Texture and Texture View creation

BUGS=[409099263]

.github/actions/dep-versions/action.yml

@@ -0,0 +1,7 @@
+name: 'Set up dependency versions'
+runs:
+  using: "composite"
+  steps:
+    - name: Set up dependency versions
+      shell: bash    
+      run: cat ./build/common/versions >> $GITHUB_ENV

.github/actions/linux-prereq/action.yml

@@ -0,0 +1,38 @@
+name: 'Linux Preqrequisites'
+runs:
+  using: "composite"
+  steps:
+    - uses: ./.github/actions/dep-versions
+    - name: Install Linux Prerequisites
+      shell: bash
+      run: |
+        set -xe
+        # See https://askubuntu.com/questions/272248/processing-triggers-for-man-db/1476024#1476024
+        echo "set man-db/auto-update false" | sudo debconf-communicate
+        sudo dpkg-reconfigure man-db
+
+        # Install ninja
+        source ./build/common/get-ninja.sh
+
+        # Install CMake
+        mkdir -p cmake
+        cd cmake
+
+        sudo wget https://github.com/Kitware/CMake/releases/download/v$GITHUB_CMAKE_VERSION/cmake-$GITHUB_CMAKE_VERSION-Linux-x86_64.sh
+        sudo chmod +x ./cmake-$GITHUB_CMAKE_VERSION-Linux-x86_64.sh
+        sudo ./cmake-$GITHUB_CMAKE_VERSION-Linux-x86_64.sh --skip-license > /dev/null
+        sudo update-alternatives --install /usr/bin/cmake cmake $(pwd)/bin/cmake 1000 --force
+
+        cd ..
+
+        sudo wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key | sudo apt-key add -
+        sudo apt-get update
+        sudo apt-get install clang-$GITHUB_CLANG_VERSION libc++-$GITHUB_CLANG_VERSION-dev libc++abi-$GITHUB_CLANG_VERSION-dev
+        sudo apt-get install mesa-common-dev libxi-dev libxxf86vm-dev
+
+        # For dawn
+        sudo apt-get install libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev libx11-xcb-dev
+
+        sudo update-alternatives --install /usr/bin/cc cc /usr/bin/clang-${GITHUB_CLANG_VERSION} 100
+        sudo update-alternatives --install /usr/bin/c++ c++ /usr/bin/clang++-${GITHUB_CLANG_VERSION} 100
+        set +xe

.github/actions/mac-prereq/action.yml

@@ -0,0 +1,23 @@
+name: 'Mac Preqrequisites'
+runs:
+  using: "composite"
+  steps:
+    - uses: ./.github/actions/dep-versions
+    - 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: Cache Brew
+      id: brew-cache
+      uses: actions/cache@v4 # Use a specific version
+      with:
+        path: $HOME/Library/Caches/Homebrew
+        key: ${{ runner.os }}-brew-20250424
+    - name: Install Mac Prerequisites
+      shell: bash
+      run: |
+        # Install ninja
+        source ./build/common/get-ninja.sh

.github/workflows/android-continuous.yml

@@ -16,6 +16,9 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
       - name: Run Android Continuous
         uses: ./.github/actions/android-continuous
         with:

.github/workflows/ios-continuous.yml

@@ -14,6 +14,9 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/mac-prereq
       - name: Run build script
         run: |
           cd build/ios && printf "y" | ./build.sh continuous

.github/workflows/linux-continuous.yml

@@ -14,6 +14,9 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
       - name: Run build script
         run: |
           cd build/linux && printf "y" | ./build.sh continuous

.github/workflows/mac-continuous.yml

@@ -14,6 +14,9 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/mac-prereq
       - name: Run build script
         run: |
           cd build/mac && printf "y" | ./build.sh continuous

.github/workflows/presubmit.yml

@@ -9,22 +9,32 @@ on:
     - main
 
 jobs:
-  build-desktop:
-    name: build-desktop
-    runs-on: ${{ matrix.os }}
-
-    strategy:
-      matrix:
-        os: [macos-14-xlarge, ubuntu-22.04-16core]
-
+  build-desktop-mac:
+    name: build-mac
+    runs-on: macos-14-xlarge
     steps:
       - uses: actions/checkout@v4.1.6
         with:
           fetch-depth: 0
+      - uses: ./.github/actions/mac-prereq
       - name: Run build script
         run: |
-          WORKFLOW_OS=`echo \`uname\` | sed "s/Darwin/mac/" | tr [:upper:] [:lower:]`
-          cd build/$WORKFLOW_OS && printf "y" | ./build.sh presubmit
+          cd build/mac && printf "y" | ./build.sh presubmit
+      - name: Test material parser
+        run: |
+          out/cmake-release/filament/test/test_material_parser
+
+  build-desktop-linux:
+    name: build-linux
+    runs-on: ubuntu-22.04-16core
+    steps:
+      - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
+      - name: Run build script
+        run: |
+          cd build/linux && printf "y" | ./build.sh presubmit
       - name: Test material parser
         run: |
           out/cmake-release/filament/test/test_material_parser
@@ -35,6 +45,8 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
       - name: Run build script
         run: |
           build\windows\build-github.bat presubmit
@@ -48,6 +60,7 @@ jobs:
       - uses: actions/checkout@v4.1.6
         with:
           fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
       - uses: actions/setup-java@v3
         with:
           distribution: 'temurin'
@@ -66,6 +79,7 @@ jobs:
       - uses: actions/checkout@v4.1.6
         with:
           fetch-depth: 0
+      - uses: ./.github/actions/mac-prereq
       - name: Run build script
         run: |
           cd build/ios && printf "y" | ./build.sh presubmit
@@ -79,6 +93,9 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
       - name: Run build script
         run: |
           cd build/web && printf "y" | ./build.sh presubmit
@@ -101,18 +118,20 @@ jobs:
     runs-on: macos-14-xlarge
     steps:
       - uses: actions/checkout@v4.1.6
-      - 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 python prereqs
-        run: pip install mako setuptools pyyaml
-      - name: Run script
-        run: |
-          bash test/renderdiff/test.sh
+          fetch-depth: 0
+      - uses: ./.github/actions/mac-prereq
+      - name: Cache Mesa and deps
+        id: mesa-cache
+        uses: actions/cache@v4 # Use a specific version
+        with:
+          path: mesa
+          key: ${{ runner.os }}-mesa-deps-2-${{ vars.MESA_VERSION }}
+      - name: Get Mesa
+        id: mesa-prereq
+        run: bash test/utils/get_mesa.sh
+      - name: Run Test
+        run: bash test/renderdiff/test.sh
       - uses: actions/upload-artifact@v4
         with:
           name: presubmit-renderdiff-result
@@ -121,11 +140,13 @@ jobs:
   validate-wgsl-webgpu:
     name: validate-wgsl-webgpu
     runs-on: 'ubuntu-24.04-8core'
-
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
       - name: Run build script
-        run: source ./build/linux/ci-common.sh && ./build.sh -W debug test_filamat filament
+        run: ./build.sh -W debug test_filamat filament
       - name: Run test
         run: ./out/cmake-debug/libs/filamat/test_filamat --gtest_filter=MaterialCompiler.Wgsl*
 
@@ -136,23 +157,16 @@ jobs:
       - 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
+      - uses: ./.github/actions/mac-prereq
+      - name: Install clang-tidy and deps
         run: |
           pip install pyyaml
-          brew install llvm
-          sudo ln -s "$(brew --prefix llvm)/bin/clang-tidy" "/usr/local/bin/clang-tidy"
+          brew install llvm@${GITHUB_LLVM_VERSION}
+          sudo ln -s "$(brew --prefix llvm)@${GITHUB_LLVM_VERSION}/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
+        run: bash test/code-correctness/test.sh

.github/workflows/release.yml

@@ -29,8 +29,42 @@ on:
     types: [created]
 
 jobs:
-  build-desktop:
-    name: build-desktop
+  build-linux:
+    name: build-linux
+    runs-on: ubuntu-22.04-32core
+    if: github.event_name == 'release' || github.event.inputs.platform == 'desktop'
+
+    steps:
+      - name: Decide Git ref
+        id: git_ref
+        run: |
+          REF=${RELEASE_TAG:-${GITHUB_REF}}
+          TAG=${REF##*/}
+          echo "ref=${REF}" >> $GITHUB_OUTPUT
+          echo "tag=${TAG}" >> $GITHUB_OUTPUT
+      - uses: actions/checkout@v4.1.6
+        with:
+          ref: ${{ steps.git_ref.outputs.ref }}
+      - uses: ./.github/actions/linux-prereq
+      - name: Run build script
+        env:
+          TAG: ${{ steps.git_ref.outputs.tag }}
+        run: |
+          cd build/linux && printf "y" | ./build.sh release
+          cd ../..
+          mv out/filament-release-linux.tgz out/filament-${TAG}-linux.tgz
+      - uses: actions/github-script@v6
+        env:
+          TAG: ${{ steps.git_ref.outputs.tag }}
+        with:
+          script: |
+            const upload = require('./build/common/upload-release-assets');
+            const { TAG } = process.env;
+            const globber = await glob.create('out/*.tgz');
+            await upload({ github, context }, await globber.glob(), TAG);
+
+  build-mac:
+    name: build-mac
     runs-on: ${{ matrix.os }}
     if: github.event_name == 'release' || github.event.inputs.platform == 'desktop'
 
@@ -49,15 +83,14 @@ jobs:
       - uses: actions/checkout@v4.1.6
         with:
           ref: ${{ steps.git_ref.outputs.ref }}
+      - uses: ./.github/actions/mac-prereq
       - name: Run build script
         env:
           TAG: ${{ steps.git_ref.outputs.tag }}
         run: |
-          WORKFLOW_OS=`echo \`uname\` | sed "s/Darwin/mac/" | tr [:upper:] [:lower:]`
-          cd build/$WORKFLOW_OS && printf "y" | ./build.sh release
+          cd build/mac && printf "y" | ./build.sh release
           cd ../..
-          if [ -f out/filament-release-darwin.tgz ]; then mv out/filament-release-darwin.tgz out/filament-${TAG}-mac.tgz; fi;
-          if [ -f out/filament-release-linux.tgz ]; then mv out/filament-release-linux.tgz out/filament-${TAG}-linux.tgz; fi;
+          mv out/filament-release-darwin.tgz out/filament-${TAG}-mac.tgz
       - uses: actions/github-script@v6
         env:
           TAG: ${{ steps.git_ref.outputs.tag }}
@@ -84,6 +117,7 @@ jobs:
       - uses: actions/checkout@v4.1.6
         with:
           ref: ${{ steps.git_ref.outputs.ref }}
+      - uses: ./.github/actions/linux-prereq
       - name: Run build script
         env:
           TAG: ${{ steps.git_ref.outputs.tag }}
@@ -121,6 +155,7 @@ jobs:
         with:
           distribution: 'temurin'
           java-version: '17'
+      - uses: ./.github/actions/linux-prereq
       - name: Run build script
         env:
           TAG: ${{ steps.git_ref.outputs.tag }}
@@ -171,6 +206,7 @@ jobs:
       - uses: actions/checkout@v4.1.6
         with:
           ref: ${{ steps.git_ref.outputs.ref }}
+      - uses: ./.github/actions/mac-prereq
       - name: Run build script
         env:
           TAG: ${{ steps.git_ref.outputs.tag }}

.github/workflows/web-continuous.yml

@@ -14,6 +14,9 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
       - name: Run build script
         run: |
           cd build/web && printf "y" | ./build.sh continuous

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/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.
 

build/android/build.sh

@@ -14,12 +14,12 @@ echo "Please refer to BUILDING.md for more information."
 read -r -p "Do you wish to proceed (y/n)? " choice
 case "${choice}" in
     y|Y)
-	      echo "Build will proceed..."
-	      ;;
+              echo "Build will proceed..."
+              ;;
     n|N)
-    	  exit 0
-    	  ;;
-	  *)
+          exit 0
+          ;;
+          *)
         exit 0
         ;;
 esac
@@ -30,30 +30,27 @@ set -x
 UNAME=`echo $(uname)`
 LC_UNAME=`echo $UNAME | tr '[:upper:]' '[:lower:]'`
 
-FILAMENT_ANDROID_CI_BUILD=true
-
 # build-common.sh will generate the following variables:
 #     $GENERATE_ARCHIVES
 #     $BUILD_DEBUG
 #     $BUILD_RELEASE
 source `dirname $0`/../common/ci-common.sh
-if [[  "$LC_UNAME" == "linux" ]]; then
-    source `dirname $0`/../linux/ci-common.sh
-elif [[ "$LC_UNAME" == "darwin" ]]; then
-    source `dirname $0`/../mac/ci-common.sh
-fi
 source `dirname $0`/../common/build-common.sh
 
 if [[ "$GITHUB_WORKFLOW" ]]; then
     java_version=$(java -version 2>&1 | head -1 | cut -d'"' -f2 | sed '/^1\./s///' | cut -d'.' -f1)
     if [[ "$java_version" < 17 ]]; then
         echo "Android builds require Java 17, found version ${java_version} instead"
-        exit 0
+        exit 1
     fi
 fi
 
 # Unless explicitly specified, NDK version will be set to match exactly the required one
-FILAMENT_NDK_VERSION=${FILAMENT_NDK_VERSION:-$(cat `dirname $0`/ndk.version)}
+FILAMENT_NDK_VERSION=${GITHUB_NDK_VERSION:-27.0.11718014}
+
+(! grep "${FILAMENT_NDK_VERSION}" `dirname $0`/../../android/build.gradle > /dev/null) &&
+    echo "Mismatch of NDK versions: want ${FILAMENT_NDK_VERSION} and not found in android/build.gradle" &&
+    exit 1
 
 # Install the required NDK version specifically (if not present)
 if [[ ! -d "${ANDROID_HOME}/ndk/$FILAMENT_NDK_VERSION" ]]; then

build/android/ndk.version

@@ -1 +0,0 @@
-27.0.11718014

build/common/check-headers.sh

@@ -56,10 +56,16 @@ popd >/dev/null
 rm -rf out/check-headers
 mkdir -p out/check-headers
 
+TMP_FILE=out/check-headers/temp.cpp
 echo "Checking that public headers compile independently..."
 for include in "${includes[@]}"; do
+    rm -f ${TMP_FILE}
     echo "Checking ${include}"
-    echo "#include <${include}>" >> out/check-headers/temp.cpp
-    clang -std=c++17 -I "${FILAMENT_HEADERS}" out/check-headers/temp.cpp -c -o /dev/null
+    if [[ "${include}" == "utils/Systrace.h" ]]; then
+        # A necessary define before we can include utils/Systrace.h
+        echo "#define SYSTRACE_TAG SYSTRACE_TAG_DISABLED" >> ${TMP_FILE}
+    fi
+    echo "#include <${include}>" >> ${TMP_FILE}
+    clang -std=c++17 -I "${FILAMENT_HEADERS}" ${TMP_FILE} -c -o /dev/null
 done
 echo "Done!"

build/common/ci-common.sh

@@ -2,5 +2,4 @@
 
 if [[ "$GITHUB_WORKFLOW" ]]; then
     echo "Running workflow $GITHUB_WORKFLOW (event: $GITHUB_EVENT_NAME, action: $GITHUB_ACTION)"
-    CONTINUOUS_INTEGRATION=true
 fi

build/common/get-ninja.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+if [[ "$GITHUB_WORKFLOW" ]]; then
+    OS_NAME=$(uname -s)
+    NINJA_DL_DIR=/tmp/ninja-dl
+    NINJA_PLATFORM=
+    if [[ "$OS_NAME" == "Linux" ]]; then
+        NINJA_PLATFORM=linux
+    elif [[ "$OS_NAME" == "Darwin" ]]; then
+        NINJA_PLATFORM=mac
+    fi
+    curl -L -o /tmp/ninja-dl.zip https://github.com/ninja-build/ninja/releases/download/v${GITHUB_NINJA_VERSION}/ninja-${NINJA_PLATFORM}.zip
+    mkdir -p $NINJA_DL_DIR
+    unzip -q /tmp/ninja-dl.zip -d $NINJA_DL_DIR
+    chmod +x ${NINJA_DL_DIR}/ninja
+    # Install ninja globally for AGP
+    sudo cp ${NINJA_DL_DIR}/ninja /usr/local/bin/
+fi

build/common/versions

@@ -0,0 +1,6 @@
+GITHUB_CLANG_VERSION=14
+GITHUB_CMAKE_VERSION=3.19.5
+GITHUB_NINJA_VERSION=1.10.2
+GITHUB_MESA_VERSION=24.2.1
+GITHUB_LLVM_VERSION=16
+GITHUB_NDK_VERSION=27.0.11718014

build/ios/build.sh

@@ -28,7 +28,6 @@ set -e
 set -x
 
 source `dirname $0`/../common/ci-common.sh
-source `dirname $0`/ci-common.sh
 source `dirname $0`/../common/build-common.sh
 
 pushd `dirname $0`/../.. > /dev/null
@@ -41,4 +40,3 @@ if [[ "${GENERATE_ARCHIVES}" ]]; then
 fi
 
 ./build.sh -i -p ios -c $BUILD_SIMULATOR $GENERATE_ARCHIVES $BUILD_DEBUG $BUILD_RELEASE
-

build/ios/ci-common.sh

@@ -1,6 +0,0 @@
-#!/bin/bash
-
-curl -OL https://github.com/ninja-build/ninja/releases/download/v1.10.2/ninja-mac.zip
-unzip -q ninja-mac.zip
-chmod +x ninja
-export PATH="$PWD:$PATH"

build/linux/build.sh

@@ -32,7 +32,6 @@ set -x
 #     $BUILD_DEBUG
 #     $BUILD_RELEASE
 source `dirname $0`/../common/ci-common.sh
-source `dirname $0`/ci-common.sh
 source `dirname $0`/../common/build-common.sh
 
 pushd `dirname $0`/../.. > /dev/null

build/linux/ci-common.sh

@@ -1,38 +0,0 @@
-#!/bin/bash
-
-# version of clang we want to use
-export GITHUB_CLANG_VERSION=14
-# version of CMake to use instead of the default one
-export GITHUB_CMAKE_VERSION=3.19.5
-# version of ninja to use
-export GITHUB_NINJA_VERSION=1.10.2
-
-# Steps for GitHub Workflows
-if [[ "$GITHUB_WORKFLOW" ]]; then
-    # Install ninja
-    wget -q https://github.com/ninja-build/ninja/releases/download/v$GITHUB_NINJA_VERSION/ninja-linux.zip
-    unzip -q ninja-linux.zip
-    export PATH="$PWD:$PATH"
-
-    # Install CMake
-    mkdir -p cmake
-    cd cmake
-
-    sudo wget https://github.com/Kitware/CMake/releases/download/v$GITHUB_CMAKE_VERSION/cmake-$GITHUB_CMAKE_VERSION-Linux-x86_64.sh
-    sudo chmod +x ./cmake-$GITHUB_CMAKE_VERSION-Linux-x86_64.sh
-    sudo ./cmake-$GITHUB_CMAKE_VERSION-Linux-x86_64.sh --skip-license > /dev/null
-    sudo update-alternatives --install /usr/bin/cmake cmake $(pwd)/bin/cmake 1000 --force
-
-    cd ..
-
-    sudo wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key | sudo apt-key add -
-    sudo apt-get update
-    sudo apt-get install clang-$GITHUB_CLANG_VERSION libc++-$GITHUB_CLANG_VERSION-dev libc++abi-$GITHUB_CLANG_VERSION-dev
-    sudo apt-get install mesa-common-dev libxi-dev libxxf86vm-dev
-
-    # For dawn
-    sudo apt-get install libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev libx11-xcb-dev
-
-    sudo update-alternatives --install /usr/bin/cc cc /usr/bin/clang-${GITHUB_CLANG_VERSION} 100
-    sudo update-alternatives --install /usr/bin/c++ c++ /usr/bin/clang++-${GITHUB_CLANG_VERSION} 100
-fi

build/mac/build.sh

@@ -28,7 +28,6 @@ set -e
 set -x
 
 source `dirname $0`/../common/ci-common.sh
-source `dirname $0`/ci-common.sh
 source `dirname $0`/../common/build-common.sh
 
 pushd `dirname $0`/../.. > /dev/null

build/mac/ci-common.sh

@@ -1,8 +0,0 @@
-#!/bin/bash
-
-curl -OL https://github.com/ninja-build/ninja/releases/download/v1.10.2/ninja-mac.zip
-unzip -q ninja-mac.zip
-chmod +x ninja
-# Install ninja globally for AGP
-cp ninja /usr/local/bin
-export PATH="$PWD:$PATH"

build/toolchain-aarch64-linux-android.cmake

@@ -31,10 +31,16 @@ set(DIST_ARCH arm64-v8a)
 string(TOLOWER ${CMAKE_HOST_SYSTEM_NAME} HOST_NAME_L)
 file(TO_CMAKE_PATH $ENV{ANDROID_HOME} ANDROID_HOME_UNIX)
 
+message(STATUS "Try using NDK \'${FILAMENT_NDK_VERSION}\'")
 if (NOT FILAMENT_NDK_VERSION)
-    file(READ "${CMAKE_CURRENT_LIST_DIR}/android/ndk.version" FILAMENT_NDK_VERSION)
-    string(REGEX MATCH "^\\d+" FILAMENT_NDK_VERSION ${FILAMENT_NDK_VERSION})
+    file(READ "${CMAKE_CURRENT_LIST_DIR}/common/versions" VERSIONS_STR)
+    string(REGEX MATCH "GITHUB_NDK_VERSION=(\\d+)" _UNUSED ${VERSIONS_STR})
+    if(CMAKE_MATCH_1)
+        set(FILAMENT_NDK_VERSION "${CMAKE_MATCH_1}")
+    endif()
 endif()
+message(STATUS "Using NDK \'${FILAMENT_NDK_VERSION}\'")
+
 file(GLOB NDK_VERSIONS LIST_DIRECTORIES true ${ANDROID_HOME_UNIX}/ndk/${FILAMENT_NDK_VERSION}*)
 list(SORT NDK_VERSIONS)
 list(GET NDK_VERSIONS -1 NDK_VERSION)

build/toolchain-arm7-linux-android.cmake

@@ -32,10 +32,16 @@ set(DIST_ARCH armeabi-v7a)
 string(TOLOWER ${CMAKE_HOST_SYSTEM_NAME} HOST_NAME_L)
 file(TO_CMAKE_PATH $ENV{ANDROID_HOME} ANDROID_HOME_UNIX)
 
