matc: initialize CustomVariable

The default values were not provided via default construction.
Caused matc to crash on Linux.

.github/workflows/android-continuous.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: macos-14
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - uses: actions/setup-java@v3
         with:
           distribution: 'temurin'

.github/workflows/ios-continuous.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: macos-14
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - name: Run build script
         run: |
           cd build/ios && printf "y" | ./build.sh continuous

.github/workflows/linux-continuous.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-22.04
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - name: Run build script
         run: |
           cd build/linux && printf "y" | ./build.sh continuous

.github/workflows/mac-continuous.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: macos-14
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - name: Run build script
         run: |
           cd build/mac && printf "y" | ./build.sh continuous

.github/workflows/npm-deploy.yml

@@ -13,7 +13,7 @@ jobs:
     name: npm-deploy
     runs-on: macos-14
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
         with:
           ref: ${{ github.event.inputs.release_tag }}
       # Setup .npmrc file to publish to npm

.github/workflows/presubmit.yml

@@ -18,7 +18,7 @@ jobs:
         os: [macos-14, ubuntu-22.04]
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - name: Run build script
         run: |
           WORKFLOW_OS=`echo \`uname\` | sed "s/Darwin/mac/" | tr [:upper:] [:lower:]`
@@ -32,7 +32,7 @@ jobs:
     runs-on: windows-2019
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - name: Run build script
         run: |
           build\windows\build-github.bat presubmit
@@ -43,7 +43,7 @@ jobs:
     runs-on: macos-14
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - uses: actions/setup-java@v3
         with:
           distribution: 'temurin'
@@ -57,7 +57,7 @@ jobs:
     runs-on: macos-14
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - name: Run build script
         run: |
           cd build/ios && printf "y" | ./build.sh presubmit
@@ -70,7 +70,7 @@ jobs:
     runs-on: macos-14
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - name: Run build script
         run: |
           cd build/web && printf "y" | ./build.sh presubmit

.github/workflows/release.yml

@@ -41,7 +41,7 @@ jobs:
           TAG=${REF##*/}
           echo "ref=${REF}" >> $GITHUB_OUTPUT
           echo "tag=${TAG}" >> $GITHUB_OUTPUT
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
         with:
           ref: ${{ steps.git_ref.outputs.ref }}
       - name: Run build script
@@ -76,7 +76,7 @@ jobs:
           TAG=${REF##*/}
           echo "ref=${REF}" >> $GITHUB_OUTPUT
           echo "tag=${TAG}" >> $GITHUB_OUTPUT
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
         with:
           ref: ${{ steps.git_ref.outputs.ref }}
       - name: Run build script
@@ -109,7 +109,7 @@ jobs:
           TAG=${REF##*/}
           echo "ref=${REF}" >> $GITHUB_OUTPUT
           echo "tag=${TAG}" >> $GITHUB_OUTPUT
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
         with:
           ref: ${{ steps.git_ref.outputs.ref }}
       - uses: actions/setup-java@v3
@@ -163,7 +163,7 @@ jobs:
           TAG=${REF##*/}
           echo "ref=${REF}" >> $GITHUB_OUTPUT
           echo "tag=${TAG}" >> $GITHUB_OUTPUT
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
         with:
           ref: ${{ steps.git_ref.outputs.ref }}
       - name: Run build script
@@ -197,7 +197,7 @@ jobs:
           echo "ref=${REF}" >> $GITHUB_OUTPUT
           echo "tag=${TAG}" >> $GITHUB_OUTPUT
         shell: bash
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
         with:
           ref: ${{ steps.git_ref.outputs.ref }}
       - name: Run build script
@@ -205,7 +205,7 @@ jobs:
           TAG: ${{ steps.git_ref.outputs.tag }}
         run: |
           build\windows\build-github.bat release
-          move out\filament-windows.tgz out\filament-$Env:TAG-windows.tgz
+          move out\filament-windows.tgz out\filament-%TAG%-windows.tgz
         shell: cmd
       - uses: actions/github-script@v6
         env:

.github/workflows/web-continuous.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: macos-14
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - name: Run build script
         run: |
           cd build/web && printf "y" | ./build.sh continuous

.github/workflows/windows-continuous.yml

@@ -13,7 +13,7 @@ jobs:
     runs-on: windows-2019
 
     steps:
-      - uses: actions/checkout@v3.3.0
+      - uses: actions/checkout@v4.1.6
       - name: Run build script
         run: |
           build\windows\build-github.bat continuous

BUILDING.md

@@ -73,7 +73,6 @@ The following CMake options are boolean options specific to Filament:
 - `FILAMENT_SUPPORTS_VULKAN`:      Include the Vulkan backend
 - `FILAMENT_INSTALL_BACKEND_TEST`: Install the backend test library so it can be consumed on iOS
 - `FILAMENT_USE_EXTERNAL_GLES3`:   Experimental: Compile Filament against OpenGL ES 3
-- `FILAMENT_USE_SWIFTSHADER`:      Compile Filament against SwiftShader
 - `FILAMENT_SKIP_SAMPLES`:         Don't build sample apps
 
 To turn an option on or off:
@@ -426,7 +425,7 @@ value is the desired roughness between 0 and 1.
 
 ## Generating C++ documentation
 
-To generate the documentation you must first install `doxygen` and `graphviz`, then run the 
+To generate the documentation you must first install `doxygen` and `graphviz`, then run the
 following commands:
 
 ```shell
@@ -436,32 +435,62 @@ doxygen docs/doxygen/filament.doxygen
 
 Finally simply open `docs/html/index.html` in your web browser.
 
-## SwiftShader
+## Software Rasterization
 
-To try out Filament's Vulkan support with SwiftShader, first build SwiftShader and set the
-`SWIFTSHADER_LD_LIBRARY_PATH` variable to the folder that contains `libvk_swiftshader.dylib`:
+We have tested swiftshader and Mesa for software rasterization on the Vulkan/GL backends.
+
+To use this for Vulkan, please first make sure that the [Vulkan SDK](https://www.lunarg.com/vulkan-sdk/) is
+installed on your machine. If you are doing a manual installation of the SDK on Linux, you will have
+to source `setup-env.sh` in the SDK's root folder to make sure the Vulkan loader is the first lib loaded.
+
+### Swiftshader (Vulkan) [tested on macOS and Linux]
+
+First, build SwiftShader
 
 ```shell
 git clone https://github.com/google/swiftshader.git
 cd swiftshader/build
 cmake .. &&  make -j
-export SWIFTSHADER_LD_LIBRARY_PATH=`pwd`
 ```
 
-Next, go to your Filament repo and use the [easy build](#easy-build) script with `-t`.
+and then set `VK_ICD_FILENAMES` to the ICD json produced in the build. For example,
+```shell
+export VK_ICD_FILENAMES=/Users/user/swiftshader/build/Darwin/vk_swiftshader_icd.json
+```
 
-## SwiftShader for CI
+Build and run Filament as usual and specify the Vulkan backend when creating the Engine.
 
-Continuous testing turnaround can be quite slow if you need to build SwiftShader from scratch, so we
-provide an Ubuntu-based Docker image that has it already built. The Docker image also includes
-everything necessary for building Filament. You can fetch and run the image as follows:
+### Mesa's LLVMPipe (GL) and Lavapipe (Vulkan) [tested on Linux]
+
+We will only cover steps that build Mesa from source. The official documentation of Mesa mentioned
+that in general precompiled libraries [are **not** made available](https://docs.mesa3d.org/precompiled.html).
+
+Download the repo and make sure you have the build depedencies. For example (assuming an Ubuntu/Debian distro),
+```shell
+git clone https://gitlab.freedesktop.org/mesa/mesa.git
+sudo apt-get build-dep mesa
+```
+
+To build both the GL and Vulkan rasterizers,
 
 ```shell
-docker pull ghcr.io/filament-assets/swiftshader
-docker run -it ghcr.io/filament-assets/swiftshader
+cd mesa
+mkdir -p out
+meson setup builddir/ -Dprefix=$(pwd)/out -Dglx=xlib -Dgallium-drivers=swrast -Dvulkan-drivers=swrast
+meson install -C builddir/
 ```
 
-To do more with the container, see the helper script at `build/swiftshader/test.sh`.
+For GL, we need to ensure that we load the GL lib from the mesa output directory.  For example, to run
+the debug `gltf_viewer`, we would execute
+```shell
+LD_LIBRARY_PATH=/Users/user/mesa/out/lib/x86_64-linux-gnu \
+    ./out/cmake-debug/samples/gltf_viewer -a opengl
+```
 
-If you are a team member, you can update the public image to the latest SwiftShader by
-following the instructions at the top of `build/swiftshader/Dockerfile`.
+For Vulkan, we need to set the path to the ICD json, which tells the loader where to find the driver
+library. To run `gltf_viewer`, we would execute
+```shell
+VK_ICD_FILENAMES=/Users/user/mesa/out/share/vulkan/icd.d/lvp_icd.x86_64.json \
+    ./out/cmake-debug/samples/gltf_viewer -a vulkan
+
+```

CMakeLists.txt

@@ -21,8 +21,6 @@ project(TNT)
 # ==================================================================================================
 option(FILAMENT_USE_EXTERNAL_GLES3 "Experimental: Compile Filament against OpenGL ES 3" OFF)
 
-option(FILAMENT_USE_SWIFTSHADER "Compile Filament against SwiftShader" OFF)
-
 option(FILAMENT_ENABLE_LTO "Enable link-time optimizations if supported by the compiler" OFF)
 
 option(FILAMENT_SKIP_SAMPLES "Don't build samples" OFF)
@@ -45,6 +43,8 @@ option(FILAMENT_ENABLE_TSAN "Enable Thread Sanitizer" OFF)
 
 option(FILAMENT_ENABLE_FEATURE_LEVEL_0 "Enable Feature Level 0" ON)
 
+option(FILAMENT_ENABLE_MULTIVIEW "Enable multiview for Filament" OFF)
+
 set(FILAMENT_NDK_VERSION "" CACHE STRING
     "Android NDK version or version prefix to be used when building for Android."
 )
@@ -69,6 +69,10 @@ set(FILAMENT_METAL_HANDLE_ARENA_SIZE_IN_MB "8" CACHE STRING
     "Size of the Metal handle arena, default 8."
 )
 
+set(FILAMENT_BACKEND_DEBUG_FLAG "" CACHE STRING
+    "A debug flag meant for enabling/disabling backend debugging paths"
+)
+
 # Enable exceptions by default in spirv-cross.
 set(SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS OFF)
 
@@ -139,11 +143,6 @@ if (LINUX)
             add_definitions(-DFILAMENT_SUPPORTS_XCB)
         endif()
 
-        # Default Swiftshader build does not enable the xlib extension
-        if (FILAMENT_SUPPORTS_XLIB AND FILAMENT_USE_SWIFTSHADER)
-            set(FILAMENT_SUPPORTS_XLIB OFF)
-        endif()
-
         if (FILAMENT_SUPPORTS_XLIB)
             add_definitions(-DFILAMENT_SUPPORTS_XLIB)
         endif()
@@ -167,15 +166,6 @@ if (NOT ANDROID AND NOT WEBGL AND NOT IOS AND NOT FILAMENT_LINUX_IS_MOBILE)
     set(IS_HOST_PLATFORM TRUE)
 endif()
 
-if (IOS)
-    # Remove the headerpad_max_install_names linker flag on iOS. It causes warnings when linking
-    # executables with bitcode.
-    string(REPLACE "-Wl,-headerpad_max_install_names" "" CMAKE_C_LINK_FLAGS ${CMAKE_C_LINK_FLAGS})
-    string(REPLACE "-Wl,-headerpad_max_install_names" "" CMAKE_CXX_LINK_FLAGS ${CMAKE_CXX_LINK_FLAGS})
-    string(REPLACE "-Wl,-headerpad_max_install_names" "" CMAKE_SHARED_LIBRARY_CREATE_C_FLAGS ${CMAKE_SHARED_LIBRARY_CREATE_C_FLAGS})
-    string(REPLACE "-Wl,-headerpad_max_install_names" "" CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS ${CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS})
-endif()
-
 if (WIN32)
     # Link statically against c/c++ lib to avoid missing redistriburable such as
     # "VCRUNTIME140.dll not found. Try reinstalling the app.", but give users
@@ -330,10 +320,6 @@ if (FILAMENT_SUPPORTS_EGL_ON_LINUX)
   set(EGL TRUE)
 endif()
 
-if (FILAMENT_USE_SWIFTSHADER)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DFILAMENT_USE_SWIFTSHADER")
-endif()
-
 if (WIN32)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D_USE_MATH_DEFINES=1")
 endif()
@@ -387,8 +373,9 @@ endif()
 if (NOT MSVC AND NOT IOS)
     # Omitting stack frame pointers prevents the generation of readable stack traces in crash reports on iOS
     set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fomit-frame-pointer")
+endif()
 
-    # These aren't compatible with -fembed-bitcode (and seem to have no effect on Apple platforms anyway)
+if (NOT MSVC)
     set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -ffunction-sections -fdata-sections")
 endif()
 
@@ -453,8 +440,13 @@ endif()
 if (NOT WEBGL)
     set(GC_SECTIONS "-Wl,--gc-sections")
 endif()
+
 set(B_SYMBOLIC_FUNCTIONS "-Wl,-Bsymbolic-functions")
 
+if (ANDROID)
+set(BINARY_ALIGNMENT "-Wl,-z,max-page-size=16384")
+endif()
+
 if (APPLE)
     set(GC_SECTIONS "-Wl,-dead_strip")
     set(B_SYMBOLIC_FUNCTIONS "")
@@ -468,7 +460,7 @@ if (APPLE)
 endif()
 
 set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${GC_SECTIONS}")
-set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} ${GC_SECTIONS} ${B_SYMBOLIC_FUNCTIONS}")
+set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} ${GC_SECTIONS} ${B_SYMBOLIC_FUNCTIONS} ${BINARY_ALIGNMENT}")
 
 if (WEBGL_PTHREADS)
     set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -pthread")
@@ -531,6 +523,29 @@ else()
     option(FILAMENT_DISABLE_MATOPT "Disable material optimizations" ON)
 endif()
 
+# This only affects the prebuilt shader files in gltfio and samples, not filament library.
+# The value can be either "instanced", "multiview", or "none"
+set(FILAMENT_SAMPLES_STEREO_TYPE "none" CACHE STRING
+    "Stereoscopic type that shader files in gltfio and samples are built for."
+)
+string(TOLOWER "${FILAMENT_SAMPLES_STEREO_TYPE}" FILAMENT_SAMPLES_STEREO_TYPE)
+if (NOT FILAMENT_SAMPLES_STEREO_TYPE STREQUAL "instanced"
+    AND NOT FILAMENT_SAMPLES_STEREO_TYPE STREQUAL "multiview"
+    AND NOT FILAMENT_SAMPLES_STEREO_TYPE STREQUAL "none")
+    message(FATAL_ERROR "Invalid stereo type: \"${FILAMENT_SAMPLES_STEREO_TYPE}\" choose either \"instanced\",  \"multiview\", or \"none\" ")
+endif ()
+
+# Compiling samples for multiview implies enabling multiview feature as well.
+if (FILAMENT_SAMPLES_STEREO_TYPE STREQUAL "multiview")
+    set(FILAMENT_ENABLE_MULTIVIEW ON)
+endif ()
+
+# Define backend flag for debug only
+if (CMAKE_BUILD_TYPE STREQUAL "Debug" AND NOT FILAMENT_BACKEND_DEBUG_FLAG STREQUAL "")
+    add_definitions(-DFILAMENT_BACKEND_DEBUG_FLAG=${FILAMENT_BACKEND_DEBUG_FLAG})
+    unset(FILAMENT_BACKEND_DEBUG_FLAG)
+endif()
+
 # ==================================================================================================
 # Material compilation flags
 # ==================================================================================================
@@ -659,21 +674,6 @@ else()
     set(IMPORT_EXECUTABLES ${FILAMENT}/${IMPORT_EXECUTABLES_DIR}/ImportExecutables-${CMAKE_BUILD_TYPE}.cmake)
 endif()
 
-# ==================================================================================================
-# Try to find Vulkan if the SDK is installed, otherwise fall back to the bundled version.
-# This needs to stay in our top-level CMakeLists because it sets up variables that are used by the
-# "bluevk" and "samples" targets.
-# ==================================================================================================
-
-if (FILAMENT_USE_SWIFTSHADER)
-    if (NOT FILAMENT_SUPPORTS_VULKAN)
-        message(ERROR "SwiftShader is only useful when Vulkan is enabled.")
-    endif()
-    find_library(SWIFTSHADER_VK NAMES vk_swiftshader HINTS "$ENV{SWIFTSHADER_LD_LIBRARY_PATH}")
-    message(STATUS "Found SwiftShader VK library in: ${SWIFTSHADER_VK}.")
-    add_definitions(-DFILAMENT_VKLIBRARY_PATH=\"${SWIFTSHADER_VK}\")
-endif()
-
 # ==================================================================================================
 # Common Functions
 # ==================================================================================================
@@ -735,7 +735,6 @@ add_subdirectory(${FILAMENT}/filament)
 add_subdirectory(${FILAMENT}/shaders)
 add_subdirectory(${EXTERNAL}/basisu/tnt)
 add_subdirectory(${EXTERNAL}/civetweb/tnt)
-add_subdirectory(${EXTERNAL}/hat-trie/tnt)
 add_subdirectory(${EXTERNAL}/imgui/tnt)
 add_subdirectory(${EXTERNAL}/robin-map/tnt)
 add_subdirectory(${EXTERNAL}/smol-v/tnt)
@@ -802,6 +801,9 @@ if (IS_HOST_PLATFORM)
     add_subdirectory(${TOOLS}/glslminifier)
     add_subdirectory(${TOOLS}/matc)
     add_subdirectory(${TOOLS}/matinfo)
+    if (NOT WIN32)  # matedit not yet supported on Windows
+        add_subdirectory(${TOOLS}/matedit)
+    endif()
     add_subdirectory(${TOOLS}/mipgen)
     add_subdirectory(${TOOLS}/normal-blending)
     add_subdirectory(${TOOLS}/resgen)

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.50.5'
+    implementation 'com.google.android.filament:filament-android:1.54.0'
 }
 ```
 
@@ -51,7 +51,7 @@ Here are all the libraries available in the group `com.google.android.filament`:
 iOS projects can use CocoaPods to install the latest release:
 
 ```shell
-pod 'Filament', '~> 1.50.5'
+pod 'Filament', '~> 1.54.0'
 ```
 
 ### Snapshots
@@ -176,6 +176,7 @@ steps:
   - [x] KHR_materials_unlit
   - [x] KHR_materials_variants
   - [x] KHR_materials_volume
+  - [x] KHR_materials_specular
   - [x] KHR_mesh_quantization
   - [x] KHR_texture_basisu
   - [x] KHR_texture_transform

RELEASE_NOTES.md

@@ -7,6 +7,94 @@ 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).
 
+## v1.54.1
+
+
+## v1.54.0
+
+- materials: add a new `stereoscopicType` material parameter. [⚠️ **New Material Version**]
+- Fix a crash when compiling shaders on IMG devices
+
+## v1.53.5
+
+- engine: Fix bug causing certain sampler parameters to not be applied correctly in GLES 2.0 and on
+  certain GLES 3.0 drivers.
+
+## v1.53.4
+
+
+## v1.53.3
+
+- Add drag and drop support for IBL files for desktop gltf_viewer.
+
+## v1.53.2
+
+
+## v1.53.1
+
+
+## v1.53.0
+
+- engine: fix skinning normals with large transforms (b/342459864) [⚠️ **New Material Version**]
+
+## v1.52.3
+
+
+## v1.52.2
+
+
+## v1.52.1
+
+- Add instructions for using Mesa for software rasterization
+
+## v1.51.9
+
+
+## v1.51.8
+
+- filagui: Fix regression which broke WebGL
+- Add a new Engine::Config setting to control preferred shader language
+- Add `getEyeIndex` vertex API
+- ios: Remove bitcode from iOS builds
+
+## v1.51.7
+
+- Add new matedit tool
+- filagui: Support rendering `GL_TEXTURE_EXTERNAL_OES` textures.
+- `setFrameScheduledCallback` now takes a `utils::Invocable`.
+- engine: Add `isPaused()`
+
+## v1.51.6
+
+- Add new matedit tool
+- filagui: Support rendering `GL_TEXTURE_EXTERNAL_OES` textures.
+
+## v1.51.5
+
+
+## v1.51.4
+
+
+## v1.51.3
+
+
+## v1.51.2
+
+- engine: Add experimental APIs `Engine::builder::paused()` and `Engine::setPaused()`
+
+## v1.51.1
+
+
+## v1.51.0
+
+- materials: add support for post-lighting mix factor (b/328498606) [⚠️ **New Material Version**]
+
+## v1.50.6
+
+- Add new API `SwapChain::getFrameScheduledCallback`
+- vulkan: fixed validation error VUID-vkAcquireNextImageKHR-semaphore-01779
+- opengl: Add support for protected content swapchains and contexts
+
 ## v1.50.5
 
 - android: NDK 26.1.10909125 is used by default

android/build.gradle

@@ -83,12 +83,12 @@ buildscript {
         'minSdk': 21,
         'targetSdk': 34,
         'compileSdk': 34,
-        'kotlin': '1.9.21',
-        'kotlin_coroutines': '1.7.3',
+        'kotlin': '2.0.0',
+        'kotlin_coroutines': '1.9.0-RC',
         'buildTools': '34.0.0',
-        'ndk': '26.1.10909125',
-        'androidx_core': '1.12.0',
-        'androidx_annotations': '1.7.0'
+        'ndk': '27.0.11718014',
+        'androidx_core': '1.13.1',
+        'androidx_annotations': '1.8.0'
     ]
 
     ext.deps = [
@@ -104,7 +104,7 @@ buildscript {
     ]
 
     dependencies {
-        classpath 'com.android.tools.build:gradle:8.2.0'
+        classpath 'com.android.tools.build:gradle:8.4.1'
         classpath "org.jetbrains.kotlin:kotlin-gradle-plugin:${versions.kotlin}"
     }
 

android/filamat-android/CMakeLists.txt

@@ -38,6 +38,7 @@ set(FILAMAT_INCLUDE_DIRS
 include_directories(${FILAMENT_DIR}/include)
 
 set(CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} -Wl,--version-script=${CMAKE_SOURCE_DIR}/libfilamat-jni.map")
+set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -Wl,-z,max-page-size=16384")
 
 add_library(filamat-jni SHARED src/main/cpp/MaterialBuilder.cpp)
 target_include_directories(filamat-jni PRIVATE ${FILAMAT_INCLUDE_DIRS})

android/filament-android/CMakeLists.txt

@@ -59,6 +59,7 @@ endif()
 
 set(VERSION_SCRIPT "${CMAKE_CURRENT_SOURCE_DIR}/libfilament-jni.map")
 set(CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} -Wl,--version-script=${VERSION_SCRIPT}")
+set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -Wl,-z,max-page-size=16384")
 
 add_library(filament-jni SHARED
     src/main/cpp/BufferObject.cpp

android/filament-android/src/main/cpp/Engine.cpp

@@ -342,6 +342,21 @@ Java_com_google_android_filament_Engine_nIsValidMaterial(JNIEnv*, jclass,
     return (jboolean)engine->isValid((Material*)nativeMaterial);
 }
 
+extern "C" JNIEXPORT jboolean JNICALL
+Java_com_google_android_filament_Engine_nIsValidMaterialInstance(JNIEnv*, jclass,
+        jlong nativeEngine, jlong nativeMaterial, jlong nativeMaterialInstance) {
+    Engine* engine = (Engine *)nativeEngine;
+    return (jboolean)engine->isValid((Material*)nativeMaterial,
+            (MaterialInstance*)nativeMaterialInstance);
+}
+
+extern "C" JNIEXPORT jboolean JNICALL
+        Java_com_google_android_filament_Engine_nIsValidExpensiveMaterialInstance(JNIEnv*, jclass,
+        jlong nativeEngine, jlong nativeMaterialInstance) {
+    Engine* engine = (Engine *)nativeEngine;
+    return (jboolean)engine->isValidExpensive((MaterialInstance*)nativeMaterialInstance);
+}
+
 extern "C" JNIEXPORT jboolean JNICALL
 Java_com_google_android_filament_Engine_nIsValidSkybox(JNIEnv*, jclass,
         jlong nativeEngine, jlong nativeSkybox) {
@@ -391,6 +406,27 @@ Java_com_google_android_filament_Engine_nFlush(JNIEnv*, jclass,
     engine->flush();
 }
 
+extern "C" JNIEXPORT jboolean JNICALL
+Java_com_google_android_filament_Engine_nIsPaused(JNIEnv*, jclass,
+        jlong nativeEngine) {
+    Engine* engine = (Engine*) nativeEngine;
+    return (jboolean)engine->isPaused();
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_Engine_nSetPaused(JNIEnv*, jclass,
+        jlong nativeEngine, jboolean paused) {
+    Engine* engine = (Engine*) nativeEngine;
+    engine->setPaused(paused);
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_Engine_nUnprotected(JNIEnv*, jclass,
+        jlong nativeEngine, jboolean paused) {
+    Engine* engine = (Engine*) nativeEngine;
+    engine->unprotected();
+}
+
 // Managers...
 
 extern "C" JNIEXPORT jlong JNICALL
@@ -487,7 +523,10 @@ extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBu
         jlong jobSystemThreadCount,
         jlong textureUseAfterFreePoolSize, jboolean disableParallelShaderCompile,
         jint stereoscopicType, jlong stereoscopicEyeCount,
-        jlong resourceAllocatorCacheSizeMB, jlong resourceAllocatorCacheMaxAge) {
+        jlong resourceAllocatorCacheSizeMB, jlong resourceAllocatorCacheMaxAge,
+        jboolean disableHandleUseAfterFreeCheck,
+        jint preferredShaderLanguage,
+        jboolean forceGLES2Context) {
     Engine::Builder* builder = (Engine::Builder*) nativeBuilder;
     Engine::Config config = {
             .commandBufferSizeMB = (uint32_t) commandBufferSizeMB,
@@ -502,6 +541,9 @@ extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBu
             .stereoscopicEyeCount = (uint8_t) stereoscopicEyeCount,
             .resourceAllocatorCacheSizeMB = (uint32_t) resourceAllocatorCacheSizeMB,
             .resourceAllocatorCacheMaxAge = (uint8_t) resourceAllocatorCacheMaxAge,
+            .disableHandleUseAfterFreeCheck = (bool) disableHandleUseAfterFreeCheck,
+            .preferredShaderLanguage = (Engine::Config::ShaderLanguage) preferredShaderLanguage,
+            .forceGLES2Context = (bool) forceGLES2Context,
     };
     builder->config(&config);
 }
@@ -518,8 +560,20 @@ extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBu
     builder->sharedContext((void*) sharedContext);
 }
 
+extern "C" JNIEXPORT void JNICALL Java_com_google_android_filament_Engine_nSetBuilderPaused(
+        JNIEnv*, jclass, jlong nativeBuilder, jboolean paused) {
+    Engine::Builder* builder = (Engine::Builder*) nativeBuilder;
+    builder->paused((bool) paused);
+}
+
 extern "C" JNIEXPORT jlong JNICALL
 Java_com_google_android_filament_Engine_nBuilderBuild(JNIEnv*, jclass, jlong nativeBuilder) {
     Engine::Builder* builder = (Engine::Builder*) nativeBuilder;
     return (jlong) builder->build();
 }
+
+extern "C"
+JNIEXPORT jlong JNICALL
+Java_com_google_android_filament_Engine_getSteadyClockTimeNano(JNIEnv *env, jclass clazz) {
+    return (jlong)Engine::getSteadyClockTimeNano();
+}

android/filament-android/src/main/cpp/Material.cpp

@@ -25,12 +25,17 @@ using namespace filament;
 
 extern "C" JNIEXPORT jlong JNICALL
 Java_com_google_android_filament_Material_nBuilderBuild(JNIEnv *env, jclass,
-        jlong nativeEngine, jobject buffer_, jint size) {
+        jlong nativeEngine, jobject buffer_, jint size, jint shBandCount) {
     Engine* engine = (Engine*) nativeEngine;
     AutoBuffer buffer(env, buffer_, size);
-    Material* material = Material::Builder()
+    auto builder = Material::Builder();
+    if (shBandCount) {
+        builder.sphericalHarmonicsBandCount(shBandCount);
+    }
+    Material* material = builder
             .package(buffer.getData(), buffer.getSize())
             .build(*engine);
+
     return (jlong) material;
 }
 

android/filament-android/src/main/cpp/RenderableManager.cpp

@@ -104,6 +104,14 @@ Java_com_google_android_filament_RenderableManager_nBuilderGeometry__JIIJJIIII(J
             (size_t) count);
 }
 
+extern "C"
+JNIEXPORT void JNICALL
+Java_com_google_android_filament_RenderableManager_nBuilderGeometryType(JNIEnv*, jclass,
+        jlong nativeBuilder, int type) {
+    RenderableManager::Builder *builder = (RenderableManager::Builder *) nativeBuilder;
+    builder->geometryType((RenderableManager::Builder::GeometryType)type);
+}
+
 extern "C"
 JNIEXPORT void JNICALL
 Java_com_google_android_filament_RenderableManager_nBuilderMaterial(JNIEnv*, jclass,
@@ -237,12 +245,18 @@ Java_com_google_android_filament_RenderableManager_nBuilderMorphing(JNIEnv*, jcl
 }
 
 extern "C" JNIEXPORT void JNICALL
-Java_com_google_android_filament_RenderableManager_nBuilderSetMorphTargetBufferAt(JNIEnv*, jclass,
-        jlong nativeBuilder, int level, int primitiveIndex, jlong nativeMorphTargetBuffer,
-        int offset, int count) {
+Java_com_google_android_filament_RenderableManager_nBuilderMorphingStandard(JNIEnv*, jclass,
+        jlong nativeBuilder, jlong nativeMorphTargetBuffer) {
     RenderableManager::Builder *builder = (RenderableManager::Builder *) nativeBuilder;
     MorphTargetBuffer *morphTargetBuffer = (MorphTargetBuffer *) nativeMorphTargetBuffer;
-    builder->morphing(level, primitiveIndex, morphTargetBuffer, offset, count);
+    builder->morphing(morphTargetBuffer);
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_RenderableManager_nBuilderSetMorphTargetBufferOffsetAt(JNIEnv*, jclass,
+        jlong nativeBuilder, int level, int primitiveIndex, int offset) {
+    RenderableManager::Builder *builder = (RenderableManager::Builder *) nativeBuilder;
+    builder->morphing(level, primitiveIndex, offset);
 }
 
 extern "C" JNIEXPORT void JNICALL
@@ -314,13 +328,12 @@ Java_com_google_android_filament_RenderableManager_nSetMorphWeights(JNIEnv* env,
 }
 
 extern "C" JNIEXPORT void JNICALL
-Java_com_google_android_filament_RenderableManager_nSetMorphTargetBufferAt(JNIEnv*,
+Java_com_google_android_filament_RenderableManager_nSetMorphTargetBufferOffsetAt(JNIEnv*,
         jclass, jlong nativeRenderableManager, jint i, int level, jint primitiveIndex,
-        jlong nativeMorphTargetBuffer, jint offset, jint count) {
+        jlong, jint offset) {
     RenderableManager *rm = (RenderableManager *) nativeRenderableManager;
-    MorphTargetBuffer *morphTargetBuffer = (MorphTargetBuffer *) nativeMorphTargetBuffer;
-    rm->setMorphTargetBufferAt((RenderableManager::Instance) i, (uint8_t) level,
-            (size_t) primitiveIndex, morphTargetBuffer, (size_t) offset, (size_t) count);
+    rm->setMorphTargetBufferOffsetAt((RenderableManager::Instance) i, (uint8_t) level,
+            (size_t) primitiveIndex, (size_t) offset);
 }
 
 extern "C" JNIEXPORT jint JNICALL

android/filament-android/src/main/cpp/Renderer.cpp

@@ -28,6 +28,14 @@
 using namespace filament;
 using namespace backend;
 
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_Renderer_nSkipFrame(JNIEnv *, jclass, jlong nativeRenderer,
+        jlong vsyncSteadyClockTimeNano) {
+    Renderer *renderer = (Renderer *) nativeRenderer;
+    renderer->skipFrame(uint64_t(vsyncSteadyClockTimeNano));
+}
+
 extern "C" JNIEXPORT jboolean JNICALL
 Java_com_google_android_filament_Renderer_nBeginFrame(JNIEnv *, jclass, jlong nativeRenderer,
         jlong nativeSwapChain, jlong frameTimeNanos) {
@@ -187,3 +195,10 @@ Java_com_google_android_filament_Renderer_nSetPresentationTime(JNIEnv *, jclass
     Renderer *renderer = (Renderer *) nativeRenderer;
     renderer->setPresentationTime(monotonicClockNanos);
 }
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_Renderer_nSetVsyncTime(JNIEnv *, jclass,
+    jlong nativeRenderer, jlong steadyClockTimeNano) {
+    Renderer *renderer = (Renderer *) nativeRenderer;
+    renderer->setVsyncTime(steadyClockTimeNano);
+}

android/filament-android/src/main/cpp/SwapChain.cpp

@@ -34,7 +34,15 @@ Java_com_google_android_filament_SwapChain_nSetFrameCompletedCallback(JNIEnv* en
 }
 
 extern "C" JNIEXPORT jboolean JNICALL
-Java_com_google_android_filament_SwapChain_nIsSRGBSwapChainSupported(JNIEnv *, jclass, jlong nativeEngine) {
+Java_com_google_android_filament_SwapChain_nIsSRGBSwapChainSupported(
+        JNIEnv *, jclass, jlong nativeEngine) {
     Engine* engine = (Engine*) nativeEngine;
-    return (bool)SwapChain::isSRGBSwapChainSupported(*engine);
+    return (jboolean)SwapChain::isSRGBSwapChainSupported(*engine);
+}
+
+extern "C" JNIEXPORT jboolean JNICALL
+Java_com_google_android_filament_SwapChain_nIsProtectedContentSupported(
+        JNIEnv *, jclass, jlong nativeEngine) {
+    Engine* engine = (Engine*) nativeEngine;
+    return (jboolean)SwapChain::isProtectedContentSupported(*engine);
 }

android/filament-android/src/main/cpp/View.cpp

@@ -49,6 +49,12 @@ Java_com_google_android_filament_View_nSetCamera(JNIEnv*, jclass,
     view->setCamera(camera);
 }
 
+extern "C" JNIEXPORT jboolean JNICALL
+Java_com_google_android_filament_View_nHasCamera(JNIEnv*, jclass, jlong nativeView) {
+    View* view = (View*) nativeView;
+    return (jboolean)view->hasCamera();
+}
+
 extern "C" JNIEXPORT void JNICALL
 Java_com_google_android_filament_View_nSetColorGrading(JNIEnv*, jclass,
         jlong nativeView, jlong nativeColorGrading) {
@@ -525,3 +531,12 @@ Java_com_google_android_filament_View_nGetFogEntity(JNIEnv *env, jclass clazz,
     View *view = (View *) nativeView;
     return (jint)view->getFogEntity().getId();
 }
+
+extern "C"
+JNIEXPORT void JNICALL
+Java_com_google_android_filament_View_nClearFrameHistory(JNIEnv *env, jclass clazz,
+        jlong nativeView, jlong nativeEngine) {
+    View *view = (View *) nativeView;
+    Engine *engine = (Engine *) nativeEngine;
+    view->clearFrameHistory(*engine);
+}

android/filament-android/src/main/java/com/google/android/filament/Engine.java

@@ -159,9 +159,12 @@ public class Engine {
     };
 
