color);
+    void setParameter(const char* name, size_t nameLength, RgbType type, math::float3 color);

    /** inline helper to provide the name as a null-terminated string literal */
    inline void setParameter(StringLiteral name, RgbType type, math::float3 color) {
@@ -206,7 +196,7 @@ public:
    }

    /** inline helper to provide the name as a null-terminated C string */
-    inline void setParameter(const char* UTILS_NONNULL name, RgbType type, math::float3 color) {
+    inline void setParameter(const char* name, RgbType type, math::float3 color) {
        setParameter(name, strlen(name), type, color);
    }

@@ -221,8 +211,7 @@ public:
     * @param color         Array of read, green, blue and alpha channels values.
     * @throws utils::PreConditionPanic if name doesn't exist or no-op if exceptions are disabled.
     */
-    void setParameter(const char* UTILS_NONNULL name, size_t nameLength,
-            RgbaType type, math::float4 color);
+    void setParameter(const char* name, size_t nameLength, RgbaType type, math::float4 color);

    /** inline helper to provide the name as a null-terminated string literal */
    inline void setParameter(StringLiteral name, RgbaType type, math::float4 color) {
@@ -230,7 +219,7 @@ public:
    }

    /** inline helper to provide the name as a null-terminated C string */
-    inline void setParameter(const char* UTILS_NONNULL name, RgbaType type, math::float4 color) {
+    inline void setParameter(const char* name, RgbaType type, math::float4 color) {
        setParameter(name, strlen(name), type, color);
    }

@@ -21,11 +21,8 @@

 #include <filament/Engine.h>

-#include <utils/compiler.h>
-
 #include <math/mathfwd.h>

-#include <stddef.h>

 namespace filament {

@@ -68,13 +65,14 @@ public:
         *
         * @param engine Reference to the filament::Engine to associate this MorphTargetBuffer with.
         *
-         * @return pointer to the newly created object.
+         * @return pointer to the newly created object or nullptr if exceptions are disabled and
+         *         an error occurred.
         *
         * @exception utils::PostConditionPanic if a runtime error occurred, such as running out of
         *            memory or other resources.
         * @exception utils::PreConditionPanic if a parameter to a builder function was invalid.
         */
-        MorphTargetBuffer* UTILS_NONNULL build(Engine& engine);
+        MorphTargetBuffer* build(Engine& engine);
    private:
        friend class FMorphTargetBuffer;
    };
@@ -94,7 +92,7 @@ public:
     * @see setTangentsAt
     */
    void setPositionsAt(Engine& engine, size_t targetIndex,
-            math::float3 const* UTILS_NONNULL positions, size_t count, size_t offset = 0);
+            math::float3 const* positions, size_t count, size_t offset = 0);

    /**
     * Updates positions for the given morph target.
@@ -109,7 +107,7 @@ public:
     * @see setTangentsAt
     */
    void setPositionsAt(Engine& engine, size_t targetIndex,
-            math::float4 const* UTILS_NONNULL positions, size_t count, size_t offset = 0);
+            math::float4 const* positions, size_t count, size_t offset = 0);

    /**
     * Updates tangents for the given morph target.
@@ -125,7 +123,7 @@ public:
     * @see setPositionsAt
     */
    void setTangentsAt(Engine& engine, size_t targetIndex,
-            math::short4 const* UTILS_NONNULL tangents, size_t count, size_t offset = 0);
+            math::short4 const* tangents, size_t count, size_t offset = 0);

    /**
     * Returns the vertex count of this MorphTargetBuffer.
@@ -19,13 +19,10 @@

 #include <filament/Color.h>

-#include <math/vec2.h>
-#include <math/vec3.h>
+#include <stdint.h>

 #include <math.h>

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

 class Texture;
@@ -441,7 +438,6 @@ struct TemporalAntiAliasingOptions {
    float filterWidth = 1.0f;   //!< reconstruction filter width typically between 0.2 (sharper, aliased) and 1.5 (smoother)
    float feedback = 0.12f;     //!< history feedback, between 0 (maximum temporal AA) and 1 (no temporal AA).
    float lodBias = -1.0f;      //!< texturing lod bias (typically -1 or -2)
-    float sharpness = 0.0f;     //!< post-TAA sharpen, especially useful when upscaling is true.
    bool enabled = false;       //!< enables or disables temporal anti-aliasing
    bool upscaling = false;     //!< 4x TAA upscaling. Disables Dynamic Resolution. [BETA]

@@ -461,8 +457,7 @@ struct TemporalAntiAliasingOptions {
        RGSS_X4,             //!  4-samples, rotated grid sampling
        UNIFORM_HELIX_X4,    //!  4-samples, uniform grid in helix sequence
        HALTON_23_X8,        //!  8-samples of halton 2,3
-        HALTON_23_X16,       //! 16-samples of halton 2,3
-        HALTON_23_X32        //! 32-samples of halton 2,3
+        HALTON_23_X16        //! 16-samples of halton 2,3
    };

    bool filterHistory = true;      //!< whether to filter the history buffer
@@ -24,10 +24,7 @@
 #include <backend/DriverEnums.h>
 #include <backend/TargetBufferInfo.h>

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

 namespace filament {

@@ -106,7 +103,7 @@ public:
         * @param texture The associated texture object.
         * @return A reference to this Builder for chaining calls.
         */
-        Builder& texture(AttachmentPoint attachment, Texture* UTILS_NULLABLE texture) noexcept;
+        Builder& texture(AttachmentPoint attachment, Texture* texture) noexcept;

        /**
         * Sets the mipmap level for a given attachment point.
@@ -138,9 +135,10 @@ public:
        /**
         * Creates the RenderTarget object and returns a pointer to it.
         *
-         * @return pointer to the newly created object.
+         * @return pointer to the newly created object or nullptr if exceptions are disabled and
+         *         an error occurred.
         */
-        RenderTarget* UTILS_NONNULL build(Engine& engine);
+        RenderTarget* build(Engine& engine);

    private:
        friend class FRenderTarget;
@@ -151,7 +149,7 @@ public:
     * @param attachment Attachment point
     * @return A Texture object or nullptr if no texture is set for this attachment point
     */
-    Texture* UTILS_NULLABLE getTexture(AttachmentPoint attachment) const noexcept;
+    Texture* getTexture(AttachmentPoint attachment) const noexcept;

    /**
     * Returns the mipmap level set on the given attachment point
@@ -29,15 +29,9 @@
 #include <utils/FixedCapacityVector.h>

 #include <math/mathfwd.h>
-#include <math/vec3.h>
-#include <math/vec4.h>

 #include <type_traits>

-#include <float.h>
-#include <stddef.h>
-#include <stdint.h>
-
 namespace utils {
    class Entity;
 } // namespace utils
@@ -129,7 +123,7 @@ public:
     * Retrieve the Entities of all the components of this manager.
     * @return A list, in no particular order, of all the entities managed by this manager.
     */
-    utils::Entity const* UTILS_NONNULL getEntities() const noexcept;
+    utils::Entity const* getEntities() const noexcept;

    /**
     * The transformation associated with a skinning joint.
@@ -190,19 +184,9 @@ public:
         * @param maxIndex specifies the maximum index contained in the index buffer
         * @param count number of indices to read (for triangles, this should be a multiple of 3)
         */
-        Builder& geometry(size_t index, PrimitiveType type,
-                VertexBuffer* UTILS_NONNULL vertices,
-                IndexBuffer* UTILS_NONNULL indices,
-                size_t offset, size_t minIndex, size_t maxIndex, size_t count) noexcept;
-
-        Builder& geometry(size_t index, PrimitiveType type,
-                VertexBuffer* UTILS_NONNULL vertices,
-                IndexBuffer* UTILS_NONNULL indices,
-                size_t offset, size_t count) noexcept; //!< \overload
-
-        Builder& geometry(size_t index, PrimitiveType type,
-                VertexBuffer* UTILS_NONNULL vertices,
-                IndexBuffer* UTILS_NONNULL indices) noexcept; //!< \overload
+        Builder& geometry(size_t index, PrimitiveType type, VertexBuffer* vertices, IndexBuffer* indices, size_t offset, size_t minIndex, size_t maxIndex, size_t count) noexcept;
+        Builder& geometry(size_t index, PrimitiveType type, VertexBuffer* vertices, IndexBuffer* indices, size_t offset, size_t count) noexcept; //!< \overload
+        Builder& geometry(size_t index, PrimitiveType type, VertexBuffer* vertices, IndexBuffer* indices) noexcept; //!< \overload

        /**
         * Binds a material instance to the specified primitive.
@@ -219,8 +203,7 @@ public:
         *
         * @see Engine::setActiveFeatureLevel
         */
-        Builder& material(size_t index,
-                MaterialInstance const* UTILS_NONNULL materialInstance) noexcept;
+        Builder& material(size_t index, MaterialInstance const* materialInstance) noexcept;

        /**
         * The axis-aligned bounding box of the renderable.
@@ -368,8 +351,7 @@ public:
         * @param count 0 to disable, otherwise the number of bone transforms (up to 255)
         * @param offset offset in the SkinningBuffer
         */
-        Builder& skinning(SkinningBuffer* UTILS_NONNULL skinningBuffer,
-                size_t count, size_t offset) noexcept;
+        Builder& skinning(SkinningBuffer* skinningBuffer, size_t count, size_t offset) noexcept;


        /**
@@ -387,8 +369,8 @@ public:
         * @param boneCount 0 to disable, otherwise the number of bone transforms (up to 255)
         * @param transforms the initial set of transforms (one for each bone)
         */
-        Builder& skinning(size_t boneCount, math::mat4f const* UTILS_NONNULL transforms) noexcept;
-        Builder& skinning(size_t boneCount, Bone const* UTILS_NONNULL bones) noexcept; //!< \overload
+        Builder& skinning(size_t boneCount, math::mat4f const* transforms) noexcept;
+        Builder& skinning(size_t boneCount, Bone const* bones) noexcept; //!< \overload
        Builder& skinning(size_t boneCount) noexcept; //!< \overload

        /**
@@ -417,8 +399,7 @@ public:
         * @see VertexBuffer:Builder:advancedSkinning
         */
        Builder& boneIndicesAndWeights(size_t primitiveIndex,
-                math::float2 const* UTILS_NONNULL indicesAndWeights,
-                size_t count, size_t bonesPerVertex) noexcept;
+                math::float2 const* indicesAndWeights, size_t count, size_t bonesPerVertex) noexcept;

        /**
         * Define bone indices and weights "pairs" for vertex skinning as a float2.
@@ -481,11 +462,10 @@ public:
         * @param count number of vertices in the morph target buffer to read, must equal the geometry's count (for triangles, this should be a multiple of 3)
         */
        Builder& morphing(uint8_t level, size_t primitiveIndex,
-                MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer,
-                size_t offset, size_t count) noexcept;
+                MorphTargetBuffer* morphTargetBuffer, size_t offset, size_t count) noexcept;

        inline Builder& morphing(uint8_t level, size_t primitiveIndex,
-                MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer) noexcept;
+                MorphTargetBuffer* morphTargetBuffer) noexcept;

        /**
         * Sets the drawing order for blended primitives. The drawing order is either global or
@@ -552,8 +532,7 @@ public:
         *                      the result of Engine::getMaxAutomaticInstances().
         * @param instanceBuffer an InstanceBuffer containing at least instanceCount transforms
         */
-        Builder& instances(size_t instanceCount,
-                InstanceBuffer* UTILS_NONNULL instanceBuffer) noexcept;
+        Builder& instances(size_t instanceCount, InstanceBuffer* instanceBuffer) noexcept;