+message(STATUS "Try using NDK \'${FILAMENT_NDK_VERSION}\'")
 if (NOT FILAMENT_NDK_VERSION)
-    file(READ "${CMAKE_CURRENT_LIST_DIR}/android/ndk.version" FILAMENT_NDK_VERSION)
-    string(REGEX MATCH "^\\d+" FILAMENT_NDK_VERSION ${FILAMENT_NDK_VERSION})
+    file(READ "${CMAKE_CURRENT_LIST_DIR}/common/versions" VERSIONS_STR)
+    string(REGEX MATCH "GITHUB_NDK_VERSION=(\\d+)" _UNUSED ${VERSIONS_STR})
+    if(CMAKE_MATCH_1)
+        set(FILAMENT_NDK_VERSION "${CMAKE_MATCH_1}")
+    endif()
 endif()
+message(STATUS "Using NDK \'${FILAMENT_NDK_VERSION}\'")
+
 file(GLOB NDK_VERSIONS LIST_DIRECTORIES true ${ANDROID_HOME_UNIX}/ndk/${FILAMENT_NDK_VERSION}*)
 list(SORT NDK_VERSIONS)
 list(GET NDK_VERSIONS -1 NDK_VERSION)

build/toolchain-x86-linux-android.cmake

@@ -31,10 +31,16 @@ set(DIST_ARCH x86)
 string(TOLOWER ${CMAKE_HOST_SYSTEM_NAME} HOST_NAME_L)
 file(TO_CMAKE_PATH $ENV{ANDROID_HOME} ANDROID_HOME_UNIX)
 
+message(STATUS "Try using NDK \'${FILAMENT_NDK_VERSION}\'")
 if (NOT FILAMENT_NDK_VERSION)
-    file(READ "${CMAKE_CURRENT_LIST_DIR}/android/ndk.version" FILAMENT_NDK_VERSION)
-    string(REGEX MATCH "^\\d+" FILAMENT_NDK_VERSION ${FILAMENT_NDK_VERSION})
+    file(READ "${CMAKE_CURRENT_LIST_DIR}/common/versions" VERSIONS_STR)
+    string(REGEX MATCH "GITHUB_NDK_VERSION=(\\d+)" _UNUSED ${VERSIONS_STR})
+    if(CMAKE_MATCH_1)
+        set(FILAMENT_NDK_VERSION "${CMAKE_MATCH_1}")
+    endif()
 endif()
+message(STATUS "Using NDK \'${FILAMENT_NDK_VERSION}\'")
+
 file(GLOB NDK_VERSIONS LIST_DIRECTORIES true ${ANDROID_HOME_UNIX}/ndk/${FILAMENT_NDK_VERSION}*)
 list(SORT NDK_VERSIONS)
 list(GET NDK_VERSIONS -1 NDK_VERSION)

build/toolchain-x86_64-linux-android.cmake

@@ -31,10 +31,16 @@ set(DIST_ARCH x86_64)
 string(TOLOWER ${CMAKE_HOST_SYSTEM_NAME} HOST_NAME_L)
 file(TO_CMAKE_PATH $ENV{ANDROID_HOME} ANDROID_HOME_UNIX)
 
+message(STATUS "Try using NDK \'${FILAMENT_NDK_VERSION}\'")
 if (NOT FILAMENT_NDK_VERSION)
-    file(READ "${CMAKE_CURRENT_LIST_DIR}/android/ndk.version" FILAMENT_NDK_VERSION)
-    string(REGEX MATCH "^\\d+" FILAMENT_NDK_VERSION ${FILAMENT_NDK_VERSION})
+    file(READ "${CMAKE_CURRENT_LIST_DIR}/common/versions" VERSIONS_STR)
+    string(REGEX MATCH "GITHUB_NDK_VERSION=(\\d+)" _UNUSED ${VERSIONS_STR})
+    if(CMAKE_MATCH_1)
+        set(FILAMENT_NDK_VERSION "${CMAKE_MATCH_1}")
+    endif()
 endif()
+message(STATUS "Using NDK \'${FILAMENT_NDK_VERSION}\'")
+
 file(GLOB NDK_VERSIONS LIST_DIRECTORIES true ${ANDROID_HOME_UNIX}/ndk/${FILAMENT_NDK_VERSION}*)
 list(SORT NDK_VERSIONS)
 list(GET NDK_VERSIONS -1 NDK_VERSION)

build/web/ci-common.sh

@@ -1,16 +1,4 @@
 #!/bin/bash
-if [ `uname` == "Linux" ];then
-    source `dirname $0`/../linux/ci-common.sh
-elif [ `uname` == "Darwin" ];then
-    curl -OL https://github.com/ninja-build/ninja/releases/download/v1.10.2/ninja-mac.zip
-    unzip -q ninja-mac.zip
-else
-    echo "Unsupported OS"
-    exit 1
-fi
-
-chmod +x ninja
-export PATH="$PWD:$PATH"
 
 # Install emscripten.
 curl -L https://github.com/emscripten-core/emsdk/archive/refs/tags/3.1.60.zip > emsdk.zip

filament/backend/CMakeLists.txt

@@ -257,8 +257,13 @@ if (FILAMENT_SUPPORTS_WEBGPU)
             src/webgpu/WebGPUConstants.h
             src/webgpu/WebGPUDriver.cpp
             src/webgpu/WebGPUDriver.h
+            src/webgpu/WebGPUHandles.cpp
+            src/webgpu/WebGPUHandles.h
+            src/webgpu/WebGPUPipelineCreation.cpp
+            src/webgpu/WebGPUPipelineCreation.h
             src/webgpu/WebGPUSwapChain.cpp
             src/webgpu/WebGPUSwapChain.h
+            src/webgpu/WGPUProgram.cpp
     )
     if (WIN32)
         list(APPEND SRCS src/webgpu/platform/WebGPUPlatformWindows.cpp)
@@ -507,8 +512,10 @@ 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
@@ -532,6 +539,9 @@ 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/Platform.h

@@ -149,13 +149,6 @@ public:
          *      - PlatformEGLAndroid
          */
         bool assertNativeWindowIsValid = false;
-
-        /**
-         * The action to take if a Drawable cannot be acquired. If true, the
-         * frame is aborted instead of panic. This is only supported for:
-         *      - PlatformMetal
-         */
-        bool metalDisablePanicOnDrawableFailure = false;
     };
 
     Platform() noexcept;

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

@@ -382,9 +382,26 @@ public:
         return {};
     }
 
-    using ImageData = std::pair<VkImage, VkDeviceMemory>;
+    struct ImageData {
+        struct Bundle {
+            VkImage image = VK_NULL_HANDLE;
+            VkDeviceMemory memory = VK_NULL_HANDLE;
+
+            inline bool valid() const noexcept {
+                return image != VK_NULL_HANDLE;
+            }
+        };
+        // It's possible for the external image to also have a known VK format. We need to create an
+        // image for that in case we are not looking to use an external "sampler" with this image.
+        Bundle internal;
+
+        // If we get a externalFormat in the metadata, then we should create an image with
+        // VK_FORMAT_UNDEFINED
+        Bundle external;
+    };
+
     virtual ImageData createVkImageFromExternal(ExternalImageHandleRef image) const {
-        return { VK_NULL_HANDLE, VK_NULL_HANDLE };
+        return {};
     }
 
 protected:

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/metal/MetalDriver.mm

@@ -1408,8 +1408,8 @@ void MetalDriver::setRenderPrimitiveBuffer(Handle<HwRenderPrimitive> rph, Primit
     auto primitive = handle_cast<MetalRenderPrimitive>(rph);
     auto vertexBuffer = handle_cast<MetalVertexBuffer>(vbh);
     auto indexBuffer = handle_cast<MetalIndexBuffer>(ibh);
-    MetalVertexBufferInfo const* const vbi = handle_cast<MetalVertexBufferInfo>(vertexBuffer->vbih);
-    primitive->setBuffers(vbi, vertexBuffer, indexBuffer);
+    primitive->vertexBuffer = vertexBuffer;
+    primitive->indexBuffer = indexBuffer;
     primitive->type = pt;
 }
 

filament/backend/src/metal/MetalHandles.h

@@ -194,12 +194,8 @@ struct MetalIndexBuffer : public HwIndexBuffer {
 };
 
 struct MetalRenderPrimitive : public HwRenderPrimitive {
-    MetalRenderPrimitive();
-    void setBuffers(MetalVertexBufferInfo const* const vbi,
-            MetalVertexBuffer* vertexBuffer, MetalIndexBuffer* indexBuffer);
     // The pointers to MetalVertexBuffer and MetalIndexBuffer are "weak".
     // The MetalVertexBuffer and MetalIndexBuffer must outlive the MetalRenderPrimitive.
-
     MetalVertexBuffer* vertexBuffer = nullptr;
     MetalIndexBuffer* indexBuffer = nullptr;
 };
@@ -380,7 +376,6 @@ public:
 
     math::uint2 getAttachmentSize() noexcept;
 
-    bool isDefaultRenderTarget() const { return defaultRenderTarget; }
     uint8_t getSamples() const { return samples; }
 
     Attachment getDrawColorAttachment(size_t index);

filament/backend/src/metal/MetalHandles.mm

@@ -536,15 +536,6 @@ MetalIndexBuffer::MetalIndexBuffer(MetalContext& context, BufferUsage usage, uin
         uint32_t indexCount) : HwIndexBuffer(elementSize, indexCount),
         buffer(context, BufferObjectBinding::VERTEX, usage, elementSize * indexCount, true) { }
 
-MetalRenderPrimitive::MetalRenderPrimitive() {
-}
-
-void MetalRenderPrimitive::setBuffers(MetalVertexBufferInfo const* const vbi,
-        MetalVertexBuffer* vertexBuffer, MetalIndexBuffer* indexBuffer) {
-    this->vertexBuffer = vertexBuffer;
-    this->indexBuffer = indexBuffer;
-}
-
 MetalProgram::MetalProgram(MetalContext& context, Program&& program) noexcept
     : HwProgram(program.getName()), mContext(context) {
     mToken = context.shaderCompiler->createProgram(program.getName(), std::move(program));

filament/backend/src/metal/PlatformMetal.mm

@@ -45,9 +45,6 @@ PlatformMetal::~PlatformMetal() noexcept {
 }
 
 Driver* PlatformMetal::createDriver(void* /*sharedContext*/, const Platform::DriverConfig& driverConfig) noexcept {
-    pImpl->mDrawableFailureBehavior = driverConfig.metalDisablePanicOnDrawableFailure
-            ? DrawableFailureBehavior::ABORT_FRAME
-            : DrawableFailureBehavior::PANIC;
     return MetalDriverFactory::create(this, driverConfig);
 }
 

filament/backend/src/opengl/GLUtils.cpp

@@ -16,9 +16,15 @@
 
 #include "GLUtils.h"
 
+#include "private/backend/Driver.h"
+
+#include <utils/compiler.h>
+#include <utils/ostream.h>
 #include <utils/trap.h>
 
-#include "private/backend/Driver.h"
+#include <string_view>
+
+#include <stddef.h>
 
 namespace filament::backend {
 
@@ -28,38 +34,31 @@ using namespace utils;
 namespace GLUtils {
 
 UTILS_NOINLINE
-const char* getGLError(GLenum error) noexcept {
-    const char* string = "unknown";
+std::string_view getGLErrorString(GLenum error) noexcept {
     switch (error) {
         case GL_NO_ERROR:
-            string = "GL_NO_ERROR";
-            break;
+            return "GL_NO_ERROR";
         case GL_INVALID_ENUM:
-            string = "GL_INVALID_ENUM";
-            break;
+            return "GL_INVALID_ENUM";
         case GL_INVALID_VALUE:
-            string = "GL_INVALID_VALUE";
-            break;
+            return "GL_INVALID_VALUE";
         case GL_INVALID_OPERATION:
-            string = "GL_INVALID_OPERATION";
-            break;
+            return "GL_INVALID_OPERATION";
         case GL_INVALID_FRAMEBUFFER_OPERATION:
-            string = "GL_INVALID_FRAMEBUFFER_OPERATION";
-            break;
+            return "GL_INVALID_FRAMEBUFFER_OPERATION";
         case GL_OUT_OF_MEMORY:
-            string = "GL_OUT_OF_MEMORY";
-            break;
+            return "GL_OUT_OF_MEMORY";
         default:
             break;
     }
-    return string;
+    return "unknown";
 }
 
 UTILS_NOINLINE
 GLenum checkGLError(io::ostream& out, const char* function, size_t line) noexcept {
     GLenum const error = glGetError();
-    if (error != GL_NO_ERROR) {
-        const char* string = getGLError(error);
+    if (UTILS_VERY_UNLIKELY(error != GL_NO_ERROR)) {
+        auto const string = getGLErrorString(error);
         out << "OpenGL error " << io::hex << error << " (" << string << ") in \""
             << function << "\" at line " << io::dec << line << io::endl;
     }
@@ -69,46 +68,39 @@ GLenum checkGLError(io::ostream& out, const char* function, size_t line) noexcep
 UTILS_NOINLINE
 void assertGLError(io::ostream& out, const char* function, size_t line) noexcept {
     GLenum const err = checkGLError(out, function, line);
-    if (err != GL_NO_ERROR) {
+    if (UTILS_VERY_UNLIKELY(err != GL_NO_ERROR)) {
         debug_trap();
     }
 }
 
 UTILS_NOINLINE
-const char* getFramebufferStatus(GLenum status) noexcept {
-    const char* string = "unknown";
+std::string_view getFramebufferStatusString(GLenum status) noexcept {
     switch (status) {
         case GL_FRAMEBUFFER_COMPLETE:
-            string = "GL_FRAMEBUFFER_COMPLETE";
-            break;
+            return "GL_FRAMEBUFFER_COMPLETE";
         case GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT:
-            string = "GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT";
-            break;
+            return "GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT";
         case GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:
-            string = "GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";
-            break;
+            return "GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";
         case GL_FRAMEBUFFER_UNSUPPORTED:
-            string = "GL_FRAMEBUFFER_UNSUPPORTED";
-            break;
+            return "GL_FRAMEBUFFER_UNSUPPORTED";
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
         case GL_FRAMEBUFFER_UNDEFINED:
-            string = "GL_FRAMEBUFFER_UNDEFINED";
-            break;
+            return "GL_FRAMEBUFFER_UNDEFINED";
         case GL_FRAMEBUFFER_INCOMPLETE_MULTISAMPLE:
-            string = "GL_FRAMEBUFFER_INCOMPLETE_MULTISAMPLE";
-            break;
+            return "GL_FRAMEBUFFER_INCOMPLETE_MULTISAMPLE";
 #endif
         default:
             break;
     }
-    return string;
+    return "unknown";
 }
 
 UTILS_NOINLINE
 GLenum checkFramebufferStatus(io::ostream& out, GLenum target, const char* function, size_t line) noexcept {
     GLenum const status = glCheckFramebufferStatus(target);
-    if (status != GL_FRAMEBUFFER_COMPLETE) {
-        const char* string = getFramebufferStatus(status);
+    if (UTILS_VERY_UNLIKELY(status != GL_FRAMEBUFFER_COMPLETE)) {
+        auto const string = getFramebufferStatusString(status);
         out << "OpenGL framebuffer error " << io::hex << status << " (" << string << ") in \""
             << function << "\" at line " << io::dec << line << io::endl;
     }
@@ -118,7 +110,7 @@ GLenum checkFramebufferStatus(io::ostream& out, GLenum target, const char* funct
 UTILS_NOINLINE
 void assertFramebufferStatus(io::ostream& out, GLenum target, const char* function, size_t line) noexcept {
     GLenum const status = checkFramebufferStatus(out, target, function, line);
-    if (status != GL_FRAMEBUFFER_COMPLETE) {
+    if (UTILS_VERY_UNLIKELY(status != GL_FRAMEBUFFER_COMPLETE)) {
         debug_trap();
     }
 }

filament/backend/src/opengl/GLUtils.h

@@ -17,29 +17,26 @@
 #ifndef TNT_FILAMENT_BACKEND_OPENGL_GLUTILS_H
 #define TNT_FILAMENT_BACKEND_OPENGL_GLUTILS_H
 
-#include <utils/compiler.h>
 #include <utils/debug.h>
-#include <utils/Log.h>
+#include <utils/ostream.h>
 
 #include <backend/DriverEnums.h>
 
 #include <string_view>
 #include <unordered_set>
 
-#include <stdint.h>
 #include <stddef.h>
-#include <string.h>
+#include <stdint.h>
 
 #include "gl_headers.h"
 
-namespace filament::backend {
-namespace GLUtils {
+namespace filament::backend::GLUtils {
 
-const char* getGLError(GLenum error) noexcept;
+std::string_view getGLErrorString(GLenum error) noexcept;
 GLenum checkGLError(utils::io::ostream& out, const char* function, size_t line) noexcept;
 void assertGLError(utils::io::ostream& out, const char* function, size_t line) noexcept;
 
-const char* getFramebufferStatus(GLenum err) noexcept;
+std::string_view getFramebufferStatusString(GLenum err) noexcept;
 GLenum checkFramebufferStatus(utils::io::ostream& out, GLenum target, const char* function, size_t line) noexcept;
 void assertFramebufferStatus(utils::io::ostream& out, GLenum target, const char* function, size_t line) noexcept;
 
@@ -53,7 +50,7 @@ void assertFramebufferStatus(utils::io::ostream& out, GLenum target, const char*
 #   define CHECK_GL_FRAMEBUFFER_STATUS(out, target) { GLUtils::checkFramebufferStatus(out, target, __func__, __LINE__); }
 #endif
 
-constexpr GLuint getComponentCount(ElementType type) noexcept {
+constexpr GLuint getComponentCount(ElementType const type) noexcept {
     using ElementType = ElementType;
     switch (type) {
         case ElementType::BYTE:
@@ -87,27 +84,29 @@ constexpr GLuint getComponentCount(ElementType type) noexcept {
         case ElementType::USHORT4:
             return 4;
     }
+    // should never happen
+    return 1;
 }
 
 // ------------------------------------------------------------------------------------------------
 // Our enums to GLenum conversions
 // ------------------------------------------------------------------------------------------------
 
-constexpr GLbitfield getAttachmentBitfield(TargetBufferFlags flags) noexcept {
+constexpr GLbitfield getAttachmentBitfield(TargetBufferFlags const flags) noexcept {
     GLbitfield mask = 0;
     if (any(flags & TargetBufferFlags::COLOR_ALL)) {
-        mask |= (GLbitfield)GL_COLOR_BUFFER_BIT;
+        mask |= GLbitfield(GL_COLOR_BUFFER_BIT);
     }
     if (any(flags & TargetBufferFlags::DEPTH)) {
-        mask |= (GLbitfield)GL_DEPTH_BUFFER_BIT;
+        mask |= GLbitfield(GL_DEPTH_BUFFER_BIT);
     }
     if (any(flags & TargetBufferFlags::STENCIL)) {
-        mask |= (GLbitfield)GL_STENCIL_BUFFER_BIT;
+        mask |= GLbitfield(GL_STENCIL_BUFFER_BIT);
     }
     return mask;
 }
 
-constexpr GLenum getBufferUsage(BufferUsage usage) noexcept {
+constexpr GLenum getBufferUsage(BufferUsage const usage) noexcept {
     switch (usage) {
         case BufferUsage::STATIC:
             return GL_STATIC_DRAW;
@@ -116,7 +115,7 @@ constexpr GLenum getBufferUsage(BufferUsage usage) noexcept {
     }
 }
 
-constexpr GLenum getBufferBindingType(BufferObjectBinding bindingType) noexcept {
+constexpr GLenum getBufferBindingType(BufferObjectBinding const bindingType) noexcept {
     switch (bindingType) {
         case BufferObjectBinding::VERTEX:
             return GL_ARRAY_BUFFER;
@@ -135,13 +134,15 @@ constexpr GLenum getBufferBindingType(BufferObjectBinding bindingType) noexcept
             return 0x90D2; // just to return something
 #endif
     }
+    // should never happen
+    return GL_ARRAY_BUFFER;
 }
 
-constexpr GLboolean getNormalization(bool normalized) noexcept {
+constexpr GLboolean getNormalization(bool const normalized) noexcept {
     return GLboolean(normalized ? GL_TRUE : GL_FALSE);
 }
 
-constexpr GLenum getComponentType(ElementType type) noexcept {
+constexpr GLenum getComponentType(ElementType const type) noexcept {
     using ElementType = ElementType;
     switch (type) {
         case ElementType::BYTE:
@@ -184,9 +185,11 @@ constexpr GLenum getComponentType(ElementType type) noexcept {
             return GL_HALF_FLOAT_OES;
 #endif
     }
+    // should never happen
+    return GL_INT;
 }
 
-constexpr GLenum getTextureTargetNotExternal(SamplerType target) noexcept {
+constexpr GLenum getTextureTargetNotExternal(SamplerType const target) noexcept {
     switch (target) {
         case SamplerType::SAMPLER_2D:
             return GL_TEXTURE_2D;
@@ -202,14 +205,16 @@ constexpr GLenum getTextureTargetNotExternal(SamplerType target) noexcept {
             // we should never be here
             return GL_TEXTURE_2D;
     }
+    // should never happen
+    return GL_TEXTURE_2D;
 }
 
-constexpr GLenum getCubemapTarget(uint16_t layer) noexcept {
+constexpr GLenum getCubemapTarget(uint16_t const layer) noexcept {
     assert_invariant(layer <= 5);
     return GL_TEXTURE_CUBE_MAP_POSITIVE_X + layer;
 }
 
