Update Gitignore to ignore generated files.

- Re-enable renderdiff test.
- Cache the mesa directory so that we're not spending time pulling
  and compiling it.
- Move python prereqs into the script and use venv.

.github/workflows/presubmit.yml

@@ -19,6 +19,8 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
       - name: Run build script
         run: |
           WORKFLOW_OS=`echo \`uname\` | sed "s/Darwin/mac/" | tr [:upper:] [:lower:]`
@@ -44,6 +46,8 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
       - uses: actions/setup-java@v3
         with:
           distribution: 'temurin'
@@ -60,6 +64,8 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
       - name: Run build script
         run: |
           cd build/ios && printf "y" | ./build.sh presubmit
@@ -102,9 +108,21 @@ jobs:
         uses: actions/setup-python@v5
         with:
           python-version: '3.x'
-      - name: Install python prereqs
-        run: pip install mako setuptools pyyaml
-      - name: Run script
+      - name: Cache Mesa and deps
+        id: mesa-cache
+        uses: actions/cache@v4 # Use a specific version
+        with:
+          path: |
+            $HOME/Library/Caches/Homebrew
+            mesa
+          key: ${{ runner.os }}-mesa-deps-${{ vars.MESA_VERSION }}
+      - name: Get Mesa
+        id: mesa-prereq
+        env:
+          MESA_VERSION: ${{ vars.MESA_VERSION }}
+        run: |
+          bash test/utils/get_mesa.sh
+      - name: Run Test
         run: |
           bash test/renderdiff/test.sh
       - uses: actions/upload-artifact@v4
@@ -122,3 +140,31 @@ jobs:
         run: source ./build/linux/ci-common.sh && ./build.sh -W debug test_filamat filament
       - name: Run test
         run: ./out/cmake-debug/libs/filamat/test_filamat --gtest_filter=MaterialCompiler.Wgsl*
+
+  code-correcteness:
+    name: code-correctness
+    runs-on: 'macos-14-xlarge'
+    steps:
+      - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - name: Set up Homebrew
+        id: set-up-homebrew
+        uses: Homebrew/actions/setup-homebrew@master
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.x'
+      - name: Install prerequisites
+        run: |
+          pip install pyyaml
+          brew install llvm
+          sudo ln -s "$(brew --prefix llvm)/bin/clang-tidy" "/usr/local/bin/clang-tidy"
+      - name: Run build script
+        # We need to build before clang-tidy can run analysis
+        run: |
+          # This will build for all three desktop backends on mac
+          ./build.sh -p desktop debug gltf_viewer
+      - name: Run test
+        run: |
+          bash test/code-correctness/test.sh

.gitignore

@@ -18,3 +18,18 @@ test*.json
 results
 /compile_commands.json
 /.cache
+build/.cmake/
+build/CMakeCache.txt
+build/CMakeFiles/
+build/Makefile
+build/SPIRV-Tools*
+build/cmake_install.cmake
+build/compile_commands.json
+build/filament/
+build/include/
+build/libs/
+build/mac/ninja
+build/samples/
+build/shaders/
+build/third_party/
+build/tools/

CMakeLists.txt

@@ -801,6 +801,7 @@ add_subdirectory(${EXTERNAL}/draco/tnt)
 add_subdirectory(${EXTERNAL}/jsmn/tnt)
 add_subdirectory(${EXTERNAL}/stb/tnt)
 add_subdirectory(${EXTERNAL}/getopt)
+add_subdirectory(${EXTERNAL}/perfetto/tnt)
 
 # Note that this has to be placed after mikktspace in order for combine_static_libs to work.
 add_subdirectory(${LIBRARIES}/geometry)

NEW_RELEASE_NOTES.md

@@ -7,5 +7,3 @@ for next branch cut* header.
 appropriate header in [RELEASE_NOTES.md](./RELEASE_NOTES.md).
 
 ## Release notes for next branch cut
-
-- Fix build/compile errors when upgrading to MacOS 15.4

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.59.0'
+    implementation 'com.google.android.filament:filament-android:1.59.3'
 }
 ```
 
@@ -51,7 +51,7 @@ Here are all the libraries available in the group `com.google.android.filament`:
 iOS projects can use CocoaPods to install the latest release:
 
 ```shell
-pod 'Filament', '~> 1.59.0'
+pod 'Filament', '~> 1.59.3'
 ```
 
 ## Documentation

RELEASE_NOTES.md

@@ -7,6 +7,16 @@ 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).
 
+## v1.59.4
+
+
+## v1.59.3
+
+
+## v1.59.2
+
+- Fix build/compile errors when upgrading to MacOS 15.4
+
 ## v1.59.1
 
 

android/buildSrc/src/main/groovy/FilamentPlugin.groovy

@@ -142,16 +142,13 @@ abstract class MaterialCompiler extends TaskWithBinary {
                 if (!exclude_vulkan) {
                     matcArgs += ['-a', 'vulkan']
                 }
-
                 def include_webgpu = providers
                         .gradleProperty("com.google.android.filament.include-webgpu")
                         .forUseAtConfigurationTime().present
-
-                if (!include_webgpu) {
-                    matcArgs += ['-a', 'webgpu']
+                if (include_webgpu) {
+                    matcArgs += ['-a', 'webgpu', '--variant-filter=skinning,stereo']
                 }
 
-
                 def mat_no_opt = providers
                         .gradleProperty("com.google.android.filament.matnopt")
                         .forUseAtConfigurationTime().present

android/filamat-android/CMakeLists.txt

@@ -26,6 +26,10 @@ add_library(utils STATIC IMPORTED)
 set_target_properties(utils PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libutils.a)
 
+add_library(perfetto STATIC IMPORTED)
+set_target_properties(perfetto PROPERTIES IMPORTED_LOCATION
+        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libperfetto.a)
+
 add_library(filabridge STATIC IMPORTED)
 set_target_properties(filabridge PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libfilabridge.a)
@@ -40,6 +44,7 @@ set_target_properties(shaders PROPERTIES IMPORTED_LOCATION
 
 set(FILAMAT_INCLUDE_DIRS
         ../../libs/utils/include
+        ../../third_party/perfetto
         )
 
 include_directories(${FILAMENT_DIR}/include)
@@ -55,6 +60,7 @@ target_link_libraries(filamat-jni
         filabridge
         shaders
         utils
+        perfetto
         log
         smol-v
         $<$<STREQUAL:${FILAMENT_SUPPORTS_WEBGPU},ON>:tint>

android/filament-android/CMakeLists.txt

@@ -21,6 +21,10 @@ add_library(utils STATIC IMPORTED)
 set_target_properties(utils PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libutils.a)
 
+add_library(perfetto STATIC IMPORTED)
+set_target_properties(perfetto PROPERTIES IMPORTED_LOCATION
+        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libperfetto.a)
+
 add_library(ibl-lite STATIC IMPORTED)
 set_target_properties(ibl-lite PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libibl-lite.a)
@@ -123,6 +127,7 @@ target_link_libraries(filament-jni
     PRIVATE android
     PRIVATE jnigraphics
     PRIVATE utils
+    PRIVATE perfetto
 
     # libgeometry is PUBLIC because gltfio uses it.
     PUBLIC geometry
@@ -141,6 +146,7 @@ target_include_directories(filament-jni PRIVATE
         ${FILAMENT_DIR}/include
         ../../filament/backend/include
         ../../third_party/robin-map
+        ../../third_party/perfetto
         ../../libs/utils/include)
 
 # Force a relink when the version script is changed:

android/gltfio-android/CMakeLists.txt

@@ -35,6 +35,10 @@ add_library(utils STATIC IMPORTED)
 set_target_properties(utils PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libutils.a)
 
+add_library(perfetto STATIC IMPORTED)
+set_target_properties(perfetto PROPERTIES IMPORTED_LOCATION
+        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libperfetto.a)
+
 add_library(uberzlib STATIC IMPORTED)
 set_target_properties(uberzlib PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libuberzlib.a)
@@ -121,6 +125,7 @@ set(GLTFIO_INCLUDE_DIRS
         ../../third_party/meshoptimizer/src
         ../../third_party/robin-map
         ../../third_party/stb
+        ../../third_party/perfetto
         ../../libs/utils/include
         ../../libs/ktxreader/include
 )
@@ -129,7 +134,7 @@ add_library(gltfio-jni SHARED ${GLTFIO_SRCS})
 target_include_directories(gltfio-jni PRIVATE ${GLTFIO_INCLUDE_DIRS})
 set_target_properties(gltfio-jni PROPERTIES LINK_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/libgltfio-jni.symbols)
 set_target_properties(gltfio-jni PROPERTIES LINK_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/libgltfio-jni.map)
-target_link_libraries(gltfio-jni filament-jni utils uberzlib log stb ktxreader basis_transcoder zstd uberarchive)
+target_link_libraries(gltfio-jni filament-jni utils perfetto uberzlib log stb ktxreader basis_transcoder zstd uberarchive)
 target_link_libraries(gltfio-jni dracodec meshoptimizer)
 target_compile_definitions(gltfio-jni PUBLIC GLTFIO_DRACO_SUPPORTED=1)
 target_include_directories(gltfio-jni PRIVATE ${DRACO_DIR}/src)

android/gradle.properties

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.59.0
+VERSION_NAME=1.59.3
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 

filament/backend/CMakeLists.txt

@@ -177,29 +177,6 @@ endif()
 if (FILAMENT_SUPPORTS_VULKAN)
     list(APPEND SRCS
             include/backend/platforms/VulkanPlatform.h
-            src/vulkan/VulkanDescriptorSetCache.cpp
-            src/vulkan/VulkanDescriptorSetCache.h
-            src/vulkan/VulkanDescriptorSetLayoutCache.cpp
-            src/vulkan/VulkanDescriptorSetLayoutCache.h
-            src/vulkan/VulkanPipelineLayoutCache.cpp
-            src/vulkan/VulkanPipelineLayoutCache.h
-            src/vulkan/memory/ResourceManager.cpp
-            src/vulkan/memory/ResourceManager.h
-            src/vulkan/memory/ResourcePointer.h
-            src/vulkan/memory/Resource.cpp
-            src/vulkan/memory/Resource.h
-            src/vulkan/platform/VulkanPlatform.cpp
-            src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp
-            src/vulkan/platform/VulkanPlatformSwapChainImpl.h
-            src/vulkan/utils/Conversion.cpp
-            src/vulkan/utils/Conversion.h
-            src/vulkan/utils/Definitions.h
-            src/vulkan/utils/Helper.h
-            src/vulkan/utils/Image.h
-            src/vulkan/utils/Image.cpp
-            src/vulkan/utils/Spirv.h
-            src/vulkan/utils/Spirv.cpp
-            src/vulkan/utils/StaticVector.h
             src/vulkan/VulkanAsyncHandles.h
             src/vulkan/VulkanBlitter.cpp
             src/vulkan/VulkanBlitter.h
@@ -210,6 +187,10 @@ if (FILAMENT_SUPPORTS_VULKAN)
             src/vulkan/VulkanConstants.h
             src/vulkan/VulkanContext.cpp
             src/vulkan/VulkanContext.h
+            src/vulkan/VulkanDescriptorSetCache.cpp
+            src/vulkan/VulkanDescriptorSetCache.h
+            src/vulkan/VulkanDescriptorSetLayoutCache.cpp
+            src/vulkan/VulkanDescriptorSetLayoutCache.h
             src/vulkan/VulkanDriver.cpp
             src/vulkan/VulkanDriver.h
             src/vulkan/VulkanDriverFactory.h
@@ -217,24 +198,43 @@ if (FILAMENT_SUPPORTS_VULKAN)
             src/vulkan/VulkanFboCache.h
             src/vulkan/VulkanHandles.cpp
             src/vulkan/VulkanHandles.h
-            src/vulkan/VulkanMemory.h
             src/vulkan/VulkanMemory.cpp
+            src/vulkan/VulkanMemory.h
             src/vulkan/VulkanPipelineCache.cpp
             src/vulkan/VulkanPipelineCache.h
+            src/vulkan/VulkanPipelineLayoutCache.cpp
+            src/vulkan/VulkanPipelineLayoutCache.h
             src/vulkan/VulkanQueryManager.cpp
             src/vulkan/VulkanQueryManager.h
+            src/vulkan/VulkanReadPixels.cpp
+            src/vulkan/VulkanReadPixels.h
             src/vulkan/VulkanSamplerCache.cpp
             src/vulkan/VulkanSamplerCache.h
             src/vulkan/VulkanStagePool.cpp
             src/vulkan/VulkanStagePool.h
             src/vulkan/VulkanSwapChain.cpp
             src/vulkan/VulkanSwapChain.h
-            src/vulkan/VulkanReadPixels.cpp
-            src/vulkan/VulkanReadPixels.h
             src/vulkan/VulkanTexture.cpp
             src/vulkan/VulkanTexture.h
             src/vulkan/VulkanYcbcrConversionCache.cpp
             src/vulkan/VulkanYcbcrConversionCache.h
+            src/vulkan/memory/Resource.cpp
+            src/vulkan/memory/Resource.h
+            src/vulkan/memory/ResourceManager.cpp
+            src/vulkan/memory/ResourceManager.h
+            src/vulkan/memory/ResourcePointer.h
+            src/vulkan/platform/VulkanPlatform.cpp
+            src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp
+            src/vulkan/platform/VulkanPlatformSwapChainImpl.h
+            src/vulkan/utils/Conversion.cpp
+            src/vulkan/utils/Conversion.h
+            src/vulkan/utils/Definitions.h
+            src/vulkan/utils/Helper.h
+            src/vulkan/utils/Image.cpp
+            src/vulkan/utils/Image.h
+            src/vulkan/utils/Spirv.cpp
+            src/vulkan/utils/Spirv.h
+            src/vulkan/utils/StaticVector.h
     )
     if (LINUX OR WIN32)
         list(APPEND SRCS src/vulkan/platform/VulkanPlatformLinuxWindows.cpp)
@@ -255,10 +255,11 @@ if (FILAMENT_SUPPORTS_WEBGPU)
             src/webgpu/WebGPUConstants.h
             src/webgpu/WebGPUDriver.cpp
             src/webgpu/WebGPUDriver.h
-            src/webgpu/WebGPUSwapChain.cpp
-            src/webgpu/WebGPUSwapChain.h
             src/webgpu/WebGPUHandles.cpp
             src/webgpu/WebGPUHandles.h
+            src/webgpu/WebGPUSwapChain.cpp
+            src/webgpu/WebGPUSwapChain.h
+            src/webgpu/WGPUProgram.cpp
     )
     if (WIN32)
         list(APPEND SRCS src/webgpu/platform/WebGPUPlatformWindows.cpp)
@@ -507,7 +508,10 @@ if (APPLE OR LINUX)
         test/Arguments.cpp
         test/ImageExpectations.cpp
         test/Lifetimes.cpp
+        test/PlatformRunner.cpp
         test/Shader.cpp
+        test/SharedShaders.cpp
+        test/Skip.cpp
         test/test_FeedbackLoops.cpp
         test/test_Blit.cpp
         test/test_MissingRequiredAttributes.cpp
@@ -531,6 +535,9 @@ if (APPLE OR LINUX)
         filamat
         SPIRV
         spirv-cross-glsl)
+    # Create input/output directories for test result images.
+    file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/images/actual_images)
+    file(COPY test/expected_images DESTINATION ${CMAKE_CURRENT_BINARY_DIR}/images)
 endif()
 
 # TODO: Disabling IOS test due to breakage wrt glslang update

filament/backend/include/backend/BufferObjectStreamDescriptor.h

@@ -55,4 +55,9 @@ public:
 
 } // namespace filament::backend
 
+#if !defined(NDEBUG)
+utils::io::ostream& operator<<(utils::io::ostream& out,
+        const filament::backend::BufferObjectStreamDescriptor& b);
+#endif
+
 #endif // TNT_FILAMENT_BACKEND_BUFFEROBJECTSTREAMDESCRIPTOR_H

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

@@ -104,10 +104,6 @@ public:
 
         // 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();

filament/backend/include/backend/platforms/WebGPUPlatform.h

@@ -38,6 +38,12 @@ public:
 
     [[nodiscard]] wgpu::Instance& getInstance() noexcept { return mInstance; }
 
+    // TODO consider that this functionality is not WebGPU-specific, and thus could be
+    //      placed in a generic place and even reused across backends. Alternatively,
+    //      a 3rd party library could be considered. However, this was a simple and
+    //      quick change and works for now.
+    // gets the size (height and width) of the surface/window
+    [[nodiscard]] wgpu::Extent2D getSurfaceExtent(void* nativeWindow) const;
     // either returns a valid surface or panics
     [[nodiscard]] wgpu::Surface createSurface(void* nativeWindow, uint64_t flags);
     // either returns a valid adapter or panics

filament/backend/src/CommandStream.cpp

@@ -84,7 +84,7 @@ void CommandStream::execute(void* buffer) {
 
     Profiler profiler;
 
-    if (SYSTRACE_TAG) {
+    if constexpr (SYSTRACE_TAG) {
         if (UTILS_UNLIKELY(mUsePerformanceCounter)) {
             // we want to remove all this when tracing is completely disabled
             profiler.resetEvents(Profiler::EV_CPU_CYCLES  | Profiler::EV_BPU_MISSES);
@@ -100,7 +100,7 @@ void CommandStream::execute(void* buffer) {
         }
     });
 
-    if (SYSTRACE_TAG) {
+    if constexpr (SYSTRACE_TAG) {
         if (UTILS_UNLIKELY(mUsePerformanceCounter)) {
             // we want to remove all this when tracing is completely disabled
             profiler.stop();

filament/backend/src/opengl/OpenGLProgram.cpp

@@ -85,6 +85,7 @@ OpenGLProgram::~OpenGLProgram() noexcept {
         delete lazyInitializationData;
 
         ShaderCompilerService::terminate(mToken);
+        assert_invariant(!mToken);
     }
 
     delete [] mUniformsRecords;

filament/backend/src/opengl/ShaderCompilerService.h

@@ -24,23 +24,23 @@
 #include "OpenGLBlobCache.h"
 
 #include <backend/CallbackHandler.h>
+#include <backend/DriverEnums.h>
 #include <backend/Program.h>
 
 #include <utils/CString.h>
 #include <utils/FixedCapacityVector.h>
-#include <utils/Invocable.h>
 #include <utils/JobSystem.h>
 
-#include <atomic>
-#include <condition_variable>
-#include <deque>
+#include <array>
 #include <functional>
 #include <memory>
 #include <mutex>
-#include <thread>
+#include <tuple>
 #include <utility>
 #include <vector>
 
+#include <stdint.h>
+
 namespace filament::backend {
 
 class OpenGLDriver;
@@ -57,6 +57,8 @@ class ShaderCompilerService {
 
 public:
     using program_token_t = std::shared_ptr<OpenGLProgramToken>;
+    using shaders_t = std::array<GLuint, Program::SHADER_TYPE_COUNT>;
+    using shaders_source_t = std::array<utils::CString, Program::SHADER_TYPE_COUNT>;
 
     explicit ShaderCompilerService(OpenGLDriver& driver);
 
@@ -82,6 +84,7 @@ public:
     void tick();
 
     // Destroys a valid token and all associated resources. Used to "cancel" a program compilation.
+    // This function is not called if `initialize(token)` is already invoked.
     static void terminate(program_token_t& token);
 
     // stores a user data pointer in the token
@@ -90,6 +93,12 @@ public:
     // retrieves the user data pointer stored in the token
     static void* getUserData(const program_token_t& token) noexcept;
 
+    // Issue one callback handle.
+    CallbackManager::Handle issueCallbackHandle() const noexcept;
+
+    // Return a callback handle to the callback manager.
+    void submitCallbackHandle(CallbackManager::Handle handle) noexcept;
+
     // call the callback when all active programs are ready
     void notifyWhenAllProgramsAreReady(
             CallbackHandler* handler, CallbackHandler::Callback callback, void* user);
@@ -97,7 +106,7 @@ public:
 private:
     struct Job {
         template<typename FUNC>
-        Job(FUNC&& fn) : fn(std::forward<FUNC>(fn)) {}
+        Job(FUNC&& fn) : fn(std::forward<FUNC>(fn)) {} // NOLINT(*-explicit-constructor)
         Job(std::function<bool(Job const& job)> fn,
                 CallbackHandler* handler, void* user, CallbackHandler::Callback callback)
                 : fn(std::move(fn)), handler(handler), user(user), callback(callback) {
@@ -126,39 +135,49 @@ private:
     using ContainerType = std::tuple<CompilerPriorityQueue, program_token_t, Job>;
     std::vector<ContainerType> mRunAtNextTickOps;
 
-    GLuint initialize(ShaderCompilerService::program_token_t& token) noexcept;
+    GLuint initialize(program_token_t& token);
+    void ensureTokenIsReady(program_token_t const& token);
 
-    static void getProgramFromCompilerPool(program_token_t& token) noexcept;
-
-    static void compileShaders(
-            OpenGLContext& context,
-            Program::ShaderSource shadersSource,
-            utils::FixedCapacityVector<Program::SpecializationConstant> const& specializationConstants,
-            bool multiview,
-            std::array<GLuint, Program::SHADER_TYPE_COUNT>& outShaders,
-            std::array<utils::CString, Program::SHADER_TYPE_COUNT>& outShaderSourceCode) noexcept;
-
-    static void process_GOOGLE_cpp_style_line_directive(OpenGLContext& context,
-            char* source, size_t len) noexcept;
-
-    static void process_OVR_multiview2(OpenGLContext& context, int32_t eyeCount,
-            char* source, size_t len) noexcept;
-
-    static std::string_view process_ARB_shading_language_packing(OpenGLContext& context) noexcept;
-
-    static std::array<std::string_view, 3> splitShaderSource(std::string_view source) noexcept;
-
-    static GLuint linkProgram(OpenGLContext& context,
-            std::array<GLuint, Program::SHADER_TYPE_COUNT> shaders,
-            utils::FixedCapacityVector<std::pair<utils::CString, uint8_t>> const& attributes) noexcept;
-
-    static bool checkProgramStatus(program_token_t const& token) noexcept;
-
-    void runAtNextTick(CompilerPriorityQueue priority,
-            const program_token_t& token, Job job) noexcept;
+    void runAtNextTick(CompilerPriorityQueue priority, program_token_t const& token,
+            Job job) noexcept;
     void executeTickOps() noexcept;
-    bool cancelTickOp(program_token_t token) noexcept;
-    // order of insertion is important
+    bool cancelTickOp(program_token_t const& token) noexcept;
+
+    // Compile shaders with the given `shaderSource`. `gl.shaders` is always populated with valid
+    // shader IDs after this method. But this doesn't necessarily mean the shaders are successfully
+    // compiled. Errors can be checked by calling `checkCompileStatus` later.