     /**
-     * The type of technique for stereoscopic rendering
+     * The type of technique for stereoscopic rendering. (Note that the materials used will need to be
+     * compatible with the chosen technique.)
      */
     public enum StereoscopicType {
+        /** No stereoscopic rendering. */
+        NONE,
         /** Stereoscopic rendering is performed using instanced rendering technique. */
         INSTANCED,
         /** Stereoscopic rendering is performed using the multiview feature from the graphics backend. */
@@ -224,7 +227,10 @@ public class Engine {
                     config.jobSystemThreadCount,
                     config.textureUseAfterFreePoolSize, config.disableParallelShaderCompile,
                     config.stereoscopicType.ordinal(), config.stereoscopicEyeCount,
-                    config.resourceAllocatorCacheSizeMB, config.resourceAllocatorCacheMaxAge);
+                    config.resourceAllocatorCacheSizeMB, config.resourceAllocatorCacheMaxAge,
+                    config.disableHandleUseAfterFreeCheck,
+                    config.preferredShaderLanguage.ordinal(),
+                    config.forceGLES2Context);
             return this;
         }
 
@@ -239,6 +245,20 @@ public class Engine {
             return this;
         }
 
+        /**
+         * Sets the initial paused state of the rendering thread.
+         *
+         * <p>Warning: This is an experimental API. See {@link Engine#setPaused(boolean)} for
+         * caveats.
+         *
+         * @param paused Whether to start the rendering thread paused.
+         * @return A reference to this Builder for chaining calls.
+         */
+        public Builder paused(boolean paused) {
+            nSetBuilderPaused(mNativeBuilder, paused);
+            return this;
+        }
+
         /**
          * Creates an instance of Engine
          *
@@ -388,7 +408,7 @@ public class Engine {
          *
          * @see View#setStereoscopicOptions
          */
-        public StereoscopicType stereoscopicType = StereoscopicType.INSTANCED;
+        public StereoscopicType stereoscopicType = StereoscopicType.NONE;
 
         /**
          * The number of eyes to render when stereoscopic rendering is enabled. Supported values are
@@ -400,16 +420,51 @@ public class Engine {
         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).
+         * @Deprecated This value is no longer used.
          */
         public long resourceAllocatorCacheSizeMB = 64;
 
         /*
-         * This value determines for how many frames are texture entries kept in the cache.
-         * The default value of 30 corresponds to about half a second at 60 fps.
+         * This value determines how many frames texture entries are kept for in the cache. This
+         * is a soft limit, meaning some texture older than this are allowed to stay in the cache.
+         * Typically only one texture is evicted per frame.
+         * The default is 1.
          */
-        public long resourceAllocatorCacheMaxAge = 30;
+        public long resourceAllocatorCacheMaxAge = 1;
+
+        /*
+         * Disable backend handles use-after-free checks.
+         */
+        public boolean disableHandleUseAfterFreeCheck = false;
+
+        /*
+         * Sets a preferred shader language for Filament to use.
+         *
+         * The Metal backend supports two shader languages: MSL (Metal Shading Language) and
+         * METAL_LIBRARY (precompiled .metallib). This option controls which shader language is
+         * used when materials contain both.
+         *
+         * By default, when preferredShaderLanguage is unset, Filament will prefer METAL_LIBRARY
+         * shaders if present within a material, falling back to MSL. Setting
+         * preferredShaderLanguage to ShaderLanguage::MSL will instead instruct Filament to check
+         * for the presence of MSL in a material first, falling back to METAL_LIBRARY if MSL is not
+         * present.
+         *
+         * When using a non-Metal backend, setting this has no effect.
+         */
+        public enum ShaderLanguage {
+            DEFAULT,
+            MSL,
+            METAL_LIBRARY,
+        };
+        public ShaderLanguage preferredShaderLanguage = ShaderLanguage.DEFAULT;
+
+        /*
+         * When the OpenGL ES backend is used, setting this value to true will force a GLES2.0
+         * context if supported by the Platform, or if not, will have the backend pretend
+         * it's a GLES2 context. Ignored on other backends.
+         */
+        public boolean forceGLES2Context = false;
     }
 
     private Engine(long nativeEngine, Config config) {
@@ -656,7 +711,7 @@ public class Engine {
      */
     @NonNull
     public SwapChain createSwapChain(@NonNull Object surface) {
-        return createSwapChain(surface, SwapChain.CONFIG_DEFAULT);
+        return createSwapChain(surface, SwapChainFlags.CONFIG_DEFAULT);
     }
 
     /**
@@ -664,15 +719,15 @@ public class Engine {
      *
      * @param surface on Android, <b>must be</b> an instance of {@link android.view.Surface}
      *
-     * @param flags configuration flags, see {@link SwapChain}
+     * @param flags configuration flags, see {@link SwapChainFlags}
      *
      * @return a newly created {@link SwapChain} object
      *
      * @exception IllegalStateException can be thrown if the SwapChain couldn't be created
      *
-     * @see SwapChain#CONFIG_DEFAULT
-     * @see SwapChain#CONFIG_TRANSPARENT
-     * @see SwapChain#CONFIG_READABLE
+     * @see SwapChainFlags#CONFIG_DEFAULT
+     * @see SwapChainFlags#CONFIG_TRANSPARENT
+     * @see SwapChainFlags#CONFIG_READABLE
      *
      */
     @NonNull
@@ -690,21 +745,22 @@ public class Engine {
      *
      * @param width  width of the rendering buffer
      * @param height height of the rendering buffer
-     * @param flags  configuration flags, see {@link SwapChain}
+     * @param flags  configuration flags, see {@link SwapChainFlags}
      *
      * @return a newly created {@link SwapChain} object
      *
      * @exception IllegalStateException can be thrown if the SwapChain couldn't be created
      *
-     * @see SwapChain#CONFIG_DEFAULT
-     * @see SwapChain#CONFIG_TRANSPARENT
-     * @see SwapChain#CONFIG_READABLE
+     * @see SwapChainFlags#CONFIG_DEFAULT
+     * @see SwapChainFlags#CONFIG_TRANSPARENT
+     * @see SwapChainFlags#CONFIG_READABLE
      *
      */
     @NonNull
     public SwapChain createSwapChain(int width, int height, long flags) {
         if (width >= 0 && height >= 0) {
-            long nativeSwapChain = nCreateSwapChainHeadless(getNativeObject(), width, height, flags);
+            long nativeSwapChain =
+                nCreateSwapChainHeadless(getNativeObject(), width, height, flags);
             if (nativeSwapChain == 0) throw new IllegalStateException("Couldn't create SwapChain");
             return new SwapChain(nativeSwapChain, null);
         }
@@ -716,11 +772,12 @@ public class Engine {
      *
      * @param surface a properly initialized {@link NativeSurface}
      *
-     * @param flags configuration flags, see {@link SwapChain}
+     * @param flags configuration flags, see {@link SwapChainFlags}
      *
      * @return a newly created {@link SwapChain} object
      *
-     * @exception IllegalStateException can be thrown if the {@link SwapChain} couldn't be created
+     * @exception IllegalStateException can be thrown if the {@link SwapChainFlags} couldn't be
+     *            created
      */
     @NonNull
     public SwapChain createSwapChainFromNativeSurface(@NonNull NativeSurface surface, long flags) {
@@ -829,6 +886,24 @@ public class Engine {
         return nIsValidMaterial(getNativeObject(), object.getNativeObject());
     }
 
+    /**
+     * Returns whether the object is valid.
+     * @param object Object to check for validity
+     * @return returns true if the specified object is valid.
+     */
+    public boolean isValidMaterialInstance(@NonNull Material ma, MaterialInstance mi) {
+        return nIsValidMaterialInstance(getNativeObject(), ma.getNativeObject(), mi.getNativeObject());
+    }
+
+    /**
+     * Returns whether the object is valid.
+     * @param object Object to check for validity
+     * @return returns true if the specified object is valid.
+     */
+    public boolean isValidExpensiveMaterialInstance(@NonNull MaterialInstance object) {
+        return nIsValidExpensiveMaterialInstance(getNativeObject(), object.getNativeObject());
+    }
+
     /**
      * Returns whether the object is valid.
      * @param object Object to check for validity
@@ -1189,6 +1264,51 @@ public class Engine {
         nFlush(getNativeObject());
     }
 
+    /**
+     * Get paused state of rendering thread.
+     *
+     * <p>Warning: This is an experimental API.
+     *
+     * @see #setPaused
+     */
+    public boolean isPaused() {
+        return nIsPaused(getNativeObject());
+    }
+
+    /**
+     * Pause or resume the rendering thread.
+     *
+     * <p>Warning: This is an experimental API. In particular, note the following caveats.
+     *
+     * <ul><li>
+     * Buffer callbacks will never be called as long as the rendering thread is paused.
+     * Do not rely on a buffer callback to unpause the thread.
+     * </li><li>
+     * While the rendering thread is paused, rendering commands will continue to be queued until the
+     * buffer limit is reached. When the limit is reached, the program will abort.
+     * </li></ul>
+     */
+    public void setPaused(boolean paused) {
+        nSetPaused(getNativeObject(), paused);
+    }
+
+    /**
+     * Switch the command queue to unprotected mode. Protected mode can be activated via
+     * Renderer::beginFrame() using a protected SwapChain.
+     * @see Renderer
+     * @see SwapChain
+     */
+    public void unprotected() {
+        nUnprotected(getNativeObject());
+    }
+
+    /**
+     * Get the current time. This is a convenience function that simply returns the
+     * time in nanosecond since epoch of std::chrono::steady_clock.
+     * @return current time in nanosecond since epoch of std::chrono::steady_clock.
+     * @see Renderer#beginFrame
+     */
+    public static native long getSteadyClockTimeNano();
 
     @UsedByReflection("TextureHelper.java")
     public long getNativeObject() {
@@ -1255,6 +1375,8 @@ public class Engine {
     private static native boolean nIsValidSkinningBuffer(long nativeEngine, long nativeSkinningBuffer);
     private static native boolean nIsValidIndirectLight(long nativeEngine, long nativeIndirectLight);
     private static native boolean nIsValidMaterial(long nativeEngine, long nativeMaterial);
+    private static native boolean nIsValidMaterialInstance(long nativeEngine, long nativeMaterial, long nativeMaterialInstance);
+    private static native boolean nIsValidExpensiveMaterialInstance(long nativeEngine, long nativeMaterialInstance);
     private static native boolean nIsValidSkybox(long nativeEngine, long nativeSkybox);
     private static native boolean nIsValidColorGrading(long nativeEngine, long nativeColorGrading);
     private static native boolean nIsValidTexture(long nativeEngine, long nativeTexture);
@@ -1263,6 +1385,9 @@ public class Engine {
     private static native void nDestroyEntity(long nativeEngine, int entity);
     private static native void nFlushAndWait(long nativeEngine);
     private static native void nFlush(long nativeEngine);
+    private static native boolean nIsPaused(long nativeEngine);
+    private static native void nSetPaused(long nativeEngine, boolean paused);
+    private static native void nUnprotected(long nativeEngine);
     private static native long nGetTransformManager(long nativeEngine);
     private static native long nGetLightManager(long nativeEngine);
     private static native long nGetRenderableManager(long nativeEngine);
@@ -1283,8 +1408,12 @@ public class Engine {
             long minCommandBufferSizeMB, long perFrameCommandsSizeMB, long jobSystemThreadCount,
             long textureUseAfterFreePoolSize, boolean disableParallelShaderCompile,
             int stereoscopicType, long stereoscopicEyeCount,
-            long resourceAllocatorCacheSizeMB, long resourceAllocatorCacheMaxAge);
+            long resourceAllocatorCacheSizeMB, long resourceAllocatorCacheMaxAge,
+            boolean disableHandleUseAfterFreeCheck,
+            int preferredShaderLanguage,
+            boolean forceGLES2Context);
     private static native void nSetBuilderFeatureLevel(long nativeBuilder, int ordinal);
     private static native void nSetBuilderSharedContext(long nativeBuilder, long sharedContext);
+    private static native void nSetBuilderPaused(long nativeBuilder, boolean paused);
     private static native long nBuilderBuild(long nativeBuilder);
 }

android/filament-android/src/main/java/com/google/android/filament/Material.java

@@ -346,6 +346,7 @@ public class Material {
     public static class Builder {
         private Buffer mBuffer;
         private int mSize;
+        private int mShBandCount = 0;
 
         /**
          * Specifies the material data. The material data is a binary blob produced by
@@ -361,6 +362,22 @@ public class Material {
             return this;
         }
 
+        /**
+         * Sets the quality of the indirect lights computations. This is only taken into account
+         * if this material is lit and in the surface domain. This setting will affect the
+         * IndirectLight computation if one is specified on the Scene and Spherical Harmonics
+         * are used for the irradiance.
+         *
+         * @param shBandCount Number of spherical harmonic bands. Must be 1, 2 or 3 (default).
+         * @return Reference to this Builder for chaining calls.
+         * @see IndirectLight
+         */
+        @NonNull
+        public Builder sphericalHarmonicsBandCount(@IntRange(from = 0) int shBandCount) {
+            mShBandCount = shBandCount;
+            return this;
+        }
+
         /**
          * Creates and returns the Material object.
          *
@@ -372,7 +389,8 @@ public class Material {
          */
         @NonNull
         public Material build(@NonNull Engine engine) {
-            long nativeMaterial = nBuilderBuild(engine.getNativeObject(), mBuffer, mSize);
+            long nativeMaterial = nBuilderBuild(engine.getNativeObject(),
+                mBuffer, mSize, mShBandCount);
             if (nativeMaterial == 0) throw new IllegalStateException("Couldn't create Material");
             return new Material(nativeMaterial);
         }
@@ -1023,7 +1041,7 @@ public class Material {
         mNativeObject = 0;
     }
 
-    private static native long nBuilderBuild(long nativeEngine, @NonNull Buffer buffer, int size);
+    private static native long nBuilderBuild(long nativeEngine, @NonNull Buffer buffer, int size, int shBandCount);
     private static native long nCreateInstance(long nativeMaterial);
     private static native long nCreateInstanceWithName(long nativeMaterial, @NonNull String name);
     private static native long nGetDefaultInstance(long nativeMaterial);

android/filament-android/src/main/java/com/google/android/filament/MorphTargetBuffer.java

@@ -74,7 +74,7 @@ public class MorphTargetBuffer {
          *
          * @exception IllegalStateException if the MorphTargetBuffer could not be created
          *
-         * @see #setMorphTargetBufferAt
+         * @see #setMorphTargetBufferOffsetAt
          */
         @NonNull
         public MorphTargetBuffer build(@NonNull Engine engine) {

android/filament-android/src/main/java/com/google/android/filament/RenderableManager.java

@@ -175,6 +175,32 @@ public class RenderableManager {
             return this;
         }
 
+        /**
+         * Type of geometry for a Renderable
+         */
+        public enum GeometryType {
+            /** dynamic gemoetry has no restriction */
+            DYNAMIC,
+            /** bounds and world space transform are immutable */
+            STATIC_BOUNDS,
+            /** skinning/morphing not allowed and Vertex/IndexBuffer immutables */
+            STATIC
+        }
+
+        /**
+         * Specify whether this renderable has static bounds. In this context his means that
+         * the renderable's bounding box cannot change and that the renderable's transform is
+         * assumed immutable. Changing the renderable's transform via the TransformManager
+         * can lead to corrupted graphics. Note that skinning and morphing are not forbidden.
+         * Disabled by default.
+         * @param enable whether this renderable has static bounds. false by default.
+         */
+        @NonNull
+        public Builder geometryType(GeometryType type) {
+            nBuilderGeometryType(mNativeBuilder, type.ordinal());
+            return this;
+        }
+
         /**
          * Binds a material instance to the specified primitive.
          *
@@ -498,14 +524,7 @@ public class RenderableManager {
         }
 
         /**
-         * Controls if the renderable has vertex morphing targets, zero by default. This is
-         * required to enable GPU morphing.
-         *
-         * <p>Filament supports two morphing modes: standard (default) and legacy.</p>
-         *
-         * <p>For standard morphing, A {@link MorphTargetBuffer} must be created and provided via
-         * {@link RenderableManager#setMorphTargetBufferAt}. Standard morphing supports up to
-         * <code>CONFIG_MAX_MORPH_TARGET_COUNT</code> morph targets.</p>
+         * Controls if the renderable has legacy vertex morphing targets, zero by default.
          *
          * For legacy morphing, the attached {@link VertexBuffer} must provide data in the
          * appropriate {@link VertexBuffer.VertexAttribute} slots (<code>MORPH_POSITION_0</code> etc).
@@ -523,6 +542,22 @@ public class RenderableManager {
             return this;
         }
 
+        /**
+         * Controls if the renderable has vertex morphing targets, zero by default.
+         *
+         * <p>For standard morphing, A {@link MorphTargetBuffer} must be provided.
+         * Standard morphing supports up to
+         * <code>CONFIG_MAX_MORPH_TARGET_COUNT</code> morph targets.</p>
+         *
+         * <p>See also {@link RenderableManager#setMorphWeights}, which can be called on a per-frame basis
+         * to advance the animation.</p>
+         */
+        @NonNull
+        public Builder morphing(@NonNull MorphTargetBuffer morphTargetBuffer) {
+            nBuilderMorphingStandard(mNativeBuilder, morphTargetBuffer.getNativeObject());
+            return this;
+        }
+
         /**
          * Specifies the morph target buffer for a primitive.
          *
@@ -534,31 +569,13 @@ public class RenderableManager {
          *
          * @param level the level of detail (lod), only 0 can be specified
          * @param primitiveIndex zero-based index of the primitive, must be less than the count passed to Builder constructor
-         * @param morphTargetBuffer specifies the morph target buffer
          * @param offset specifies where in the morph target buffer to start reading (expressed as a number of vertices)
-         * @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)
          */
         @NonNull
         public Builder morphing(@IntRange(from = 0) int level,
                                 @IntRange(from = 0) int primitiveIndex,
-                                @NonNull MorphTargetBuffer morphTargetBuffer,
-                                @IntRange(from = 0) int offset,
-                                @IntRange(from = 0) int count) {
-            nBuilderSetMorphTargetBufferAt(mNativeBuilder, level, primitiveIndex,
-                    morphTargetBuffer.getNativeObject(), offset, count);
-            return this;
-        }
-
-        /**
-         * Utility method to specify morph target buffer for a primitive.
-         * For details, see the {@link RenderableManager.Builder#morphing}.
-         */
-        @NonNull
-        public Builder morphing(@IntRange(from = 0) int level,
-                                @IntRange(from = 0) int primitiveIndex,
-                                @NonNull MorphTargetBuffer morphTargetBuffer) {
-            nBuilderSetMorphTargetBufferAt(mNativeBuilder, level, primitiveIndex,
-                    morphTargetBuffer.getNativeObject(), 0, morphTargetBuffer.getVertexCount());
+                                @IntRange(from = 0) int offset) {
+            nBuilderSetMorphTargetBufferOffsetAt(mNativeBuilder, level, primitiveIndex, offset);
             return this;
         }
 
@@ -661,26 +678,11 @@ public class RenderableManager {
      *
      * @see Builder#morphing
      */
-    public void setMorphTargetBufferAt(@EntityInstance int i,
+    public void setMorphTargetBufferOffsetAt(@EntityInstance int i,
                                        @IntRange(from = 0) int level,
                                        @IntRange(from = 0) int primitiveIndex,
-                                       @NonNull MorphTargetBuffer morphTargetBuffer,
-                                       @IntRange(from = 0) int offset,
-                                       @IntRange(from = 0) int count) {
-        nSetMorphTargetBufferAt(mNativeObject, i, level, primitiveIndex,
-                morphTargetBuffer.getNativeObject(), offset, count);
-    }
-
-    /**
-     * Utility method to change morph target buffer for the given primitive.
-     * For details, see the {@link RenderableManager#setMorphTargetBufferAt}.
-     */
-    public void setMorphTargetBufferAt(@EntityInstance int i,
-                                       @IntRange(from = 0) int level,
-                                       @IntRange(from = 0) int primitiveIndex,
-                                       @NonNull MorphTargetBuffer morphTargetBuffer) {
-        nSetMorphTargetBufferAt(mNativeObject, i, level, primitiveIndex,
-                morphTargetBuffer.getNativeObject(), 0, morphTargetBuffer.getVertexCount());
+                                       @IntRange(from = 0) int offset) {
+        nSetMorphTargetBufferOffsetAt(mNativeObject, i, level, primitiveIndex, 0, offset);
     }
 
     /**
@@ -964,6 +966,7 @@ public class RenderableManager {
     private static native void nBuilderGeometry(long nativeBuilder, int index, int value, long nativeVertexBuffer, long nativeIndexBuffer);
     private static native void nBuilderGeometry(long nativeBuilder, int index, int value, long nativeVertexBuffer, long nativeIndexBuffer, int offset, int count);
     private static native void nBuilderGeometry(long nativeBuilder, int index, int value, long nativeVertexBuffer, long nativeIndexBuffer, int offset, int minIndex, int maxIndex, int count);
+    private static native void nBuilderGeometryType(long nativeBuilder, int type);
     private static native void nBuilderMaterial(long nativeBuilder, int index, long nativeMaterialInstance);
     private static native void nBuilderBlendOrder(long nativeBuilder, int index, int blendOrder);
     private static native void nBuilderGlobalBlendOrderEnabled(long nativeBuilder, int index, boolean enabled);
@@ -979,7 +982,8 @@ public class RenderableManager {
     private static native int nBuilderSkinningBones(long nativeBuilder, int boneCount, Buffer bones, int remaining);
     private static native void nBuilderSkinningBuffer(long nativeBuilder, long nativeSkinningBuffer, int boneCount, int offset);
     private static native void nBuilderMorphing(long nativeBuilder, int targetCount);
-    private static native void nBuilderSetMorphTargetBufferAt(long nativeBuilder, int level, int primitiveIndex, long nativeMorphTargetBuffer, int offset, int count);
+    private static native void nBuilderMorphingStandard(long nativeBuilder, long nativeMorphTargetBuffer);
+    private static native void nBuilderSetMorphTargetBufferOffsetAt(long nativeBuilder, int level, int primitiveIndex, int offset);
     private static native void nBuilderEnableSkinningBuffers(long nativeBuilder, boolean enabled);
     private static native void nBuilderFog(long nativeBuilder, boolean enabled);
     private static native void nBuilderLightChannel(long nativeRenderableManager, int channel, boolean enable);
@@ -989,7 +993,7 @@ public class RenderableManager {
     private static native int nSetBonesAsMatrices(long nativeObject, int i, Buffer matrices, int remaining, int boneCount, int offset);
     private static native int nSetBonesAsQuaternions(long nativeObject, int i, Buffer quaternions, int remaining, int boneCount, int offset);
     private static native void nSetMorphWeights(long nativeObject, int instance, float[] weights, int offset);
-    private static native void nSetMorphTargetBufferAt(long nativeObject, int i, int level, int primitiveIndex, long nativeMorphTargetBuffer, int offset, int count);
+    private static native void nSetMorphTargetBufferOffsetAt(long nativeObject, int i, int level, int primitiveIndex, long nativeMorphTargetBuffer, int offset);
     private static native int nGetMorphTargetCount(long nativeObject, int i);
     private static native void nSetAxisAlignedBoundingBox(long nativeRenderableManager, int i, float cx, float cy, float cz, float ex, float ey, float ez);
     private static native void nSetLayerMask(long nativeRenderableManager, int i, int select, int value);

android/filament-android/src/main/java/com/google/android/filament/Renderer.java

@@ -284,6 +284,33 @@ public class Renderer {
         nSetPresentationTime(getNativeObject(), monotonicClockNanos);
     }
 
+    /**
+     * The use of this method is optional. It sets the VSYNC time expressed as the duration in
+     * nanosecond since epoch of std::chrono::steady_clock.
+     * If called, passing 0 to frameTimeNanos in Renderer.BeginFrame will use this
+     * time instead.
+     * @param steadyClockTimeNano duration in nanosecond since epoch of std::chrono::steady_clock
+     * @see Engine#getSteadyClockTimeNano
+     * @see Renderer#beginFrame
+     */
+    public void setVsyncTime(long steadyClockTimeNano) {
+        nSetVsyncTime(getNativeObject(), steadyClockTimeNano);
+    }
+
+    /**
+     * Call skipFrame when momentarily skipping frames, for instance if the content of the
+     * scene doesn't change.
+     *
+     * @param vsyncSteadyClockTimeNano The time in nanoseconds when the frame started being rendered,
+     *                       in the {@link System#nanoTime()} timebase. Divide this value by 1000000 to
+     *                       convert it to the {@link android.os.SystemClock#uptimeMillis()}
+     *                       time base. This typically comes from
+     *                       {@link android.view.Choreographer.FrameCallback}.
+     */
+    public void skipFrame(long vsyncSteadyClockTimeNano) {
+        nSkipFrame(getNativeObject(), vsyncSteadyClockTimeNano);
+    }
+
     /**
      * Sets up a frame for this <code>Renderer</code>.
      * <p><code>beginFrame</code> manages frame pacing, and returns whether or not a frame should be
@@ -702,6 +729,8 @@ public class Renderer {
     }
 
     private static native void nSetPresentationTime(long nativeObject, long monotonicClockNanos);
+    private static native void nSetVsyncTime(long nativeObject, long steadyClockTimeNano);
+    private static native void nSkipFrame(long nativeObject, long vsyncSteadyClockTimeNano);
     private static native boolean nBeginFrame(long nativeRenderer, long nativeSwapChain, long frameTimeNanos);
     private static native void nEndFrame(long nativeRenderer);
     private static native void nRender(long nativeRenderer, long nativeView);

android/filament-android/src/main/java/com/google/android/filament/SwapChain.java

@@ -68,77 +68,30 @@ public class SwapChain {
     private final Object mSurface;
     private long mNativeObject;
 
-    public static final long CONFIG_DEFAULT = 0x0;
-
-    /**
-     * This flag indicates that the <code>SwapChain</code> must be allocated with an
-     * alpha-channel.
-     */
-    public static final long CONFIG_TRANSPARENT = 0x1;
-
-    /**
-     * This flag indicates that the <code>SwapChain</code> may be used as a source surface
-     * for reading back render results.  This config must be set when creating
-     * any <code>SwapChain</code>  that will be used as the source for a blit operation.
-     *
-     * @see Renderer#copyFrame
-     */
-    public static final long CONFIG_READABLE = 0x2;
-
-    /**
-     * Indicates that the native X11 window is an XCB window rather than an XLIB window.
-     * This is ignored on non-Linux platforms and in builds that support only one X11 API.
-     */
-    public static final long CONFIG_ENABLE_XCB = 0x4;
-
-    /**
-     * Indicates that the SwapChain must automatically perform linear to sRGB encoding.
-     *
-     * This flag is ignored if isSRGBSwapChainSupported() is false.
-     *
-     * When using this flag, post-processing should be disabled.
-     *
-     * @see SwapChain#isSRGBSwapChainSupported
-     * @see View#setPostProcessingEnabled
-     */
-    public static final long CONFIG_SRGB_COLORSPACE = 0x10;
-
-    /**
-     * Indicates that this SwapChain should allocate a stencil buffer in addition to a depth buffer.
-     *
-     * This flag is necessary when using View::setStencilBufferEnabled and rendering directly into
-     * the SwapChain (when post-processing is disabled).
-     *
-     * The specific format of the stencil buffer depends on platform support. The following pixel
-     * formats are tried, in order of preference:
-     *
-     * Depth only (without CONFIG_HAS_STENCIL_BUFFER):
-     * - DEPTH32F
-     * - DEPTH24
-     *
-     * Depth + stencil (with CONFIG_HAS_STENCIL_BUFFER):
-     * - DEPTH32F_STENCIL8
-     * - DEPTH24F_STENCIL8
-     *
-     * Note that enabling the stencil buffer may hinder depth precision and should only be used if
-     * necessary.
-     *
-     * @see View#setStencilBufferEnabled
-     * @see View#setPostProcessingEnabled
-     */
-    public static final long CONFIG_HAS_STENCIL_BUFFER = 0x20;
-
     SwapChain(long nativeSwapChain, Object surface) {
         mNativeObject = nativeSwapChain;
         mSurface = surface;
     }
 
     /**
-     * Return whether createSwapChain supports the SWAP_CHAIN_CONFIG_SRGB_COLORSPACE flag.
+     * Return whether createSwapChain supports the CONFIG_PROTECTED_CONTENT flag.
      * The default implementation returns false.
      *
      * @param engine A reference to the filament Engine
-     * @return true if SWAP_CHAIN_CONFIG_SRGB_COLORSPACE is supported, false otherwise.
+     * @return true if CONFIG_PROTECTED_CONTENT is supported, false otherwise.
+     * @see SwapChainFlags#CONFIG_PROTECTED_CONTENT
+     */
+    public static boolean isProtectedContentSupported(@NonNull Engine engine) {
+        return nIsProtectedContentSupported(engine.getNativeObject());
+    }
+
+    /**
+     * Return whether createSwapChain supports the CONFIG_SRGB_COLORSPACE flag.
+     * The default implementation returns false.
+     *
+     * @param engine A reference to the filament Engine
+     * @return true if CONFIG_SRGB_COLORSPACE is supported, false otherwise.
+     * @see SwapChainFlags#CONFIG_SRGB_COLORSPACE
      */
     public static boolean isSRGBSwapChainSupported(@NonNull Engine engine) {
         return nIsSRGBSwapChainSupported(engine.getNativeObject());
@@ -186,4 +139,5 @@ public class SwapChain {
 
     private static native void nSetFrameCompletedCallback(long nativeSwapChain, Object handler, Runnable callback);
     private static native boolean nIsSRGBSwapChainSupported(long nativeEngine);
+    private static native boolean nIsProtectedContentSupported(long nativeEngine);
 }

android/filament-android/src/main/java/com/google/android/filament/SwapChainFlags.java

@@ -0,0 +1,97 @@
+/*
+ * Copyright (C) 2024 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.google.android.filament;
+
+// Note: SwapChainFlags is kept separate from SwapChain so that UiHelper does not need to depend
+// on SwapChain. This allows clients to use UiHelper without requiring all of Filament's Java
+// classes.
+
+/**
+ * Flags that a <code>SwapChain</code> can be created with to control behavior.
+ *
+ * @see Engine#createSwapChain
+ * @see Engine#createSwapChainFromNativeSurface
+ */
+public final class SwapChainFlags {
+
+    public static final long CONFIG_DEFAULT = 0x0;
+
+    /**
+     * This flag indicates that the <code>SwapChain</code> must be allocated with an
+     * alpha-channel.
+     */
+    public static final long CONFIG_TRANSPARENT = 0x1;
+
+    /**
+     * This flag indicates that the <code>SwapChain</code> may be used as a source surface
+     * for reading back render results.  This config must be set when creating
+     * any <code>SwapChain</code>  that will be used as the source for a blit operation.
+     *
+     * @see Renderer#copyFrame
+     */
+    public static final long CONFIG_READABLE = 0x2;
+
+    /**
+     * Indicates that the native X11 window is an XCB window rather than an XLIB window.
+     * This is ignored on non-Linux platforms and in builds that support only one X11 API.
+     */
+    public static final long CONFIG_ENABLE_XCB = 0x4;
+
+    /**
+     * Indicates that the SwapChain must automatically perform linear to sRGB encoding.
+     *
+     * This flag is ignored if isSRGBSwapChainSupported() is false.
+     *
+     * When using this flag, post-processing should be disabled.
+     *
+     * @see SwapChain#isSRGBSwapChainSupported
+     * @see View#setPostProcessingEnabled
+     */
+    public static final long CONFIG_SRGB_COLORSPACE = 0x10;
+
+    /**
+     * Indicates that this SwapChain should allocate a stencil buffer in addition to a depth buffer.
+     *
+     * This flag is necessary when using View::setStencilBufferEnabled and rendering directly into
+     * the SwapChain (when post-processing is disabled).
+     *
+     * The specific format of the stencil buffer depends on platform support. The following pixel
+     * formats are tried, in order of preference:
+     *
+     * Depth only (without CONFIG_HAS_STENCIL_BUFFER):
+     * - DEPTH32F
+     * - DEPTH24
+     *
+     * Depth + stencil (with CONFIG_HAS_STENCIL_BUFFER):
+     * - DEPTH32F_STENCIL8
+     * - DEPTH24F_STENCIL8
+     *
+     * Note that enabling the stencil buffer may hinder depth precision and should only be used if
+     * necessary.
+     *
+     * @see View#setStencilBufferEnabled
+     * @see View#setPostProcessingEnabled
+     */
+    public static final long CONFIG_HAS_STENCIL_BUFFER = 0x20;
+
+    /**
+     * The SwapChain contains protected content. Only supported when isProtectedContentSupported()
+     * is true.
+     */
+    public static final long CONFIG_PROTECTED_CONTENT   = 0x40;
+}
+

android/filament-android/src/main/java/com/google/android/filament/View.java

@@ -241,6 +241,15 @@ public class View {
         nSetCamera(getNativeObject(), camera == null ? 0 : camera.getNativeObject());
     }
 
+    /**
+     * Query whether a camera is set.
+     * @return true if a camera is set, false otherwise
+     * @see #setCamera
+     */
+    public boolean hasCamera() {
+        return nHasCamera(getNativeObject());
+    }
+
     /**
      * Gets this View's associated Camera, or null if none has been assigned.
      *
@@ -1224,6 +1233,18 @@ public class View {
         return nGetFogEntity(getNativeObject());
     }
 
+    /**
+     * When certain temporal features are used (e.g.: TAA or Screen-space reflections), the view
+     * keeps a history of previous frame renders associated with the Renderer the view was last
+     * used with. When switching Renderer, it may be necessary to clear that history by calling
+     * this method. Similarly, if the whole content of the screen change, like when a cut-scene
+     * starts, clearing the history might be needed to avoid artifacts due to the previous frame
+     * being very different.
+     */
+    public void clearFrameHistory(Engine engine) {
+        nClearFrameHistory(getNativeObject(), engine.getNativeObject());
+    }
+
     public long getNativeObject() {
         if (mNativeObject == 0) {
             throw new IllegalStateException("Calling method on destroyed View");
@@ -1238,6 +1259,7 @@ public class View {
     private static native void nSetName(long nativeView, String name);
     private static native void nSetScene(long nativeView, long nativeScene);
     private static native void nSetCamera(long nativeView, long nativeCamera);
+    private static native boolean nHasCamera(long nativeView);
     private static native void nSetViewport(long nativeView, int left, int bottom, int width, int height);
     private static native void nSetVisibleLayers(long nativeView, int select, int value);
     private static native void nSetShadowingEnabled(long nativeView, boolean enabled);
@@ -1284,7 +1306,7 @@ public class View {
     private static native void nSetMaterialGlobal(long nativeView, int index, float x, float y, float z, float w);
     private static native void nGetMaterialGlobal(long nativeView, int index, float[] out);
     private static native int nGetFogEntity(long nativeView);
-
+    private static native void nClearFrameHistory(long nativeView, long nativeEngine);
 
     /**
      * List of available ambient occlusion techniques.

android/filament-android/src/main/java/com/google/android/filament/android/UiHelper.java

@@ -27,7 +27,7 @@ import android.view.SurfaceHolder;
 import android.view.SurfaceView;
 import android.view.TextureView;
 
-import com.google.android.filament.SwapChain;
+import com.google.android.filament.SwapChainFlags;
 
 /**
  * UiHelper is a simple class that can manage either a SurfaceView, TextureView, or a SurfaceHolder
@@ -538,7 +538,7 @@ public class UiHelper {
      * the options set on this UiHelper.
      */
     public long getSwapChainFlags() {
-        return isOpaque() ? SwapChain.CONFIG_DEFAULT : SwapChain.CONFIG_TRANSPARENT;
+        return isOpaque() ? SwapChainFlags.CONFIG_DEFAULT : SwapChainFlags.CONFIG_TRANSPARENT;
     }
 