-constexpr GLenum getWrapMode(SamplerWrapMode mode) noexcept {
+constexpr GLenum getWrapMode(SamplerWrapMode const mode) noexcept {
     using SamplerWrapMode = SamplerWrapMode;
     switch (mode) {
         case SamplerWrapMode::REPEAT:
@@ -219,6 +224,8 @@ constexpr GLenum getWrapMode(SamplerWrapMode mode) noexcept {
         case SamplerWrapMode::MIRRORED_REPEAT:
             return GL_MIRRORED_REPEAT;
     }
+    // should never happen
+    return GL_CLAMP_TO_EDGE;
 }
 
 constexpr GLenum getTextureFilter(SamplerMinFilter filter) noexcept {
@@ -234,6 +241,8 @@ constexpr GLenum getTextureFilter(SamplerMinFilter filter) noexcept {
             return GL_NEAREST_MIPMAP_NEAREST
                    - GLenum(SamplerMinFilter::NEAREST_MIPMAP_NEAREST) + GLenum(filter);
     }
+    // should never happen
+    return GL_NEAREST;
 }
 
 constexpr GLenum getTextureFilter(SamplerMagFilter filter) noexcept {
@@ -241,7 +250,7 @@ constexpr GLenum getTextureFilter(SamplerMagFilter filter) noexcept {
 }
 
 
-constexpr GLenum getBlendEquationMode(BlendEquation mode) noexcept {
+constexpr GLenum getBlendEquationMode(BlendEquation const mode) noexcept {
     using BlendEquation = BlendEquation;
     switch (mode) {
         case BlendEquation::ADD:               return GL_FUNC_ADD;
@@ -250,9 +259,11 @@ constexpr GLenum getBlendEquationMode(BlendEquation mode) noexcept {
         case BlendEquation::MIN:               return GL_MIN;
         case BlendEquation::MAX:               return GL_MAX;
     }
+    // should never happen
+    return GL_FUNC_ADD;
 }
 
-constexpr GLenum getBlendFunctionMode(BlendFunction mode) noexcept {
+constexpr GLenum getBlendFunctionMode(BlendFunction const mode) noexcept {
     using BlendFunction = BlendFunction;
     switch (mode) {
         case BlendFunction::ZERO:                  return GL_ZERO;
@@ -267,9 +278,11 @@ constexpr GLenum getBlendFunctionMode(BlendFunction mode) noexcept {
         case BlendFunction::ONE_MINUS_DST_ALPHA:   return GL_ONE_MINUS_DST_ALPHA;
         case BlendFunction::SRC_ALPHA_SATURATE:    return GL_SRC_ALPHA_SATURATE;
     }
+    // should never happen
+    return GL_ONE;
 }
 
-constexpr GLenum getCompareFunc(SamplerCompareFunc func) noexcept {
+constexpr GLenum getCompareFunc(SamplerCompareFunc const func) noexcept {
     switch (func) {
         case SamplerCompareFunc::LE:    return GL_LEQUAL;
         case SamplerCompareFunc::GE:    return GL_GEQUAL;
@@ -280,28 +293,30 @@ constexpr GLenum getCompareFunc(SamplerCompareFunc func) noexcept {
         case SamplerCompareFunc::A:     return GL_ALWAYS;
         case SamplerCompareFunc::N:     return GL_NEVER;
     }
+    // should never happen
+    return GL_LEQUAL;
 }
 
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
-constexpr GLenum getTextureCompareMode(SamplerCompareMode mode) noexcept {
+constexpr GLenum getTextureCompareMode(SamplerCompareMode const mode) noexcept {
     return mode == SamplerCompareMode::NONE ?
            GL_NONE : GL_COMPARE_REF_TO_TEXTURE;
 }
 
-constexpr GLenum getTextureCompareFunc(SamplerCompareFunc func) noexcept {
+constexpr GLenum getTextureCompareFunc(SamplerCompareFunc const func) noexcept {
     return getCompareFunc(func);
 }
 #endif
 
-constexpr GLenum getDepthFunc(SamplerCompareFunc func) noexcept {
+constexpr GLenum getDepthFunc(SamplerCompareFunc const func) noexcept {
     return getCompareFunc(func);
 }
 
-constexpr GLenum getStencilFunc(SamplerCompareFunc func) noexcept {
+constexpr GLenum getStencilFunc(SamplerCompareFunc const func) noexcept {
     return getCompareFunc(func);
 }
 
-constexpr GLenum getStencilOp(StencilOperation op) noexcept {
+constexpr GLenum getStencilOp(StencilOperation const op) noexcept {
     switch (op) {
         case StencilOperation::KEEP:        return GL_KEEP;
         case StencilOperation::ZERO:        return GL_ZERO;
@@ -312,9 +327,11 @@ constexpr GLenum getStencilOp(StencilOperation op) noexcept {
         case StencilOperation::DECR_WRAP:   return GL_DECR_WRAP;
         case StencilOperation::INVERT:      return GL_INVERT;
     }
+    // should never happen
+    return GL_KEEP;
 }
 
-constexpr GLenum getFormat(PixelDataFormat format) noexcept {
+constexpr GLenum getFormat(PixelDataFormat const format) noexcept {
     using PixelDataFormat = PixelDataFormat;
     switch (format) {
         case PixelDataFormat::RGB:              return GL_RGB;
@@ -336,9 +353,11 @@ constexpr GLenum getFormat(PixelDataFormat format) noexcept {
         default: return GL_NONE;
 #endif
     }
+    // should never happen
+    return GL_RGBA;
 }
 
-constexpr GLenum getType(PixelDataType type) noexcept {
+constexpr GLenum getType(PixelDataType const type) noexcept {
     using PixelDataType = PixelDataType;
     switch (type) {
         case PixelDataType::UBYTE:                return GL_UNSIGNED_BYTE;
@@ -360,10 +379,12 @@ constexpr GLenum getType(PixelDataType type) noexcept {
         default: return GL_NONE;
 #endif
     }
+    // should never happen
+    return GL_UNSIGNED_INT;
 }
 
 #if !defined(__EMSCRIPTEN__)  && !defined(FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2)
-constexpr GLenum getSwizzleChannel(TextureSwizzle c) noexcept {
+constexpr GLenum getSwizzleChannel(TextureSwizzle const c) noexcept {
     using TextureSwizzle = TextureSwizzle;
     switch (c) {
         case TextureSwizzle::SUBSTITUTE_ZERO:   return GL_ZERO;
@@ -373,10 +394,12 @@ constexpr GLenum getSwizzleChannel(TextureSwizzle c) noexcept {
         case TextureSwizzle::CHANNEL_2:         return GL_BLUE;
         case TextureSwizzle::CHANNEL_3:         return GL_ALPHA;
     }
+    // should never happen
+    return GL_RED;
 }
 #endif
 
-constexpr GLenum getCullingMode(CullingMode mode) noexcept {
+constexpr GLenum getCullingMode(CullingMode const mode) noexcept {
     switch (mode) {
         case CullingMode::NONE:
             // should never happen
@@ -388,11 +411,13 @@ constexpr GLenum getCullingMode(CullingMode mode) noexcept {
         case CullingMode::FRONT_AND_BACK:
             return GL_FRONT_AND_BACK;
     }
+    // should never happen
+    return GL_FRONT_AND_BACK;
 }
 
 // ES2 supported internal formats for texturing and how they  map to a format/type
 constexpr std::pair<GLenum, GLenum> textureFormatToFormatAndType(
-        TextureFormat format) noexcept {
+        TextureFormat const format) noexcept {
     switch (format) {
         case TextureFormat::R8:         return { 0x1909 /*GL_LUMINANCE*/, GL_UNSIGNED_BYTE };
         case TextureFormat::RGB8:       return { GL_RGB,                  GL_UNSIGNED_BYTE };
@@ -413,7 +438,7 @@ constexpr std::pair<GLenum, GLenum> textureFormatToFormatAndType(
 // clang loses it on this one, and generates a huge jump table when
 // inlined. So we don't  mark it as inline (only constexpr) which solves the problem,
 // strangely, when not inlined, clang simply generates an array lookup.
-constexpr /* inline */ GLenum getInternalFormat(TextureFormat format) noexcept {
+constexpr /* inline */ GLenum getInternalFormat(TextureFormat const format) noexcept {
     switch (format) {
 
         /* Formats supported by our ES2 implementations */
@@ -661,7 +686,7 @@ public:
 
 unordered_string_set split(const char* extensions) noexcept;
 
-} // namespace GLUtils
-} // namespace filament::backend
+} // namespace filament::backend::GLUtils
+
 
 #endif // TNT_FILAMENT_BACKEND_OPENGL_GLUTILS_H

filament/backend/src/opengl/OpenGLDriver.cpp

@@ -2098,6 +2098,7 @@ void OpenGLDriver::setAcquiredImage(Handle<HwStream> sh, void* hwbuffer, const m
 
     glstream->user_thread.pending = mPlatform.transformAcquiredImage({
             hwbuffer, cb, userData, handler });
+    glstream->user_thread.transform = transform;
 
     if (glstream->user_thread.pending.image != nullptr) {
         // If there's no pending image, do nothing. Note that GL_OES_EGL_image does not let you pass

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/opengl/platforms/CocoaExternalImage.mm

@@ -17,9 +17,11 @@
 #define COREVIDEO_SILENCE_GL_DEPRECATION
 
 #include "CocoaExternalImage.h"
-#include <utils/Panic.h>
 #include "../GLUtils.h"
 
+#include <utils/Panic.h>
+#include <utils/Log.h>
+
 namespace filament::backend {
 
 static const char *s_vertex = R"SHADER(#version 410 core

filament/backend/src/opengl/platforms/CocoaTouchExternalImage.mm

@@ -28,6 +28,7 @@
 #include <utils/compiler.h>
 #include <utils/Panic.h>
 #include <utils/debug.h>
+#include <utils/Log.h>
 
 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/VulkanCommands.cpp

@@ -295,6 +295,8 @@ VulkanCommandBuffer& CommandBufferPool::getRecording() {
 }
 
 void CommandBufferPool::gc() {
+    FVK_SYSTRACE_CONTEXT();
+    FVK_SYSTRACE_START("CommandBufferPool::gc");
     ActiveBuffers reclaimed;
     mSubmitted.forEachSetBit([this,&reclaimed] (size_t index) {
         auto& buffer = mBuffers[index];
@@ -304,6 +306,7 @@ void CommandBufferPool::gc() {
         }
     });
     mSubmitted &= ~reclaimed;
+    FVK_SYSTRACE_END();
 }
 
 void CommandBufferPool::update() {
@@ -333,7 +336,9 @@ void CommandBufferPool::wait() {
     mSubmitted.forEachSetBit([this, &count, &fences] (size_t index) {
         fences[count++] = mBuffers[index]->getVkFence();
     });
-    vkWaitForFences(mDevice, count, fences, VK_TRUE, UINT64_MAX);
+    if (count) {
+        vkWaitForFences(mDevice, count, fences, VK_TRUE, UINT64_MAX);
+    }
     update();
 }
 

filament/backend/src/vulkan/VulkanConstants.h

@@ -93,7 +93,7 @@
 #endif
 
 #ifndef NDEBUG
-#define FVK_DEBUG_FLAGS (FVK_DEBUG_PERFORMANCE | FVK_DEBUG_FORWARDED_FLAG | FVK_DEBUG_VALIDATION)
+#define FVK_DEBUG_FLAGS (FVK_DEBUG_PERFORMANCE | FVK_DEBUG_FORWARDED_FLAG)
 #else
 #define FVK_DEBUG_FLAGS 0
 #endif
@@ -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/VulkanDescriptorSetCache.cpp

@@ -31,8 +31,8 @@ namespace filament::backend {
 
 namespace {
 
-using DescriptorCount = VulkanDescriptorSetLayout::Count;
 using DescriptorSetLayoutArray = VulkanDescriptorSetCache::DescriptorSetLayoutArray;
+using DescriptorCount = VulkanDescriptorSetCache::DescriptorCount;
 
 // We create a pool for each layout as defined by the number of descriptors of each type. For
 // example, a layout of
@@ -203,11 +203,10 @@ public:
     DescriptorInfinitePool(VkDevice device)
         : mDevice(device) {}
 
-    VkDescriptorSet obtainSet(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-        auto const vklayout = layout->getVkLayout();
+    VkDescriptorSet obtainSet(DescriptorCount const& count, VkDescriptorSetLayout vklayout) {
         DescriptorPool* sameTypePool = nullptr;
         for (auto& pool: mPools) {
-            if (!pool->canAllocate(layout->count)) {
+            if (!pool->canAllocate(count)) {
                 continue;
             }
             if (auto set = pool->obtainSet(vklayout); set != VK_NULL_HANDLE) {
@@ -225,8 +224,7 @@ public:
         }
 
         // We need to increase the set of pools by one.
-        mPools.push_back(std::make_unique<DescriptorPool>(mDevice,
-                DescriptorCount::fromLayoutBitmask(layout->bitmask), capacity));
+        mPools.push_back(std::make_unique<DescriptorPool>(mDevice, count, capacity));
         auto& pool = mPools.back();
         auto ret = pool->obtainSet(vklayout);
         assert_invariant(ret != VK_NULL_HANDLE && "failed to obtain a set?");
@@ -276,39 +274,36 @@ void VulkanDescriptorSetCache::unbind(uint8_t setIndex) {
 }
 
 void VulkanDescriptorSetCache::commit(VulkanCommandBuffer* commands,
-        VkPipelineLayout pipelineLayout, fvkutils::DescriptorSetMask const& setMask) {
+        VkPipelineLayout pipelineLayout, fvkutils::DescriptorSetMask const& useExternalSamplers,
+        fvkutils::DescriptorSetMask const& setMask) {
     // setMask indicates the set of descriptor sets the driver wants to bind, curMask is the
     // actual set of sets that *needs* to be bound.
     fvkutils::DescriptorSetMask curMask = setMask;
 
-    auto& updateSets = mStashedSets;
-    bool const pipelineLayoutIsSame = mLastBoundInfo.pipelineLayout == pipelineLayout;
-
-    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;
+    auto const& updateSets = mStashedSets;
+    curMask.forEachSetBit([&](size_t index) {
+        if (!updateSets[index]) {
+            curMask.unset(index);
         }
-    } else {
-        setMask.forEachSetBit([&](size_t index) {
-            if (!updateSets[index]) {
+    });
+
+    if (mLastBoundInfo.pipelineLayout == pipelineLayout) {
+        auto& lastBoundSets = mLastBoundInfo.boundSets;
+        curMask.forEachSetBit([&](size_t index) {
+            if (updateSets[index] == lastBoundSets[index] && !useExternalSamplers[index]) {
                 curMask.unset(index);
             }
         });
     }
 
-    curMask.forEachSetBit([&updateSets, commands, pipelineLayout](size_t index) {
+    curMask.forEachSetBit([&](size_t index) {
         // This code actually binds the descriptor sets.
         auto set = updateSets[index];
         VkCommandBuffer const cmdbuffer = commands->buffer();
+        VkDescriptorSet vkset = useExternalSamplers[index] ? set->getExternalSamplerVkSet() :
+            set->getVkSet();
         vkCmdBindDescriptorSets(cmdbuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, index,
-                1, &set->getVkSet(), set->uniqueDynamicUboCount, set->getOffsets()->data());
+                1, &vkset, set->uniqueDynamicUboCount, set->getOffsets()->data());
         commands->acquire(set);
     });
 
@@ -334,7 +329,7 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
     if (set->dynamicUboMask.test(binding)) {
         type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC;
     }
-    VkWriteDescriptorSet const descriptorWrite = {
+    VkWriteDescriptorSet descriptorWrite = {
         .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
         .dstSet = set->getVkSet(),
         .dstBinding = binding,
@@ -343,12 +338,17 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
         .pBufferInfo = &info,
     };
     vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
+
+    if (auto externalSamplerSet = set->getExternalSamplerVkSet();
+            externalSamplerSet != VK_NULL_HANDLE) {
+        descriptorWrite.dstSet = externalSamplerSet;
+        vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
+    }
     set->acquire(bufferObject);
 }
 
-void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
-        VkSampler sampler) noexcept {
+void VulkanDescriptorSetCache::updateSamplerImpl(VkDescriptorSet vkset, uint8_t binding,
+        fvkmemory::resource_ptr<VulkanTexture> texture, VkSampler sampler) noexcept {
     VkImageSubresourceRange range = texture->getPrimaryViewRange();
     VkImageViewType const expectedType = texture->getViewType();
     if (any(texture->usage & TextureUsage::DEPTH_ATTACHMENT) &&
@@ -364,16 +364,29 @@ void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescr
         .imageLayout = fvkutils::getVkLayout(texture->getDefaultLayout()),
     };
 
-    VkWriteDescriptorSet const descriptorWrite = {
+    VkWriteDescriptorSet descriptorWrite = {
         .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
         .pNext = nullptr,
-        .dstSet = set->getVkSet(),
+        .dstSet = vkset,
         .dstBinding = binding,
         .descriptorCount = 1,
         .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
         .pImageInfo = &info,
     };
     vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
+}
+
+void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+        uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
+        VkSampler sampler) noexcept {
+    updateSamplerImpl(set->getVkSet(), binding, texture, sampler);
+    set->acquire(texture);
+}
+
+void VulkanDescriptorSetCache::updateSamplerForExternalSamplerSet(
+        fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t binding,
+        fvkmemory::resource_ptr<VulkanTexture> texture) noexcept {
+    updateSamplerImpl(set->getExternalSamplerVkSet(), binding, texture, VK_NULL_HANDLE);
     set->acquire(texture);
 }
 
@@ -383,32 +396,32 @@ void VulkanDescriptorSetCache::updateInputAttachment(
     // TOOD: fill this in.
 }
 
+
 fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetCache::createSet(
         Handle<HwDescriptorSet> handle, fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-    auto const vkSet = getVkSet(layout);
+    auto const vkSet = mDescriptorPool->obtainSet(layout->count, layout->getVkLayout());
     auto const& count = layout->count;
     auto const vklayout = layout->getVkLayout();
-    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
-                // down).
-                if (mDescriptorPool) {
-                    mDescriptorPool->recycle(count, vklayout, vkSet);
-                }
-            });
-    set->setVkSet(vkSet);
+    auto set = fvkmemory::resource_ptr<VulkanDescriptorSet>::make(
+            mResourceManager, handle, layout->bitmask.dynamicUbo, layout->count.dynamicUbo,
+            [vkSet, count, vklayout, this](
+                    VulkanDescriptorSet*) { this->manualRecycle(count, vklayout, vkSet); },
+            vkSet);
     return set;
 }
 
-VkDescriptorSet VulkanDescriptorSetCache::getVkSet(
-        fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-    return mDescriptorPool->obtainSet(layout);
+VkDescriptorSet VulkanDescriptorSetCache::getVkSet(DescriptorCount const& count,
+        VkDescriptorSetLayout vklayout) {
+    return mDescriptorPool->obtainSet(count, vklayout);
 }
 
-void VulkanDescriptorSetCache::manualRecyle(VulkanDescriptorSetLayout::Count const& count,
+void VulkanDescriptorSetCache::manualRecycle(VulkanDescriptorSetLayout::Count const& count,
         VkDescriptorSetLayout vklayout, VkDescriptorSet vkSet) {
-    mDescriptorPool->recycle(count, vklayout, vkSet);
+    // Note that mDescriptorPool could be gone due to terminate (when the backend shuts
+    // down).
+    if (mDescriptorPool) {
+        mDescriptorPool->recycle(count, vklayout, vkSet);
+    }
 }
 
 void VulkanDescriptorSetCache::gc() { mStashedSets = {}; }

filament/backend/src/vulkan/VulkanDescriptorSetCache.h

@@ -43,6 +43,7 @@ public:
     using DescriptorSetLayoutArray = VulkanDescriptorSetLayout::DescriptorSetLayoutArray;
     using DescriptorSetArray =
             std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>, UNIQUE_DESCRIPTOR_SET_COUNT>;
+    using DescriptorCount = VulkanDescriptorSetLayout::Count;
 
     VulkanDescriptorSetCache(VkDevice device, fvkmemory::ResourceManager* resourceManager);
     ~VulkanDescriptorSetCache();
@@ -56,6 +57,10 @@ public:
     void updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t binding,
             fvkmemory::resource_ptr<VulkanTexture> texture, VkSampler sampler) noexcept;
 
+    void updateSamplerForExternalSamplerSet(fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t binding,
+            fvkmemory::resource_ptr<VulkanTexture> texture) noexcept;
+
+
     void updateInputAttachment(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
             VulkanAttachment const& attachment) noexcept;
 
@@ -65,17 +70,17 @@ public:
     void unbind(uint8_t setIndex);
 
     void commit(VulkanCommandBuffer* commands, VkPipelineLayout pipelineLayout,
+            fvkutils::DescriptorSetMask const& useExternalSamplerMask,
             fvkutils::DescriptorSetMask const& setMask);
 
     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 meant to be used with external samplers
+    VkDescriptorSet getVkSet(DescriptorCount const& count, VkDescriptorSetLayout vklayout);
 
-    // This method is only meant to be used with external samplers.
-    void manualRecyle(VulkanDescriptorSetLayout::Count const& count, VkDescriptorSetLayout vklayout,
+    // This method is meant to be used with external samplers
+    void manualRecycle(VulkanDescriptorSetLayout::Count const& count, VkDescriptorSetLayout vklayout,
             VkDescriptorSet vkSet);
 
     DescriptorSetArray const& getBoundSets() const { return mStashedSets; }
@@ -83,6 +88,9 @@ public:
     void gc();
 
 private:
+    void updateSamplerImpl(VkDescriptorSet set, uint8_t binding,
+            fvkmemory::resource_ptr<VulkanTexture> texture, VkSampler sampler) noexcept;
+
     class DescriptorInfinitePool;
 
     VkDevice mDevice;

filament/backend/src/vulkan/VulkanDescriptorSetLayoutCache.cpp

@@ -127,6 +127,7 @@ void VulkanDescriptorSetLayoutCache::terminate() noexcept {
 
 VkDescriptorSetLayout VulkanDescriptorSetLayoutCache::getVkLayout(
         VulkanDescriptorSetLayout::Bitmask const& bitmasks,
+        fvkutils::SamplerBitmask externalSamplers,
         utils::FixedCapacityVector<VkSampler> immutableSamplers) {
     LayoutKey key = {
         .bitmask = bitmasks,
@@ -141,7 +142,7 @@ VkDescriptorSetLayout VulkanDescriptorSetLayoutCache::getVkLayout(
     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,
+    count += appendSamplerBindings(&toBind[count], bitmasks.sampler, externalSamplers,
             immutableSamplers);
     count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
             bitmasks.inputAttachment);
@@ -160,9 +161,9 @@ VkDescriptorSetLayout VulkanDescriptorSetLayoutCache::getVkLayout(
 
 fvkmemory::resource_ptr<VulkanDescriptorSetLayout> VulkanDescriptorSetLayoutCache::createLayout(
         Handle<HwDescriptorSetLayout> handle, backend::DescriptorSetLayout&& info) {
+    BitmaskGroup maskGroup = VulkanDescriptorSetLayout::Bitmask::fromLayoutDescription(info);
     auto layout = fvkmemory::resource_ptr<VulkanDescriptorSetLayout>::make(mResourceManager, handle,
-            info);
-    layout->setVkLayout(getVkLayout(layout->bitmask));
+            std::move(info), getVkLayout(maskGroup, maskGroup.externalSampler));
     return layout;
 }
 

filament/backend/src/vulkan/VulkanDescriptorSetLayoutCache.h

@@ -46,6 +46,7 @@ public:
 