+    static void compileShaders(OpenGLContext& context, Program::ShaderSource shadersSource,
+            utils::FixedCapacityVector<Program::SpecializationConstant> const&
+                    specializationConstants,
+            bool multiview, program_token_t const& token) noexcept;
+
+    // Check if the shader compilation is completed. You may want to call this when the extension
+    // `KHR_parallel_shader_compile` is enabled.
+    static bool isCompileCompleted(program_token_t const& token) noexcept;
+
+    // Check compilation status of the shaders and log errors on failure.
+    static void checkCompileStatus(program_token_t const& token) noexcept;
+
+    // Create a program by linking the compiled shaders. `gl.program` is always populated with a
+    // valid program ID after this method. But this doesn't necessarily mean the program is
+    // successfully linked. Errors can be checked by calling `checkLinkStatusAndCleanupShaders`
+    // later.
+    static void linkProgram(OpenGLContext const& context, program_token_t const& token) noexcept;
+
+    // Check if the program link is completed. You may want to call this when the extension
+    // `KHR_parallel_shader_compile` is enabled.
+    static bool isLinkCompleted(program_token_t const& token) noexcept;
+
+    // Check link status of the program and log errors on failure. Return the result of the link.
+    // Also cleanup shaders regardless of the result.
+    static bool checkLinkStatusAndCleanupShaders(program_token_t const& token) noexcept;
+
+    // Try caching the program if we haven't done it yet. Cache it only when the program is valid.
+    static void tryCachingProgram(OpenGLBlobCache& cache, OpenGLPlatform& platform,
+            program_token_t const& token) noexcept;
+
+    // Cleanup GL resources.
+    static void cleanupProgramAndShaders(program_token_t const& token) noexcept;
 };
 
 } // namespace filament::backend

filament/backend/src/ostream.cpp

@@ -15,6 +15,7 @@
  */
 
 #include <backend/BufferDescriptor.h>
+#include <backend/BufferObjectStreamDescriptor.h>
 #include <backend/DescriptorSetOffsetArray.h>
 #include <backend/DriverEnums.h>
 #include <backend/PipelineState.h>
@@ -437,6 +438,10 @@ io::ostream& operator<<(io::ostream& out, BufferDescriptor const& b) {
     << ", user=" << b.getUser() << " }";
 }
 
+io::ostream& operator<<(io::ostream& out, const BufferObjectStreamDescriptor& b) {
+    return out << "BufferObjectStreamDescriptor{ streams(" << b.mStreams.size() << ")=... }";
+}
+
 io::ostream& operator<<(io::ostream& out, PixelBufferDescriptor const& b) {
     BufferDescriptor const& base = static_cast<BufferDescriptor const&>(b);
     return out << "PixelBufferDescriptor{ " << base

filament/backend/src/vulkan/VulkanConstants.h

@@ -152,12 +152,12 @@ static_assert(FVK_ENABLED(FVK_DEBUG_VALIDATION));
 #elif FVK_ENABLED(FVK_DEBUG_SYSTRACE)
 
     #include <utils/Systrace.h>
-    
+
     #define FVK_SYSTRACE_CONTEXT()      SYSTRACE_CONTEXT()
     #define FVK_SYSTRACE_START(marker)  SYSTRACE_NAME_BEGIN(marker)
     #define FVK_SYSTRACE_END()          SYSTRACE_NAME_END()
-    #define FVK_SYSTRACE_SCOPE()        SYSTRACE_NAME(__func__)
-    #define FVK_PROFILE_MARKER(marker)  FVK_SYSTRACE_SCOPE()
+    #define FVK_SYSTRACE_SCOPE()        SYSTRACE_CALL()
+    #define FVK_PROFILE_MARKER(marker)  SYSTRACE_CALL()
 
 #else
     #define FVK_SYSTRACE_CONTEXT()

filament/backend/src/vulkan/VulkanContext.h

@@ -70,13 +70,11 @@ struct VulkanRenderPass {
 // context are stored in VulkanPlatform.
 struct VulkanContext {
 public:
-    inline uint32_t selectMemoryType(uint32_t flags, VkFlags reqs) const {
-        if ((reqs & VK_MEMORY_PROPERTY_PROTECTED_BIT) != 0) {
-            assert_invariant(isProtectedMemorySupported() == true);
-        }
+    static uint32_t selectMemoryType(VkPhysicalDeviceMemoryProperties const& memoryProperties,
+            uint32_t flags, VkFlags reqs) {
         for (uint32_t i = 0; i < VK_MAX_MEMORY_TYPES; i++) {
             if (flags & 1) {
-                if ((mMemoryProperties.memoryTypes[i].propertyFlags & reqs) == reqs) {
+                if ((memoryProperties.memoryTypes[i].propertyFlags & reqs) == reqs) {
                     return i;
                 }
             }
@@ -85,6 +83,13 @@ public:
         return (uint32_t) VK_MAX_MEMORY_TYPES;
     }
 
+    inline uint32_t selectMemoryType(uint32_t flags, VkFlags reqs) const {
+        if ((reqs & VK_MEMORY_PROPERTY_PROTECTED_BIT) != 0) {
+            assert_invariant(isProtectedMemorySupported());
+        }
+        return selectMemoryType(mMemoryProperties, flags, reqs);
+    }
+
     inline fvkutils::VkFormatList const& getAttachmentDepthStencilFormats() const {
         return mDepthStencilFormats;
     }
@@ -118,7 +123,7 @@ public:
     }
 
     inline bool isMultiviewEnabled() const noexcept {
-        return mMultiviewEnabled;
+        return mPhysicalDeviceVk11Features.multiview == VK_TRUE;
     }
 
     inline bool isClipDistanceSupported() const noexcept {
@@ -142,6 +147,9 @@ private:
     VkPhysicalDeviceProperties2 mPhysicalDeviceProperties = {
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2,
     };
+    VkPhysicalDeviceVulkan11Features mPhysicalDeviceVk11Features = {
+        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES,
+    };
     VkPhysicalDeviceFeatures2 mPhysicalDeviceFeatures = {
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2,
     };
@@ -154,7 +162,6 @@ private:
     };
     bool mDebugMarkersSupported = false;
     bool mDebugUtilsSupported = false;
-    bool mMultiviewEnabled = false;
     bool mLazilyAllocatedMemorySupported = false;
     bool mProtectedMemorySupported = false;
 

filament/backend/src/vulkan/VulkanDescriptorSetCache.cpp

@@ -282,18 +282,25 @@ void VulkanDescriptorSetCache::commit(VulkanCommandBuffer* commands,
     fvkutils::DescriptorSetMask curMask = setMask;
 
     auto& updateSets = mStashedSets;
-    auto& lastBoundSets = mLastBoundInfo.boundSets;
+    bool const pipelineLayoutIsSame = mLastBoundInfo.pipelineLayout == pipelineLayout;
 
-    setMask.forEachSetBit([&](size_t index) {
-        if (!updateSets[index] || updateSets[index] == lastBoundSets[index]) {
-            curMask.unset(index);
+    if (pipelineLayoutIsSame) {
+        auto& lastBoundSets = mLastBoundInfo.boundSets;
+        setMask.forEachSetBit([&](size_t index) {
+            if (!updateSets[index] || updateSets[index] == lastBoundSets[index]) {
+                curMask.unset(index);
+            }
+        });
+        if (curMask.none() &&
+                mLastBoundInfo.setMask == setMask && mLastBoundInfo.boundSets == updateSets) {
+            return;
         }
-    });
-
-    if (curMask.none() &&
-            (mLastBoundInfo.pipelineLayout == pipelineLayout && mLastBoundInfo.setMask == setMask &&
-                    mLastBoundInfo.boundSets == updateSets)) {
-        return;
+    } else {
+        setMask.forEachSetBit([&](size_t index) {
+            if (!updateSets[index]) {
+                curMask.unset(index);
+            }
+        });
     }
 
     curMask.forEachSetBit([&updateSets, commands, pipelineLayout](size_t index) {
@@ -301,7 +308,7 @@ void VulkanDescriptorSetCache::commit(VulkanCommandBuffer* commands,
         auto set = updateSets[index];
         VkCommandBuffer const cmdbuffer = commands->buffer();
         vkCmdBindDescriptorSets(cmdbuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, index,
-                1, &set->vkSet, set->uniqueDynamicUboCount, set->getOffsets()->data());
+                1, &set->getVkSet(), set->uniqueDynamicUboCount, set->getOffsets()->data());
         commands->acquire(set);
     });
 
@@ -329,7 +336,7 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
     }
     VkWriteDescriptorSet const descriptorWrite = {
         .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
-        .dstSet = set->vkSet,
+        .dstSet = set->getVkSet(),
         .dstBinding = binding,
         .descriptorCount = 1,
         .descriptorType = type,
@@ -342,10 +349,6 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
 void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
         uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
         VkSampler sampler) noexcept {
-    VkDescriptorImageInfo info{
-        .sampler = sampler,
-        .imageLayout = fvkutils::getVkLayout(texture->getDefaultLayout()),
-    };
     VkImageSubresourceRange range = texture->getPrimaryViewRange();
     VkImageViewType const expectedType = texture->getViewType();
     if (any(texture->usage & TextureUsage::DEPTH_ATTACHMENT) &&
@@ -355,12 +358,16 @@ void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescr
         range.levelCount = 1;
         range.layerCount = 1;
     }
-    info.imageView = texture->getView(range);
+    VkDescriptorImageInfo info{
+        .sampler = sampler,
+        .imageView = texture->getView(range),
+        .imageLayout = fvkutils::getVkLayout(texture->getDefaultLayout()),
+    };
 
     VkWriteDescriptorSet const descriptorWrite = {
         .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
         .pNext = nullptr,
-        .dstSet = set->vkSet,
+        .dstSet = set->getVkSet(),
         .dstBinding = binding,
         .descriptorCount = 1,
         .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
@@ -378,10 +385,10 @@ void VulkanDescriptorSetCache::updateInputAttachment(
 
 fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetCache::createSet(
         Handle<HwDescriptorSet> handle, fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-    auto const vkSet = mDescriptorPool->obtainSet(layout);
+    auto const vkSet = getVkSet(layout);
     auto const& count = layout->count;
     auto const vklayout = layout->getVkLayout();
-    return fvkmemory::resource_ptr<VulkanDescriptorSet>::make(mResourceManager, handle, vkSet,
+    auto set = fvkmemory::resource_ptr<VulkanDescriptorSet>::make(mResourceManager, handle,
             layout->bitmask.dynamicUbo, layout->count.dynamicUbo,
             [vkSet, count, vklayout, this](VulkanDescriptorSet*) {
                 // Note that mDescriptorPool could be gone due to terminate (when the backend shuts
@@ -390,10 +397,20 @@ fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetCache::createSet
                     mDescriptorPool->recycle(count, vklayout, vkSet);
                 }
             });
+    set->setVkSet(vkSet);
+    return set;
 }
 
-void VulkanDescriptorSetCache::gc() {
-    mStashedSets = {};
+VkDescriptorSet VulkanDescriptorSetCache::getVkSet(
+        fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
+    return mDescriptorPool->obtainSet(layout);
 }
 
+void VulkanDescriptorSetCache::manualRecyle(VulkanDescriptorSetLayout::Count const& count,
+        VkDescriptorSetLayout vklayout, VkDescriptorSet vkSet) {
+    mDescriptorPool->recycle(count, vklayout, vkSet);
+}
+
+void VulkanDescriptorSetCache::gc() { mStashedSets = {}; }
+
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanDescriptorSetCache.h

@@ -41,6 +41,8 @@ public:
             VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT;
 
     using DescriptorSetLayoutArray = VulkanDescriptorSetLayout::DescriptorSetLayoutArray;
+    using DescriptorSetArray =
+            std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>, UNIQUE_DESCRIPTOR_SET_COUNT>;
 
     VulkanDescriptorSetCache(VkDevice device, fvkmemory::ResourceManager* resourceManager);
     ~VulkanDescriptorSetCache();
@@ -68,14 +70,21 @@ public:
     fvkmemory::resource_ptr<VulkanDescriptorSet> createSet(Handle<HwDescriptorSet> handle,
             fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
 
+    // This method is only meant to be used with external samplers (or internally within this
+    // class).
+    VkDescriptorSet getVkSet(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
+
+    // This method is only meant to be used with external samplers.
+    void manualRecyle(VulkanDescriptorSetLayout::Count const& count, VkDescriptorSetLayout vklayout,
+            VkDescriptorSet vkSet);
+
+    DescriptorSetArray const& getBoundSets() const { return mStashedSets; }
+
     void gc();
 
 private:
     class DescriptorInfinitePool;
 
-    using DescriptorSetArray =
-            std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>, UNIQUE_DESCRIPTOR_SET_COUNT>;
-
     VkDevice mDevice;
     fvkmemory::ResourceManager* mResourceManager;
     std::unique_ptr<DescriptorInfinitePool> mDescriptorPool;

filament/backend/src/vulkan/VulkanDescriptorSetLayoutCache.cpp

@@ -18,6 +18,8 @@
 
 #include "VulkanHandles.h"
 
+#include <utils/Hash.h>
+
 namespace filament::backend {
 
 namespace {
@@ -58,6 +60,56 @@ uint32_t appendBindings(VkDescriptorSetLayoutBinding* toBind, VkDescriptorType t
     return count;
 }
 
+uint32_t appendSamplerBindings(VkDescriptorSetLayoutBinding* toBind,
+        fvkutils::SamplerBitmask const& mask, fvkutils::SamplerBitmask const& external,
+        utils::FixedCapacityVector<VkSampler> const& immutableSamplers) {
+    using Bitmask = fvkutils::SamplerBitmask;
+    uint32_t count = 0;
+    Bitmask alreadySeen;
+    uint8_t immutableIndex = 0;
+    size_t const immutableSamplerCount = immutableSamplers.size();
+    mask.forEachSetBit([&](size_t index) {
+        VkShaderStageFlags stages = 0;
+        uint32_t binding = 0;
+        if (index < fvkutils::getFragmentStageShift<Bitmask>()) {
+            binding = (uint32_t) index;
+            stages |= VK_SHADER_STAGE_VERTEX_BIT;
+            auto fragIndex = index + fvkutils::getFragmentStageShift<Bitmask>();
+            if (mask.test(fragIndex)) {
+                stages |= VK_SHADER_STAGE_FRAGMENT_BIT;
+                alreadySeen.set(fragIndex);
+            }
+        } else if (!alreadySeen.test(index)) {
+            // We are in fragment stage bits
+            binding = (uint32_t) (index - fvkutils::getFragmentStageShift<Bitmask>());
+            stages |= VK_SHADER_STAGE_FRAGMENT_BIT;
+        }
+
+        if (stages) {
+            toBind[count++] = {
+                .binding = binding,
+                .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+                .descriptorCount = 1,
+                .stageFlags = stages,
+                .pImmutableSamplers = external[index] && immutableSamplerCount > immutableIndex
+                                              ? &immutableSamplers[immutableIndex++]
+                                              : nullptr,
+            };
+        }
+    });
+    return count;
+}
+
+uint64_t computeImmutableSamplerHash(utils::FixedCapacityVector<VkSampler> const& samplers) {
+    size_t const size = samplers.size();
+    if (size == 0) {
+        return 0;
+    } else if (size == 1) {
+        return (uint64_t) samplers[0];
+    }
+    return utils::hash::murmur3((uint32_t*) samplers.data(), samplers.size() * 2, 0);
+}
+
 } // anonymous namespace
 
 VulkanDescriptorSetLayoutCache::VulkanDescriptorSetLayoutCache(VkDevice device,
@@ -73,37 +125,44 @@ void VulkanDescriptorSetLayoutCache::terminate() noexcept {
     }
 }
 
+VkDescriptorSetLayout VulkanDescriptorSetLayoutCache::getVkLayout(
+        VulkanDescriptorSetLayout::Bitmask const& bitmasks,
+        utils::FixedCapacityVector<VkSampler> immutableSamplers) {
+    LayoutKey key = {
+        .bitmask = bitmasks,
+        .immutableSamplerHash = computeImmutableSamplerHash(immutableSamplers),
+    };
+    if (auto itr = mVkLayouts.find(key); itr != mVkLayouts.end()) {
+        return itr->second;
+    }
+
+    VkDescriptorSetLayoutBinding toBind[VulkanDescriptorSetLayout::MAX_BINDINGS];
+    uint32_t count = 0;
+    count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC,
+            bitmasks.dynamicUbo);
+    count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, bitmasks.ubo);
+    count += appendSamplerBindings(&toBind[count], bitmasks.sampler, bitmasks.externalSampler,
+            immutableSamplers);
+    count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
+            bitmasks.inputAttachment);
+
+    assert_invariant(count != 0 && "Need at least one binding for descriptor set layout.");
+    VkDescriptorSetLayoutCreateInfo dlinfo = {
+        .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
+        .bindingCount = count,
+        .pBindings = toBind,
+    };
+    VkDescriptorSetLayout vklayout;
+    vkCreateDescriptorSetLayout(mDevice, &dlinfo, VKALLOC, &vklayout);
+    mVkLayouts[key] = vklayout;
+    return vklayout;
+}
+
 fvkmemory::resource_ptr<VulkanDescriptorSetLayout> VulkanDescriptorSetLayoutCache::createLayout(
         Handle<HwDescriptorSetLayout> handle, backend::DescriptorSetLayout&& info) {
     auto layout = fvkmemory::resource_ptr<VulkanDescriptorSetLayout>::make(mResourceManager, handle,
             info);
-    VkDescriptorSetLayout vklayout = VK_NULL_HANDLE;
-    auto const& bitmasks = layout->bitmask;
-    if (auto itr = mVkLayouts.find(bitmasks); itr != mVkLayouts.end()) {
-        vklayout = itr->second;
-    } else {
-        VkDescriptorSetLayoutBinding toBind[VulkanDescriptorSetLayout::MAX_BINDINGS];
-        uint32_t count = 0;
-        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC,
-                bitmasks.dynamicUbo);
-        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, bitmasks.ubo);
-        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
-                bitmasks.sampler);
-        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
-                bitmasks.inputAttachment);
-
-        assert_invariant(count != 0 && "Need at least one binding for descriptor set layout.");
-        VkDescriptorSetLayoutCreateInfo dlinfo = {
-            .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
-            .pNext = nullptr,
-            .bindingCount = count,
-            .pBindings = toBind,
-        };
-
-        vkCreateDescriptorSetLayout(mDevice, &dlinfo, VKALLOC, &vklayout);
-        mVkLayouts[bitmasks] = vklayout;
-    }
-    layout->setVkLayout(vklayout);
+    layout->setVkLayout(getVkLayout(layout->bitmask));
     return layout;
 }
 

filament/backend/src/vulkan/VulkanDescriptorSetLayoutCache.h

@@ -26,12 +26,11 @@
 #include <backend/TargetBufferInfo.h>
 
 #include <utils/bitset.h>
+#include <utils/FixedCapacityVector.h>
 
 #include <bluevk/BlueVK.h>
 #include <tsl/robin_map.h>
 
-#include <memory>
-
 namespace filament::backend {
 
 class VulkanDescriptorSetLayoutCache {
@@ -41,21 +40,34 @@ public:
 
     void terminate() noexcept;
 
+    // Just a wrapper around getVkLayout()
     fvkmemory::resource_ptr<VulkanDescriptorSetLayout> createLayout(
             Handle<HwDescriptorSetLayout> handle, backend::DescriptorSetLayout&& info);
 
+    // This method is meant to be used with external samplers
+    VkDescriptorSetLayout getVkLayout(VulkanDescriptorSetLayout::Bitmask const& bitmasks,
+            utils::FixedCapacityVector<VkSampler> immutableSamplers = {});
+
 private:
     VkDevice mDevice;
     fvkmemory::ResourceManager* mResourceManager;
 
-    using BitmaskGroup = VulkanDescriptorSetLayout::Bitmask;
-    using BitmaskGroupHashFn = utils::hash::MurmurHashFn<BitmaskGroup>;
-    struct BitmaskGroupEqual {
-        bool operator()(BitmaskGroup const& k1, BitmaskGroup const& k2) const { return k1 == k2; }
+    struct LayoutKey {
+        // this describes the layout using bitset.
+        VulkanDescriptorSetLayout::Bitmask bitmask = {};
+        // number of immutable samplers can be arbitrary; so we hash them into 64-bit.