        /**
         * Adds the Renderable component to an entity.
@@ -580,18 +559,18 @@ public:
        friend class FRenderPrimitive;
        friend class FRenderableManager;
        struct Entry {
-            VertexBuffer* UTILS_NULLABLE vertices = nullptr;
-            IndexBuffer* UTILS_NULLABLE indices = nullptr;
+            VertexBuffer* vertices = nullptr;
+            IndexBuffer* indices = nullptr;
            size_t offset = 0;
            size_t minIndex = 0;
            size_t maxIndex = 0;
            size_t count = 0;
-            MaterialInstance const* UTILS_NULLABLE materialInstance = nullptr;
+            MaterialInstance const* materialInstance = nullptr;
            PrimitiveType type = PrimitiveType::TRIANGLES;
            uint16_t blendOrder = 0;
            bool globalBlendOrderEnabled = false;
            struct {
-                MorphTargetBuffer* UTILS_NULLABLE buffer = nullptr;
+                MorphTargetBuffer* buffer = nullptr;
                size_t offset = 0;
                size_t count = 0;
            } morphing;
@@ -709,11 +688,8 @@ public:
     * Updates the bone transforms in the range [offset, offset + boneCount).
     * The bones must be pre-allocated using Builder::skinning().
     */
-    void setBones(Instance instance, Bone const* UTILS_NONNULL transforms,
-            size_t boneCount = 1, size_t offset = 0);
-
-    void setBones(Instance instance, math::mat4f const* UTILS_NONNULL transforms,
-            size_t boneCount = 1, size_t offset = 0); //!< \overload
+    void setBones(Instance instance, Bone const* transforms, size_t boneCount = 1, size_t offset = 0);
+    void setBones(Instance instance, math::mat4f const* transforms, size_t boneCount = 1, size_t offset = 0); //!< \overload

    /**
     * Associates a region of a SkinningBuffer to a renderable instance
@@ -726,7 +702,7 @@ public:
     * @param count             Size of the region in bones, must be smaller or equal to 256.
     * @param offset            Start offset of the region in bones
     */
-    void setSkinningBuffer(Instance instance, SkinningBuffer* UTILS_NONNULL skinningBuffer,
+    void setSkinningBuffer(Instance instance, SkinningBuffer* skinningBuffer,
            size_t count, size_t offset);

    /**
@@ -741,25 +717,24 @@ public:
     * @param offset Index of the first morph target weight to set at instance.
     */
    void setMorphWeights(Instance instance,
-            float const* UTILS_NONNULL weights, size_t count, size_t offset = 0);
+            float const* weights, size_t count, size_t offset = 0);

    /**
     * Associates a MorphTargetBuffer to the given primitive.
     */
    void setMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex,
-            MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer, size_t offset, size_t count);
+            MorphTargetBuffer* morphTargetBuffer, size_t offset, size_t count);

    /**
     * Utility method to change a MorphTargetBuffer to the given primitive
     */
    inline void setMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex,
-            MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer);
+            MorphTargetBuffer* morphTargetBuffer);

    /**
     * Get a MorphTargetBuffer to the given primitive or null if it doesn't exist.
     */
-    MorphTargetBuffer* UTILS_NULLABLE getMorphTargetBufferAt(Instance instance,
-            uint8_t level, size_t primitiveIndex) const noexcept;
+    MorphTargetBuffer* getMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex) const noexcept;

    /**
     * Gets the number of morphing in the given entity.
@@ -801,22 +776,20 @@ public:
     * @see Engine::setActiveFeatureLevel
     */
    void setMaterialInstanceAt(Instance instance,
-            size_t primitiveIndex, MaterialInstance const* UTILS_NONNULL materialInstance);
+            size_t primitiveIndex, MaterialInstance const* materialInstance);

    /**
     * Retrieves the material instance that is bound to the given primitive.
     */
-    MaterialInstance* UTILS_NULLABLE getMaterialInstanceAt(
-            Instance instance, size_t primitiveIndex) const noexcept;
+    MaterialInstance* getMaterialInstanceAt(Instance instance, size_t primitiveIndex) const noexcept;

    /**
     * Changes the geometry for the given primitive.
     *
     * \see Builder::geometry()
     */
-    void setGeometryAt(Instance instance, size_t primitiveIndex, PrimitiveType type,
-            VertexBuffer* UTILS_NONNULL vertices,
-            IndexBuffer* UTILS_NONNULL indices,
+    void setGeometryAt(Instance instance, size_t primitiveIndex,
+            PrimitiveType type, VertexBuffer* vertices, IndexBuffer* indices,
            size_t offset, size_t count) noexcept;

    /**
@@ -877,9 +850,7 @@ public:
    template<typename VECTOR, typename INDEX,
            typename = typename is_supported_vector_type<VECTOR>::type,
            typename = typename is_supported_index_type<INDEX>::type>
-    static Box computeAABB(
-            VECTOR const* UTILS_NONNULL vertices,
-            INDEX const* UTILS_NONNULL indices, size_t count,
+    static Box computeAABB(VECTOR const* vertices, INDEX const* indices, size_t count,
            size_t stride = sizeof(VECTOR)) noexcept;

 protected:
@@ -887,27 +858,23 @@ protected:
    ~RenderableManager() = default;
 };

-RenderableManager::Builder& RenderableManager::Builder::morphing(
-        uint8_t level, size_t primitiveIndex,
-        MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer) noexcept {
+RenderableManager::Builder& RenderableManager::Builder::morphing(uint8_t level, size_t primitiveIndex,
+        MorphTargetBuffer* morphTargetBuffer) noexcept {
    return morphing(level, primitiveIndex, morphTargetBuffer, 0,
            morphTargetBuffer->getVertexCount());
 }

-void RenderableManager::setMorphTargetBufferAt(
-        Instance instance, uint8_t level, size_t primitiveIndex,
-        MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer) {
+void RenderableManager::setMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex,
+        MorphTargetBuffer* morphTargetBuffer) {
    setMorphTargetBufferAt(instance, level, primitiveIndex, morphTargetBuffer, 0,
            morphTargetBuffer->getVertexCount());
 }

 template<typename VECTOR, typename INDEX, typename, typename>
-Box RenderableManager::computeAABB(
-        VECTOR const* UTILS_NONNULL vertices,
-        INDEX const* UTILS_NONNULL indices,
-        size_t count, size_t stride) noexcept {
-    math::float3 bmin(FLT_MAX);
-    math::float3 bmax(-FLT_MAX);
+Box RenderableManager::computeAABB(VECTOR const* vertices, INDEX const* indices, size_t count,
+        size_t stride) noexcept {
+    math::float3 bmin(std::numeric_limits<float>::max());
+    math::float3 bmax(std::numeric_limits<float>::lowest());
    for (size_t i = 0; i < count; ++i) {
        VECTOR const* p = reinterpret_cast<VECTOR const*>(
                (char const*)vertices + indices[i] * stride);
@@ -23,6 +23,9 @@

 #include <utils/compiler.h>

+#include <backend/PresentCallable.h>
+#include <backend/DriverEnums.h>
+
 #include <math/vec4.h>

 #include <stdint.h>
@@ -181,14 +184,14 @@ public:
     *
     * @return A pointer to the Engine instance this Renderer is associated to.
     */
-    Engine* UTILS_NONNULL getEngine() noexcept;
+    Engine* getEngine() noexcept;

    /**
     * Get the Engine that created this Renderer.
     *
     * @return A constant pointer to the Engine instance this Renderer is associated to.
     */
-    inline Engine const* UTILS_NONNULL getEngine() const noexcept {
+    inline Engine const* getEngine() const noexcept {
        return const_cast<Renderer *>(this)->getEngine();
    }

@@ -264,7 +267,7 @@ public:
     * @see
     * endFrame()
     */
-    bool beginFrame(SwapChain* UTILS_NONNULL swapChain,
+    bool beginFrame(SwapChain* swapChain,
            uint64_t vsyncSteadyClockTimeNano = 0u);

    /**
@@ -335,7 +338,7 @@ public:
     * beginFrame(), endFrame(), View
     *
     */
-    void render(View const* UTILS_NONNULL view);
+    void render(View const* view);

    /**
     * Copy the currently rendered view to the indicated swap chain, using the
@@ -350,7 +353,7 @@ public:
     * copyFrame() should be called after a frame is rendered using render()
     * but before endFrame() is called.
     */
-    void copyFrame(SwapChain* UTILS_NONNULL dstSwapChain, Viewport const& dstViewport,
+    void copyFrame(SwapChain* dstSwapChain, Viewport const& dstViewport,
            Viewport const& srcViewport, uint32_t flags = 0);

    /**
@@ -490,7 +493,7 @@ public:
     * readPixels() is intended for debugging and testing. It will impact performance significantly.
     *
     */
-    void readPixels(RenderTarget* UTILS_NONNULL renderTarget,
+    void readPixels(RenderTarget* renderTarget,
            uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
            backend::PixelBufferDescriptor&& buffer);

@@ -517,7 +520,7 @@ public:
     * However, internally, renderStandaloneView() is highly multi-threaded to both improve
     * performance in mitigate the call's latency.
     */
-    void renderStandaloneView(View const* UTILS_NONNULL view);
+    void renderStandaloneView(View const* view);


    /**
@@ -24,8 +24,6 @@
 #include <utils/compiler.h>
 #include <utils/Invocable.h>

-#include <stddef.h>
-
 namespace utils {
    class Entity;
 } // namespace utils
@@ -75,14 +73,14 @@ public:
     *
     * @param skybox The Skybox to use to fill untouched pixels, or nullptr to unset the Skybox.
     */
-    void setSkybox(Skybox* UTILS_NULLABLE skybox) noexcept;
+    void setSkybox(Skybox* skybox) noexcept;

    /**
     * Returns the Skybox associated with the Scene.
     *
     * @return The associated Skybox, or nullptr if there is none.
     */
-    Skybox* UTILS_NULLABLE getSkybox() const noexcept;
+    Skybox* getSkybox() const noexcept;

    /**
     * Set the IndirectLight to use when rendering the Scene.
@@ -93,7 +91,7 @@ public:
     * @param ibl The IndirectLight to use when rendering the Scene or nullptr to unset.
     * @see getIndirectLight
     */
-    void setIndirectLight(IndirectLight* UTILS_NULLABLE ibl) noexcept;
+    void setIndirectLight(IndirectLight* ibl) noexcept;

    /**
     * Get the IndirectLight or nullptr if none is set.
@@ -101,7 +99,7 @@ public:
     * @return the the IndirectLight or nullptr if none is set
     * @see setIndirectLight
     */
-    IndirectLight* UTILS_NULLABLE getIndirectLight() const noexcept;
+    IndirectLight* getIndirectLight() const noexcept;

    /**
     * Adds an Entity to the Scene.
@@ -120,7 +118,7 @@ public:
     * @param entities Array containing entities to add to the scene.
     * @param count Size of the entity array.
     */
-    void addEntities(const utils::Entity* UTILS_NONNULL entities, size_t count);
+    void addEntities(const utils::Entity* entities, size_t count);

    /**
     * Removes the Renderable from the Scene.
@@ -139,7 +137,7 @@ public:
     * @param entities Array containing entities to remove from the scene.
     * @param count Size of the entity array.
     */
-    void removeEntities(const utils::Entity* UTILS_NONNULL entities, size_t count);
+    void removeEntities(const utils::Entity* entities, size_t count);

    /**
     * Returns the total number of Entities in the Scene, whether alive or not.
@@ -26,7 +26,7 @@
 #include <math/mathfwd.h>

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

 namespace filament {

@@ -74,7 +74,8 @@ public:
         *
         * @param engine Reference to the filament::Engine to associate this SkinningBuffer with.
         *
-         * @return pointer to the newly created object.
+         * @return pointer to the newly created object or nullptr if exceptions are disabled and
+         *         an error occurred.
         *
         * @exception utils::PostConditionPanic if a runtime error occurred, such as running out of
         *            memory or other resources.
@@ -82,7 +83,7 @@ public:
         *
         * @see SkinningBuffer::setBones
         */
-        SkinningBuffer* UTILS_NONNULL build(Engine& engine);
+        SkinningBuffer* build(Engine& engine);
    private:
        friend class FSkinningBuffer;
    };
@@ -95,7 +96,7 @@ public:
     * @param offset offset in elements (not bytes) in the SkinningBuffer (not in transforms)
     * @see RenderableManager::setSkinningBuffer
     */
-    void setBones(Engine& engine, RenderableManager::Bone const* UTILS_NONNULL transforms,
+    void setBones(Engine& engine, RenderableManager::Bone const* transforms,
            size_t count, size_t offset = 0);