     // This method is meant to be used with external samplers
     VkDescriptorSetLayout getVkLayout(VulkanDescriptorSetLayout::Bitmask const& bitmasks,
+            fvkutils::SamplerBitmask externalSamplers,
             utils::FixedCapacityVector<VkSampler> immutableSamplers = {});
 
 private:

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -30,7 +30,6 @@
 #include "vulkan/memory/ResourcePointer.h"
 #include "vulkan/utils/Conversion.h"
 #include "vulkan/utils/Definitions.h"
-#include "vulkan/vulkan_core.h"
 
 #include <backend/DriverEnums.h>
 #include <backend/platforms/VulkanPlatform.h>
@@ -198,7 +197,7 @@ Dispatcher VulkanDriver::getDispatcher() const noexcept {
 }
 
 VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext const& context,
-        Platform::DriverConfig const& driverConfig) noexcept
+        Platform::DriverConfig const& driverConfig)
     : mPlatform(platform),
       mResourceManager(driverConfig.handleArenaSize, driverConfig.disableHandleUseAfterFreeCheck,
               driverConfig.disableHeapHandleTags),
@@ -251,7 +250,7 @@ VulkanDriver::~VulkanDriver() noexcept = default;
 
 UTILS_NOINLINE
 Driver* VulkanDriver::create(VulkanPlatform* platform, VulkanContext const& context,
-         Platform::DriverConfig const& driverConfig) noexcept {
+         Platform::DriverConfig const& driverConfig) {
 #if 0
     // this is useful for development, but too verbose even for debug builds
     // For reference on a 64-bits machine in Release mode:
@@ -331,7 +330,7 @@ void VulkanDriver::terminate() {
 
     mStagePool.terminate();
     mPipelineCache.terminate();
-    mFramebufferCache.reset();
+    mFramebufferCache.terminate();
     mSamplerCache.terminate();
     mDescriptorSetLayoutCache.terminate();
     mDescriptorSetCache.terminate();
@@ -438,6 +437,7 @@ void VulkanDriver::updateDescriptorSetTexture(
         };
         VkSampler const vksampler = mSamplerCache.getSampler(cacheParams);
         mDescriptorSetCache.updateSampler(set, binding, texture, vksampler);
+        mExternalImageManager.clearTextureBinding(set, binding);
     }
 }
 
@@ -458,6 +458,7 @@ void VulkanDriver::finish(int dummy) {
 void VulkanDriver::createRenderPrimitiveR(Handle<HwRenderPrimitive> rph,
         Handle<HwVertexBuffer> vbh, Handle<HwIndexBuffer> ibh,
         PrimitiveType pt) {
+    FVK_SYSTRACE_SCOPE();
     auto vb = resource_ptr<VulkanVertexBuffer>::cast(&mResourceManager, vbh);
     auto ib = resource_ptr<VulkanIndexBuffer>::cast(&mResourceManager, ibh);
     auto ptr = resource_ptr<VulkanRenderPrimitive>::make(&mResourceManager, rph, pt, vb, ib);
@@ -468,12 +469,14 @@ void VulkanDriver::destroyRenderPrimitive(Handle<HwRenderPrimitive> rph) {
     if (!rph) {
         return;
     }
+    FVK_SYSTRACE_SCOPE();
     auto ptr = resource_ptr<VulkanRenderPrimitive>::cast(&mResourceManager, rph);
     ptr.dec();
 }
 
 void VulkanDriver::createVertexBufferInfoR(Handle<HwVertexBufferInfo> vbih, uint8_t bufferCount,
         uint8_t attributeCount, AttributeArray attributes) {
+    FVK_SYSTRACE_SCOPE();
     auto vbi = resource_ptr<VulkanVertexBufferInfo>::make(&mResourceManager, vbih, bufferCount,
             attributeCount, attributes);
     vbi.inc();
@@ -483,12 +486,14 @@ void VulkanDriver::destroyVertexBufferInfo(Handle<HwVertexBufferInfo> vbih) {
     if (!vbih) {
         return;
     }
+    FVK_SYSTRACE_SCOPE();
     auto vbi = resource_ptr<VulkanVertexBufferInfo>::cast(&mResourceManager, vbih);
     vbi.dec();
 }
 
 void VulkanDriver::createVertexBufferR(Handle<HwVertexBuffer> vbh, uint32_t vertexCount,
         Handle<HwVertexBufferInfo> vbih) {
+    FVK_SYSTRACE_SCOPE();
     auto vbi = resource_ptr<VulkanVertexBufferInfo>::cast(&mResourceManager, vbih);
     auto vb = resource_ptr<VulkanVertexBuffer>::make(&mResourceManager, vbh, mContext, mStagePool,
             vertexCount, vbi);
@@ -499,12 +504,14 @@ void VulkanDriver::destroyVertexBuffer(Handle<HwVertexBuffer> vbh) {
     if (!vbh) {
         return;
     }
+    FVK_SYSTRACE_SCOPE();
     auto vb = resource_ptr<VulkanVertexBuffer>::cast(&mResourceManager, vbh);
     vb.dec();
 }
 
 void VulkanDriver::createIndexBufferR(Handle<HwIndexBuffer> ibh, ElementType elementType,
         uint32_t indexCount, BufferUsage usage) {
+    FVK_SYSTRACE_SCOPE();
     auto elementSize = (uint8_t) getElementTypeSize(elementType);
     auto ib = resource_ptr<VulkanIndexBuffer>::make(&mResourceManager, ibh, mAllocator, mStagePool,
             elementSize, indexCount);
@@ -515,12 +522,14 @@ void VulkanDriver::destroyIndexBuffer(Handle<HwIndexBuffer> ibh) {
     if (!ibh) {
         return;
     }
+    FVK_SYSTRACE_SCOPE();
     auto ib = resource_ptr<VulkanIndexBuffer>::cast(&mResourceManager, ibh);
     ib.dec();
 }
 
 void VulkanDriver::createBufferObjectR(Handle<HwBufferObject> boh, uint32_t byteCount,
         BufferObjectBinding bindingType, BufferUsage usage) {
+    FVK_SYSTRACE_SCOPE();
     auto bo = resource_ptr<VulkanBufferObject>::make(&mResourceManager, boh, mAllocator, mStagePool,
             byteCount, bindingType);
     bo.inc();
@@ -530,6 +539,7 @@ void VulkanDriver::destroyBufferObject(Handle<HwBufferObject> boh) {
     if (!boh) {
         return;
     }
+    FVK_SYSTRACE_SCOPE();
     auto bo = resource_ptr<VulkanBufferObject>::cast(&mResourceManager, boh);
     bo.dec();
 }
@@ -575,36 +585,49 @@ void VulkanDriver::createTextureExternalImage2R(Handle<HwTexture> th, backend::S
         backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
         Platform::ExternalImageHandleRef externalImage) {
     FVK_SYSTRACE_SCOPE();
-    auto const& metadata = mPlatform->extractExternalImageMetadata(externalImage);
+    auto metadata = mPlatform->extractExternalImageMetadata(externalImage);
 
-    // In theory the following are reasonable expectations, but in practice it's hard for client's
-    // to match up the dimensions of the texture with that of the AHB.
-    // assert_invariant(width == metadata.width);
-    // assert_invariant(height == metadata.height);
+    assert_invariant(width == metadata.width);
+    assert_invariant(height == metadata.height);
+
+    // We do not check the format since AHB could return both a known format and an external format.
+    // In which case, we choose one or the other, but this choice is not known to the client.
+    // Therefore the following lines are commented out.
     // assert_invariant(format == metadata.filamentFormat);
     // assert_invariant(fvkutils::getVkFormat(format) == metadata.format);
 
-    VkImage vkimg;
-    VkDeviceMemory deviceMemory;
-    std::tie(vkimg, deviceMemory) = mPlatform->createVkImageFromExternal(externalImage);
+    auto imgData = mPlatform->createVkImageFromExternal(externalImage);
 
-    VkSamplerYcbcrConversion conversion =
-            mExternalImageManager.getVkSamplerYcbcrConversion(metadata);
+    assert_invariant(imgData.internal.valid() || imgData.external.valid());
 
-    auto texture = resource_ptr<VulkanTexture>::make(&mResourceManager, th, mContext,
-            mPlatform->getDevice(), mAllocator, &mResourceManager, &mCommands, vkimg, deviceMemory,
-            metadata.format, conversion, metadata.samples, metadata.width, metadata.height,
-            metadata.layers, usage, mStagePool);
-
-    if (conversion != VK_NULL_HANDLE) {
-        mExternalImageManager.addExternallySampledTexture(texture, externalImage);
+    VkFormat vkformat = metadata.format;
+    VkImage vkimage = VK_NULL_HANDLE;
+    VkDeviceMemory memory = VK_NULL_HANDLE;
+    if (imgData.internal.valid()) {
+        metadata.externalFormat = 0;
+        vkimage = imgData.internal.image;
+        memory = imgData.internal.memory;
+    } else { // imgData.external.valid()
+        vkformat = VK_FORMAT_UNDEFINED;
+        vkimage = imgData.external.image;
+        memory = imgData.external.memory;
     }
 
+    VkSamplerYcbcrConversion const conversion =
+            mExternalImageManager.getVkSamplerYcbcrConversion(metadata);
+    auto texture = resource_ptr<VulkanTexture>::make(&mResourceManager, th, mContext,
+            mPlatform->getDevice(), mAllocator, &mResourceManager, &mCommands, vkimage, memory,
+            vkformat, conversion, metadata.samples, metadata.width, metadata.height,
+            metadata.layers, usage, mStagePool);
+    auto& commands = mCommands.get();
     // Unlike uploaded textures or swapchains, we need to explicit transition this
     // texture into the read layout.
-    auto& commands = mCommands.get();
     texture->transitionLayout(&commands, texture->getPrimaryViewRange(), VulkanLayout::READ_ONLY);
 
+    if (imgData.external.valid()) {
+        mExternalImageManager.addExternallySampledTexture(texture, externalImage);
+    }
+
     texture.inc();
 }
 
@@ -754,6 +777,7 @@ void VulkanDriver::createFenceR(Handle<HwFence> fh, int) {
 }
 
 void VulkanDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow, uint64_t flags) {
+    FVK_SYSTRACE_SCOPE();
     // Running gc() to guard against an edge case where the old swapchains need to have been
     // destroyed before the new swapchain can be created. Otherwise, we would fail
     // vkCreateSwapchainKHR with VK_ERROR_NATIVE_WINDOW_IN_USE_KHR.
@@ -795,6 +819,7 @@ void VulkanDriver::createTimerQueryR(Handle<HwTimerQuery> tqh, int) {
 
 void VulkanDriver::createDescriptorSetLayoutR(Handle<HwDescriptorSetLayout> dslh,
         backend::DescriptorSetLayout&& info) {
+    FVK_SYSTRACE_SCOPE();
     auto layout = mDescriptorSetLayoutCache.createLayout(dslh, std::move(info));
     layout.inc();
 }
@@ -809,7 +834,6 @@ void VulkanDriver::createDescriptorSetR(Handle<HwDescriptorSet> dsh,
 
     if (layout->hasExternalSamplers()) {
         mAppState.hasExternalSamplerLayouts = true;
-        mExternalImageManager.addDescriptorSet(layout, set);
     }
 }
 
@@ -930,17 +954,13 @@ void VulkanDriver::destroyTimerQuery(Handle<HwTimerQuery> tqh) {
 void VulkanDriver::destroyDescriptorSetLayout(Handle<HwDescriptorSetLayout> dslh) {
     auto layout = resource_ptr<VulkanDescriptorSetLayout>::cast(&mResourceManager, dslh);
     layout.dec();
-
-    if (layout->hasExternalSamplers()) {
-        mExternalImageManager.removeDescriptorSetLayout(layout);
-    }
 }
 
 void VulkanDriver::destroyDescriptorSet(Handle<HwDescriptorSet> dsh) {
     auto set = resource_ptr<VulkanDescriptorSet>::cast(&mResourceManager, dsh);
     set.dec();
 
-    if (mAppState.hasExternalSamplers()) {
+    if (mAppState.hasExternalSamplers() && set->getExternalSamplerVkSet() != VK_NULL_HANDLE) {
         mExternalImageManager.removeDescriptorSet(set);
     }
 }
@@ -1543,7 +1563,7 @@ void VulkanDriver::makeCurrent(Handle<HwSwapChain> drawSch, Handle<HwSwapChain>
     swapChain->acquire(resized);
 
     if (resized) {
-        mFramebufferCache.reset();
+        mFramebufferCache.resetFramebuffers();
     }
 
     if (UTILS_LIKELY(mDefaultRenderTarget)) {
@@ -1744,37 +1764,69 @@ void VulkanDriver::blitDEPRECATED(TargetBufferFlags buffers,
 }
 
 void VulkanDriver::bindPipeline(PipelineState const& pipelineState) {
-    // We need to determine whether to delay bindning until draw().
-    mPipelineState.bindInDraw.first = false;
+    // This resets all of the pipeline states; the most relevant (needing reset) is .bindInDraw.
+    mPipelineState = {};
+
+    auto& setLayouts = pipelineState.pipelineLayout.setLayout;
+    DescriptorSetLayoutHandleList layoutHandles;
+    uint8_t layoutCount = 0;
+    std::transform(setLayouts.begin(), setLayouts.end(), layoutHandles.begin(),
+            [&](auto const& handle) -> resource_ptr<VulkanDescriptorSetLayout> {
+                if (!handle) {
+                    return {};
+                }
+                layoutCount++;
+                return resource_ptr<VulkanDescriptorSetLayout>::cast(&mResourceManager, handle);
+            });
+
+    constexpr uint8_t descriptorSetMaskTable[4] = {0x1, 0x3, 0x7, 0xF};
+    fvkutils::DescriptorSetMask const descriptorSetMask =
+            fvkutils::DescriptorSetMask(descriptorSetMaskTable[layoutCount]);
+
     if (mAppState.hasExternalSamplers()) {
-        auto& layouts = pipelineState.pipelineLayout.setLayout;
-        auto haveExternalSamplers = [&](auto hwHandle) {
-            if (!hwHandle) {
+        auto const haveExternalSamplers = [&](auto layoutHandle) {
+            if (!layoutHandle) {
                 return false;
             }
-            auto layout =
-                    resource_ptr<VulkanDescriptorSetLayout>::cast(&mResourceManager, hwHandle);
-            return layout->hasExternalSamplers();
+            return layoutHandle->hasExternalSamplers();
         };
-        if (std::any_of(layouts.begin(), layouts.end(), haveExternalSamplers)) {
-            mPipelineState.bindInDraw = { true, pipelineState };
+        if (std::any_of(layoutHandles.begin(), layoutHandles.end(), haveExternalSamplers)) {
+            BindInDrawBundle bundle = {
+                .pipelineState = pipelineState,
+                .dsLayoutHandles = layoutHandles,
+                .descriptorSetMask = descriptorSetMask,
+            };
+            mPipelineState.bindInDraw = { true, bundle };
             return;
         }
     }
-    bindPipelineImpl(pipelineState);
+
+    // The normal, non-external sampler path
+    using VkDescriptorSetLayoutArray = VulkanPipelineLayoutCache::DescriptorSetLayoutArray;
+    VkDescriptorSetLayoutArray vkLayouts;
+    std::transform(layoutHandles.begin(), layoutHandles.end(), vkLayouts.begin(),
+            [](auto const& layout) -> VkDescriptorSetLayout {
+                if (!layout) {
+                    return VK_NULL_HANDLE;
+                }
+                return layout->getVkLayout();
+            });
+    auto program = resource_ptr<VulkanProgram>::cast(&mResourceManager, pipelineState.program);
+    auto pipelineLayout = mPipelineLayoutCache.getLayout(vkLayouts, program);
+    bindPipelineImpl(pipelineState, pipelineLayout, descriptorSetMask);
 }
 
-void VulkanDriver::bindPipelineImpl(PipelineState const& pipelineState) {
+void VulkanDriver::bindPipelineImpl(PipelineState const& pipelineState,
+        VkPipelineLayout pipelineLayout, fvkutils::DescriptorSetMask descriptorSetMask) {
     FVK_SYSTRACE_SCOPE();
     auto commands = mCurrentRenderPass.commandBuffer;
     auto vbi = resource_ptr<VulkanVertexBufferInfo>::cast(&mResourceManager,
             pipelineState.vertexBufferInfo);
 
-    Handle<HwProgram> programHandle = pipelineState.program;
     RasterState const& rasterState = pipelineState.rasterState;
     PolygonOffset const& depthOffset = pipelineState.polygonOffset;
 
-    auto program = resource_ptr<VulkanProgram>::cast(&mResourceManager, programHandle);
+    auto program = resource_ptr<VulkanProgram>::cast(&mResourceManager, pipelineState.program);
     commands->acquire(program);
 
     // Update the VK raster state.
@@ -1816,29 +1868,11 @@ void VulkanDriver::bindPipelineImpl(PipelineState const& pipelineState) {
     mPipelineCache.bindPrimitiveTopology(topology);
     mPipelineCache.bindVertexArray(attribDesc, bufferDesc, vbi->getAttributeCount());
 
-    auto& setLayouts = pipelineState.pipelineLayout.setLayout;
-    VulkanDescriptorSetLayout::DescriptorSetLayoutArray layoutList;
-    uint8_t layoutCount = 0;
-    std::transform(setLayouts.begin(), setLayouts.end(), layoutList.begin(),
-            [&](Handle<HwDescriptorSetLayout> handle) -> VkDescriptorSetLayout {
-                if (!handle) {
-                    return VK_NULL_HANDLE;
-                }
-                auto layout =
-                        resource_ptr<VulkanDescriptorSetLayout>::cast(&mResourceManager, handle);
-                layoutCount++;
-                return layout->getVkLayout();
-            });
-    auto pipelineLayout = mPipelineLayoutCache.getLayout(layoutList, program);
-
-    constexpr uint8_t descriptorSetMaskTable[4] = {0x1, 0x3, 0x7, 0xF};
-
-    mPipelineState = {
-        .program = program,
-        .pipelineLayout = pipelineLayout,
-        .descriptorSetMask = fvkutils::DescriptorSetMask(descriptorSetMaskTable[layoutCount]),
-        .bindInDraw = {false, {}},
-    };
+    // Note that we cannot reinit mPipeline because the .bindInDraw metadata that needs to carry
+    // over even on bind.
+    mPipelineState.program = program;
+    mPipelineState.pipelineLayout = pipelineLayout;
+    mPipelineState.descriptorSetMask = descriptorSetMask;
 
     mPipelineCache.bindLayout(pipelineLayout);
     mPipelineCache.bindPipeline(mCurrentRenderPass.commandBuffer);
@@ -1883,20 +1917,36 @@ void VulkanDriver::bindDescriptorSet(
 void VulkanDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
     FVK_SYSTRACE_SCOPE();
     VkCommandBuffer cmdbuffer = mCurrentRenderPass.commandBuffer->buffer();
+    auto const& [doBindInDraw, bundle] = mPipelineState.bindInDraw;
 
-    if (mAppState.hasExternalSamplers()) {
-        auto const& [bindInDraw, pipelineSt] = mPipelineState.bindInDraw;
-        bool const hasUpdated =
-                mExternalImageManager.prepareBindSets(mDescriptorSetCache.getBoundSets());
-        if (bindInDraw || hasUpdated) {
-            bindPipelineImpl(pipelineSt);
+    fvkutils::DescriptorSetMask setsWithExternalSamplers = {};
+    if (doBindInDraw) {
+        auto& layoutHandles = bundle.dsLayoutHandles;
+        setsWithExternalSamplers = mExternalImageManager.prepareBindSets(layoutHandles,
+                mDescriptorSetCache.getBoundSets());
+
+        VulkanDescriptorSetLayout::DescriptorSetLayoutArray vklayouts;
+        for (size_t i = 0; i < layoutHandles.size(); i++) {
+            if (!layoutHandles[i]) {
+                vklayouts[i] = VK_NULL_HANDLE;
+                continue;
+            }
+            if (setsWithExternalSamplers[i]) {
+                vklayouts[i] = layoutHandles[i]->getExternalSamplerVkLayout();
+            } else {
+                vklayouts[i] = layoutHandles[i]->getVkLayout();
+            }
+        }
+        auto program =
+                resource_ptr<VulkanProgram>::cast(&mResourceManager, bundle.pipelineState.program);
+        VkPipelineLayout const pipelineLayout = mPipelineLayoutCache.getLayout(vklayouts, program);
+        if (pipelineLayout != mPipelineState.pipelineLayout) {
+            bindPipelineImpl(bundle.pipelineState, pipelineLayout, bundle.descriptorSetMask);
         }
         mPipelineState.bindInDraw.first = false;
     }
-
-    mDescriptorSetCache.commit(mCurrentRenderPass.commandBuffer,
-            mPipelineState.pipelineLayout,
-            mPipelineState.descriptorSetMask);
+    mDescriptorSetCache.commit(mCurrentRenderPass.commandBuffer, mPipelineState.pipelineLayout,
+            setsWithExternalSamplers, mPipelineState.descriptorSetMask);
 
     // Finally, make the actual draw call. TODO: support subranges
     uint32_t const firstIndex = indexOffset;

filament/backend/src/vulkan/VulkanDriver.h

@@ -55,7 +55,7 @@ constexpr uint8_t MAX_RENDERTARGET_ATTACHMENT_TEXTURES =
 class VulkanDriver final : public DriverBase {
 public:
     static Driver* create(VulkanPlatform* platform, VulkanContext const& context,
-            Platform::DriverConfig const& driverConfig) noexcept;
+            Platform::DriverConfig const& driverConfig);
 