     /**

android/filament-utils-android/CMakeLists.txt

@@ -31,6 +31,7 @@ set_target_properties(iblprefilter PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libfilament-iblprefilter.a)
 
 set(CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} -Wl,--version-script=${CMAKE_CURRENT_SOURCE_DIR}/libfilament-utils-jni.map")
+set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -Wl,-z,max-page-size=16384")
 
 add_library(filament-utils-jni SHARED
         src/main/cpp/AutomationEngine.cpp

android/filament-utils-android/src/main/cpp/Manipulator.cpp

@@ -125,6 +125,11 @@ extern "C" JNIEXPORT void Java_com_google_android_filament_utils_Manipulator_nBu
     builder->groundPlane(a, b, c, d);
 }
 
+extern "C" JNIEXPORT void Java_com_google_android_filament_utils_Manipulator_nBuilderPanning(JNIEnv*, jclass, jlong nativeBuilder, jboolean enabled) {
+    Builder* builder = (Builder*) nativeBuilder;
+    builder->panning(enabled);
+}
+
 extern "C" JNIEXPORT long Java_com_google_android_filament_utils_Manipulator_nBuilderBuild(JNIEnv*, jclass, jlong nativeBuilder, jint mode) {
     Builder* builder = (Builder*) nativeBuilder;
     return (jlong) builder->build((Mode) mode);

android/filament-utils-android/src/main/java/com/google/android/filament/utils/Manipulator.java

@@ -274,6 +274,17 @@ public class Manipulator {
             return this;
         }
 
+        /**
+         * Sets whether panning is enabled in the manipulator.
+         *
+         * @return this <code>Builder</code> object for chaining calls
+         */
+        @NonNull
+        public Builder panning(Boolean enabled) {
+            nBuilderPanning(mNativeBuilder, enabled);
+            return this;
+        }
+
         /**
          * Creates and returns the <code>Manipulator</code> object.
          *
@@ -483,6 +494,7 @@ public class Manipulator {
     private static native void nBuilderFlightPanSpeed(long nativeBuilder, float x, float y);
     private static native void nBuilderFlightMoveDamping(long nativeBuilder, float damping);
     private static native void nBuilderGroundPlane(long nativeBuilder, float a, float b, float c, float d);
+    private static native void nBuilderPanning(long nativeBuilder, Boolean enabled);
     private static native long nBuilderBuild(long nativeBuilder, int mode);
 
     private static native void nDestroyManipulator(long nativeManip);

android/gltfio-android/CMakeLists.txt

@@ -44,6 +44,7 @@ set_target_properties(uberarchive PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libuberarchive.a)
 
 set(CMAKE_SHARED_LINKER_FLAGS_RELEASE "${CMAKE_SHARED_LINKER_FLAGS_RELEASE} -Wl,--version-script=${CMAKE_CURRENT_SOURCE_DIR}/libgltfio-jni.map")
+set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -Wl,-z,max-page-size=16384")
 
 set(GLTFIO_SRCS
         ${GLTFIO_DIR}/include/gltfio/Animator.h
@@ -81,9 +82,20 @@ set(GLTFIO_SRCS
         ${GLTFIO_DIR}/src/TangentsJob.cpp
         ${GLTFIO_DIR}/src/TangentsJob.h
         ${GLTFIO_DIR}/src/UbershaderProvider.cpp
+        ${GLTFIO_DIR}/src/Utility.cpp
+        ${GLTFIO_DIR}/src/Utility.h
         ${GLTFIO_DIR}/src/Wireframe.cpp
         ${GLTFIO_DIR}/src/Wireframe.h
         ${GLTFIO_DIR}/src/downcast.h
+        ${GLTFIO_DIR}/src/extended/AssetLoaderExtended.cpp
+        ${GLTFIO_DIR}/src/extended/AssetLoaderExtended.h
+        ${GLTFIO_DIR}/src/extended/ResourceLoaderExtended.cpp
+        ${GLTFIO_DIR}/src/extended/ResourceLoaderExtended.h
+        ${GLTFIO_DIR}/src/extended/TangentsJobExtended.cpp
+        ${GLTFIO_DIR}/src/extended/TangentsJobExtended.h
+        ${GLTFIO_DIR}/src/extended/TangentSpaceMeshWrapper.cpp
+        ${GLTFIO_DIR}/src/extended/TangentSpaceMeshWrapper.h
+
 
         src/main/cpp/Animator.cpp
         src/main/cpp/AssetLoader.cpp
@@ -108,7 +120,6 @@ set(GLTFIO_INCLUDE_DIRS
         ../../third_party/cgltf
         ../../third_party/meshoptimizer/src
         ../../third_party/robin-map
-        ../../third_party/hat-trie
         ../../third_party/stb
         ../../libs/utils/include
         ../../libs/ktxreader/include

android/gradle.properties

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.50.5
+VERSION_NAME=1.54.0
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 

android/gradle/wrapper/gradle-wrapper.properties

@@ -1,6 +1,6 @@
 #Wed Nov 17 10:40:18 PST 2021
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-8.2-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-8.6-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

android/samples/sample-hello-triangle/src/main/java/com/google/android/filament/hellotriangle/MainActivity.kt

@@ -321,7 +321,7 @@ class MainActivity : Activity() {
             var flags = uiHelper.swapChainFlags
             if (engine.activeFeatureLevel == Engine.FeatureLevel.FEATURE_LEVEL_0) {
                 if (SwapChain.isSRGBSwapChainSupported(engine)) {
-                    flags = flags or SwapChain.CONFIG_SRGB_COLORSPACE
+                    flags = flags or SwapChainFlags.CONFIG_SRGB_COLORSPACE
                 }
             }
 

build.sh

@@ -44,8 +44,6 @@ function print_help {
     echo "        Exclude Vulkan support from the Android build."
     echo "    -s"
     echo "        Add iOS simulator support to the iOS build."
-    echo "    -t"
-    echo "        Enable SwiftShader support for Vulkan in desktop builds."
     echo "    -e"
     echo "        Enable EGL on Linux support for desktop builds."
     echo "    -l"
@@ -61,6 +59,14 @@ function print_help {
     echo "    -b"
     echo "        Enable Address and Undefined Behavior Sanitizers (asan/ubsan) for debugging."
     echo "        This is only for the desktop build."
+    echo "    -x value"
+    echo "        Define a preprocessor flag FILAMENT_BACKEND_DEBUG_FLAG with [value]. This is useful for"
+    echo "        enabling debug paths in the backend from the build script. For example, make a"
+    echo "        systrace-enabled build without directly changing #defines. Remember to add -f when"
+    echo "        changing this option."
+    echo "    -S type"
+    echo "        Enable stereoscopic rendering where type is one of [instanced|multiview]. This is only"
+    echo "        meant for building the samples."
     echo ""
     echo "Build types:"
     echo "    release"
@@ -160,8 +166,6 @@ INSTALL_COMMAND=
 VULKAN_ANDROID_OPTION="-DFILAMENT_SUPPORTS_VULKAN=ON"
 VULKAN_ANDROID_GRADLE_OPTION=""
 
-SWIFTSHADER_OPTION="-DFILAMENT_USE_SWIFTSHADER=OFF"
-
 EGL_ON_LINUX_OPTION="-DFILAMENT_SUPPORTS_EGL_ON_LINUX=OFF"
 
 MATDBG_OPTION="-DFILAMENT_ENABLE_MATDBG=OFF"
@@ -172,6 +176,10 @@ MATOPT_GRADLE_OPTION=""
 
 ASAN_UBSAN_OPTION=""
 
+BACKEND_DEBUG_FLAG_OPTION=""
+
+STEREOSCOPIC_OPTION=""
+
 IOS_BUILD_SIMULATOR=false
 BUILD_UNIVERSAL_LIBRARIES=false
 
@@ -226,11 +234,12 @@ function build_desktop_target {
             -DIMPORT_EXECUTABLES_DIR=out \
             -DCMAKE_BUILD_TYPE="$1" \
             -DCMAKE_INSTALL_PREFIX="../${lc_target}/filament" \
-            ${SWIFTSHADER_OPTION} \
             ${EGL_ON_LINUX_OPTION} \
             ${MATDBG_OPTION} \
             ${MATOPT_OPTION} \
             ${ASAN_UBSAN_OPTION} \
+            ${BACKEND_DEBUG_FLAG_OPTION} \
+            ${STEREOSCOPIC_OPTION} \
             ${architectures} \
             ../..
         ln -sf "out/cmake-${lc_target}/compile_commands.json" \
@@ -289,6 +298,7 @@ function build_webgl_with_target {
             -DCMAKE_BUILD_TYPE="$1" \
             -DCMAKE_INSTALL_PREFIX="../webgl-${lc_target}/filament" \
             -DWEBGL=1 \
+            ${BACKEND_DEBUG_FLAG_OPTION} \
             ../..
         ln -sf "out/cmake-webgl-${lc_target}/compile_commands.json" \
            ../../compile_commands.json
@@ -363,6 +373,8 @@ function build_android_target {
             ${MATDBG_OPTION} \
             ${MATOPT_OPTION} \
             ${VULKAN_ANDROID_OPTION} \
+            ${BACKEND_DEBUG_FLAG_OPTION} \
+            ${STEREOSCOPIC_OPTION} \
             ../..
         ln -sf "out/cmake-android-${lc_target}-${arch}/compile_commands.json" \
            ../../compile_commands.json
@@ -516,7 +528,7 @@ function build_android {
             if [[ "${BUILD_ANDROID_SAMPLES}" == "true" ]]; then
                 for sample in ${ANDROID_SAMPLES}; do
                     echo "Installing out/${sample}-debug.apk"
-                    cp samples/${sample}/build/outputs/apk/debug/${sample}-debug-unsigned.apk \
+                    cp samples/${sample}/build/outputs/apk/debug/${sample}-debug.apk \
                         ../out/${sample}-debug.apk
                 done
             fi
@@ -597,6 +609,7 @@ function build_ios_target {
             -DCMAKE_TOOLCHAIN_FILE=../../third_party/clang/iOS.cmake \
             ${MATDBG_OPTION} \
             ${MATOPT_OPTION} \
+            ${STEREOSCOPIC_OPTION} \
             ../..
         ln -sf "out/cmake-ios-${lc_target}-${arch}/compile_commands.json" \
            ../../compile_commands.json
@@ -730,6 +743,13 @@ function validate_build_command {
             exit 1
         fi
     fi
+
+    # Make sure FILAMENT_BACKEND_DEBUG_FLAG is only meant for debug builds
+    if [[ "${ISSUE_DEBUG_BUILD}" != "true" ]] && [[ ! -z "${BACKEND_DEBUG_FLAG_OPTION}" ]]; then
+        echo "Error: cannot specify FILAMENT_BACKEND_DEBUG_FLAG in non-debug build"
+        exit 1
+    fi
+
     set -e
 }
 
@@ -776,7 +796,7 @@ function check_debug_release_build {
 
 pushd "$(dirname "$0")" > /dev/null
 
-while getopts ":hacCfgijmp:q:uvslwtedk:b" opt; do
+while getopts ":hacCfgijmp:q:uvslwedk:bx:S:" opt; do
     case ${opt} in
         h)
             print_help
@@ -840,7 +860,7 @@ while getopts ":hacCfgijmp:q:uvslwtedk:b" opt; do
                         echo "Platform must be one of [desktop|android|ios|webgl|all]"
                         echo ""
                         exit 1
-                    ;;    
+                    ;;
                 esac
             done
             ;;
@@ -895,10 +915,6 @@ while getopts ":hacCfgijmp:q:uvslwtedk:b" opt; do
             IOS_BUILD_SIMULATOR=true
             echo "iOS simulator support enabled."
             ;;
-        t)
-            SWIFTSHADER_OPTION="-DFILAMENT_USE_SWIFTSHADER=ON"
-            echo "SwiftShader support enabled."
-            ;;
         e)
             EGL_ON_LINUX_OPTION="-DFILAMENT_SUPPORTS_EGL_ON_LINUX=ON -DFILAMENT_SKIP_SDL2=ON -DFILAMENT_SKIP_SAMPLES=ON"
             echo "EGL on Linux support enabled; skipping SDL2."
@@ -918,6 +934,22 @@ while getopts ":hacCfgijmp:q:uvslwtedk:b" opt; do
         b)  ASAN_UBSAN_OPTION="-DFILAMENT_ENABLE_ASAN_UBSAN=ON"
             echo "Enabled ASAN/UBSAN"
             ;;
+        x)  BACKEND_DEBUG_FLAG_OPTION="-DFILAMENT_BACKEND_DEBUG_FLAG=${OPTARG}"
+            ;;
+        S)  case $(echo "${OPTARG}" | tr '[:upper:]' '[:lower:]') in
+                instanced)
+                    STEREOSCOPIC_OPTION="-DFILAMENT_SAMPLES_STEREO_TYPE=instanced"
+                    ;;
+                multiview)
+                    STEREOSCOPIC_OPTION="-DFILAMENT_SAMPLES_STEREO_TYPE=multiview"
+                    ;;
+                *)
+                    echo "Unknown stereoscopic type ${OPTARG}"
+                    echo "Type must be one of [instanced|multiview]"
+                    echo ""
+                    exit 1
+            esac
+            ;;
         \?)
             echo "Invalid option: -${OPTARG}" >&2
             echo ""

build/android/build.sh

@@ -57,7 +57,8 @@ FILAMENT_NDK_VERSION=${FILAMENT_NDK_VERSION:-$(cat `dirname $0`/ndk.version)}
 
 # Install the required NDK version specifically (if not present)
 if [[ ! -d "${ANDROID_HOME}/ndk/$FILAMENT_NDK_VERSION" ]]; then
-    ${ANDROID_HOME}/cmdline-tools/latest/bin/sdkmanager "ndk;$FILAMENT_NDK_VERSION" > /dev/null
+    yes | ${ANDROID_HOME}/cmdline-tools/latest/bin/sdkmanager --licenses
+    ${ANDROID_HOME}/cmdline-tools/latest/bin/sdkmanager "ndk;$FILAMENT_NDK_VERSION"
 fi
 
 # Only build 1 64 bit target during presubmit to cut down build times during presubmit

build/android/ndk.version

@@ -1 +1 @@
-26.1.10909125
+27.0.11718014

build/swiftshader/Dockerfile

@@ -1,53 +0,0 @@
-# Build the image:
-#   docker build --no-cache --tag ssfilament -f build/swiftshader/Dockerfile .
-#   docker tag ssfilament ghcr.io/filament-assets/swiftshader
-#
-# Publish the image:
-#   docker login ghcr.io --username <user> --password <token>
-#   docker push ghcr.io/filament-assets/swiftshader
-#
-# Run the image and mount the current directory:
-#   docker run -it -v `pwd`:/trees/filament -t ssfilament
-
-FROM ubuntu:focal
-WORKDIR /trees
-ARG DEBIAN_FRONTEND=noninteractive
-ENV SWIFTSHADER_LD_LIBRARY_PATH=/trees/swiftshader/build
-ENV CXXFLAGS='-fno-builtin -Wno-pass-failed'
-
-RUN apt-get update && \
-    apt-get --no-install-recommends install -y \
-	apt-transport-https \
-	apt-utils \
-	build-essential \
-	cmake \
-	ca-certificates \
-	git \
-	ninja-build \
-	python \
-	python3 \
-	xorg-dev \
-	clang-7 \
-	libc++-7-dev \
-	libc++abi-7-dev \
-	lldb
-
-# Ensure that clang is used instead of gcc.
-RUN set -eux ;\
-    update-alternatives --install /usr/bin/clang clang /usr/bin/clang-7 100 ;\
-    update-alternatives --install /usr/bin/clang++ clang++ /usr/bin/clang++-7 100 ;\
-    update-alternatives --install /usr/bin/cc cc /usr/bin/clang 100 ;\
-    update-alternatives --install /usr/bin/c++ c++ /usr/bin/clang++ 100
-
-# Get patch files from the local Filament tree.
-COPY build/swiftshader/*.diff .
-
-# Clone SwiftShader, apply patches, and build it.
-RUN set -eux ;\
-	git clone https://swiftshader.googlesource.com/SwiftShader swiftshader ;\
-    cd swiftshader ;\
-	git checkout 139f5c3 ;\
-	git apply /trees/*.diff ;\
-    cd build ;\
-    cmake .. -GNinja -DCMAKE_BUILD_TYPE=Release ;\
-    ninja

build/swiftshader/gallery.py

@@ -1,56 +0,0 @@
-#!/usr/bin/env python3
-
-from pathlib import Path
-import os
-
-spath = os.path.dirname(os.path.realpath(__file__))
-
-path = Path(spath)
-
-folder = "../../results/"
-
-images = list(path.glob(folder + '*.png'))
-
-images.sort()
-
-gallery = open(path.absolute().joinpath(folder + 'index.html'), 'w')
-
-gallery.write("""<html>
-<head>
-<script type="module" src="https://unpkg.com/img-comparison-slider@latest/dist/component/component.esm.js"></script>
-<script nomodule="" src="https://unpkg.com/img-comparison-slider@latest/dist/component/component.js"></script>
-<link rel="stylesheet" href="https://unpkg.com/img-comparison-slider@latest/dist/collection/styles/initial.css"/>
-<style>
-    h2 {
-        font-weight: normal;
-        margin-top: 150px;
-        margin-bottom: 20px;
-    }
-    a {
-        text-decoration: none;
-        font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
-        color: blue;
-    }
-    a:hover {
-        font-weight: bold;
-    }
-</style>
-</head>
-<body>
-""")
-
-tag = ''
-
-for image in images:
-    group = image.stem.rstrip('0123456789')
-    before = f'https://filament-assets.github.io/golden/{group}/{image.name}'
-    after = image.name
-    gallery.write('\n')
-    gallery.write(f'<h2><a href="{image.stem}.json">{image.stem}.json</a></h2>\n')
-    gallery.write('<img-comparison-slider>\n')
-    gallery.write(f'<img slot="before" src="{before}" /> <img slot="after" src="{after}" />\n')
-    gallery.write('</img-comparison-slider>\n')
-
-gallery.write("""</body>
-</html>
-""")

build/swiftshader/patch_00.diff

@@ -1,62 +0,0 @@
-diff --git a/src/Vulkan/VkPipeline.cpp b/src/Vulkan/VkPipeline.cpp
-index 86913ec72..3b35345af 100644
---- a/src/Vulkan/VkPipeline.cpp
-+++ b/src/Vulkan/VkPipeline.cpp
-@@ -71,7 +71,56 @@ std::vector<uint32_t> preprocessSpirv(
- 	if(optimize)
- 	{
- 		// Full optimization list taken from spirv-opt.
--		opt.RegisterPerformancePasses();
-+
-+		// We have removed CreateRedundancyEliminationPass because it segfaults when encountering:
-+		//     %389 = OpCompositeConstruct %7 %386 %387 %388 %86
-+		// When inserting an entry into instruction_to_value_ (which is an unordered_map)
-+		// This could perhaps be investigated further with help from asan.
-+
-+		using namespace spvtools;
-+		opt.RegisterPass(CreateWrapOpKillPass())
-+			.RegisterPass(CreateDeadBranchElimPass())
-+			.RegisterPass(CreateMergeReturnPass())
-+			.RegisterPass(CreateInlineExhaustivePass())
-+			.RegisterPass(CreateAggressiveDCEPass())
-+			.RegisterPass(CreatePrivateToLocalPass())
-+			.RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
-+			.RegisterPass(CreateLocalSingleStoreElimPass())
-+			.RegisterPass(CreateAggressiveDCEPass())
-+			.RegisterPass(CreateScalarReplacementPass())
-+			.RegisterPass(CreateLocalAccessChainConvertPass())
-+			.RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
-+			.RegisterPass(CreateLocalSingleStoreElimPass())
-+			.RegisterPass(CreateAggressiveDCEPass())
-+			.RegisterPass(CreateLocalMultiStoreElimPass())
-+			.RegisterPass(CreateAggressiveDCEPass())
-+			.RegisterPass(CreateCCPPass())
-+			.RegisterPass(CreateAggressiveDCEPass())
-+			.RegisterPass(CreateLoopUnrollPass(true))
-+			.RegisterPass(CreateDeadBranchElimPass())
-+			.RegisterPass(CreateRedundancyEliminationPass()) // workaround for SEGFAULT
-+			.RegisterPass(CreateCombineAccessChainsPass())
-+			.RegisterPass(CreateSimplificationPass())
-+			.RegisterPass(CreateScalarReplacementPass())
-+			.RegisterPass(CreateLocalAccessChainConvertPass())
-+			.RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
-+			.RegisterPass(CreateLocalSingleStoreElimPass())
-+			.RegisterPass(CreateAggressiveDCEPass())
-+			.RegisterPass(CreateSSARewritePass())
-+			.RegisterPass(CreateAggressiveDCEPass())
-+			.RegisterPass(CreateVectorDCEPass())
-+			.RegisterPass(CreateDeadInsertElimPass())
-+			.RegisterPass(CreateDeadBranchElimPass())
-+			.RegisterPass(CreateSimplificationPass())
-+			.RegisterPass(CreateIfConversionPass())
-+			.RegisterPass(CreateCopyPropagateArraysPass())
-+			.RegisterPass(CreateReduceLoadSizePass())
-+			.RegisterPass(CreateAggressiveDCEPass())
-+			.RegisterPass(CreateBlockMergePass())
-+			.RegisterPass(CreateRedundancyEliminationPass()) // workaround for SEGFAULT
-+			.RegisterPass(CreateDeadBranchElimPass())
-+			.RegisterPass(CreateBlockMergePass())
-+			.RegisterPass(CreateSimplificationPass());
- 	}
- 
- 	std::vector<uint32_t> optimized;

build/swiftshader/test.sh

@@ -1,127 +0,0 @@
-#!/bin/bash
-set -e
-
-function print_help {
-    local self_name=$(basename "$0")
-    echo "This script issues docker commands for testing Filament with SwiftShader."
-    echo "The usual sequence of commands is: fetch, start, build filament release, and run."
-    echo ""
-    echo "Usage:"
-    echo "    $self_name [command]"
-    echo ""
-    echo "Commands:"
-    echo "    build filament [debug | release]"
-    echo "        Use the container to build Filament."
-    echo "    build swiftshader [debug | release]"
-    echo "        Use the container to do a clean rebuild of SwiftShader."
-    echo "        (Note that the container already has SwiftShader built.)"
-    echo "    fetch"
-    echo "        Download the docker image from the central repository."
-    echo "    help"
-    echo "        Print this help message."
-    echo "    logs"
-    echo "        Print messages from the container's kernel ring buffer."
-    echo "        This is useful for diagnosing OOM issues."
-    echo "    run [lldb]"
-    echo "        Launch a test inside the container, optionally via lldb."
-    echo "    shell"
-    echo "        Interact with a bash prompt in the container."
-    echo "    start"
-    echo "        Start a container from the image."
-    echo "    stop"
-    echo "        Stop the container."
-    echo ""
-}
-
-# Change the current working directory to the Filament root.
-pushd "$(dirname "$0")/../.." > /dev/null
-
-if [[ "$1" == "build" ]] && [[ "$2" == "filament" ]]; then
-    docker exec runner filament/build.sh -t $3 gltf_viewer
-    exit $?
-fi
-
-if [[ "$1" == "build" ]] && [[ "$2" == "swiftshader" ]]; then
-    BUILD_TYPE="$3"
-    BUILD_TYPE="$(tr '[:lower:]' '[:upper:]' <<< ${BUILD_TYPE:0:1})${BUILD_TYPE:1}"
-    docker exec --workdir /trees/swiftshader runner rm -rf build
-    docker exec --workdir /trees/swiftshader runner mkdir build
-    docker exec --workdir /trees/swiftshader/build runner cmake -GNinja -DCMAKE_BUILD_TYPE="$BUILD_TYPE" ..
-    docker exec --workdir /trees/swiftshader/build runner ninja
-    exit $?
-fi
-
-if [[ "$1" == "fetch" ]]; then
-    docker pull ghcr.io/filament-assets/swiftshader:latest
-    docker tag ghcr.io/filament-assets/swiftshader:latest ssfilament
-    exit $?
-fi
-
-if [[ "$1" == "help" ]]; then
-    print_help
-    exit 0
-fi
-
-if [[ "$1" == "logs" ]]; then
-    docker exec runner dmesg --human --read-clear
-    exit $?
-fi
-
-if [[ "$1" == "run" ]] && [[ "$2" == "lldb" ]]; then
-    docker exec -i --workdir /trees/filament/results runner \
-          lldb --batch -o run -o bt -- \
-          ../out/cmake-release/samples/gltf_viewer \
-          --headless \
-          --batch ../libs/viewer/tests/basic.json \
-          --api vulkan
-    docker exec runner /trees/filament/build/swiftshader/gallery.py
-    exit $?
-fi
-
-if [[ "$1" == "run" ]]; then
-    docker exec --tty  --workdir /trees/filament/results runner \
-          /usr/bin/catchsegv \
-          ../out/cmake-release/samples/gltf_viewer \
-          --headless \
-          --batch ../libs/viewer/tests/basic.json \
-          --api vulkan
-    docker exec runner /trees/filament/build/swiftshader/gallery.py
-    exit $?
-fi
-
-if [[ "$1" == "shell" ]]; then
-    docker exec --interactive --tty runner /bin/bash
-    exit $?
-fi
-
-# Notes on options being passed to docker's run command:
-#
-# - The memory constraint seems to prevent an OOM signal in GitHub Actions.
-# - The cap / security args allow use of lldb and creation of core dumps.
-# - The privileged arg allows use of dmesg for examining OOM logs.
-#
-# Currently, a GitHub Actions VM has 2 CPUs, 7 GB RAM, and 14 GB of SSD disk space.
-#
-# Please be aware that Docker Desktop might impose additional resource constraints, and that those
-# settings can only be controlled with its GUI. We recommend at least 7 GB of memory and 2 GB swap.
-if [[ "$1" == "start" ]]; then
-    mkdir -p results
-    docker run --tty --rm --detach --privileged \
-          --memory 6.5g \
-          --name runner \
-          --cap-add=SYS_PTRACE \
-          --security-opt seccomp=unconfined \
-          --security-opt apparmor=unconfined \
-          --volume `pwd`:/trees/filament \
-          --workdir /trees \
-          ssfilament
-    exit $?
-fi
-
-if [[ "$1" == "stop" ]]; then
-    docker container rm runner --force
-    exit $?
-fi
-
-print_help
-exit 1

docs/Filament.html

@@ -476,7 +476,7 @@ Energy conservation is one of the key components of a good BRDF for physically b
 <p>
 
 
-For the specular term, \(f_m\) is a mirror BRDF that can be modeled with the Fresnel law, noted \(F\) in the Cook-Torrance approximation of the microfacet model integration:
+For the specular term, \(f_r\) is a mirror BRDF that can be modeled with the Fresnel law, noted \(F\) in the Cook-Torrance approximation of the microfacet model integration:
 
 </p><p>
 
@@ -857,7 +857,7 @@ The full GLSL implementation of the standard model is shown in <a href="#listing
 <span class="line">    <span class="hljs-comment">// perceptually linear roughness to roughness (see parameterization)</span></span>
 <span class="line">    <span class="hljs-type">float</span> roughness = perceptualRoughness * perceptualRoughness;</span>
 <span class="line"></span>
-<span class="line">    <span class="hljs-type">float</span> D = D_GGX(NoH, a);</span>
+<span class="line">    <span class="hljs-type">float</span> D = D_GGX(NoH, roughness);</span>
 <span class="line">    <span class="hljs-type">vec3</span>  F = F_Schlick(LoH, f0);</span>
 <span class="line">    <span class="hljs-type">float</span> V = V_SmithGGXCorrelated(NoV, NoL, roughness);</span>
 <span class="line"></span>

docs/Materials.md.html

@@ -1308,7 +1308,12 @@ Description
       declare a variable called `eyeDirection` you can access it in the fragment shader using
       `variable_eyeDirection`. In the vertex shader, the interpolant name is simply a member of
       the `MaterialVertexInputs` structure (`material.eyeDirection` in your example). Each
-      interpolant is of type `float4` (`vec4`) in the shaders.
+      interpolant is of type `float4` (`vec4`) in the shaders. By default the precision of the
+      interpolant is `highp` in *both* the vertex and fragment shaders.
+      An alternate syntax can be used to specify both the name and precision of the interpolant.
+      In this case the specified precision is used as-is in both fragment and vertex stages, in
+      particular if `default` is specified the default precision is used is the fragment shader
+      (`mediump`) and in the vertex shader (`highp`).
 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ JSON
 material {
@@ -1320,7 +1325,11 @@ material {
         }
     ],
     variables : [
-         eyeDirection
+         eyeDirection,
+         {
+            name : eyeColor,
+            precision : medium
+         }
     ],
     vertexDomain : device,
     depthWrite : false,
@@ -1397,7 +1406,7 @@ Type
 :    `string`
 
 Value
-:     Any of `opaque`, `transparent`, `fade`, `add`, `masked`, `multiply`, `screen`. Defaults to `opaque`.
+:     Any of `opaque`, `transparent`, `fade`, `add`, `masked`, `multiply`, `screen`, `custom`. Defaults to `opaque`.
 
 Description
 :     Defines how/if the rendered object is blended with the content of the render target.
@@ -1420,6 +1429,7 @@ Description
       of the material's output defines whether a fragment is discarded or not. Additionally,
       ALPHA_TO_COVERAGE is enabled for non-translucent views. See the maskThreshold section for more
       information.
+    - **Custom**: blending is enabled. But the blending function is user specified. See `blendFunction`.
 
 !!! Note
     When `blending` is set to `masked`, alpha to coverage is automatically enabled for the material.
@@ -1432,6 +1442,36 @@ material {
 }
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
+### Blending and transparency: blendFunction
+
+Type
+:    `object`
+
+Fields
+:     `srcRGB`, `srcA`, `dstRGB`, `dstA`
+
+Description
+:       - *srcRGB*: source function applied to the RGB channels
+        - *srcA*: source function applied to the alpha channel
+        - *srcRGB*: destination function applied to the RGB channels
+        - *srcRGB*: destination function applied to the alpha channel
+        The values possible for each functions are one of `zero`, `one`, `srcColor`, `oneMinusSrcColor`,
+               `dstColor`, `oneMinusDstColor`, `srcAlpha`, `oneMinusSrcAlpha`, `dstAlpha`,
+               `oneMinusDstAlpha`, `srcAlphaSaturate`
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ JSON
+material {
+    blending : custom,
+    blendFunction :
+    {
+        srcRGB: one,
+        srcA: one,
+        dstRGB: oneMinusSrcColor,
+        dstA: oneMinusSrcAlpha
+    }
+ }
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
 ### Blending and transparency: postLightingBlending
 
 Type
@@ -2394,6 +2434,7 @@ The following APIs are only available from the vertex block:
 **getWorldFromModelMatrix()**        | float4x4 |  Matrix that converts from model (object) space to world space
 **getWorldFromModelNormalMatrix()**  | float3x3 |  Matrix that converts normals from model (object) space to world space
 **getVertexIndex()**                 | int      |  Index of the current vertex
+**getEyeIndex()**                    | int      |  Index of the eye being rendered, starting at 0
 
 ### Fragment only
 

docs/viewer/index.html

@@ -43,7 +43,7 @@ filament-viewer::part(canvas) {
 </p>
 
 </main>
-<script src="https://unpkg.com/filament@1.44.0/filament.js"></script>
+<script src="https://unpkg.com/filament@1.51.6/filament.js"></script>
 <script src="https://unpkg.com/gltumble"></script>
 <script src="filament-viewer.js" type="module"></script>
 </body>

filament/CMakeLists.txt

@@ -163,6 +163,7 @@ set(PRIVATE_HDRS
         src/ResourceList.h
         src/ShadowMap.h
         src/ShadowMapManager.h
+        src/SharedHandle.h
         src/TypedUniformBuffer.h
         src/UniformBuffer.h
         src/components/CameraManager.h
@@ -214,6 +215,7 @@ set(PRIVATE_HDRS
 set(MATERIAL_SRCS
         src/materials/antiAliasing/fxaa.mat
         src/materials/antiAliasing/taa.mat
+        src/materials/blitDepth.mat
         src/materials/blitLow.mat
         src/materials/blitArray.mat
         src/materials/bloom/bloomDownsample.mat
@@ -252,8 +254,13 @@ set(MATERIAL_SRCS
 )
 
 set(MATERIAL_FL0_SRCS
-        src/materials/defaultMaterial0.mat
-        src/materials/skybox0.mat
+        src/materials/defaultMaterial.mat
+        src/materials/skybox.mat
+)
+
+set(MATERIAL_MULTIVIEW_SRCS
+        src/materials/defaultMaterial.mat
+        src/materials/skybox.mat
 )
 
 # Embed the binary resource blob for materials.
@@ -285,6 +292,11 @@ if (FILAMENT_ENABLE_FEATURE_LEVEL_0)
     add_definitions(-DFILAMENT_ENABLE_FEATURE_LEVEL_0)
 endif()
 
+# Whether to include MULTIVIEW materials.
+if (FILAMENT_ENABLE_MULTIVIEW)
+    add_definitions(-DFILAMENT_ENABLE_MULTIVIEW)
+endif()
+
 # ==================================================================================================
 # Definitions
 # ==================================================================================================
@@ -315,33 +327,42 @@ foreach (mat_src ${MATERIAL_SRCS})
     get_filename_component(localname "${mat_src}" NAME_WE)
     get_filename_component(fullname "${mat_src}" ABSOLUTE)
     set(output_path "${MATERIAL_DIR}/${localname}.filamat")
-
     add_custom_command(
             OUTPUT ${output_path}
             COMMAND matc ${MATC_BASE_FLAGS} -o ${output_path} ${fullname}
             MAIN_DEPENDENCY ${fullname}
             DEPENDS matc
-            COMMENT "Compiling material ${mat_src} to ${output_path}"
+            COMMENT "Compiling material ${fullname} to ${output_path}"
     )
     list(APPEND MATERIAL_BINS ${output_path})
-endforeach()
-
-if (FILAMENT_ENABLE_FEATURE_LEVEL_0)
-    foreach (mat_src ${MATERIAL_FL0_SRCS})
-        get_filename_component(localname "${mat_src}" NAME_WE)
-        get_filename_component(fullname "${mat_src}" ABSOLUTE)
-        set(output_path "${MATERIAL_DIR}/${localname}.filamat")
 
+    list(FIND MATERIAL_FL0_SRCS ${mat_src} index)
+    if (${index} GREATER -1 AND FILAMENT_ENABLE_FEATURE_LEVEL_0)
+        string(REGEX REPLACE "[.]filamat$" "_fl0.filamat" output_path_fl0 ${output_path})
         add_custom_command(
-                OUTPUT ${output_path}
-                COMMAND matc ${MATC_BASE_FLAGS} -o ${output_path} ${fullname}
+                OUTPUT ${output_path_fl0}
+                COMMAND matc ${MATC_BASE_FLAGS} -PfeatureLevel=0 -o ${output_path_fl0} ${fullname}
                 MAIN_DEPENDENCY ${fullname}
                 DEPENDS matc
-                COMMENT "Compiling material ${mat_src} to ${output_path}"
+                COMMENT "Compiling material ${fullname} to ${output_path_fl0}"
         )
-        list(APPEND MATERIAL_BINS ${output_path})
-    endforeach ()
-endif ()
+        list(APPEND MATERIAL_BINS ${output_path_fl0})
+    endif ()
+
+    list(FIND MATERIAL_MULTIVIEW_SRCS ${mat_src} index)
+    if (${index} GREATER -1 AND FILAMENT_ENABLE_MULTIVIEW)
+        string(REGEX REPLACE "[.]filamat$" "_multiview.filamat" output_path_multiview ${output_path})
+        add_custom_command(
+                OUTPUT ${output_path_multiview}
+                COMMAND matc ${MATC_BASE_FLAGS} -PstereoscopicType=multiview -o ${output_path_multiview} ${fullname}
+                MAIN_DEPENDENCY ${fullname}
+                DEPENDS matc
+                COMMENT "Compiling material ${fullname} to ${output_path_multiview}"
+        )
+        list(APPEND MATERIAL_BINS ${output_path_multiview})
+    endif ()
+
+endforeach()
 
 # Additional dependencies on included files for materials
 

filament/backend/CMakeLists.txt

@@ -66,11 +66,13 @@ set(PRIVATE_HDRS
 # OpenGL / OpenGL ES Sources
 # ==================================================================================================
 
-if (FILAMENT_SUPPORTS_OPENGL AND NOT FILAMENT_USE_EXTERNAL_GLES3 AND NOT FILAMENT_USE_SWIFTSHADER)
+if (FILAMENT_SUPPORTS_OPENGL AND NOT FILAMENT_USE_EXTERNAL_GLES3)
     list(APPEND SRCS
             include/backend/platforms/OpenGLPlatform.h
             src/opengl/gl_headers.cpp
             src/opengl/gl_headers.h
+            src/opengl/GLBufferObject.h
+            src/opengl/GLTexture.h
             src/opengl/GLUtils.cpp
             src/opengl/GLUtils.h
             src/opengl/OpenGLBlobCache.cpp
@@ -166,6 +168,10 @@ endif()
 if (FILAMENT_SUPPORTS_VULKAN)
     list(APPEND SRCS
             include/backend/platforms/VulkanPlatform.h
+            src/vulkan/caching/VulkanDescriptorSetManager.cpp
+            src/vulkan/caching/VulkanDescriptorSetManager.h
+            src/vulkan/caching/VulkanPipelineLayoutCache.cpp
+            src/vulkan/caching/VulkanPipelineLayoutCache.h
             src/vulkan/platform/VulkanPlatform.cpp
             src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp
             src/vulkan/platform/VulkanPlatformSwapChainImpl.h
@@ -411,7 +417,9 @@ if (APPLE OR LINUX)
         test/test_MissingRequiredAttributes.cpp
         test/test_ReadPixels.cpp
         test/test_BufferUpdates.cpp
+        test/test_Callbacks.cpp
         test/test_MRT.cpp
+        test/test_PushConstants.cpp
         test/test_LoadImage.cpp
         test/test_StencilBuffer.cpp
         test/test_Scissor.cpp
@@ -472,6 +480,10 @@ if (APPLE)
        # linker from removing "unused" symbols.
        target_link_libraries(backend_test_mac PRIVATE -force_load backend_test)
        set_target_properties(backend_test_mac PROPERTIES FOLDER Tests)
+
+       # This is needed after XCode 15.3
+       set_target_properties(backend_test_mac PROPERTIES BUILD_WITH_INSTALL_RPATH TRUE)
+       set_target_properties(backend_test_mac PROPERTIES INSTALL_RPATH /usr/local/lib)
     endif()
 endif()
 

filament/backend/include/backend/DriverEnums.h

@@ -22,14 +22,17 @@
 #include <utils/BitmaskEnum.h>
 #include <utils/unwindows.h> // Because we define ERROR in the FenceStatus enum.
 