    /**
@@ -106,7 +107,7 @@ public:
     * @param offset offset in elements (not bytes) in the SkinningBuffer (not in transforms)
     * @see RenderableManager::setSkinningBuffer
     */
-    void setBones(Engine& engine, math::mat4f const* UTILS_NONNULL transforms,
+    void setBones(Engine& engine, math::mat4f const* transforms,
            size_t count, size_t offset = 0);

    /**
@@ -23,9 +23,8 @@

 #include <utils/compiler.h>

-#include <math/mathfwd.h>
-
 #include <stdint.h>
+#include <math/mathfwd.h>

 namespace filament {

@@ -92,7 +91,7 @@ public:
         *
         * @see Texture
         */
-        Builder& environment(Texture* UTILS_NONNULL cubemap) noexcept;
+        Builder& environment(Texture* cubemap) noexcept;

        /**
         * Indicates whether the sun should be rendered. The sun can only be
@@ -136,9 +135,9 @@ public:
         *
         * @param engine Reference to the filament::Engine to associate this Skybox with.
         *
-         * @return pointer to the newly created object.
+         * @return pointer to the newly created object, or nullptr if the light couldn't be created.
         */
-        Skybox* UTILS_NONNULL build(Engine& engine);
+        Skybox* build(Engine& engine);

    private:
        friend class FSkybox;
@@ -172,9 +171,9 @@ public:
    float getIntensity() const noexcept;

    /**
-     * @return the associated texture
+     * @return the associated texture, or null if it does not exist
     */
-    Texture const* UTILS_NONNULL getTexture() const noexcept;
+    Texture const* getTexture() const noexcept;

 protected:
    // prevent heap allocation
@@ -20,12 +20,12 @@
 #include <filament/FilamentAPI.h>

 #include <backend/DriverEnums.h>
+
+#include <backend/PixelBufferDescriptor.h>
 #include <backend/CallbackHandler.h>

 #include <utils/compiler.h>

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

 class FStream;
@@ -114,7 +114,7 @@ public:
         *
         * @return This Builder, for chaining calls.
         */
-        Builder& stream(void* UTILS_NULLABLE stream) noexcept;
+        Builder& stream(void* stream) noexcept;

        /**
         *
@@ -141,9 +141,9 @@ public:
         *
         * @param engine Reference to the filament::Engine to associate this Stream with.
         *
-         * @return pointer to the newly created object.
+         * @return pointer to the newly created object, or nullptr if the stream couldn't be created.
         */
-        Stream* UTILS_NONNULL build(Engine& engine);
+        Stream* build(Engine& engine);

    private:
        friend class FStream;
@@ -171,12 +171,11 @@ public:
     *
     * @param image      Pointer to AHardwareBuffer, casted to void* since this is a public header.
     * @param callback   This is triggered by Filament when it wishes to release the image.
-     *                   The callback tales two arguments: the AHardwareBuffer and the userdata.
+     *                   It callback tales two arguments: the AHardwareBuffer and the userdata.
     * @param userdata   Optional closure data. Filament will pass this into the callback when it
     *                   releases the image.
     */
-    void setAcquiredImage(void* UTILS_NONNULL image,
-            Callback UTILS_NONNULL callback, void* UTILS_NULLABLE userdata) noexcept;
+    void setAcquiredImage(void* image, Callback callback, void* userdata) noexcept;

    /**
     * @see setAcquiredImage(void*, Callback, void*)
@@ -188,9 +187,7 @@ public:
     * @param userdata   Optional closure data. Filament will pass this into the callback when it
     *                   releases the image.
     */
-    void setAcquiredImage(void* UTILS_NONNULL image,
-            backend::CallbackHandler* UTILS_NULLABLE handler,
-            Callback UTILS_NONNULL callback, void* UTILS_NULLABLE userdata) noexcept;
+    void setAcquiredImage(void* image, backend::CallbackHandler* handler, Callback callback, void* userdata) noexcept;

    /**
     * Updates the size of the incoming stream. Whether this value is used is
@@ -21,12 +21,11 @@

 #include <backend/CallbackHandler.h>
 #include <backend/DriverEnums.h>
+#include <backend/PresentCallable.h>

 #include <utils/compiler.h>
 #include <utils/Invocable.h>

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

 class Engine;
@@ -152,7 +151,7 @@ class Engine;
 class UTILS_PUBLIC SwapChain : public FilamentAPI {
 public:
    using FrameScheduledCallback = backend::FrameScheduledCallback;
-    using FrameCompletedCallback = utils::Invocable<void(SwapChain* UTILS_NONNULL)>;
+    using FrameCompletedCallback = utils::Invocable<void(SwapChain*)>;

    /**
     * Requests a SwapChain with an alpha channel.
@@ -236,7 +235,7 @@ public:
     */
    static bool isSRGBSwapChainSupported(Engine& engine) noexcept;

-    void* UTILS_NULLABLE getNativeWindow() const noexcept;
+    void* getNativeWindow() const noexcept;

    /**
     * FrameScheduledCallback is a callback function that notifies an application when Filament has
@@ -269,8 +268,7 @@ public:
     *
     * @see PresentCallable
     */
-    void setFrameScheduledCallback(FrameScheduledCallback UTILS_NULLABLE callback,
-            void* UTILS_NULLABLE user = nullptr);
+    void setFrameScheduledCallback(FrameScheduledCallback callback, void* user = nullptr);

    /**
     * FrameCompletedCallback is a callback function that notifies an application when a frame's
@@ -291,7 +289,7 @@ public:
     *
     * @see CallbackHandler
     */
-    void setFrameCompletedCallback(backend::CallbackHandler* UTILS_NULLABLE handler = nullptr,
+    void setFrameCompletedCallback(backend::CallbackHandler* handler = nullptr,
            FrameCompletedCallback&& callback = {}) noexcept;


@@ -20,16 +20,12 @@
 #define TNT_FILAMENT_TEXTURE_H

 #include <filament/FilamentAPI.h>
-
 #include <backend/DriverEnums.h>
 #include <backend/PixelBufferDescriptor.h>

 #include <utils/compiler.h>

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

 namespace filament {

@@ -204,13 +200,14 @@ public:
         *
         * @param engine Reference to the filament::Engine to associate this Texture with.
         *
-         * @return pointer to the newly created object.
+         * @return pointer to the newly created object or nullptr if exceptions are disabled and
+         *         an error occurred.
         *
         * @exception utils::PostConditionPanic if a runtime error occurred, such as running out of
         *            memory or other resources.
         * @exception utils::PreConditionPanic if a parameter to a builder function was invalid.
         */
-        Texture* UTILS_NONNULL build(Engine& engine);
+        Texture* build(Engine& engine);

        /* no user serviceable parts below */

@@ -399,7 +396,7 @@ public:
     * @see Builder::sampler()
     *
     */
-    void setExternalImage(Engine& engine, void* UTILS_NONNULL image) noexcept;
+    void setExternalImage(Engine& engine, void* image) noexcept;

    /**
     * Specify the external image and plane to associate with this Texture. Typically the external
@@ -430,7 +427,7 @@ public:
     *                      kCVPixelFormatType_420YpCbCr8BiPlanarFullRange images. On platforms
     *                      other than iOS, this method is a no-op.
     */
-    void setExternalImage(Engine& engine, void* UTILS_NONNULL image, size_t plane) noexcept;
+    void setExternalImage(Engine& engine, void* image, size_t plane) noexcept;

    /**
     * Specify the external stream to associate with this Texture. Typically the external
@@ -449,7 +446,7 @@ public:
     * @see Builder::sampler(), Stream
     *
     */
-    void setExternalStream(Engine& engine, Stream* UTILS_NULLABLE stream) noexcept;
+    void setExternalStream(Engine& engine, Stream* stream) noexcept;

    /**
     * Generates all the mipmap levels automatically. This requires the texture to have a
@@ -499,7 +496,7 @@ public:
     */
    void generatePrefilterMipmap(Engine& engine,
            PixelBufferDescriptor&& buffer, const FaceOffsets& faceOffsets,
-            PrefilterOptions const* UTILS_NULLABLE options = nullptr);
+            PrefilterOptions const* options = nullptr);


    /** @deprecated */
@@ -25,8 +25,6 @@

 #include <math.h>

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

 /**
@@ -21,8 +21,6 @@

 #include <math/mathfwd.h>

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

 /**
@@ -26,7 +26,6 @@

 #include <iterator>

-#include <stddef.h>

 namespace utils {
 class Entity;
@@ -140,7 +139,7 @@ public:
     * Retrieve the Entities of all the components of this manager.
     * @return A list, in no particular order, of all the entities managed by this manager.
     */
-    utils::Entity const* UTILS_NONNULL getEntities() const noexcept;
+    utils::Entity const* getEntities() const noexcept;

    /**
     * Enables or disable the accurate translation mode. Disabled by default.
@@ -224,7 +223,7 @@ public:
     * @param count The maximum number of children to retrieve.
     * @return The number of children written to the pointer.
     */
-    size_t getChildren(Instance i, utils::Entity* UTILS_NONNULL children, size_t count) const noexcept;
+    size_t getChildren(Instance i, utils::Entity* children, size_t count) const noexcept;

    /**
     * Returns an iterator to the Instance of the first child of the given parent.
@@ -27,9 +27,6 @@

 #include <utils/compiler.h>

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

 class FVertexBuffer;
@@ -163,13 +160,14 @@ public:
         *
         * @param engine Reference to the filament::Engine to associate this VertexBuffer with.
         *
-         * @return pointer to the newly created object.
+         * @return pointer to the newly created object or nullptr if exceptions are disabled and
+         *         an error occurred.
         *
         * @exception utils::PostConditionPanic if a runtime error occurred, such as running out of
         *            memory or other resources.
         * @exception utils::PreConditionPanic if a parameter to a builder function was invalid.
         */
-        VertexBuffer* UTILS_NONNULL build(Engine& engine);
+        VertexBuffer* build(Engine& engine);

    private:
        friend class FVertexBuffer;
@@ -208,8 +206,7 @@ public:
     *                    and Builder::bufferCount() - 1.
     * @param bufferObject The handle to the GPU data that will be used in this buffer slot.
     */
-    void setBufferObjectAt(Engine& engine, uint8_t bufferIndex,
-            BufferObject const*  UTILS_NONNULL bufferObject);
+    void setBufferObjectAt(Engine& engine, uint8_t bufferIndex, BufferObject const* bufferObject);

 protected:
    // prevent heap allocation
@@ -19,19 +19,17 @@
 #ifndef TNT_FILAMENT_VIEW_H
 #define TNT_FILAMENT_VIEW_H

+#include <filament/Color.h>
 #include <filament/FilamentAPI.h>
 #include <filament/Options.h>

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

 #include <math/mathfwd.h>

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

 namespace backend {
@@ -93,7 +91,7 @@ public:
     * Sets the View's name. Only useful for debugging.
     * @param name Pointer to the View's name. The string is copied.
     */
-    void setName(const char* UTILS_NONNULL name) noexcept;
+    void setName(const char* name) noexcept;

    /**
     * Returns the View's name
@@ -102,7 +100,7 @@ public:
     *
     * @attention Do *not* free the pointer or modify its content.
     */
-    const char* UTILS_NULLABLE getName() const noexcept;
+    const char* getName() const noexcept;

    /**
     * Set this View instance's Scene.
@@ -118,19 +116,19 @@ public:
     *  There is no reference-counting.
     *  Make sure to dissociate a Scene from all Views before destroying it.
     */
-    void setScene(Scene* UTILS_NULLABLE scene);
+    void setScene(Scene* scene);

    /**
     * Returns the Scene currently associated with this View.
     * @return A pointer to the Scene associated to this View. nullptr if no Scene is set.
     */
-    Scene* UTILS_NULLABLE getScene() noexcept;
+    Scene* getScene() noexcept;

    /**
     * Returns the Scene currently associated with this View.
     * @return A pointer to the Scene associated to this View. nullptr if no Scene is set.
     */
-    Scene const* UTILS_NULLABLE getScene() const noexcept {
+    Scene const* getScene() const noexcept {
        return const_cast<View*>(this)->getScene();
    }

@@ -145,7 +143,7 @@ public:
     *
     * @param renderTarget Render target associated with view, or nullptr for the swap chain.
     */
-    void setRenderTarget(RenderTarget* UTILS_NULLABLE renderTarget) noexcept;
+    void setRenderTarget(RenderTarget* renderTarget) noexcept;