 #if FVK_ENABLED(FVK_DEBUG_DEBUG_UTILS)
     // Encapsulates the VK_EXT_debug_utils extension.  In particular, we use
@@ -90,8 +90,8 @@ private:
     void debugCommandBegin(CommandStream* cmds, bool synchronous,
             const char* methodName) noexcept override;
 
-    inline VulkanDriver(VulkanPlatform* platform, VulkanContext const& context,
-            Platform::DriverConfig const& driverConfig) noexcept;
+    VulkanDriver(VulkanPlatform* platform, VulkanContext const& context,
+            Platform::DriverConfig const& driverConfig);
 
     ~VulkanDriver() noexcept override;
 
@@ -120,7 +120,8 @@ private:
 
 private:
     void collectGarbage();
-    void bindPipelineImpl(PipelineState const& pipelineState);
+    void bindPipelineImpl(PipelineState const& pipelineState, VkPipelineLayout pipelineLayout,
+            fvkutils::DescriptorSetMask descriptorSetMask);
 
     VulkanPlatform* mPlatform = nullptr;
     fvkmemory::ResourceManager mResourceManager;
@@ -148,14 +149,24 @@ private:
     VulkanExternalImageManager mExternalImageManager;
 
     // This is necessary for us to write to push constants after binding a pipeline.
+    using DescriptorSetLayoutHandleList = std::array<resource_ptr<VulkanDescriptorSetLayout>,
+            VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT>;
+
+    struct BindInDrawBundle {
+        PipelineState pipelineState = {};
+        DescriptorSetLayoutHandleList dsLayoutHandles = {};
+        fvkutils::DescriptorSetMask descriptorSetMask = {};
+        resource_ptr<VulkanProgram> program = {};
+    };
+
     struct {
         // For push constant
-        resource_ptr<VulkanProgram> program;
+        resource_ptr<VulkanProgram> program = {};
         // For push commiting dynamic ubos in draw()
-        VkPipelineLayout pipelineLayout;
-        fvkutils::DescriptorSetMask descriptorSetMask;
+        VkPipelineLayout pipelineLayout = VK_NULL_HANDLE;
+        fvkutils::DescriptorSetMask descriptorSetMask = {};
 
-        std::pair<bool, PipelineState> bindInDraw = {false, {}};
+        std::pair<bool, BindInDrawBundle> bindInDraw = {false, {}};
     } mPipelineState = {};
 
     struct {

filament/backend/src/vulkan/VulkanExternalImageManager.cpp

@@ -31,13 +31,51 @@ namespace filament::backend {
 
 namespace {
 
+using Bitmask = fvkutils::UniformBufferBitmask;
+static_assert(sizeof(Bitmask) * 8 == fvkutils::MAX_DESCRIPTOR_SET_BITMASK_BITS);
+
 template<typename T>
-void erase(std::vector<T>& v, std::function<bool(T const&)> f) {
+void erasep(std::vector<T>& v, std::function<bool(T const&)> f) {
     auto newEnd = std::remove_if(v.begin(), v.end(), f);
     v.erase(newEnd, v.end());
 }
 
-} // anonymous
+using ImageData = VulkanExternalImageManager::VulkanExternalImageManager::ImageData;
+ImageData& findImage(std::vector<ImageData>& images,
+        fvkmemory::resource_ptr<VulkanTexture> texture) {
+    auto itr = std::find_if(images.begin(), images.end(), [&](ImageData const& data) {
+        return data.image == texture;
+    });
+    assert_invariant(itr != images.end());
+    return *itr;
+}
+
+void copySet(VkDevice device, VkDescriptorSet srcSet, VkDescriptorSet dstSet, Bitmask bindings) {
+    // TODO: fix the size for better memory management
+    std::vector<VkCopyDescriptorSet> copies;
+    bindings.forEachSetBit([&](size_t index) {
+        copies.push_back({
+            .sType = VK_STRUCTURE_TYPE_COPY_DESCRIPTOR_SET,
+            .srcSet = srcSet,
+            .srcBinding = (uint32_t) index,
+            .dstSet = dstSet,
+            .dstBinding = (uint32_t) index,
+            .descriptorCount = 1,
+        });
+    });
+    vkUpdateDescriptorSets(device, 0, nullptr, copies.size(), copies.data());
+}
+
+Bitmask foldBitsInHalf(Bitmask bitset) {
+    Bitmask outBitset;
+    bitset.forEachSetBit([&](size_t index) {
+        constexpr size_t BITMASK_LOWER_BITS_LEN = sizeof(outBitset) * 4;
+        outBitset.set(index % BITMASK_LOWER_BITS_LEN);
+    });
+    return outBitset;
+}
+
+}// namespace
 
 VulkanExternalImageManager::VulkanExternalImageManager(VulkanPlatform* platform,
         VulkanSamplerCache* samplerCache, VulkanYcbcrConversionCache* ycbcrConversionCache,
@@ -52,7 +90,6 @@ VulkanExternalImageManager::VulkanExternalImageManager(VulkanPlatform* platform,
 VulkanExternalImageManager::~VulkanExternalImageManager() = default;
 
 void VulkanExternalImageManager::terminate() {
-    mSetAndLayouts.clear();
     mSetBindings.clear();
     mImages.clear();
 }
@@ -61,45 +98,50 @@ void VulkanExternalImageManager::onBeginFrame() {
     std::for_each(mImages.begin(), mImages.end(), [](ImageData& image) {
         image.hasBeenValidated = false;
     });
+    std::for_each(mSetBindings.begin(), mSetBindings.end(), [](SetBindingInfo& info) {
+        info.bound = false;
+    });
 }
 
-bool VulkanExternalImageManager::prepareBindSets(SetArray const& sets) {
-    bool hasUpdated = false;
-    for (auto set: sets) {
-        if (!set) {
+fvkutils::DescriptorSetMask VulkanExternalImageManager::prepareBindSets(LayoutArray const& layouts,
+        SetArray const& sets) {
+    fvkutils::DescriptorSetMask shouldUseExternalSampler{};
+    for (uint8_t i = 0; i < sets.size(); i++) {
+        auto set = sets[i];
+        auto layout = layouts[i];
+        if (!set || !layout) {
             continue;
         }
-        if (auto itr = std::find_if(mSetAndLayouts.begin(), mSetAndLayouts.end(),
-                    [&](auto const& setAndLayout) { return setAndLayout.first == set; });
-                itr != mSetAndLayouts.end()) {
-            hasUpdated = updateSetAndLayout(itr->first, itr->second) || hasUpdated;
+        if (hasExternalSampler(set)) {
+            updateSetAndLayout(set, layout);
+            shouldUseExternalSampler.set(i);
         }
     }
-    return hasUpdated;
+    return shouldUseExternalSampler;
 }
 
-bool VulkanExternalImageManager::updateSetAndLayout(
+bool VulkanExternalImageManager::hasExternalSampler(
+        fvkmemory::resource_ptr<VulkanDescriptorSet> set) {
+    auto itr = std::find_if(mSetBindings.begin(), mSetBindings.end(),
+            [&](SetBindingInfo const& info) { return info.set == set; });
+    return itr != mSetBindings.end();
+}
+
+void VulkanExternalImageManager::updateSetAndLayout(
         fvkmemory::resource_ptr<VulkanDescriptorSet> set,
         fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-    auto findImage = [&](fvkmemory::resource_ptr<VulkanTexture> texture) -> ImageData* {
-        auto itr = std::find_if(mImages.begin(), mImages.end(), [&](ImageData const& data) {
-            return data.ptr == texture;
-        });
-        assert_invariant(itr != mImages.end());
-        return &(*itr);
-    };
-
-    //std::vector<std::pair<uint8_t, ImageData*>> externalImages;
-    utils::FixedCapacityVector<std::pair<uint8_t, VkSampler>> samplerAndBindings;
+    utils::FixedCapacityVector<
+            std::tuple<uint8_t, VkSampler, fvkmemory::resource_ptr<VulkanTexture>>>
+            samplerAndBindings;
     samplerAndBindings.reserve(MAX_SAMPLER_COUNT);
 
-    bool hasImageUpdates = false;
+    fvkutils::SamplerBitmask actualExternalSamplers;
     for (auto& bindingInfo : mSetBindings) {
-        if (bindingInfo.set != set) {
+        if (bindingInfo.set != set || bindingInfo.bound) {
             continue;
         }
-        auto imageData = findImage(bindingInfo.image);
-        hasImageUpdates = updateImage(imageData) || hasImageUpdates;
+        auto& imageData = findImage(mImages, bindingInfo.image);
+        updateImage(&imageData);
 
         auto samplerParams = bindingInfo.samplerParams;
         // according to spec, these must match chromaFilter
@@ -109,88 +151,64 @@ bool VulkanExternalImageManager::updateSetAndLayout(
 
         auto sampler = mSamplerCache->getSampler({
             .sampler = samplerParams,
-            .conversion = imageData->conversion,
+            .conversion = imageData.conversion,
         });
-        samplerAndBindings.push_back({ bindingInfo.binding, sampler });
+        actualExternalSamplers.set(bindingInfo.binding);
+        samplerAndBindings.push_back({ bindingInfo.binding, sampler, bindingInfo.image });
+        bindingInfo.bound = true;
     }
 
-    // We need to sort by binding number
-    std::sort(samplerAndBindings.begin(), samplerAndBindings.end());
+    if (samplerAndBindings.empty()) {
+        return;
+    }
+
+    // Sort by binding number
+    std::sort(samplerAndBindings.begin(), samplerAndBindings.end(), [](auto const& a, auto const& b) {
+        return std::get<0>(a) < std::get<0>(b);
+    });
 
     utils::FixedCapacityVector<VkSampler> outSamplers;
     outSamplers.reserve(MAX_SAMPLER_COUNT);
     std::for_each(samplerAndBindings.begin(), samplerAndBindings.end(),
-            [&](auto const& b) { outSamplers.push_back(b.second); });
+            [&](auto const& b) { outSamplers.push_back(std::get<1>(b)); });
 
-    VkDescriptorSetLayout const oldLayout = layout->getVkLayout();
-    VkDescriptorSetLayout const newLayout =
-            mDescriptorSetLayoutCache->getVkLayout(layout->bitmask, outSamplers);
-    bool const hasLayoutUpdate = oldLayout != newLayout;
-    layout->setVkLayout(newLayout);
+    VkDescriptorSetLayout const oldLayout = layout->getExternalSamplerVkLayout();
+    VkDescriptorSetLayout const newLayout = mDescriptorSetLayoutCache->getVkLayout(layout->bitmask,
+            actualExternalSamplers, outSamplers);
 
-    assert_invariant(
-            (!hasImageUpdates && !hasLayoutUpdate) ||
-            (hasImageUpdates && hasLayoutUpdate));
+    // Need to copy the set
+    VkDescriptorSet const oldSet = set->getExternalSamplerVkSet();
+    if (oldLayout != newLayout || oldSet == VK_NULL_HANDLE) {
+        // Build a new descriptor set from the new layout
+        VkDescriptorSet const newSet = mDescriptorSetCache->getVkSet(layout->count, newLayout);
+        auto const ubo = layout->bitmask.ubo | layout->bitmask.dynamicUbo;
+        auto const samplers = layout->bitmask.sampler & (~actualExternalSamplers);
 
-    if (!hasLayoutUpdate) {
-        return false;
-    }
+        // Each bitmask denotes a binding index, and separated into two stages - vertex and buffer
+        // We fold the two stages into just the lower half of the bits to denote a combined set of
+        // bindings.
+        Bitmask const copyBindings = foldBitsInHalf(ubo | samplers);
+        VkDescriptorSet const srcSet = oldSet != VK_NULL_HANDLE ? oldSet : set->getVkSet();
+        copySet(mPlatform->getDevice(), srcSet, newSet, copyBindings);
 
-    auto foldBitsInHalf = [](auto bitset) {
-        constexpr size_t BITMASK_LOWER_BITS_LEN = sizeof(bitset) * 4;
-        decltype(bitset) outBitset;
-        bitset.forEachSetBit([&](size_t index) { outBitset.set(index % BITMASK_LOWER_BITS_LEN); });
-        return outBitset;
-    };
-    // We need to build a new descriptor set from the new layout
-    VkDescriptorSet oldSet = set->getVkSet();
-    VkDescriptorSet newSet = mDescriptorSetCache->getVkSet(layout);
-
-    using Bitmask = fvkutils::UniformBufferBitmask;
-    static_assert(sizeof(Bitmask) * 8 == fvkutils::MAX_DESCRIPTOR_SET_BITMASK_BITS);
-
-    auto const ubo = layout->bitmask.ubo | layout->bitmask.dynamicUbo;
-    auto const samplers = layout->bitmask.sampler & (~layout->bitmask.externalSampler);
-
-    // each bitmask denotes a binding index, and separated into two stages - vertex and buffer
-    // We fold the two stages into just the lower half of the bits to denote a combined set of
-    // bindings.
-    Bitmask const copyBindings = foldBitsInHalf(ubo | samplers);
-
-    // TODO: fix the size for better memory
-    std::vector<VkCopyDescriptorSet> copies;
-    copyBindings.forEachSetBit([&](size_t index) {
-        copies.push_back({
-            .sType = VK_STRUCTURE_TYPE_COPY_DESCRIPTOR_SET,
-            .srcSet = oldSet,
-            .srcBinding = (uint32_t) index,
-            .dstSet = newSet,
-            .dstBinding = (uint32_t) index,
-            .descriptorCount = 1,
+        set->setExternalSamplerVkSet(newSet, [&](VulkanDescriptorSet*) {
+            mDescriptorSetCache->manualRecycle(layout->count, newLayout, newSet);
         });
-    });
-    vkUpdateDescriptorSets(mPlatform->getDevice(), 0, nullptr, copies.size(), copies.data());
-
-    set->setVkSet(newSet);
-
-    // We need to release the vkset, which is no longer used, back into the pool.
-    mDescriptorSetCache->manualRecyle(layout->count, oldLayout, oldSet);
-
-    // We need to update the external samplers in the set
-    for (auto& bindingInfo: mSetBindings) {
-        if (bindingInfo.set != set) {
-            continue;
+        if (oldLayout != newLayout) {
+            layout->setExternalSamplerVkLayout(newLayout);
         }
-        mDescriptorSetCache->updateSampler(set, bindingInfo.binding, bindingInfo.image,
-                VK_NULL_HANDLE);
     }
-    return true;
+
+    // Update the external samplers in the set
+    for (auto& [binding, sampler, image]: samplerAndBindings) {
+        mDescriptorSetCache->updateSamplerForExternalSamplerSet(set, binding, image);
+    }
 }
 
 VkSamplerYcbcrConversion VulkanExternalImageManager::getVkSamplerYcbcrConversion(
         VulkanPlatform::ExternalImageMetadata const& metadata) {
     // This external image does not require external sampler (YUV conversion).
-    if (metadata.externalFormat == 0) {
+    if (metadata.externalFormat == 0 && !fvkutils::isVKYcbcrConversionFormat(metadata.format)) {
         return VK_NULL_HANDLE;
     }
     VulkanYcbcrConversionCache::Params ycbcrParams = {
@@ -207,74 +225,70 @@ VkSamplerYcbcrConversion VulkanExternalImageManager::getVkSamplerYcbcrConversion
             // Unclear where to get the chromaFilter, we just assume it's nearest.
             .chromaFilter = SamplerMagFilter::NEAREST,
         },
-        .format = metadata.filamentFormat,
+        .format = metadata.format,
         .externalFormat = metadata.externalFormat,
     };
     return mYcbcrConversionCache->getConversion(ycbcrParams);
 }
 
-bool VulkanExternalImageManager::updateImage(ImageData* image) {
+void VulkanExternalImageManager::updateImage(ImageData* image) {
     if (image->hasBeenValidated) {
-        return false;
+        return;
     }
     image->hasBeenValidated = true;
 
     auto metadata = mPlatform->extractExternalImageMetadata(image->platformHandle);
     auto vkYcbcr = getVkSamplerYcbcrConversion(metadata);
     if (vkYcbcr == image->conversion) {
-        return false;
+        return;
     }
 
-    image->ptr->setYcbcrConversion(vkYcbcr, metadata.externalFormat != 0);
+    image->image->setYcbcrConversion(vkYcbcr);
     image->conversion = vkYcbcr;
-    return true;
-}
-
-void VulkanExternalImageManager::addDescriptorSet(
-        fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout,
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set) {
-    mSetAndLayouts.push_back({set, layout});
+    return;
 }
 
 void VulkanExternalImageManager::removeDescriptorSet(
         fvkmemory::resource_ptr<VulkanDescriptorSet> inSet) {
-    erase<SetAndLayout>(mSetAndLayouts,
-            [&](auto const& setLayout) { return (setLayout.first == inSet); });
-    erase<SetBindingInfo>(mSetBindings,
+    erasep<SetBindingInfo>(mSetBindings,
             [&](auto const& bindingInfo) { return (bindingInfo.set == inSet); });
 }
 
-void VulkanExternalImageManager::removeDescriptorSetLayout(
-        fvkmemory::resource_ptr<VulkanDescriptorSetLayout> inLayout) {
-    erase<SetAndLayout>(mSetAndLayouts,
-            [&](auto const& setLayout) { return (setLayout.second == inLayout); });
-}
-
 void VulkanExternalImageManager::bindExternallySampledTexture(
         fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t bindingPoint,
         fvkmemory::resource_ptr<VulkanTexture> image, SamplerParams samplerParams) {
     // Should we do duplicate validation here?
-    mSetBindings.push_back({ bindingPoint, image, set, samplerParams });
+    auto& imageData = findImage(mImages, image);
+    mSetBindings.push_back({ bindingPoint, imageData.image, set, samplerParams });
 }
 
 void VulkanExternalImageManager::addExternallySampledTexture(
-        fvkmemory::resource_ptr<VulkanTexture> image,
+       fvkmemory::resource_ptr<VulkanTexture> image,
         Platform::ExternalImageHandleRef platformHandleRef) {
     mImages.push_back({ image, platformHandleRef, false });
 }
 
 void VulkanExternalImageManager::removeExternallySampledTexture(
         fvkmemory::resource_ptr<VulkanTexture> image) {
-    erase<SetBindingInfo>(mSetBindings,
+    erasep<SetBindingInfo>(mSetBindings,
             [&](auto const& bindingInfo) { return (bindingInfo.image == image); });
-    erase<ImageData>(mImages, [&](auto const& imageData) { return imageData.ptr == image; });
+    erasep<ImageData>(mImages, [&](auto const& imageData) {
+        return imageData.image == image;
+    });
 }
 
 bool VulkanExternalImageManager::isExternallySampledTexture(
         fvkmemory::resource_ptr<VulkanTexture> image) const {
-    return std::find_if(mImages.begin(), mImages.end(),
-                   [&](auto const& imageData) { return imageData.ptr == image; }) != mImages.end();
+    return std::find_if(mImages.begin(), mImages.end(), [&](auto const& imageData) {
+        return imageData.image == image;
+    }) != mImages.end();
 }
 
+void VulkanExternalImageManager::clearTextureBinding(
+        fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t bindingPoint) {
+    erasep<SetBindingInfo>(mSetBindings, [&](auto const& bindingInfo) {
+        return (bindingInfo.set == set && bindingInfo.binding == bindingPoint);
+    });
+}
 
 } // namesapce filament::backend

filament/backend/src/vulkan/VulkanExternalImageManager.h

@@ -51,13 +51,14 @@ public:
     using SetArray = std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>,
             VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT>;
 
-    // This sets the currently bound layouts objects for the pipeline
-    bool prepareBindSets(SetArray const& layouts);
+    using LayoutArray = std::array<fvkmemory::resource_ptr<VulkanDescriptorSetLayout>,
+            VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT>;
 
-    void addDescriptorSet(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout,
-            fvkmemory::resource_ptr<VulkanDescriptorSet> set);
+    using VkLayoutArray = VulkanDescriptorSetLayout::DescriptorSetLayoutArray;
 
-    void removeDescriptorSetLayout(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
+    // Returns  bitmask to indicate whether or not to use the external sampler version of each
+    // descriptor set.
+    fvkutils::DescriptorSetMask prepareBindSets(LayoutArray const& layouts, SetArray const& sets);
 
     void removeDescriptorSet(fvkmemory::resource_ptr<VulkanDescriptorSet> set);
 
@@ -65,7 +66,10 @@ public:
             uint8_t bindingPoint, fvkmemory::resource_ptr<VulkanTexture> image,
             SamplerParams samplerParams);
 
-    void addExternallySampledTexture(fvkmemory::resource_ptr<VulkanTexture> image,
+    void clearTextureBinding(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+                             uint8_t bindingPoint);
+
+    void addExternallySampledTexture(fvkmemory::resource_ptr<VulkanTexture> external,
             Platform::ExternalImageHandleRef platformHandleRef);
 
     void removeExternallySampledTexture(fvkmemory::resource_ptr<VulkanTexture> image);
@@ -75,18 +79,20 @@ public:
     VkSamplerYcbcrConversion getVkSamplerYcbcrConversion(
             VulkanPlatform::ExternalImageMetadata const& metadata);
 
-private:
-
     struct ImageData {
-        fvkmemory::resource_ptr<VulkanTexture> ptr;
+        fvkmemory::resource_ptr<VulkanTexture> image;
         Platform::ExternalImageHandle platformHandle;
         bool hasBeenValidated = false; // indicates whether the image has been validated *this frame*
         VkSamplerYcbcrConversion conversion = VK_NULL_HANDLE;
     };
 
-    bool updateSetAndLayout(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+private:
+    bool hasExternalSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set);
+
+    void updateSetAndLayout(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
             fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
-    bool updateImage(ImageData* imageData);
+
+    void updateImage(ImageData* imageData);
 
     VulkanPlatform* mPlatform;
     VulkanSamplerCache* mSamplerCache;
@@ -102,10 +108,10 @@ private:
         fvkmemory::resource_ptr<VulkanTexture> image;
         fvkmemory::resource_ptr<VulkanDescriptorSet> set;
         SamplerParams samplerParams;
+        bool bound = false;
     };
 