+#include <backend/Platform.h>
 #include <backend/PresentCallable.h>
 
+#include <utils/Invocable.h>
 #include <utils/ostream.h>
 
 #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 <variant>      // FIXME: STL headers are not allowed in public headers
 
 #include <stddef.h>
 #include <stdint.h>
@@ -81,14 +84,24 @@ static constexpr uint64_t SWAP_CHAIN_CONFIG_SRGB_COLORSPACE     = 0x10;
 /**
  * Indicates that the SwapChain should also contain a stencil component.
  */
-static constexpr uint64_t SWAP_CHAIN_HAS_STENCIL_BUFFER         = 0x20;
+static constexpr uint64_t SWAP_CHAIN_CONFIG_HAS_STENCIL_BUFFER  = 0x20;
+static constexpr uint64_t SWAP_CHAIN_HAS_STENCIL_BUFFER         = SWAP_CHAIN_CONFIG_HAS_STENCIL_BUFFER;
 
+/**
+ * The SwapChain contains protected content. Currently only supported by OpenGLPlatform and
+ * only when OpenGLPlatform::isProtectedContextSupported() is true.
+ */
+static constexpr uint64_t SWAP_CHAIN_CONFIG_PROTECTED_CONTENT   = 0x40;
 
 static constexpr size_t MAX_VERTEX_ATTRIBUTE_COUNT  = 16;   // This is guaranteed by OpenGL ES.
 static constexpr size_t MAX_SAMPLER_COUNT           = 62;   // Maximum needed at feature level 3.
 static constexpr size_t MAX_VERTEX_BUFFER_COUNT     = 16;   // Max number of bound buffer objects.
 static constexpr size_t MAX_SSBO_COUNT              = 4;    // This is guaranteed by OpenGL ES.
 
+static constexpr size_t MAX_PUSH_CONSTANT_COUNT     = 32;   // Vulkan 1.1 spec allows for 128-byte
+                                                            // of push constant (we assume 4-byte
+                                                            // types).
+
 // Per feature level caps
 // Use (int)FeatureLevel to index this array
 static constexpr struct {
@@ -105,7 +118,7 @@ static_assert(MAX_VERTEX_BUFFER_COUNT <= MAX_VERTEX_ATTRIBUTE_COUNT,
         "The number of buffer objects that can be attached to a VertexBuffer must be "
         "less than or equal to the maximum number of vertex attributes.");
 
-static constexpr size_t CONFIG_UNIFORM_BINDING_COUNT = 10;  // This is guaranteed by OpenGL ES.
+static constexpr size_t CONFIG_UNIFORM_BINDING_COUNT = 9;   // This is guaranteed by OpenGL ES.
 static constexpr size_t CONFIG_SAMPLER_BINDING_COUNT = 4;   // This is guaranteed by OpenGL ES.
 
 /**
@@ -129,6 +142,12 @@ enum class Backend : uint8_t {
     NOOP = 4,     //!< Selects the no-op driver for testing purposes.
 };
 
+enum class TimerQueryResult : int8_t {
+    ERROR = -1,     // an error occurred, result won't be available
+    NOT_READY = 0,  // result to ready yet
+    AVAILABLE = 1,  // result is available
+};
+
 static constexpr const char* backendToString(Backend backend) {
     switch (backend) {
         case Backend::NOOP:
@@ -145,14 +164,16 @@ static constexpr const char* backendToString(Backend backend) {
 }
 
 /**
- * Defines the shader language. Similar to the above backend enum, but the OpenGL backend can select
- * between two shader languages: ESSL 1.0 and ESSL 3.0.
+ * Defines the shader language. Similar to the above backend enum, with some differences:
+ * - The OpenGL backend can select between two shader languages: ESSL 1.0 and ESSL 3.0.
+ * - The Metal backend can prefer precompiled Metal libraries, while falling back to MSL.
  */
 enum class ShaderLanguage {
     ESSL1 = 0,
     ESSL3 = 1,
     SPIRV = 2,
     MSL = 3,
+    METAL_LIBRARY = 4,
 };
 
 static constexpr const char* shaderLanguageToString(ShaderLanguage shaderLanguage) {
@@ -165,6 +186,8 @@ static constexpr const char* shaderLanguageToString(ShaderLanguage shaderLanguag
             return "SPIR-V";
         case ShaderLanguage::MSL:
             return "MSL";
+        case ShaderLanguage::METAL_LIBRARY:
+            return "Metal precompiled library";
     }
 }
 
@@ -314,7 +337,7 @@ enum class UniformType : uint8_t {
 /**
  * Supported constant parameter types
  */
- enum class ConstantType : uint8_t {
+enum class ConstantType : uint8_t {
   INT,
   FLOAT,
   BOOL
@@ -659,16 +682,17 @@ enum class TextureFormat : uint16_t {
 };
 
 //! Bitmask describing the intended Texture Usage
-enum class TextureUsage : uint8_t {
-    NONE                = 0x00,
-    COLOR_ATTACHMENT    = 0x01,                     //!< Texture can be used as a color attachment
-    DEPTH_ATTACHMENT    = 0x02,                     //!< Texture can be used as a depth attachment
-    STENCIL_ATTACHMENT  = 0x04,                     //!< Texture can be used as a stencil attachment
-    UPLOADABLE          = 0x08,                     //!< Data can be uploaded into this texture (default)
-    SAMPLEABLE          = 0x10,                     //!< Texture can be sampled (default)
-    SUBPASS_INPUT       = 0x20,                     //!< Texture can be used as a subpass input
-    BLIT_SRC            = 0x40,                     //!< Texture can be used the source of a blit()
-    BLIT_DST            = 0x80,                     //!< Texture can be used the destination of a blit()
+enum class TextureUsage : uint16_t {
+    NONE                = 0x0000,
+    COLOR_ATTACHMENT    = 0x0001,            //!< Texture can be used as a color attachment
+    DEPTH_ATTACHMENT    = 0x0002,            //!< Texture can be used as a depth attachment
+    STENCIL_ATTACHMENT  = 0x0004,            //!< Texture can be used as a stencil attachment
+    UPLOADABLE          = 0x0008,            //!< Data can be uploaded into this texture (default)
+    SAMPLEABLE          = 0x0010,            //!< Texture can be sampled (default)
+    SUBPASS_INPUT       = 0x0020,            //!< Texture can be used as a subpass input
+    BLIT_SRC            = 0x0040,            //!< Texture can be used the source of a blit()
+    BLIT_DST            = 0x0080,            //!< Texture can be used the destination of a blit()
+    PROTECTED           = 0x0100,            //!< Texture can be used for protected content
     DEFAULT             = UPLOADABLE | SAMPLEABLE   //!< Default texture usage
 };
 
@@ -1034,7 +1058,7 @@ struct RasterState {
             bool inverseFrontFaces                      : 1;        // 31
 
             //! padding, must be 0
-            uint8_t padding                             : 1;        // 32
+            bool depthClamp                             : 1;        // 32
         };
         uint32_t u = 0;
     };
@@ -1172,11 +1196,27 @@ struct StencilState {
 
     //! Stencil operations for front-facing polygons
     StencilOperations front = {
-            .stencilFunc = StencilFunction::A, .ref = 0, .readMask = 0xff, .writeMask = 0xff };
+            .stencilFunc = StencilFunction::A,
+            .stencilOpStencilFail = StencilOperation::KEEP,
+            .padding0 = 0,
+            .stencilOpDepthFail = StencilOperation::KEEP,
+            .stencilOpDepthStencilPass = StencilOperation::KEEP,
+            .padding1 = 0,
+            .ref = 0,
+            .readMask = 0xff,
+            .writeMask = 0xff };
 
     //! Stencil operations for back-facing polygons
     StencilOperations back  = {
-            .stencilFunc = StencilFunction::A, .ref = 0, .readMask = 0xff, .writeMask = 0xff };
+            .stencilFunc = StencilFunction::A,
+            .stencilOpStencilFail = StencilOperation::KEEP,
+            .padding0 = 0,
+            .stencilOpDepthFail = StencilOperation::KEEP,
+            .stencilOpDepthStencilPass = StencilOperation::KEEP,
+            .padding1 = 0,
+            .ref = 0,
+            .readMask = 0xff,
+            .writeMask = 0xff };
 
     //! Whether stencil-buffer writes are enabled
     bool stencilWrite = false;
@@ -1184,13 +1224,15 @@ struct StencilState {
     uint8_t padding = 0;
 };
 
+using PushConstantVariant = std::variant<int32_t, float, bool>;
+
 static_assert(sizeof(StencilState::StencilOperations) == 5u,
         "StencilOperations size not what was intended");
 
 static_assert(sizeof(StencilState) == 12u,
         "StencilState size not what was intended");
 
-using FrameScheduledCallback = void(*)(PresentCallable callable, void* user);
+using FrameScheduledCallback = utils::Invocable<void(backend::PresentCallable)>;
 
 enum class Workaround : uint16_t {
     // The EASU pass must split because shader compiler flattens early-exit branch
@@ -1202,23 +1244,14 @@ enum class Workaround : uint16_t {
     ADRENO_UNIFORM_ARRAY_CRASH,
     // Workaround a Metal pipeline compilation error with the message:
     // "Could not statically determine the target of a texture". See light_indirect.fs
-    A8X_STATIC_TEXTURE_TARGET_ERROR,
+    METAL_STATIC_TEXTURE_TARGET_ERROR,
     // Adreno drivers sometimes aren't able to blit into a layer of a texture array.
     DISABLE_BLIT_INTO_TEXTURE_ARRAY,
     // Multiple workarounds needed for PowerVR GPUs
     POWER_VR_SHADER_WORKAROUNDS,
-    // The driver has some threads pinned, and we can't easily know on which core, it can hurt
-    // performance more if we end-up pinned on the same one.
-    DISABLE_THREAD_AFFINITY
 };
 
-//! The type of technique for stereoscopic rendering
-enum class StereoscopicType : uint8_t {
-    // Stereoscopic rendering is performed using instanced rendering technique.
-    INSTANCED,
-    // Stereoscopic rendering is performed using the multiview feature from the graphics backend.
-    MULTIVIEW,
-};
+using StereoscopicType = backend::Platform::StereoscopicType;
 
 } // namespace filament::backend
 

filament/backend/include/backend/Handle.h

@@ -75,6 +75,19 @@ protected:
     HandleBase(HandleBase const& rhs) noexcept = default;
     HandleBase& operator=(HandleBase const& rhs) noexcept = default;
 
+    HandleBase(HandleBase&& rhs) noexcept
+            : object(rhs.object) {
+        rhs.object = nullid;
+    }
+
+    HandleBase& operator=(HandleBase&& rhs) noexcept {
+        if (this != &rhs) {
+            object = rhs.object;
+            rhs.object = nullid;
+        }
+        return *this;
+    }
+
 private:
     HandleId object;
 };
@@ -89,8 +102,10 @@ struct Handle : public HandleBase {
     Handle() noexcept = default;
 
     Handle(Handle const& rhs) noexcept = default;
+    Handle(Handle&& rhs) noexcept = default;
 
     Handle& operator=(Handle const& rhs) noexcept = default;
+    Handle& operator=(Handle&& rhs) noexcept = default;
 
     explicit Handle(HandleId id) noexcept : HandleBase(id) { }
 

filament/backend/include/backend/PipelineState.h

@@ -29,10 +29,13 @@ namespace filament::backend {
 //! \privatesection
 
 struct PipelineState {
-    Handle<HwProgram> program;
-    RasterState rasterState;
-    StencilState stencilState;
-    PolygonOffset polygonOffset;
+    Handle<HwProgram> program;                                              //  4
+    Handle<HwVertexBufferInfo> vertexBufferInfo;                            //  4
+    RasterState rasterState;                                                //  4
+    StencilState stencilState;                                              // 12
+    PolygonOffset polygonOffset;                                            //  8
+    PrimitiveType primitiveType = PrimitiveType::TRIANGLES;                 //  1
+    uint8_t padding[3] = {};                                                //  3
 };
 
 } // namespace filament::backend

filament/backend/include/backend/Platform.h

@@ -23,6 +23,7 @@
 #include <utils/Invocable.h>
 
 #include <stddef.h>
+#include <stdint.h>
 
 namespace filament::backend {
 
@@ -40,6 +41,26 @@ public:
     struct Fence {};
     struct Stream {};
 
+    /**
+     * The type of technique for stereoscopic rendering. (Note that the materials used will need to
+     * be compatible with the chosen technique.)
+     */
+    enum class StereoscopicType : uint8_t {
+        /**
+         * No stereoscopic rendering
+         */
+        NONE,
+        /**
+         * Stereoscopic rendering is performed using instanced rendering technique.
+         */
+        INSTANCED,
+        /**
+         * Stereoscopic rendering is performed using the multiview feature from the graphics
+         * backend.
+         */
+        MULTIVIEW,
+    };
+
     struct DriverConfig {
         /**
          * Size of handle arena in bytes. Setting to 0 indicates default value is to be used.
@@ -54,11 +75,29 @@ public:
          */
         size_t textureUseAfterFreePoolSize = 0;
 
+        size_t metalUploadBufferSizeBytes = 512 * 1024;
+
         /**
          * Set to `true` to forcibly disable parallel shader compilation in the backend.
-         * Currently only honored by the GL backend.
+         * Currently only honored by the GL and Metal backends.
          */
         bool disableParallelShaderCompile = false;
+
+        /**
+         * Disable backend handles use-after-free checks.
+         */
+        bool disableHandleUseAfterFreeCheck = false;
+
+        /**
+         * Force GLES2 context if supported, or pretend the context is ES2. Only meaningful on
+         * GLES 3.x backends.
+         */
+        bool forceGLES2Context = false;
+
+        /**
+         * Sets the technique for stereoscopic rendering.
+         */
+        StereoscopicType stereoscopicType = StereoscopicType::NONE;
     };
 
     Platform() noexcept;
@@ -188,9 +227,40 @@ public:
     size_t retrieveBlob(const void* UTILS_NONNULL key, size_t keySize,
             void* UTILS_NONNULL value, size_t valueSize);
 
+    using DebugUpdateStatFunc = utils::Invocable<void(const char* UTILS_NONNULL key, uint64_t value)>;
+
+    /**
+     * Sets the callback function that the backend can use to update backend-specific statistics
+     * to aid with debugging. This callback is guaranteed to be called on the Filament driver
+     * thread.
+     *
+     * @param debugUpdateStat   an Invocable that updates debug statistics
+     */
+    void setDebugUpdateStatFunc(DebugUpdateStatFunc&& debugUpdateStat) noexcept;
+
+    /**
+     * @return true if debugUpdateStat is valid.
+     */
+    bool hasDebugUpdateStatFunc() const noexcept;
+
+    /**
+     * To track backend-specific statistics, the backend implementation can call the
+     * application-provided callback function debugUpdateStatFunc to associate or update a value
+     * with a given key. It is possible for this function to be called multiple times with the
+     * same key, in which case newer values should overwrite older values.
+     *
+     * This function is guaranteed to be called only on a single thread, the Filament driver
+     * thread.
+     *
+     * @param key          a null-terminated C-string with the key of the debug statistic
+     * @param value        the updated value of key
+     */
+    void debugUpdateStat(const char* UTILS_NONNULL key, uint64_t value);
+
 private:
     InsertBlobFunc mInsertBlob;
     RetrieveBlobFunc mRetrieveBlob;
+    DebugUpdateStatFunc mDebugUpdateStat;
 };
 
 } // namespace filament

filament/backend/include/backend/PresentCallable.h

@@ -48,7 +48,7 @@ namespace filament::backend {
  * and optional user data:
  *
  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
- * swapChain->setFrameScheduledCallback(myFrameScheduledCallback, nullptr);
+ * swapChain->setFrameScheduledCallback(nullptr, myFrameScheduledCallback);
  * if (renderer->beginFrame(swapChain)) {
  *     renderer->render(view);
  *     renderer->endFrame();
@@ -58,8 +58,6 @@ namespace filament::backend {
  * @remark Only Filament's Metal backend supports PresentCallables and frame callbacks. Other
  * backends ignore the callback (which will never be called) and proceed normally.
  *
- * @remark The SwapChain::FrameScheduledCallback is called on an arbitrary thread.
- *
  * Applications *must* call each PresentCallable they receive. Each PresentCallable represents a
  * frame that is waiting to be presented. If an application fails to call a PresentCallable, a
  * memory leak could occur. To "cancel" the presentation of a frame, pass false to the

filament/backend/include/backend/Program.h

@@ -84,6 +84,9 @@ public:
     // null terminating character.
     Program& shader(ShaderStage shader, void const* data, size_t size);
 
+    // sets the language of the shader sources provided with shader() (defaults to ESSL3)
+    Program& shaderLanguage(ShaderLanguage shaderLanguage);
+
     // Note: This is only needed for GLES3.0 backends, because the layout(binding=) syntax is
     //       not permitted in glsl. The backend needs a way to associate a uniform block
     //       to a binding point.
@@ -114,8 +117,18 @@ public:
     Program& specializationConstants(
             utils::FixedCapacityVector<SpecializationConstant> specConstants) noexcept;
 
+    struct PushConstant {
+        utils::CString name;
+        ConstantType type;
+    };
+
+    Program& pushConstants(ShaderStage stage,
+            utils::FixedCapacityVector<PushConstant> constants) noexcept;
+
     Program& cacheId(uint64_t cacheId) noexcept;
 
+    Program& multiview(bool multiview) noexcept;
+
     ShaderSource const& getShadersSource() const noexcept { return mShadersSource; }
     ShaderSource& getShadersSource() noexcept { return mShadersSource; }
 
@@ -134,6 +147,8 @@ public:
     utils::CString const& getName() const noexcept { return mName; }
     utils::CString& getName() noexcept { return mName; }
 
+    auto const& getShaderLanguage() const { return mShaderLanguage; }
+
     utils::FixedCapacityVector<SpecializationConstant> const& getSpecializationConstants() const noexcept {
         return mSpecializationConstants;
     }
@@ -141,8 +156,19 @@ public:
         return mSpecializationConstants;
     }
 
+    utils::FixedCapacityVector<PushConstant> const& getPushConstants(
+            ShaderStage stage) const noexcept {
+        return mPushConstants[static_cast<uint8_t>(stage)];
+    }
+
+    utils::FixedCapacityVector<PushConstant>& getPushConstants(ShaderStage stage) noexcept {
+        return mPushConstants[static_cast<uint8_t>(stage)];
+    }
+
     uint64_t getCacheId() const noexcept { return mCacheId; }
 
+    bool isMultiview() const noexcept { return mMultiview; }
+
     CompilerPriorityQueue getPriorityQueue() const noexcept { return mPriorityQueue; }
 
 private:
@@ -151,13 +177,20 @@ private:
     UniformBlockInfo mUniformBlocks = {};
     SamplerGroupInfo mSamplerGroups = {};
     ShaderSource mShadersSource;
+    ShaderLanguage mShaderLanguage = ShaderLanguage::ESSL3;
     utils::CString mName;
     uint64_t mCacheId{};
     utils::Invocable<utils::io::ostream&(utils::io::ostream& out)> mLogger;
     utils::FixedCapacityVector<SpecializationConstant> mSpecializationConstants;
+    std::array<utils::FixedCapacityVector<PushConstant>, SHADER_TYPE_COUNT> mPushConstants;
     utils::FixedCapacityVector<std::pair<utils::CString, uint8_t>> mAttributes;
     std::array<UniformInfo, Program::UNIFORM_BINDING_COUNT> mBindingUniformInfo;
     CompilerPriorityQueue mPriorityQueue = CompilerPriorityQueue::HIGH;
+    // Indicates the current engine was initialized with multiview stereo, and the variant for this
+    // program contains STE flag. This will be referred later for the OpenGL shader compiler to
+    // determine whether shader code replacement for the num_views should be performed.
+    // This variable could be promoted as a more generic variable later if other similar needs occur.
+    bool mMultiview = false;
 };
 
 } // namespace filament::backend

filament/backend/include/backend/TargetBufferInfo.h

@@ -29,13 +29,36 @@ namespace filament::backend {
 //! \privatesection
 
 struct TargetBufferInfo {
+    // note: the parameters of this constructor are not in the order of this structure's fields
+    TargetBufferInfo(Handle<HwTexture> handle, uint8_t level, uint16_t layer, uint8_t baseViewIndex) noexcept
+        : handle(handle), baseViewIndex(baseViewIndex), level(level), layer(layer) {
+    }
+
+    TargetBufferInfo(Handle<HwTexture> handle, uint8_t level, uint16_t layer) noexcept
+            : handle(handle), level(level), layer(layer) {
+    }
+
+    TargetBufferInfo(Handle<HwTexture> handle, uint8_t level) noexcept
+            : handle(handle), level(level) {
+    }
+
+    TargetBufferInfo(Handle<HwTexture> handle) noexcept // NOLINT(*-explicit-constructor)
+            : handle(handle) {
+    }
+
+    TargetBufferInfo() noexcept = default;
+
     // texture to be used as render target
     Handle<HwTexture> handle;
 
+    // Starting layer index for multiview. This value is only used when the `layerCount` for the
+    // render target is greater than 1.
+    uint8_t baseViewIndex = 0;
+
     // level to be used
     uint8_t level = 0;
 
-    // for cubemaps and 3D textures. See TextureCubemapFace for the face->layer mapping
+    // For cubemaps and 3D textures. See TextureCubemapFace for the face->layer mapping
     uint16_t layer = 0;
 };
 
@@ -60,7 +83,7 @@ public:
 
     MRT() noexcept = default;
 
-    MRT(TargetBufferInfo const& color) noexcept // NOLINT(hicpp-explicit-conversions)
+    MRT(TargetBufferInfo const& color) noexcept // NOLINT(hicpp-explicit-conversions, *-explicit-constructor)
             : mInfos{ color } {
     }
 
@@ -80,7 +103,7 @@ public:
 
     // this is here for backward compatibility
     MRT(Handle<HwTexture> handle, uint8_t level, uint16_t layer) noexcept
-            : mInfos{{ handle, level, layer }} {
+            : mInfos{{ handle, level, layer, 0 }} {
     }
 };
 

filament/backend/include/backend/platforms/OpenGLPlatform.h

@@ -22,7 +22,9 @@
 #include <backend/Platform.h>
 
 #include <utils/compiler.h>
+#include <utils/Invocable.h>
 
+#include <stddef.h>
 #include <stdint.h>
 
 namespace filament::backend {
@@ -62,6 +64,22 @@ public:
      */
     virtual void terminate() noexcept = 0;
 
+    /**
+     * Return whether createSwapChain supports the SWAP_CHAIN_CONFIG_SRGB_COLORSPACE flag.
+     * The default implementation returns false.
+     *
+     * @return true if SWAP_CHAIN_CONFIG_SRGB_COLORSPACE is supported, false otherwise.
+     */
+    virtual bool isSRGBSwapChainSupported() const noexcept;
+
+    /**
+     * Return whether protected contexts are supported by this backend.
+     * If protected context are supported, the SWAP_CHAIN_CONFIG_PROTECTED_CONTENT flag can be
+     * used when creating a SwapChain.
+     * The default implementation returns false.
+     */
+    virtual bool isProtectedContextSupported() const noexcept;
+
     /**
      * Called by the driver to create a SwapChain for this driver.
      *
@@ -71,17 +89,9 @@ public:
      * @return              The driver's SwapChain object.
      *
      */
-    virtual SwapChain* UTILS_NONNULL createSwapChain(
+    virtual SwapChain* UTILS_NULLABLE createSwapChain(
             void* UTILS_NULLABLE nativeWindow, uint64_t flags) noexcept = 0;
 
-    /**
-     * Return whether createSwapChain supports the SWAP_CHAIN_CONFIG_SRGB_COLORSPACE flag.
-     * The default implementation returns false.
-     *
-     * @return true if SWAP_CHAIN_CONFIG_SRGB_COLORSPACE is supported, false otherwise.
-     */
-    virtual bool isSRGBSwapChainSupported() const noexcept;
-
     /**
      * Called by the driver create a headless SwapChain.
      *
@@ -116,20 +126,93 @@ public:
     virtual TargetBufferFlags getPreservedFlags(SwapChain* UTILS_NONNULL swapChain) noexcept;
 
     /**
-     * Called by the driver to establish the default FBO. The default implementation returns 0.
-      * @return a GLuint casted to a uint32_t that is an OpenGL framebuffer object.
+     * Returns true if the swapchain is protected
      */
-    virtual uint32_t createDefaultRenderTarget() noexcept;
+    virtual bool isSwapChainProtected(Platform::SwapChain* UTILS_NONNULL swapChain) noexcept;
+
+    /**
+     * Called by the driver to establish the default FBO. The default implementation returns 0.
+     *
+     * This method can be called either on the regular or protected OpenGL contexts and can return
+     * a different or identical name, since these names exist in different namespaces.
+     *
+     * @return a GLuint casted to a uint32_t that is an OpenGL framebuffer object.
+     */
+    virtual uint32_t getDefaultFramebufferObject() noexcept;
+
+    /**
+     * Called by the backend when a frame starts.
+     * @param steady_clock_ns vsync time point on the monotonic clock
+     * @param refreshIntervalNs refresh interval in nanosecond
+     * @param frameId a frame id
+     */
+    virtual void beginFrame(
+            int64_t monotonic_clock_ns,
+            int64_t refreshIntervalNs,
+            uint32_t frameId) noexcept;
+
+    /**
+     * Called by the backend when a frame ends.
+     * @param frameId the frame id used in beginFrame
+     */
+    virtual void endFrame(
+            uint32_t frameId) noexcept;
+
+    /**
+     * Type of contexts available
+     */
+    enum class ContextType {
+        NONE,           //!< No current context
+        UNPROTECTED,    //!< current context is unprotected
+        PROTECTED       //!< current context supports protected content
+    };
+
+    /**
+     * Returns the type of the context currently in use. This value is updated by makeCurrent()
+     * and therefore can be cached between calls. ContextType::PROTECTED can only be returned
+     * if isProtectedContextSupported() is true.
+     * @return ContextType
+     */
+    virtual ContextType getCurrentContextType() const noexcept;
+
+    /**
+     * Binds the requested context to the current thread and drawSwapChain to the default FBO
+     * returned by getDefaultFramebufferObject().
+     *
+     * @param type type of context to bind to the current thread.
+     * @param drawSwapChain SwapChain to draw to. It must be bound to the default FBO.
+     * @param readSwapChain SwapChain to read from (for operation like `glBlitFramebuffer`)
+     * @return true on success, false on error.
+     */
+    virtual bool makeCurrent(ContextType type,
+            SwapChain* UTILS_NONNULL drawSwapChain,
+            SwapChain* UTILS_NONNULL readSwapChain) noexcept = 0;
 
     /**
      * Called by the driver to make the OpenGL context active on the calling thread and bind
-     * the drawSwapChain to the default render target (FBO) created with createDefaultRenderTarget.
+     * the drawSwapChain to the default FBO returned by getDefaultFramebufferObject().
+     * The context used is either the default context or the protected context. When a context
+     * change is necessary, the preContextChange and postContextChange callbacks are called,
+     * before and after the context change respectively. postContextChange is given the index
+     * of the new context (0 for default and 1 for protected).
+     * The default implementation just calls makeCurrent(getCurrentContextType(), SwapChain*, SwapChain*).
+     *
      * @param drawSwapChain SwapChain to draw to. It must be bound to the default FBO.
      * @param readSwapChain SwapChain to read from (for operation like `glBlitFramebuffer`)
+     * @param preContextChange called before the context changes
+     * @param postContextChange called after the context changes
      */
     virtual void makeCurrent(
             SwapChain* UTILS_NONNULL drawSwapChain,
-            SwapChain* UTILS_NONNULL readSwapChain) noexcept = 0;
+            SwapChain* UTILS_NONNULL readSwapChain,
+            utils::Invocable<void()> preContextChange,
+            utils::Invocable<void(size_t index)> postContextChange) noexcept;
+
+    /**
+     * Called by the backend just before calling commit()
+     * @see commit()
+     */
+    virtual void preCommit() noexcept;
 
     /**
      * Called by the driver once the current frame finishes drawing. Typically, this should present

filament/backend/include/backend/platforms/PlatformCocoaGL.h

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

filament/backend/include/backend/platforms/PlatformCocoaTouchGL.h

@@ -45,7 +45,7 @@ public:
 
     void terminate() noexcept override;
 
-    uint32_t createDefaultRenderTarget() noexcept override;
+    uint32_t getDefaultFramebufferObject() noexcept override;
 
     bool isExtraContextSupported() const noexcept override;
     void createContext(bool shared) override;
@@ -53,7 +53,7 @@ public:
     SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
     SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
     void destroySwapChain(SwapChain* swapChain) noexcept override;
-    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
+    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
     void commit(SwapChain* swapChain) noexcept override;
 
     OpenGLPlatform::ExternalTexture* createExternalImageTexture() noexcept override;

filament/backend/include/backend/platforms/PlatformEGL.h

@@ -25,6 +25,8 @@
 #include <EGL/eglext.h>
 #include <EGL/eglplatform.h>
 
+#include <utils/Invocable.h>
+
 #include <initializer_list>
 #include <utility>
 #include <vector>
@@ -41,15 +43,11 @@ class PlatformEGL : public OpenGLPlatform {
 public:
 
     PlatformEGL() noexcept;
-    bool isExtraContextSupported() const noexcept override;
-    void createContext(bool shared) override;
-    void releaseContext() noexcept override;
 
     // Return true if we're on an OpenGL platform (as opposed to OpenGL ES). false by default.
     virtual bool isOpenGL() const noexcept;
 
 protected:
-
     // --------------------------------------------------------------------------------------------
     // Helper for EGL configs and attributes parameters
 
@@ -89,13 +87,30 @@ protected:
     // --------------------------------------------------------------------------------------------
     // OpenGLPlatform Interface
 
+    bool isExtraContextSupported() const noexcept override;
+    void createContext(bool shared) override;
+    void releaseContext() noexcept override;
+
     void terminate() noexcept override;
 
+    bool isProtectedContextSupported() const noexcept override;
+
     bool isSRGBSwapChainSupported() const noexcept override;
     SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
     SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
     void destroySwapChain(SwapChain* swapChain) noexcept override;
-    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
+    bool isSwapChainProtected(SwapChain* swapChain) noexcept override;
+
+    ContextType getCurrentContextType() const noexcept override;
+
+    bool makeCurrent(ContextType type,
+            SwapChain* drawSwapChain,
+            SwapChain* readSwapChain) noexcept override;
+
+    void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain,
+            utils::Invocable<void()> preContextChange,
+            utils::Invocable<void(size_t index)> postContextChange) noexcept override;
+
     void commit(SwapChain* swapChain) noexcept override;
 
     bool canCreateFence() noexcept override;
@@ -122,16 +137,27 @@ protected:
     static void clearGlError() noexcept;
 
     /**
-     * Always use this instead of eglMakeCurrent().
+     * Always use this instead of eglMakeCurrent(), as it tracks some state.
      */
-    EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) noexcept;
+
+    EGLContext getContextForType(ContextType type) const noexcept;
+
+    // makes the draw and read surface current without changing the current context
+    EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) noexcept {
+        return egl.makeCurrent(drawSurface, readSurface);
+    }
+
+    // makes context current and set draw and read surfaces to EGL_NO_SURFACE
+    EGLBoolean makeCurrent(EGLContext context) noexcept {
+        return egl.makeCurrent(context, mEGLDummySurface, mEGLDummySurface);
+    }
 
     // TODO: this should probably use getters instead.
     EGLDisplay mEGLDisplay = EGL_NO_DISPLAY;
     EGLContext mEGLContext = EGL_NO_CONTEXT;
-    EGLSurface mCurrentDrawSurface = EGL_NO_SURFACE;
-    EGLSurface mCurrentReadSurface = EGL_NO_SURFACE;
+    EGLContext mEGLContextProtected = EGL_NO_CONTEXT;
     EGLSurface mEGLDummySurface = EGL_NO_SURFACE;
+    ContextType mCurrentContextType = ContextType::NONE;
     // mEGLConfig is valid only if ext.egl.KHR_no_config_context is false
     EGLConfig mEGLConfig = EGL_NO_CONFIG_KHR;
     Config mContextAttribs;
@@ -148,6 +174,7 @@ protected:
             bool KHR_gl_colorspace = false;
             bool KHR_no_config_context = false;
             bool KHR_surfaceless_context = false;
+            bool EXT_protected_content = false;
         } egl;
     } ext;
 