    /**
     * Gets the offscreen render target associated with this view.
@@ -154,7 +152,7 @@ public:
     *
     * @see setRenderTarget
     */
-    RenderTarget* UTILS_NULLABLE getRenderTarget() const noexcept;
+    RenderTarget* getRenderTarget() const noexcept;

    /**
     * Sets the rectangular region to render to.
@@ -187,7 +185,7 @@ public:
     *  There is no reference-counting.
     *  Make sure to dissociate a Camera from all Views before destroying it.
     */
-    void setCamera(Camera* UTILS_NONNULL camera) noexcept;
+    void setCamera(Camera* camera) noexcept;

    /**
     * Returns the Camera currently associated with this View.
@@ -404,13 +402,13 @@ public:
     *  There is no reference-counting.
     *  Make sure to dissociate a ColorGrading from all Views before destroying it.
     */
-    void setColorGrading(ColorGrading* UTILS_NULLABLE colorGrading) noexcept;
+    void setColorGrading(ColorGrading* colorGrading) noexcept;

    /**
     * Returns the color grading transforms currently associated to this view.
     * @return A pointer to the ColorGrading associated to this View.
     */
-    const ColorGrading* UTILS_NULLABLE getColorGrading() const noexcept;
+    const ColorGrading* getColorGrading() const noexcept;

    /**
     * Sets ambient occlusion options.
@@ -716,10 +714,10 @@ public:
    bool isFrustumCullingEnabled() const noexcept;

    //! debugging: sets the Camera used for rendering. It may be different from the culling camera.
-    void setDebugCamera(Camera* UTILS_NULLABLE camera) noexcept;
+    void setDebugCamera(Camera* camera) noexcept;

    //! debugging: returns a Camera from the point of view of *the* dominant directional light used for shadowing.
-    Camera const* UTILS_NULLABLE getDirectionalLightCamera() const noexcept;
+    Camera const* getDirectionalLightCamera() const noexcept;


    /** Result of a picking query */
@@ -748,12 +746,11 @@ public:
    /** User data for PickingQueryResultCallback */
    struct PickingQuery {
        // note: this is enough to store a std::function<> -- just saying...
-        void* UTILS_NULLABLE storage[4];
+        void* storage[4];
    };

    /** callback type used for picking queries. */
-    using PickingQueryResultCallback =
-            void(*)(PickingQueryResult const& result, PickingQuery* UTILS_NONNULL pq);
+    using PickingQueryResultCallback = void(*)(PickingQueryResult const& result, PickingQuery* pq);

    /**
     * Helper for creating a picking query from Foo::method, by pointer.
@@ -767,8 +764,8 @@ public:
     * @param handler   Handler to dispatch the callback or nullptr for the default handler.
     */
    template<typename T, void(T::*method)(PickingQueryResult const&)>
-    void pick(uint32_t x, uint32_t y, T* UTILS_NONNULL instance,
-            backend::CallbackHandler* UTILS_NULLABLE handler = nullptr) noexcept {
+    void pick(uint32_t x, uint32_t y, T* instance,
+            backend::CallbackHandler* handler = nullptr) noexcept {
        PickingQuery& query = pick(x, y, [](PickingQueryResult const& result, PickingQuery* pq) {
            (static_cast<T*>(pq->storage[0])->*method)(result);
        }, handler);
@@ -788,7 +785,7 @@ public:
     */
    template<typename T, void(T::*method)(PickingQueryResult const&)>
    void pick(uint32_t x, uint32_t y, T instance,
-            backend::CallbackHandler* UTILS_NULLABLE handler = nullptr) noexcept {
+            backend::CallbackHandler* handler = nullptr) noexcept {
        static_assert(sizeof(instance) <= sizeof(PickingQuery::storage), "user data too large");
        PickingQuery& query = pick(x, y, [](PickingQueryResult const& result, PickingQuery* pq) {
            T* const that = static_cast<T*>(reinterpret_cast<void*>(pq->storage));
@@ -809,7 +806,7 @@ public:
     */
    template<typename T>
    void pick(uint32_t x, uint32_t y, T functor,
-            backend::CallbackHandler* UTILS_NULLABLE handler = nullptr) noexcept {
+            backend::CallbackHandler* handler = nullptr) noexcept {
        static_assert(sizeof(functor) <= sizeof(PickingQuery::storage), "functor too large");
        PickingQuery& query = pick(x, y, handler,
                (PickingQueryResultCallback)[](PickingQueryResult const& result, PickingQuery* pq) {
@@ -835,9 +832,8 @@ public:
     *                  8*sizeof(void*) bytes of user data. This user data is later accessible
     *                  in the PickingQueryResultCallback callback 3rd parameter.
     */
-    PickingQuery& pick(uint32_t x, uint32_t y,
-            backend::CallbackHandler* UTILS_NULLABLE handler,
-            PickingQueryResultCallback UTILS_NONNULL callback) noexcept;
+    PickingQuery& pick(uint32_t x, uint32_t y, backend::CallbackHandler* handler,
+            PickingQueryResultCallback callback) noexcept;

    /**
     * Set the value of material global variables. There are up-to four such variable each of
@@ -23,6 +23,9 @@

 #include <utils/compiler.h>

+#include <math/scalar.h>
+#include <math/mathfwd.h>
+
 #include <stddef.h>
 #include <stdint.h>

@@ -54,7 +54,7 @@ using LinearAllocatorArena = utils::Arena<

 #endif

-using RootArenaScope = utils::ArenaScope<LinearAllocatorArena>;
+using ArenaScope = utils::ArenaScope<LinearAllocatorArena>;

 } // namespace filament

@@ -168,8 +168,7 @@ void Froxelizer::setProjection(const mat4f& projection,
 }

 bool Froxelizer::prepare(
-        FEngine::DriverApi& driverApi, RootArenaScope& rootArenaScope,
-        filament::Viewport const& viewport,
+        FEngine::DriverApi& driverApi, ArenaScope& arena, filament::Viewport const& viewport,
        const mat4f& projection, float projectionNear, float projectionFar) noexcept {
    setViewport(viewport);
    setProjection(projection, projectionNear, projectionFar);
@@ -200,12 +199,12 @@ bool Froxelizer::prepare(

    // light records per froxel (~256 KiB)
    mLightRecords = {
-            rootArenaScope.allocate<LightRecord>(getFroxelBufferEntryCount(), CACHELINE_SIZE),
+            arena.allocate<LightRecord>(getFroxelBufferEntryCount(), CACHELINE_SIZE),
            getFroxelBufferEntryCount() };

    // froxel thread data (~256 KiB)
    mFroxelShardedData = {
-            rootArenaScope.allocate<FroxelThreadData>(GROUP_COUNT, CACHELINE_SIZE),
+            arena.allocate<FroxelThreadData>(GROUP_COUNT, CACHELINE_SIZE),
            uint32_t(GROUP_COUNT)
    };

@@ -110,7 +110,7 @@ public:
     *
     * return true if updateUniforms() needs to be called
     */
-    bool prepare(backend::DriverApi& driverApi, RootArenaScope& rootArenaScope, Viewport const& viewport,
+    bool prepare(backend::DriverApi& driverApi, ArenaScope& arena, Viewport const& viewport,
            const math::mat4f& projection, float projectionNear, float projectionFar) noexcept;

    Froxel getFroxelAt(size_t x, size_t y, size_t z) const noexcept;
@@ -25,7 +25,7 @@ MaterialInstance* Material::createInstance(const char* name) const noexcept {
 }

 const char* Material::getName() const noexcept {
-    return downcast(this)->getName().c_str_safe();
+    return downcast(this)->getName().c_str();
 }

 Shading Material::getShading()  const noexcept {
@@ -64,7 +64,6 @@
 #include <math/mat3.h>
 #include <math/mat4.h>
 #include <math/scalar.h>
-#include <math/vec2.h>
 #include <math/vec3.h>
 #include <math/vec4.h>

@@ -75,7 +74,6 @@
 #include <utils/FixedCapacityVector.h>

 #include <algorithm>
-#include <array>
 #include <cmath>
 #include <limits>
 #include <string_view>
@@ -115,7 +113,7 @@ PostProcessManager::PostProcessMaterial::PostProcessMaterial() noexcept {
 }

 PostProcessManager::PostProcessMaterial::PostProcessMaterial(MaterialInfo const& info) noexcept
-        : PostProcessMaterial() {
+    : PostProcessMaterial() {
    mData = info.data; // aliased to mMaterial
    mSize = info.size;
    mConstants = info.constants;
@@ -167,7 +165,7 @@ void PostProcessManager::PostProcessMaterial::loadMaterial(FEngine& engine) cons
    mHasMaterial = true;
    auto builder = Material::Builder();
    builder.package(mData, mSize);
-    for (auto const& constant: mConstants) {
+    for (auto const& constant : mConstants) {
        std::visit([&](auto&& arg) {
            builder.constant(constant.name.data(), constant.name.size(), arg);
        }, constant.value);
@@ -189,60 +187,63 @@ PipelineState PostProcessManager::PostProcessMaterial::getPipelineState(
    FMaterial* const material = getMaterial(engine);
    material->prepareProgram(Variant{ variantKey });
    return {
-            .program = material->getProgram(Variant{ variantKey }),
+            .program = material->getProgram(Variant{variantKey}),
            .rasterState = material->getRasterState(),
            .scissor = material->getDefaultInstance()->getScissor()
    };
 }

 UTILS_NOINLINE
-FMaterialInstance* PostProcessManager::PostProcessMaterial::getMaterialInstance(
-        FEngine& engine) const noexcept {
+FMaterialInstance* PostProcessManager::PostProcessMaterial::getMaterialInstance(FEngine& engine) const noexcept {
    FMaterial* const material = getMaterial(engine);
    return material->getDefaultInstance();
 }

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

-const PostProcessManager::JitterSequence<4> PostProcessManager::sRGSS4 = {{{
+const PostProcessManager::JitterSequence<4> PostProcessManager::sRGSS4 = {{
        { 0.625f, 0.125f },
        { 0.125f, 0.375f },
        { 0.875f, 0.625f },
        { 0.375f, 0.875f }
-}}};
+}};

-const PostProcessManager::JitterSequence<4> PostProcessManager::sUniformHelix4 = {{{
+const PostProcessManager::JitterSequence<4> PostProcessManager::sUniformHelix4 = {{
        { 0.25f, 0.25f },
        { 0.75f, 0.75f },
        { 0.25f, 0.75f },
        { 0.75f, 0.25f },
-}}};
+}};

-template<size_t COUNT>
-constexpr auto halton() {
-    std::array<float2, COUNT> h;
-    for (size_t i = 0; i < COUNT; i++) {
-        h[i] = {
-                filament::halton(i, 2),
-                filament::halton(i, 3) };
-    }
-    return h;
-}
-
-const PostProcessManager::JitterSequence<32>
-        PostProcessManager::sHaltonSamples = { halton<32>() };
+const PostProcessManager::JitterSequence<16> PostProcessManager::sHaltonSamples = {{
+        { filament::halton( 0, 2), filament::halton( 0, 3) },
+        { filament::halton( 1, 2), filament::halton( 1, 3) },
+        { filament::halton( 2, 2), filament::halton( 2, 3) },
+        { filament::halton( 3, 2), filament::halton( 3, 3) },
+        { filament::halton( 4, 2), filament::halton( 4, 3) },
+        { filament::halton( 5, 2), filament::halton( 5, 3) },
+        { filament::halton( 6, 2), filament::halton( 6, 3) },
+        { filament::halton( 7, 2), filament::halton( 7, 3) },
+        { filament::halton( 8, 2), filament::halton( 8, 3) },
+        { filament::halton( 9, 2), filament::halton( 9, 3) },
+        { filament::halton(10, 2), filament::halton(10, 3) },
+        { filament::halton(11, 2), filament::halton(11, 3) },
+        { filament::halton(12, 2), filament::halton(12, 3) },
+        { filament::halton(13, 2), filament::halton(13, 3) },
+        { filament::halton(14, 2), filament::halton(14, 3) },
+        { filament::halton(15, 2), filament::halton(15, 3) }
+}};