     // Use vectors instead of hash maps because we only expect small number of entries.
-    std::vector<SetAndLayout> mSetAndLayouts;
     std::vector<SetBindingInfo> mSetBindings;
     std::vector<ImageData> mImages;
 };

filament/backend/src/vulkan/VulkanFboCache.cpp

@@ -330,25 +330,39 @@ VkRenderPass VulkanFboCache::getRenderPass(RenderPassKey const& config) noexcept
     mRenderPassCache[config] = {renderPass, mCurrentTime};
 
 #if FVK_ENABLED(FVK_DEBUG_FBO_CACHE)
-    FVK_LOGD << "Created render pass " << renderPass << " with "
-        << "samples = " << int(config.samples) << ", "
-        << "depth = " << (hasDepth ? 1 : 0) << ", "
-        << "colorAttachmentCount[0] = " << subpasses[0].colorAttachmentCount
-        << utils::io::endl;
-    #endif
+    FVK_LOGD << "Created render pass " << renderPass << " with ";
+    for (int i = 0; i < MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT; ++i) {
+        FVK_LOGD << (int) config.colorFormat[i] << " ";
+    }
+    FVK_LOGD << ", "
+             << "depth = " << config.depthFormat << ", "
+             << "initialDepthLayout = " << (int) config.initialDepthLayout << ", "
+             << "samples = " << int(config.samples) << ", "
+             << "needsResolveMask = " << int(config.needsResolveMask) << ", "
+             << "usesLazilyAllocatedMemory = " << int(config.usesLazilyAllocatedMemory) << ", "
+             << "viewCount = " << int(config.viewCount) << ", "
+             << "colorAttachmentCount[0] = " << subpasses[0].colorAttachmentCount
+             << utils::io::endl;
+#endif
 
     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

@@ -149,11 +149,18 @@ void VulkanDescriptorSet::acquire(fvkmemory::resource_ptr<VulkanBufferObject> ob
     mResources.push_back(obj);
 }
 
-VulkanDescriptorSetLayout::VulkanDescriptorSetLayout(DescriptorSetLayout const& layout)
+VulkanDescriptorSetLayout::VulkanDescriptorSetLayout(DescriptorSetLayout&& layout,
+        VkDescriptorSetLayout vkLayout)
     : bitmask(fromBackendLayout(layout)),
-      count(Count::fromLayoutBitmask(bitmask)) {}
+      count(Count::fromLayoutBitmask(bitmask)),
+      mVkLayout(vkLayout) {}
 
-PushConstantDescription::PushConstantDescription(backend::Program const& program) noexcept {
+VulkanDescriptorSetLayout::Bitmask VulkanDescriptorSetLayout::Bitmask::fromLayoutDescription(
+        DescriptorSetLayout const& layout) {
+    return fromBackendLayout(layout);
+}
+
+PushConstantDescription::PushConstantDescription(backend::Program const& program) {
     mRangeCount = 0;
     for (auto stage : { ShaderStage::VERTEX, ShaderStage::FRAGMENT, ShaderStage::COMPUTE }) {
         auto const& constants = program.getPushConstants(stage);

filament/backend/src/vulkan/VulkanHandles.h

@@ -71,7 +71,7 @@ struct VulkanDescriptorSetLayout : public HwDescriptorSetLayout, fvkmemory::Reso
         fvkutils::SamplerBitmask sampler;           // 8 bytes
         fvkutils::InputAttachmentBitmask inputAttachment; // 8 bytes
 
-        // This is a subset of the bitmask.sampler field.
+        // This is a subset of the sampler field.
         fvkutils::SamplerBitmask externalSampler; // 8 bytes
 
         bool operator==(Bitmask const& right) const {
@@ -79,6 +79,8 @@ struct VulkanDescriptorSetLayout : public HwDescriptorSetLayout, fvkmemory::Reso
                    inputAttachment == right.inputAttachment &&
                    externalSampler == right.externalSampler;
         }
+
+        static Bitmask fromLayoutDescription(DescriptorSetLayout const& layout);
     };
     static_assert(sizeof(Bitmask) == 40);
 
@@ -120,16 +122,20 @@ struct VulkanDescriptorSetLayout : public HwDescriptorSetLayout, fvkmemory::Reso
         }
     };
 
-    VulkanDescriptorSetLayout(DescriptorSetLayout const& layout);
+    VulkanDescriptorSetLayout(DescriptorSetLayout&& layout, VkDescriptorSetLayout vkLayout);
 
     // Note that we don't destroy the vklayout. This is done by the layout cache.
     ~VulkanDescriptorSetLayout() = default;
 
-    VkDescriptorSetLayout const& getVkLayout() const noexcept { return mVkLayout; }
+    VkDescriptorSetLayout 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; }
+    VkDescriptorSetLayout getExternalSamplerVkLayout() const noexcept {
+        return mExternalSamplerVkLayout;
+    }
+
+    void setExternalSamplerVkLayout(VkDescriptorSetLayout vklayout) noexcept {
+        mExternalSamplerVkLayout = vklayout;
+    }
 
     bool hasExternalSamplers() const noexcept { return bitmask.externalSampler.count() > 0; }
 
@@ -137,7 +143,11 @@ struct VulkanDescriptorSetLayout : public HwDescriptorSetLayout, fvkmemory::Reso
     Count const count;
 
 private:
-    VkDescriptorSetLayout mVkLayout = VK_NULL_HANDLE;
+    // This is the layout without any immutable samplers.
+    VkDescriptorSetLayout const mVkLayout = VK_NULL_HANDLE;
+
+    // This is the layout with immutable samplers, and can be updated.
+    VkDescriptorSetLayout mExternalSamplerVkLayout = VK_NULL_HANDLE;
 };
 
 struct VulkanDescriptorSet : public HwDescriptorSet, fvkmemory::Resource {
@@ -149,23 +159,37 @@ public:
     VulkanDescriptorSet(
             fvkutils::UniformBufferBitmask const& dynamicUboMask,
             uint8_t uniqueDynamicUboCount,
-            OnRecycle&& onRecycleFn)
+            OnRecycle&& onRecycleFn, VkDescriptorSet vkSet)
         : dynamicUboMask(dynamicUboMask),
           uniqueDynamicUboCount(uniqueDynamicUboCount),
+          mVkSet(vkSet),
           mOnRecycleFn(std::move(onRecycleFn)) {}
 
+    // NOLINTNEXTLINE(bugprone-exception-escape)
     ~VulkanDescriptorSet() {
         if (mOnRecycleFn) {
             mOnRecycleFn(this);
         }
+        if (mOnRecycleExternalSamplerFn) {
+            mOnRecycleExternalSamplerFn(this);
+        }
     }
 
-    VkDescriptorSet const& getVkSet() const noexcept {
+    VkDescriptorSet 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; }
+    VkDescriptorSet getExternalSamplerVkSet() const noexcept {
+        return mExternalSamplerVkSet;
+    }
+
+    void setExternalSamplerVkSet(VkDescriptorSet vkset, OnRecycle onRecycle) {
+        mExternalSamplerVkSet = vkset;
+        if (mOnRecycleExternalSamplerFn) {
+            mOnRecycleExternalSamplerFn(this);
+        }
+        mOnRecycleExternalSamplerFn = onRecycle;
+    }
 
     void setOffsets(backend::DescriptorSetOffsetArray&& offsets) noexcept {
         mOffsets = std::move(offsets);
@@ -182,17 +206,20 @@ public:
     uint8_t const uniqueDynamicUboCount;
 
 private:
-    VkDescriptorSet mVkSet = VK_NULL_HANDLE;
+    VkDescriptorSet const mVkSet;
+    VkDescriptorSet mExternalSamplerVkSet = VK_NULL_HANDLE;
+
     backend::DescriptorSetOffsetArray mOffsets;
     std::vector<fvkmemory::resource_ptr<fvkmemory::Resource>> mResources;
     OnRecycle mOnRecycleFn;
+    OnRecycle mOnRecycleExternalSamplerFn;
 };
 
 using PushConstantNameArray = utils::FixedCapacityVector<char const*>;
 using PushConstantNameByStage = std::array<PushConstantNameArray, Program::SHADER_TYPE_COUNT>;
 
 struct PushConstantDescription {
-    explicit PushConstantDescription(backend::Program const& program) noexcept;
+    explicit PushConstantDescription(backend::Program const& program);
 
     VkPushConstantRange const* getVkRanges() const noexcept { return mRanges; }
     uint32_t getVkRangeCount() const noexcept { return mRangeCount; }

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;
@@ -61,8 +66,12 @@ void VulkanPipelineCache::bindPipeline(VulkanCommandBuffer* commands) {
     // If an error occurred, allow higher levels to handle it gracefully.
     assert_invariant(cacheEntry != nullptr && "Failed to create/find pipeline");
 
-    mBoundPipeline = mPipelineRequirements;
-    vkCmdBindPipeline(cmdbuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, cacheEntry->handle);
+
+    static PipelineEqual equal;
+    if (!equal(mBoundPipeline, mPipelineRequirements)) {
+        mBoundPipeline = mPipelineRequirements;
+        vkCmdBindPipeline(cmdbuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, cacheEntry->handle);
+    }
 }
 
 VulkanPipelineCache::PipelineCacheEntry* VulkanPipelineCache::createPipeline() noexcept {
@@ -215,7 +224,7 @@ 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 +280,8 @@ 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/VulkanReadPixels.cpp

@@ -335,7 +335,7 @@ void VulkanReadPixels::run(fvkmemory::resource_ptr<VulkanRenderTarget> srcTarget
     mTaskHandler->post(std::move(waitFenceFunc), std::move(cleanPbdFunc));
 }
 
-void VulkanReadPixels::runUntilComplete() noexcept {
+void VulkanReadPixels::runUntilComplete() {
     if (!mTaskHandler) {
         return;
     }

filament/backend/src/vulkan/VulkanReadPixels.h

@@ -79,7 +79,7 @@ public:
             OnReadCompleteFunction const& readCompleteFunc);
 
     // This method will block until all of the in-flight requests are complete.
-    void runUntilComplete() noexcept;
+    void runUntilComplete();
 
 private:
     VkDevice mDevice = VK_NULL_HANDLE;

filament/backend/src/vulkan/VulkanSamplerCache.cpp

@@ -18,7 +18,6 @@
 
 #include "VulkanConstants.h"
 #include "vulkan/utils/Conversion.h"
-#include "vulkan/vulkan_core.h"
 
 #include <utils/Panic.h>
 
@@ -29,7 +28,7 @@ namespace filament::backend {
 VulkanSamplerCache::VulkanSamplerCache(VkDevice device)
     : mDevice(device) {}
 
-VkSampler VulkanSamplerCache::getSampler(Params params) noexcept {
+VkSampler VulkanSamplerCache::getSampler(Params params) {
     auto iter = mCache.find(params);
     if (UTILS_LIKELY(iter != mCache.end())) {
         return iter->second;

filament/backend/src/vulkan/VulkanSamplerCache.h

@@ -38,7 +38,7 @@ public:
     static_assert(sizeof(Params) == 16);
 
     explicit VulkanSamplerCache(VkDevice device);
-    VkSampler getSampler(Params params) noexcept;
+    VkSampler getSampler(Params params);
     void terminate() noexcept;
 private:
     VkDevice mDevice;

filament/backend/src/vulkan/VulkanStagePool.cpp

@@ -24,7 +24,7 @@
 
 #include <utils/Panic.h>
 
-static constexpr uint32_t TIME_BEFORE_EVICTION = FVK_MAX_COMMAND_BUFFERS;
+static constexpr uint32_t TIME_BEFORE_EVICTION = 3;
 
 namespace filament::backend {
 
@@ -39,7 +39,7 @@ VulkanStage const* VulkanStagePool::acquireStage(uint32_t numBytes) {
         auto stage = iter->second;
         mFreeStages.erase(iter);
         stage->lastAccessed = mCurrentFrame;
-        mUsedStages.insert(stage);
+        mUsedStages.push_back(stage);
         return stage;
     }
     // We were not able to find a sufficiently large stage, so create a new one.
@@ -51,7 +51,7 @@ VulkanStage const* VulkanStagePool::acquireStage(uint32_t numBytes) {
     });
 
     // Create the VkBuffer.
-    mUsedStages.insert(stage);
+    mUsedStages.push_back(stage);
     VkBufferCreateInfo bufferInfo {
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
         .size = numBytes,
@@ -77,7 +77,7 @@ VulkanStageImage const* VulkanStagePool::acquireImage(PixelDataFormat format, Pi
         if (image->format == vkformat && image->width == width && image->height == height) {
             mFreeImages.erase(image);
             image->lastAccessed = mCurrentFrame;
-            mUsedImages.insert(image);
+            mUsedImages.push_back(image);
             return image;
         }
     }
@@ -89,7 +89,7 @@ VulkanStageImage const* VulkanStagePool::acquireImage(PixelDataFormat format, Pi
         .lastAccessed = mCurrentFrame,
     });
 
-    mUsedImages.insert(image);
+    mUsedImages.push_back(image);
 
     const VkImageCreateInfo imageInfo = {
         .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
@@ -161,7 +161,7 @@ void VulkanStagePool::gc() noexcept {
             stage->lastAccessed = mCurrentFrame;
             mFreeStages.insert(std::make_pair(stage->capacity, stage));
         } else {
-            mUsedStages.insert(stage);
+            mUsedStages.push_back(stage);
         }
     }
 
@@ -185,7 +185,7 @@ void VulkanStagePool::gc() noexcept {
             image->lastAccessed = mCurrentFrame;
             mFreeImages.insert(image);
         } else {
-            mUsedImages.insert(image);
+            mUsedImages.push_back(image);
         }
     }
     FVK_SYSTRACE_END();

filament/backend/src/vulkan/VulkanStagePool.h

@@ -22,6 +22,7 @@
 
 #include <map>
 #include <unordered_set>
+#include <vector>
 
 namespace filament::backend {
 
@@ -73,10 +74,10 @@ private:
     std::multimap<uint32_t, VulkanStage const*> mFreeStages;
 
     // Simple unordered set for stashing a list of in-use stages that can be reclaimed later.
-    std::unordered_set<VulkanStage const*> mUsedStages;
+    std::vector<VulkanStage const*> mUsedStages;
 
     std::unordered_set<VulkanStageImage const*> mFreeImages;
-    std::unordered_set<VulkanStageImage const*> mUsedImages;
+    std::vector<VulkanStageImage const*> mUsedImages;
 
     // Store the current "time" (really just a frame count) and LRU eviction parameters.
     uint64_t mCurrentFrame = 0;

filament/backend/src/vulkan/VulkanTexture.cpp

@@ -223,8 +223,7 @@ VkImageUsageFlags getUsage(VulkanContext const& context, uint8_t samples,
 VulkanTextureState::VulkanTextureState(VulkanStagePool& stagePool, VulkanCommands* commands,
         VmaAllocator allocator, VkDevice device, VkImage image, VkDeviceMemory deviceMemory,
         VkFormat format, VkImageViewType viewType, uint8_t levels, uint8_t layerCount,
-        VkSamplerYcbcrConversion ycbcrConversion, bool isExternalFormat, VkImageUsageFlags usage,
-        bool isProtected)
+        VkSamplerYcbcrConversion ycbcrConversion, VkImageUsageFlags usage, bool isProtected)
     : mStagePool(stagePool),
       mCommands(commands),
       mAllocator(allocator),
@@ -234,17 +233,17 @@ VulkanTextureState::VulkanTextureState(VulkanStagePool& stagePool, VulkanCommand
       mVkFormat(format),
       mViewType(viewType),
       mFullViewRange{ fvkutils::getImageAspect(format), 0, levels, 0, layerCount },
-      mYcbcr{ ycbcrConversion, isExternalFormat },
+      mYcbcr{ ycbcrConversion },
       mDefaultLayout(getDefaultLayoutImpl(usage)),
       mUsage(usage),
       mIsProtected(isProtected) {}
 
 VulkanTextureState::~VulkanTextureState() {
+    clearCachedImageViews();
     if (mTextureImageMemory != VK_NULL_HANDLE) {
         vkDestroyImage(mDevice, mTextureImage, VKALLOC);
         vkFreeMemory(mDevice, mTextureImageMemory, VKALLOC);
     }
-    clearCachedImageViews();
 }
 
 void VulkanTextureState::clearCachedImageViews() noexcept {
@@ -272,7 +271,7 @@ VkImageView VulkanTextureState::getImageView(VkImageSubresourceRange range, VkIm
         .flags = 0,
         .image = mTextureImage,
         .viewType = viewType,
-        .format = mYcbcr.isExternalFormat ? VK_FORMAT_UNDEFINED : mVkFormat,
+        .format = mYcbcr.conversion != VK_NULL_HANDLE ? VK_FORMAT_UNDEFINED : mVkFormat,
         .components = swizzle,
         .subresourceRange = range,
     };
@@ -294,7 +293,6 @@ VulkanTexture::VulkanTexture(VulkanContext const& context, VkDevice device, VmaA
               commands, allocator, device, image, memory, format,
               fvkutils::getViewType(SamplerType::SAMPLER_2D),
               /*mipLevels=*/1, getLayerCountFromDepth(depth), conversion,
-              /*isExternalFormat=*/false,
               getUsage(context, samples, VK_NULL_HANDLE, format, tusage),
               any(usage & TextureUsage::PROTECTED))) {
     mPrimaryViewRange = mState->mFullViewRange;
@@ -424,8 +422,7 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
     mState = fvkmemory::resource_ptr<VulkanTextureState>::construct(resourceManager, stagePool,
             commands, allocator, device, textureImage, textureImageMemory, vkFormat,
             fvkutils::getViewType(target), levels, getLayerCount(target, depth),
-            VK_NULL_HANDLE /* ycbcrConversion */, false /*isExternalFormat*/, imageInfo.usage,
-            isProtected);
+            VK_NULL_HANDLE /* ycbcrConversion */, imageInfo.usage, isProtected);
 
     // Spec out the "primary" VkImageView that shaders use to sample from the image.
     mPrimaryViewRange = mState->mFullViewRange;
@@ -764,15 +761,13 @@ void VulkanTexture::setLayout(VkImageSubresourceRange const& range, VulkanLayout
     }
 }
 
-void VulkanTexture::setYcbcrConversion(VkSamplerYcbcrConversion conversion, bool isExternalFormat) {
+void VulkanTexture::setYcbcrConversion(VkSamplerYcbcrConversion conversion) {
     // Note that this comparison is valid because we only ever create VkSamplerYcbcrConversion from
     // a cache.  So for each set of parameters, there is exactly one conversion (similar to
     // samplers).
     VulkanTextureState::Ycbcr ycbcr = {
         .conversion = conversion,
-        .isExternalFormat = isExternalFormat,
     };
-
     if (mState->mYcbcr != ycbcr) {
         mState->mYcbcr = ycbcr;
         mState->clearCachedImageViews();

filament/backend/src/vulkan/VulkanTexture.h

@@ -40,8 +40,7 @@ struct VulkanTextureState : public fvkmemory::Resource {
     VulkanTextureState(VulkanStagePool& stagePool, VulkanCommands* commands, VmaAllocator allocator,
             VkDevice device, VkImage image, VkDeviceMemory deviceMemory, VkFormat format,
             VkImageViewType viewType, uint8_t levels, uint8_t layerCount,
-            VkSamplerYcbcrConversion ycbcrConversion, bool isExternalFormat,
-            VkImageUsageFlags usage, bool isProtected);
+            VkSamplerYcbcrConversion ycbcrConversion, VkImageUsageFlags usage, bool isProtected);
 
     ~VulkanTextureState();
 
@@ -86,10 +85,9 @@ private:
     // conversion matrix per-frame.
     struct Ycbcr {
         VkSamplerYcbcrConversion conversion;
-        bool isExternalFormat;
 
         bool operator==(Ycbcr const& other) const {
-            return conversion == other.conversion && isExternalFormat == other.isExternalFormat;
+            return conversion == other.conversion;
         }
 
         bool operator!=(Ycbcr const& other) const {
@@ -209,7 +207,7 @@ struct VulkanTexture : public HwTexture, fvkmemory::Resource {
     // This is used in the case of external images and external samplers. AHB might update the
     // conversion per-frame. This implies that we need to invalidate the view cache when that
     // happens.
-    void setYcbcrConversion(VkSamplerYcbcrConversion conversion, bool isExternal);
+    void setYcbcrConversion(VkSamplerYcbcrConversion conversion);
 
 #if FVK_ENABLED(FVK_DEBUG_TEXTURE)
     void print() const;

filament/backend/src/vulkan/VulkanYcbcrConversionCache.cpp

@@ -33,7 +33,7 @@ VulkanYcbcrConversionCache::VulkanYcbcrConversionCache(VkDevice device)
     : mDevice(device) {}
 