+        uint64_t immutableSamplerHash = 0;
+    };
+    static_assert(sizeof(LayoutKey) == 48);
+
+    using LayoutKeyHashFn = utils::hash::MurmurHashFn<LayoutKey>;
+    struct LayoutKeyEqual {
+        bool operator()(LayoutKey const& k1, LayoutKey const& k2) const {
+            return k1.bitmask == k2.bitmask && k1.immutableSamplerHash == k2.immutableSamplerHash;
+        }
     };
 
-    tsl::robin_map<BitmaskGroup, VkDescriptorSetLayout, BitmaskGroupHashFn, BitmaskGroupEqual>
-            mVkLayouts;
+    tsl::robin_map<LayoutKey, VkDescriptorSetLayout, LayoutKeyHashFn, LayoutKeyEqual> mVkLayouts;
 };
 
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -327,7 +327,7 @@ void VulkanDriver::terminate() {
 
     mStagePool.terminate();
     mPipelineCache.terminate();
-    mFramebufferCache.reset();
+    mFramebufferCache.terminate();
     mSamplerCache.terminate();
     mDescriptorSetLayoutCache.terminate();
     mDescriptorSetCache.terminate();
@@ -1523,7 +1523,7 @@ void VulkanDriver::makeCurrent(Handle<HwSwapChain> drawSch, Handle<HwSwapChain>
     swapChain->acquire(resized);
 
     if (resized) {
-        mFramebufferCache.reset();
+        mFramebufferCache.resetFramebuffers();
     }
 
     if (UTILS_LIKELY(mDefaultRenderTarget)) {

filament/backend/src/vulkan/VulkanFboCache.cpp

@@ -340,15 +340,21 @@ VkRenderPass VulkanFboCache::getRenderPass(RenderPassKey const& config) noexcept
     return renderPass;
 }
 
-void VulkanFboCache::reset() noexcept {
-    for (auto pair : mFramebufferCache) {
+void VulkanFboCache::resetFramebuffers() noexcept {
+    for (const auto& pair: mFramebufferCache) {
         mRenderPassRefCount[pair.first.renderPass]--;
         vkDestroyFramebuffer(mDevice, pair.second.handle, VKALLOC);
     }
     mFramebufferCache.clear();
-    for (auto pair : mRenderPassCache) {
+}
+
+void VulkanFboCache::terminate() noexcept {
+    resetFramebuffers();
+
+    for (const auto& pair: mRenderPassCache) {
         vkDestroyRenderPass(mDevice, pair.second.handle, VKALLOC);
     }
+    mRenderPassRefCount.clear();
     mRenderPassCache.clear();
 }
 

filament/backend/src/vulkan/VulkanFboCache.h

@@ -106,8 +106,11 @@ public:
     // Evicts old unused Vulkan objects. Call this once per frame.
     void gc() noexcept;
 
+    // Frees all Framebuffer objects. Call this every time a the swapchain is resized
+    void resetFramebuffers() noexcept;
+
     // Frees all Vulkan objects. Call this during shutdown before the device is destroyed.
-    void reset() noexcept;
+    void terminate() noexcept;
 
 private:
     VkDevice mDevice;

filament/backend/src/vulkan/VulkanHandles.cpp

@@ -30,6 +30,7 @@
 
 #include <backend/platforms/VulkanPlatform.h>
 
+#include <utils/compiler.h> // UTILS_FALLTHROUGH
 #include <utils/Panic.h>    // ASSERT_POSTCONDITION
 
 using namespace bluevk;
@@ -89,8 +90,9 @@ BitmaskGroup fromBackendLayout(DescriptorSetLayout const& layout) {
                 }
                 break;
             }
-            // TODO: properly handle external sampler
             case DescriptorType::SAMPLER_EXTERNAL:
+                fromStageFlags(binding.stageFlags, binding.binding, mask.externalSampler);
+                UTILS_FALLTHROUGH;
             case DescriptorType::SAMPLER: {
                 fromStageFlags(binding.stageFlags, binding.binding, mask.sampler);
                 break;

filament/backend/src/vulkan/VulkanHandles.h

@@ -66,18 +66,21 @@ struct VulkanDescriptorSetLayout : public HwDescriptorSetLayout, fvkmemory::Reso
 
     // The bitmask representation of a set layout.
     struct Bitmask {
-        // TODO: better utiltize the space below and use bitset instead.
         fvkutils::UniformBufferBitmask ubo;         // 8 bytes
         fvkutils::UniformBufferBitmask dynamicUbo;  // 8 bytes
         fvkutils::SamplerBitmask sampler;           // 8 bytes
         fvkutils::InputAttachmentBitmask inputAttachment; // 8 bytes
 
+        // This is a subset of the bitmask.sampler field.
+        fvkutils::SamplerBitmask externalSampler; // 8 bytes
+
         bool operator==(Bitmask const& right) const {
             return ubo == right.ubo && dynamicUbo == right.dynamicUbo && sampler == right.sampler &&
-                   inputAttachment == right.inputAttachment;
+                   inputAttachment == right.inputAttachment &&
+                   externalSampler == right.externalSampler;
         }
     };
-    static_assert(sizeof(Bitmask) == 32);
+    static_assert(sizeof(Bitmask) == 40);
 
     // This is a convenience struct to quickly check layout compatibility in terms of descriptor set
     // pools.
@@ -119,10 +122,16 @@ struct VulkanDescriptorSetLayout : public HwDescriptorSetLayout, fvkmemory::Reso
 
     VulkanDescriptorSetLayout(DescriptorSetLayout const& layout);
 
+    // Note that we don't destroy the vklayout. This is done by the layout cache.
     ~VulkanDescriptorSetLayout() = default;
 
-    VkDescriptorSetLayout getVkLayout() const { return mVkLayout; }
-    void setVkLayout(VkDescriptorSetLayout vklayout) { mVkLayout = vklayout; }
+    VkDescriptorSetLayout const& getVkLayout() const noexcept { return mVkLayout; }
+
+    // It is possible to have the layout switch out due to AHardwarebuffer (external image) format
+    // changes.
+    void setVkLayout(VkDescriptorSetLayout vklayout) noexcept { mVkLayout = vklayout; }
+
+    bool hasExternalSamplers() const noexcept { return bitmask.externalSampler.count() > 0; }
 
     Bitmask const bitmask;
     Count const count;
@@ -137,12 +146,11 @@ public:
     // can use to repackage the vk handle.
     using OnRecycle = std::function<void(VulkanDescriptorSet*)>;
 
-    VulkanDescriptorSet(VkDescriptorSet rawSet,
+    VulkanDescriptorSet(
             fvkutils::UniformBufferBitmask const& dynamicUboMask,
             uint8_t uniqueDynamicUboCount,
             OnRecycle&& onRecycleFn)
-        : vkSet(rawSet),
-          dynamicUboMask(dynamicUboMask),
+        : dynamicUboMask(dynamicUboMask),
           uniqueDynamicUboCount(uniqueDynamicUboCount),
           mOnRecycleFn(std::move(onRecycleFn)) {}
 
@@ -152,6 +160,13 @@ public:
         }
     }
 
+    VkDescriptorSet const& getVkSet() const noexcept {
+        return mVkSet;
+    }
+
+    // Note that the only case where you'd set it more than once is with external images/samplers.
+    void setVkSet(VkDescriptorSet vkset) noexcept { mVkSet = vkset; }
+
     void setOffsets(backend::DescriptorSetOffsetArray&& offsets) noexcept {
         mOffsets = std::move(offsets);
     }
@@ -163,11 +178,11 @@ public:
     void acquire(fvkmemory::resource_ptr<VulkanTexture> texture);
     void acquire(fvkmemory::resource_ptr<VulkanBufferObject> buffer);
 
-    VkDescriptorSet const vkSet;
     fvkutils::UniformBufferBitmask const dynamicUboMask;
     uint8_t const uniqueDynamicUboCount;
 
 private:
+    VkDescriptorSet mVkSet = VK_NULL_HANDLE;
     backend::DescriptorSetOffsetArray mOffsets;
     std::vector<fvkmemory::resource_ptr<fvkmemory::Resource>> mResources;
     OnRecycle mOnRecycleFn;

filament/backend/src/vulkan/VulkanPipelineCache.cpp

@@ -35,7 +35,12 @@ using namespace bluevk;
 namespace filament::backend {
 
 VulkanPipelineCache::VulkanPipelineCache(VkDevice device)
-        : mDevice(device) {}
+    : mDevice(device) {
+    VkPipelineCacheCreateInfo createInfo = {
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO,
+    };
+    bluevk::vkCreatePipelineCache(mDevice, &createInfo, VKALLOC, &mPipelineCache);
+}
 
 void VulkanPipelineCache::bindLayout(VkPipelineLayout layout) noexcept {
     mPipelineRequirements.layout = layout;
@@ -215,7 +220,7 @@ VulkanPipelineCache::PipelineCacheEntry* VulkanPipelineCache::createPipeline() n
     PipelineCacheEntry cacheEntry = {
         .lastUsed = mCurrentTime,
     };
-    VkResult error = vkCreateGraphicsPipelines(mDevice, VK_NULL_HANDLE, 1, &pipelineCreateInfo,
+    VkResult error = vkCreateGraphicsPipelines(mDevice, mPipelineCache, 1, &pipelineCreateInfo,
             VKALLOC, &cacheEntry.handle);
     assert_invariant(error == VK_SUCCESS);
     if (error != VK_SUCCESS) {
@@ -271,6 +276,8 @@ void VulkanPipelineCache::terminate() noexcept {
     }
     mPipelines.clear();
     mBoundPipeline = {};
+
+    vkDestroyPipelineCache(mDevice, mPipelineCache, VKALLOC);
 }
 
 void VulkanPipelineCache::gc() noexcept {

filament/backend/src/vulkan/VulkanPipelineCache.h

@@ -198,6 +198,10 @@ private:
     // Immutable state.
     VkDevice mDevice = VK_NULL_HANDLE;
 
+    // Vuklan Driver pipeline cache handle. In the cases a pipeline has been  evicted by the `gc`,
+    // recreating the same pipeline is cheaper, helping with frame stalling.
+    VkPipelineCache mPipelineCache = VK_NULL_HANDLE;
+
     // Current requirements for the pipeline layout, pipeline, and descriptor sets.
     PipelineKey mPipelineRequirements = {};
 

filament/backend/src/vulkan/VulkanSamplerCache.cpp

@@ -33,8 +33,15 @@ VkSampler VulkanSamplerCache::getSampler(Params params) noexcept {
     if (UTILS_LIKELY(iter != mCache.end())) {
         return iter->second;
     }
+    VkSamplerYcbcrConversionInfo ycbcrConversion = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO,
+        .conversion = params.conversion,
+    };
+
     auto const& samplerParams = params.sampler;
-    VkSamplerCreateInfo samplerInfo{ .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
+    VkSamplerCreateInfo samplerInfo{
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
+        .pNext = params.conversion != VK_NULL_HANDLE ? &ycbcrConversion : VK_NULL_HANDLE,
         .magFilter = fvkutils::getFilter(samplerParams.filterMag),
         .minFilter = fvkutils::getFilter(samplerParams.filterMin),
         .mipmapMode = fvkutils::getMipmapMode(samplerParams.filterMin),
@@ -48,7 +55,8 @@ VkSampler VulkanSamplerCache::getSampler(Params params) noexcept {
         .minLod = 0.0f,
         .maxLod = fvkutils::getMaxLod(samplerParams.filterMin),
         .borderColor = VK_BORDER_COLOR_INT_OPAQUE_BLACK,
-        .unnormalizedCoordinates = VK_FALSE };
+        .unnormalizedCoordinates = VK_FALSE,
+    };
     VkSampler sampler;
     VkResult result = vkCreateSampler(mDevice, &samplerInfo, VKALLOC, &sampler);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "Unable to create sampler."

filament/backend/src/vulkan/VulkanSwapChain.cpp

@@ -108,11 +108,6 @@ void VulkanSwapChain::present() {
 
     mCommands->flush();
 
-    // call the image ready wait function
-    if (mExplicitImageReadyWait != nullptr) {
-        mExplicitImageReadyWait(swapChain);
-    }
-
     // We only present if it is not headless. No-op for headless.
     if (!mHeadless) {
         VkSemaphore const finishedDrawing = mCommands->acquireFinishedSignal();
@@ -146,7 +141,6 @@ void VulkanSwapChain::acquire(bool& resized) {
     VulkanPlatform::ImageSyncData imageSyncData;
     VkResult const result = mPlatform->acquire(swapChain, &imageSyncData);
     mCurrentSwapIndex = imageSyncData.imageIndex;
-    mExplicitImageReadyWait = imageSyncData.explicitImageReadyWait;
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS || result == VK_SUBOPTIMAL_KHR)
             << "Cannot acquire in swapchain. error=" << static_cast<int32_t>(result);
     if (imageSyncData.imageReadySemaphore != VK_NULL_HANDLE) {

filament/backend/src/vulkan/VulkanSwapChain.h

@@ -49,7 +49,9 @@ struct VulkanSwapChain : public HwSwapChain, fvkmemory::Resource {
 
     void present();
 
-    void acquire(bool& reized);
+    // Acquire a new image from the swapchain. If the image is not available it would wait until it
+    // is.
+    void acquire(bool& resized);
 
     fvkmemory::resource_ptr<VulkanTexture> getCurrentColor() const noexcept {
         uint32_t const imageIndex = mCurrentSwapIndex;
@@ -99,7 +101,6 @@ private:
     VkExtent2D mExtent;
     uint32_t mLayerCount;
     uint32_t mCurrentSwapIndex;
-    std::function<void(Platform::SwapChain* handle)> mExplicitImageReadyWait = nullptr;
     bool mAcquired;
     bool mIsFirstRenderPass;
 };

filament/backend/src/vulkan/VulkanYcbcrConversionCache.cpp

@@ -43,7 +43,7 @@ VkSamplerYcbcrConversion VulkanYcbcrConversionCache::getConversion(
     TextureSwizzle const swizzleArray[] = { chroma.r, chroma.g, chroma.b, chroma.a };
     VkSamplerYcbcrConversionCreateInfo conversionInfo = {
         .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_CREATE_INFO,
-        .format = VK_FORMAT_UNDEFINED,
+        .format = fvkutils::getVkFormat(params.format),
         .ycbcrModel = fvkutils::getYcbcrModelConversion(chroma.ycbcrModel),
         .ycbcrRange = fvkutils::getYcbcrRange(chroma.ycbcrRange),
         .components = fvkutils::getSwizzleMap(swizzleArray),
@@ -52,6 +52,7 @@ VkSamplerYcbcrConversion VulkanYcbcrConversionCache::getConversion(
         .chromaFilter = fvkutils::getFilter(chroma.chromaFilter),
     };
 
+    // We could put this in the platform class, but that seems like a bit of an overkill
 #if defined(__ANDROID__)
     VkExternalFormatANDROID externalFormat = {
         .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
@@ -59,6 +60,7 @@ VkSamplerYcbcrConversion VulkanYcbcrConversionCache::getConversion(
     };
     if (params.externalFormat) {
         conversionInfo.pNext = &externalFormat;
+        conversionInfo.format = VK_FORMAT_UNDEFINED;
     }
 #endif
 

filament/backend/src/vulkan/VulkanYcbcrConversionCache.h

@@ -31,7 +31,8 @@ class VulkanYcbcrConversionCache {
 public:
     struct Params {
         SamplerYcbcrConversion conversion = {};
-        uint32_t padding = 0;
+        TextureFormat format = {};
+        uint16_t padding = 0;
         uint64_t externalFormat = 0;
     };
     static_assert(sizeof(Params) == 16);

filament/backend/src/vulkan/platform/VulkanPlatform.cpp

@@ -211,6 +211,7 @@ ExtensionSet getDeviceExtensions(VkPhysicalDevice device) {
         VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME,
         VK_KHR_SAMPLER_YCBCR_CONVERSION_EXTENSION_NAME,
         VK_ANDROID_EXTERNAL_MEMORY_ANDROID_HARDWARE_BUFFER_EXTENSION_NAME,
+        VK_EXT_QUEUE_FAMILY_FOREIGN_EXTENSION_NAME,
 #endif
         // MoltenVk is the only non-conformant implementation we're interested in.
 #if defined(__APPLE__)
@@ -326,7 +327,9 @@ VkInstance createInstance(ExtensionSet const& requiredExts) {
 }
 
 VkDevice createLogicalDevice(VkPhysicalDevice physicalDevice,
-        VkPhysicalDeviceFeatures2 const& features, uint32_t graphicsQueueFamilyIndex,
+        VkPhysicalDeviceFeatures2 const& features,
+        VkPhysicalDeviceVulkan11Features const& vk11Features,
+        uint32_t graphicsQueueFamilyIndex,
         uint32_t protectedGraphicsQueueFamilyIndex, ExtensionSet const& deviceExtensions,
         bool requestImageView2DOn3DImage) {
     VkDevice device;
@@ -359,14 +362,28 @@ VkDevice createLogicalDevice(VkPhysicalDevice physicalDevice,
 
     // We could simply enable all supported features, but since that may have performance
     // consequences let's just enable the features we need.
-    VkPhysicalDeviceFeatures enabledFeatures{
+    VkPhysicalDeviceFeatures enabledFeatures = {
         .depthClamp = features.features.depthClamp,
         .samplerAnisotropy = features.features.samplerAnisotropy,
         .textureCompressionETC2 = features.features.textureCompressionETC2,
         .textureCompressionBC = features.features.textureCompressionBC,
         .shaderClipDistance = features.features.shaderClipDistance,
     };
-    deviceCreateInfo.pEnabledFeatures = &enabledFeatures;
+
+    VkPhysicalDeviceFeatures2 enabledFeatures2 = {
+        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2,
+        .features = enabledFeatures,
+    };
+    chainStruct(&deviceCreateInfo, &enabledFeatures2);
+
+    VkPhysicalDeviceVulkan11Features enabledVk11Features = {
+        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES,
+        .multiview = vk11Features.multiview,
+#if defined(__ANDROID__)
+        .samplerYcbcrConversion = vk11Features.samplerYcbcrConversion,
+#endif
+    };
+    chainStruct(&deviceCreateInfo, &enabledVk11Features);
 
     deviceCreateInfo.enabledExtensionCount = (uint32_t) requestExtensions.size();
     deviceCreateInfo.ppEnabledExtensionNames = requestExtensions.data();
@@ -383,7 +400,7 @@ VkDevice createLogicalDevice(VkPhysicalDevice physicalDevice,
 
     VkPhysicalDeviceMultiviewFeaturesKHR multiview = {
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES_KHR,
-        .multiview = VK_TRUE,
+        .multiview = vk11Features.multiview,
         .multiviewGeometryShader = VK_FALSE,
         .multiviewTessellationShader = VK_FALSE,
     };
@@ -734,6 +751,7 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_PROPERTIES,
     };
     chainStruct(&context.mPhysicalDeviceFeatures, &queryProtectedMemoryFeatures);
+    chainStruct(&context.mPhysicalDeviceFeatures, &context.mPhysicalDeviceVk11Features);
     chainStruct(&context.mPhysicalDeviceProperties, &protectedMemoryProperties);
 
     // Initialize the following fields: physicalDeviceProperties, memoryProperties,
@@ -795,10 +813,10 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
     }
 
     if (mImpl->mDevice == VK_NULL_HANDLE) {
-        mImpl->mDevice =
-                createLogicalDevice(mImpl->mPhysicalDevice, context.mPhysicalDeviceFeatures,
-                        mImpl->mGraphicsQueueFamilyIndex, mImpl->mProtectedGraphicsQueueFamilyIndex,
-                        deviceExts, requestPortabilitySubsetImageView2DOn3DImage);
+        mImpl->mDevice = createLogicalDevice(mImpl->mPhysicalDevice,
+                context.mPhysicalDeviceFeatures, context.mPhysicalDeviceVk11Features,
+                mImpl->mGraphicsQueueFamilyIndex, mImpl->mProtectedGraphicsQueueFamilyIndex,
+                deviceExts, requestPortabilitySubsetImageView2DOn3DImage);
     }
 
     assert_invariant(mImpl->mDevice != VK_NULL_HANDLE);
@@ -826,12 +844,10 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
     if (!mImpl->mSharedContext) {
         context.mDebugUtilsSupported = setContains(instExts, VK_EXT_DEBUG_UTILS_EXTENSION_NAME);
         context.mDebugMarkersSupported = setContains(deviceExts, VK_EXT_DEBUG_MARKER_EXTENSION_NAME);
-        context.mMultiviewEnabled = setContains(deviceExts, VK_KHR_MULTIVIEW_EXTENSION_NAME);
     } else {
         VulkanSharedContext const* scontext = (VulkanSharedContext const*) sharedContext;
         context.mDebugUtilsSupported = scontext->debugUtilsSupported;
         context.mDebugMarkersSupported = scontext->debugMarkersSupported;
-        context.mMultiviewEnabled = scontext->multiviewSupported;
     }
 
     // Check the availability of lazily allocated memory

filament/backend/src/vulkan/platform/VulkanPlatformApple.mm

@@ -24,28 +24,12 @@
 #include <bluevk/BlueVK.h>
 
 // Platform specific includes and defines
-#if defined(__APPLE__)
-    #include <Cocoa/Cocoa.h>
-    #import <Metal/Metal.h>
-    #import <QuartzCore/CAMetalLayer.h>
+#include <Cocoa/Cocoa.h>
+#import <Metal/Metal.h>
+#import <QuartzCore/CAMetalLayer.h>
 
-    #ifndef VK_MVK_macos_surface
-         #error VK_MVK_macos_surface is not defined
-    #endif
-#elif defined(FILAMENT_IOS)
-    // Metal is not available when building for the iOS simulator on Desktop.