 PostProcessManager::PostProcessManager(FEngine& engine) noexcept
        : mEngine(engine),
-          mWorkaroundSplitEasu(false),
-          mWorkaroundAllowReadOnlyAncillaryFeedbackLoop(false) {
+         mWorkaroundSplitEasu(false),
+         mWorkaroundAllowReadOnlyAncillaryFeedbackLoop(false) {
 }

 PostProcessManager::~PostProcessManager() noexcept = default;

 UTILS_NOINLINE
-void PostProcessManager::registerPostProcessMaterial(std::string_view name,
-        MaterialInfo const& info) {
+void PostProcessManager::registerPostProcessMaterial(std::string_view name, MaterialInfo const& info) {
    mMaterialRegistry.try_emplace(name, info);
 }

@@ -307,7 +308,6 @@ static const PostProcessManager::MaterialInfo sMaterialList[] = {
        { "fsr_rcas",                   MATERIAL(FSR_RCAS) },
        { "debugShadowCascades",        MATERIAL(DEBUGSHADOWCASCADES) },
        { "resolveDepth",               MATERIAL(RESOLVEDEPTH) },
-        { "shadowmap",                  MATERIAL(SHADOWMAP) },
 };

 void PostProcessManager::init() noexcept {
@@ -326,13 +326,13 @@ void PostProcessManager::init() noexcept {
            driver.isWorkaroundNeeded(Workaround::ALLOW_READ_ONLY_ANCILLARY_FEEDBACK_LOOP);

    #pragma nounroll
-    for (auto const& info: sMaterialListFeatureLevel0) {
+    for (auto const& info : sMaterialListFeatureLevel0) {
        registerPostProcessMaterial(info.name, info);
    }

    if (mEngine.getActiveFeatureLevel() >= FeatureLevel::FEATURE_LEVEL_1) {
        #pragma nounroll
-        for (auto const& info: sMaterialList) {
+        for (auto const& info : sMaterialList) {
            registerPostProcessMaterial(info.name, info);
        }
    }
@@ -414,7 +414,7 @@ void PostProcessManager::commitAndRender(FrameGraphResources::RenderPassInfo con
 // ------------------------------------------------------------------------------------------------

 PostProcessManager::StructurePassOutput PostProcessManager::structure(FrameGraph& fg,
-        RenderPassBuilder const& passBuilder, uint8_t structureRenderFlags,
+        RenderPass const& pass, uint8_t structureRenderFlags,
        uint32_t width, uint32_t height,
        StructurePassConfig const& config) noexcept {

@@ -466,19 +466,17 @@ PostProcessManager::StructurePassOutput PostProcessManager::structure(FrameGraph
                        .clearFlags = TargetBufferFlags::COLOR0 | TargetBufferFlags::DEPTH
                });
            },
-            [=, passBuilder = passBuilder](FrameGraphResources const& resources,
+            [=, renderPass = pass](FrameGraphResources const& resources,
                    auto const&, DriverApi&) mutable {
                Variant structureVariant(Variant::DEPTH_VARIANT);
                structureVariant.setPicking(config.picking);

                auto out = resources.getRenderPassInfo();
-
-                passBuilder.renderFlags(structureRenderFlags);
-                passBuilder.variant(structureVariant);
-                passBuilder.commandTypeFlags(RenderPass::CommandTypeFlags::SSAO);
-
-                RenderPass const pass{ passBuilder.build(mEngine) };
-                RenderPass::execute(pass, mEngine, resources.getPassName(), out.target, out.params);
+                renderPass.setRenderFlags(structureRenderFlags);
+                renderPass.setVariant(structureVariant);
+                renderPass.appendCommands(mEngine, RenderPass::CommandTypeFlags::SSAO);
+                renderPass.sortCommands(mEngine);
+                renderPass.execute(mEngine, resources.getPassName(), out.target, out.params);
            });

    auto depth = structurePass->depth;
@@ -525,7 +523,7 @@ PostProcessManager::StructurePassOutput PostProcessManager::structure(FrameGraph
 // ------------------------------------------------------------------------------------------------

 FrameGraphId<FrameGraphTexture> PostProcessManager::ssr(FrameGraph& fg,
-        RenderPassBuilder const& passBuilder,
+        RenderPass const& pass,
        FrameHistory const& frameHistory,
        CameraInfo const& cameraInfo,
        PerViewUniforms& uniforms,
@@ -588,7 +586,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::ssr(FrameGraph& fg,
            },
            [this, projection = cameraInfo.projection,
                    userViewMatrix = cameraInfo.getUserViewMatrix(), uvFromClipMatrix, historyProjection,
-                    options, &uniforms, passBuilder = passBuilder]
+                    options, &uniforms, renderPass = pass]
            (FrameGraphResources const& resources, auto const& data, DriverApi& driver) mutable {
                // set structure sampler
                uniforms.prepareStructure(data.structure ?
@@ -609,17 +607,17 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::ssr(FrameGraph& fg,
                auto out = resources.getRenderPassInfo();

                // Remove the HAS_SHADOWING RenderFlags, since it's irrelevant when rendering reflections
-                passBuilder.renderFlags(~RenderPass::HAS_SHADOWING, 0);
+                RenderPass::RenderFlags flags = renderPass.getRenderFlags();
+                flags &= ~RenderPass::HAS_SHADOWING;
+                renderPass.setRenderFlags(flags);

                // use our special SSR variant, it can only be applied to object that have
                // the SCREEN_SPACE ReflectionMode.
-                passBuilder.variant(Variant{ Variant::SPECIAL_SSR });
-
+                renderPass.setVariant(Variant{Variant::SPECIAL_SSR});
                // generate all our drawing commands, except blended objects.
-                passBuilder.commandTypeFlags(RenderPass::CommandTypeFlags::SCREEN_SPACE_REFLECTIONS);
-
-                RenderPass const pass{ passBuilder.build(mEngine) };
-                RenderPass::execute(pass, mEngine, resources.getPassName(), out.target, out.params);
+                renderPass.appendCommands(mEngine, RenderPass::CommandTypeFlags::SCREEN_SPACE_REFLECTIONS);
+                renderPass.sortCommands(mEngine);
+                renderPass.execute(mEngine, resources.getPassName(), out.target, out.params);
            });

    return ssrPass->reflections;
@@ -1685,8 +1683,8 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::dof(FrameGraph& fg,

    // we assume the width/height is already multiple of 16
    assert_invariant(!(colorDesc.width  & 0xF) && !(colorDesc.height & 0xF));
-    const uint32_t tileBufferWidth  = width;
-    const uint32_t tileBufferHeight = height;
+    const uint32_t tileBufferWidth  = colorDesc.width  / dofResolution;
+    const uint32_t tileBufferHeight = colorDesc.height / dofResolution;
    const size_t tileReductionCount = ctz(tileSize / dofResolution);

    struct PostProcessDofTiling1 {
@@ -2571,8 +2569,6 @@ void PostProcessManager::prepareTaa(FrameGraph& fg,
            case JitterPattern::HALTON_23_X8:
                return sHaltonSamples(frameIndex % 8);
            case JitterPattern::HALTON_23_X16:
-                return sHaltonSamples(frameIndex % 16);
-            case JitterPattern::HALTON_23_X32:
                return sHaltonSamples(frameIndex);
        }
    };
@@ -2622,7 +2618,6 @@ void PostProcessManager::configureTemporalAntiAliasingMaterial(
        }
    };

-    setConstantParameter(ma, "upscaling", taaOptions.upscaling);
    setConstantParameter(ma, "historyReprojection", taaOptions.historyReprojection);
    setConstantParameter(ma, "filterHistory", taaOptions.filterHistory);
    setConstantParameter(ma, "filterInput", taaOptions.filterInput);
@@ -2772,77 +2767,24 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
            });

    input = colorGradingConfig.asSubpass ? taaPass->tonemappedOutput : taaPass->output;
-    auto history = input;
-
-    // optional sharpen pass from FSR1
-    if (taaOptions.sharpness > 0.0f) {
-        input = rcas(fg, taaOptions.sharpness,
-                input, fg.getDescriptor(input), colorGradingConfig.translucent);
-    }

    struct ExportColorHistoryData {
        FrameGraphId<FrameGraphTexture> color;
    };
-    fg.addPass<ExportColorHistoryData>("Export TAA history",
+    auto& exportHistoryPass = fg.addPass<ExportColorHistoryData>("Export TAA history",
            [&](FrameGraph::Builder& builder, auto& data) {
                // We need to use sideEffect here to ensure this pass won't be culled.
                // The "output" of this pass is going to be used during the next frame as
                // an "import".
                builder.sideEffect();
-                data.color = builder.sample(history); // FIXME: an access must be declared for detach(), why?
-            }, [&current](FrameGraphResources const& resources, auto const& data, auto&) {
-                resources.detach(data.color, &current.color, &current.desc);
+                data.color = builder.sample(input); // FIXME: an access must be declared for detach(), why?
+            }, [&current](FrameGraphResources const& resources, auto const& data,
+                    backend::DriverApi&) {
+                resources.detach(data.color,
+                        &current.color, &current.desc);
            });

-    return input;
-}
-
-FrameGraphId<FrameGraphTexture> PostProcessManager::rcas(
-        FrameGraph& fg,
-        float sharpness,
-        FrameGraphId<FrameGraphTexture> input,
-        FrameGraphTexture::Descriptor const& outDesc,
-        bool translucent) {
-
-    struct QuadBlitData {
-        FrameGraphId<FrameGraphTexture> input;
-        FrameGraphId<FrameGraphTexture> output;
-    };
-
-    auto& ppFsrRcas = fg.addPass<QuadBlitData>("FidelityFX FSR1 Rcas",
-            [&](FrameGraph::Builder& builder, auto& data) {
-                data.input = builder.sample(input);
-                data.output = builder.createTexture("FFX FSR1 Rcas output", outDesc);
-                data.output = builder.declareRenderPass(data.output);
-            },
-            [=](FrameGraphResources const& resources,
-                    auto const& data, DriverApi& driver) {
-
-                auto input = resources.getTexture(data.input);
-                auto out = resources.getRenderPassInfo();
-                auto const& outputDesc = resources.getDescriptor(data.input);
-
-                auto& material = getPostProcessMaterial("fsr_rcas");
-                FMaterialInstance* const mi = material.getMaterialInstance(mEngine);
-
-                FSRUniforms uniforms;
-                FSR_SharpeningSetup(&uniforms, { .sharpness = 2.0f - 2.0f * sharpness });
-                mi->setParameter("RcasCon", uniforms.RcasCon);
-                mi->setParameter("color", input, {}); // uses texelFetch
-                mi->setParameter("resolution", float4{
-                        outputDesc.width, outputDesc.height,
-                        1.0f / outputDesc.width, 1.0f / outputDesc.height });
-                mi->commit(driver);
-                mi->use(driver);
-
-                const uint8_t variant = uint8_t(
-                        translucent ? PostProcessVariant::TRANSLUCENT : PostProcessVariant::OPAQUE);
-
-                PipelineState const pipeline(material.getPipelineState(mEngine, variant));
-                render(out, pipeline, driver);
-            });
-
-    return ppFsrRcas->output;
+    return exportHistoryPass->color;
 }

 FrameGraphId<FrameGraphTexture> PostProcessManager::upscale(FrameGraph& fg, bool translucent,
@@ -3005,9 +2947,41 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::upscale(FrameGraph& fg, bool
    auto output = ppQuadBlit->output;