 VkSamplerYcbcrConversion VulkanYcbcrConversionCache::getConversion(
-        VulkanYcbcrConversionCache::Params params) noexcept {
+        VulkanYcbcrConversionCache::Params params) {
     auto iter = mCache.find(params);
     if (UTILS_LIKELY(iter != mCache.end())) {
         return iter->second;
@@ -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 = fvkutils::getVkFormat(params.format),
+        .format = params.format,
         .ycbcrModel = fvkutils::getYcbcrModelConversion(chroma.ycbcrModel),
         .ycbcrRange = fvkutils::getYcbcrRange(chroma.ycbcrRange),
         .components = fvkutils::getSwizzleMap(swizzleArray),

filament/backend/src/vulkan/VulkanYcbcrConversionCache.h

@@ -30,15 +30,14 @@ namespace filament::backend {
 class VulkanYcbcrConversionCache {
 public:
     struct Params {
-        SamplerYcbcrConversion conversion = {};
-        TextureFormat format = {};
-        uint16_t padding = 0;
-        uint64_t externalFormat = 0;
+        SamplerYcbcrConversion conversion = {}; // 4
+        VkFormat format;                        // 4
+        uint64_t externalFormat = 0;            // 8
     };
     static_assert(sizeof(Params) == 16);
 
     explicit VulkanYcbcrConversionCache(VkDevice device);
-    VkSamplerYcbcrConversion getConversion(Params params) noexcept;
+    VkSamplerYcbcrConversion getConversion(Params params);
     void terminate() noexcept;
 
 private:
@@ -48,7 +47,8 @@ private:
         bool operator()(Params lhs, Params rhs) const noexcept {
             SamplerYcbcrConversion::EqualTo equal;
             return equal(lhs.conversion, rhs.conversion) &&
-                   lhs.externalFormat == rhs.externalFormat;
+                   lhs.externalFormat == rhs.externalFormat &&
+                   lhs.format == rhs.format;
         }
     };
     using ConversionHashFn = utils::hash::MurmurHashFn<Params>;

filament/backend/src/vulkan/memory/ResourceManager.cpp

@@ -31,6 +31,8 @@ ResourceManager::ResourceManager(size_t arenaSize, bool disableUseAfterFreeCheck
     : mHandleAllocatorImpl("Handles", arenaSize, disableUseAfterFreeCheck, disablePoolHandleTags) {}
 
 void ResourceManager::gc() noexcept {
+    FVK_SYSTRACE_CONTEXT();
+    FVK_SYSTRACE_START("ResourceManager::gc");
     auto destroyAll = [this](GcList& list) {
         for (auto const& [type, id]: list) {
             destroyWithType(type, id);
@@ -49,6 +51,7 @@ void ResourceManager::gc() noexcept {
     GcList gcs;
     std::swap(gcs, mGcList);
     destroyAll(gcs);
+    FVK_SYSTRACE_END();
 }
 
 void ResourceManager::terminate() noexcept {

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

@@ -212,6 +212,8 @@ ExtensionSet getDeviceExtensions(VkPhysicalDevice device) {
         VK_KHR_SAMPLER_YCBCR_CONVERSION_EXTENSION_NAME,
         VK_ANDROID_EXTERNAL_MEMORY_ANDROID_HARDWARE_BUFFER_EXTENSION_NAME,
         VK_EXT_QUEUE_FAMILY_FOREIGN_EXTENSION_NAME,
+        // This is needed for external images.  See VulkanPlatformAndroid
+        VK_KHR_IMAGE_FORMAT_LIST_EXTENSION_NAME,
 #endif
         // MoltenVk is the only non-conformant implementation we're interested in.
 #if defined(__APPLE__)

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

@@ -17,12 +17,12 @@
 
 #include "vulkan/VulkanConstants.h"
 #include "vulkan/VulkanContext.h"
-#include "vulkan/vulkan_core.h"
 
 #include <backend/DriverEnums.h>
 #include <private/backend/BackendUtilsAndroid.h>
 
 #include <utils/Panic.h>
+#include "vulkan/utils/Image.h"
 
 #include <bluevk/BlueVK.h>
 
@@ -39,6 +39,14 @@ namespace {
 
 VkFormat transformVkFormat(VkFormat format, bool sRGB) {
     if (!sRGB) {
+        switch (format) {
+            case VK_FORMAT_R8G8B8A8_SRGB:
+               return VK_FORMAT_R8G8B8A8_UNORM;
+            case VK_FORMAT_R8G8B8_SRGB:
+                return VK_FORMAT_R8G8B8_UNORM;
+            default:
+                break;
+        }
         return format;
     }
 
@@ -147,7 +155,6 @@ std::pair<TextureFormat, TextureUsage> getFilamentFormatAndUsage(const AHardware
     };
 }
 
-
 }// namespace
 
 VulkanPlatformAndroid::ExternalImageVulkanAndroid::~ExternalImageVulkanAndroid() {
@@ -220,7 +227,6 @@ VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::extractExternalImag
     }
     metadata.samples = VK_SAMPLE_COUNT_1_BIT;
 
-
     VkAndroidHardwareBufferFormatPropertiesANDROID formatInfo = {
         .sType = VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_FORMAT_PROPERTIES_ANDROID,
     };
@@ -236,13 +242,16 @@ VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::extractExternalImag
     FILAMENT_CHECK_POSTCONDITION(metadata.format == bufferPropertiesFormat)
             << "mismatched image format( " << metadata.format << ") and queried format("
             << bufferPropertiesFormat << ") for external image (AHB)";
-    metadata.externalFormat = formatInfo.externalFormat;
 
-    // Choose either externalFormat > 0 or metadata.format and prefer the latter.
-    if (metadata.externalFormat > 0 &&  metadata.format != VK_FORMAT_UNDEFINED) {
-        // See VUID-VkImageCreateInfo-pNext-09457
-        metadata.usage = VK_IMAGE_USAGE_SAMPLED_BIT;
-        metadata.externalFormat = 0;
+    bool const requiresConversion =
+        metadata.format == VK_FORMAT_UNDEFINED ||
+        fvkutils::isVKYcbcrConversionFormat(metadata.format);
+    if (requiresConversion) {
+      metadata.format = VK_FORMAT_UNDEFINED;
+      metadata.usage = VK_IMAGE_USAGE_SAMPLED_BIT;
+      metadata.externalFormat = formatInfo.externalFormat;
+    } else {
+      metadata.externalFormat = 0;
     }
 
     metadata.allocationSize = properties.allocationSize;
@@ -257,91 +266,127 @@ VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::extractExternalImag
     return metadata;
 }
 
-VulkanPlatformAndroid::ImageData VulkanPlatformAndroid::createVkImageFromExternal(
+VulkanPlatform::ImageData VulkanPlatformAndroid::createVkImageFromExternal(
         ExternalImageHandleRef externalImage) const {
-    auto const& metadata = extractExternalImageMetadata(externalImage);
+    auto metadata = extractExternalImageMetadata(externalImage);
 
     auto const* fvkExternalImage =
             static_cast<ExternalImageVulkanAndroid const*>(externalImage.get());
     AHardwareBuffer* buffer = fvkExternalImage->aHardwareBuffer;
 
-    // if external format we need to specifiy it in the allocation
-    bool const useExternalFormat = metadata.format == VK_FORMAT_UNDEFINED;
-
-    VkExternalFormatANDROID const externalFormat = {
-        .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
-        .pNext = nullptr,
-        .externalFormat = metadata.externalFormat,
-    };
-    VkExternalMemoryImageCreateInfo const externalCreateInfo = {
-        .sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO,
-        .pNext = useExternalFormat ? &externalFormat : nullptr,
-        .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID,
-    };
-
-    VkImageCreateInfo const imageInfo{
-        .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
-        .pNext = &externalCreateInfo,
-        .flags = useExternalFormat ? VK_IMAGE_CREATE_ALIAS_BIT : 0u,
-        .imageType = VK_IMAGE_TYPE_2D,
-        .format = metadata.format,
-        .extent = {
-            metadata.width,
-            metadata.height,
-            1u,
-        },
-        .mipLevels = 1,
-        .arrayLayers = metadata.layers,
-        .samples = metadata.samples,
-        .usage = metadata.usage,
-    };
-
     VkDevice const device = getDevice();
+    VkPhysicalDevice const physicalDevice = getPhysicalDevice();
+    auto buildImage = [&](ExternalImageMetadata const& metadata) {
+        bool const isExternal = metadata.externalFormat != 0;
+        VkExternalFormatANDROID const externalFormat = {
+            .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
+            .pNext = nullptr,
+            .externalFormat = metadata.externalFormat,
+        };
+        VkExternalMemoryImageCreateInfo externalCreateInfo = {
+            .sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO,
+            .pNext = isExternal ? &externalFormat : nullptr,
+            .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID,
+        };
 
-    VkImage image;
-    VkResult result = vkCreateImage(device, &imageInfo, VKALLOC, &image);
-    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+        VkFormat formats[2] = {};
+        VkImageFormatListCreateInfo imageFormatListInfo = {
+            .sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_LIST_CREATE_INFO,
+            .pNext = nullptr,
+            .viewFormatCount = 2,
+            .pViewFormats = formats,
+        };
+
+        if (fvkExternalImage->sRGB) {
+            formats[0] = metadata.format;
+            formats[1] = transformVkFormat(metadata.format, /*sRGB=*/false);
+            imageFormatListInfo.pNext = externalCreateInfo.pNext;
+            externalCreateInfo.pNext = &imageFormatListInfo;
+        }
+
+        VkImageCreateInfo const imageInfo = {
+            .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
+            .pNext = &externalCreateInfo,
+            .flags = fvkExternalImage->sRGB ? VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT : 0u,
+            .imageType = VK_IMAGE_TYPE_2D,
+            // For non external images, use the same format as the AHB, which isn't in SRGB
+            // Fix VUID-VkMemoryAllocateInfo-pNext-02387
+            .format = transformVkFormat(metadata.format, /*sRGB=*/false),
+            .extent = {
+                metadata.width,
+                metadata.height,
+                1u,
+            },
+            .mipLevels = 1,
+            .arrayLayers = metadata.layers,
+            .samples = metadata.samples,
+            .usage = metadata.usage,
+        };
+        VkImage image;
+        VkResult result = vkCreateImage(device, &imageInfo, VKALLOC, &image);
+        FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
             << "vkCreateImage failed with error=" << static_cast<int32_t>(result);
-
-    // Allocate the memory
-    VkImportAndroidHardwareBufferInfoANDROID const androidHardwareBufferInfo = {
-        .sType = VK_STRUCTURE_TYPE_IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID,
-        .buffer = buffer,
-    };
-    VkMemoryDedicatedAllocateInfo const memoryDedicatedAllocateInfo = {
-        .sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO,
-        .pNext = &androidHardwareBufferInfo,
-        .image = image,
-        .buffer = VK_NULL_HANDLE,
+        return image;
     };
 
-    VkPhysicalDeviceMemoryProperties memoryProperties;
-    vkGetPhysicalDeviceMemoryProperties(getPhysicalDevice(), &memoryProperties);
+    auto allocMem = [&](VkImage image, ExternalImageMetadata const& metadata) {
+        bool const isExternal = metadata.externalFormat != 0;
+        // Allocate the memory
+        VkImportAndroidHardwareBufferInfoANDROID const androidHardwareBufferInfo = {
+            .sType = VK_STRUCTURE_TYPE_IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID,
+            .buffer = buffer,
+        };
+        VkMemoryDedicatedAllocateInfo const memoryDedicatedAllocateInfo = {
+            .sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO,
+            .pNext = &androidHardwareBufferInfo,
+            .image = image,
+            .buffer = VK_NULL_HANDLE,
+        };
+        VkPhysicalDeviceMemoryProperties memoryProperties;
+        vkGetPhysicalDeviceMemoryProperties(physicalDevice, &memoryProperties);
+        VkMemoryPropertyFlags requiredMemoryFlags =
+                !isExternal && any(metadata.filamentUsage & TextureUsage::UPLOADABLE)
+                        ? VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
+                        : VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
 
+        if (any(metadata.filamentUsage & TextureUsage::PROTECTED)) {
+            requiredMemoryFlags |= VK_MEMORY_PROPERTY_PROTECTED_BIT;
+        }
 
-    VkMemoryPropertyFlags const requiredMemoryFlags =
-            any(metadata.filamentUsage & TextureUsage::UPLOADABLE)
-                    ? VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
-                    : VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+        uint32_t const memoryTypeIndex = VulkanContext::selectMemoryType(memoryProperties,
+                metadata.memoryTypeBits, requiredMemoryFlags);
 
-    uint32_t const memoryTypeIndex = VulkanContext::selectMemoryType(memoryProperties,
-            metadata.memoryTypeBits, requiredMemoryFlags);
-
-    VkMemoryAllocateInfo const allocInfo = {
-        .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
-        .pNext = &memoryDedicatedAllocateInfo,
-        .allocationSize = metadata.allocationSize,
-        .memoryTypeIndex = memoryTypeIndex,
+        VkMemoryAllocateInfo const allocInfo = {
+            .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
+            .pNext = &memoryDedicatedAllocateInfo,
+            .allocationSize = metadata.allocationSize,
+            .memoryTypeIndex = memoryTypeIndex,
+        };
+        VkDeviceMemory memory;
+        VkResult result = vkAllocateMemory(device, &allocInfo, VKALLOC, &memory);
+        FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+                << "vkAllocateMemory failed with error=" << static_cast<int32_t>(result);
+        result = vkBindImageMemory(getDevice(), image, memory, 0);
+        FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+                << "vkBindImageMemory error=" << static_cast<int32_t>(result);
+        return memory;
     };
-    VkDeviceMemory memory;
-    result = vkAllocateMemory(device, &allocInfo, VKALLOC, &memory);
-    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-            << "vkAllocateMemory failed with error=" << static_cast<int32_t>(result);
 
-    result = vkBindImageMemory(getDevice(), image, memory, 0);
-    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-            << "vkBindImageMemory error=" << static_cast<int32_t>(result);
-    return { image, memory };
+    VulkanPlatform::ImageData::Bundle internal = {}, external = {};
+    auto img = buildImage(metadata);
+    auto mem = allocMem(img, metadata);
+
+    // Note that we're always choosing a non-externally sampled format if it exists.
+    if (metadata.externalFormat == 0) {
+        internal = { img, mem };
+    } else {
+        external = { img, mem };
+    }
+
+    return {
+        .internal = internal,
+        .external = external,
+    };
 }
 
 VulkanPlatform::ExtensionSet VulkanPlatformAndroid::getSwapchainInstanceExtensions() const {
@@ -355,7 +400,7 @@ VulkanPlatform::SurfaceBundle VulkanPlatformAndroid::createVkSurfaceKHR(void* na
     VkSurfaceKHR surface;
     VkExtent2D extent;
 
-    VkAndroidSurfaceCreateInfoKHR const createInfo{
+    VkAndroidSurfaceCreateInfoKHR const createInfo = {
         .sType = VK_STRUCTURE_TYPE_ANDROID_SURFACE_CREATE_INFO_KHR,
         .window = (ANativeWindow*) nativeWindow,
     };

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;
 }
@@ -66,36 +46,20 @@ VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl(
 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

@@ -51,6 +51,8 @@ VkFormat getVkFormat(ElementType type, bool normalized, bool integer) {
                 return VK_FORMAT_UNDEFINED;
         }
     }
+
+    // Non-normalized case
     switch (type) {
         // Single Component Types
         case ElementType::BYTE: return integer ? VK_FORMAT_R8_SINT : VK_FORMAT_R8_SSCALED;

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

@@ -184,6 +184,46 @@ bool isVkStencilFormat(VkFormat format) {
     return (getImageAspect(format) & VK_IMAGE_ASPECT_STENCIL_BIT) != 0;
 }
 
+bool isVKYcbcrConversionFormat(VkFormat format) {
+    switch (format) {
+        case VK_FORMAT_G8B8G8R8_422_UNORM:
+        case VK_FORMAT_B8G8R8G8_422_UNORM:
+        case VK_FORMAT_G8_B8_R8_3PLANE_420_UNORM:
+        case VK_FORMAT_G8_B8R8_2PLANE_420_UNORM:
+        case VK_FORMAT_G8_B8_R8_3PLANE_422_UNORM:
+        case VK_FORMAT_G8_B8R8_2PLANE_422_UNORM:
+        case VK_FORMAT_G8_B8_R8_3PLANE_444_UNORM:
+        case VK_FORMAT_G10X6B10X6G10X6R10X6_422_UNORM_4PACK16:
+        case VK_FORMAT_B10X6G10X6R10X6G10X6_422_UNORM_4PACK16:
+        case VK_FORMAT_G10X6_B10X6_R10X6_3PLANE_420_UNORM_3PACK16:
+        case VK_FORMAT_G10X6_B10X6R10X6_2PLANE_420_UNORM_3PACK16:
+        case VK_FORMAT_G10X6_B10X6_R10X6_3PLANE_422_UNORM_3PACK16:
+        case VK_FORMAT_G10X6_B10X6R10X6_2PLANE_422_UNORM_3PACK16:
+        case VK_FORMAT_G10X6_B10X6_R10X6_3PLANE_444_UNORM_3PACK16:
+        case VK_FORMAT_G12X4B12X4G12X4R12X4_422_UNORM_4PACK16:
+        case VK_FORMAT_B12X4G12X4R12X4G12X4_422_UNORM_4PACK16:
+        case VK_FORMAT_G12X4_B12X4_R12X4_3PLANE_420_UNORM_3PACK16:
+        case VK_FORMAT_G12X4_B12X4R12X4_2PLANE_420_UNORM_3PACK16:
+        case VK_FORMAT_G12X4_B12X4_R12X4_3PLANE_422_UNORM_3PACK16:
+        case VK_FORMAT_G12X4_B12X4R12X4_2PLANE_422_UNORM_3PACK16:
+        case VK_FORMAT_G12X4_B12X4_R12X4_3PLANE_444_UNORM_3PACK16:
+        case VK_FORMAT_G16B16G16R16_422_UNORM:
+        case VK_FORMAT_B16G16R16G16_422_UNORM:
+        case VK_FORMAT_G16_B16_R16_3PLANE_420_UNORM:
+        case VK_FORMAT_G16_B16R16_2PLANE_420_UNORM:
+        case VK_FORMAT_G16_B16_R16_3PLANE_422_UNORM:
+        case VK_FORMAT_G16_B16R16_2PLANE_422_UNORM:
+        case VK_FORMAT_G16_B16_R16_3PLANE_444_UNORM:
+        case VK_FORMAT_G8_B8R8_2PLANE_444_UNORM:
+        case VK_FORMAT_G10X6_B10X6R10X6_2PLANE_444_UNORM_3PACK16:
+        case VK_FORMAT_G12X4_B12X4R12X4_2PLANE_444_UNORM_3PACK16:
+        case VK_FORMAT_G16_B16R16_2PLANE_444_UNORM:
+            return true;
+        default:
+            return false;
+    }
+}
+
 static uint32_t mostSignificantBit(uint32_t x) { return 1ul << (31ul - utils::clz(x)); }
 
 uint8_t reduceSampleCount(uint8_t sampleCount, VkSampleCountFlags mask) {

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

@@ -98,6 +98,8 @@ bool isVkDepthFormat(VkFormat format);
 
 bool isVkStencilFormat(VkFormat format);
 
+bool isVKYcbcrConversionFormat(VkFormat format);
+
 VkImageAspectFlags getImageAspect(VkFormat format);
 
 uint8_t reduceSampleCount(uint8_t sampleCount, VkSampleCountFlags mask);

filament/backend/src/webgpu/WGPUProgram.cpp

@@ -0,0 +1,159 @@
+/*
+ * 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;
+
+    wgpu::Instance instance = device.GetAdapter().GetInstance();
+    instance.WaitAny(
+            module.GetCompilationInfo(wgpu::CallbackMode::WaitAnyOnly,
+                    [&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;
+                    }),
+            UINT16_MAX);
+    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

@@ -16,7 +16,8 @@
 
 #include "webgpu/WebGPUDriver.h"
 
-#include "webgpu/WebGPUConstants.h"
+#include "WebGPUPipelineCreation.h"
+#include "WebGPUSwapChain.h"
 #include <backend/platforms/WebGPUPlatform.h>
 
 #include "CommandStreamDispatcher.h"
@@ -24,18 +25,20 @@
 #include "private/backend/Dispatcher.h"
 #include <backend/DriverEnums.h>
 #include <backend/Handle.h>
+#include <backend/TargetBufferInfo.h>
 
 #include <math/mat3.h>
 #include <utils/CString.h>
+#include <utils/Panic.h>
 #include <utils/ostream.h>
 
 #include <dawn/webgpu_cpp_print.h>
 #include <webgpu/webgpu_cpp.h>
 
 #include <algorithm>
+#include <array>
 #include <cstddef>
 #include <cstdint>
-#include <memory>
 #include <sstream>
 #include <string_view>
 #include <utility>
@@ -183,8 +186,6 @@ void printAdapterDetails(wgpu::Adapter const& adapter) {
 }
 #endif
 
-
-
 #if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
 void printDeviceDetails(wgpu::Device const& device) {
     wgpu::SupportedFeatures supportedFeatures{};
@@ -228,6 +229,15 @@ WebGPUDriver::WebGPUDriver(WebGPUPlatform& platform, const Platform::DriverConfi
 #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
+    mQueue = mDevice.GetQueue();
 }
 
 WebGPUDriver::~WebGPUDriver() noexcept = default;
@@ -256,6 +266,7 @@ void WebGPUDriver::terminate() {
 }
 
 void WebGPUDriver::tick(int) {
+    mDevice.Tick();
 }
 
 void WebGPUDriver::beginFrame(int64_t monotonic_clock_ns,
@@ -285,37 +296,52 @@ 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) {
+    if (boh) {
+        destructHandle<WGPUBufferObject>(boh);
+    }
 }
 
 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) {
@@ -325,16 +351,19 @@ void WebGPUDriver::destroyTimerQuery(Handle<HwTimerQuery> tqh) {
 }
 
 void WebGPUDriver::destroyDescriptorSetLayout(Handle<HwDescriptorSetLayout> tqh) {
+    if (tqh) {
+        destructHandle<WebGPUDescriptorSetLayout>(tqh);
+    }
 }
 
 void WebGPUDriver::destroyDescriptorSet(Handle<HwDescriptorSet> tqh) {
+    if (tqh) {
+        destructHandle<WebGPUDescriptorSet>(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 {
@@ -342,15 +371,13 @@ Handle<HwSwapChain> WebGPUDriver::createSwapChainHeadlessS() noexcept {
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureS() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUTexture>();
 }
 
-Handle<HwTexture> WebGPUDriver::importTextureS() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
-}
+Handle<HwTexture> WebGPUDriver::importTextureS() noexcept { return allocHandle<WGPUTexture>(); }
 
 Handle<HwProgram> WebGPUDriver::createProgramS() noexcept {
-    return Handle<HwProgram>((Handle<HwProgram>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUProgram>();
 }
 
 Handle<HwFence> WebGPUDriver::createFenceS() noexcept {
@@ -366,74 +393,66 @@ Handle<HwIndexBuffer> WebGPUDriver::createIndexBufferS() noexcept {
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureViewS() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUTexture>();
 }
 
 Handle<HwBufferObject> WebGPUDriver::createBufferObjectS() noexcept {
-    return Handle<HwBufferObject>((Handle<HwBufferObject>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUBufferObject>();
 }
 
 Handle<HwRenderTarget> WebGPUDriver::createRenderTargetS() noexcept {
-    return Handle<HwRenderTarget>((Handle<HwRenderTarget>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPURenderTarget>();
 }
 
 Handle<HwVertexBuffer> WebGPUDriver::createVertexBufferS() noexcept {
-    return Handle<HwVertexBuffer>((Handle<HwVertexBuffer>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUVertexBuffer>();
 }
 