-    #define METAL_AVAILABLE __has_include(<QuartzCore/CAMetalLayer.h>)
-    #if METAL_AVAILABLE
-        #import <Metal/Metal.h>
-        #import <QuartzCore/CAMetalLayer.h>
-    #endif
-
-    #ifndef VK_MVK_ios_surface
-        #error VK_MVK_ios_surface is not defined
-    #endif
-    #define METALVIEW_TAG 255
-#else
-    #error Not a supported Apple + Vulkan platform
+#ifndef VK_MVK_macos_surface
+    #error VK_MVK_macos_surface is not defined
 #endif
 
 using namespace bluevk;
@@ -54,11 +38,7 @@ namespace filament::backend {
 
 VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl() {
     ExtensionSet const ret = {
-    #if defined(__APPLE__)
         VK_MVK_MACOS_SURFACE_EXTENSION_NAME,  // TODO: replace with VK_EXT_metal_surface
-    #elif defined(FILAMENT_IOS) && defined(METAL_AVAILABLE)
-        VK_MVK_IOS_SURFACE_EXTENSION_NAME,
-    #endif
     };
     return ret;
 }
@@ -90,36 +70,20 @@ VkImageView VulkanPlatform::createExternalImageViewImpl(VkDevice device,
 VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHRImpl(void* nativeWindow,
         VkInstance instance, uint64_t flags) noexcept {
     VkSurfaceKHR surface;
-    #if defined(__APPLE__)
-        NSView* nsview = (__bridge NSView*) nativeWindow;
-        FILAMENT_CHECK_POSTCONDITION(nsview) << "Unable to obtain Metal-backed NSView.";
+    NSView* nsview = (__bridge NSView*) nativeWindow;
+    FILAMENT_CHECK_POSTCONDITION(nsview) << "Unable to obtain Metal-backed NSView.";
 
-        // Create the VkSurface.
-        FILAMENT_CHECK_POSTCONDITION(vkCreateMacOSSurfaceMVK)
-                << "Unable to load vkCreateMacOSSurfaceMVK.";
-        VkMacOSSurfaceCreateInfoMVK createInfo = {};
-        createInfo.sType = VK_STRUCTURE_TYPE_MACOS_SURFACE_CREATE_INFO_MVK;
-        createInfo.pView = (__bridge void*) nsview;
-        VkResult result = vkCreateMacOSSurfaceMVK((VkInstance) instance, &createInfo, VKALLOC,
-                (VkSurfaceKHR*) &surface);
-        FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-                << "vkCreateMacOSSurfaceMVK. error=" << static_cast<int32_t>(result);
-#elif defined(FILAMENT_IOS) && defined(METAL_AVAILABLE)
-    CAMetalLayer* metalLayer = (CAMetalLayer*) nativeWindow;
     // Create the VkSurface.
-    FILAMENT_CHECK_POSTCONDITION(vkCreateIOSSurfaceMVK)
-            << "Unable to load vkCreateIOSSurfaceMVK function.";
-    VkIOSSurfaceCreateInfoMVK createInfo = {};
-    createInfo.sType = VK_STRUCTURE_TYPE_IOS_SURFACE_CREATE_INFO_MVK;
-    createInfo.pNext = NULL;
-    createInfo.flags = 0;
-    createInfo.pView = metalLayer;
-    VkResult result = vkCreateIOSSurfaceMVK((VkInstance) instance, &createInfo, VKALLOC,
+    FILAMENT_CHECK_POSTCONDITION(vkCreateMacOSSurfaceMVK)
+            << "Unable to load vkCreateMacOSSurfaceMVK.";
+    VkMacOSSurfaceCreateInfoMVK createInfo = {};
+    createInfo.sType = VK_STRUCTURE_TYPE_MACOS_SURFACE_CREATE_INFO_MVK;
+    createInfo.pView = (__bridge void*) nsview;
+    VkResult result = vkCreateMacOSSurfaceMVK((VkInstance) instance, &createInfo, VKALLOC,
             (VkSurfaceKHR*) &surface);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-            << "vkCreateIOSSurfaceMVK failed. error=" << static_cast<int32_t>(result);
-#endif
-        return std::make_tuple(surface, VkExtent2D{});
+            << "vkCreateMacOSSurfaceMVK. error=" << static_cast<int32_t>(result);
+    return std::make_tuple(surface, VkExtent2D{});
 }
 
 } // namespace filament::backend

filament/backend/src/vulkan/utils/Conversion.cpp

@@ -668,4 +668,111 @@ VkShaderStageFlags getShaderStageFlags(ShaderStageFlags stageFlags) {
     return flags;
 }
 
+VkSamplerYcbcrModelConversion getYcbcrModelConversion(
+    SamplerYcbcrModelConversion model) {
+    switch (model) {
+    case SamplerYcbcrModelConversion::RGB_IDENTITY:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY;
+    case SamplerYcbcrModelConversion::YCBCR_IDENTITY:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY;
+    case SamplerYcbcrModelConversion::YCBCR_709:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709;
+    case SamplerYcbcrModelConversion::YCBCR_601:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601;
+    case SamplerYcbcrModelConversion::YCBCR_2020:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020;
+    default:
+        assert_invariant(false &&
+            "Unknown data type, conversion is not supported.");
+    }
+}
+
+VkSamplerYcbcrRange getYcbcrRange(SamplerYcbcrRange range) {
+    switch (range) {
+    case SamplerYcbcrRange::ITU_FULL:
+        return VK_SAMPLER_YCBCR_RANGE_ITU_FULL;
+    case SamplerYcbcrRange::ITU_NARROW:
+        return VK_SAMPLER_YCBCR_RANGE_ITU_NARROW;
+    default:
+        assert_invariant(false &&
+            "Unknown data type, conversion is not supported.");
+    }
+}
+
+VkChromaLocation getChromaLocation(ChromaLocation loc) {
+    switch (loc) {
+    case ChromaLocation::COSITED_EVEN:
+        return VK_CHROMA_LOCATION_COSITED_EVEN;
+    case ChromaLocation::MIDPOINT:
+        return VK_CHROMA_LOCATION_MIDPOINT;
+    default:
+        assert_invariant(false &&
+            "Unknown data type, conversion is not supported.");
+    }
+}
+
+SamplerYcbcrModelConversion getYcbcrModelConversionFilament(VkSamplerYcbcrModelConversion model) {
+    switch (model) {
+        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY:
+            return SamplerYcbcrModelConversion::RGB_IDENTITY;
+        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY:
+            return SamplerYcbcrModelConversion::YCBCR_IDENTITY;
+        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709:
+            return SamplerYcbcrModelConversion::YCBCR_709;
+        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601:
+            return SamplerYcbcrModelConversion::YCBCR_601;
+        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020:
+            return SamplerYcbcrModelConversion::YCBCR_2020;
+        default:
+            assert_invariant(false && "Unknown data type, conversion is not supported.");
+            return {};
+    }
+}
+
+SamplerYcbcrRange getYcbcrRangeFilament(VkSamplerYcbcrRange range) {
+    switch (range) {
+        case VK_SAMPLER_YCBCR_RANGE_ITU_FULL:
+            return SamplerYcbcrRange::ITU_FULL;
+        case VK_SAMPLER_YCBCR_RANGE_ITU_NARROW:
+            return SamplerYcbcrRange::ITU_NARROW;
+        default:
+            assert_invariant(false && "Unknown data type, conversion is not supported.");
+            return {};
+    }
+}
+
+ChromaLocation getChromaLocationFilament(VkChromaLocation loc) {
+    switch (loc) {
+        case VK_CHROMA_LOCATION_COSITED_EVEN:
+            return ChromaLocation::COSITED_EVEN;
+        case VK_CHROMA_LOCATION_MIDPOINT:
+            return ChromaLocation::MIDPOINT;
+        default:
+            assert_invariant(false && "Unknown data type, conversion is not supported.");
+            return {};
+    }
+}
+
+TextureSwizzle getSwizzleFilament(VkComponentSwizzle c, uint8_t rgbaIndex) {
+    switch (c) {
+        case VK_COMPONENT_SWIZZLE_ZERO:
+            return TextureSwizzle::SUBSTITUTE_ZERO;
+        case VK_COMPONENT_SWIZZLE_ONE:
+            return TextureSwizzle::SUBSTITUTE_ONE;
+        case VK_COMPONENT_SWIZZLE_IDENTITY:
+            return (TextureSwizzle) (((uint8_t) TextureSwizzle::CHANNEL_0) + rgbaIndex);
+        case VK_COMPONENT_SWIZZLE_R:
+            return TextureSwizzle::CHANNEL_0;
+        case VK_COMPONENT_SWIZZLE_G:
+            return TextureSwizzle::CHANNEL_1;
+        case VK_COMPONENT_SWIZZLE_B:
+            return TextureSwizzle::CHANNEL_2;
+        case VK_COMPONENT_SWIZZLE_A:
+            return TextureSwizzle::CHANNEL_3;
+        default:
+            assert_invariant(false && "Unknown data type, conversion is not supported.");
+            return {};
+    }
+}
+
 } // namespace filament::backend::fvkutils

filament/backend/src/vulkan/utils/Conversion.h

@@ -69,6 +69,12 @@ VkSamplerYcbcrModelConversion getYcbcrModelConversion(SamplerYcbcrModelConversio
 VkSamplerYcbcrRange getYcbcrRange(SamplerYcbcrRange range);
 VkChromaLocation getChromaLocation(ChromaLocation loc);
 
+// Ycbcr related functions
+SamplerYcbcrModelConversion getYcbcrModelConversionFilament(VkSamplerYcbcrModelConversion model);
+SamplerYcbcrRange getYcbcrRangeFilament(VkSamplerYcbcrRange range);
+ChromaLocation getChromaLocationFilament(VkChromaLocation loc);
+TextureSwizzle getSwizzleFilament(VkComponentSwizzle c, uint8_t rgbaIndex);
+
 inline VkImageViewType getViewType(SamplerType target) {
     switch (target) {
         case SamplerType::SAMPLER_CUBEMAP:

filament/backend/src/vulkan/utils/Definitions.h

@@ -348,6 +348,10 @@ using SamplerBitmask = utils::bitset64;
 // general.
 using InputAttachmentBitmask = utils::bitset64;
 
+constexpr uint8_t MAX_DESCRIPTOR_SET_BITMASK_BITS =
+        std::max(std::max(sizeof(UniformBufferBitmask), sizeof(SamplerBitmask)),
+                sizeof(InputAttachmentBitmask)) * 8;
+
 template<typename Bitmask>
 static constexpr uint8_t getVertexStageShift() noexcept {
     // We assume the bottom half of bits are for vertex stages.

filament/backend/src/vulkan/utils/Image.cpp

@@ -193,49 +193,6 @@ uint8_t reduceSampleCount(uint8_t sampleCount, VkSampleCountFlags mask) {
     return mostSignificantBit((sampleCount - 1) & mask);
 }
 
-VkSamplerYcbcrModelConversion getYcbcrModelConversion(
-    SamplerYcbcrModelConversion model) {
-    switch (model) {
-    case SamplerYcbcrModelConversion::RGB_IDENTITY:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY;
-    case SamplerYcbcrModelConversion::YCBCR_IDENTITY:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY;
-    case SamplerYcbcrModelConversion::YCBCR_709:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709;
-    case SamplerYcbcrModelConversion::YCBCR_601:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601;
-    case SamplerYcbcrModelConversion::YCBCR_2020:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020;
-    default:
-        assert_invariant(false &&
-            "Unknown data type, conversion is not supported.");
-    }
-}
-
-VkSamplerYcbcrRange getYcbcrRange(SamplerYcbcrRange range) {
-    switch (range) {
-    case SamplerYcbcrRange::ITU_FULL:
-        return VK_SAMPLER_YCBCR_RANGE_ITU_FULL;
-    case SamplerYcbcrRange::ITU_NARROW:
-        return VK_SAMPLER_YCBCR_RANGE_ITU_NARROW;
-    default:
-        assert_invariant(false &&
-            "Unknown data type, conversion is not supported.");
-    }
-}
-
-VkChromaLocation getChromaLocation(ChromaLocation loc) {
-    switch (loc) {
-    case ChromaLocation::COSITED_EVEN:
-        return VK_CHROMA_LOCATION_COSITED_EVEN;
-    case ChromaLocation::MIDPOINT:
-        return VK_CHROMA_LOCATION_MIDPOINT;
-    default:
-        assert_invariant(false &&
-            "Unknown data type, conversion is not supported.");
-    }
-}
-
 } // namespace filament::backend::fvkutils
 
 bool operator<(const VkImageSubresourceRange& a, const VkImageSubresourceRange& b) {

filament/backend/src/webgpu/WGPUProgram.cpp

@@ -0,0 +1,150 @@
+/*
+ * Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "WebGPUHandles.h"
+
+#include "WebGPUConstants.h"
+
+#include "DriverBase.h"
+#include <backend/DriverEnums.h>
+#include <backend/Program.h>
+
+#include <utils/Panic.h>
+#include <utils/ostream.h>
+
+#include <webgpu/webgpu_cpp.h>
+
+#include <sstream>
+#include <string_view>
+#include <vector>
+
+namespace filament::backend {
+
+namespace {
+
+[[nodiscard]] constexpr std::string_view toString(ShaderStage stage) {
+    switch (stage) {
+        case ShaderStage::VERTEX:
+            return "vertex";
+        case ShaderStage::FRAGMENT:
+            return "fragment";
+        case ShaderStage::COMPUTE:
+            return "compute";
+    }
+}
+
+[[nodiscard]] wgpu::ShaderModule createShaderModule(wgpu::Device& device, const char* programName,
+        std::array<utils::FixedCapacityVector<uint8_t>, Program::SHADER_TYPE_COUNT> const&
+                shaderSource,
+        ShaderStage stage) {
+    utils::FixedCapacityVector<uint8_t> const& sourceBytes =
+            shaderSource[static_cast<size_t>(stage)];
+    if (sourceBytes.empty()) {
+        return nullptr;// nothing to compile, the shader was not provided
+    }
+    wgpu::ShaderModuleWGSLDescriptor wgslDescriptor{};
+    wgslDescriptor.code = wgpu::StringView(reinterpret_cast<const char*>(sourceBytes.data()));
+    std::stringstream labelStream;
+    labelStream << programName << " " << toString(stage) << " shader";
+    auto label = labelStream.str();
+    wgpu::ShaderModuleDescriptor descriptor{
+        .nextInChain = &wgslDescriptor,
+        .label = label.data()
+    };
+    wgpu::ShaderModule module = device.CreateShaderModule(&descriptor);
+    FILAMENT_CHECK_POSTCONDITION(module != nullptr) << "Failed to create " << descriptor.label;
+    module.GetCompilationInfo(wgpu::CallbackMode::AllowSpontaneous,
+            [&descriptor](auto const& status, wgpu::CompilationInfo const* info) {
+                switch (status) {
+                    case wgpu::CompilationInfoRequestStatus::CallbackCancelled:
+                        FWGPU_LOGW << "Shader compilation info callback cancelled for "
+                                   << descriptor.label << "?" << utils::io::endl;
+                        return;
+                    case wgpu::CompilationInfoRequestStatus::Success:
+                        break;
+                }
+                if (info != nullptr) {
+                    std::stringstream errorStream;
+                    int errorCount = 0;
+                    for (size_t msgIndex = 0; msgIndex < info->messageCount; msgIndex++) {
+                        wgpu::CompilationMessage const& message = info->messages[msgIndex];
+                        switch (message.type) {
+                            case wgpu::CompilationMessageType::Info:
+                                FWGPU_LOGI << descriptor.label << ": " << message.message
+                                           << " line#:" << message.lineNum
+                                           << " linePos:" << message.linePos
+                                           << " offset:" << message.offset
+                                           << " length:" << message.length << utils::io::endl;
+                                break;
+                            case wgpu::CompilationMessageType::Warning:
+                                FWGPU_LOGW << "Warning compiling " << descriptor.label << ": "
+                                           << message.message << " line#:" << message.lineNum
+                                           << " linePos:" << message.linePos
+                                           << " offset:" << message.offset
+                                           << " length:" << message.length << utils::io::endl;
+                                break;
+                            case wgpu::CompilationMessageType::Error:
+                                errorCount++;
+                                errorStream << "Error " << errorCount << " : "
+                                            << std::string_view(message.message)
+                                            << " line#:" << message.lineNum
+                                            << " linePos:" << message.linePos
+                                            << " offset:" << message.offset
+                                            << " length:" << message.length << "\n";
+                                break;
+                        }
+                    }
+                    FILAMENT_CHECK_POSTCONDITION(errorCount < 1)
+                            << errorCount << " error(s) compiling " << descriptor.label << ":\n"
+                            << errorStream.str();
+                }
+                FWGPU_LOGD << descriptor.label << " compiled successfully" << utils::io::endl;
+            });
+    return module;
+}
+
+std::vector<wgpu::ConstantEntry> convertConstants(
+        utils::FixedCapacityVector<filament::backend::Program::SpecializationConstant> const&
+                constantsInfo) {
+    std::vector<wgpu::ConstantEntry> constants(constantsInfo.size());
+    for (size_t i = 0; i < constantsInfo.size(); i++) {
+        filament::backend::Program::SpecializationConstant const& specConstant = constantsInfo[i];
+        wgpu::ConstantEntry& constantEntry = constants[i];
+        constantEntry.key = wgpu::StringView(std::to_string(specConstant.id));
+        if (auto* v = std::get_if<int32_t>(&specConstant.value)) {
+            constantEntry.value = static_cast<double>(*v);
+        } else if (auto* f = std::get_if<float>(&specConstant.value)) {
+            constantEntry.value = static_cast<double>(*f);
+        } else if (auto* b = std::get_if<bool>(&specConstant.value)) {
+            constantEntry.value = *b ? 0.0 : 1.0;
+        }
+    }
+    return constants;
+}
+
+}// namespace
+
+WGPUProgram::WGPUProgram(wgpu::Device& device, Program& program)
+    : HwProgram(program.getName()),
+      vertexShaderModule(createShaderModule(device, name.c_str_safe(), program.getShadersSource(),
+              ShaderStage::VERTEX)),
+      fragmentShaderModule(createShaderModule(device, name.c_str_safe(), program.getShadersSource(),
+              ShaderStage::FRAGMENT)),
+      computeShaderModule(createShaderModule(device, name.c_str_safe(), program.getShadersSource(),
+              ShaderStage::COMPUTE)),
+      constants(convertConstants(program.getSpecializationConstants())) {}
+
+}// namespace filament::backend

filament/backend/src/webgpu/WebGPUDriver.cpp

@@ -16,13 +16,13 @@
 
 #include "webgpu/WebGPUDriver.h"
 
+#include "WebGPUSwapChain.h"
 #include "webgpu/WebGPUConstants.h"
 #include <backend/platforms/WebGPUPlatform.h>
 
 #include "CommandStreamDispatcher.h"
 #include "DriverBase.h"
 #include "private/backend/Dispatcher.h"
-
 #include <backend/DriverEnums.h>
 #include <backend/Handle.h>
 
@@ -228,6 +228,14 @@ WebGPUDriver::WebGPUDriver(WebGPUPlatform& platform, const Platform::DriverConfi
               driverConfig.disableHeapHandleTags) {
 #if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
     printInstanceDetails(mPlatform.getInstance());
+#endif
+    mAdapter = mPlatform.requestAdapter(nullptr);
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+    printAdapterDetails(mAdapter);
+#endif
+    mDevice = mPlatform.requestDevice(mAdapter);
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+    printDeviceDetails(mDevice);
 #endif
 }
 
@@ -262,9 +270,6 @@ void WebGPUDriver::tick(int) {
 
 void WebGPUDriver::beginFrame(int64_t monotonic_clock_ns,
         int64_t refreshIntervalNs, uint32_t frameId) {
-    wgpu::CommandEncoderDescriptor commandEncoderDescriptor{};
-    commandEncoderDescriptor.nextInChain = nullptr;
-    mCommandEncoder = mDevice.CreateCommandEncoder(&commandEncoderDescriptor);
 }
 
 void WebGPUDriver::setFrameScheduledCallback(Handle<HwSwapChain> sch,
@@ -281,12 +286,6 @@ void WebGPUDriver::setPresentationTime(int64_t monotonic_clock_ns) {
 }
 
 void WebGPUDriver::endFrame(uint32_t frameId) {
-    // FWGPU_LOGW << __FUNCTION__<< "\n";
-    mQueue.Submit(1, &mCommandBuffer);
-    mCommandEncoder = nullptr;
-    mCommandBuffer = nullptr;
-    mTextureView = nullptr;
-    mSwapChain->Present();
 }
 
 void WebGPUDriver::flush(int) {
@@ -296,15 +295,27 @@ void WebGPUDriver::finish(int) {
 }
 
 void WebGPUDriver::destroyRenderPrimitive(Handle<HwRenderPrimitive> rph) {
+   if (rph) {
+        destructHandle<WGPURenderPrimitive>(rph);
+    }
 }
 
 void WebGPUDriver::destroyVertexBufferInfo(Handle<HwVertexBufferInfo> vbih) {
+   if (vbih) {
+        destructHandle<WGPUVertexBufferInfo>(vbih);
+    }
 }
 
 void WebGPUDriver::destroyVertexBuffer(Handle<HwVertexBuffer> vbh) {
+    if (vbh) {
+        destructHandle<WGPUVertexBuffer>(vbh);
+    }
 }
 
 void WebGPUDriver::destroyIndexBuffer(Handle<HwIndexBuffer> ibh) {
+    if (ibh) {
+        destructHandle<WGPUIndexBuffer>(ibh);
+    }
 }
 
 void WebGPUDriver::destroyBufferObject(Handle<HwBufferObject> boh) {
@@ -314,19 +325,19 @@ void WebGPUDriver::destroyTexture(Handle<HwTexture> th) {
 }
 
 void WebGPUDriver::destroyProgram(Handle<HwProgram> ph) {
+    if (ph) {
+        destructHandle<WGPUProgram>(ph);
+    }
 }
 
 void WebGPUDriver::destroyRenderTarget(Handle<HwRenderTarget> rth) {
 }
 
 void WebGPUDriver::destroySwapChain(Handle<HwSwapChain> sch) {
+    if (sch) {
+        destructHandle<WebGPUSwapChain>(sch);
+    }
     mSwapChain = nullptr;
-    // TODO:  use webgpu handle allocator from
-    //        https://github.com/google/filament/pull/8566
-//    if (sch) {
-//        HwSwapChain* hwSwapChain = handleCast<HwSwapChain*>(sch);
-//        destruct(sch, hwSwapChain);
-//    }
 }
 
 void WebGPUDriver::destroyStream(Handle<HwStream> sh) {
@@ -336,16 +347,16 @@ void WebGPUDriver::destroyTimerQuery(Handle<HwTimerQuery> tqh) {
 }
 
 void WebGPUDriver::destroyDescriptorSetLayout(Handle<HwDescriptorSetLayout> tqh) {
+    if (tqh) {
+        destructHandle<WebGPUDescriptorSetLayout>(tqh);
+    }
 }
 
 void WebGPUDriver::destroyDescriptorSet(Handle<HwDescriptorSet> tqh) {
 }
 
 Handle<HwSwapChain> WebGPUDriver::createSwapChainS() noexcept {
-    // TODO:  use webgpu handle allocator from.