@@ -156,12 +183,29 @@ protected:
         Config attribs{};
         EGLNativeWindowType nativeWindow{};
         EGLConfig config{};
+        uint64_t flags{};
     };
 
     void initializeGlExtensions() noexcept;
 
 protected:
     EGLConfig findSwapChainConfig(uint64_t flags, bool window, bool pbuffer) const;
+
+private:
+    class EGL {
+        EGLDisplay& mEGLDisplay;
+        EGLSurface mCurrentDrawSurface = EGL_NO_SURFACE;
+        EGLSurface mCurrentReadSurface = EGL_NO_SURFACE;
+        EGLContext mCurrentContext = EGL_NO_CONTEXT;
+    public:
+        explicit EGL(EGLDisplay& dpy) : mEGLDisplay(dpy) {}
+        EGLBoolean makeCurrent(EGLContext context,
+                EGLSurface drawSurface, EGLSurface readSurface) noexcept;
+
+        EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) noexcept {
+            return makeCurrent(mCurrentContext, drawSurface, readSurface);
+        }
+    } egl{ mEGLDisplay };
 };
 
 } // namespace filament::backend

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

@@ -22,6 +22,10 @@
 #include <backend/platforms/OpenGLPlatform.h>
 #include <backend/platforms/PlatformEGL.h>
 
+#include <utils/android/PerformanceHintManager.h>
+
+#include <chrono>
+
 #include <stddef.h>
 #include <stdint.h>
 
@@ -58,6 +62,13 @@ protected:
 
     void terminate() noexcept override;
 
+    void beginFrame(
+            int64_t monotonic_clock_ns,
+            int64_t refreshIntervalNs,
+            uint32_t frameId) noexcept override;
+
+    void preCommit() noexcept override;
+
     /**
      * Set the presentation time using `eglPresentationTimeANDROID`
      * @param presentationTimeInNanosecond
@@ -79,8 +90,18 @@ protected:
     AcquiredImage transformAcquiredImage(AcquiredImage source) noexcept override;
 
 private:
+    struct InitializeJvmForPerformanceManagerIfNeeded {
+        InitializeJvmForPerformanceManagerIfNeeded();
+    };
+
     int mOSVersion;
     ExternalStreamManagerAndroid& mExternalStreamManager;
+    InitializeJvmForPerformanceManagerIfNeeded const mInitializeJvmForPerformanceManagerIfNeeded;
+    utils::PerformanceHintManager mPerformanceHintManager;
+    utils::PerformanceHintManager::Session mPerformanceHintSession;
+
+    using clock = std::chrono::high_resolution_clock;
+    clock::time_point mStartTimeOfActualWork;
 };
 
 } // namespace filament::backend

filament/backend/include/backend/platforms/PlatformGLX.h

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

filament/backend/include/backend/platforms/PlatformWGL.h

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

filament/backend/include/backend/platforms/PlatformWebGL.h

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

filament/backend/include/backend/platforms/VulkanPlatform.h

@@ -23,9 +23,9 @@
 
 #include <utils/CString.h>
 #include <utils/FixedCapacityVector.h>
+#include <utils/Hash.h>
 #include <utils/PrivateImplementation.h>
 
-#include <string_view>
 #include <tuple>
 #include <unordered_set>
 
@@ -47,6 +47,14 @@ struct VulkanPlatformPrivate;
 class VulkanPlatform : public Platform, utils::PrivateImplementation<VulkanPlatformPrivate> {
 public:
 
+    struct ExtensionHashFn {
+        std::size_t operator()(utils::CString const& s) const noexcept {
+            return std::hash<std::string>{}(s.data());
+        }
+    };
+    // Utility for managing device or instance extensions during initialization.
+    using ExtensionSet = std::unordered_set<utils::CString, ExtensionHashFn>;
+
     /**
      * A collection of handles to objects and metadata that comprises a Vulkan context. The client
      * can instantiate this struct and pass to Engine::Builder::sharedContext if they wishes to
@@ -82,6 +90,20 @@ public:
         VkExtent2D extent = {0, 0};
     };
 
+    struct ImageSyncData {
+        static constexpr uint32_t INVALID_IMAGE_INDEX = UINT32_MAX;
+
+        // The index of the next image as returned by vkAcquireNextImage or equivalent.
+        uint32_t imageIndex = INVALID_IMAGE_INDEX;
+
+        // Semaphore to be signaled once the image is available.
+        VkSemaphore imageReadySemaphore = VK_NULL_HANDLE;
+
+        // A function called right before vkQueueSubmit. After this call, the image must be 
+        // available. This pointer can be null if imageReadySemaphore is not VK_NULL_HANDLE.
+        std::function<void(SwapChainPtr handle)> explicitImageReadyWait = nullptr;
+    };
+
     VulkanPlatform();
 
     ~VulkanPlatform() override;
@@ -119,6 +141,12 @@ public:
          * before recreating the swapchain. Default is true.
          */
         bool flushAndWaitOnWindowResize = true;
+
+        /**
+         * Whether the swapchain image should be transitioned to a layout suitable for
+         * presentation. Default is true.
+         */
+        bool transitionSwapChainImageLayoutForPresent = true;
     };
 
     /**
@@ -147,13 +175,10 @@ public:
      * corresponding VkImage will be used as the output color attachment. The client should signal
      * the `clientSignal` semaphore when the image is ready to be used by the backend.
      * @param handle         The handle returned by createSwapChain()
-     * @param clientSignal   The semaphore that the client will signal to indicate that the backend
-     *                       may render into the image.
-     * @param index          Pointer to memory that will be filled with the index that corresponding
-     *                       to an image in the `SwapChainBundle.colors` array.
+     * @param outImageSyncData The synchronization data used for image readiness
      * @return               Result of acquire
      */
-    virtual VkResult acquire(SwapChainPtr handle, VkSemaphore clientSignal, uint32_t* index);
+    virtual VkResult acquire(SwapChainPtr handle, ImageSyncData* outImageSyncData);
 
     /**
      * Present the image corresponding to `index` to the display. The client should wait on
@@ -192,6 +217,13 @@ public:
     virtual SwapChainPtr createSwapChain(void* nativeWindow, uint64_t flags = 0,
             VkExtent2D extent = {0, 0});
 
+    /**
+     * Allows implementers to provide instance extensions that they'd like to include in the
+     * instance creation.
+     * @return          A set of extensions to enable for the instance.
+     */
+    virtual ExtensionSet getRequiredInstanceExtensions() { return {}; }
+
     /**
      * Destroy the swapchain.
      * @param handle    The handle returned by createSwapChain()
@@ -236,10 +268,9 @@ public:
     VkQueue getGraphicsQueue() const noexcept;
 
 private:
-    // Platform dependent helper methods
-    using ExtensionSet = std::unordered_set<std::string_view>;
-    static ExtensionSet getRequiredInstanceExtensions();
+    static ExtensionSet getSwapchainInstanceExtensions();
 
+    // Platform dependent helper methods
     using SurfaceBundle = std::tuple<VkSurfaceKHR, VkExtent2D>;
     static SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
             uint64_t flags) noexcept;

filament/backend/include/private/backend/CommandBufferQueue.h

@@ -50,12 +50,13 @@ class CommandBufferQueue {
     size_t mFreeSpace = 0;
     size_t mHighWatermark = 0;
     uint32_t mExitRequested = 0;
+    bool mPaused = false;
 
     static constexpr uint32_t EXIT_REQUESTED = 0x31415926;
 
 public:
     // requiredSize: guaranteed available space after flush()
-    CommandBufferQueue(size_t requiredSize, size_t bufferSize);
+    CommandBufferQueue(size_t requiredSize, size_t bufferSize, bool paused);
     ~CommandBufferQueue();
 
     CircularBuffer& getCircularBuffer() noexcept { return mCircularBuffer; }
@@ -80,6 +81,10 @@ public:
     // returns from waitForCommands() immediately.
     void requestExit();
 
+    // suspend or unsuspend the queue.
+    bool isPaused() const noexcept;
+    void setPaused(bool paused);
+
     bool isExitRequested() const;
 };
 

filament/backend/include/private/backend/CommandStream.h

@@ -134,7 +134,7 @@ struct CommandType<void (Driver::*)(ARGS...)> {
 
     public:
         template<typename M, typename D>
-        static inline void execute(M&& method, D&& driver, CommandBase* base, intptr_t* next) noexcept {
+        static inline void execute(M&& method, D&& driver, CommandBase* base, intptr_t* next) {
             Command* self = static_cast<Command*>(base);
             *next = align(sizeof(Command));
 #if DEBUG_COMMAND_STREAM
@@ -168,7 +168,7 @@ struct CommandType<void (Driver::*)(ARGS...)> {
 
 class CustomCommand : public CommandBase {
     std::function<void()> mCommand;
-    static void execute(Driver&, CommandBase* base, intptr_t* next) noexcept;
+    static void execute(Driver&, CommandBase* base, intptr_t* next);
 public:
     inline CustomCommand(CustomCommand&& rhs) = default;
     inline explicit CustomCommand(std::function<void()> cmd)

filament/backend/include/private/backend/Driver.h

@@ -76,7 +76,7 @@ public:
     // the fn function will execute a batch of driver commands
     // this gives the driver a chance to wrap their execution in a meaningful manner
     // the default implementation simply calls fn
-    virtual void execute(std::function<void(void)> const& fn) noexcept;
+    virtual void execute(std::function<void(void)> const& fn);
 
     // This is called on debug build, or when enabled manually on the backend thread side.
     virtual void debugCommandBegin(CommandStream* cmds,

filament/backend/include/private/backend/DriverAPI.inc

@@ -133,18 +133,18 @@ DECL_DRIVER_API_0(tick)
 
 DECL_DRIVER_API_N(beginFrame,
         int64_t, monotonic_clock_ns,
+        int64_t, refreshIntervalNs,
         uint32_t, frameId)
 
 DECL_DRIVER_API_N(setFrameScheduledCallback,
         backend::SwapChainHandle, sch,
-        backend::FrameScheduledCallback, callback,
-        void*, user)
+        backend::CallbackHandler*, handler,
+        backend::FrameScheduledCallback&&, callback)
 
 DECL_DRIVER_API_N(setFrameCompletedCallback,
         backend::SwapChainHandle, sch,
         backend::CallbackHandler*, handler,
-        backend::CallbackHandler::Callback, callback,
-        void*, user)
+        utils::Invocable<void(void)>&&, callback)
 
 DECL_DRIVER_API_N(setPresentationTime,
         int64_t, monotonic_clock_ns)
@@ -239,6 +239,7 @@ DECL_DRIVER_API_R_N(backend::RenderTargetHandle, createRenderTarget,
         uint32_t, width,
         uint32_t, height,
         uint8_t, samples,
+        uint8_t, layerCount,
         backend::MRT, color,
         backend::TargetBufferInfo, depth,
         backend::TargetBufferInfo, stencil)
@@ -298,14 +299,18 @@ DECL_DRIVER_API_SYNCHRONOUS_0(bool, isFrameBufferFetchMultiSampleSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isFrameTimeSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isAutoDepthResolveSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isSRGBSwapChainSupported)
-DECL_DRIVER_API_SYNCHRONOUS_N(bool, isStereoSupported, backend::StereoscopicType, stereoscopicType)
+DECL_DRIVER_API_SYNCHRONOUS_0(bool, isProtectedContentSupported)
+DECL_DRIVER_API_SYNCHRONOUS_0(bool, isStereoSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isParallelShaderCompileSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isDepthStencilResolveSupported)
+DECL_DRIVER_API_SYNCHRONOUS_N(bool, isDepthStencilBlitSupported, backend::TextureFormat, format)
+DECL_DRIVER_API_SYNCHRONOUS_0(bool, isProtectedTexturesSupported)
+DECL_DRIVER_API_SYNCHRONOUS_0(bool, isDepthClampSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(uint8_t, getMaxDrawBuffers)
 DECL_DRIVER_API_SYNCHRONOUS_0(size_t, getMaxUniformBufferSize)
 DECL_DRIVER_API_SYNCHRONOUS_0(math::float2, getClipSpaceParams)
 DECL_DRIVER_API_SYNCHRONOUS_N(void, setupExternalImage, void*, image)
-DECL_DRIVER_API_SYNCHRONOUS_N(bool, getTimerQueryValue, backend::TimerQueryHandle, query, uint64_t*, elapsedTime)
+DECL_DRIVER_API_SYNCHRONOUS_N(backend::TimerQueryResult, getTimerQueryValue, backend::TimerQueryHandle, query, uint64_t*, elapsedTime)
 DECL_DRIVER_API_SYNCHRONOUS_N(bool, isWorkaroundNeeded, backend::Workaround, workaround)
 DECL_DRIVER_API_SYNCHRONOUS_0(backend::FeatureLevel, getFeatureLevel)
 
@@ -429,6 +434,11 @@ DECL_DRIVER_API_N(bindSamplers,
         uint32_t, index,
         backend::SamplerGroupHandle, sbh)
 
+DECL_DRIVER_API_N(setPushConstant,
+        backend::ShaderStage, stage,
+        uint8_t, index,
+        backend::PushConstantVariant, value)
+
 DECL_DRIVER_API_N(insertEventMarker,
         const char*, string,
         uint32_t, len = 0)
@@ -491,6 +501,17 @@ DECL_DRIVER_API_N(blit,
         math::uint2, srcOrigin,
         math::uint2, size)
 
+DECL_DRIVER_API_N(bindPipeline,
+        backend::PipelineState const&, state)
+
+DECL_DRIVER_API_N(bindRenderPrimitive,
+        backend::RenderPrimitiveHandle, rph)
+
+DECL_DRIVER_API_N(draw2,
+        uint32_t, indexOffset,
+        uint32_t, indexCount,
+        uint32_t, instanceCount)
+
 DECL_DRIVER_API_N(draw,
         backend::PipelineState, state,
         backend::RenderPrimitiveHandle, rph,
@@ -498,6 +519,7 @@ DECL_DRIVER_API_N(draw,
         uint32_t, indexCount,
         uint32_t, instanceCount)
 
+
 DECL_DRIVER_API_N(dispatchCompute,
         backend::ProgramHandle, program,
         math::uint3, workGroupCount)

filament/backend/include/private/backend/HandleAllocator.h

@@ -37,9 +37,9 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#define HandleAllocatorGL  HandleAllocator<24,  64, 136>    // ~3640 / pool / MiB
-#define HandleAllocatorVK  HandleAllocator<80, 176, 320>    // ~1820 / pool / MiB
-#define HandleAllocatorMTL HandleAllocator<48, 160, 592>    // ~1310 / pool / MiB
+#define HandleAllocatorGL  HandleAllocator<32,  64, 136>    // ~4520 / pool / MiB
+#define HandleAllocatorVK  HandleAllocator<64, 160, 312>    // ~1820 / pool / MiB
+#define HandleAllocatorMTL HandleAllocator<32,  64, 552>    // ~1660 / pool / MiB
 
 namespace filament::backend {
 
@@ -49,7 +49,7 @@ namespace filament::backend {
 template<size_t P0, size_t P1, size_t P2>
 class HandleAllocator {
 public:
-    HandleAllocator(const char* name, size_t size) noexcept;
+    HandleAllocator(const char* name, size_t size, bool disableUseAfterFreeCheck) noexcept;
     HandleAllocator(HandleAllocator const& rhs) = delete;
     HandleAllocator& operator=(HandleAllocator const& rhs) = delete;
     ~HandleAllocator();
@@ -65,7 +65,7 @@ public:
      *
      */
     template<typename D, typename ... ARGS>
-    Handle<D> allocateAndConstruct(ARGS&& ... args) noexcept {
+    Handle<D> allocateAndConstruct(ARGS&& ... args) {
         Handle<D> h{ allocateHandle<D>() };
         D* addr = handle_cast<D*>(h);
         new(addr) D(std::forward<ARGS>(args)...);
@@ -97,7 +97,7 @@ public:
      */
     template<typename D, typename B, typename ... ARGS>
     typename std::enable_if_t<std::is_base_of_v<B, D>, D>*
-    destroyAndConstruct(Handle<B> const& handle, ARGS&& ... args) noexcept {
+    destroyAndConstruct(Handle<B> const& handle, ARGS&& ... args) {
         assert_invariant(handle);
         D* addr = handle_cast<D*>(const_cast<Handle<B>&>(handle));
         assert_invariant(addr);
@@ -163,31 +163,44 @@ public:
     inline typename std::enable_if_t<
             std::is_pointer_v<Dp> &&
             std::is_base_of_v<B, typename std::remove_pointer_t<Dp>>, Dp>
-    handle_cast(Handle<B>& handle) noexcept {
+    handle_cast(Handle<B>& handle) {
         assert_invariant(handle);
         auto [p, tag] = handleToPointer(handle.getId());
 
         if (isPoolHandle(handle.getId())) {
             // check for use after free
-            uint8_t const age = (tag & HANDLE_AGE_MASK) >> HANDLE_AGE_SHIFT;
-            auto const pNode = static_cast<typename Allocator::Node*>(p);
-            uint8_t const expectedAge = pNode[-1].age;
-            ASSERT_POSTCONDITION(expectedAge == age,
-                    "use-after-free of Handle with id=%d", handle.getId());
+            if (UTILS_UNLIKELY(!mUseAfterFreeCheckDisabled)) {
+                uint8_t const age = (tag & HANDLE_AGE_MASK) >> HANDLE_AGE_SHIFT;
+                auto const pNode = static_cast<typename Allocator::Node*>(p);
+                uint8_t const expectedAge = pNode[-1].age;
+                FILAMENT_CHECK_POSTCONDITION(expectedAge == age) <<
+                        "use-after-free of Handle with id=" << handle.getId();
+            }
         }
 
         return static_cast<Dp>(p);
     }
 
+    template<typename B>
+    bool is_valid(Handle<B>& handle) {
+        if (handle && isPoolHandle(handle.getId())) {
+            auto [p, tag] = handleToPointer(handle.getId());
+            uint8_t const age = (tag & HANDLE_AGE_MASK) >> HANDLE_AGE_SHIFT;
+            auto const pNode = static_cast<typename Allocator::Node*>(p);
+            uint8_t const expectedAge = pNode[-1].age;
+            return expectedAge == age;
+        }
+        return true;
+    }
+
     template<typename Dp, typename B>
     inline typename std::enable_if_t<
             std::is_pointer_v<Dp> &&
             std::is_base_of_v<B, typename std::remove_pointer_t<Dp>>, Dp>
-    handle_cast(Handle<B> const& handle) noexcept {
+    handle_cast(Handle<B> const& handle) {
         return handle_cast<Dp>(const_cast<Handle<B>&>(handle));
     }
 
-
 private:
 
     template<typename D>
@@ -210,8 +223,9 @@ private:
         Pool<P1> mPool1;
         Pool<P2> mPool2;
         UTILS_UNUSED_IN_RELEASE const utils::AreaPolicy::HeapArea& mArea;
+        bool mUseAfterFreeCheckDisabled;
     public:
-        explicit Allocator(const utils::AreaPolicy::HeapArea& area);
+        explicit Allocator(const utils::AreaPolicy::HeapArea& area, bool disableUseAfterFreeCheck);
 
         static constexpr size_t getAlignment() noexcept { return MIN_ALIGNMENT; }
 
@@ -237,8 +251,10 @@ private:
             // check for double-free
             Node* const pNode = static_cast<Node*>(p);
             uint8_t& expectedAge = pNode[-1].age;
-            ASSERT_POSTCONDITION(expectedAge == age,
-                    "double-free of Handle of size %d at %p", size, p);
+            if (UTILS_UNLIKELY(!mUseAfterFreeCheckDisabled)) {
+                FILAMENT_CHECK_POSTCONDITION(expectedAge == age) <<
+                        "double-free of Handle of size " << size << " at " << p;
+            }
             expectedAge = (expectedAge + 1) & 0xF; // fixme
 
             if (size <= mPool0.getSize()) { mPool0.free(p); return; }
@@ -313,7 +329,7 @@ private:
         return (id & HANDLE_HEAP_FLAG) == 0u;
     }
 
-    HandleBase::HandleId allocateHandleSlow(size_t size) noexcept;
+    HandleBase::HandleId allocateHandleSlow(size_t size);
     void deallocateHandleSlow(HandleBase::HandleId id, size_t size) noexcept;
 
     // We inline this because it's just 4 instructions in the fast case
@@ -348,6 +364,7 @@ private:
     mutable utils::Mutex mLock;
     tsl::robin_map<HandleBase::HandleId, void*> mOverflowMap;
     HandleBase::HandleId mId = 0;
+    bool mUseAfterFreeCheckDisabled = false;
 };
 
 } // namespace filament::backend

filament/backend/src/Callable.cpp

@@ -27,8 +27,8 @@ PresentCallable::PresentCallable(PresentFn fn, void* user) noexcept
 }
 
 void PresentCallable::operator()(bool presentFrame) noexcept {
-    ASSERT_PRECONDITION(mPresentFn, "This PresentCallable was already called. " \
-            "PresentCallables should be called exactly once.");
+    FILAMENT_CHECK_PRECONDITION(mPresentFn) << "This PresentCallable was already called. "
+                                               "PresentCallables should be called exactly once.";
     mPresentFn(presentFrame, mUser);
     // Set mPresentFn to nullptr to denote that the callable has been called.
     mPresentFn = nullptr;

filament/backend/src/CircularBuffer.cpp

@@ -32,10 +32,11 @@
 #    define HAS_MMAP 0
 #endif
 
-#include <stdint.h>
 #include <stddef.h>
-#include <stdlib.h>
+#include <stdint.h>
 #include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 using namespace utils;
 
@@ -81,6 +82,9 @@ void* CircularBuffer::alloc(size_t size) noexcept {
             // map the circular buffer once...
             vaddr = mmap(reserve_vaddr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
             if (vaddr != MAP_FAILED) {
+                // populate the address space with pages (because this is a circular buffer,
+                // all the pages will be allocated eventually, might as well do it now)
+                memset(vaddr, 0, size);
                 // and map the circular buffer again, behind the previous copy...
                 vaddr_shadow = mmap((char*)vaddr + size, size,
                         PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
@@ -101,7 +105,7 @@ void* CircularBuffer::alloc(size_t size) noexcept {
     if (UTILS_UNLIKELY(mAshmemFd < 0)) {
         // ashmem failed
         if (vaddr_guard != MAP_FAILED) {
-            munmap(vaddr_guard, size);
+            munmap(vaddr_guard, BLOCK_SIZE);
         }
 
         if (vaddr_shadow != MAP_FAILED) {
@@ -119,12 +123,11 @@ void* CircularBuffer::alloc(size_t size) noexcept {
         data = mmap(nullptr, size * 2 + BLOCK_SIZE,
                 PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 
-        ASSERT_POSTCONDITION(data,
-                "couldn't allocate %u KiB of virtual address space for the command buffer",
-                (size * 2 / 1024));
+        FILAMENT_CHECK_POSTCONDITION(data != MAP_FAILED) <<
+                "couldn't allocate " << (size * 2 / 1024) <<
+                " KiB of virtual address space for the command buffer";
 
-        slog.d << "WARNING: Using soft CircularBuffer (" << (size * 2 / 1024) << " KiB)"
-               << io::endl;
+        slog.w << "Using 'soft' CircularBuffer (" << (size * 2 / 1024) << " KiB)" << io::endl;
 
         // guard page at the end
         void* guard = (void*)(uintptr_t(data) + size * 2);

filament/backend/src/CommandBufferQueue.cpp

@@ -39,10 +39,11 @@ using namespace utils;
 
 namespace filament::backend {
 
-CommandBufferQueue::CommandBufferQueue(size_t requiredSize, size_t bufferSize)
+CommandBufferQueue::CommandBufferQueue(size_t requiredSize, size_t bufferSize, bool paused)
         : mRequiredSize((requiredSize + (CircularBuffer::getBlockSize() - 1u)) & ~(CircularBuffer::getBlockSize() -1u)),
           mCircularBuffer(bufferSize),
-          mFreeSpace(mCircularBuffer.size()) {
+          mFreeSpace(mCircularBuffer.size()),
+          mPaused(paused) {
     assert_invariant(mCircularBuffer.size() > requiredSize);
 }
 
@@ -56,10 +57,23 @@ void CommandBufferQueue::requestExit() {
     mCondition.notify_one();
 }
 
+bool CommandBufferQueue::isPaused() const noexcept {
+    std::lock_guard<utils::Mutex> const lock(mLock);
+    return mPaused;
+}
+
+void CommandBufferQueue::setPaused(bool paused) {
+    std::lock_guard<utils::Mutex> const lock(mLock);
+    if (paused) {
+        mPaused = true;
+    } else {
+        mPaused = false;
+        mCondition.notify_one();
+    }
+}
+
 bool CommandBufferQueue::isExitRequested() const {
     std::lock_guard<utils::Mutex> const lock(mLock);
-    ASSERT_PRECONDITION( mExitRequested == 0 || mExitRequested == EXIT_REQUESTED,
-            "mExitRequested is corrupted (value = 0x%08x)!", mExitRequested);
     return (bool)mExitRequested;
 }
 
@@ -87,22 +101,23 @@ void CommandBufferQueue::flush() noexcept {
     size_t const used = std::distance(
             static_cast<char const*>(begin), static_cast<char const*>(end));
 
+
     std::unique_lock<utils::Mutex> lock(mLock);
+
+    // circular buffer is too small, we corrupted the stream
+    FILAMENT_CHECK_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: " << used <<
+            " bytes, overflow: " << used - mFreeSpace << " bytes";
+
+    mFreeSpace -= used;
     mCommandBuffersToExecute.push_back({ begin, end });
     mCondition.notify_one();
 
-    // 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));
-
     // wait until there is enough space in the buffer
-    mFreeSpace -= used;
     if (UTILS_UNLIKELY(mFreeSpace < requiredSize)) {
 
-
 #ifndef NDEBUG
         size_t const totalUsed = circularBuffer.size() - mFreeSpace;
         slog.d << "CommandStream used too much space (will block): "
@@ -115,6 +130,11 @@ void CommandBufferQueue::flush() noexcept {
 #endif
 
         SYSTRACE_NAME("waiting: CircularBuffer::flush()");
+
+        FILAMENT_CHECK_POSTCONDITION(!mPaused) <<
+                "CommandStream is full, but since the rendering thread is paused, "
+                "the buffer cannot flush and we will deadlock. Instead, abort.";
+
         mCondition.wait(lock, [this, requiredSize]() -> bool {
             // TODO: on macOS, we need to call pumpEvents from time to time
             return mFreeSpace >= requiredSize;
@@ -127,19 +147,17 @@ std::vector<CommandBufferQueue::Range> CommandBufferQueue::waitForCommands() con
         return std::move(mCommandBuffersToExecute);
     }
     std::unique_lock<utils::Mutex> lock(mLock);
-    while (mCommandBuffersToExecute.empty() && !mExitRequested) {
+    while ((mCommandBuffersToExecute.empty() || mPaused) && !mExitRequested) {
         mCondition.wait(lock);
     }
-
-    ASSERT_PRECONDITION( mExitRequested == 0 || mExitRequested == EXIT_REQUESTED,
-            "mExitRequested is corrupted (value = 0x%08x)!", mExitRequested);
-
     return std::move(mCommandBuffersToExecute);
 }
 
 void CommandBufferQueue::releaseBuffer(CommandBufferQueue::Range const& buffer) {
+    size_t const used = std::distance(
+            static_cast<char const*>(buffer.begin), static_cast<char const*>(buffer.end));
     std::lock_guard<utils::Mutex> const lock(mLock);
-    mFreeSpace += uintptr_t(buffer.end) - uintptr_t(buffer.begin);
+    mFreeSpace += used;
     mCondition.notify_one();
 }
 

filament/backend/src/CommandStream.cpp

@@ -149,7 +149,7 @@ void CommandType<void (Driver::*)(ARGS...)>::Command<METHOD>::log() noexcept  {
 
 // ------------------------------------------------------------------------------------------------
 
-void CustomCommand::execute(Driver&, CommandBase* base, intptr_t* next) noexcept {
+void CustomCommand::execute(Driver&, CommandBase* base, intptr_t* next) {
     *next = CustomCommand::align(sizeof(CustomCommand));
     static_cast<CustomCommand*>(base)->mCommand();
     static_cast<CustomCommand*>(base)->~CustomCommand();

filament/backend/src/Driver.cpp

@@ -214,7 +214,7 @@ size_t Driver::getElementTypeSize(ElementType type) noexcept {
 
 Driver::~Driver() noexcept = default;
 
-void Driver::execute(std::function<void(void)> const& fn) noexcept {
+void Driver::execute(std::function<void(void)> const& fn) {
     fn();
 }
 

filament/backend/src/DriverBase.h

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

filament/backend/src/HandleAllocator.cpp

@@ -39,8 +39,10 @@ using namespace utils;
 
 template <size_t P0, size_t P1, size_t P2>
 UTILS_NOINLINE
-HandleAllocator<P0, P1, P2>::Allocator::Allocator(AreaPolicy::HeapArea const& area)
-        : mArea(area) {
+HandleAllocator<P0, P1, P2>::Allocator::Allocator(AreaPolicy::HeapArea const& area,
+        bool disableUseAfterFreeCheck)
+        : mArea(area),
+          mUseAfterFreeCheckDisabled(disableUseAfterFreeCheck) {
 
     // The largest handle this allocator can generate currently depends on the architecture's
     // min alignment, typically 8 or 16 bytes.
@@ -74,8 +76,10 @@ HandleAllocator<P0, P1, P2>::Allocator::Allocator(AreaPolicy::HeapArea const& ar
 // ------------------------------------------------------------------------------------------------
 
 template <size_t P0, size_t P1, size_t P2>
-HandleAllocator<P0, P1, P2>::HandleAllocator(const char* name, size_t size) noexcept
-    : mHandleArena(name, size) {
+HandleAllocator<P0, P1, P2>::HandleAllocator(const char* name, size_t size,
+        bool disableUseAfterFreeCheck) noexcept
+    : mHandleArena(name, size, disableUseAfterFreeCheck),
+      mUseAfterFreeCheckDisabled(disableUseAfterFreeCheck) {
 }
 
 template <size_t P0, size_t P1, size_t P2>
@@ -103,15 +107,15 @@ void* HandleAllocator<P0, P1, P2>::handleToPointerSlow(HandleBase::HandleId id)
 }
 
 template <size_t P0, size_t P1, size_t P2>
-HandleBase::HandleId HandleAllocator<P0, P1, P2>::allocateHandleSlow(size_t size) noexcept {
+HandleBase::HandleId HandleAllocator<P0, P1, P2>::allocateHandleSlow(size_t size) {
     void* p = ::malloc(size);
     std::unique_lock lock(mLock);
 
     HandleBase::HandleId id = (++mId) | HANDLE_HEAP_FLAG;
 
-    ASSERT_POSTCONDITION(mId < HANDLE_HEAP_FLAG,
+    FILAMENT_CHECK_POSTCONDITION(mId < HANDLE_HEAP_FLAG) <<
             "No more Handle ids available! This can happen if HandleAllocator arena has been full"
-            " for a while. Please increase FILAMENT_OPENGL_HANDLE_ARENA_SIZE_IN_MB");
+            " for a while. Please increase FILAMENT_OPENGL_HANDLE_ARENA_SIZE_IN_MB";
 
     mOverflowMap.emplace(id, p);
     lock.unlock();

filament/backend/src/Platform.cpp

@@ -53,4 +53,18 @@ size_t Platform::retrieveBlob(void const* key, size_t keySize, void* value, size
     return 0;
 }
 
+void Platform::setDebugUpdateStatFunc(DebugUpdateStatFunc&& debugUpdateStat) noexcept {
+    mDebugUpdateStat = std::move(debugUpdateStat);
+}
+
+bool Platform::hasDebugUpdateStatFunc() const noexcept {
+    return bool(mDebugUpdateStat);
+}
+
+void Platform::debugUpdateStat(const char* key, uint64_t value) {
+    if (mDebugUpdateStat) {
+        mDebugUpdateStat(key, value);
+    }
+}
+
 } // namespace filament::backend

filament/backend/src/PlatformFactory.cpp

@@ -29,21 +29,21 @@
         #include "backend/platforms/PlatformCocoaTouchGL.h"
     #endif
 #elif defined(__APPLE__)
-    #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3) && !defined(FILAMENT_USE_SWIFTSHADER)
+    #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3)
         #include <backend/platforms/PlatformCocoaGL.h>
     #endif
 #elif defined(__linux__)
     #if defined(FILAMENT_SUPPORTS_X11)
-        #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3) && !defined(FILAMENT_USE_SWIFTSHADER)
+        #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3)
             #include "backend/platforms/PlatformGLX.h"
         #endif
     #elif defined(FILAMENT_SUPPORTS_EGL_ON_LINUX)
-        #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3) && !defined(FILAMENT_USE_SWIFTSHADER)
+        #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3)
             #include "backend/platforms/PlatformEGLHeadless.h"
         #endif
     #endif
 #elif defined(WIN32)
-    #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3) && !defined(FILAMENT_USE_SWIFTSHADER)
+    #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3)
         #include "backend/platforms/PlatformWGL.h"
     #endif
 #elif defined(__EMSCRIPTEN__)
@@ -111,8 +111,7 @@ Platform* PlatformFactory::create(Backend* backend) noexcept {
     }
     assert_invariant(*backend == Backend::OPENGL);
     #if defined(FILAMENT_SUPPORTS_OPENGL)
-        #if defined(FILAMENT_USE_EXTERNAL_GLES3) || defined(FILAMENT_USE_SWIFTSHADER)
-            // Swiftshader OpenGLES support is deprecated and incomplete
+        #if defined(FILAMENT_USE_EXTERNAL_GLES3)
             return nullptr;
         #elif defined(__ANDROID__)
             return new PlatformEGLAndroid();

filament/backend/src/Program.cpp

@@ -21,7 +21,7 @@ namespace filament::backend {
 using namespace utils;
 
 // We want these in the .cpp file, so they're not inlined (not worth it)
-Program::Program() noexcept { // NOLINT(modernize-use-equals-default)
+Program::Program() noexcept {  // NOLINT(modernize-use-equals-default)
 }
 
 Program::Program(Program&& rhs) noexcept = default;
@@ -47,6 +47,11 @@ Program& Program::shader(ShaderStage shader, void const* data, size_t size) {
     return *this;
 }
 
+Program& Program::shaderLanguage(ShaderLanguage shaderLanguage) {
+    mShaderLanguage = shaderLanguage;
+    return *this;
+}
+
 Program& Program::uniformBlockBindings(
         FixedCapacityVector<std::pair<utils::CString, uint8_t>> const& uniformBlockBindings) noexcept {
     for (auto const& item : uniformBlockBindings) {
@@ -86,11 +91,22 @@ Program& Program::specializationConstants(
     return *this;
 }
 
+Program& Program::pushConstants(ShaderStage stage,
+        utils::FixedCapacityVector<PushConstant> constants) noexcept {
+    mPushConstants[static_cast<uint8_t>(stage)] = std::move(constants);
+    return *this;
+}
+
 Program& Program::cacheId(uint64_t cacheId) noexcept {
     mCacheId = cacheId;
     return *this;
 }
 
+Program& Program::multiview(bool multiview) noexcept {
+    mMultiview = multiview;
+    return *this;
+}
+
 io::ostream& operator<<(io::ostream& out, const Program& builder) {
     out << "Program{";
     builder.mLogger(out);

filament/backend/src/VirtualMachineEnv.cpp

@@ -16,13 +16,23 @@
 
 #include "private/backend/VirtualMachineEnv.h"
 
+#include <utils/compiler.h>
 #include <utils/debug.h>
 
+#include <jni.h>
+
 namespace filament {
 
 JavaVM* VirtualMachineEnv::sVirtualMachine = nullptr;
 