    // if we had to take the low quality fallback, we still do the "sharpen pass"
-    if (dsrOptions.sharpness > 0.0f &&
-            (dsrOptions.quality != QualityLevel::LOW || lowQualityFallback)) {
-        output = rcas(fg, dsrOptions.sharpness, output, outDesc, translucent);
+    if (dsrOptions.sharpness > 0.0f && (dsrOptions.quality != QualityLevel::LOW || lowQualityFallback)) {
+        auto& ppFsrRcas = fg.addPass<QuadBlitData>("FidelityFX FSR1 Rcas",
+                [&](FrameGraph::Builder& builder, auto& data) {
+                    data.input = builder.sample(output);
+                    data.output = builder.createTexture("FFX FSR1 Rcas output", outDesc);
+                    data.output = builder.declareRenderPass(data.output);
+                },
+                [=](FrameGraphResources const& resources,
+                        auto const& data, DriverApi& driver) {
+
+                    auto color = resources.getTexture(data.input);
+                    auto out = resources.getRenderPassInfo();
+                    auto const& outputDesc = resources.getDescriptor(data.output);
+
+                    auto& material = getPostProcessMaterial("fsr_rcas");
+                    FMaterialInstance* const mi = material.getMaterialInstance(mEngine);
+
+                    FSRUniforms uniforms;
+                    FSR_SharpeningSetup(&uniforms, { .sharpness = 2.0f - 2.0f * dsrOptions.sharpness });
+                    mi->setParameter("RcasCon", uniforms.RcasCon);
+                    mi->setParameter("color", color, { }); // uses texelFetch
+                    mi->setParameter("resolution", float4{
+                            outputDesc.width, outputDesc.height,
+                            1.0f / outputDesc.width, 1.0f / outputDesc.height });
+                    mi->commit(driver);
+                    mi->use(driver);
+
+                    const uint8_t variant = uint8_t(translucent ?
+                            PostProcessVariant::TRANSLUCENT : PostProcessVariant::OPAQUE);
+
+                    PipelineState const pipeline(material.getPipelineState(mEngine, variant));
+                    render(out, pipeline, driver);
+                });
+
+        output = ppFsrRcas->output;
    }

    // we rely on automatic culling of unused render passes
@@ -3276,51 +3250,4 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::debugShadowCascades(FrameGra
    return debugShadowCascadePass->output;
 }

-FrameGraphId<FrameGraphTexture> PostProcessManager::debugDisplayShadowTexture(
-        FrameGraph& fg,
-        FrameGraphId<FrameGraphTexture> input,
-        FrameGraphId<FrameGraphTexture> shadowmap, float scale,
-        uint8_t layer, uint8_t level, uint8_t channel, float power) noexcept {
-    if (shadowmap) {
-        struct ShadowMapData {
-            FrameGraphId<FrameGraphTexture> color;
-            FrameGraphId<FrameGraphTexture> depth;
-        };
-
-        auto const& desc = fg.getDescriptor(input);
-        float const ratio = float(desc.height) / float(desc.width);
-        float const screenScale = float(fg.getDescriptor(shadowmap).height) / float(desc.height);
-        float2 const s = { screenScale * scale * ratio, screenScale * scale };
-
-        auto& shadomapDebugPass = fg.addPass<ShadowMapData>("shadowmap debug pass",
-                [&](FrameGraph::Builder& builder, auto& data) {
-                    data.color = builder.read(input,
-                            FrameGraphTexture::Usage::COLOR_ATTACHMENT);
-                    data.color = builder.write(data.color,
-                            FrameGraphTexture::Usage::COLOR_ATTACHMENT);
-                    data.depth = builder.sample(shadowmap);
-                    builder.declareRenderPass("color target", {
-                            .attachments = { .color = { data.color }}
-                    });
-                },
-                [=](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {
-                    auto out = resources.getRenderPassInfo();
-                    auto in = resources.getTexture(data.depth);
-                    auto const& material = getPostProcessMaterial("shadowmap");
-                    FMaterialInstance* const mi = material.getMaterialInstance(mEngine);
-                    mi->setParameter("shadowmap", in, {
-                        .filterMin = SamplerMinFilter::NEAREST_MIPMAP_NEAREST });
-                    mi->setParameter("scale", s);
-                    mi->setParameter("layer", (uint32_t)layer);
-                    mi->setParameter("level", (uint32_t)level);
-                    mi->setParameter("channel", (uint32_t)channel);
-                    mi->setParameter("power", power);
-                    commitAndRender(out, material, driver);
-                });
-        input = shadomapDebugPass->color;
-    }
-    return input;
-}
-
-
 } // namespace filament
@@ -36,7 +36,6 @@

 #include <tsl/robin_map.h>

-#include <array>
 #include <random>
 #include <string_view>
 #include <variant>
@@ -50,7 +49,6 @@ class FMaterialInstance;
 class FrameGraph;
 class PerViewUniforms;
 class RenderPass;
-class RenderPassBuilder;
 struct CameraInfo;

 class PostProcessManager {
@@ -100,12 +98,12 @@ public:
        FrameGraphId<FrameGraphTexture> picking;
    };
    StructurePassOutput structure(FrameGraph& fg,
-            RenderPassBuilder const& passBuilder, uint8_t structureRenderFlags,
+            RenderPass const& pass, uint8_t structureRenderFlags,
            uint32_t width, uint32_t height, StructurePassConfig const& config) noexcept;

    // reflections pass
    FrameGraphId<FrameGraphTexture> ssr(FrameGraph& fg,
-            RenderPassBuilder const& passBuilder,
+            RenderPass const& pass,
            FrameHistory const& frameHistory,
            CameraInfo const& cameraInfo,
            PerViewUniforms& uniforms,
@@ -243,13 +241,6 @@ public:
            filament::Viewport const& vp, FrameGraphTexture::Descriptor const& outDesc,
            backend::SamplerMagFilter filter) noexcept;

-    FrameGraphId<FrameGraphTexture> rcas(
-            FrameGraph& fg,
-            float sharpness,
-            FrameGraphId<FrameGraphTexture> input,
-            FrameGraphTexture::Descriptor const& outDesc,
-            bool translucent);
-
    // upscale/downscale blitter using shaders
    FrameGraphId<FrameGraphTexture> blit(FrameGraph& fg, bool translucent,
            FrameGraphId<FrameGraphTexture> input,
@@ -283,11 +274,6 @@ public:
            FrameGraphId<FrameGraphTexture> input,
            FrameGraphId<FrameGraphTexture> depth) noexcept;

-    FrameGraphId<FrameGraphTexture> debugDisplayShadowTexture(FrameGraph& fg,
-            FrameGraphId<FrameGraphTexture> input,
-            FrameGraphId<FrameGraphTexture> shadowmap, float scale,
-            uint8_t layer, uint8_t level, uint8_t channel, float power) noexcept;
-
    backend::Handle<backend::HwTexture> getOneTexture() const;
    backend::Handle<backend::HwTexture> getZeroTexture() const;
    backend::Handle<backend::HwTexture> getOneTextureArray() const;
@@ -376,12 +362,12 @@ private:
    template<size_t SIZE>
    struct JitterSequence {
        auto operator()(size_t i) const noexcept { return positions[i % SIZE] - 0.5f; }
-        const std::array<math::float2, SIZE> positions;
+        const math::float2 positions[SIZE];
    };

    static const JitterSequence<4> sRGSS4;
    static const JitterSequence<4> sUniformHelix4;
-    static const JitterSequence<32> sHaltonSamples;
+    static const JitterSequence<16> sHaltonSamples;

    bool mWorkaroundSplitEasu : 1;
    bool mWorkaroundAllowReadOnlyAncillaryFeedbackLoop : 1;
@@ -19,43 +19,17 @@
 #include "RenderPrimitive.h"
 #include "ShadowMap.h"

-#include "details/Camera.h"
 #include "details/Material.h"
 #include "details/MaterialInstance.h"
 #include "details/View.h"

-#include "components/RenderableManager.h"
-
-#include <private/filament/EngineEnums.h>
 #include <private/filament/UibStructs.h>
-#include <private/filament/Variant.h>

-#include <filament/MaterialEnums.h>
-
-#include <backend/DriverApiForward.h>
-#include <backend/DriverEnums.h>
-#include <backend/Handle.h>
-#include <backend/PipelineState.h>
-
-#include "private/backend/CircularBuffer.h"
-
-#include <utils/compiler.h>
-#include <utils/debug.h>
 #include <utils/JobSystem.h>
-#include <utils/Panic.h>
-#include <utils/Slice.h>
 #include <utils/Systrace.h>
-#include <utils/Range.h>

-#include <algorithm>
-#include <functional>
-#include <limits>
 #include <utility>

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

@@ -63,103 +37,60 @@ namespace filament {

 using namespace backend;

-RenderPassBuilder& RenderPassBuilder::customCommand(
-        FEngine& engine,
-        uint8_t channel,
-        RenderPass::Pass pass,
-        RenderPass::CustomCommand custom,
-        uint32_t order,
-        RenderPass::Executor::CustomCommandFn const& command) {
-    if (!mCustomCommands.has_value()) {
-        // construct the vector the first time
-        mCustomCommands.emplace(engine.getPerRenderPassArena());
-    }
-    mCustomCommands->emplace_back(channel, pass, custom, order, command);
-    return *this;
+RenderPass::RenderPass(FEngine& engine,
+        RenderPass::Arena& arena) noexcept
+        : mCommandArena(arena),
+          mCustomCommands(engine.getPerRenderPassAllocator()) {
 }

-RenderPass RenderPassBuilder::build(FEngine& engine) {
-    ASSERT_POSTCONDITION(mRenderableSoa, "RenderPassBuilder::geometry() hasn't been called");
-    assert_invariant(mScissorViewport.width  <= std::numeric_limits<int32_t>::max());
-    assert_invariant(mScissorViewport.height <= std::numeric_limits<int32_t>::max());
-    return RenderPass{ engine, *this };
-}
-
-// ------------------------------------------------------------------------------------------------
-
-RenderPass::RenderPass(FEngine& engine, RenderPassBuilder const& builder) noexcept
-        : mRenderableSoa(*builder.mRenderableSoa),
-          mVisibleRenderables(builder.mVisibleRenderables),
-          mUboHandle(builder.mUboHandle),
-          mCameraPosition(builder.mCameraPosition),
-          mCameraForwardVector(builder.mCameraForwardVector),
-          mFlags(builder.mFlags),
-          mVariant(builder.mVariant),
-          mVisibilityMask(builder.mVisibilityMask),
-          mScissorViewport(builder.mScissorViewport),
-          mCustomCommands(engine.getPerRenderPassArena()) {
-
-    // compute the number of commands we need
-    updateSummedPrimitiveCounts(
-            const_cast<FScene::RenderableSoa&>(mRenderableSoa), mVisibleRenderables);
-
-    uint32_t commandCount =
-            FScene::getPrimitiveCount(mRenderableSoa, mVisibleRenderables.last);
-    const bool colorPass  = bool(builder.mCommandTypeFlags & CommandTypeFlags::COLOR);
-    const bool depthPass  = bool(builder.mCommandTypeFlags & CommandTypeFlags::DEPTH);
-    commandCount *= uint32_t(colorPass * 2 + depthPass);
-    commandCount += 1; // for the sentinel
-
-    uint32_t const customCommandCount =
-            builder.mCustomCommands.has_value() ? builder.mCustomCommands->size() : 0;
-
-    Command* const curr = builder.mArena.alloc<Command>(commandCount + customCommandCount);
-    assert_invariant(curr);
-
-    if (UTILS_UNLIKELY(builder.mArena.getAllocator().isHeapAllocation(curr))) {
-        static bool sLogOnce = true;
-        if (UTILS_UNLIKELY(sLogOnce)) {
-            sLogOnce = false;
-            PANIC_LOG("RenderPass arena is full, using slower system heap. Please increase "
-                      "the appropriate constant (e.g. FILAMENT_PER_RENDER_PASS_ARENA_SIZE_IN_MB).");
-        }
-    }
-
-    mCommandBegin = curr;
-    mCommandEnd = curr + commandCount + customCommandCount;
-
-    appendCommands(engine, { curr, commandCount }, builder.mCommandTypeFlags);
-
-    if (builder.mCustomCommands.has_value()) {
-        Command* p = curr + commandCount;
-        for (auto [channel, passId, command, order, fn]: builder.mCustomCommands.value()) {
-            appendCustomCommand(p++, channel, passId, command, order, fn);
-        }
-    }
-
-    // sort commands once we're done adding commands
-    sortCommands(builder.mArena);
-
-    if (engine.isAutomaticInstancingEnabled()) {
-        instanceify(engine, builder.mArena);
-    }
-}
+RenderPass::RenderPass(RenderPass const& rhs) = default;

 // this destructor is actually heavy because it inlines ~vector<>
 RenderPass::~RenderPass() noexcept = default;