-    //        https://github.com/google/filament/pull/8566
-    // return allocAndConstructHandle<HwSwapChain>();
-    return Handle<HwSwapChain>((Handle<HwSwapChain>::HandleId) mNextFakeHandle++);
+    return allocHandle<WebGPUSwapChain>();
 }
 
 Handle<HwSwapChain> WebGPUDriver::createSwapChainHeadlessS() noexcept {
@@ -361,7 +372,7 @@ Handle<HwTexture> WebGPUDriver::importTextureS() noexcept {
 }
 
 Handle<HwProgram> WebGPUDriver::createProgramS() noexcept {
-    return Handle<HwProgram>((Handle<HwProgram>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUProgram>();
 }
 
 Handle<HwFence> WebGPUDriver::createFenceS() noexcept {
@@ -377,7 +388,7 @@ Handle<HwIndexBuffer> WebGPUDriver::createIndexBufferS() noexcept {
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureViewS() noexcept {
-    return allocHandle<WGPUTexture>();
+    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
 }
 
 Handle<HwBufferObject> WebGPUDriver::createBufferObjectS() noexcept {
@@ -397,7 +408,7 @@ Handle<HwDescriptorSet> WebGPUDriver::createDescriptorSetS() noexcept {
 }
 
 Handle<HwRenderPrimitive> WebGPUDriver::createRenderPrimitiveS() noexcept {
-    return Handle<HwRenderPrimitive>((Handle<HwRenderPrimitive>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPURenderPrimitive>();
 }
 
 Handle<HwVertexBufferInfo> WebGPUDriver::createVertexBufferInfoS() noexcept {
@@ -405,7 +416,7 @@ Handle<HwVertexBufferInfo> WebGPUDriver::createVertexBufferInfoS() noexcept {
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureViewSwizzleS() noexcept {
-    return allocHandle<WGPUTexture>();
+    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
 }
 
 Handle<HwRenderTarget> WebGPUDriver::createDefaultRenderTargetS() noexcept {
@@ -413,8 +424,7 @@ Handle<HwRenderTarget> WebGPUDriver::createDefaultRenderTargetS() noexcept {
 }
 
 Handle<HwDescriptorSetLayout> WebGPUDriver::createDescriptorSetLayoutS() noexcept {
-    return Handle<HwDescriptorSetLayout>(
-            (Handle<HwDescriptorSetLayout>::HandleId) mNextFakeHandle++);
+    return allocHandle<WebGPUDescriptorSetLayout>();
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureExternalImageS() noexcept {
@@ -430,43 +440,15 @@ Handle<HwTexture> WebGPUDriver::createTextureExternalImagePlaneS() noexcept {
 }
 
 void WebGPUDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow, uint64_t flags) {
-    // TODO:  use webgpu handle allocator from.
-    //        https://github.com/google/filament/pull/8566
-    // HwSwapChain* hwSwapChain = handleCast<HwSwapChain*>(sch);
+    mNativeWindow = nativeWindow;
     assert_invariant(!mSwapChain);
     wgpu::Surface surface = mPlatform.createSurface(nativeWindow, flags);
-    mAdapter = mPlatform.requestAdapter(surface);
-#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
-    printAdapterDetails(mAdapter);
-#endif
-    mDevice = mPlatform.requestDevice(mAdapter);
-#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
-    printDeviceDetails(mDevice);
-#endif
+
     mQueue = mDevice.GetQueue();
-    mSwapChain = std::make_unique<WebGPUSwapChain>(std::move(surface), mAdapter, mDevice, flags);
-    // TODO configure the surface (maybe before or after creating the swapchain?
-    //                             how do we get the surface extent?)
-    // TODO actually create the swapchain
-    // auto onQueueWorkDone = [](wgpu::QueueWorkDoneStatus status, void* /* pUserData */) {
-        // FWGPU_LOGW << "Queued work finished with status: " << status << std::endl;
-    // };
-    // mQueue.OnSubmittedWorkDone(wgpu::CallbackMode::WaitAnyOnly, [](wgpu::QueueWorkDoneStatus status, void* pUserdata) {
-    //     FWGPU_LOGW << "Queued work finished with status:\n";
-    // }, nullptr /* pUserData */);
-    void* userDataPtr = nullptr;
-    mQueue.OnSubmittedWorkDone(
-        wgpu::CallbackMode::AllowProcessEvents,
-        [](wgpu::QueueWorkDoneStatus status, void* pUserData) {
-            FWGPU_LOGW << "Queued work finished with status: " << static_cast<int>(status) << "\n";
-            if (pUserData == nullptr) {
-                // Expected case
-            } else {
-                FWGPU_LOGW << "Unexpected non-null pUserData received.\n";
-            }
-        },
-        userDataPtr /* pUserData */
-    );
+    wgpu::Extent2D surfaceSize = mPlatform.getSurfaceExtent(mNativeWindow);
+    mSwapChain = constructHandle<WebGPUSwapChain>(sch, std::move(surface), surfaceSize, mAdapter,
+            mDevice, flags);
+    assert_invariant(mSwapChain);
     FWGPU_LOGW << "WebGPU support is still essentially a no-op at this point in development (only "
                   "background components have been instantiated/selected, such as surface/screen, "
                   "graphics device/GPU, etc.), thus nothing is being drawn to the screen."
@@ -479,9 +461,6 @@ void WebGPUDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow,
                   "rebuilding Filament with that flag, e.g. ./build.sh -x "
                << FWGPU_PRINT_SYSTEM << " ..." << utils::io::endl;
 #endif
-    // TODO:  use webgpu handle allocator from.
-    //        https://github.com/google/filament/pull/8566
-    // hwSwapChain->swapChain = mSwapChain.get();
 }
 
 void WebGPUDriver::createSwapChainHeadlessR(Handle<HwSwapChain> sch, uint32_t width,
@@ -504,13 +483,7 @@ void WebGPUDriver::createTextureR(Handle<HwTexture> th, SamplerType target, uint
         TextureUsage usage) {}
 
 void WebGPUDriver::createTextureViewR(Handle<HwTexture> th, Handle<HwTexture> srch,
-        uint8_t baseLevel, uint8_t levelCount) {
-    // FWGPU_LOGW << __FUNCTION__<< "\n";
-    WGPUTexture const* src = handleCast<WGPUTexture>(srch);
-    (void) src;
-    // textures.insert(
-    //         constructHandle<WGPUTexture>(th, src, baseLevel, levelCount));
-}
+        uint8_t baseLevel, uint8_t levelCount) {}
 
 void WebGPUDriver::createTextureViewSwizzleR(Handle<HwTexture> th, Handle<HwTexture> srch,
         backend::TextureSwizzle r, backend::TextureSwizzle g, backend::TextureSwizzle b,
@@ -535,7 +508,9 @@ void WebGPUDriver::importTextureR(Handle<HwTexture> th, intptr_t id, SamplerType
 void WebGPUDriver::createRenderPrimitiveR(Handle<HwRenderPrimitive> rph, Handle<HwVertexBuffer> vbh,
         Handle<HwIndexBuffer> ibh, PrimitiveType pt) {}
 
-void WebGPUDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {}
+void WebGPUDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {
+    constructHandle<WGPUProgram>(ph, mDevice, program);
+}
 
 void WebGPUDriver::createDefaultRenderTargetR(Handle<HwRenderTarget> rth, int) {
     assert_invariant(!mDefaultRenderTarget);
@@ -552,7 +527,9 @@ void WebGPUDriver::createFenceR(Handle<HwFence> fh, int) {}
 void WebGPUDriver::createTimerQueryR(Handle<HwTimerQuery> tqh, int) {}
 
 void WebGPUDriver::createDescriptorSetLayoutR(Handle<HwDescriptorSetLayout> dslh,
-        backend::DescriptorSetLayout&& info) {}
+        backend::DescriptorSetLayout&& info) {
+    constructHandle<WebGPUDescriptorSetLayout>(dslh, std::move(info), mDevice);
+}
 
 void WebGPUDriver::createDescriptorSetR(Handle<HwDescriptorSet> dsh,
         Handle<HwDescriptorSetLayout> dslh) {}
@@ -704,9 +681,8 @@ void WebGPUDriver::setVertexBufferObject(Handle<HwVertexBuffer> vbh, uint32_t in
         Handle<HwBufferObject> boh) {
     auto* vertexBuffer = handleCast<WGPUVertexBuffer>(vbh);
     auto* bufferObject = handleCast<WGPUBufferObject>(boh);
-    assert_invariant(index < vertexBuffer->mBuffers.size());
+    assert_invariant(index < vertexBuffer->buffers.size());
     vertexBuffer->setBuffer(bufferObject, index);
-
 }
 
 void WebGPUDriver::update3DImage(Handle<HwTexture> th,
@@ -739,48 +715,71 @@ void WebGPUDriver::compilePrograms(CompilerPriorityQueue priority,
 }
 
 void WebGPUDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassParams& params) {
-    // FWGPU_LOGW << __FUNCTION__<< "\n";
-    mTextureView = mSwapChain->GetNextSurfaceTextureView(params.viewport.width, params.viewport.height);
-    wgpu::RenderPassColorAttachment renderPassColorAttachment = {};
-    renderPassColorAttachment.view = mTextureView;
-    renderPassColorAttachment.resolveTarget = nullptr;
-    renderPassColorAttachment.loadOp = wgpu::LoadOp::Clear;
-    renderPassColorAttachment.storeOp = wgpu::StoreOp::Store;
-    renderPassColorAttachment.clearValue = wgpu::Color{1, 0 , 0 , 1};
-    renderPassColorAttachment.depthSlice = WGPU_DEPTH_SLICE_UNDEFINED;
+    wgpu::CommandEncoderDescriptor commandEncoderDescriptor = {
+        .label = "command_encoder"
+    };
+    mCommandEncoder = mDevice.CreateCommandEncoder(&commandEncoderDescriptor);
+    assert_invariant(mCommandEncoder);
+    // TODO: Remove this code once WebGPU pipeline is implemented
+    static float red = 1.0f;
+    if (red - 0.01 > 0) {
+        red -= 0.01;
+    } else {
+        red = 1.0f;
+    }
+    assert_invariant(mTextureView);
+    wgpu::RenderPassColorAttachment renderPassColorAttachment = {
+        .view = mTextureView,
+        // TODO: remove this code once WebGPU Pipeline is implemented with render targets, pipeline and buffers.
+        .depthSlice = wgpu::kDepthSliceUndefined,
+        .loadOp = wgpu::LoadOp::Clear,
+        .storeOp = wgpu::StoreOp::Store,
+        .clearValue = wgpu::Color{red, 0 , 0 , 1},
+    };
 
-    wgpu::RenderPassDescriptor renderPassDescriptor = {};
-    renderPassDescriptor.nextInChain = nullptr;
-    renderPassDescriptor.colorAttachmentCount = 1;
-    renderPassDescriptor.colorAttachments = &renderPassColorAttachment;
-    renderPassDescriptor.depthStencilAttachment = nullptr;
-    renderPassDescriptor.timestampWrites = nullptr;
+    wgpu::RenderPassDescriptor renderPassDescriptor = {
+        .colorAttachmentCount = 1,
+        .colorAttachments = &renderPassColorAttachment,
+        .depthStencilAttachment = nullptr,
+        .timestampWrites = nullptr,
+    };
 
     mRenderPassEncoder = mCommandEncoder.BeginRenderPass(&renderPassDescriptor);
-    mRenderPassEncoder.SetViewport((float)params.viewport.left, (float)params.viewport.bottom,
-            (float) params.viewport.width, (float) params.viewport.height, params.depthRange.near, params.depthRange.far);
-            // (float) 1024/*params.viewport.width*/, (float) 640/*params.viewport.height*/, params.depthRange.near, params.depthRange.far);
+    mRenderPassEncoder.SetViewport(params.viewport.left, params.viewport.bottom,
+            params.viewport.width, params.viewport.height, params.depthRange.near, params.depthRange.far);
 }
 
 void WebGPUDriver::endRenderPass(int) {
-    // FWGPU_LOGW << __FUNCTION__<< "\n";
     mRenderPassEncoder.End();
     mRenderPassEncoder = nullptr;
+    wgpu::CommandBufferDescriptor commandBufferDescriptor {
+        .label = "command_buffer",
+    };
+    mCommandBuffer = mCommandEncoder.Finish(&commandBufferDescriptor);
+    assert_invariant(mCommandBuffer);
 }
 
 void WebGPUDriver::nextSubpass(int) {
 }
 
 void WebGPUDriver::makeCurrent(Handle<HwSwapChain> drawSch, Handle<HwSwapChain> readSch) {
+    ASSERT_PRECONDITION_NON_FATAL(drawSch == readSch,
+            "WebGPU driver does not support distinct draw/read swap chains.");
+    auto* swapChain = handleCast<WebGPUSwapChain>(drawSch);
+    mSwapChain = swapChain;
+    assert_invariant(mSwapChain);
+    wgpu::Extent2D surfaceSize = mPlatform.getSurfaceExtent(mNativeWindow);
+    mTextureView = mSwapChain->getCurrentSurfaceTextureView(surfaceSize);
+    assert_invariant(mTextureView);
 }
 
 void WebGPUDriver::commit(Handle<HwSwapChain> sch) {
-    wgpu::CommandBufferDescriptor commandBufferDescriptor = {};
-    commandBufferDescriptor.nextInChain = nullptr;
-    mCommandBuffer = mCommandEncoder.Finish(&commandBufferDescriptor);
-
-    // mQueue.Submit(1, &mCommandBuffer);
-    // mCommandBuffer = nullptr;
+    mCommandEncoder = nullptr;
+    mQueue.Submit(1, &mCommandBuffer);
+    mCommandBuffer = nullptr;
+    mTextureView = nullptr;
+    assert_invariant(mSwapChain);
+    mSwapChain->present();
 }
 
 void WebGPUDriver::setPushConstant(backend::ShaderStage stage, uint8_t index,
@@ -808,7 +807,7 @@ void WebGPUDriver::readPixels(Handle<HwRenderTarget> src,
     scheduleDestroy(std::move(p));
 }
 
-void WebGPUDriver::readBufferSubData(backend::BufferObjectHandle boh,
+void WebGPUDriver::readBufferSubData(Handle<HwBufferObject> boh,
         uint32_t offset, uint32_t size, backend::BufferDescriptor&& p) {
     scheduleDestroy(std::move(p));
 }
@@ -837,8 +836,6 @@ void WebGPUDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
 }
 
 void WebGPUDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
-    // mRenderPassEncoder.DrawIndexed(indexCount, instanceCount, indexOffset, 0, 0);
-
 }
 
 void WebGPUDriver::draw(PipelineState pipelineState, Handle<HwRenderPrimitive> rph,
@@ -862,22 +859,22 @@ void WebGPUDriver::resetState(int) {
 }
 
 void WebGPUDriver::updateDescriptorSetBuffer(
-        backend::DescriptorSetHandle dsh,
+        Handle<HwDescriptorSet> dsh,
         backend::descriptor_binding_t binding,
-        backend::BufferObjectHandle boh,
+        Handle<HwBufferObject> boh,
         uint32_t offset,
         uint32_t size) {
 }
 
 void WebGPUDriver::updateDescriptorSetTexture(
-        backend::DescriptorSetHandle dsh,
+        Handle<HwDescriptorSet> dsh,
         backend::descriptor_binding_t binding,
-        backend::TextureHandle th,
+        Handle<HwTexture> th,
         SamplerParams params) {
 }
 
 void WebGPUDriver::bindDescriptorSet(
-        backend::DescriptorSetHandle dsh,
+        Handle<HwDescriptorSet> dsh,
         backend::descriptor_set_t set,
         backend::DescriptorSetOffsetArray&& offsets) {
 }

filament/backend/src/webgpu/WebGPUDriver.h

@@ -17,7 +17,7 @@
 #ifndef TNT_FILAMENT_BACKEND_WEBGPUDRIVER_H
 #define TNT_FILAMENT_BACKEND_WEBGPUDRIVER_H
 
-#include "webgpu/WebGPUSwapChain.h"
+#include "WebGPUHandles.h"
 #include <backend/platforms/WebGPUPlatform.h>
 
 #include "DriverBase.h"
@@ -30,8 +30,6 @@
 
 #include <webgpu/webgpu_cpp.h>
 
-#include "WebGPUHandles.h"
-
 #include <cstdint>
 #include <memory>
 
@@ -41,6 +39,8 @@
 
 namespace filament::backend {
 
+class WebGPUSwapChain;
+
 /**
  * WebGPU backend (driver) implementation
  */
@@ -62,8 +62,8 @@ private:
     wgpu::Adapter mAdapter = nullptr;
     wgpu::Device mDevice = nullptr;
     wgpu::Queue mQueue = nullptr;
-    // TODO consider moving to handle allocator when ready
-    std::unique_ptr<WebGPUSwapChain> mSwapChain = nullptr;
+    void* mNativeWindow = nullptr;
+    WebGPUSwapChain* mSwapChain = nullptr;
     uint64_t mNextFakeHandle = 1;
     wgpu::CommandEncoder mCommandEncoder = nullptr;
     wgpu::TextureView mTextureView = nullptr;
@@ -103,11 +103,17 @@ private:
     D* constructHandle(Handle<B>& handle, ARGS&& ... args) noexcept {
         return mHandleAllocator.construct<D>(handle, std::forward<ARGS>(args)...);
     }
+
     template<typename D, typename B>
     D* handleCast(Handle<B> handle) noexcept {
         return mHandleAllocator.handle_cast<D*>(handle);
     }
 
+    template<typename D, typename B>
+    void destructHandle(Handle<B>& handle) noexcept {
+        auto* p = mHandleAllocator.handle_cast<D*>(handle);
+        return mHandleAllocator.deallocate(handle, p);
+    }
 };
 
 }// namespace filament::backend

filament/backend/src/webgpu/WebGPUHandles.cpp

@@ -1,19 +1,152 @@
-//
-// Created by Idris Idris Shah on 3/21/25.
-//
+/*
+* Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
 
 #include "WebGPUHandles.h"
 
+#include <utility>
+
+namespace {
+
+wgpu::Buffer createIndexBuffer(wgpu::Device const& device, uint8_t elementSize, uint32_t indexCount) {
+    wgpu::BufferDescriptor descriptor{ .label = "index_buffer",
+        .usage = wgpu::BufferUsage::CopyDst | wgpu::BufferUsage::Index,
+        .size = elementSize * indexCount,
+        .mappedAtCreation = false };
+    return device.CreateBuffer(&descriptor);
+}
+} // namespace
+
 namespace filament::backend {
 
-WGPUVertexBuffer::WGPUVertexBuffer(uint32_t vextexCount, uint32_t bufferCount,
-        Handle<WGPUVertexBufferInfo> vbih)
-    : HwVertexBuffer(vextexCount),
-      vbih(vbih),
-      mBuffers(MAX_VERTEX_BUFFER_COUNT) {}
+
+WGPUIndexBuffer::WGPUIndexBuffer(wgpu::Device const& device, uint8_t elementSize,
+        uint32_t indexCount)
+    : buffer(createIndexBuffer(device, elementSize, indexCount)) {}
+
+
+WGPUVertexBuffer::WGPUVertexBuffer(wgpu::Device const &device, uint32_t vextexCount, uint32_t bufferCount,
+                                   Handle<WGPUVertexBufferInfo> vbih)
+        : HwVertexBuffer(vextexCount),
+          vbih(vbih),
+          buffers(bufferCount) {
+    wgpu::BufferDescriptor descriptor {
+            .usage = wgpu::BufferUsage::CopyDst | wgpu::BufferUsage::Vertex,
+            .size = vextexCount * bufferCount,
+            .mappedAtCreation = false };
+
+    for (uint32_t i = 0; i < bufferCount; ++i) {
+        descriptor.label = ("vertex_buffer_" + std::to_string(i)).c_str();
+        buffers[i] = device.CreateBuffer(&descriptor);
+    }
+}
+
+// TODO: Empty function is a place holder for verxtex buffer updates and should be
+// updated for that purpose.
 void WGPUVertexBuffer::setBuffer(WGPUBufferObject* bufferObject, uint32_t index) {}
 
 WGPUBufferObject::WGPUBufferObject(BufferObjectBinding bindingType, uint32_t byteCount)
     : HwBufferObject(byteCount),
-      mBindingType(bindingType) {}
+      bufferObjectBinding(bindingType) {}
+
+wgpu::ShaderStage WebGPUDescriptorSetLayout::filamentStageToWGPUStage(ShaderStageFlags fFlags) {
+    wgpu::ShaderStage retStages = wgpu::ShaderStage::None;
+    if (any(ShaderStageFlags::VERTEX & fFlags)) {
+        retStages |= wgpu::ShaderStage::Vertex;
+    }
+    if (any(ShaderStageFlags::FRAGMENT & fFlags)) {
+        retStages |= wgpu::ShaderStage::Fragment;
+    }
+    if (any(ShaderStageFlags::COMPUTE & fFlags)) {
+        retStages |= wgpu::ShaderStage::Compute;
+    }
+    return retStages;
+}
+
+WebGPUDescriptorSetLayout::WebGPUDescriptorSetLayout(DescriptorSetLayout const& layout,
+        wgpu::Device const& device) {
+    assert_invariant(device);
+
+    // TODO: layoutDescriptor has a "Label". Ideally we can get info on what this layout is for
+    // debugging. For now, hack an incrementing value.