-// This is called when the library is loaded. We need this to get a reference to the global VM
+/*
+ * This is typically called by filament_jni.so when it is loaded. If filament_jni.so is not used,
+ * then this must be called manually -- however, this is a problem because VirtualMachineEnv.h
+ * is currently private and part of backend.
+ * For now, we authorize this usage, but we will need to fix it; by making a proper public
+ * API for this.
+ */
+UTILS_PUBLIC
 UTILS_NOINLINE
 jint VirtualMachineEnv::JNI_OnLoad(JavaVM* vm) noexcept {
     JNIEnv* env = nullptr;

filament/backend/src/metal/MetalBlitter.mm

@@ -109,9 +109,8 @@ inline bool MTLSizeEqual(T a, T b) noexcept {
 MetalBlitter::MetalBlitter(MetalContext& context) noexcept : mContext(context) { }
 
 void MetalBlitter::blit(id<MTLCommandBuffer> cmdBuffer, const BlitArgs& args, const char* label) {
-
-    ASSERT_PRECONDITION(args.source.region.size.depth == args.destination.region.size.depth,
-            "Blitting requires the source and destination regions to have the same depth.");
+    FILAMENT_CHECK_PRECONDITION(args.source.region.size.depth == args.destination.region.size.depth)
+            << "Blitting requires the source and destination regions to have the same depth.";
 
     // Determine if the blit for color or depth are eligible to use a MTLBlitCommandEncoder.
     // blitFastPath returns true upon success.
@@ -327,7 +326,8 @@ id<MTLFunction> MetalBlitter::compileFragmentFunction(BlitFunctionKey key) const
             utils::slog.e << description << utils::io::endl;
         }
     }
-    ASSERT_POSTCONDITION(library && function, "Unable to compile fragment shader for MetalBlitter.");
+    FILAMENT_CHECK_POSTCONDITION(library && function)
+            << "Unable to compile fragment shader for MetalBlitter.";
 
     return function;
 }
@@ -352,7 +352,8 @@ id<MTLFunction> MetalBlitter::getBlitVertexFunction() {
             utils::slog.e << description << utils::io::endl;
         }
     }
-    ASSERT_POSTCONDITION(library && function, "Unable to compile vertex shader for MetalBlitter.");
+    FILAMENT_CHECK_POSTCONDITION(library && function)
+            << "Unable to compile vertex shader for MetalBlitter.";
 
     mVertexFunction = function;
 

filament/backend/src/metal/MetalBuffer.h

@@ -18,7 +18,7 @@
 #define TNT_FILAMENT_DRIVER_METALBUFFER_H
 
 #include "MetalContext.h"
-#include "MetalBufferPool.h"
+#include "MetalPlatform.h"
 
 #include <backend/DriverEnums.h>
 
@@ -28,9 +28,131 @@
 
 #include <utility>
 #include <memory>
+#include <atomic>
+#include <chrono>
 
 namespace filament::backend {
 
+class ScopedAllocationTimer {
+public:
+    ScopedAllocationTimer(const char* name) : mBeginning(clock_t::now()), mName(name) {}
+    ~ScopedAllocationTimer() {
+        using namespace std::literals::chrono_literals;
+        static constexpr std::chrono::seconds LONG_TIME_THRESHOLD = 10s;
+
+        auto end = clock_t::now();
+        std::chrono::duration<double, std::micro> allocationTimeMicroseconds = end - mBeginning;
+
+        if (UTILS_UNLIKELY(allocationTimeMicroseconds > LONG_TIME_THRESHOLD)) {
+            if (platform && platform->hasDebugUpdateStatFunc()) {
+                char buffer[64];
+                snprintf(buffer, sizeof(buffer), "filament.metal.long_buffer_allocation_time.%s",
+                        mName);
+                platform->debugUpdateStat(
+                        buffer, static_cast<uint64_t>(allocationTimeMicroseconds.count()));
+            }
+        }
+    }
+
+    static void setPlatform(MetalPlatform* p) { platform = p; }
+
+private:
+    typedef std::chrono::steady_clock clock_t;
+
+    static MetalPlatform* platform;
+
+    std::chrono::time_point<clock_t> mBeginning;
+    const char* mName;
+};
+
+class TrackedMetalBuffer {
+public:
+
+    static constexpr size_t EXCESS_BUFFER_COUNT = 30000;
+
+    enum class Type {
+        NONE = 0,
+        GENERIC = 1,
+        RING = 2,
+        STAGING = 3,
+    };
+    static constexpr size_t TypeCount = 3;
+
+    static constexpr auto toIndex(Type t) {
+        assert_invariant(t != Type::NONE);
+        switch (t) {
+            case Type::NONE:
+            case Type::GENERIC:
+                return 0;
+            case Type::RING:
+                return 1;
+            case Type::STAGING:
+                return 2;
+        }
+    }
+
+    TrackedMetalBuffer() noexcept : mBuffer(nil) {}
+    TrackedMetalBuffer(nullptr_t) noexcept : mBuffer(nil) {}
+    TrackedMetalBuffer(id<MTLBuffer> buffer, Type type) : mBuffer(buffer), mType(type) {
+        assert_invariant(type != Type::NONE);
+        if (buffer) {
+            aliveBuffers[toIndex(type)]++;
+            mType = type;
+            if (getAliveBuffers() >= EXCESS_BUFFER_COUNT) {
+                if (platform && platform->hasDebugUpdateStatFunc()) {
+                    platform->debugUpdateStat("filament.metal.excess_buffers_allocated",
+                            TrackedMetalBuffer::getAliveBuffers());
+                }
+            }
+        }
+    }
+
+    ~TrackedMetalBuffer() {
+        if (mBuffer) {
+            assert_invariant(mType != Type::NONE);
+            aliveBuffers[toIndex(mType)]--;
+        }
+    }
+
+    TrackedMetalBuffer(TrackedMetalBuffer&&) = delete;
+    TrackedMetalBuffer(TrackedMetalBuffer const&) = delete;
+    TrackedMetalBuffer& operator=(TrackedMetalBuffer const&) = delete;
+
+    TrackedMetalBuffer& operator=(TrackedMetalBuffer&& rhs) noexcept {
+        swap(rhs);
+        return *this;
+    }
+
+    id<MTLBuffer> get() const noexcept { return mBuffer; }
+    operator bool() const noexcept { return bool(mBuffer); }
+
+    static uint64_t getAliveBuffers() {
+        uint64_t sum = 0;
+        for (const auto& v : aliveBuffers) {
+            sum += v;
+        }
+        return sum;
+    }
+
+    static uint64_t getAliveBuffers(Type type) {
+        assert_invariant(type != Type::NONE);
+        return aliveBuffers[toIndex(type)];
+    }
+    static void setPlatform(MetalPlatform* p) { platform = p; }
+
+private:
+    void swap(TrackedMetalBuffer& other) noexcept {
+        std::swap(mBuffer, other.mBuffer);
+        std::swap(mType, other.mType);
+    }
+
+    id<MTLBuffer> mBuffer;
+    Type mType = Type::NONE;
+
+    static MetalPlatform* platform;
+    static std::array<uint64_t, TypeCount> aliveBuffers;
+};
+
 class MetalBuffer {
 public:
 
@@ -82,7 +204,16 @@ public:
 
 private:
 
-    id<MTLBuffer> mBuffer = nil;
+    enum class UploadStrategy {
+        POOL,
+        BUMP_ALLOCATOR,
+    };
+
+    void uploadWithPoolBuffer(void* src, size_t size, size_t byteOffset) const;
+    void uploadWithBumpAllocator(void* src, size_t size, size_t byteOffset) const;
+
+    UploadStrategy mUploadStrategy;
+    TrackedMetalBuffer mBuffer;
     size_t mBufferSize = 0;
     void* mCpuBuffer = nullptr;
     MetalContext& mContext;
@@ -131,7 +262,9 @@ public:
           mBufferOptions(options),
           mSlotSizeBytes(computeSlotSize(layout)),
           mSlotCount(slotCount) {
-        mBuffer = [device newBufferWithLength:mSlotSizeBytes * mSlotCount options:mBufferOptions];
+        ScopedAllocationTimer timer("ring");
+        mBuffer = { [device newBufferWithLength:mSlotSizeBytes * mSlotCount options:mBufferOptions],
+            TrackedMetalBuffer::Type::RING };
         assert_invariant(mBuffer);
     }
 
@@ -149,9 +282,13 @@ public:
             // If we already have an aux buffer, it will get freed here, unless it has been retained
             // by a MTLCommandBuffer. In that case, it will be freed when the command buffer
             // finishes executing.
-            mAuxBuffer = [mDevice newBufferWithLength:mSlotSizeBytes options:mBufferOptions];
+            {
+                ScopedAllocationTimer timer("ring");
+                mAuxBuffer = { [mDevice newBufferWithLength:mSlotSizeBytes options:mBufferOptions],
+                    TrackedMetalBuffer::Type::RING };
+            }
             assert_invariant(mAuxBuffer);
-            return {mAuxBuffer, 0};
+            return { mAuxBuffer.get(), 0 };
         }
         mCurrentSlot = (mCurrentSlot + 1) % mSlotCount;
         mOccupiedSlots->fetch_add(1, std::memory_order_relaxed);
@@ -180,9 +317,9 @@ public:
      */
     std::pair<id<MTLBuffer>, NSUInteger> getCurrentAllocation() const {
         if (UTILS_UNLIKELY(mAuxBuffer)) {
-            return { mAuxBuffer, 0 };
+            return { mAuxBuffer.get(), 0 };
         }
-        return { mBuffer, mCurrentSlot * mSlotSizeBytes };
+        return { mBuffer.get(), mCurrentSlot * mSlotSizeBytes };
     }
 
     bool canAccomodateLayout(MTLSizeAndAlign layout) const {
@@ -191,8 +328,8 @@ public:
 
 private:
     id<MTLDevice> mDevice;
-    id<MTLBuffer> mBuffer;
-    id<MTLBuffer> mAuxBuffer;
+    TrackedMetalBuffer mBuffer;
+    TrackedMetalBuffer mAuxBuffer;
 
     MTLResourceOptions mBufferOptions;
 

filament/backend/src/metal/MetalBuffer.mm

@@ -15,14 +15,28 @@
  */
 
 #include "MetalBuffer.h"
+#include "MetalBufferPool.h"
 
 #include "MetalContext.h"
 
 namespace filament {
 namespace backend {
 
+std::array<uint64_t, TrackedMetalBuffer::TypeCount> TrackedMetalBuffer::aliveBuffers = { 0 };
+MetalPlatform* TrackedMetalBuffer::platform = nullptr;
+MetalPlatform* ScopedAllocationTimer::platform = nullptr;
+
 MetalBuffer::MetalBuffer(MetalContext& context, BufferObjectBinding bindingType, BufferUsage usage,
-        size_t size, bool forceGpuBuffer) : mBufferSize(size), mContext(context) {
+        size_t size, bool forceGpuBuffer)
+    : mBufferSize(size), mContext(context) {
+    const MetalBumpAllocator& allocator = *mContext.bumpAllocator;
+    // VERTEX is also used for index buffers
+    if (allocator.getCapacity() > 0 && bindingType == BufferObjectBinding::VERTEX) {
+        mUploadStrategy = UploadStrategy::BUMP_ALLOCATOR;
+    } else {
+        mUploadStrategy = UploadStrategy::POOL;
+    }
+
     // If the buffer is less than 4K in size and is updated frequently, we don't use an explicit
     // buffer. Instead, we use immediate command encoder methods like setVertexBytes:length:atIndex:.
     // This won't work for SSBOs, since they are read/write.
@@ -34,8 +48,13 @@ MetalBuffer::MetalBuffer(MetalContext& context, BufferObjectBinding bindingType,
     }
 
     // Otherwise, we allocate a private GPU buffer.
-    mBuffer = [context.device newBufferWithLength:size options:MTLResourceStorageModePrivate];
-    ASSERT_POSTCONDITION(mBuffer, "Could not allocate Metal buffer of size %zu.", size);
+    {
+        ScopedAllocationTimer timer("generic");
+        mBuffer = { [context.device newBufferWithLength:size options:MTLResourceStorageModePrivate],
+            TrackedMetalBuffer::Type::GENERIC };
+    }
+    FILAMENT_CHECK_POSTCONDITION(mBuffer)
+            << "Could not allocate Metal buffer of size " << size << ".";
 }
 
 MetalBuffer::~MetalBuffer() {
@@ -48,37 +67,26 @@ void MetalBuffer::copyIntoBuffer(void* src, size_t size, size_t byteOffset) {
     if (size <= 0) {
         return;
     }
-    ASSERT_PRECONDITION(size + byteOffset <= mBufferSize,
-            "Attempting to copy %zu bytes into a buffer of size %zu at offset %zu",
-            size, mBufferSize, byteOffset);
+    FILAMENT_CHECK_PRECONDITION(size + byteOffset <= mBufferSize)
+            << "Attempting to copy " << size << " bytes into a buffer of size " << mBufferSize
+            << " at offset " << byteOffset;
+    // The copy blit requires that byteOffset be a multiple of 4.
+    FILAMENT_CHECK_PRECONDITION(!(byteOffset & 0x3)) << "byteOffset must be a multiple of 4";
 
-    // Either copy into the Metal buffer or into our cpu buffer.
+    // If we have a cpu buffer, we can directly copy into it.
     if (mCpuBuffer) {
         memcpy(static_cast<uint8_t*>(mCpuBuffer) + byteOffset, src, size);
         return;
     }
 
-    // Acquire a staging buffer to hold the contents of this update.
-    MetalBufferPool* bufferPool = mContext.bufferPool;
-    const MetalBufferPoolEntry* const staging = bufferPool->acquireBuffer(size);
-    memcpy(staging->buffer.contents, src, size);
-
-    // The blit below requires that byteOffset be a multiple of 4.
-    ASSERT_PRECONDITION(!(byteOffset & 0x3u), "byteOffset must be a multiple of 4");
-
-    // Encode a blit from the staging buffer into the private GPU buffer.
-    id<MTLCommandBuffer> cmdBuffer = getPendingCommandBuffer(&mContext);
-    id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
-    blitEncoder.label = @"Buffer upload blit";
-    [blitEncoder copyFromBuffer:staging->buffer
-                   sourceOffset:0
-                       toBuffer:mBuffer
-              destinationOffset:byteOffset
-                           size:size];
-    [blitEncoder endEncoding];
-    [cmdBuffer addCompletedHandler:^(id<MTLCommandBuffer> cb) {
-        bufferPool->releaseBuffer(staging);
-    }];
+    switch (mUploadStrategy) {
+        case UploadStrategy::BUMP_ALLOCATOR:
+            uploadWithBumpAllocator(src, size, byteOffset);
+            break;
+        case UploadStrategy::POOL:
+            uploadWithPoolBuffer(src, size, byteOffset);
+            break;
+    }
 }
 
 void MetalBuffer::copyIntoBufferUnsynchronized(void* src, size_t size, size_t byteOffset) {
@@ -93,7 +101,7 @@ id<MTLBuffer> MetalBuffer::getGpuBufferForDraw(id<MTLCommandBuffer> cmdBuffer) n
         return nil;
     }
     assert_invariant(mBuffer);
-    return mBuffer;
+    return mBuffer.get();
 }
 
 void MetalBuffer::bindBuffers(id<MTLCommandBuffer> cmdBuffer, id<MTLCommandEncoder> encoder,
@@ -189,5 +197,42 @@ void MetalBuffer::bindBuffers(id<MTLCommandBuffer> cmdBuffer, id<MTLCommandEncod
     }
 }
 
+void MetalBuffer::uploadWithPoolBuffer(void* src, size_t size, size_t byteOffset) const {
+    MetalBufferPool* bufferPool = mContext.bufferPool;
+    const MetalBufferPoolEntry* const staging = bufferPool->acquireBuffer(size);
+    memcpy(staging->buffer.get().contents, src, size);
+
+    // Encode a blit from the staging buffer into the private GPU buffer.
+    id<MTLCommandBuffer> cmdBuffer = getPendingCommandBuffer(&mContext);
+    id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
+    blitEncoder.label = @"Buffer upload blit - pool buffer";
+    [blitEncoder copyFromBuffer:staging->buffer.get()
+                   sourceOffset:0
+                       toBuffer:mBuffer.get()
+              destinationOffset:byteOffset
+                           size:size];
+    [blitEncoder endEncoding];
+    [cmdBuffer addCompletedHandler:^(id<MTLCommandBuffer> cb) {
+        bufferPool->releaseBuffer(staging);
+    }];
+}
+
+void MetalBuffer::uploadWithBumpAllocator(void* src, size_t size, size_t byteOffset) const {
+    MetalBumpAllocator& allocator = *mContext.bumpAllocator;
+    auto [buffer, offset] = allocator.allocateStagingArea(size);
+    memcpy(static_cast<char*>(buffer.contents) + offset, src, size);
+
+    // Encode a blit from the staging buffer into the private GPU buffer.
+    id<MTLCommandBuffer> cmdBuffer = getPendingCommandBuffer(&mContext);
+    id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
+    blitEncoder.label = @"Buffer upload blit - bump allocator";
+    [blitEncoder copyFromBuffer:buffer
+                   sourceOffset:offset
+                       toBuffer:mBuffer.get()
+              destinationOffset:byteOffset
+                           size:size];
+    [blitEncoder endEncoding];
+}
+
 } // namespace backend
 } // namespace filament

filament/backend/src/metal/MetalBufferPool.h

@@ -19,6 +19,8 @@
 
 #include <Metal/Metal.h>
 
+#include "MetalBuffer.h"
+
 #include <map>
 #include <mutex>
 #include <unordered_set>
@@ -30,12 +32,34 @@ struct MetalContext;
 
 // Immutable POD representing a shared CPU-GPU buffer.
 struct MetalBufferPoolEntry {
-    id<MTLBuffer> buffer;
+    TrackedMetalBuffer buffer;
     size_t capacity;
     mutable uint64_t lastAccessed;
     mutable uint32_t referenceCount;
 };
 
+class MetalBumpAllocator {
+public:
+    MetalBumpAllocator(id<MTLDevice> device, size_t capacity);
+
+    /**
+     * Allocates a staging area of the given size. Returns a pair of the buffer and the offset
+     * within the buffer. The buffer is guaranteed to be at least the given size, but may be larger.
+     * Clients must not write to the buffer beyond the returned offset + size.
+     * Clients are responsible for holding a reference to the returned buffer.
+     * Allocations are guaranteed to be aligned to 4 bytes.
+     */
+    std::pair<id<MTLBuffer>, size_t> allocateStagingArea(size_t size);
+
+    size_t getCapacity() const noexcept { return mCapacity; }
+
+private:
+    id<MTLDevice> mDevice;
+    TrackedMetalBuffer mCurrentUploadBuffer = nil;
+    size_t mHead = 0;
+    size_t mCapacity;
+};
+
 // Manages a pool of Metal buffers, periodically releasing ones that have been unused for awhile.
 class MetalBufferPool {
 public:

filament/backend/src/metal/MetalBufferPool.mm

@@ -42,15 +42,20 @@ MetalBufferPoolEntry const* MetalBufferPool::acquireBuffer(size_t numBytes) {
     }
 
     // We were not able to find a sufficiently large stage, so create a new one.
-    id<MTLBuffer> buffer = [mContext.device newBufferWithLength:numBytes
-                                                        options:MTLResourceStorageModeShared];
-    ASSERT_POSTCONDITION(buffer, "Could not allocate Metal staging buffer of size %zu.", numBytes);
-    MetalBufferPoolEntry* stage = new MetalBufferPoolEntry({
-        .buffer = buffer,
+    id<MTLBuffer> buffer = nil;
+    {
+        ScopedAllocationTimer timer("staging");
+        buffer = [mContext.device newBufferWithLength:numBytes
+                                              options:MTLResourceStorageModeShared];
+    }
+    FILAMENT_CHECK_POSTCONDITION(buffer)
+            << "Could not allocate Metal staging buffer of size " << numBytes << ".";
+    MetalBufferPoolEntry* stage = new MetalBufferPoolEntry {
+        .buffer = { buffer, TrackedMetalBuffer::Type::STAGING },
         .capacity = numBytes,
         .lastAccessed = mCurrentFrame,
         .referenceCount = 1
-    });
+    };
     mUsedStages.insert(stage);
 
     return stage;
@@ -111,5 +116,39 @@ void MetalBufferPool::reset() noexcept {
     mFreeStages.clear();
 }
 
+MetalBumpAllocator::MetalBumpAllocator(id<MTLDevice> device, size_t capacity)
+    : mDevice(device), mCapacity(capacity) {
+    if (mCapacity > 0) {
+        mCurrentUploadBuffer = { [device newBufferWithLength:capacity options:MTLStorageModeShared],
+            TrackedMetalBuffer::Type::STAGING };
+    }
+}
+
+std::pair<id<MTLBuffer>, size_t> MetalBumpAllocator::allocateStagingArea(size_t size) {
+    if (size == 0) {
+        return { nil, 0 };
+    }
+    if (size > mCapacity) {
+        return { [mDevice newBufferWithLength:size options:MTLStorageModeShared], 0 };
+    }
+    assert_invariant(mCurrentUploadBuffer);
+
+    // Align the head to a 4-byte boundary.
+    mHead = (mHead + 3) & ~3;
+
+    if (UTILS_LIKELY(mHead + size <= mCapacity)) {
+        const size_t oldHead = mHead;
+        mHead += size;
+        return { mCurrentUploadBuffer.get(), oldHead };
+    }
+
+    // We're finished with the current allocation.
+    mCurrentUploadBuffer = { [mDevice newBufferWithLength:mCapacity options:MTLStorageModeShared],
+        TrackedMetalBuffer::Type::STAGING };
+    mHead = size;
+
+    return { mCurrentUploadBuffer.get(), 0 };
+}
+
 } // namespace backend
 } // namespace filament

filament/backend/src/metal/MetalContext.h

@@ -44,6 +44,7 @@ namespace backend {
 class MetalDriver;
 class MetalBlitter;
 class MetalBufferPool;
+class MetalBumpAllocator;
 class MetalRenderTarget;
 class MetalSamplerGroup;
 class MetalSwapChain;
@@ -55,6 +56,18 @@ struct MetalVertexBuffer;
 
 constexpr static uint8_t MAX_SAMPLE_COUNT = 8;  // Metal devices support at most 8 MSAA samples
 
+class MetalPushConstantBuffer {
+public:
+    void setPushConstant(PushConstantVariant value, uint8_t index);
+    bool isDirty() const { return mDirty; }
+    void setBytes(id<MTLCommandEncoder> encoder, ShaderStage stage);
+    void clear();
+
+private:
+    std::vector<PushConstantVariant> mPushConstants;
+    bool mDirty = false;
+};
+
 struct MetalContext {
     explicit MetalContext(size_t metalFreedTextureListSize)
         : texturesToDestroy(metalFreedTextureListSize) {}
@@ -80,7 +93,7 @@ struct MetalContext {
     } highestSupportedGpuFamily;
 
     struct {
-        bool a8xStaticTextureTargetError;
+        bool staticTextureTargetError;
     } bugs;
 
     // sampleCountLookup[requestedSamples] gives a <= sample count supported by the device.
@@ -99,6 +112,8 @@ struct MetalContext {
     std::array<BufferState, MAX_SSBO_COUNT> ssboState;
     CullModeStateTracker cullModeState;
     WindingStateTracker windingState;
+    DepthClampStateTracker depthClampState;
+    Handle<HwRenderPrimitive> currentRenderPrimitive;
 
     // State caches.
     DepthStencilStateCache depthStencilStateCache;
@@ -108,6 +123,8 @@ struct MetalContext {
 
     PolygonOffset currentPolygonOffset = {0.0f, 0.0f};
 
+    std::array<MetalPushConstantBuffer, Program::SHADER_TYPE_COUNT> currentPushConstants;
+
     MetalSamplerGroup* samplerBindings[Program::SAMPLER_BINDING_COUNT] = {};
 
     // Keeps track of sampler groups we've finalized for the current render pass.
@@ -126,6 +143,7 @@ struct MetalContext {
     utils::FixedCircularBuffer<Handle<HwTexture>> texturesToDestroy;
 
     MetalBufferPool* bufferPool;
+    MetalBumpAllocator* bumpAllocator;
 
     MetalSwapChain* currentDrawSwapChain = nil;
     MetalSwapChain* currentReadSwapChain = nil;

filament/backend/src/metal/MetalContext.mm

@@ -113,7 +113,8 @@ id<MTLCommandBuffer> getPendingCommandBuffer(MetalContext* context) {
             }
         }
     }];
-    ASSERT_POSTCONDITION(context->pendingCommandBuffer, "Could not obtain command buffer.");
+    FILAMENT_CHECK_POSTCONDITION(context->pendingCommandBuffer)
+            << "Could not obtain command buffer.";
     return context->pendingCommandBuffer;
 }
 
@@ -153,5 +154,68 @@ bool isInRenderPass(MetalContext* context) {
     return context->currentRenderPassEncoder != nil;
 }
 
+void MetalPushConstantBuffer::setPushConstant(PushConstantVariant value, uint8_t index) {
+    if (mPushConstants.size() <= index) {
+        mPushConstants.resize(index + 1);
+        mDirty = true;
+    }
+    if (UTILS_LIKELY(mPushConstants[index] != value)) {
+        mDirty = true;
+        mPushConstants[index] = value;
+    }
+}
+
+void MetalPushConstantBuffer::setBytes(id<MTLCommandEncoder> encoder, ShaderStage stage) {
+    constexpr size_t PUSH_CONSTANT_SIZE_BYTES = 4;
+    constexpr size_t PUSH_CONSTANT_BUFFER_INDEX = 26;
+
+    static char buffer[MAX_PUSH_CONSTANT_COUNT * PUSH_CONSTANT_SIZE_BYTES];
+    assert_invariant(mPushConstants.size() <= MAX_PUSH_CONSTANT_COUNT);
+
+    size_t bufferSize = PUSH_CONSTANT_SIZE_BYTES * mPushConstants.size();
+    for (size_t i = 0; i < mPushConstants.size(); i++) {
+        const auto& constant = mPushConstants[i];
+        std::visit(
+                [i](auto arg) {
+                    if constexpr (std::is_same_v<decltype(arg), bool>) {
+                        // bool push constants are converted to uints in MSL.
+                        // We must ensure we write all the bytes for boolean values to work
+                        // correctly.
+                        uint32_t boolAsUint = arg ? 0x00000001 : 0x00000000;
+                        *(reinterpret_cast<uint32_t*>(buffer + PUSH_CONSTANT_SIZE_BYTES * i)) =
+                                boolAsUint;
+                    } else {
+                        *(decltype(arg)*)(buffer + PUSH_CONSTANT_SIZE_BYTES * i) = arg;
+                    }
+                },
+                constant);
+    }
+
+    switch (stage) {
+        case ShaderStage::VERTEX:
+            [(id<MTLRenderCommandEncoder>)encoder setVertexBytes:buffer
+                                                          length:bufferSize
+                                                         atIndex:PUSH_CONSTANT_BUFFER_INDEX];
+            break;
+        case ShaderStage::FRAGMENT:
+            [(id<MTLRenderCommandEncoder>)encoder setFragmentBytes:buffer
+                                                            length:bufferSize
+                                                           atIndex:PUSH_CONSTANT_BUFFER_INDEX];
+            break;
+        case ShaderStage::COMPUTE:
+            [(id<MTLComputeCommandEncoder>)encoder setBytes:buffer
+                                                     length:bufferSize
+                                                    atIndex:PUSH_CONSTANT_BUFFER_INDEX];
+            break;
+    }
+
+    mDirty = false;
+}
+
+void MetalPushConstantBuffer::clear() {
+    mPushConstants.clear();
+    mDirty = false;
+}
+
 } // namespace backend
 } // namespace filament

filament/backend/src/metal/MetalDriver.h

@@ -17,6 +17,7 @@
 #ifndef TNT_FILAMENT_DRIVER_METALDRIVER_H
 #define TNT_FILAMENT_DRIVER_METALDRIVER_H
 
+#include <backend/DriverEnums.h>
 #include "private/backend/Driver.h"
 #include "DriverBase.h"
 
@@ -140,6 +141,7 @@ private:
     void enumerateBoundBuffers(BufferObjectBinding bindingType,
             const std::function<void(const BufferState&, MetalBuffer*, uint32_t)>& f);
 
+    backend::StereoscopicType const mStereoscopicType;
 };
 
 } // namespace backend

filament/backend/src/metal/MetalDriver.mm

@@ -20,6 +20,7 @@
 #include "metal/MetalDriver.h"
 
 #include "MetalBlitter.h"
+#include "MetalBufferPool.h"
 #include "MetalContext.h"
 #include "MetalDriverFactory.h"
 #include "MetalEnums.h"
@@ -36,6 +37,7 @@
 
 #include <utils/Log.h>
 #include <utils/Panic.h>
+#include <utils/sstream.h>
 
 #include <algorithm>
 
@@ -48,24 +50,26 @@ Driver* MetalDriverFactory::create(MetalPlatform* const platform, const Platform
     // For reference on a 64-bits machine in Release mode:
     //    MetalTimerQuery              :  16       few
     //    HwStream                     :  24       few
-    //    MetalIndexBuffer             :  40       moderate
+    //    MetalRenderPrimitive         :  24       many
+    //    MetalVertexBuffer            :  32       moderate
+    // -- less than or equal 32 bytes
     //    MetalFence                   :  48       few
-    //    MetalBufferObject            :  48       many
-    // -- less than or equal 48 bytes
+    //    MetalIndexBuffer             :  56       moderate
+    //    MetalBufferObject            :  64       many
+    // -- less than or equal 64 bytes
     //    MetalSamplerGroup            : 112       few
-    //    MetalProgram                 : 144       moderate
+    //    MetalProgram                 : 152       moderate
     //    MetalTexture                 : 152       moderate
-    //    MetalVertexBuffer            : 152       moderate
-    // -- less than or equal 160 bytes
-    //    MetalSwapChain               : 184       few
+    //    MetalSwapChain               : 208       few
     //    MetalRenderTarget            : 272       few
-    //    MetalRenderPrimitive         : 584       many
-    // -- less than or equal to 592 bytes
+    //    MetalVertexBufferInfo        : 552       moderate
+    // -- less than or equal to 552 bytes
 
     utils::slog.d
            << "\nMetalSwapChain: " << sizeof(MetalSwapChain)
            << "\nMetalBufferObject: " << sizeof(MetalBufferObject)
            << "\nMetalVertexBuffer: " << sizeof(MetalVertexBuffer)
+           << "\nMetalVertexBufferInfo: " << sizeof(MetalVertexBufferInfo)
            << "\nMetalIndexBuffer: " << sizeof(MetalIndexBuffer)
            << "\nMetalSamplerGroup: " << sizeof(MetalSamplerGroup)
            << "\nMetalRenderPrimitive: " << sizeof(MetalRenderPrimitive)
@@ -96,9 +100,15 @@ Dispatcher MetalDriver::getDispatcher() const noexcept {
 MetalDriver::MetalDriver(MetalPlatform* platform, const Platform::DriverConfig& driverConfig) noexcept
         : mPlatform(*platform),
           mContext(new MetalContext(driverConfig.textureUseAfterFreePoolSize)),
-          mHandleAllocator("Handles", driverConfig.handleArenaSize) {
+          mHandleAllocator("Handles",
+                  driverConfig.handleArenaSize,
+                  driverConfig.disableHandleUseAfterFreeCheck),
+          mStereoscopicType(driverConfig.stereoscopicType) {
     mContext->driver = this;
 
+    TrackedMetalBuffer::setPlatform(platform);
+    ScopedAllocationTimer::setPlatform(platform);
+
     mContext->device = mPlatform.createDevice();
     assert_invariant(mContext->device);
 
@@ -152,8 +162,10 @@ MetalDriver::MetalDriver(MetalPlatform* platform, const Platform::DriverConfig&
         sc[s] = [mContext->device supportsTextureSampleCount:s] ? s : sc[s - 1];
     }
 
-    mContext->bugs.a8xStaticTextureTargetError =
-            [mContext->device.name containsString:@"Apple A8X GPU"];
+    mContext->bugs.staticTextureTargetError =
+            [mContext->device.name containsString:@"Apple A8X GPU"] ||
+            [mContext->device.name containsString:@"Apple A8 GPU"] ||
+            [mContext->device.name containsString:@"Apple A7 GPU"];
 
     mContext->commandQueue = mPlatform.createCommandQueue(mContext->device);
     mContext->pipelineStateCache.setDevice(mContext->device);
@@ -161,6 +173,8 @@ MetalDriver::MetalDriver(MetalPlatform* platform, const Platform::DriverConfig&
     mContext->samplerStateCache.setDevice(mContext->device);
     mContext->argumentEncoderCache.setDevice(mContext->device);
     mContext->bufferPool = new MetalBufferPool(*mContext);
+    mContext->bumpAllocator =
+            new MetalBumpAllocator(mContext->device, driverConfig.metalUploadBufferSizeBytes);
     mContext->blitter = new MetalBlitter(*mContext);
 
     if (@available(iOS 12, *)) {
@@ -171,14 +185,18 @@ MetalDriver::MetalDriver(MetalPlatform* platform, const Platform::DriverConfig&
 
     CVReturn success = CVMetalTextureCacheCreate(kCFAllocatorDefault, nullptr, mContext->device,
             nullptr, &mContext->textureCache);
-    ASSERT_POSTCONDITION(success == kCVReturnSuccess, "Could not create Metal texture cache.");
+    FILAMENT_CHECK_POSTCONDITION(success == kCVReturnSuccess)
+            << "Could not create Metal texture cache.";
 
     if (@available(iOS 12, *)) {
         dispatch_queue_t queue = dispatch_get_global_queue(QOS_CLASS_DEFAULT, 0);
         mContext->eventListener = [[MTLSharedEventListener alloc] initWithDispatchQueue:queue];
     }
 
-    mContext->shaderCompiler = new MetalShaderCompiler(mContext->device, *this);
+    const MetalShaderCompiler::Mode compilerMode = driverConfig.disableParallelShaderCompile
+            ? MetalShaderCompiler::Mode::SYNCHRONOUS
+            : MetalShaderCompiler::Mode::ASYNCHRONOUS;
+    mContext->shaderCompiler = new MetalShaderCompiler(mContext->device, *this, compilerMode);
     mContext->shaderCompiler->init();
 
 #if defined(FILAMENT_METAL_PROFILING)
@@ -189,10 +207,13 @@ MetalDriver::MetalDriver(MetalPlatform* platform, const Platform::DriverConfig&
 }
 
 MetalDriver::~MetalDriver() noexcept {
+    TrackedMetalBuffer::setPlatform(nullptr);
+    ScopedAllocationTimer::setPlatform(nullptr);
     mContext->device = nil;
     mContext->emptyTexture = nil;
     CFRelease(mContext->textureCache);
     delete mContext->bufferPool;
+    delete mContext->bumpAllocator;
     delete mContext->blitter;
     delete mContext->timerQueryImpl;
     delete mContext->shaderCompiler;
@@ -203,22 +224,32 @@ void MetalDriver::tick(int) {
     executeTickOps();
 }
 
-void MetalDriver::beginFrame(int64_t monotonic_clock_ns, uint32_t frameId) {
+void MetalDriver::beginFrame(int64_t monotonic_clock_ns,
+        int64_t refreshIntervalNs, uint32_t frameId) {
 #if defined(FILAMENT_METAL_PROFILING)
     os_signpost_interval_begin(mContext->log, mContext->signpostId, "Frame encoding", "%{public}d", frameId);
 #endif
+    if (mPlatform.hasDebugUpdateStatFunc()) {
+        mPlatform.debugUpdateStat("filament.metal.alive_buffers", TrackedMetalBuffer::getAliveBuffers());
+        mPlatform.debugUpdateStat("filament.metal.alive_buffers.generic",
+                TrackedMetalBuffer::getAliveBuffers(TrackedMetalBuffer::Type::GENERIC));
+        mPlatform.debugUpdateStat("filament.metal.alive_buffers.ring",
+                TrackedMetalBuffer::getAliveBuffers(TrackedMetalBuffer::Type::RING));
+        mPlatform.debugUpdateStat("filament.metal.alive_buffers.staging",
+                TrackedMetalBuffer::getAliveBuffers(TrackedMetalBuffer::Type::STAGING));
+    }
 }
 