-void RenderPass::resize(Arena& arena, size_t count) noexcept {
+RenderPass::Command* RenderPass::append(size_t count) noexcept {
+    // this is like an "in-place" realloc(). Works only with LinearAllocator.
+    Command* const curr = mCommandArena.alloc<Command>(count);
+    assert_invariant(curr);
+    assert_invariant(mCommandBegin == nullptr || curr == mCommandEnd);
+    if (mCommandBegin == nullptr) {
+        mCommandBegin = mCommandEnd = curr;
+    }
+    mCommandEnd += count;
+    return curr;
+}
+
+void RenderPass::resize(size_t count) noexcept {
    if (mCommandBegin) {
        mCommandEnd = mCommandBegin + count;
-        arena.rewind(mCommandEnd);
+        mCommandArena.rewind(mCommandEnd);
    }
 }

-void RenderPass::appendCommands(FEngine& engine,
-        Slice<Command> commands, CommandTypeFlags const commandTypeFlags) noexcept {
+void RenderPass::setGeometry(FScene::RenderableSoa const& soa, Range<uint32_t> vr,
+        backend::Handle<backend::HwBufferObject> uboHandle) noexcept {
+    mRenderableSoa = &soa;
+    mVisibleRenderables = vr;
+    mUboHandle = uboHandle;
+}
+
+void RenderPass::setCamera(const CameraInfo& camera) noexcept {
+    mCameraPosition = camera.getPosition();
+    mCameraForwardVector = camera.getForwardVector();
+}
+
+void RenderPass::setScissorViewport(backend::Viewport viewport) noexcept {
+    assert_invariant(viewport.width  <= std::numeric_limits<int32_t>::max());
+    assert_invariant(viewport.height <= std::numeric_limits<int32_t>::max());
+    mScissorViewport = viewport;
+}
+
+void RenderPass::appendCommands(FEngine& engine, CommandTypeFlags const commandTypeFlags) noexcept {
    SYSTRACE_CALL();
    SYSTRACE_CONTEXT();

+    assert_invariant(mRenderableSoa);
+
    utils::Range<uint32_t> const vr = mVisibleRenderables;
    // trace the number of visible renderables
    SYSTRACE_VALUE32("visibleRenderables", vr.size());
@@ -173,10 +104,17 @@ void RenderPass::appendCommands(FEngine& engine,
    const FScene::VisibleMaskType visibilityMask = mVisibilityMask;

    // up-to-date summed primitive counts needed for generateCommands()
-    FScene::RenderableSoa const& soa = mRenderableSoa;
+    FScene::RenderableSoa const& soa = *mRenderableSoa;
+    updateSummedPrimitiveCounts(const_cast<FScene::RenderableSoa&>(soa), vr);

-    Command* curr = commands.data();
-    size_t const commandCount = commands.size();
+    // compute how much maximum storage we need for this pass
+    uint32_t commandCount = FScene::getPrimitiveCount(soa, vr.last);
+    // double the color pass for transparent objects that need to render twice
+    const bool colorPass  = bool(commandTypeFlags & CommandTypeFlags::COLOR);
+    const bool depthPass  = bool(commandTypeFlags & CommandTypeFlags::DEPTH);
+    commandCount *= uint32_t(colorPass * 2 + depthPass);
+    commandCount += 1; // for the sentinel
+    Command* const curr = append(commandCount);

    auto stereoscopicEyeCount =
            renderFlags & IS_STEREOSCOPIC ? engine.getConfig().stereoscopicEyeCount : 1;
@@ -214,8 +152,7 @@ void RenderPass::appendCommands(FEngine& engine,
    }
 }

-void RenderPass::appendCustomCommand(Command* commands,
-        uint8_t channel, Pass pass, CustomCommand custom, uint32_t order,
+void RenderPass::appendCustomCommand(uint8_t channel, Pass pass, CustomCommand custom, uint32_t order,
        Executor::CustomCommandFn command) {

    assert_invariant((uint64_t(order) << CUSTOM_ORDER_SHIFT) <=  CUSTOM_ORDER_MASK);
@@ -231,10 +168,11 @@ void RenderPass::appendCustomCommand(Command* commands,
    cmd |= uint64_t(order) << CUSTOM_ORDER_SHIFT;
    cmd |= uint64_t(index);

-    commands->key = cmd;
+    Command* const curr = append(1);
+    curr->key = cmd;
 }

-void RenderPass::sortCommands(Arena& arena) noexcept {
+void RenderPass::sortCommands(FEngine& engine) noexcept {
    SYSTRACE_NAME("sort and trim commands");

    std::sort(mCommandBegin, mCommandEnd);
@@ -245,20 +183,30 @@ void RenderPass::sortCommands(Arena& arena) noexcept {
                return c.key != uint64_t(Pass::SENTINEL);
            });

-    resize(arena, uint32_t(last - mCommandBegin));
+    resize(uint32_t(last - mCommandBegin));
+
+    if (engine.isAutomaticInstancingEnabled()) {
+        instanceify(engine);
+    }
 }

-void RenderPass::execute(RenderPass const& pass,
-        FEngine& engine, const char* name,
+void RenderPass::execute(FEngine& engine, const char* name,
        backend::Handle<backend::HwRenderTarget> renderTarget,
-        backend::RenderPassParams params) noexcept {
+        backend::RenderPassParams params) const noexcept {
+
    DriverApi& driver = engine.getDriverApi();
+
+    // this is a good time to flush the CommandStream, because we're about to potentially
+    // output a lot of commands. This guarantees here that we have at least
+    // FILAMENT_MIN_COMMAND_BUFFERS_SIZE_IN_MB bytes (1MiB by default).
+    engine.flush();
+
    driver.beginRenderPass(renderTarget, params);
-    pass.getExecutor().execute(engine, name);
+    getExecutor().execute(engine, name);
    driver.endRenderPass();
 }

-void RenderPass::instanceify(FEngine& engine, Arena& arena) noexcept {
+void RenderPass::instanceify(FEngine& engine) noexcept {
    SYSTRACE_NAME("instanceify");

    // instanceify works by scanning the **sorted** command stream, looking for repeat draw
@@ -314,8 +262,7 @@ void RenderPass::instanceify(FEngine& engine, Arena& arena) noexcept {
                // buffer large enough for all instances data
                stagingBufferSize = sizeof(PerRenderableData) * (last - curr);
                stagingBuffer = (PerRenderableData*)::malloc(stagingBufferSize);
-                uboData = mRenderableSoa.data<FScene::UBO>();
-                assert_invariant(uboData);
+                uboData = mRenderableSoa->data<FScene::UBO>();
            }

            // copy the ubo data to a staging buffer
@@ -368,7 +315,7 @@ void RenderPass::instanceify(FEngine& engine, Arena& arena) noexcept {
            return command.key == uint64_t(Pass::SENTINEL);
        });

-        resize(arena, uint32_t(lastCommand - mCommandBegin));
+        resize(uint32_t(lastCommand - mCommandBegin));
    }

    assert_invariant(stagingBuffer == nullptr);
@@ -376,7 +323,7 @@ void RenderPass::instanceify(FEngine& engine, Arena& arena) noexcept {


 /* static */
-UTILS_ALWAYS_INLINE // This function exists only to make the code more readable. we want it inlined.
+UTILS_ALWAYS_INLINE // this function exists only to make the code more readable. we want it inlined.
 inline              // and we don't need it in the compilation unit
 void RenderPass::setupColorCommand(Command& cmdDraw, Variant variant,
        FMaterialInstance const* const UTILS_RESTRICT mi, bool inverseFrontFaces) noexcept {
@@ -427,7 +374,7 @@ void RenderPass::setupColorCommand(Command& cmdDraw, Variant variant,

 /* static */
 UTILS_NOINLINE
-void RenderPass::generateCommands(CommandTypeFlags commandTypeFlags, Command* const commands,
+void RenderPass::generateCommands(uint32_t commandTypeFlags, Command* const commands,
        FScene::RenderableSoa const& soa, Range<uint32_t> range,
        Variant variant, RenderFlags renderFlags,
        FScene::VisibleMaskType visibilityMask, float3 cameraPosition, float3 cameraForward,
@@ -485,9 +432,9 @@ void RenderPass::generateCommands(CommandTypeFlags commandTypeFlags, Command* co
 }

 /* static */
-template<RenderPass::CommandTypeFlags commandTypeFlags>
+template<uint32_t commandTypeFlags>
 UTILS_NOINLINE
-RenderPass::Command* RenderPass::generateCommandsImpl(RenderPass::CommandTypeFlags extraFlags,
+RenderPass::Command* RenderPass::generateCommandsImpl(uint32_t extraFlags,
        Command* UTILS_RESTRICT curr,
        FScene::RenderableSoa const& UTILS_RESTRICT soa, Range<uint32_t> range,
        Variant const variant, RenderFlags renderFlags, FScene::VisibleMaskType visibilityMask,
@@ -790,13 +737,13 @@ void RenderPass::updateSummedPrimitiveCounts(
 // ------------------------------------------------------------------------------------------------

 void RenderPass::Executor::overridePolygonOffset(backend::PolygonOffset const* polygonOffset) noexcept {
-    if ((mPolygonOffsetOverride = (polygonOffset != nullptr))) { // NOLINT(*-assignment-in-if-condition)
+    if ((mPolygonOffsetOverride = (polygonOffset != nullptr))) {
        mPolygonOffset = *polygonOffset;
    }
 }

 void RenderPass::Executor::overrideScissor(backend::Viewport const* scissor) noexcept {
-    if ((mScissorOverride = (scissor != nullptr))) { // NOLINT(*-assignment-in-if-condition)
+    if ((mScissorOverride = (scissor != nullptr))) {
        mScissor = *scissor;
    }
 }
@@ -807,20 +754,15 @@ void RenderPass::Executor::overrideScissor(backend::Viewport const& scissor) noe
 }

 void RenderPass::Executor::execute(FEngine& engine, const char*) const noexcept {
-    execute(engine, mCommands.begin(), mCommands.end());
+    execute(engine.getDriverApi(), mCommands.begin(), mCommands.end());
 }

 UTILS_NOINLINE // no need to be inlined
-void RenderPass::Executor::execute(FEngine& engine,
+void RenderPass::Executor::execute(backend::DriverApi& driver,
        const Command* first, const Command* last) const noexcept {
-
    SYSTRACE_CALL();
    SYSTRACE_CONTEXT();

-    DriverApi& driver = engine.getDriverApi();
-    size_t const capacity = engine.getMinCommandBufferSize();
-    CircularBuffer const& circularBuffer = driver.getCircularBuffer();
-
    if (first != last) {
        SYSTRACE_VALUE32("commandCount", last - first);

@@ -839,163 +781,126 @@ void RenderPass::Executor::execute(FEngine& engine,
        FMaterial const* UTILS_RESTRICT ma = nullptr;
        auto const* UTILS_RESTRICT pCustomCommands = mCustomCommands.data();

-        // Maximum space occupied in the CircularBuffer by a single `Command`. This must be
-        // reevaluated when the inner loop below adds DriverApi commands or when we change the
-        // CommandStream protocol. Currently, the maximum is 240 bytes, and we use 256 to be on
-        // the safer side.
-        size_t const maxCommandSizeInBytes = 256;
+        first--;
+        while (++first != last) {
+            assert_invariant(first->key != uint64_t(Pass::SENTINEL));

-        // Number of Commands that can be issued and guaranteed to fit in the current
-        // CircularBuffer allocation. In practice, we'll have tons of headroom especially if
-        // skinning and morphing aren't used. With a 2 MiB buffer (the default) a batch is
-        // 8192 commands (i.e. draw calls).
-        size_t const batchCommandCount = capacity / maxCommandSizeInBytes;
-        while(first != last) {
-            Command const* const batchLast = std::min(first + batchCommandCount, last);
+            /*
+             * Be careful when changing code below, this is the hot inner-loop
+             */