+    static int layoutNum = 0;
+
+    uint samplerCount =
+            std::count_if(layout.bindings.begin(), layout.bindings.end(), [](auto& fEntry) {
+                return fEntry.type == DescriptorType::SAMPLER ||
+                       fEntry.type == DescriptorType::SAMPLER_EXTERNAL;
+            });
+
+
+    std::vector<wgpu::BindGroupLayoutEntry> wEntries;
+    wEntries.reserve(layout.bindings.size() + samplerCount);
+
+    for (auto fEntry: layout.bindings) {
+        auto& wEntry = wEntries.emplace_back();
+        wEntry.visibility = filamentStageToWGPUStage(fEntry.stageFlags);
+        wEntry.binding = fEntry.binding * 2;
+
+        switch (fEntry.type) {
+            // TODO Metal treats these the same. Is this fine?
+            case DescriptorType::SAMPLER_EXTERNAL:
+            case DescriptorType::SAMPLER: {
+                // Sampler binding is 2n+1 due to split.
+                auto& samplerEntry = wEntries.emplace_back();
+                samplerEntry.binding = fEntry.binding * 2 + 1;
+                samplerEntry.visibility = wEntry.visibility;
+                // We are simply hoping that undefined and defaults suffices here.
+                samplerEntry.sampler.type = wgpu::SamplerBindingType::Undefined;
+                wEntry.texture.sampleType = wgpu::TextureSampleType::Undefined;
+                break;
+            }
+            case DescriptorType::UNIFORM_BUFFER: {
+                wEntry.buffer.hasDynamicOffset =
+                        any(fEntry.flags & DescriptorFlags::DYNAMIC_OFFSET);
+                wEntry.buffer.type = wgpu::BufferBindingType::Uniform;
+                // TODO: Ideally we fill minBindingSize
+                break;
+            }
+
+            case DescriptorType::INPUT_ATTACHMENT: {
+                // TODO: support INPUT_ATTACHMENT. Metal does not currently.
+                PANIC_POSTCONDITION("Input Attachment is not supported");
+                break;
+            }
+
+            case DescriptorType::SHADER_STORAGE_BUFFER: {
+                // TODO: Vulkan does not support this, can we?
+                PANIC_POSTCONDITION("Shader storage is not supported");
+                break;
+            }
+        }
+
+        // Currently flags are only used to specify dynamic offset.
+
+        // UNUSED
+        // fEntry.count
+    }
+
+    wgpu::BindGroupLayoutDescriptor layoutDescriptor{
+        // TODO: layoutDescriptor has a "Label". Ideally we can get info on what this layout is for
+        // debugging. For now, hack an incrementing value.
+        .label{ "layout_" + std::to_string(++layoutNum) },
+        .entryCount = wEntries.size(),
+        .entries = wEntries.data()
+    };
+    // TODO Do we need to defer this until we have more info on textures and samplers??
+    mLayout = device.CreateBindGroupLayout(&layoutDescriptor);
+}
+WebGPUDescriptorSetLayout::~WebGPUDescriptorSetLayout() {}
 }// namespace filament::backend

filament/backend/src/webgpu/WebGPUHandles.h

@@ -1,6 +1,19 @@
-//
-// Created by Idris Idris Shah on 3/21/25.
-//
+/*
+* Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
 
 #ifndef TNT_FILAMENT_BACKEND_WEBGPUHANDLES_H
 #define TNT_FILAMENT_BACKEND_WEBGPUHANDLES_H
@@ -9,13 +22,29 @@
 
 #include <backend/DriverEnums.h>
 #include <backend/Handle.h>
+
 #include <utils/FixedCapacityVector.h>
 
 #include <webgpu/webgpu_cpp.h>
 
+#include <cstdint>
+#include <vector>
+
 namespace filament::backend {
 
+class WGPUProgram final : public HwProgram {
+public:
+    WGPUProgram(wgpu::Device&, Program&);
+
+    wgpu::ShaderModule vertexShaderModule = nullptr;
+    wgpu::ShaderModule fragmentShaderModule = nullptr;
+    wgpu::ShaderModule computeShaderModule = nullptr;
+    std::vector<wgpu::ConstantEntry> constants;
+};
+
 struct WGPUBufferObject;
+// TODO: Currently WGPUVertexBufferInfo is not used by WebGPU for useful task.
+// Update the struct when used by WebGPU driver.
 struct WGPUVertexBufferInfo : public HwVertexBufferInfo {
     WGPUVertexBufferInfo(uint8_t bufferCount, uint8_t attributeCount,
             AttributeArray const& attributes)
@@ -25,28 +54,46 @@ struct WGPUVertexBufferInfo : public HwVertexBufferInfo {
 };
 
 struct WGPUVertexBuffer : public HwVertexBuffer {
-    WGPUVertexBuffer(uint32_t vextexCount, uint32_t bufferCount, Handle<WGPUVertexBufferInfo> vbih);
-    void setBuffer(WGPUBufferObject* bufferObject, uint32_t index);
+    WGPUVertexBuffer(wgpu::Device const &device, uint32_t vextexCount, uint32_t bufferCount,
+                     Handle<WGPUVertexBufferInfo> vbih);
+
+    void setBuffer(WGPUBufferObject *bufferObject, uint32_t index);
 
     Handle<WGPUVertexBufferInfo> vbih;
-    utils::FixedCapacityVector<wgpu::Buffer> mBuffers;
+    utils::FixedCapacityVector<wgpu::Buffer> buffers;
 };
 
 struct WGPUIndexBuffer : public HwIndexBuffer {
-    WGPUIndexBuffer(BufferUsage usage, uint8_t elementSize, uint32_t indexCount);
+    WGPUIndexBuffer(wgpu::Device const &device, uint8_t elementSize,
+                    uint32_t indexCount);
 
     wgpu::Buffer buffer;
 };
 
+// TODO: Currently WGPUVertexBufferInfo is not used by WebGPU for useful task.
+// Update the struct when used by WebGPU driver.
 struct WGPUBufferObject : HwBufferObject {
     WGPUBufferObject(BufferObjectBinding bindingType, uint32_t byteCount);
 
-    wgpu::Buffer mBuffer;
-    const BufferObjectBinding mBindingType;
+    wgpu::Buffer buffer;
+    const BufferObjectBinding bufferObjectBinding;
+};
+class WebGPUDescriptorSetLayout : public HwDescriptorSetLayout {
+public:
+    WebGPUDescriptorSetLayout(DescriptorSetLayout const& layout, wgpu::Device const& device);
+    ~WebGPUDescriptorSetLayout();
+
+private:
+    // TODO: If this is useful elsewhere, remove it from this class
+    // Convert Filament Shader Stage Flags bitmask to webgpu equivilant
+    static wgpu::ShaderStage filamentStageToWGPUStage(ShaderStageFlags fFlags);
+
+    wgpu::BindGroupLayout mLayout;
 };
 
-class WGPUTexture : public HwTexture {
-public:
+// TODO: Currently WGPUTexture is not used by WebGPU for useful task.
+// Update the struct when used by WebGPU driver.
+struct WGPUTexture : public HwTexture {
     WGPUTexture(SamplerType target, uint8_t levels, TextureFormat format, uint8_t samples,
             uint32_t width, uint32_t height, uint32_t depth, TextureUsage usage) noexcept;
 
@@ -56,11 +103,22 @@ public:
     wgpu::Texture texture = nullptr;
 };
 
-class WGPURenderTarget : public HwRenderTarget {
-public:
+struct WGPURenderPrimitive : public HwRenderPrimitive {
+    WGPURenderPrimitive();
+
+    void setBuffers(WGPUVertexBufferInfo const* const vbi,
+            WGPUVertexBuffer* vertexBuffer, WGPUIndexBuffer* indexBuffer);
+
+    WGPUVertexBuffer* vertexBuffer = nullptr;
+    WGPUIndexBuffer* indexBuffer = nullptr;
+};
+
+// TODO: Currently WGPURenderTarget is not used by WebGPU for useful task.
+// Update the struct when used by WebGPU driver.
+struct WGPURenderTarget : public HwRenderTarget {
     class Attachment {
     public:
-        friend class WGPURenderTarget;
+        friend struct WGPURenderTarget;
 
         Attachment() = default;
         Attachment(WGPUTexture* gpuTexture, uint8_t level = 0, uint16_t layer = 0)

filament/backend/src/webgpu/WebGPUSwapChain.cpp

@@ -190,10 +190,13 @@ wgpu::CompositeAlphaMode selectAlphaMode(size_t availableAlphaModesCount,
     }
 }
 
-void initConfig(wgpu::SurfaceConfiguration& config, wgpu::Device& device,
-        wgpu::SurfaceCapabilities const& capabilities, bool useSRGBColorSpace) {
+void initConfig(wgpu::SurfaceConfiguration& config, wgpu::Device const& device,
+        wgpu::SurfaceCapabilities const& capabilities, wgpu::Extent2D const& surfaceSize,
+        bool useSRGBColorSpace) {
     config.device = device;
     config.usage = wgpu::TextureUsage::RenderAttachment;
+    config.width = surfaceSize.width;
+    config.height = surfaceSize.height;
     config.format =
             selectColorFormat(capabilities.formatCount, capabilities.formats, useSRGBColorSpace);
     config.presentMode =
@@ -205,8 +208,8 @@ void initConfig(wgpu::SurfaceConfiguration& config, wgpu::Device& device,
 
 namespace filament::backend {
 
-WebGPUSwapChain::WebGPUSwapChain(wgpu::Surface&& surface, wgpu::Adapter& adapter,
-        wgpu::Device& device, uint64_t flags)
+WebGPUSwapChain::WebGPUSwapChain(wgpu::Surface&& surface, wgpu::Extent2D const& surfaceSize,
+        wgpu::Adapter& adapter, wgpu::Device& device, uint64_t flags)
     : mSurface(surface) {
     wgpu::SurfaceCapabilities capabilities = {};
     if (!mSurface.GetCapabilities(adapter, &capabilities)) {
@@ -217,59 +220,55 @@ WebGPUSwapChain::WebGPUSwapChain(wgpu::Surface&& surface, wgpu::Adapter& adapter
 #endif
     }
     const bool useSRGBColorSpace = (flags & SWAP_CHAIN_CONFIG_SRGB_COLORSPACE) != 0;
-    initConfig(mConfig, device, capabilities, useSRGBColorSpace);
+    initConfig(mConfig, device, capabilities, surfaceSize, useSRGBColorSpace);
+    mSurface.Configure(&mConfig);
 }
 
 WebGPUSwapChain::~WebGPUSwapChain() {
-    if (mConfigured) {
-        mSurface.Unconfigure();
-        mConfigured = false;
-    }
+    mSurface.Unconfigure();
 }
 
-void WebGPUSwapChain::GetCurrentTexture(uint32_t width, uint32_t height, wgpu::SurfaceTexture* texture) {
-    if (width < 1 || height < 1) {
-        PANIC_LOG("WebGPUSwapChain::GetCurrentTexture: Invalid width and/or height requested.");
-        return;
-    }
-    if (mConfig.width != width || mConfig.height != height || !mConfigured) {
-        mConfig.width = width;
-        mConfig.height = height;
+void WebGPUSwapChain::setExtent(wgpu::Extent2D const& currentSurfaceSize) {
+    FILAMENT_CHECK_POSTCONDITION(currentSurfaceSize.width > 0 || currentSurfaceSize.height > 0)
+            << "WebGPUSwapChain::setExtent: Invalid width " << currentSurfaceSize.width
+            << " and/or height " << currentSurfaceSize.height << " requested.";
+    if (mConfig.width != currentSurfaceSize.width || mConfig.height != currentSurfaceSize.height) {
+        mConfig.width = currentSurfaceSize.width;
+        mConfig.height = currentSurfaceSize.height;
 #if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
         printSurfaceConfiguration(mConfig);
 #endif
+        FWGPU_LOGD << "Resizing to width " << mConfig.width << " height " << mConfig.height
+                   << utils::io::endl;
+        // TODO we may need to ensure no surface texture is flight when we do this. some
+        //      synchronization may be necessary
         mSurface.Configure(&mConfig);
-        mConfigured = true;
-        return;
     }
-
-    mSurface.GetCurrentTexture(texture);
 }
 
-wgpu::TextureView WebGPUSwapChain::GetNextSurfaceTextureView(uint32_t width, uint32_t height) {
+wgpu::TextureView WebGPUSwapChain::getCurrentSurfaceTextureView(
+        wgpu::Extent2D const& currentSurfaceSize) {
+    setExtent(currentSurfaceSize);
     wgpu::SurfaceTexture surfaceTexture;
-    GetCurrentTexture(width, height, &surfaceTexture);
-    if (surfaceTexture.status != wgpu::SurfaceGetCurrentTextureStatus::Success) {
+    mSurface.GetCurrentTexture(&surfaceTexture);
+    if (surfaceTexture.status != wgpu::SurfaceGetCurrentTextureStatus::SuccessOptimal) {
         return nullptr;
     }
-
     // Create a view for this surface texture
-    wgpu::TextureViewDescriptor textureViewDescriptor;
-    textureViewDescriptor.nextInChain = nullptr;
-    textureViewDescriptor.label = "Surface texture view";
-    textureViewDescriptor.format = surfaceTexture.texture.GetFormat();
-    textureViewDescriptor.dimension = wgpu::TextureViewDimension::e2D;
-    textureViewDescriptor.baseMipLevel = 0;
-    textureViewDescriptor.mipLevelCount = 1;
-    textureViewDescriptor.baseArrayLayer = 0;
-    textureViewDescriptor.arrayLayerCount = 1;
-    textureViewDescriptor.aspect = wgpu::TextureAspect::All;
-    wgpu::TextureView textureView = surfaceTexture.texture.CreateView(&textureViewDescriptor);
-
-    return textureView;
+    // TODO: review these initiliazations as webgpu pipeline gets mature
+    wgpu::TextureViewDescriptor textureViewDescriptor = {
+        .label = "texture_view",
+        .format = surfaceTexture.texture.GetFormat(),
+        .dimension = wgpu::TextureViewDimension::e2D,
+        .baseMipLevel = 0,
+        .mipLevelCount = 1,
+        .baseArrayLayer = 0,
+        .arrayLayerCount = 1
+    };
+    return surfaceTexture.texture.CreateView(&textureViewDescriptor);
 }
 
-void WebGPUSwapChain::Present() {
+void WebGPUSwapChain::present() {
     assert_invariant(mSurface);
     mSurface.Present();
 }

filament/backend/src/webgpu/WebGPUSwapChain.h

@@ -19,26 +19,28 @@
 
 #include <webgpu/webgpu_cpp.h>
 
+#include "DriverBase.h"
 #include <backend/Platform.h>
 
 #include <cstdint>
 
 namespace filament::backend {
 
-class WebGPUSwapChain : public Platform::SwapChain {
+class WebGPUSwapChain final : public Platform::SwapChain, HwSwapChain {
 public:
-    WebGPUSwapChain(wgpu::Surface&& surface, wgpu::Adapter& adapter, wgpu::Device& device,
-            uint64_t flags);
+    WebGPUSwapChain(wgpu::Surface&& surface, wgpu::Extent2D const& surfaceSize,
+            wgpu::Adapter& adapter, wgpu::Device& device, uint64_t flags);
     ~WebGPUSwapChain();
 
-    void GetCurrentTexture(uint32_t width, uint32_t height, wgpu::SurfaceTexture*);
-    wgpu::TextureView GetNextSurfaceTextureView(uint32_t width, uint32_t height);
-    void Present();
+    wgpu::TextureView getCurrentSurfaceTextureView(wgpu::Extent2D const&);
+
+    void present();
 
 private:
+    void setExtent(wgpu::Extent2D const&);
+
     wgpu::Surface mSurface = {};
     wgpu::SurfaceConfiguration mConfig = {};
-    bool mConfigured = false;
 };
 
 } // namespace filament::backend

filament/backend/src/webgpu/platform/WebGPUPlatformAndroid.cpp

@@ -18,6 +18,7 @@
 
 #include <utils/Panic.h>
 
+#include <android/native_window.h>
 #include <webgpu/webgpu_cpp.h>
 
 #include <cstdint>
@@ -28,6 +29,14 @@
 
 namespace filament::backend {
 
+wgpu::Extent2D WebGPUPlatform::getSurfaceExtent(void* nativeWindow) const {
+    ANativeWindow* window = static_cast<ANativeWindow*>(nativeWindow);
+    return wgpu::Extent2D{
+        .width = static_cast<uint32_t>(ANativeWindow_getWidth(window)),
+        .height = static_cast<uint32_t>(ANativeWindow_getHeight(window))
+    };
+}
+
 wgpu::Surface WebGPUPlatform::createSurface(void* nativeWindow, uint64_t /*flags*/) {
     wgpu::SurfaceSourceAndroidNativeWindow surfaceSourceAndroidWindow{};
     surfaceSourceAndroidWindow.window = nativeWindow;

filament/backend/src/webgpu/platform/WebGPUPlatformApple.mm

@@ -24,21 +24,8 @@
 #include <cstdint>
 
 // Platform specific includes and defines
-#if defined(__APPLE__)
-    #include <Cocoa/Cocoa.h>
-    #import <QuartzCore/CAMetalLayer.h>
-#elif defined(FILAMENT_IOS)
-    // Metal is not available when building for the iOS simulator on Desktop.
-    #define METAL_AVAILABLE __has_include(<QuartzCore/CAMetalLayer.h>)
-    #if METAL_AVAILABLE
-        #import <Metal/Metal.h>
-        #import <QuartzCore/CAMetalLayer.h>
-    #endif
-    // is this needed?
-    #define METALVIEW_TAG 255
-#else
-    #error Not a supported Apple + WebGPU platform
-#endif
+#include <Cocoa/Cocoa.h>
+#import <QuartzCore/CAMetalLayer.h>
 
 /**
  * Apple (Mac OS and IOS) specific implementation aspects of the WebGPU backend
@@ -46,14 +33,19 @@
 
 namespace filament::backend {
 
+wgpu::Extent2D WebGPUPlatform::getSurfaceExtent(void* nativeWindow) const {
+    // Both IOS and MacOS expects CAMetalLayer.
+    CAMetalLayer* metalLayer = (__bridge CAMetalLayer*) nativeWindow;
+    return wgpu::Extent2D{
+        .width = static_cast<uint32_t>(metalLayer.drawableSize.width),
+        .height = static_cast<uint32_t>(metalLayer.drawableSize.height)
+    };
+}
+
 wgpu::Surface WebGPUPlatform::createSurface(void* nativeWindow, uint64_t /*flags*/) {
     wgpu::Surface surface = nullptr;
-#if defined(__APPLE__)
-    auto nsView = (__bridge NSView*) nativeWindow;
-    FILAMENT_CHECK_POSTCONDITION(nsView) << "Unable to obtain Metal-backed NSView.";
-    [nsView setWantsLayer:YES];
-    id metalLayer = [CAMetalLayer layer];
-    [nsView setLayer:metalLayer];
+    // Both IOS and MacOS expects CAMetalLayer.