-void MetalDriver::setFrameScheduledCallback(Handle<HwSwapChain> sch,
-        FrameScheduledCallback callback, void* user) {
+void MetalDriver::setFrameScheduledCallback(
+        Handle<HwSwapChain> sch, CallbackHandler* handler, FrameScheduledCallback&& callback) {
     auto* swapChain = handle_cast<MetalSwapChain>(sch);
-    swapChain->setFrameScheduledCallback(callback, user);
+    swapChain->setFrameScheduledCallback(handler, std::move(callback));
 }
 
-void MetalDriver::setFrameCompletedCallback(Handle<HwSwapChain> sch,
-        CallbackHandler* handler, CallbackHandler::Callback callback, void* user) {
+void MetalDriver::setFrameCompletedCallback(
+        Handle<HwSwapChain> sch, CallbackHandler* handler, utils::Invocable<void(void)>&& callback) {
     auto* swapChain = handle_cast<MetalSwapChain>(sch);
-    swapChain->setFrameCompletedCallback(handler, callback, user);
+    swapChain->setFrameCompletedCallback(handler, std::move(callback));
 }
 
 void MetalDriver::execute(std::function<void(void)> const& fn) noexcept {
@@ -264,14 +295,14 @@ void MetalDriver::endFrame(uint32_t frameId) {
 }
 
 void MetalDriver::flush(int) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-                        "flush must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "flush must be called outside of a render pass.";
     submitPendingCommands(mContext);
 }
 
 void MetalDriver::finish(int) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "finish must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "finish must be called outside of a render pass.";
     // Wait for all frames to finish by submitting and waiting on a dummy command buffer.
     submitPendingCommands(mContext);
     id<MTLCommandBuffer> oneOffBuffer = [mContext->commandQueue commandBuffer];
@@ -332,19 +363,19 @@ void MetalDriver::importTextureR(Handle<HwTexture> th, intptr_t i,
         TextureFormat format, uint8_t samples, uint32_t width, uint32_t height,
         uint32_t depth, TextureUsage usage) {
     id<MTLTexture> metalTexture = (id<MTLTexture>) CFBridgingRelease((void*) i);
-    ASSERT_PRECONDITION(metalTexture.width == width,
-            "Imported id<MTLTexture> width (%d) != Filament texture width (%d)",
-            metalTexture.width, width);
-    ASSERT_PRECONDITION(metalTexture.height == height,
-            "Imported id<MTLTexture> height (%d) != Filament texture height (%d)",
-            metalTexture.height, height);
-    ASSERT_PRECONDITION(metalTexture.mipmapLevelCount == levels,
-            "Imported id<MTLTexture> levels (%d) != Filament texture levels (%d)",
-            metalTexture.mipmapLevelCount, levels);
+    FILAMENT_CHECK_PRECONDITION(metalTexture.width == width)
+            << "Imported id<MTLTexture> width (" << metalTexture.width
+            << ") != Filament texture width (" << width << ")";
+    FILAMENT_CHECK_PRECONDITION(metalTexture.height == height)
+            << "Imported id<MTLTexture> height (" << metalTexture.height
+            << ") != Filament texture height (" << height << ")";
+    FILAMENT_CHECK_PRECONDITION(metalTexture.mipmapLevelCount == levels)
+            << "Imported id<MTLTexture> levels (" << metalTexture.mipmapLevelCount
+            << ") != Filament texture levels (" << levels << ")";
     MTLTextureType filamentMetalType = getMetalType(target);
-    ASSERT_PRECONDITION(metalTexture.textureType == filamentMetalType,
-            "Imported id<MTLTexture> type (%d) != Filament texture type (%d)",
-            metalTexture.textureType, filamentMetalType);
+    FILAMENT_CHECK_PRECONDITION(metalTexture.textureType == filamentMetalType)
+            << "Imported id<MTLTexture> type (" << metalTexture.textureType
+            << ") != Filament texture type (" << filamentMetalType << ")";
     mContext->textures.insert(construct_handle<MetalTexture>(th, *mContext,
         target, levels, format, samples, width, height, depth, usage, metalTexture));
 }
@@ -371,10 +402,10 @@ void MetalDriver::createDefaultRenderTargetR(Handle<HwRenderTarget> rth, int dum
 
 void MetalDriver::createRenderTargetR(Handle<HwRenderTarget> rth,
         TargetBufferFlags targetBufferFlags, uint32_t width, uint32_t height,
-        uint8_t samples, MRT color,
+        uint8_t samples, uint8_t layerCount, MRT color,
         TargetBufferInfo depth, TargetBufferInfo stencil) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "createRenderTarget must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "createRenderTarget must be called outside of a render pass.";
     // Clamp sample count to what the device supports.
     auto& sc = mContext->sampleCountLookup;
     samples = sc[std::min(MAX_SAMPLE_COUNT, samples)];
@@ -385,33 +416,33 @@ void MetalDriver::createRenderTargetR(Handle<HwRenderTarget> rth,
             continue;
         }
         const auto& buffer = color[i];
-        ASSERT_PRECONDITION(buffer.handle,
-                "The COLOR%u flag was specified, but invalid color handle provided.", i);
+        FILAMENT_CHECK_PRECONDITION(buffer.handle)
+                << "The COLOR" << i << " flag was specified, but invalid color handle provided.";
         auto colorTexture = handle_cast<MetalTexture>(buffer.handle);
-        ASSERT_PRECONDITION(colorTexture->getMtlTextureForWrite(),
-                "Color texture passed to render target has no texture allocation");
+        FILAMENT_CHECK_PRECONDITION(colorTexture->getMtlTextureForWrite())
+                << "Color texture passed to render target has no texture allocation";
         colorTexture->extendLodRangeTo(buffer.level);
         colorAttachments[i] = { colorTexture, color[i].level, color[i].layer };
     }
 
     MetalRenderTarget::Attachment depthAttachment = {};
     if (any(targetBufferFlags & TargetBufferFlags::DEPTH)) {
-        ASSERT_PRECONDITION(depth.handle,
-                "The DEPTH flag was specified, but invalid depth handle provided.");
+        FILAMENT_CHECK_PRECONDITION(depth.handle)
+                << "The DEPTH flag was specified, but invalid depth handle provided.";
         auto depthTexture = handle_cast<MetalTexture>(depth.handle);
-        ASSERT_PRECONDITION(depthTexture->getMtlTextureForWrite(),
-                "Depth texture passed to render target has no texture allocation.");
+        FILAMENT_CHECK_PRECONDITION(depthTexture->getMtlTextureForWrite())
+                << "Depth texture passed to render target has no texture allocation.";
         depthTexture->extendLodRangeTo(depth.level);
         depthAttachment = { depthTexture, depth.level, depth.layer };
     }
 
     MetalRenderTarget::Attachment stencilAttachment = {};
     if (any(targetBufferFlags & TargetBufferFlags::STENCIL)) {
-        ASSERT_PRECONDITION(stencil.handle,
-                "The STENCIL flag was specified, but invalid stencil handle provided.");
+        FILAMENT_CHECK_PRECONDITION(stencil.handle)
+                << "The STENCIL flag was specified, but invalid stencil handle provided.";
         auto stencilTexture = handle_cast<MetalTexture>(stencil.handle);
-        ASSERT_PRECONDITION(stencilTexture->getMtlTextureForWrite(),
-                "Stencil texture passed to render target has no texture allocation.");
+        FILAMENT_CHECK_PRECONDITION(stencilTexture->getMtlTextureForWrite())
+                << "Stencil texture passed to render target has no texture allocation.";
         stencilTexture->extendLodRangeTo(stencil.level);
         stencilAttachment = { stencilTexture, stencil.level, stencil.layer };
     }
@@ -770,24 +801,43 @@ bool MetalDriver::isSRGBSwapChainSupported() {
     return false;
 }
 
-bool MetalDriver::isStereoSupported(backend::StereoscopicType stereoscopicType) {
-    switch (stereoscopicType) {
-    case backend::StereoscopicType::INSTANCED:
-        return true;
-    case backend::StereoscopicType::MULTIVIEW:
-        // TODO: implement multiview feature in Metal.
-        return false;
+bool MetalDriver::isProtectedContentSupported() {
+    // the SWAP_CHAIN_CONFIG_PROTECTED_CONTENT flag is not supported
+    return false;
+}
+
+bool MetalDriver::isStereoSupported() {
+    switch (mStereoscopicType) {
+        case backend::StereoscopicType::INSTANCED:
+            return true;
+        case backend::StereoscopicType::MULTIVIEW:
+            // TODO: implement multiview feature in Metal.
+            return false;
+        case backend::StereoscopicType::NONE:
+            return false;
     }
 }
 
 bool MetalDriver::isParallelShaderCompileSupported() {
-    return true;
+    return mContext->shaderCompiler->isParallelShaderCompileSupported();
 }
 
 bool MetalDriver::isDepthStencilResolveSupported() {
     return false;
 }
 
+bool MetalDriver::isDepthStencilBlitSupported(TextureFormat format) {
+    return true;
+}
+
+bool MetalDriver::isProtectedTexturesSupported() {
+    return false;
+}
+
+bool MetalDriver::isDepthClampSupported() {
+    return true;
+}
+
 bool MetalDriver::isWorkaroundNeeded(Workaround workaround) {
     switch (workaround) {
         case Workaround::SPLIT_EASU:
@@ -796,8 +846,8 @@ bool MetalDriver::isWorkaroundNeeded(Workaround workaround) {
             return true;
         case Workaround::ADRENO_UNIFORM_ARRAY_CRASH:
             return false;
-        case Workaround::A8X_STATIC_TEXTURE_TARGET_ERROR:
-            return mContext->bugs.a8xStaticTextureTargetError;
+        case Workaround::METAL_STATIC_TEXTURE_TARGET_ERROR:
+            return mContext->bugs.staticTextureTargetError;
         case Workaround::DISABLE_BLIT_INTO_TEXTURE_ARRAY:
             return false;
         default:
@@ -841,8 +891,8 @@ void MetalDriver::updateIndexBuffer(Handle<HwIndexBuffer> ibh, BufferDescriptor&
 
 void MetalDriver::updateBufferObject(Handle<HwBufferObject> boh, BufferDescriptor&& data,
         uint32_t byteOffset) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "updateBufferObject must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "updateBufferObject must be called outside of a render pass.";
     auto* bo = handle_cast<MetalBufferObject>(boh);
     bo->updateBuffer(data.buffer, data.size, byteOffset);
     scheduleDestroy(std::move(data));
@@ -881,8 +931,8 @@ void MetalDriver::update3DImage(Handle<HwTexture> th, uint32_t level,
         uint32_t xoffset, uint32_t yoffset, uint32_t zoffset,
         uint32_t width, uint32_t height, uint32_t depth,
         PixelBufferDescriptor&& data) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "update3DImage must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "update3DImage must be called outside of a render pass.";
     auto tex = handle_cast<MetalTexture>(th);
     tex->loadImage(level, MTLRegionMake3D(xoffset, yoffset, zoffset, width, height, depth), data);
     scheduleDestroy(std::move(data));
@@ -898,15 +948,15 @@ void MetalDriver::setupExternalImage(void* image) {
 }
 
 void MetalDriver::setExternalImage(Handle<HwTexture> th, void* image) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "setExternalImage must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "setExternalImage must be called outside of a render pass.";
     auto texture = handle_cast<MetalTexture>(th);
     texture->externalImage.set((CVPixelBufferRef) image);
 }
 
 void MetalDriver::setExternalImagePlane(Handle<HwTexture> th, void* image, uint32_t plane) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "setExternalImagePlane must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "setExternalImagePlane must be called outside of a render pass.";
     auto texture = handle_cast<MetalTexture>(th);
     texture->externalImage.set((CVPixelBufferRef) image, plane);
 }
@@ -914,21 +964,22 @@ void MetalDriver::setExternalImagePlane(Handle<HwTexture> th, void* image, uint3
 void MetalDriver::setExternalStream(Handle<HwTexture> th, Handle<HwStream> sh) {
 }
 
-bool MetalDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint64_t* elapsedTime) {
+TimerQueryResult MetalDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint64_t* elapsedTime) {
     auto* tq = handle_cast<MetalTimerQuery>(tqh);
-    return mContext->timerQueryImpl->getQueryResult(tq, elapsedTime);
+    return mContext->timerQueryImpl->getQueryResult(tq, elapsedTime) ?
+           TimerQueryResult::AVAILABLE : TimerQueryResult::NOT_READY;
 }
 
 void MetalDriver::generateMipmaps(Handle<HwTexture> th) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-                        "generateMipmaps must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "generateMipmaps must be called outside of a render pass.";
     auto tex = handle_cast<MetalTexture>(th);
     tex->generateMipmaps();
 }
 
 void MetalDriver::updateSamplerGroup(Handle<HwSamplerGroup> sbh, BufferDescriptor&& data) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "updateSamplerGroup must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "updateSamplerGroup must be called outside of a render pass.";
 
     auto sb = handle_cast<MetalSamplerGroup>(sbh);
     assert_invariant(sb->size == data.size / sizeof(SamplerDescriptor));
@@ -1067,6 +1118,10 @@ void MetalDriver::beginRenderPass(Handle<HwRenderTarget> rth,
     mContext->currentPolygonOffset = {0.0f, 0.0f};
 
     mContext->finalizedSamplerGroups.clear();
+
+    for (auto& pc : mContext->currentPushConstants) {
+        pc.clear();
+    }
 }
 
 void MetalDriver::nextSubpass(int dummy) {}
@@ -1214,6 +1269,16 @@ void MetalDriver::bindSamplers(uint32_t index, Handle<HwSamplerGroup> sbh) {
     mContext->samplerBindings[index] = sb;
 }
 
+void MetalDriver::setPushConstant(backend::ShaderStage stage, uint8_t index,
+        backend::PushConstantVariant value) {
+    FILAMENT_CHECK_PRECONDITION(isInRenderPass(mContext))
+            << "setPushConstant must be called inside a render pass.";
+    assert_invariant(static_cast<size_t>(stage) < mContext->currentPushConstants.size());
+    MetalPushConstantBuffer& pushConstants =
+            mContext->currentPushConstants[static_cast<size_t>(stage)];
+    pushConstants.setPushConstant(value, index);
+}
+
 void MetalDriver::insertEventMarker(const char* string, uint32_t len) {
 
 }
@@ -1264,8 +1329,8 @@ void MetalDriver::stopCapture(int) {
 
 void MetalDriver::readPixels(Handle<HwRenderTarget> src, uint32_t x, uint32_t y, uint32_t width,
         uint32_t height, PixelBufferDescriptor&& data) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-                        "readPixels must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "readPixels must be called outside of a render pass.";
 
     auto srcTarget = handle_cast<MetalRenderTarget>(src);
     // We always readPixels from the COLOR0 attachment.
@@ -1279,17 +1344,19 @@ void MetalDriver::readPixels(Handle<HwRenderTarget> src, uint32_t x, uint32_t y,
     width = std::min(static_cast<uint32_t>(srcTextureSize.width), width);
 
     const MTLPixelFormat format = getMetalFormat(data.format, data.type);
-    ASSERT_PRECONDITION(format != MTLPixelFormatInvalid,
-            "The chosen combination of PixelDataFormat (%d) and PixelDataType (%d) is not supported for "
-            "readPixels.", (int) data.format, (int) data.type);
+    FILAMENT_CHECK_PRECONDITION(format != MTLPixelFormatInvalid)
+            << "The chosen combination of PixelDataFormat (" << (int)data.format
+            << ") and PixelDataType (" << (int)data.type
+            << ") is not supported for "
+               "readPixels.";
 
     const bool formatConversionNecessary = srcTexture.pixelFormat != format;
 
     // TODO: MetalBlitter does not currently support format conversions to integer types.
     // The format and type must match the source pixel format exactly.
-    ASSERT_PRECONDITION(!formatConversionNecessary || !isMetalFormatInteger(format),
-            "readPixels does not support integer format conversions from MTLPixelFormat (%d) to (%d).",
-            (int) srcTexture.pixelFormat, (int) format);
+    FILAMENT_CHECK_PRECONDITION(!formatConversionNecessary || !isMetalFormatInteger(format))
+            << "readPixels does not support integer format conversions from MTLPixelFormat ("
+            << (int)srcTexture.pixelFormat << ") to (" << (int)format << ").";
 
     MTLTextureDescriptor* textureDescriptor =
             [MTLTextureDescriptor texture2DDescriptorWithPixelFormat:format
@@ -1355,31 +1422,31 @@ void MetalDriver::resolve(
     assert_invariant(srcTexture);
     assert_invariant(dstTexture);
 
-    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder == nil,
-            "resolve() cannot be invoked inside a render pass.");
+    FILAMENT_CHECK_PRECONDITION(mContext->currentRenderPassEncoder == nil)
+            << "resolve() cannot be invoked inside a render pass.";
 
-    ASSERT_PRECONDITION(
-            dstTexture->width == srcTexture->width && dstTexture->height == srcTexture->height,
-            "invalid resolve: src and dst sizes don't match");
+    FILAMENT_CHECK_PRECONDITION(
+            dstTexture->width == srcTexture->width && dstTexture->height == srcTexture->height)
+            << "invalid resolve: src and dst sizes don't match";
 
-    ASSERT_PRECONDITION(srcTexture->samples > 1 && dstTexture->samples == 1,
-            "invalid resolve: src.samples=%u, dst.samples=%u",
-            +srcTexture->samples, +dstTexture->samples);
+    FILAMENT_CHECK_PRECONDITION(srcTexture->samples > 1 && dstTexture->samples == 1)
+            << "invalid resolve: src.samples=" << +srcTexture->samples
+            << ", dst.samples=" << +dstTexture->samples;
 
-    ASSERT_PRECONDITION(srcTexture->format == dstTexture->format,
-            "src and dst texture format don't match");
+    FILAMENT_CHECK_PRECONDITION(srcTexture->format == dstTexture->format)
+            << "src and dst texture format don't match";
 
-    ASSERT_PRECONDITION(!isDepthFormat(srcTexture->format),
-            "can't resolve depth formats");
+    FILAMENT_CHECK_PRECONDITION(!isDepthFormat(srcTexture->format))
+            << "can't resolve depth formats";
 
-    ASSERT_PRECONDITION(!isStencilFormat(srcTexture->format),
-            "can't resolve stencil formats");
+    FILAMENT_CHECK_PRECONDITION(!isStencilFormat(srcTexture->format))
+            << "can't resolve stencil formats";
 
-    ASSERT_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST),
-            "texture doesn't have BLIT_DST");
+    FILAMENT_CHECK_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST))
+            << "texture doesn't have BLIT_DST";
 
-    ASSERT_PRECONDITION(any(srcTexture->usage & TextureUsage::BLIT_SRC),
-            "texture doesn't have BLIT_SRC");
+    FILAMENT_CHECK_PRECONDITION(any(srcTexture->usage & TextureUsage::BLIT_SRC))
+            << "texture doesn't have BLIT_SRC";
 
     // FIXME: on metal the blit() call below always take the slow path (using a shader)
 
@@ -1404,21 +1471,22 @@ void MetalDriver::blit(
     assert_invariant(srcTexture);
     assert_invariant(dstTexture);
 
-    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder == nil,
-            "blit() cannot be invoked inside a render pass.");
+    FILAMENT_CHECK_PRECONDITION(mContext->currentRenderPassEncoder == nil)
+            << "blit() cannot be invoked inside a render pass.";
 
-    ASSERT_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST),
-            "texture doesn't have BLIT_DST");
+    FILAMENT_CHECK_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST))
+            << "texture doesn't have BLIT_DST";
 
-    ASSERT_PRECONDITION(any(srcTexture->usage & TextureUsage::BLIT_SRC),
-            "texture doesn't have BLIT_SRC");
+    FILAMENT_CHECK_PRECONDITION(any(srcTexture->usage & TextureUsage::BLIT_SRC))
+            << "texture doesn't have BLIT_SRC";
 
-    ASSERT_PRECONDITION(srcTexture->format == dstTexture->format,
-            "src and dst texture format don't match");
+    FILAMENT_CHECK_PRECONDITION(srcTexture->format == dstTexture->format)
+            << "src and dst texture format don't match";
 
-    ASSERT_PRECONDITION(isBlitableTextureType(srcTexture->getMtlTextureForRead().textureType) &&
-                        isBlitableTextureType(dstTexture->getMtlTextureForWrite().textureType),
-            "Metal does not support blitting to/from non-2D textures.");
+    FILAMENT_CHECK_PRECONDITION(
+            isBlitableTextureType(srcTexture->getMtlTextureForRead().textureType) &&
+            isBlitableTextureType(dstTexture->getMtlTextureForWrite().textureType))
+            << "Metal does not support blitting to/from non-2D textures.";
 
     MetalBlitter::BlitArgs args{};
     args.filter = SamplerMagFilter::NEAREST;
@@ -1456,18 +1524,18 @@ void MetalDriver::blitDEPRECATED(TargetBufferFlags buffers,
     // It is called between beginFrame and endFrame, but should never be called in the middle of
     // a render pass.
 
-    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder == nil,
-            "blitDEPRECATED() cannot be invoked inside a render pass.");
+    FILAMENT_CHECK_PRECONDITION(mContext->currentRenderPassEncoder == nil)
+            << "blitDEPRECATED() cannot be invoked inside a render pass.";
 
     auto srcTarget = handle_cast<MetalRenderTarget>(src);
     auto dstTarget = handle_cast<MetalRenderTarget>(dst);
 
-    ASSERT_PRECONDITION(buffers == TargetBufferFlags::COLOR0,
-            "blitDEPRECATED only supports COLOR0");
+    FILAMENT_CHECK_PRECONDITION(buffers == TargetBufferFlags::COLOR0)
+            << "blitDEPRECATED only supports COLOR0";
 
-    ASSERT_PRECONDITION(srcRect.left >= 0 && srcRect.bottom >= 0 &&
-                        dstRect.left >= 0 && dstRect.bottom >= 0,
-            "Source and destination rects must be positive.");
+    FILAMENT_CHECK_PRECONDITION(
+            srcRect.left >= 0 && srcRect.bottom >= 0 && dstRect.left >= 0 && dstRect.bottom >= 0)
+            << "Source and destination rects must be positive.";
 
     auto isBlitableTextureType = [](MTLTextureType t) {
         return t == MTLTextureType2D || t == MTLTextureType2DMultisample ||
@@ -1479,9 +1547,10 @@ void MetalDriver::blitDEPRECATED(TargetBufferFlags buffers,
     MetalRenderTarget::Attachment const dstColorAttachment = dstTarget->getDrawColorAttachment(0);
 
     if (srcColorAttachment && dstColorAttachment) {
-        ASSERT_PRECONDITION(isBlitableTextureType(srcColorAttachment.getTexture().textureType) &&
-                            isBlitableTextureType(dstColorAttachment.getTexture().textureType),
-                           "Metal does not support blitting to/from non-2D textures.");
+        FILAMENT_CHECK_PRECONDITION(
+                isBlitableTextureType(srcColorAttachment.getTexture().textureType) &&
+                isBlitableTextureType(dstColorAttachment.getTexture().textureType))
+                << "Metal does not support blitting to/from non-2D textures.";
 
         MetalBlitter::BlitArgs args{};
         args.filter = filter;
@@ -1592,11 +1661,13 @@ void MetalDriver::finalizeSamplerGroup(MetalSamplerGroup* samplerGroup) {
     }
 }
 
-void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph,
-        uint32_t const indexOffset, uint32_t const indexCount, uint32_t const instanceCount) {
-    ASSERT_PRECONDITION(mContext->currentRenderPassEncoder != nullptr,
-            "Attempted to draw without a valid command encoder.");
-    auto primitive = handle_cast<MetalRenderPrimitive>(rph);
+void MetalDriver::bindPipeline(PipelineState const& ps) {
+    FILAMENT_CHECK_PRECONDITION(mContext->currentRenderPassEncoder != nullptr)
+            << "bindPipeline() without a valid command encoder.";
+
+    MetalVertexBufferInfo const* const vbi =
+            handle_cast<MetalVertexBufferInfo>(ps.vertexBufferInfo);
+
     auto program = handle_cast<MetalProgram>(ps.program);
     const auto& rs = ps.rasterState;
 
@@ -1637,7 +1708,7 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph,
     MetalPipelineState const pipelineState {
         .vertexFunction = vertex,
         .fragmentFunction = fragment,
-        .vertexDescription = primitive->vertexDescription,
+        .vertexDescription = vbi->vertexDescription,
         .colorAttachmentPixelFormat = {
             colorPixelFormat[0],
             colorPixelFormat[1],
@@ -1684,6 +1755,13 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph,
         [mContext->currentRenderPassEncoder setFrontFacingWinding:winding];
     }
 
+    // depth clip mode
+    MTLDepthClipMode depthClipMode = rs.depthClamp ? MTLDepthClipModeClamp : MTLDepthClipModeClip;
+    mContext->depthClampState.updateState(depthClipMode);
+    if (mContext->depthClampState.stateChanged()) {
+        [mContext->currentRenderPassEncoder setDepthClipMode:depthClipMode];
+    }
+
     // Set the depth-stencil state, if a state change is needed.
     DepthStencilState depthState;
     if (depthAttachment) {
@@ -1731,20 +1809,6 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph,
         mContext->currentPolygonOffset = ps.polygonOffset;
     }
 
-    // Bind uniform buffers.
-    MetalBuffer* uniformsToBind[Program::UNIFORM_BINDING_COUNT] = { nil };
-    NSUInteger offsets[Program::UNIFORM_BINDING_COUNT] = { 0 };
-
-    enumerateBoundBuffers(BufferObjectBinding::UNIFORM,
-            [&uniformsToBind, &offsets](const BufferState& state, MetalBuffer* buffer,
-                    uint32_t index) {
-        uniformsToBind[index] = buffer;
-        offsets[index] = state.offset;
-    });
-    MetalBuffer::bindBuffers(getPendingCommandBuffer(mContext), mContext->currentRenderPassEncoder,
-            UNIFORM_BUFFER_BINDING_START, MetalBuffer::Stage::VERTEX | MetalBuffer::Stage::FRAGMENT,
-            uniformsToBind, offsets, Program::UNIFORM_BINDING_COUNT);
-
     // Bind sampler groups (argument buffers).
     for (size_t s = 0; s < Program::SAMPLER_BINDING_COUNT; s++) {
         MetalSamplerGroup* const samplerGroup = mContext->samplerBindings[s];
@@ -1775,19 +1839,29 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph,
                                                           atIndex:(SAMPLER_GROUP_BINDING_START + s)];
         }
     }
+}
+
+void MetalDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
+    FILAMENT_CHECK_PRECONDITION(mContext->currentRenderPassEncoder != nullptr)
+            << "bindRenderPrimitive() without a valid command encoder.";
 
     // Bind the user vertex buffers.
-
     MetalBuffer* vertexBuffers[MAX_VERTEX_BUFFER_COUNT] = {};
     size_t vertexBufferOffsets[MAX_VERTEX_BUFFER_COUNT] = {};
     size_t maxBufferIndex = 0;
 
+    MetalRenderPrimitive const* const primitive = handle_cast<MetalRenderPrimitive>(rph);
+    MetalVertexBufferInfo const* const vbi =
+            handle_cast<MetalVertexBufferInfo>(primitive->vertexBuffer->vbih);
+
+    mContext->currentRenderPrimitive = rph;
+
     auto vb = primitive->vertexBuffer;
-    for (auto m : primitive->bufferMapping) {
+    for (auto m : vbi->bufferMapping) {
         assert_invariant(
                 m.bufferArgumentIndex >= USER_VERTEX_BUFFER_BINDING_START &&
                 m.bufferArgumentIndex < USER_VERTEX_BUFFER_BINDING_START + MAX_VERTEX_BUFFER_COUNT);
-        size_t vertexBufferIndex = m.bufferArgumentIndex - USER_VERTEX_BUFFER_BINDING_START;
+        size_t const vertexBufferIndex = m.bufferArgumentIndex - USER_VERTEX_BUFFER_BINDING_START;
         vertexBuffers[vertexBufferIndex] = vb->buffers[m.sourceBufferIndex];
         maxBufferIndex = std::max(maxBufferIndex, vertexBufferIndex);
     }
@@ -1802,6 +1876,35 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph,
     [mContext->currentRenderPassEncoder setVertexBytes:bytes
                                                 length:16
                                                atIndex:ZERO_VERTEX_BUFFER_BINDING];
+}
+
+void MetalDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
+    FILAMENT_CHECK_PRECONDITION(mContext->currentRenderPassEncoder != nullptr)
+            << "draw() without a valid command encoder.";
+
+    // Bind uniform buffers.
+    MetalBuffer* uniformsToBind[Program::UNIFORM_BINDING_COUNT] = { nil };
+    NSUInteger offsets[Program::UNIFORM_BINDING_COUNT] = { 0 };
+
+    enumerateBoundBuffers(BufferObjectBinding::UNIFORM,
+            [&uniformsToBind, &offsets](const BufferState& state, MetalBuffer* buffer,
+                    uint32_t index) {
+                uniformsToBind[index] = buffer;
+                offsets[index] = state.offset;
+            });
+    MetalBuffer::bindBuffers(getPendingCommandBuffer(mContext), mContext->currentRenderPassEncoder,
+            UNIFORM_BUFFER_BINDING_START, MetalBuffer::Stage::VERTEX | MetalBuffer::Stage::FRAGMENT,
+            uniformsToBind, offsets, Program::UNIFORM_BINDING_COUNT);
+
+    // Update push constants.
+    for (size_t i = 0; i < Program::SHADER_TYPE_COUNT; i++) {
+        auto& pushConstants = mContext->currentPushConstants[i];
+        if (UTILS_UNLIKELY(pushConstants.isDirty())) {
+            pushConstants.setBytes(mContext->currentRenderPassEncoder, static_cast<ShaderStage>(i));
+        }
+    }
+
+    auto primitive = handle_cast<MetalRenderPrimitive>(mContext->currentRenderPrimitive);
 
     MetalIndexBuffer* indexBuffer = primitive->indexBuffer;
 
@@ -1815,9 +1918,19 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph,
                                                 instanceCount:instanceCount];
 }
 
+void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph,
+        uint32_t const indexOffset, uint32_t const indexCount, uint32_t const instanceCount) {
+    MetalRenderPrimitive const* const rp = handle_cast<MetalRenderPrimitive>(rph);
+    ps.primitiveType = rp->type;
+    ps.vertexBufferInfo = rp->vertexBuffer->vbih;
+    bindPipeline(ps);
+    bindRenderPrimitive(rph);
+    draw2(indexOffset, indexCount, instanceCount);
+}
+
 void MetalDriver::dispatchCompute(Handle<HwProgram> program, math::uint3 workGroupCount) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "dispatchCompute must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "dispatchCompute must be called outside of a render pass.";
 
     auto mtlProgram = handle_cast<MetalProgram>(program);
 
@@ -1917,15 +2030,15 @@ void MetalDriver::scissor(Viewport scissorBox) {
 }
 
 void MetalDriver::beginTimerQuery(Handle<HwTimerQuery> tqh) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "beginTimerQuery must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "beginTimerQuery must be called outside of a render pass.";
     auto* tq = handle_cast<MetalTimerQuery>(tqh);
     mContext->timerQueryImpl->beginTimeElapsedQuery(tq);
 }
 
 void MetalDriver::endTimerQuery(Handle<HwTimerQuery> tqh) {
-    ASSERT_PRECONDITION(!isInRenderPass(mContext),
-            "endTimerQuery must be called outside of a render pass.");
+    FILAMENT_CHECK_PRECONDITION(!isInRenderPass(mContext))
+            << "endTimerQuery must be called outside of a render pass.";
     auto* tq = handle_cast<MetalTimerQuery>(tqh);
     mContext->timerQueryImpl->endTimeElapsedQuery(tq);
 }

filament/backend/src/metal/MetalEnums.h

@@ -71,7 +71,7 @@ constexpr inline MTLIndexType getIndexType(size_t elementSize) noexcept {
     } else if (elementSize == 4) {
         return MTLIndexTypeUInt32;
     }
-    ASSERT_POSTCONDITION(false, "Index element size not supported.");
+    FILAMENT_CHECK_POSTCONDITION(false) << "Index element size not supported.";
 }
 
 constexpr inline MTLVertexFormat getMetalFormat(ElementType type, bool normalized) noexcept {
@@ -100,7 +100,7 @@ constexpr inline MTLVertexFormat getMetalFormat(ElementType type, bool normalize
             case ElementType::SHORT4: return MTLVertexFormatShort4Normalized;
             case ElementType::USHORT4: return MTLVertexFormatUShort4Normalized;
             default:
-                ASSERT_POSTCONDITION(false, "Normalized format does not exist.");
+                FILAMENT_CHECK_POSTCONDITION(false) << "Normalized format does not exist.";
                 return MTLVertexFormatInvalid;
         }
     }
@@ -326,7 +326,8 @@ constexpr inline MTLCullMode getMetalCullMode(CullingMode cullMode) noexcept {
         case CullingMode::FRONT: return MTLCullModeFront;
         case CullingMode::BACK: return MTLCullModeBack;
         case CullingMode::FRONT_AND_BACK:
-            ASSERT_POSTCONDITION(false, "FRONT_AND_BACK culling is not supported in Metal.");
+            FILAMENT_CHECK_POSTCONDITION(false)
+                    << "FRONT_AND_BACK culling is not supported in Metal.";
     }
 }
 

filament/backend/src/metal/MetalExternalImage.mm

@@ -29,7 +29,7 @@
         auto description = [error.localizedDescription cStringUsingEncoding:NSUTF8StringEncoding]; \
         utils::slog.e << description << utils::io::endl;                                           \
     }                                                                                              \
-    ASSERT_POSTCONDITION(error == nil, message);
+    FILAMENT_CHECK_POSTCONDITION(error == nil) << message;
 
 namespace filament {
 namespace backend {
@@ -86,14 +86,14 @@ void MetalExternalImage::set(CVPixelBufferRef image) noexcept {
     }
 
     OSType formatType = CVPixelBufferGetPixelFormatType(image);
-    ASSERT_POSTCONDITION(formatType == kCVPixelFormatType_32BGRA ||
-                         formatType == kCVPixelFormatType_420YpCbCr8BiPlanarFullRange,
-            "Metal external images must be in either 32BGRA or 420f format.");
+    FILAMENT_CHECK_POSTCONDITION(formatType == kCVPixelFormatType_32BGRA ||
+            formatType == kCVPixelFormatType_420YpCbCr8BiPlanarFullRange)
+            << "Metal external images must be in either 32BGRA or 420f format.";
 
     size_t planeCount = CVPixelBufferGetPlaneCount(image);
-    ASSERT_POSTCONDITION(planeCount == 0 || planeCount == 2,
-            "The Metal backend does not support images with plane counts of %d.", planeCount);
-
+    FILAMENT_CHECK_POSTCONDITION(planeCount == 0 || planeCount == 2)
+            << "The Metal backend does not support images with plane counts of " << planeCount
+            << ".";
 
     if (planeCount == 0) {
         mImage = image;
@@ -138,8 +138,8 @@ void MetalExternalImage::set(CVPixelBufferRef image, size_t plane) noexcept {
     }
 
     const OSType formatType = CVPixelBufferGetPixelFormatType(image);
-    ASSERT_POSTCONDITION(formatType == kCVPixelFormatType_420YpCbCr8BiPlanarFullRange,
-            "Metal planar external images must be in the 420f format.");
+    FILAMENT_CHECK_POSTCONDITION(formatType == kCVPixelFormatType_420YpCbCr8BiPlanarFullRange)
+            << "Metal planar external images must be in the 420f format.";
 
     mImage = image;
 