-            // actual number of commands we need to write (can be smaller than batchCommandCount)
-            size_t const commandCount = batchLast - first;
-            size_t const commandSizeInBytes = commandCount * maxCommandSizeInBytes;
-
-            // check we have enough capacity to write these commandCount commands, if not,
-            // request a new CircularBuffer allocation of `capacity` bytes.
-            if (UTILS_UNLIKELY(circularBuffer.getUsed() > capacity - commandSizeInBytes)) {
-                engine.flush(); // TODO: we should use a "fast" flush if possible
+            if (UTILS_UNLIKELY((first->key & CUSTOM_MASK) != uint64_t(CustomCommand::PASS))) {
+                mi = nullptr; // custom command could change the currently bound MaterialInstance
+                uint32_t const index = (first->key & CUSTOM_INDEX_MASK) >> CUSTOM_INDEX_SHIFT;
+                assert_invariant(index < mCustomCommands.size());
+                pCustomCommands[index]();
+                continue;
            }

-            first--;
-            while (++first != batchLast) {
-                assert_invariant(first->key != uint64_t(Pass::SENTINEL));
+            // primitiveHandle may be invalid if no geometry was set on the renderable.
+            if (UTILS_UNLIKELY(!first->primitive.primitiveHandle)) {
+                continue;
+            }

-                /*
-                 * Be careful when changing code below, this is the hot inner-loop
-                 */
+            // per-renderable uniform
+            const PrimitiveInfo info = first->primitive;
+            pipeline.rasterState = info.rasterState;

-                if (UTILS_UNLIKELY((first->key & CUSTOM_MASK) != uint64_t(CustomCommand::PASS))) {
-                    mi = nullptr; // custom command could change the currently bound MaterialInstance
-                    uint32_t const index = (first->key & CUSTOM_INDEX_MASK) >> CUSTOM_INDEX_SHIFT;
-                    assert_invariant(index < mCustomCommands.size());
-                    pCustomCommands[index]();
-                    continue;
+            if (UTILS_UNLIKELY(mi != info.mi)) {
+                // this is always taken the first time
+                mi = info.mi;
+                ma = mi->getMaterial();
+
+                auto const& scissor = mi->getScissor();
+                if (UTILS_UNLIKELY(mi->hasScissor())) {
+                    // scissor is set, we need to apply the offset/clip
+                    // clang vectorizes this!
+                    constexpr int32_t maxvali = std::numeric_limits<int32_t>::max();
+                    const backend::Viewport scissorViewport = mScissorViewport;
+                    // compute new left/bottom, assume no overflow
+                    int32_t l = scissor.left + scissorViewport.left;
+                    int32_t b = scissor.bottom + scissorViewport.bottom;
+                    // compute right/top without overflowing, scissor.width/height guaranteed
+                    // to convert to int32
+                    int32_t r = (l > maxvali - int32_t(scissor.width)) ?
+                            maxvali : l + int32_t(scissor.width);
+                    int32_t t = (b > maxvali - int32_t(scissor.height)) ?
+                            maxvali : b + int32_t(scissor.height);
+                    // clip to the viewport
+                    l = std::max(l, scissorViewport.left);
+                    b = std::max(b, scissorViewport.bottom);
+                    r = std::min(r, scissorViewport.left + int32_t(scissorViewport.width));
+                    t = std::min(t, scissorViewport.bottom + int32_t(scissorViewport.height));
+                    assert_invariant(r >= l && t >= b);
+                    *pScissor = { l, b, uint32_t(r - l), uint32_t(t - b) };
+                } else {
+                    // no scissor set (common case), 'scissor' has its default value, use that.
+                    *pScissor = scissor;
                }

-                // primitiveHandle may be invalid if no geometry was set on the renderable.
-                if (UTILS_UNLIKELY(!first->primitive.primitiveHandle)) {
-                    continue;
+                *pPipelinePolygonOffset = mi->getPolygonOffset();
+                pipeline.stencilState = mi->getStencilState();
+                mi->use(driver);
+            }
+
+            pipeline.program = ma->getProgram(info.materialVariant);
+
+            uint16_t const instanceCount = info.instanceCount & PrimitiveInfo::INSTANCE_COUNT_MASK;
+            auto getPerObjectUboHandle =
+                    [this, &info, &instanceCount]() -> std::pair<Handle<backend::HwBufferObject>, uint32_t> {
+                if (info.instanceBufferHandle) {
+                    // "hybrid" instancing -- instanceBufferHandle takes the place of the UBO
+                    return { info.instanceBufferHandle, 0 };
                }
-
-                // per-renderable uniform
-                const PrimitiveInfo info = first->primitive;
-                pipeline.rasterState = info.rasterState;
-
-                if (UTILS_UNLIKELY(mi != info.mi)) {
-                    // this is always taken the first time
-                    mi = info.mi;
-                    assert_invariant(mi);
-
-                    ma = mi->getMaterial();
-
-                    auto const& scissor = mi->getScissor();
-                    if (UTILS_UNLIKELY(mi->hasScissor())) {
-                        // scissor is set, we need to apply the offset/clip
-                        // clang vectorizes this!
-                        constexpr int32_t maxvali = std::numeric_limits<int32_t>::max();
-                        const backend::Viewport scissorViewport = mScissorViewport;
-                        // compute new left/bottom, assume no overflow
-                        int32_t l = scissor.left + scissorViewport.left;
-                        int32_t b = scissor.bottom + scissorViewport.bottom;
-                        // compute right/top without overflowing, scissor.width/height guaranteed
-                        // to convert to int32
-                        int32_t r = (l > maxvali - int32_t(scissor.width)) ?
-                                    maxvali : l + int32_t(scissor.width);
-                        int32_t t = (b > maxvali - int32_t(scissor.height)) ?
-                                    maxvali : b + int32_t(scissor.height);
-                        // clip to the viewport
-                        l = std::max(l, scissorViewport.left);
-                        b = std::max(b, scissorViewport.bottom);
-                        r = std::min(r, scissorViewport.left + int32_t(scissorViewport.width));
-                        t = std::min(t, scissorViewport.bottom + int32_t(scissorViewport.height));
-                        assert_invariant(r >= l && t >= b);
-                        *pScissor = { l, b, uint32_t(r - l), uint32_t(t - b) };
-                    } else {
-                        // no scissor set (common case), 'scissor' has its default value, use that.
-                        *pScissor = scissor;
-                    }
-
-                    *pPipelinePolygonOffset = mi->getPolygonOffset();
-                    pipeline.stencilState = mi->getStencilState();
-                    mi->use(driver);
+                bool const userInstancing =
+                        (info.instanceCount & PrimitiveInfo::USER_INSTANCE_MASK) != 0u;
+                if (!userInstancing && instanceCount > 1) {
+                    // automatic instancing
+                    return { mInstancedUboHandle, info.index * sizeof(PerRenderableData) };
+                } else {
+                    // manual instancing
+                    return { mUboHandle, info.index * sizeof(PerRenderableData) };
                }
+            };

-                assert_invariant(ma);
-                pipeline.program = ma->getProgram(info.materialVariant);
+            // bind per-renderable uniform block. there is no need to attempt to skip this command
+            // because the backends already do this.
+            auto const [perObjectUboHandle, offset] = getPerObjectUboHandle();
+            assert_invariant(perObjectUboHandle);
+            driver.bindBufferRange(BufferObjectBinding::UNIFORM,
+                    +UniformBindingPoints::PER_RENDERABLE,
+                    perObjectUboHandle,
+                    offset,
+                    sizeof(PerRenderableUib));

-                uint16_t const instanceCount =
-                        info.instanceCount & PrimitiveInfo::INSTANCE_COUNT_MASK;
-                auto getPerObjectUboHandle =
-                        [this, &info, &instanceCount]() -> std::pair<Handle<backend::HwBufferObject>, uint32_t> {
-                            if (info.instanceBufferHandle) {
-                                // "hybrid" instancing -- instanceBufferHandle takes the place of the UBO
-                                return { info.instanceBufferHandle, 0 };
-                            }
-                            bool const userInstancing =
-                                    (info.instanceCount & PrimitiveInfo::USER_INSTANCE_MASK) != 0u;
-                            if (!userInstancing && instanceCount > 1) {
-                                // automatic instancing
-                                return {
-                                        mInstancedUboHandle,
-                                        info.index * sizeof(PerRenderableData) };
-                            } else {
-                                // manual instancing
-                                return { mUboHandle, info.index * sizeof(PerRenderableData) };
-                            }
-                        };
-
-                // Bind per-renderable uniform block. There is no need to attempt to skip this command
-                // because the backends already do this.
-                auto const [perObjectUboHandle, offset] = getPerObjectUboHandle();
-                assert_invariant(perObjectUboHandle);
+            if (UTILS_UNLIKELY(info.skinningHandle)) {
+                // note: we can't bind less than sizeof(PerRenderableBoneUib) due to glsl limitations
                driver.bindBufferRange(BufferObjectBinding::UNIFORM,
-                        +UniformBindingPoints::PER_RENDERABLE,
-                        perObjectUboHandle,
-                        offset,
-                        sizeof(PerRenderableUib));
+                        +UniformBindingPoints::PER_RENDERABLE_BONES,
+                        info.skinningHandle,
+                        info.skinningOffset * sizeof(PerRenderableBoneUib::BoneData),
+                        sizeof(PerRenderableBoneUib));
+                // note: always bind the skinningTexture because the shader needs it.
+                driver.bindSamplers(+SamplerBindingPoints::PER_RENDERABLE_SKINNING,
+                        info.skinningTexture);
+                // note: even if only skinning is enabled, binding morphTargetBuffer is needed.
+                driver.bindSamplers(+SamplerBindingPoints::PER_RENDERABLE_MORPHING,
+                        info.morphTargetBuffer);
+           }

-                if (UTILS_UNLIKELY(info.skinningHandle)) {
-                    // note: we can't bind less than sizeof(PerRenderableBoneUib) due to glsl limitations
-                    driver.bindBufferRange(BufferObjectBinding::UNIFORM,
-                            +UniformBindingPoints::PER_RENDERABLE_BONES,
-                            info.skinningHandle,
-                            info.skinningOffset * sizeof(PerRenderableBoneUib::BoneData),
-                            sizeof(PerRenderableBoneUib));
-                    // note: always bind the skinningTexture because the shader needs it.
-                    driver.bindSamplers(+SamplerBindingPoints::PER_RENDERABLE_SKINNING,
-                            info.skinningTexture);
-                    // note: even if only skinning is enabled, binding morphTargetBuffer is needed.
-                    driver.bindSamplers(+SamplerBindingPoints::PER_RENDERABLE_MORPHING,
-                            info.morphTargetBuffer);
-                }
-
-                if (UTILS_UNLIKELY(info.morphWeightBuffer)) {
-                    // Instead of using a UBO per primitive, we could also have a single UBO for all
-                    // primitives and use bindUniformBufferRange which might be more efficient.
-                    driver.bindUniformBuffer(+UniformBindingPoints::PER_RENDERABLE_MORPHING,
-                            info.morphWeightBuffer);
-                    driver.bindSamplers(+SamplerBindingPoints::PER_RENDERABLE_MORPHING,
-                            info.morphTargetBuffer);
-                    // note: even if only morphing is enabled, binding skinningTexture is needed.
-                    driver.bindSamplers(+SamplerBindingPoints::PER_RENDERABLE_SKINNING,
-                            info.skinningTexture);
-                }
-
-                driver.draw(pipeline, info.primitiveHandle, instanceCount);
+            if (UTILS_UNLIKELY(info.morphWeightBuffer)) {
+                // Instead of using a UBO per primitive, we could also have a single UBO for all
+                // primitives and use bindUniformBufferRange which might be more efficient.
+                driver.bindUniformBuffer(+UniformBindingPoints::PER_RENDERABLE_MORPHING,
+                        info.morphWeightBuffer);
+                driver.bindSamplers(+SamplerBindingPoints::PER_RENDERABLE_MORPHING,
+                        info.morphTargetBuffer);
+                // note: even if only morphing is enabled, binding skinningTexture is needed.
+                driver.bindSamplers(+SamplerBindingPoints::PER_RENDERABLE_SKINNING,
+                        info.skinningTexture);
            }
-        }

-        // If the remaining space is less than half the capacity, we flush right away to
-        // allow some headroom for commands that might come later.
-        if (UTILS_UNLIKELY(circularBuffer.getUsed() > capacity / 2)) {
-            engine.flush();
+            driver.draw(pipeline, info.primitiveHandle, instanceCount);
        }
    }

--- a/Show More
+++ b/Show More