+    CAMetalLayer* metalLayer = (__bridge CAMetalLayer*) nativeWindow;
     wgpu::SurfaceSourceMetalLayer surfaceSourceMetalLayer{};
     surfaceSourceMetalLayer.layer = (__bridge void*) metalLayer;
     wgpu::SurfaceDescriptor surfaceDescriptor = {
@@ -62,19 +54,6 @@ wgpu::Surface WebGPUPlatform::createSurface(void* nativeWindow, uint64_t /*flags
     };
     surface = mInstance.CreateSurface(&surfaceDescriptor);
     FILAMENT_CHECK_POSTCONDITION(surface != nullptr) << "Unable to create Metal-backed surface.";
-#elif defined(FILAMENT_IOS)
-    CAMetalLayer* metalLayer = (CAMetalLayer*) nativeWindow;
-    wgpu::SurfaceSourceMetalLayer surfaceSourceMetalLayer{};
-    surfaceSourceMetalLayer.layer = (__bridge void*) metalLayer;
-    wgpu::SurfaceDescriptor surfaceDescriptor = {
-        .nextInChain = &surfaceSourceMetalLayer,
-        .label = "metal_surface",
-    };
-    surface = mInstance.CreateSurface(&surfaceDescriptor);
-    FILAMENT_CHECK_POSTCONDITION(surface != nullptr) << "Unable to create Metal-backed surface.";
-#else
-    #error Not a supported Apple + WebGPU platform
-#endif
     return surface;
 }
 

filament/backend/src/webgpu/platform/WebGPUPlatformLinux.cpp

@@ -79,6 +79,74 @@
 
 namespace filament::backend {
 
+wgpu::Extent2D WebGPUPlatform::getSurfaceExtent(void* nativeWindow) const {
+    auto surfaceExtent = wgpu::Extent2D{};
+#if defined(__linux__) && defined(FILAMENT_SUPPORTS_WAYLAND)
+    wl* ptrval = reinterpret_cast<wl*>(nativeWindow);
+    surfaceExtent.width = ptrval->width;
+    surfaceExtent.height = ptrval->height;
+    FILAMENT_CHECK_POSTCONDITION(surfaceExtent.width != 0 && surfaceExtent.height != 0)
+            << "Unable to get window size for Linux Wayland-backed surface.";
+#elif defined(LINUX_OR_FREEBSD) && defined(FILAMENT_SUPPORTS_X11)
+    if (g_x11.library == nullptr) {
+        g_x11.library = dlopen(LIBRARY_X11, RTLD_LOCAL | RTLD_NOW);
+        FILAMENT_CHECK_PRECONDITION(g_x11.library) << "Unable to open X11 library.";
+        #if defined(FILAMENT_SUPPORTS_XCB)
+            g_x11.xcbConnect = (XCB_CONNECT) dlsym(g_x11.library, "xcb_connect");
+            int screen = 0;
+            g_x11.connection = g_x11.xcbConnect(nullptr, &screen);
+        #endif
+        #if defined(FILAMENT_SUPPORTS_XLIB)
+            g_x11.openDisplay = (X11_OPEN_DISPLAY) dlsym(g_x11.library, "XOpenDisplay");
+            g_x11.display = g_x11.openDisplay(NULL);
+            FILAMENT_CHECK_PRECONDITION(g_x11.display) << "Unable to open X11 display.";
+        #endif
+    }
+    #if defined(FILAMENT_SUPPORTS_XCB) || defined(FILAMENT_SUPPORTS_XLIB)
+        bool useXcb = false;
+    #endif
+    #if defined(FILAMENT_SUPPORTS_XCB)
+        #if defined(FILAMENT_SUPPORTS_XLIB)
+            useXcb = (SWAP_CHAIN_CONFIG_ENABLE_XCB) != 0;
+        #else
+            useXcb = true;
+        #endif
+        if (useXcb) {
+            const xcb_setup_t* setup = xcb_get_setup(g_x11.connection);
+            xcb_screen_iterator_t screen_iter = xcb_setup_roots_iterator(setup);
+            xcb_screen_t* screen = screen_iter.data;
+            surfaceExtent.width = static_cast<uint32_t>(screen->width_in_pixels);
+            surfaceExtent.height = static_cast<uint32_t>(screen->height_in_pixels);
+            FILAMENT_CHECK_POSTCONDITION(surfaceExtent.width != 0 && surfaceExtent.height != 0)
+                    << "Unable to get window surface size for Linux (or FreeBSD) "
+                       "XCB-backed surface.";
+        }
+    #endif
+    #if defined(FILAMENT_SUPPORTS_XLIB)
+        if (!useXcb) {
+            int screenNumber = DefaultScreen(g_x11.display);
+            Screen* screen = ScreenOfDisplay(g_x11.display, screenNumber);
+            surfaceExtent.width = static_cast<uint32_t>(WidthOfScreen(screen));
+            surfaceExtent.height = static_cast<uint32_t>(HeightOfScreen(screen));
+            FILAMENT_CHECK_POSTCONDITION(surfaceExtent.width != 0 && surfaceExtent.height != 0)
+                    << "Unable to get window surface size for Linux (or FreeBSD) "
+                       "XLib-backed surface.";
+        }
+    #endif
+    FILAMENT_CHECK_POSTCONDITION(surfaceExtent.width != 0 && surfaceExtent.height != 0)
+            << "Cannot get window surface size for X11 surface for Linux (or FreeBSD) OS "
+               "(not built with support for XCB or XLIB?)";
+#elif defined(__linux__)
+    FILAMENT_CHECK_POSTCONDITION(surfaceExtent.width != 0 && surfaceExtent.height != 0)
+            << "Cannot get window surface size for Linux (or FreeBSD) OS "
+               "(not built with support for Wayland or X11?)";
+#else
+    FILAMENT_CHECK_POSTCONDITION(surfaceExtent.width != 0 && surfaceExtent.height != 0)
+            << "Not a supported (Linux) OS + WebGPU platform";
+#endif
+    return surfaceExtent;
+}
+
 wgpu::Surface WebGPUPlatform::createSurface(void* nativeWindow, uint64_t flags) {
     wgpu::Surface surface = nullptr;
 #if defined(__linux__) && defined(FILAMENT_SUPPORTS_WAYLAND)

filament/backend/src/webgpu/platform/WebGPUPlatformWindows.cpp

@@ -30,6 +30,16 @@
 
 namespace filament::backend {
 
+wgpu::Extent2D WebGPUPlatform::getSurfaceExtent(void* nativeWindow) const {
+    HWND window = static_cast<HWND>(nativeWindow);
+    RECT windowRect;
+    GetWindowRect(window, &windowRect);
+    return wgpu::Extent2D{
+        .width = static_cast<uint32_t>(windowRect.right - windowRect.left),
+        .height = static_cast<uint32_t>(windowRect.bottom - windowRect.top)
+    };
+}
+
 wgpu::Surface WebGPUPlatform::createSurface(void* nativeWindow, uint64_t /*flags*/) {
     // TODO verify this is necessary for Dawn implementation as well:
     // On (at least) NVIDIA drivers, the Vulkan implementation (specifically the call to

filament/backend/test/BackendTest.cpp

@@ -43,24 +43,29 @@ using namespace image;
 namespace test {
 
 Backend BackendTest::sBackend = Backend::NOOP;
+OperatingSystem BackendTest::sOperatingSystem = OperatingSystem::OTHER;
 bool BackendTest::sIsMobilePlatform = false;
 
-void BackendTest::init(Backend backend, bool isMobilePlatform) {
+void BackendTest::init(Backend backend, OperatingSystem operatingSystem, bool isMobilePlatform) {
     sBackend = backend;
+    sOperatingSystem = operatingSystem;
     sIsMobilePlatform = isMobilePlatform;
 }
 
 BackendTest::BackendTest() : commandBufferQueue(CONFIG_MIN_COMMAND_BUFFERS_SIZE,
         CONFIG_COMMAND_BUFFERS_SIZE, /*mPaused=*/false) {
     initializeDriver();
+    mImageExpectations.emplace(getDriverApi());
 }
 
 BackendTest::~BackendTest() {
+    // Ensure all graphics commands and callbacks are finished.
+    flushAndWait();
+    mImageExpectations->evaluate();
     // Note: Don't terminate the driver for OpenGL, as it wipes away the context and removes the buffer from the screen.
     if (sBackend == Backend::OPENGL) {
         return;
     }
-    flushAndWait();
     driver->terminate();
     delete driver;
 }
@@ -154,49 +159,16 @@ void BackendTest::renderTriangle(
     api.endRenderPass();
 }
 
-void BackendTest::readPixelsAndAssertHash(const char* testName, size_t width, size_t height,
-        Handle<HwRenderTarget> rt, uint32_t expectedHash, bool exportScreenshot) {
-    void* buffer = calloc(1, width * height * 4);
+bool BackendTest::matchesEnvironment(Backend backend) {
+    return sBackend == backend;
+}
 
-    struct Capture {
-        uint32_t expectedHash;
-        char* name;
-        bool exportScreenshot;
-        size_t width, height;
-    };
-    auto* c = new Capture();
-    c->expectedHash = expectedHash;
-    c->name = strdup(testName);
-    c->exportScreenshot = exportScreenshot;
-    c->width = width;
-    c->height = height;
+bool BackendTest::matchesEnvironment(OperatingSystem operatingSystem) {
+    return sOperatingSystem == operatingSystem;
+}
 
-    PixelBufferDescriptor pbd(buffer, width * height * 4, PixelDataFormat::RGBA, PixelDataType::UBYTE,
-            1, 0, 0, width, [](void* buffer, size_t size, void* user) {
-                auto* c = (Capture*)user;
-
-                // Export a screenshot, if requested.
-                if (c->exportScreenshot) {
-#ifndef FILAMENT_IOS
-                    LinearImage image(c->width, c->height, 4);
-                    image = toLinearWithAlpha<uint8_t>(c->width, c->height, c->width * 4,
-                            (uint8_t*) buffer);
-                    const std::string png = std::string(c->name) + ".png";
-                    std::ofstream outputStream(png.c_str(), std::ios::binary | std::ios::trunc);
-                    ImageEncoder::encode(outputStream, ImageEncoder::Format::PNG, image, "",
-                            png);
-#endif
-                }
-
-                // Hash the contents of the buffer and check that they match.
-                uint32_t hash = utils::hash::murmur3((const uint32_t*) buffer, size / 4, 0);
-                ASSERT_EQ(hash, c->expectedHash) << c->name << " failed: hashes do not match." << std::endl;
-
-                free(buffer);
-                free(c->name);
-                free(c);
-            }, (void*)c);
-    getDriverApi().readPixels(rt, 0, 0, width, height, std::move(pbd));
+bool BackendTest::matchesEnvironment(OperatingSystem operatingSystem, Backend backend) {
+    return matchesEnvironment(operatingSystem) && matchesEnvironment(backend);
 }
 
 class Environment : public ::testing::Environment {
@@ -210,8 +182,8 @@ public:
     }
 };
 
-void initTests(Backend backend, bool isMobile, int& argc, char* argv[]) {
-    BackendTest::init(backend, isMobile);
+void initTests(Backend backend, OperatingSystem operatingSystem, bool isMobile, int& argc, char* argv[]) {
+    BackendTest::init(backend, operatingSystem, isMobile);
     ::testing::InitGoogleTest(&argc, argv);
     ::testing::AddGlobalTestEnvironment(new Environment);
 }

filament/backend/test/BackendTest.h

@@ -25,15 +25,17 @@
 #include "private/backend/DriverApi.h"
 
 #include "PlatformRunner.h"
+#include "ImageExpectations.h"
 
 namespace test {
 
 class BackendTest : public ::testing::Test {
 public:
 
-    static void init(Backend backend, bool isMobilePlatform);
+    static void init(Backend backend, OperatingSystem operatingSystem, bool isMobilePlatform);
 
     static Backend sBackend;
+    static OperatingSystem sOperatingSystem;
     static bool sIsMobilePlatform;
 
 protected:
@@ -64,13 +66,14 @@ protected:
             filament::backend::Handle<filament::backend::HwProgram> program,
             const filament::backend::RenderPassParams& params);
 
-    void readPixelsAndAssertHash(const char* testName, size_t width, size_t height,
-            filament::backend::Handle<filament::backend::HwRenderTarget> rt, uint32_t expectedHash,
-            bool exportScreenshot = false);
-
     filament::backend::DriverApi& getDriverApi() { return *commandStream; }
     filament::backend::Driver& getDriver() { return *driver; }
 
+    ImageExpectations& getExpectations() { return *mImageExpectations; }
+
+    static bool matchesEnvironment(Backend backend);
+    static bool matchesEnvironment(OperatingSystem operatingSystem);
+    static bool matchesEnvironment(OperatingSystem operatingSystem, Backend backend);
 private:
 
     filament::backend::Driver* driver = nullptr;
@@ -78,6 +81,10 @@ private:
     std::unique_ptr<filament::backend::DriverApi> commandStream;
 
     filament::backend::Handle<filament::backend::HwBufferObject> uniform;
+
+    // This isn't truly optional, it just needs to delay construction until after the driver has
+    // been initialized
+    std::optional<ImageExpectations> mImageExpectations;
 };
 
 } // namespace test

filament/backend/test/ImageExpectations.cpp

@@ -16,7 +16,6 @@
 
 #include "ImageExpectations.h"
 
-#include "gtest/gtest.h"
 #include "gmock/gmock.h"
 #include "absl/strings/str_format.h"
 #include "utils/Hash.h"
@@ -28,14 +27,17 @@
 #ifndef FILAMENT_IOS
 
 #include <imageio/ImageEncoder.h>
+#include <imageio/ImageDecoder.h>
 #include <image/ColorTransform.h>
 
 #endif
 
 ScreenshotParams::ScreenshotParams(int width, int height, std::string fileName,
-        uint32_t expectedPixelHash)
-        : mWidth(width), mHeight(height), mFileName(std::move(fileName)),
-          mExpectedPixelHash(expectedPixelHash) {}
+        uint32_t expectedHash)
+    : mWidth(width),
+      mHeight(height),
+      mExpectedPixelHash(expectedHash),
+      mFileName(std::move(fileName)) {}
 
 int ScreenshotParams::width() const {
     return mWidth;
@@ -49,24 +51,28 @@ uint32_t ScreenshotParams::expectedHash() const {
     return mExpectedPixelHash;
 }
 
-std::string ScreenshotParams::outputDirectoryPath() const {
-    return ".";
+std::string ScreenshotParams::actualDirectoryPath() {
+    return "images/actual_images";
 }
 
-std::string ScreenshotParams::generatedActualFileName() const {
+std::string ScreenshotParams::actualFileName() const {
     return absl::StrFormat("%s_actual.png", mFileName);
 }
 
-std::string ScreenshotParams::generatedActualFilePath() const {
-    return absl::StrFormat("%s/%s", outputDirectoryPath(), generatedActualFileName());
+std::string ScreenshotParams::actualFilePath() const {
+    return absl::StrFormat("%s/%s", actualDirectoryPath(), actualFileName());
 }
 
-std::string ScreenshotParams::goldenFileName() const {
-    return absl::StrFormat("%s_golden.png", mFileName);
+std::string ScreenshotParams::expectedDirectoryPath() {
+    return "images/expected_images";
 }
 
-std::string ScreenshotParams::goldenFilePath() const {
-    return absl::StrFormat("%s/%s", outputDirectoryPath(), goldenFileName());
+std::string ScreenshotParams::expectedFileName() const {
+    return absl::StrFormat("%s.png", mFileName);
+}
+
+std::string ScreenshotParams::expectedFilePath() const {
+    return absl::StrFormat("%s/%s", expectedDirectoryPath(), expectedFileName());
 }
 
 ImageExpectation::ImageExpectation(const char* fileName, int lineNumber,
@@ -91,11 +97,22 @@ void ImageExpectation::evaluate() {
 
 void ImageExpectation::compareImage() const {
     bool bytesFilled = mResult.bytesFilled();
+    // If this fails, it likely means that BackendTest::flushAndWait needs to be called before
+    // ImageExpectations is evaluated or destroyed.
     EXPECT_THAT(bytesFilled, testing::IsTrue())
                         << "Render target wasn't copied to the buffer for " << mFileName;
     if (bytesFilled) {
+        // Rather than directly compare the two images compare their hashes because comparing very
+        // large arrays generates way too much debug output to be useful.
         uint32_t actualHash = mResult.hash();
-        EXPECT_THAT(actualHash, testing::Eq(mParams.expectedHash()));
+#ifndef FILAMENT_IOS
+        LoadedPng loadedImage(mParams.expectedFilePath());
+        uint32_t loadedImageHash = loadedImage.hash();
+        EXPECT_THAT(actualHash, testing::Eq(loadedImageHash)) << mParams.expectedFileName();
+#endif
+        // For builds that can't load PNGs (currently iOS only) use the expected hash.
+        EXPECT_THAT(actualHash, testing::Eq(mParams.expectedHash())) << mParams.expectedFileName();
+        // TODO: Add better debug output, such as generating a diff image.
     }
 }
 
@@ -109,12 +126,13 @@ ImageExpectations::~ImageExpectations() {
 
 void ImageExpectations::addExpectation(const char* fileName, int lineNumber,
         filament::backend::RenderTargetHandle renderTarget, ScreenshotParams params) {
-    mExpectations.emplace_back(fileName, lineNumber, mApi, std::move(params), renderTarget);
+    mExpectations.emplace_back(std::make_unique<ImageExpectation>(fileName, lineNumber, mApi,
+            std::move(params), renderTarget));
 }
 
 void ImageExpectations::evaluate() {
     for (auto& expectation: mExpectations) {
-        expectation.evaluate();
+        expectation->evaluate();
     }
     mExpectations.clear();
 }
@@ -122,32 +140,28 @@ void ImageExpectations::evaluate() {
 RenderTargetDump::RenderTargetDump(filament::backend::DriverApi& api,
         filament::backend::RenderTargetHandle renderTarget, const ScreenshotParams& params)
         : mInternal(std::make_unique<RenderTargetDump::Internal>(params)) {
-#ifdef FILAMENT_IOS
-    bytesFilled_ = true;
-    bytes_.resize(size);
-    std::fill(bytes_.begin(), bytes_.end(), 0);
-#else
     const size_t size = mInternal->params.width() * mInternal->params.height() * 4;
     mInternal->bytes.resize(size);
 
     auto cb = [](void* buffer, size_t size, void* user) {
         auto* internal = static_cast<RenderTargetDump::Internal*>(user);
+        internal->bytesFilled = true;
+#ifndef FILAMENT_IOS
         image::LinearImage image(internal->params.width(), internal->params.width(), 4);
         image = image::toLinearWithAlpha<uint8_t>(internal->params.width(),
                 internal->params.height(),
                 internal->params.width() * 4, (uint8_t*)buffer);
-        std::string filePath = internal->params.generatedActualFilePath();
+        std::string filePath = internal->params.actualFilePath();
         std::ofstream pngStream(filePath, std::ios::binary | std::ios::trunc);
         image::ImageEncoder::encode(pngStream, image::ImageEncoder::Format::PNG, image, "",
                 filePath);
-        internal->bytesFilled = true;
+#endif
     };
     filament::backend::PixelBufferDescriptor pb(mInternal->bytes.data(), size,
             filament::backend::PixelDataFormat::RGBA, filament::backend::PixelDataType::UBYTE, cb,
             (void*)mInternal.get());
     api.readPixels(renderTarget, 0, 0, mInternal->params.width(), mInternal->params.height(),
             std::move(pb));
-#endif
 }
 
 RenderTargetDump::~RenderTargetDump() {
@@ -169,4 +183,30 @@ bool RenderTargetDump::bytesFilled() const {
     return mInternal->bytesFilled;
 }
 
-RenderTargetDump::Internal::Internal(const ScreenshotParams& params) : params(params) {}
+RenderTargetDump::Internal::Internal(const ScreenshotParams& params) : params(params) {}
+
+LoadedPng::LoadedPng(std::string filePath) : mFilePath(std::move(filePath)) {
+#ifndef FILAMENT_IOS
+    std::ifstream pngStream(mFilePath, std::ios::binary);
+    image::LinearImage loadedImage = image::ImageDecoder::decode(pngStream, filePath,
+            image::ImageDecoder::ColorSpace::LINEAR);
+    size_t valuesInImage = loadedImage.getWidth() * loadedImage.getHeight() *
+                          loadedImage.getChannels();
+    // The linear image is loaded with each component as [0.0, 1.0] but should be [0, 255], so
+    // convert them.
+    mBytes = std::vector<unsigned char>(valuesInImage);
+    for (int i = 0; i < valuesInImage; ++i) {
+        mBytes[i] = static_cast<uint8_t>(loadedImage.get<float>()[i] * 255.0f);
+    }
+#endif
+    // For platforms that don't support the image loading library, leave the loaded data blank.
+}
+
+uint32_t LoadedPng::hash() const {
+    EXPECT_THAT(mBytes, testing::Not(testing::IsEmpty()))
+            << "Failed to load expected test result: " << mFilePath;
+    if (mBytes.empty()) {
+        return 0;
+    }
+    return utils::hash::murmur3((uint32_t*)mBytes.data(), mBytes.size() / 4, 0);
+}

filament/backend/test/ImageExpectations.h

@@ -46,11 +46,12 @@ public:
     int height() const;
     uint32_t expectedHash() const;
 
-    std::string outputDirectoryPath() const;
-    std::string generatedActualFileName() const;
-    std::string generatedActualFilePath() const;
-    std::string goldenFileName() const;
-    std::string goldenFilePath() const;
+    static std::string actualDirectoryPath();
+    std::string actualFileName() const;
+    std::string actualFilePath() const;
+    static std::string expectedDirectoryPath();
+    std::string expectedFileName() const;
+    std::string expectedFilePath() const;
 
 private:
     int mWidth;
@@ -98,6 +99,17 @@ private:
     std::unique_ptr<Internal> mInternal;
 };
 
+class LoadedPng {
+public:
+    explicit LoadedPng(std::string filePath);
+
+    uint32_t hash() const;
+
+private:
+    std::string mFilePath;
+    std::vector<unsigned char> mBytes;
+};
+
 class ImageExpectation {
 public:
     ImageExpectation(const char* fileName, int lineNumber, filament::backend::DriverApi& api,
@@ -130,7 +142,8 @@ public:
 
 private:
     filament::backend::DriverApi& mApi;
-    std::vector<ImageExpectation> mExpectations;
+    // Store expectations in unique pointers because they are self referential.
+    std::vector<std::unique_ptr<ImageExpectation>> mExpectations;
 };
 
 #endif //TNT_IMAGE_EXPECTATIONS_H

filament/backend/test/PlatformRunner.cpp

@@ -0,0 +1,60 @@
+/*
+ * Copyright (C) 2019 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "PlatformRunner.h"
+
+namespace utils {
+
+template<>
+CString to_string<test::Backend>(test::Backend backend) noexcept {
+    switch (backend) {
+        case test::Backend::OPENGL: {
+            return "OpenGL";
+        }
+        case test::Backend::VULKAN: {
+            return "Vulkan";
+        }
+        case test::Backend::METAL: {
+            return "Metal";
+        }
+        case test::Backend::WEBGPU: {
+            return "WebGPU";
+        }
+        case test::Backend::NOOP:
+        default: {
+            return "No-op";
+        }
+    }
+}
+
+template<>
+CString to_string(test::OperatingSystem os) noexcept {
+    switch (os) {
+        case test::OperatingSystem::LINUX: {
+            return "Linux";
+        }
+        case test::OperatingSystem::APPLE: {
+            return "Apple";
+        }
+        case test::OperatingSystem::OTHER:
+        default: {
+            return "Other";
+        }
+    }
+}
+
+} // namespace utils
+

filament/backend/test/PlatformRunner.h

@@ -19,6 +19,7 @@
 
 #include <stdint.h>
 #include <stddef.h>
+#include "utils/CString.h"
 
 namespace test {
 
@@ -34,6 +35,15 @@ enum class Backend : uint8_t {
     NOOP = 5,
 };
 
+enum class OperatingSystem: uint8_t {
+    OTHER = 1,
+    // Also represents android phones.
+    LINUX = 2,
+    // Also represents iOS phones.
+    APPLE = 3,
+    // TODO: When tests support windows add it here.
+};
+
 struct NativeView {
     void* ptr = nullptr;
     size_t width = 0, height = 0;
@@ -51,9 +61,10 @@ NativeView getNativeView();
  * No tests will be run yet.