@@ -191,8 +191,8 @@ CVMetalTextureRef MetalExternalImage::createTextureFromImage(CVPixelBufferRef im
     CVMetalTextureRef texture;
     CVReturn result = CVMetalTextureCacheCreateTextureFromImage(kCFAllocatorDefault,
             mContext.textureCache, image, nullptr, format, width, height, plane, &texture);
-    ASSERT_POSTCONDITION(result == kCVReturnSuccess,
-            "Could not create a CVMetalTexture from CVPixelBuffer.");
+    FILAMENT_CHECK_POSTCONDITION(result == kCVReturnSuccess)
+            << "Could not create a CVMetalTexture from CVPixelBuffer.";
 
     return texture;
 }
@@ -203,8 +203,8 @@ void MetalExternalImage::shutdown(MetalContext& context) noexcept {
 
 void MetalExternalImage::assertWritableImage(CVPixelBufferRef image) {
     OSType formatType = CVPixelBufferGetPixelFormatType(image);
-    ASSERT_PRECONDITION(formatType == kCVPixelFormatType_32BGRA,
-            "Metal SwapChain images must be in the 32BGRA format.");
+    FILAMENT_CHECK_PRECONDITION(formatType == kCVPixelFormatType_32BGRA)
+            << "Metal SwapChain images must be in the 32BGRA format.";
 }
 
 void MetalExternalImage::unset() {

filament/backend/src/metal/MetalHandles.h

@@ -31,6 +31,8 @@
 
 #include "private/backend/SamplerGroup.h"
 
+#include <backend/DriverEnums.h>
+
 #include <utils/bitset.h>
 #include <utils/CString.h>
 #include <utils/FixedCapacityVector.h>
@@ -71,9 +73,9 @@ public:
 
     void releaseDrawable();
 
-    void setFrameScheduledCallback(FrameScheduledCallback callback, void* user);
-    void setFrameCompletedCallback(CallbackHandler* handler,
-            CallbackHandler::Callback callback, void* user);
+    void setFrameScheduledCallback(CallbackHandler* handler, FrameScheduledCallback&& callback);
+    void setFrameCompletedCallback(
+            CallbackHandler* handler, utils::Invocable<void(void)>&& callback);
 
     // For CAMetalLayer-backed SwapChains, presents the drawable or schedules a
     // FrameScheduledCallback.
@@ -107,22 +109,23 @@ private:
     NSUInteger headlessWidth = 0;
     NSUInteger headlessHeight = 0;
     CAMetalLayer* layer = nullptr;
+    std::shared_ptr<std::mutex> layerDrawableMutex;
     MetalExternalImage externalImage;
     SwapChainType type;
 
-    // These two fields store a callback and user data to notify the client that a frame is ready
-    // for presentation.
-    // If frameScheduledCallback is nullptr, then the Metal backend automatically calls
-    // presentDrawable when the frame is committed.
-    // Otherwise, the Metal backend will not automatically present the frame. Instead, clients bear
-    // the responsibility of presenting the frame by calling the PresentCallable object.
-    FrameScheduledCallback frameScheduledCallback = nullptr;
-    void* frameScheduledUserData = nullptr;
+    // These fields store a callback to notify the client that a frame is ready for presentation. If
+    // !frameScheduled.callback, then the Metal backend automatically calls presentDrawable when the
+    // frame is committed. Otherwise, the Metal backend will not automatically present the frame.
+    // Instead, clients bear the responsibility of presenting the frame by calling the
+    // PresentCallable object.
+    struct {
+        CallbackHandler* handler = nullptr;
+        std::shared_ptr<FrameScheduledCallback> callback = nullptr;
+    } frameScheduled;
 
     struct {
         CallbackHandler* handler = nullptr;
-        CallbackHandler::Callback callback = {};
-        void* user = nullptr;
+        std::shared_ptr<utils::Invocable<void(void)>> callback = nullptr;
     } frameCompleted;
 };
 
@@ -149,7 +152,21 @@ struct MetalVertexBufferInfo : public HwVertexBufferInfo {
     MetalVertexBufferInfo(MetalContext& context,
             uint8_t bufferCount, uint8_t attributeCount, AttributeArray const& attributes);
 
-    AttributeArray attributes;
+    // This struct is used to create the pipeline description to describe vertex assembly.
+    VertexDescription vertexDescription = {};
+
+    struct Entry {
+        uint8_t sourceBufferIndex = 0;
+        uint8_t stride = 0;
+        // maps to ->
+        uint8_t bufferArgumentIndex = 0;
+
+        Entry(uint8_t sourceBufferIndex, uint8_t stride, uint8_t bufferArgumentIndex)
+                : sourceBufferIndex(sourceBufferIndex),
+                  stride(stride),
+                  bufferArgumentIndex(bufferArgumentIndex) {}
+    };
+    utils::FixedCapacityVector<Entry> bufferMapping;
 };
 
 struct MetalVertexBuffer : public HwVertexBuffer {
@@ -176,22 +193,6 @@ struct MetalRenderPrimitive : public HwRenderPrimitive {
 
     MetalVertexBuffer* vertexBuffer = nullptr;
     MetalIndexBuffer* indexBuffer = nullptr;
-
-    // This struct is used to create the pipeline description to describe vertex assembly.
-    VertexDescription vertexDescription = {};
-
-    struct Entry {
-        uint8_t sourceBufferIndex = 0;
-        uint8_t stride = 0;
-        // maps to ->
-        uint8_t bufferArgumentIndex = 0;
-
-        Entry(uint8_t sourceBufferIndex, uint8_t stride, uint8_t bufferArgumentIndex)
-                : sourceBufferIndex(sourceBufferIndex),
-                  stride(stride),
-                  bufferArgumentIndex(bufferArgumentIndex) {}
-    };
-    utils::FixedCapacityVector<Entry> bufferMapping;
 };
 
 class MetalProgram : public HwProgram {

filament/backend/src/metal/MetalHandles.mm

@@ -19,6 +19,7 @@
 #include "MetalBlitter.h"
 #include "MetalEnums.h"
 #include "MetalUtils.h"
+#include "MetalBufferPool.h"
 
 #include <filament/SwapChain.h>
 
@@ -72,6 +73,7 @@ MetalSwapChain::MetalSwapChain(MetalContext& context, CAMetalLayer* nativeWindow
     : context(context),
       depthStencilFormat(decideDepthStencilFormat(flags)),
       layer(nativeWindow),
+      layerDrawableMutex(std::make_shared<std::mutex>()),
       externalImage(context),
       type(SwapChainType::CAMETALLAYER) {
 
@@ -173,14 +175,24 @@ id<MTLTexture> MetalSwapChain::acquireDrawable() {
     }
 
     assert_invariant(isCaMetalLayer());
-    drawable = [layer nextDrawable];
 
-    ASSERT_POSTCONDITION(drawable != nil, "Could not obtain drawable.");
+    // CAMetalLayer's drawable pool is not thread safe. Use a mutex when
+    // calling -nextDrawable, or when releasing the last known reference
+    // to any CAMetalDrawable returned from a previous -nextDrawable.
+    {
+        std::lock_guard<std::mutex> lock(*layerDrawableMutex);
+        drawable = [layer nextDrawable];
+    }
+
+    FILAMENT_CHECK_POSTCONDITION(drawable != nil) << "Could not obtain drawable.";
     return drawable.texture;
 }
 
 void MetalSwapChain::releaseDrawable() {
-    drawable = nil;
+    if (drawable) {
+        std::lock_guard<std::mutex> lock(*layerDrawableMutex);
+        drawable = nil;
+    }
 }
 
 id<MTLTexture> MetalSwapChain::acquireDepthTexture() {
@@ -220,16 +232,16 @@ void MetalSwapChain::ensureDepthStencilTexture() {
     depthStencilTexture = [context.device newTextureWithDescriptor:descriptor];
 }
 
-void MetalSwapChain::setFrameScheduledCallback(FrameScheduledCallback callback, void* user) {
-    frameScheduledCallback = callback;
-    frameScheduledUserData = user;
+void MetalSwapChain::setFrameScheduledCallback(
+        CallbackHandler* handler, FrameScheduledCallback&& callback) {
+    frameScheduled.handler = handler;
+    frameScheduled.callback = std::make_shared<FrameScheduledCallback>(std::move(callback));
 }
 
-void MetalSwapChain::setFrameCompletedCallback(CallbackHandler* handler,
-        CallbackHandler::Callback callback, void* user) {
+void MetalSwapChain::setFrameCompletedCallback(
+        CallbackHandler* handler, utils::Invocable<void(void)>&& callback) {
     frameCompleted.handler = handler;
-    frameCompleted.callback = callback;
-    frameCompleted.user = user;
+    frameCompleted.callback = std::make_shared<utils::Invocable<void(void)>>(std::move(callback));
 }
 
 void MetalSwapChain::present() {
@@ -237,7 +249,7 @@ void MetalSwapChain::present() {
         scheduleFrameCompletedCallback();
     }
     if (drawable) {
-        if (frameScheduledCallback) {
+        if (frameScheduled.callback) {
             scheduleFrameScheduledCallback();
         } else  {
             [getPendingCommandBuffer(&context) presentDrawable:drawable];
@@ -255,9 +267,11 @@ public:
     PresentDrawableData(const PresentDrawableData&) = delete;
     PresentDrawableData& operator=(const PresentDrawableData&) = delete;
 
-    static PresentDrawableData* create(id<CAMetalDrawable> drawable, MetalDriver* driver) {
+    static PresentDrawableData* create(id<CAMetalDrawable> drawable,
+            std::shared_ptr<std::mutex> drawableMutex, MetalDriver* driver) {
+        assert_invariant(drawableMutex);
         assert_invariant(driver);
-        return new PresentDrawableData(drawable, driver);
+        return new PresentDrawableData(drawable, drawableMutex, driver);
     }
 
     static void maybePresentAndDestroyAsync(PresentDrawableData* that, bool shouldPresent) {
@@ -276,16 +290,22 @@ public:
     }
 
 private:
-    PresentDrawableData(id<CAMetalDrawable> drawable, MetalDriver* driver)
-        : mDrawable(drawable), mDriver(driver) {}
+    PresentDrawableData(id<CAMetalDrawable> drawable, std::shared_ptr<std::mutex> drawableMutex,
+            MetalDriver* driver)
+        : mDrawable(drawable), mDrawableMutex(drawableMutex), mDriver(driver) {}
 
     static void cleanupAndDestroy(PresentDrawableData *that) {
-        that->mDrawable = nil;
+        if (that->mDrawable) {
+            std::lock_guard<std::mutex> lock(*(that->mDrawableMutex));
+            that->mDrawable = nil;
+        }
+        that->mDrawableMutex.reset();
         that->mDriver = nullptr;
         delete that;
     }
 
     id<CAMetalDrawable> mDrawable;
+    std::shared_ptr<std::mutex> mDrawableMutex;
     MetalDriver* mDriver = nullptr;
 };
 
@@ -295,21 +315,38 @@ void presentDrawable(bool presentFrame, void* user) {
 }
 
 void MetalSwapChain::scheduleFrameScheduledCallback() {
-    if (!frameScheduledCallback) {
+    if (!frameScheduled.callback) {
         return;
     }
 
     assert_invariant(drawable);
 
-    // Destroy this by calling maybePresentAndDestroyAsync() later.
-    auto* presentData = PresentDrawableData::create(drawable, context.driver);
+    struct Callback {
+        Callback(std::shared_ptr<FrameScheduledCallback> callback, id<CAMetalDrawable> drawable,
+                 std::shared_ptr<std::mutex> drawableMutex, MetalDriver* driver)
+            : f(callback), data(PresentDrawableData::create(drawable, drawableMutex, driver)) {}
+        std::shared_ptr<FrameScheduledCallback> f;
+        // PresentDrawableData* is destroyed by maybePresentAndDestroyAsync() later.
+        std::unique_ptr<PresentDrawableData> data;
+        static void func(void* user) {
+            auto* const c = reinterpret_cast<Callback*>(user);
+            PresentDrawableData* presentDrawableData = c->data.release();
+            PresentCallable presentCallable(presentDrawable, presentDrawableData);
+            c->f->operator()(presentCallable);
+            delete c;
+        }
+    };
 
-    FrameScheduledCallback userCallback = frameScheduledCallback;
-    void* userData = frameScheduledUserData;
+    // This callback pointer will be captured by the block. Even if the scheduled handler is never
+    // called, the unique_ptr will still ensure we don't leak memory.
+    __block auto callback = std::make_unique<Callback>(
+        frameScheduled.callback, drawable, layerDrawableMutex, context.driver);
 
+    backend::CallbackHandler* handler = frameScheduled.handler;
+    MetalDriver* driver = context.driver;
     [getPendingCommandBuffer(&context) addScheduledHandler:^(id<MTLCommandBuffer> cb) {
-        PresentCallable callable(presentDrawable, static_cast<void*>(presentData));
-        userCallback(callable, userData);
+        Callback* user = callback.release();
+        driver->scheduleCallback(handler, user, &Callback::func);
     }];
 }
 
@@ -318,13 +355,25 @@ void MetalSwapChain::scheduleFrameCompletedCallback() {
         return;
     }
 
-    CallbackHandler* handler = frameCompleted.handler;
-    void* user = frameCompleted.user;
-    CallbackHandler::Callback callback = frameCompleted.callback;
+    struct Callback {
+        Callback(std::shared_ptr<utils::Invocable<void(void)>> callback) : f(callback) {}
+        std::shared_ptr<utils::Invocable<void(void)>> f;
+        static void func(void* user) {
+            auto* const c = reinterpret_cast<Callback*>(user);
+            c->f->operator()();
+            delete c;
+        }
+    };
 
+    // This callback pointer will be captured by the block. Even if the completed handler is never
+    // called, the unique_ptr will still ensure we don't leak memory.
+    __block auto callback = std::make_unique<Callback>(frameCompleted.callback);
+
+    CallbackHandler* handler = frameCompleted.handler;
     MetalDriver* driver = context.driver;
     [getPendingCommandBuffer(&context) addCompletedHandler:^(id<MTLCommandBuffer> cb) {
-        driver->scheduleCallback(handler, user, callback);
+        Callback* user = callback.release();
+        driver->scheduleCallback(handler, user, &Callback::func);
     }];
 }
 
@@ -343,27 +392,9 @@ void MetalBufferObject::updateBufferUnsynchronized(void* data, size_t size, uint
 MetalVertexBufferInfo::MetalVertexBufferInfo(MetalContext& context, uint8_t bufferCount,
         uint8_t attributeCount, AttributeArray const& attributes)
         : HwVertexBufferInfo(bufferCount, attributeCount),
-          attributes(attributes) {
-}
+          bufferMapping(utils::FixedCapacityVector<Entry>::with_capacity(MAX_VERTEX_BUFFER_COUNT)) {
 
-MetalVertexBuffer::MetalVertexBuffer(MetalContext& context,
-        uint32_t vertexCount, uint32_t bufferCount, Handle<HwVertexBufferInfo> vbih)
-    : HwVertexBuffer(vertexCount), vbih(vbih), buffers(bufferCount, nullptr) {
-}
-
-MetalIndexBuffer::MetalIndexBuffer(MetalContext& context, BufferUsage usage, uint8_t elementSize,
-        uint32_t indexCount) : HwIndexBuffer(elementSize, indexCount),
-        buffer(context, BufferObjectBinding::VERTEX, usage, elementSize * indexCount, true) { }
-
-MetalRenderPrimitive::MetalRenderPrimitive()
-    : bufferMapping(utils::FixedCapacityVector<Entry>::with_capacity(MAX_VERTEX_BUFFER_COUNT)) {}
-
-void MetalRenderPrimitive::setBuffers(MetalVertexBufferInfo const* const vbi,
-        MetalVertexBuffer* vertexBuffer, MetalIndexBuffer* indexBuffer) {
-    this->vertexBuffer = vertexBuffer;
-    this->indexBuffer = indexBuffer;
-
-    const size_t attributeCount = vbi->attributes.size();
+    const size_t maxAttributeCount = attributes.size();
 
     auto& mapping = bufferMapping;
     mapping.clear();
@@ -403,8 +434,8 @@ void MetalRenderPrimitive::setBuffers(MetalVertexBufferInfo const* const vbi,
         }
     };
 
-    for (uint32_t attributeIndex = 0; attributeIndex < attributeCount; attributeIndex++) {
-        const auto& attribute = vbi->attributes[attributeIndex];
+    for (uint32_t attributeIndex = 0; attributeIndex < maxAttributeCount; attributeIndex++) {
+        const auto& attribute = attributes[attributeIndex];
 
         // If the attribute is unused, bind it to the zero buffer. It's a Metal error for a shader
         // to read from missing vertex attributes.
@@ -431,6 +462,24 @@ void MetalRenderPrimitive::setBuffers(MetalVertexBufferInfo const* const vbi,
     }
 }
 
+MetalVertexBuffer::MetalVertexBuffer(MetalContext& context,
+        uint32_t vertexCount, uint32_t bufferCount, Handle<HwVertexBufferInfo> vbih)
+    : HwVertexBuffer(vertexCount), vbih(vbih), buffers(bufferCount, nullptr) {
+}
+
+MetalIndexBuffer::MetalIndexBuffer(MetalContext& context, BufferUsage usage, uint8_t elementSize,
+        uint32_t indexCount) : HwIndexBuffer(elementSize, indexCount),
+        buffer(context, BufferObjectBinding::VERTEX, usage, elementSize * indexCount, true) { }
+
+MetalRenderPrimitive::MetalRenderPrimitive() {
+}
+
+void MetalRenderPrimitive::setBuffers(MetalVertexBufferInfo const* const vbi,
+        MetalVertexBuffer* vertexBuffer, MetalIndexBuffer* indexBuffer) {
+    this->vertexBuffer = vertexBuffer;
+    this->indexBuffer = indexBuffer;
+}
+
 MetalProgram::MetalProgram(MetalContext& context, Program&& program) noexcept
     : HwProgram(program.getName()), mContext(context) {
 
@@ -463,15 +512,16 @@ MetalTexture::MetalTexture(MetalContext& context, SamplerType target, uint8_t le
         externalImage(context, r, g, b, a) {
 
     devicePixelFormat = decidePixelFormat(&context, format);
-    ASSERT_POSTCONDITION(devicePixelFormat != MTLPixelFormatInvalid, "Texture format not supported.");
+    FILAMENT_CHECK_POSTCONDITION(devicePixelFormat != MTLPixelFormatInvalid)
+            << "Texture format not supported.";
 
     const BOOL mipmapped = levels > 1;
     const BOOL multisampled = samples > 1;
 
 #if defined(IOS)
     const BOOL textureArray = target == SamplerType::SAMPLER_2D_ARRAY;
-    ASSERT_PRECONDITION(!textureArray || !multisampled,
-            "iOS does not support multisampled texture arrays.");
+    FILAMENT_CHECK_PRECONDITION(!textureArray || !multisampled)
+            << "iOS does not support multisampled texture arrays.";
 #endif
 
     const auto get2DTextureType = [](SamplerType target, bool isMultisampled) {
@@ -506,12 +556,12 @@ MetalTexture::MetalTexture(MetalContext& context, SamplerType target, uint8_t le
             descriptor.usage = getMetalTextureUsage(usage);
             descriptor.storageMode = MTLStorageModePrivate;
             texture = [context.device newTextureWithDescriptor:descriptor];
-            ASSERT_POSTCONDITION(texture != nil, "Could not create Metal texture. Out of memory?");
             break;
         case SamplerType::SAMPLER_CUBEMAP:
         case SamplerType::SAMPLER_CUBEMAP_ARRAY:
-            ASSERT_POSTCONDITION(!multisampled, "Multisampled cubemap faces not supported.");
-            ASSERT_POSTCONDITION(width == height, "Cubemap faces must be square.");
+            FILAMENT_CHECK_POSTCONDITION(!multisampled)
+                    << "Multisampled cubemap faces not supported.";
+            FILAMENT_CHECK_POSTCONDITION(width == height) << "Cubemap faces must be square.";
             descriptor = [MTLTextureDescriptor textureCubeDescriptorWithPixelFormat:devicePixelFormat
                                                                                size:width
                                                                           mipmapped:mipmapped];
@@ -520,7 +570,6 @@ MetalTexture::MetalTexture(MetalContext& context, SamplerType target, uint8_t le
             descriptor.usage = getMetalTextureUsage(usage);
             descriptor.storageMode = MTLStorageModePrivate;
             texture = [context.device newTextureWithDescriptor:descriptor];
-            ASSERT_POSTCONDITION(texture != nil, "Could not create Metal texture. Out of memory?");
             break;
         case SamplerType::SAMPLER_3D:
             descriptor = [MTLTextureDescriptor new];
@@ -533,7 +582,6 @@ MetalTexture::MetalTexture(MetalContext& context, SamplerType target, uint8_t le
             descriptor.usage = getMetalTextureUsage(usage);
             descriptor.storageMode = MTLStorageModePrivate;
             texture = [context.device newTextureWithDescriptor:descriptor];
-            ASSERT_POSTCONDITION(texture != nil, "Could not create Metal texture. Out of memory?");
             break;
         case SamplerType::SAMPLER_EXTERNAL:
             // If we're using external textures (CVPixelBufferRefs), we don't need to make any
@@ -542,6 +590,12 @@ MetalTexture::MetalTexture(MetalContext& context, SamplerType target, uint8_t le
             break;
     }
 
+    FILAMENT_CHECK_POSTCONDITION(target == SamplerType::SAMPLER_EXTERNAL || texture != nil)
+            << "Could not create Metal texture (SamplerType = " << int(target)
+            << ", levels = " << int(levels) << ", MTLPixelFormat = " << int(devicePixelFormat)
+            << ", width = " << width << ", height = " << height << ", depth = " << depth
+            << "). Out of memory?";
+
     // If swizzling is set, set up a swizzled texture view that we'll use when sampling this texture.
     const bool isDefaultSwizzle =
             r == TextureSwizzle::CHANNEL_0 &&
@@ -735,9 +789,9 @@ void MetalTexture::loadSlice(uint32_t level, MTLRegion region, uint32_t byteOffs
         PixelBufferDescriptor const& data) noexcept {
     const PixelBufferShape shape = PixelBufferShape::compute(data, format, region.size, byteOffset);
 
-    ASSERT_PRECONDITION(data.size >= shape.totalBytes,
-            "Expected buffer size of at least %d but "
-            "received PixelBufferDescriptor with size %d.", shape.totalBytes, data.size);
+    FILAMENT_CHECK_PRECONDITION(data.size >= shape.totalBytes)
+            << "Expected buffer size of at least " << shape.totalBytes
+            << " but received PixelBufferDescriptor with size " << data.size << ".";
 
     // Earlier versions of iOS don't have the maxBufferLength query, but 256 MB is a safe bet.
     NSUInteger deviceMaxBufferLength = 256 * 1024 * 1024;   // 256 MB
@@ -770,13 +824,13 @@ void MetalTexture::loadWithCopyBuffer(uint32_t level, uint32_t slice, MTLRegion
         PixelBufferDescriptor const& data, const PixelBufferShape& shape) {
     const size_t stagingBufferSize = shape.totalBytes;
     auto entry = context.bufferPool->acquireBuffer(stagingBufferSize);
-    memcpy(entry->buffer.contents,
+    memcpy(entry->buffer.get().contents,
             static_cast<uint8_t*>(data.buffer) + shape.sourceOffset,
             stagingBufferSize);
     id<MTLCommandBuffer> blitCommandBuffer = getPendingCommandBuffer(&context);
     id<MTLBlitCommandEncoder> blitCommandEncoder = [blitCommandBuffer blitCommandEncoder];
     blitCommandEncoder.label = @"Texture upload buffer blit";
-    [blitCommandEncoder copyFromBuffer:entry->buffer
+    [blitCommandEncoder copyFromBuffer:entry->buffer.get()
                           sourceOffset:0
                      sourceBytesPerRow:shape.bytesPerRow
                    sourceBytesPerImage:shape.bytesPerSlice
@@ -958,9 +1012,9 @@ MetalRenderTarget::MetalRenderTarget(MetalContext* context, uint32_t width, uint
         }
         color[i] = colorAttachments[i];
 
-        ASSERT_PRECONDITION(color[i].getSampleCount() <= samples,
-                "MetalRenderTarget was initialized with a MSAA COLOR%d texture, but sample count is %d.",
-                i, samples);
+        FILAMENT_CHECK_PRECONDITION(color[i].getSampleCount() <= samples)
+                << "MetalRenderTarget was initialized with a MSAA COLOR" << i
+                << " texture, but sample count is " << samples << ".";
 
         auto t = color[i].metalTexture;
         const auto twidth = std::max(1u, t->width >> color[i].level);
@@ -983,9 +1037,10 @@ MetalRenderTarget::MetalRenderTarget(MetalContext* context, uint32_t width, uint
     if (depthAttachment) {
         depth = depthAttachment;
 
-        ASSERT_PRECONDITION(depth.getSampleCount() <= samples,
-                "MetalRenderTarget was initialized with a MSAA DEPTH texture, but sample count is %d.",
-                samples);
+        FILAMENT_CHECK_PRECONDITION(depth.getSampleCount() <= samples)
+                << "MetalRenderTarget was initialized with a MSAA DEPTH texture, but sample count "
+                   "is "
+                << samples << ".";
 
         auto t = depth.metalTexture;
         const auto twidth = std::max(1u, t->width >> depth.level);
@@ -1008,9 +1063,10 @@ MetalRenderTarget::MetalRenderTarget(MetalContext* context, uint32_t width, uint
     if (stencilAttachment) {
         stencil = stencilAttachment;
 
-        ASSERT_PRECONDITION(stencil.getSampleCount() <= samples,
-                "MetalRenderTarget was initialized with a MSAA STENCIL texture, but sample count is %d.",
-                samples);
+        FILAMENT_CHECK_PRECONDITION(stencil.getSampleCount() <= samples)
+                << "MetalRenderTarget was initialized with a MSAA STENCIL texture, but sample "
+                   "count is "
+                << samples << ".";
 
         auto t = stencil.metalTexture;
         const auto twidth = std::max(1u, t->width >> stencil.level);

filament/backend/src/metal/MetalShaderCompiler.h

@@ -41,6 +41,11 @@ class MetalShaderCompiler {
     struct MetalProgramToken;
 
 public:
+    enum class Mode {
+        SYNCHRONOUS,    // synchronous shader compilation
+        ASYNCHRONOUS    // asynchronous shader compilation
+    };
+
     class MetalFunctionBundle {
     public:
         using Raster = std::tuple<id<MTLFunction>, id<MTLFunction>>;
@@ -110,7 +115,7 @@ public:
 
     using program_token_t = std::shared_ptr<MetalProgramToken>;
 
-    explicit MetalShaderCompiler(id<MTLDevice> device, MetalDriver& driver);
+    explicit MetalShaderCompiler(id<MTLDevice> device, MetalDriver& driver, Mode mode);
 
     MetalShaderCompiler(MetalShaderCompiler const& rhs) = delete;
     MetalShaderCompiler(MetalShaderCompiler&& rhs) = delete;
@@ -120,15 +125,15 @@ public:
     void init() noexcept;
     void terminate() noexcept;
 
-    // Creates a program asynchronously
+    bool isParallelShaderCompileSupported() const noexcept;
+
+    // Creates a program, either synchronously or asynchronously, depending on the Mode
+    // MetalShaderCompiler was constructed with.
     program_token_t createProgram(utils::CString const& name, Program&& program);
 
     // Returns the functions, blocking if necessary. The Token is destroyed and becomes invalid.
     MetalFunctionBundle getProgram(program_token_t& token);
 
-    // Destroys a valid token and all associated resources. Used to "cancel" a program compilation.
-    static void terminate(program_token_t& token);
-
     void notifyWhenAllProgramsAreReady(
             CallbackHandler* handler, CallbackHandler::Callback callback, void* user);
 
@@ -138,6 +143,7 @@ private:
     CompilerThreadPool mCompilerThreadPool;
     id<MTLDevice> mDevice;
     CallbackManager mCallbackManager;
+    Mode mMode;
 };
 
 } // namespace filament::backend

filament/backend/src/metal/MetalShaderCompiler.mm

@@ -70,22 +70,31 @@ struct MetalShaderCompiler::MetalProgramToken : ProgramToken {
 
 MetalShaderCompiler::MetalProgramToken::~MetalProgramToken() = default;
 
-MetalShaderCompiler::MetalShaderCompiler(id<MTLDevice> device, MetalDriver& driver)
+MetalShaderCompiler::MetalShaderCompiler(id<MTLDevice> device, MetalDriver& driver, Mode mode)
         : mDevice(device),
-          mCallbackManager(driver) {
+          mCallbackManager(driver),
+          mMode(mode) {
 
 }
 
 void MetalShaderCompiler::init() noexcept {
-    const uint32_t poolSize = 2;
-    mCompilerThreadPool.init(poolSize, []() {}, []() {});
+    const uint32_t poolSize = 1;
+    if (mMode == Mode::ASYNCHRONOUS) {
+        mCompilerThreadPool.init(poolSize, []() {}, []() {});
+    }
 }
 
 void MetalShaderCompiler::terminate() noexcept {
-    mCompilerThreadPool.terminate();
+    if (mMode == Mode::ASYNCHRONOUS) {
+        mCompilerThreadPool.terminate();
+    }
     mCallbackManager.terminate();
 }
 
+bool MetalShaderCompiler::isParallelShaderCompileSupported() const noexcept {
+    return mMode == Mode::ASYNCHRONOUS;
+}
+
 /* static */ MetalShaderCompiler::MetalFunctionBundle MetalShaderCompiler::compileProgram(
         const Program& program, id<MTLDevice> device) {
     std::array<id<MTLFunction>, Program::SHADER_TYPE_COUNT> functions = { nil };
@@ -97,26 +106,41 @@ void MetalShaderCompiler::terminate() noexcept {
             continue;
         }
 
-        assert_invariant(source[source.size() - 1] == '\0');
-
-        // the shader string is null terminated and the length includes the null character
-        NSString* objcSource = [[NSString alloc] initWithBytes:source.data()
-                                                        length:source.size() - 1
-                                                      encoding:NSUTF8StringEncoding];
-
-        // By default, Metal uses the most recent language version.
-        MTLCompileOptions* options = [MTLCompileOptions new];
-
-        // Disable Fast Math optimizations.
-        // This ensures that operations adhere to IEEE standards for floating-point arithmetic,
-        // which is crucial for half precision floats in scenarios where fast math optimizations
-        // lead to inaccuracies, such as in handling special values like NaN or Infinity.
-        options.fastMathEnabled = NO;
-
         NSError* error = nil;
-        id<MTLLibrary> library = [device newLibraryWithSource:objcSource
-                                                      options:options
-                                                        error:&error];
+        id<MTLLibrary> library = nil;
+        switch (program.getShaderLanguage()) {
+            case ShaderLanguage::MSL: {
+                // By default, Metal uses the most recent language version.
+                MTLCompileOptions* options = [MTLCompileOptions new];
+
+                // Disable Fast Math optimizations.
+                // This ensures that operations adhere to IEEE standards for floating-point
+                // arithmetic, which is crucial for half precision floats in scenarios where fast
+                // math optimizations lead to inaccuracies, such as in handling special values like
+                // NaN or Infinity.
+                options.fastMathEnabled = NO;
+
+                assert_invariant(source[source.size() - 1] == '\0');
+                // the shader string is null terminated and the length includes the null character
+                NSString* objcSource = [[NSString alloc] initWithBytes:source.data()
+                                                                length:source.size() - 1
+                                                              encoding:NSUTF8StringEncoding];
+                library = [device newLibraryWithSource:objcSource options:options error:&error];
+                break;
+            }
+            case ShaderLanguage::METAL_LIBRARY: {
+                dispatch_data_t data = dispatch_data_create(source.data(), source.size(),
+                        dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT, 0),
+                        DISPATCH_DATA_DESTRUCTOR_DEFAULT);
+                library = [device newLibraryWithData:data error:&error];
+                break;
+            }
+            case ShaderLanguage::ESSL1:
+            case ShaderLanguage::ESSL3:
+            case ShaderLanguage::SPIRV:
+                break;
+        }
+
         if (library == nil) {
             NSString* errorMessage = @"unknown error";
             if (error) {
@@ -145,6 +169,23 @@ void MetalShaderCompiler::terminate() noexcept {
         id<MTLFunction> function = [library newFunctionWithName:@"main0"
                                                  constantValues:constants
                                                           error:&error];
+        if (function == nil) {
+            // If the library loads but functions within it fail to load, it usually means the
+            // GPU backend crashed. (This can happen if it's a Metallib shader that was compiled
+            // with a minimum iOS version that's newer than this device.)
+            NSString* errorMessage = @"unknown error";
+            if (error) {
+                auto description =
+                        [error.localizedDescription cStringUsingEncoding:NSUTF8StringEncoding];
+                utils::slog.w << description << utils::io::endl;
+                errorMessage = error.localizedDescription;
+            }
+            PANIC_LOG("Failed to load main0 in Metal program.");
+            NSString* programName =
+                [NSString stringWithFormat:@"%s::main0", program.getName().c_str_safe()];
+            return MetalFunctionBundle::error(errorMessage, programName);
+        }
+
         if (!program.getName().empty()) {
             function.label = @(program.getName().c_str());
         }
@@ -180,14 +221,26 @@ MetalShaderCompiler::program_token_t MetalShaderCompiler::createProgram(
 
     token->handle = mCallbackManager.get();
 
-    CompilerPriorityQueue const priorityQueue = program.getPriorityQueue();
-    mCompilerThreadPool.queue(priorityQueue, token,
-            [this, name, device = mDevice, program = std::move(program), token]() {
-                MetalFunctionBundle compiledProgram = compileProgram(program, device);
+    switch (mMode) {
+        case Mode::ASYNCHRONOUS: {
+            CompilerPriorityQueue const priorityQueue = program.getPriorityQueue();
+            mCompilerThreadPool.queue(priorityQueue, token,
+                    [this, name, device = mDevice, program = std::move(program), token]() {
+                        MetalFunctionBundle compiledProgram = compileProgram(program, device);
+                        token->set(compiledProgram);
+                        mCallbackManager.put(token->handle);
+                    });
 
-                token->set(compiledProgram);
-                mCallbackManager.put(token->handle);
-            });
+            break;
+        }
+
+        case Mode::SYNCHRONOUS: {
+            MetalFunctionBundle compiledProgram = compileProgram(program, mDevice);
+            token->set(compiledProgram);
+            mCallbackManager.put(token->handle);
+            break;
+        }
+    }
 
     return token;
 }
@@ -195,35 +248,23 @@ MetalShaderCompiler::program_token_t MetalShaderCompiler::createProgram(
 MetalShaderCompiler::MetalFunctionBundle MetalShaderCompiler::getProgram(program_token_t& token) {
     assert_invariant(token);
 
-    if (!token->isReady()) {
-        auto job = mCompilerThreadPool.dequeue(token);
-        if (job) {
-            job();
+    if (mMode == Mode::ASYNCHRONOUS) {
+        if (!token->isReady()) {
+            auto job = mCompilerThreadPool.dequeue(token);
+            if (job) {
+                job();
+            }
         }
     }
 
+    // The job isn't guaranteed to have finished yet. We may have failed to dequeue it above,
+    // which means it's currently running. In that case get() will block until it finishes.
+
     MetalShaderCompiler::MetalFunctionBundle program = token->get();
-
     token = nullptr;
-
     return program;
 }
 
-/* static */ void MetalShaderCompiler::terminate(program_token_t& token) {
-    assert_invariant(token);
-
-    auto job = token->compiler.mCompilerThreadPool.dequeue(token);
-    if (!job) {
-        // The job is being executed right now (or has already executed).
-        token->wait();
-    } else {
-        // The job has not executed yet.
-        token->compiler.mCallbackManager.put(token->handle);
-    }
-
-    token.reset();
-}
-
 void MetalShaderCompiler::notifyWhenAllProgramsAreReady(
         CallbackHandler* handler, CallbackHandler::Callback callback, void* user) {
     mCallbackManager.setCallback(handler, callback, user);