  *
  * @param backend The backend to run the tests on.
+ * @param operatingSystem The operating system the tests are being run on.
  * @param isMobile True if the platform is a mobile platform (iOS or Android).
  */
-void initTests(Backend backend, bool isMobile, int& argc, char* argv[]);
+void initTests(Backend backend, OperatingSystem operatingSystem, bool isMobile, int& argc, char* argv[]);
 
 /**
  * Test runners should call runTests when they are ready for tests to be run.
@@ -68,6 +79,6 @@ int runTests();
  */
 Backend parseArgumentsForBackend(int argc, char* argv[]);
 
-}
+} // namespace test
 
 #endif

filament/backend/test/Shader.cpp

@@ -23,30 +23,44 @@ namespace test {
 
 using namespace filament::backend;
 
-Shader::Shader(DriverApi& api, Cleanup& cleanup, ShaderConfig config) {
-    utils::FixedCapacityVector<DescriptorSetLayoutBinding> kLayouts(config.uniformNames.size());
-    for (unsigned char i = 0; i < config.uniformNames.size(); ++i) {
-        kLayouts[i] =
-                { DescriptorType::UNIFORM_BUFFER, ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, i };
-    };
+Shader::Shader(DriverApi& api, Cleanup& cleanup, ShaderConfig config) : mCleanup(cleanup) {
+    utils::FixedCapacityVector<DescriptorSetLayoutBinding> kLayouts(config.uniforms.size());
+    for (unsigned char i = 0; i < config.uniforms.size(); ++i) {
+        kLayouts[i] = {
+                config.uniforms[i].type.value_or(DescriptorType::UNIFORM_BUFFER),
+                ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, i };
+    }
 
+    // This assumes that the uniforms will all be in a single descriptor set at index 1.
+    // If there are shaders with uniforms in other sets then ShaderConfig will need to be expanded
+    // to accommodate that.
+    size_t kDescriptorSetIndex = 1;
     filamat::DescriptorSets descriptors;
-    for (unsigned char i = 0; i < config.uniformNames.size(); ++i) {
-        descriptors[i + 1] = {{ config.uniformNames[i], kLayouts[i], {}}};
+    descriptors[kDescriptorSetIndex] = filamat::DescriptorSetInfo(config.uniforms.size());
+    for (unsigned char i = 0; i < config.uniforms.size(); ++i) {
+        descriptors[kDescriptorSetIndex][i] = {
+                config.uniforms[i].name, kLayouts[i], config.uniforms[i].samplerInfo };
     }
     ShaderGenerator shaderGen(
             std::move(config.vertexShader), std::move(config.fragmentShader), BackendTest::sBackend,
             BackendTest::sIsMobilePlatform, std::move(descriptors));
     Program prog = shaderGen.getProgram(api);
-    for (unsigned char i = 0; i < config.uniformNames.size(); ++i) {
-        prog.descriptorBindings(1, {{ config.uniformNames[i], DescriptorType::UNIFORM_BUFFER, i }});
+
+    Program::DescriptorBindingsInfo bindingsInfo(config.uniforms.size());
+    for (unsigned char i = 0; i < config.uniforms.size(); ++i) {
+        bindingsInfo[i] = {
+                config.uniforms[i].name,
+                config.uniforms[i].type.value_or(DescriptorType::UNIFORM_BUFFER), i };
     }
+    prog.descriptorBindings(1, bindingsInfo);
     mProgram = cleanup.add(api.createProgram(std::move(prog)));
 
     mDescriptorSetLayout = cleanup.add(
             api.createDescriptorSetLayout(DescriptorSetLayout{ kLayouts }));
+}
 
-    mDescriptorSet = cleanup.add(api.createDescriptorSet(mDescriptorSetLayout));
+filament::backend::DescriptorSetHandle Shader::createDescriptorSet(DriverApi& api) const {
+    return mCleanup.add(api.createDescriptorSet(mDescriptorSetLayout));
 }
 
 filament::backend::ProgramHandle Shader::getProgram() const {
@@ -57,8 +71,4 @@ filament::backend::DescriptorSetLayoutHandle Shader::getDescriptorSetLayout() co
     return mDescriptorSetLayout;
 }
 
-filament::backend::DescriptorSetHandle Shader::getDescriptorSet() const {
-    return mDescriptorSet;
-}
-
 } // namespace test

filament/backend/test/Shader.h

@@ -18,14 +18,22 @@
 #define TNT_SHADER_H
 
 #include "Lifetimes.h"
+#include "private/filament/SamplerInterfaceBlock.h"
 
 namespace test {
 
+struct UniformConfig {
+    utils::CString name;
+    // If not specified this will be DescriptorType::UNIFORM_BUFFER
+    std::optional<filament::backend::DescriptorType> type;
+    std::optional<filament::SamplerInterfaceBlock::SamplerInfo> samplerInfo;
+};
+
 // All describing a shader that should be created.
 struct ShaderConfig {
     std::string vertexShader;
     std::string fragmentShader;
-    std::vector<utils::CString> uniformNames;
+    std::vector<UniformConfig> uniforms;
 };
 
 // All values describing a uniform.
@@ -35,6 +43,7 @@ struct ResolvedUniformBindingConfig {
     uint32_t byteOffset;
     filament::backend::descriptor_set_t set;
     filament::backend::descriptor_binding_t binding;
+    std::optional<filament::backend::DescriptorSetHandle> descriptorSet;
 };
 
 // An equivalent to ResolvedUniformBindingConfig with all fields optional.
@@ -46,6 +55,7 @@ struct UniformBindingConfig {
     std::optional<uint32_t> byteOffset;
     std::optional<filament::backend::descriptor_set_t> set;
     std::optional<filament::backend::descriptor_binding_t> binding;
+    std::optional<filament::backend::DescriptorSetHandle> descriptorSet;
 
     template<typename UniformType>
     ResolvedUniformBindingConfig resolve();
@@ -79,12 +89,14 @@ public:
 
     filament::backend::ProgramHandle getProgram() const;
     filament::backend::DescriptorSetLayoutHandle getDescriptorSetLayout() const;
-    filament::backend::DescriptorSetHandle getDescriptorSet() const;
+
+    filament::backend::DescriptorSetHandle createDescriptorSet(
+            filament::backend::DriverApi& api) const;
 
 protected:
+    Cleanup& mCleanup;
     filament::backend::ProgramHandle mProgram;
     filament::backend::DescriptorSetLayoutHandle mDescriptorSetLayout;
-    filament::backend::DescriptorSetHandle mDescriptorSet;
 };
 
 template<typename UniformType>
@@ -95,7 +107,8 @@ ResolvedUniformBindingConfig UniformBindingConfig::resolve() {
             .bufferSize = bufferSize.value_or(resolvedDataSize),
             .byteOffset = byteOffset.value_or(0),
             .set = set.value_or(1),
-            .binding = binding.value_or(0)
+            .binding = binding.value_or(0),
+            .descriptorSet = descriptorSet
     };
 }
 
@@ -120,9 +133,16 @@ void Shader::bindUniform(filament::backend::DriverApi& api,
         UniformBindingConfig config) const {
     auto resolvedConfig = config.resolve<UniformType>();
 
-    api.updateDescriptorSetBuffer(getDescriptorSet(), resolvedConfig.binding, hwBuffer, 0,
+    filament::backend::DescriptorSetHandle descriptorSet;
+    if (resolvedConfig.descriptorSet.has_value()) {
+        descriptorSet = *resolvedConfig.descriptorSet;
+    } else {
+        descriptorSet = createDescriptorSet(api);
+    }
+
+    api.updateDescriptorSetBuffer(descriptorSet, resolvedConfig.binding, hwBuffer, 0,
             resolvedConfig.bufferSize);
-    api.bindDescriptorSet(getDescriptorSet(), resolvedConfig.set, {});
+    api.bindDescriptorSet(descriptorSet, resolvedConfig.set, {});
 }
 
 template<typename UniformType>

filament/backend/test/SharedShaders.cpp

@@ -0,0 +1,253 @@
+/*
+ * Copyright (C) 2021 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "SharedShaders.h"
+
+#include "Shader.h"
+#include "absl/strings/str_format.h"
+#include "gtest/gtest.h"
+
+namespace test {
+
+using namespace filament::backend;
+
+namespace {
+
+// A shader stored in pieces so that uniform declarations can be injected.
+struct ShaderText {
+    std::string mPrefix;
+    std::string mBody;
+
+    std::string withUniform(const std::string& uniformText) const {
+        return absl::StrFormat("%s\n%s\n%s", mPrefix.c_str(), uniformText.c_str(), mBody.c_str());
+    }
+};
+
+std::optional<ShaderText> GetGlslVertexShader(VertexShaderType type) {
+    switch (type) {
+        case VertexShaderType::Noop: {
+            return ShaderText{
+                    R"(
+#version 450 core
+layout(location = 0) in vec4 mesh_position;
+)", R"(
+void main() {
+    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
+})" };
+        }
+        case VertexShaderType::Simple: {
+            return ShaderText{
+                    R"(
+#version 450 core
+layout(location = 0) in vec4 mesh_position;
+)", R"(
+void main() {
+    gl_Position = vec4(
+            mesh_position.xy * (params.scaleMinusOne.xy + 1.0) + params.offset.xy,
+            params.scaleMinusOne.z + 1.0,
+            1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
+})" };
+        }
+        case VertexShaderType::Textured: {
+            return ShaderText{
+                    R"(
+#version 450 core
+layout(location = 0) in vec4 mesh_position;
+layout(location = 0) out vec2 uv;
+)", R"(
+void main() {
+    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+    uv = (mesh_position.xy * 0.5 + 0.5);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
+})" };
+        }
+        default:
+            return std::nullopt;
+    }
+}
+
+std::optional<ShaderText> GetGlslFragmentShader(FragmentShaderType type) {
+    switch (type) {
+        case FragmentShaderType::White: {
+            return ShaderText{
+                    R"(
+#version 450 core
+precision mediump int; precision highp float;
+layout(location = 0) out vec4 fragColor;
+)", R"(
+void main() {
+    fragColor = vec4(1.0, 1.0, 1.0, 1.0);
+})" };
+        }
+        case FragmentShaderType::SolidColored: {
+            return ShaderText{
+                    R"(
+#version 450 core
+precision mediump int; precision highp float;
+layout(location = 0) out vec4 fragColor;
+)", R"(
+void main() {
+    fragColor = params.color;
+})" };
+        }
+        case FragmentShaderType::Textured: {
+            return ShaderText{
+                    R"(
+#version 450 core
+precision mediump int; precision highp float;
+layout(location = 0) out vec4 fragColor;
+layout(location = 0) in vec2 uv;
+)", R"(
+void main() {
+    fragColor = texture(test_tex, uv);
+})" };
+        }
+        default:
+            return std::nullopt;
+    }
+}
+
+std::optional<std::string> GetGlslUniform(ShaderUniformType type) {
+    switch (type) {
+        case ShaderUniformType::None: {
+            return "";
+        }
+        case ShaderUniformType::Simple: {
+            return R"(
+layout(binding = 0, set = 1) uniform Params {
+    highp vec4 color;
+    // Use scaleMinusOne instead of scale so that a 0 initialized value is a good default
+    highp vec4 scaleMinusOne;
+    highp vec4 offset;
+} params;
+)";
+        }
+        case ShaderUniformType::SimpleWithPadding: {
+            return R"(
+layout(binding = 0, set = 1) uniform Params {
+    highp vec4 padding[4];  // offset of 64 bytes
+
+    highp vec4 color;
+    // Use scaleMinusOne instead of scale so that a 0 initialized value is a good default
+    highp vec4 scaleMinusOne;
+    highp vec4 offset;
+} params;
+)";
+        }
+        case ShaderUniformType::Sampler: {
+            return R"(
+layout(location = 0, set = 1) uniform sampler2D test_tex;
+)";
+        }
+        default:
+            return std::nullopt;
+    }
+}
+
+std::vector<UniformConfig> GetUniformConfig(ShaderUniformType type) {
+    switch (type) {
+        case ShaderUniformType::None: {
+            return {};
+        }
+        case ShaderUniformType::Simple: {
+            return {{ "Params" }};
+        }
+        case ShaderUniformType::SimpleWithPadding: {
+            return {{ "Params" }};
+        }
+        case ShaderUniformType::Sampler: {
+            filament::SamplerInterfaceBlock::SamplerInfo samplerInfo{
+                    "backend_test", "test_tex", 0,
+                    SamplerType::SAMPLER_2D, SamplerFormat::FLOAT, Precision::HIGH, false };
+            return {{
+                            "test_tex", DescriptorType::SAMPLER, samplerInfo
+                    }};
+        }
+        default:
+            abort();
+    }
+}
+
+ShaderLanguage getShaderLanguage(const Backend& backend) {
+    switch (backend) {
+        case Backend::METAL:
+            return ShaderLanguage::MSL;
+        case Backend::WEBGPU:
+            return ShaderLanguage::WGSL;
+        case Backend::VULKAN:
+        case Backend::NOOP:
+        case Backend::OPENGL:
+        default: {
+            return ShaderLanguage::GLSL;
+        }
+    }
+}
+
+} // namespace
+
+Shader SharedShaders::makeShader(filament::backend::DriverApi& api, Cleanup& cleanup,
+        ShaderRequest request) {
+    std::optional<ShaderText> vertex;
+    std::optional<ShaderText> fragment;
+    std::optional<std::string> uniform;
+    if (getShaderLanguage(BackendTest::sBackend) != ShaderLanguage::GLSL) {
+        // TODO: If any shaders need backend/shader language specific shaders rather than transpiled
+        // versions of the GLSL shader, check environment.
+    }
+    vertex = GetGlslVertexShader(request.mVertexType);
+    fragment = GetGlslFragmentShader(request.mFragmentType);
+    uniform = GetGlslUniform(request.mUniformType);
+    if (vertex.has_value() && fragment.has_value() && uniform.has_value()) {
+        return Shader(
+                api, cleanup, ShaderConfig{
+                        vertex->withUniform(*uniform), fragment->withUniform(*uniform),
+                        GetUniformConfig(request.mUniformType)}
+        );
+    }
+    abort();
+}
+
+std::string SharedShaders::getVertexShaderText(VertexShaderType vertex, ShaderUniformType uniform) {
+    std::optional<ShaderText> vertexText = GetGlslVertexShader(vertex);
+    std::optional<std::string> uniformText = GetGlslUniform(uniform);
+    if (!vertexText.has_value() || !uniformText.has_value()) {
+        abort();
+    }
+    return vertexText->withUniform(*uniformText);
+}
+
+std::string SharedShaders::getFragmentShaderText(FragmentShaderType fragment,
+        ShaderUniformType uniform) {
+    std::optional<ShaderText> fragmentText = GetGlslFragmentShader(fragment);
+    std::optional<std::string> uniformText = GetGlslUniform(uniform);
+    if (!fragmentText.has_value() || !uniformText.has_value()) {
+        abort();
+    }
+    return fragmentText->withUniform(*uniformText);
+}
+
+} // namespace test

filament/backend/test/SharedShaders.h

@@ -0,0 +1,50 @@
+/*
+ * Copyright (C) 2021 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef TNT_SHAREDSHADERS_H
+#define TNT_SHAREDSHADERS_H
+
+#include "Shader.h"
+#include "SharedShadersConstants.h"
+#include "Lifetimes.h"
+#include "PlatformRunner.h"
+
+namespace test {
+
+enum class ShaderLanguage : uint8_t {
+    GLSL,
+    MSL,
+    WGSL
+};
+
+struct ShaderRequest {
+    VertexShaderType mVertexType;
+    FragmentShaderType mFragmentType;
+    ShaderUniformType mUniformType;
+};
+
+class SharedShaders {
+public:
+    static Shader makeShader(filament::backend::DriverApi& api, Cleanup& cleanup,
+            ShaderRequest request);
+    static std::string getVertexShaderText(VertexShaderType vertex, ShaderUniformType uniform);
+    static std::string getFragmentShaderText(FragmentShaderType fragment,
+            ShaderUniformType uniform);
+};
+
+} // namespace test
+
+#endif //TNT_SHAREDSHADERS_H

filament/backend/test/SharedShadersConstants.h

@@ -0,0 +1,63 @@
+/*
+ * Copyright (C) 2021 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef TNT_SHAREDSHADERSCONSTANTS_H
+#define TNT_SHAREDSHADERSCONSTANTS_H
+
+#include "math/mathfwd.h"
+
+enum class ShaderUniformType : uint8_t {
+    None,
+    Simple,
+    SimpleWithPadding,
+    Sampler,
+};
+
+struct SimpleMaterialParams {
+    filament::math::float4 color;
+    // 1.0 will be added to this value before use.
+    // The XY values are used to scale position inputs and the Z value is used to set the output
+    // position's Z value.
+    filament::math::float4 scaleMinusOne;
+    // Offset will be applied after scale
+    filament::math::float4 offset;
+};
+
+struct SimpleWithPaddingMaterialParams {
+    // The associated uniform structure in the shader will have 64 bytes of padding at the beginning
+    // So users of this struct will need to add 64 bytes to its size and offset all uniform writes.
+    filament::math::float4 color;
+    // 1.0 will be added to this value before use.
+    // The XY values are used to scale position inputs and the Z value is used to set the output
+    // position's Z value.
+    filament::math::float4 scaleMinusOne;
+    // Offset will be applied after scale
+    filament::math::float4 offset;
+};
+
+enum class VertexShaderType : uint8_t {
+    Noop,
+    Simple,
+    Textured
+};
+
+enum class FragmentShaderType : uint8_t {
+    White,
+    SolidColored,
+    Textured
+};
+
+#endif //TNT_SHAREDSHADERSCONSTANTS_H

filament/backend/test/Skip.cpp

@@ -0,0 +1,96 @@
+/*
+ * Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "Skip.h"
+
+#include <sstream>
+
+namespace test {
+
+SkipEnvironment::SkipEnvironment(test::Backend backend) : backend(backend) {}
+SkipEnvironment::SkipEnvironment(test::OperatingSystem os) : os(os) {}
+SkipEnvironment::SkipEnvironment(test::OperatingSystem os, test::Backend backend)
+    : backend(backend),
+      os(os) {}
+
+bool SkipEnvironment::matches() {
+    bool backendMatches = !backend.has_value() || *backend == BackendTest::sBackend;
+    bool osMatches = !os.has_value() || *os == BackendTest::sOperatingSystem;
+    bool isMobileMatches = !isMobile.has_value() || *isMobile == BackendTest::sIsMobilePlatform;
+    return backendMatches && osMatches && isMobileMatches;
+}
+
+std::string SkipEnvironment::describe() {
+    std::stringstream result;
+    if (matches()) {
+        result << "environment matches because " << describe_actual_environment() << ".";
+    } else {
+        result << "environment does not match because " << describe_requirements() << " but "
+               << describe_actual_environment() << ".";
+    }
+    return result.str();
+}
+
+std::string SkipEnvironment::describe_actual_environment() {
+    bool resultWritten = false;
+    std::stringstream reality;
+    if (backend.has_value()) {
+        reality << "backend was " << utils::to_string(BackendTest::sBackend).c_str();
+        resultWritten = true;
+    }
+    if (os.has_value()) {
+        if (resultWritten) {
+            reality << ", and ";
+        }
+        reality << "operating system was "
+                << utils::to_string(BackendTest::sOperatingSystem).c_str();
+        resultWritten = true;
+    }
+    if (isMobile.has_value()) {
+        if (resultWritten) {
+            reality << ", and ";
+        }
+        reality << "device " << (BackendTest::sIsMobilePlatform ? "was" : "was not") << " mobile";
+        resultWritten = true;
+    }
+    return reality.str();
+}
+
+std::string SkipEnvironment::describe_requirements() {
+    bool resultWritten = false;
+    std::stringstream requirement;
+    if (backend.has_value()) {
+        requirement << "backend needs to be " << utils::to_string(*backend).c_str();
+        resultWritten = true;
+    }
+    if (os.has_value()) {
+        if (resultWritten) {
+            requirement << ", and ";
+        }
+        requirement << "operating system needs to be " << utils::to_string(*os).c_str();
+        resultWritten = true;
+    }
+    if (isMobile.has_value() && BackendTest::sIsMobilePlatform != isMobile) {
+        if (resultWritten) {
+            requirement << ", and ";
+        }
+        requirement << "device needs to " << (*isMobile ? "be" : "not be") << " mobile";
+        resultWritten = true;
+    }
+    return requirement.str();
+}
+
+} // namespace test

filament/backend/test/Skip.h

@@ -0,0 +1,55 @@
+/*
+ * Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef TNT_SKIP_H
+#define TNT_SKIP_H
+
+#include <gtest/gtest.h>
+#include "BackendTest.h"
+
+// skipEnvironment must be a test::SkipEnvironment
+#define SKIP_IF(skipEnvironment)                                                                   \
+do {                                                                                               \
+    SkipEnvironment skip(skipEnvironment);                                                         \
+    if (skip.matches()) {                                                                          \
+        GTEST_SKIP() << "Skipping test as the " << skip.describe();                                \
+    }                                                                                              \
+} while (false)
+
+namespace test {
+
+struct SkipEnvironment {
+    SkipEnvironment(const SkipEnvironment&) = default;
+    explicit SkipEnvironment(test::Backend backend);
+    explicit SkipEnvironment(test::OperatingSystem os);
+    SkipEnvironment(test::OperatingSystem os, test::Backend backend);
+
+    std::optional<test::Backend> backend;
+    std::optional<test::OperatingSystem> os;
+    std::optional<bool> isMobile;
+
+    bool matches();
+    // Describes the current state of either matching or mismatching.
+    std::string describe();
+    // Describe all the non-null requirements.
+    std::string describe_requirements();
+    // Describes the environment's status for all the attributes that are non-null.
+    std::string describe_actual_environment();
+};
+
+} // namespace test
+
+#endif// TNT_SKIP_H