 Handle<HwDescriptorSet> WebGPUDriver::createDescriptorSetS() noexcept {
-    return Handle<HwDescriptorSet>((Handle<HwDescriptorSet>::HandleId) mNextFakeHandle++);
+    return allocHandle<WebGPUDescriptorSet>();
 }
 
 Handle<HwRenderPrimitive> WebGPUDriver::createRenderPrimitiveS() noexcept {
-    return Handle<HwRenderPrimitive>((Handle<HwRenderPrimitive>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPURenderPrimitive>();
 }
 
 Handle<HwVertexBufferInfo> WebGPUDriver::createVertexBufferInfoS() noexcept {
-    return Handle<HwVertexBufferInfo>((Handle<HwVertexBufferInfo>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUVertexBufferInfo>();
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureViewSwizzleS() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUTexture>();
 }
 
 Handle<HwRenderTarget> WebGPUDriver::createDefaultRenderTargetS() noexcept {
-    return Handle<HwRenderTarget>((Handle<HwRenderTarget>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPURenderTarget>();
 }
 
 Handle<HwDescriptorSetLayout> WebGPUDriver::createDescriptorSetLayoutS() noexcept {
-    return Handle<HwDescriptorSetLayout>(
-            (Handle<HwDescriptorSetLayout>::HandleId) mNextFakeHandle++);
+    return allocHandle<WebGPUDescriptorSetLayout>();
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureExternalImageS() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUTexture>();
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureExternalImage2S() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUTexture>();
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureExternalImagePlaneS() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUTexture>();
 }
 
 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);
-    mSwapChain = nullptr;
+    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);
+
+    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."
@@ -446,59 +465,98 @@ 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,
         uint32_t height, uint64_t flags) {}
 
 void WebGPUDriver::createVertexBufferInfoR(Handle<HwVertexBufferInfo> vbih, uint8_t bufferCount,
-        uint8_t attributeCount, AttributeArray attributes) {}
+        uint8_t attributeCount, AttributeArray attributes) {
+    constructHandle<WGPUVertexBufferInfo>(vbih, bufferCount, attributeCount, attributes);
+}
 
 void WebGPUDriver::createVertexBufferR(Handle<HwVertexBuffer> vbh, uint32_t vertexCount,
-        Handle<HwVertexBufferInfo> vbih) {}
+        Handle<HwVertexBufferInfo> vbih) {
+    auto* vertexBufferInfo = handleCast<WGPUVertexBufferInfo>(vbih);
+    constructHandle<WGPUVertexBuffer>(vbh, mDevice, vertexCount, vertexBufferInfo->bufferCount,
+            vbih);
+}
 
 void WebGPUDriver::createIndexBufferR(Handle<HwIndexBuffer> ibh, ElementType elementType,
-        uint32_t indexCount, BufferUsage usage) {}
+        uint32_t indexCount, BufferUsage usage) {
+    auto elementSize = static_cast<uint8_t>(getElementTypeSize(elementType));
+    constructHandle<WGPUIndexBuffer>(ibh, mDevice, elementSize, indexCount);
+}
 
 void WebGPUDriver::createBufferObjectR(Handle<HwBufferObject> boh, uint32_t byteCount,
-        BufferObjectBinding bindingType, BufferUsage usage) {}
+        BufferObjectBinding bindingType, BufferUsage usage) {
+    constructHandle<WGPUBufferObject>(boh, mDevice, bindingType, byteCount);
+}
 
 void WebGPUDriver::createTextureR(Handle<HwTexture> th, SamplerType target, uint8_t levels,
         TextureFormat format, uint8_t samples, uint32_t w, uint32_t h, uint32_t depth,
-        TextureUsage usage) {}
+        TextureUsage usage) {
+    constructHandle<WGPUTexture>(th, target, levels, format, samples, w, h, depth, usage, mDevice);
+}
 
 void WebGPUDriver::createTextureViewR(Handle<HwTexture> th, Handle<HwTexture> srch,
-        uint8_t baseLevel, uint8_t levelCount) {}
+        uint8_t baseLevel, uint8_t levelCount) {
+    auto source = handleCast<WGPUTexture>(srch);
+
+    constructHandle<WGPUTexture>(th, source, baseLevel, levelCount);
+}
 
 void WebGPUDriver::createTextureViewSwizzleR(Handle<HwTexture> th, Handle<HwTexture> srch,
         backend::TextureSwizzle r, backend::TextureSwizzle g, backend::TextureSwizzle b,
-        backend::TextureSwizzle a) {}
+        backend::TextureSwizzle a) {
+    PANIC_POSTCONDITION("Swizzle WebGPU Texture is not supported");
+}
 
 void WebGPUDriver::createTextureExternalImage2R(Handle<HwTexture> th, backend::SamplerType target,
         backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
-        Platform::ExternalImageHandleRef externalImage) {}
+        Platform::ExternalImageHandleRef externalImage) {
+    PANIC_POSTCONDITION("External WebGPU Texture is not supported");
+}
 
 void WebGPUDriver::createTextureExternalImageR(Handle<HwTexture> th, backend::SamplerType target,
         backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
-        void* externalImage) {}
+        void* externalImage) {
+    PANIC_POSTCONDITION("External WebGPU Texture is not supported");
+}
 
 void WebGPUDriver::createTextureExternalImagePlaneR(Handle<HwTexture> th,
         backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
-        void* image, uint32_t plane) {}
+        void* image, uint32_t plane) {
+    PANIC_POSTCONDITION("External WebGPU Texture is not supported");
+}
 
 void WebGPUDriver::importTextureR(Handle<HwTexture> th, intptr_t id, SamplerType target,
         uint8_t levels, TextureFormat format, uint8_t samples, uint32_t w, uint32_t h,
-        uint32_t depth, TextureUsage usage) {}
+        uint32_t depth, TextureUsage usage) {
+    PANIC_POSTCONDITION("Import WebGPU Texture is not supported");
+}
 
 void WebGPUDriver::createRenderPrimitiveR(Handle<HwRenderPrimitive> rph, Handle<HwVertexBuffer> vbh,
-        Handle<HwIndexBuffer> ibh, PrimitiveType pt) {}
+        Handle<HwIndexBuffer> ibh, PrimitiveType pt) {
+    assert_invariant(mDevice);
 
-void WebGPUDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {}
+    auto* renderPrimitive = constructHandle<WGPURenderPrimitive>(rph);
+    auto* vertexBuffer = handleCast<WGPUVertexBuffer>(vbh);
+    auto* indexBuffer = handleCast<WGPUIndexBuffer>(ibh);
+    renderPrimitive->vertexBuffer = vertexBuffer;
+    renderPrimitive->indexBuffer = indexBuffer;
+    renderPrimitive->type = pt;
+}
 
-void WebGPUDriver::createDefaultRenderTargetR(Handle<HwRenderTarget> rth, int) {}
+void WebGPUDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {
+    constructHandle<WGPUProgram>(ph, mDevice, program);
+}
+
+void WebGPUDriver::createDefaultRenderTargetR(Handle<HwRenderTarget> rth, int) {
+    assert_invariant(!mDefaultRenderTarget);
+    mDefaultRenderTarget = constructHandle<WGPURenderTarget>(rth);
+    assert_invariant(mDefaultRenderTarget);
+}
 
 void WebGPUDriver::createRenderTargetR(Handle<HwRenderTarget> rth, TargetBufferFlags targets,
         uint32_t width, uint32_t height, uint8_t samples, uint8_t layerCount, MRT color,
@@ -509,10 +567,15 @@ 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) {}
+        Handle<HwDescriptorSetLayout> dslh) {
+    auto layout = handleCast<WebGPUDescriptorSetLayout>(dslh);
+    constructHandle<WebGPUDescriptorSet>(dsh, layout->getLayout(), layout->getLayoutSize());
+}
 
 Handle<HwStream> WebGPUDriver::createStreamNative(void* nativeStream) {
     return {};
@@ -548,11 +611,11 @@ FenceStatus WebGPUDriver::getFenceStatus(Handle<HwFence> fh) {
 // We create all textures using VK_IMAGE_TILING_OPTIMAL, so our definition of "supported" is that
 // the GPU supports the given texture format with non-zero optimal tiling features.
 bool WebGPUDriver::isTextureFormatSupported(TextureFormat format) {
-    return true;
+    return WGPUTexture::fToWGPUTextureFormat(format) != wgpu::TextureFormat::Undefined;
 }
 
 bool WebGPUDriver::isTextureSwizzleSupported() {
-    return true;
+    return false;
 }
 
 bool WebGPUDriver::isTextureFormatMipmappable(TextureFormat format) {
@@ -641,24 +704,30 @@ size_t WebGPUDriver::getMaxArrayTextureLayers() {
 
 void WebGPUDriver::updateIndexBuffer(Handle<HwIndexBuffer> ibh, BufferDescriptor&& p,
         uint32_t byteOffset) {
-    scheduleDestroy(std::move(p));
+    updateGPUBuffer(handleCast<WGPUIndexBuffer>(ibh), std::move(p), byteOffset);
 }
 
 void WebGPUDriver::updateBufferObject(Handle<HwBufferObject> ibh, BufferDescriptor&& p,
         uint32_t byteOffset) {
-    scheduleDestroy(std::move(p));
+    updateGPUBuffer(handleCast<WGPUBufferObject>(ibh), std::move(p), byteOffset);
 }
 
-void WebGPUDriver::updateBufferObjectUnsynchronized(Handle<HwBufferObject> ibh, BufferDescriptor&& p,
-        uint32_t byteOffset) {
-    scheduleDestroy(std::move(p));
+void WebGPUDriver::updateBufferObjectUnsynchronized(Handle<HwBufferObject> ibh,
+        BufferDescriptor&& p, uint32_t byteOffset) {
+    updateGPUBuffer(handleCast<WGPUBufferObject>(ibh), std::move(p), byteOffset);
 }
 
 void WebGPUDriver::resetBufferObject(Handle<HwBufferObject> boh) {
+    // Is there something that needs to be done here? Vulkan has left it unimplemented.
 }
 
 void WebGPUDriver::setVertexBufferObject(Handle<HwVertexBuffer> vbh, uint32_t index,
         Handle<HwBufferObject> boh) {
+    auto* vertexBuffer = handleCast<WGPUVertexBuffer>(vbh);
+    auto* bufferObject = handleCast<WGPUBufferObject>(boh);
+    assert_invariant(index < vertexBuffer->buffers.size());
+    assert_invariant(bufferObject->buffer.GetUsage() & wgpu::BufferUsage::Vertex);
+    vertexBuffer->buffers[index] = bufferObject->buffer;
 }
 
 void WebGPUDriver::update3DImage(Handle<HwTexture> th,
@@ -691,18 +760,72 @@ void WebGPUDriver::compilePrograms(CompilerPriorityQueue priority,
 }
 
 void WebGPUDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassParams& params) {
+    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 = {
+        .colorAttachmentCount = 1,
+        .colorAttachments = &renderPassColorAttachment,
+        .depthStencilAttachment = nullptr,
+        .timestampWrites = nullptr,
+    };
+
+    mRenderPassEncoder = mCommandEncoder.BeginRenderPass(&renderPassDescriptor);
+    mRenderPassEncoder.SetViewport(params.viewport.left, params.viewport.bottom,
+            params.viewport.width, params.viewport.height, params.depthRange.near, params.depthRange.far);
 }
 
 void WebGPUDriver::endRenderPass(int) {
+    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) {
+    mCommandEncoder = nullptr;
+    mQueue.Submit(1, &mCommandBuffer);
+    mCommandBuffer = nullptr;
+    mTextureView = nullptr;
+    assert_invariant(mSwapChain);
+    mSwapChain->present();
 }
 
 void WebGPUDriver::setPushConstant(backend::ShaderStage stage, uint8_t index,
@@ -730,7 +853,7 @@ 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));
 }
@@ -753,9 +876,60 @@ void WebGPUDriver::blit(
 }
 
 void WebGPUDriver::bindPipeline(PipelineState const& pipelineState) {
+    const auto* program = handleCast<WGPUProgram>(pipelineState.program);
+    assert_invariant(program);
+    assert_invariant(program->computeShaderModule == nullptr &&
+                     "WebGPU backend does not (yet) support compute pipelines.");
+    FILAMENT_CHECK_POSTCONDITION(program->vertexShaderModule)
+            << "WebGPU backend requires a vertex shader module for a render pipeline";
+    std::array<wgpu::BindGroupLayout, MAX_DESCRIPTOR_SET_COUNT> bindGroupLayouts{};
+    assert_invariant(bindGroupLayouts.size() >= pipelineState.pipelineLayout.setLayout.size());
+    size_t bindGroupLayoutCount = 0;
+    for (size_t i = 0; i < bindGroupLayouts.size(); i++) {
+        const auto handle = pipelineState.pipelineLayout.setLayout[bindGroupLayoutCount];
+        if (handle.getId() == HandleBase::nullid) {
+            continue;
+        }
+        bindGroupLayouts[bindGroupLayoutCount++] =
+                handleCast<WebGPUDescriptorSetLayout>(handle)->getLayout();
+    }
+    std::stringstream layoutLabelStream;
+    layoutLabelStream << program->name.c_str() << " layout";
+    const auto layoutLabel = layoutLabelStream.str();
+    const wgpu::PipelineLayoutDescriptor layoutDescriptor{
+        .label = wgpu::StringView(layoutLabel),
+        .bindGroupLayoutCount = bindGroupLayoutCount,
+        .bindGroupLayouts = bindGroupLayouts.data()
+        // TODO investigate immediateDataRangeByteSize
+    };
+    const wgpu::PipelineLayout layout = mDevice.CreatePipelineLayout(&layoutDescriptor);
+    FILAMENT_CHECK_POSTCONDITION(layout)
+            << "Failed to create wgpu::PipelineLayout for render pipeline for "
+            << layoutDescriptor.label;
+    auto const* vertexBufferInfo = handleCast<WGPUVertexBufferInfo>(pipelineState.vertexBufferInfo);
+    assert_invariant(vertexBufferInfo);
+    const wgpu::RenderPipeline pipeline = createWebGPURenderPipeline(mDevice, *program,
+            *vertexBufferInfo, layout, pipelineState.rasterState, pipelineState.stencilState,
+            pipelineState.polygonOffset, pipelineState.primitiveType, mSwapChain->getColorFormat(),
+            mSwapChain->getDepthFormat());
+    // TODO: uncomment once we have a valid pipeline to set
+    // mRenderPassEncoder.SetPipeline(pipeline);
 }
 
 void WebGPUDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
+    auto* renderPrimitive = handleCast<WGPURenderPrimitive>(rph);
+
+    // This *must* match the WGPUVertexBufferInfo that was bound in bindPipeline(). But we want
+    // to allow to call this before bindPipeline(), so the validation can only happen in draw()
+    auto vbi = handleCast<WGPUVertexBufferInfo>(renderPrimitive->vertexBuffer->vbih);
+    assert_invariant(
+            vbi->getVertexBufferLayoutSize() == renderPrimitive->vertexBuffer->buffers.size());
+    for (uint32_t i = 0; i < vbi->getVertexBufferLayoutSize(); i++) {
+        mRenderPassEncoder.SetVertexBuffer(i, renderPrimitive->vertexBuffer->buffers[i]);
+    }
+
+    mRenderPassEncoder.SetIndexBuffer(renderPrimitive->indexBuffer->buffer,
+            renderPrimitive->indexBuffer->indexFormat);
 }
 
 void WebGPUDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
@@ -781,25 +955,48 @@ void WebGPUDriver::endTimerQuery(Handle<HwTimerQuery> tqh) {
 void WebGPUDriver::resetState(int) {
 }
 
-void WebGPUDriver::updateDescriptorSetBuffer(
-        backend::DescriptorSetHandle dsh,
-        backend::descriptor_binding_t binding,
-        backend::BufferObjectHandle boh,
-        uint32_t offset,
+void WebGPUDriver::updateDescriptorSetBuffer(Handle<HwDescriptorSet> dsh,
+        backend::descriptor_binding_t binding, Handle<HwBufferObject> boh, uint32_t offset,
         uint32_t size) {
+    auto bindGroup = handleCast<WebGPUDescriptorSet>(dsh);
+    auto buffer = handleCast<WGPUBufferObject>(boh);
+    if (!bindGroup->getIsLocked()) {
+        // TODO making assumptions that size and offset mean the same thing here.
+        wgpu::BindGroupEntry entry{ .binding = static_cast<uint32_t>(binding * 2),
+            .buffer = buffer->buffer,
+            .offset = offset,
+            .size = size };
+        bindGroup->addEntry(entry.binding, std::move(entry));
+    }
 }
 
-void WebGPUDriver::updateDescriptorSetTexture(
-        backend::DescriptorSetHandle dsh,
-        backend::descriptor_binding_t binding,
-        backend::TextureHandle th,
-        SamplerParams params) {
+void WebGPUDriver::updateDescriptorSetTexture(Handle<HwDescriptorSet> dsh,
+        backend::descriptor_binding_t binding, Handle<HwTexture> th, SamplerParams params) {
+    /*
+    auto bindGroup = handleCast<WebGPUDescriptorSet>(dsh);
+    auto texture = handleCast<WGPUTexture>(th);
+
+    // TODO very high odds badd assumptions are in here about handling HwTexture. Revisit with more
+    // understanding. Right now assuming there is a wgpu::TextureView filled in
+    if (!bindGroup->getIsLocked()) {
+        // TODO making assumptions that size and offset mean the same thing here.
+        wgpu::BindGroupEntry tEntry{ .binding = static_cast<uint32_t>(binding * 2),
+            .textureView = texture->texView };
+        bindGroup->addEntry(tEntry.binding, std::move(tEntry));
+
+        wgpu::BindGroupEntry sEntry{ .binding = static_cast<uint32_t>(binding * 2 + 1),
+            .sampler = texture->sampler };
+        bindGroup->addEntry(sEntry.binding, std::move(sEntry));
+    }
+    //TODO Just the setup, this function stilll needs the rest of logic implemented
+     */
 }
 
-void WebGPUDriver::bindDescriptorSet(
-        backend::DescriptorSetHandle dsh,
-        backend::descriptor_set_t set,
+void WebGPUDriver::bindDescriptorSet(Handle<HwDescriptorSet> dsh, backend::descriptor_set_t set,
         backend::DescriptorSetOffsetArray&& offsets) {
+    auto bindGroup = handleCast<WebGPUDescriptorSet>(dsh);
+    // TODO: presume we need this, use it. Probably Encoder::SetBindGroup
+    auto wbg = bindGroup->lockAndReturn(mDevice);
 }
 
 void WebGPUDriver::setDebugTag(HandleBase::HandleId handleId, utils::CString tag) {

filament/backend/src/webgpu/WebGPUDriver.h

@@ -17,7 +17,8 @@
 #ifndef TNT_FILAMENT_BACKEND_WEBGPUDRIVER_H
 #define TNT_FILAMENT_BACKEND_WEBGPUDRIVER_H
 
-#include "webgpu/WebGPUSwapChain.h"
+#include "WebGPUHandles.h"
+#include "webgpu/WebGPUConstants.h"
 #include <backend/platforms/WebGPUPlatform.h>
 
 #include "DriverBase.h"
@@ -34,11 +35,13 @@
 #include <memory>
 
 #ifndef FILAMENT_WEBGPU_HANDLE_ARENA_SIZE_IN_MB
-#    define FILAMENT_WEBGPU_HANDLE_ARENA_SIZE_IN_MB 8
+#define FILAMENT_WEBGPU_HANDLE_ARENA_SIZE_IN_MB 8
 #endif
 
 namespace filament::backend {
 
+class WebGPUSwapChain;
+
 /**
  * WebGPU backend (driver) implementation
  */
@@ -54,16 +57,41 @@ private:
     [[nodiscard]] ShaderModel getShaderModel() const noexcept final;
     [[nodiscard]] ShaderLanguage getShaderLanguage() const noexcept final;
 
+    template<typename GPUBufferObject>
+    void updateGPUBuffer(GPUBufferObject* gpuBufferObject, BufferDescriptor&& bufferDescriptor,
+            uint32_t byteOffset) {
+        FILAMENT_CHECK_PRECONDITION(bufferDescriptor.buffer)
+                << "copyIntoBuffer called with a null buffer";
+        FILAMENT_CHECK_PRECONDITION(
+                bufferDescriptor.size + byteOffset <= gpuBufferObject->buffer.GetSize())
+                << "Attempting to copy " << bufferDescriptor.size << " bytes into a buffer of size "
+                << gpuBufferObject->buffer.GetSize() << " at offset " << byteOffset;
+
+        // TODO: All buffer objects are created with CopyDst usage.
+        // This may have some performance implications. That should be investigated later.
+        assert_invariant(gpuBufferObject->buffer.GetUsage() & wgpu::BufferUsage::CopyDst);
+
+        // WriteBuffer is an async call. But cpu buffer data is already written to the staging
+        // buffer on return from the WriteBuffer.
+        mQueue.WriteBuffer(gpuBufferObject->buffer, byteOffset, bufferDescriptor.buffer,
+                bufferDescriptor.size);
+        scheduleDestroy(std::move(bufferDescriptor));
+    }
+
     // the platform (e.g. OS) specific aspects of the WebGPU backend are strictly only
     // handled in the WebGPUPlatform
     WebGPUPlatform& mPlatform;
     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;
+    wgpu::RenderPassEncoder mRenderPassEncoder = nullptr;
+    wgpu::CommandBuffer mCommandBuffer = nullptr;
+    WGPURenderTarget* mDefaultRenderTarget = nullptr;
     /*
      * Driver interface
      */
@@ -93,6 +121,21 @@ private:
         return mHandleAllocator.allocate<D>();
     }
 
+    template<typename D, typename B, typename... ARGS>
+    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);
+        mHandleAllocator.deallocate(handle, p);
+    }
 };
 
 }// namespace filament::backend

filament/backend/src/webgpu/WebGPUHandles.cpp

@@ -0,0 +1,655 @@
+/*
+ * 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 {
+constexpr wgpu::BufferUsage getBufferObjectUsage(
+        filament::backend::BufferObjectBinding bindingType) noexcept {
+    switch (bindingType) {
+        case filament::backend::BufferObjectBinding::VERTEX:
+            return wgpu::BufferUsage::Vertex;
+        case filament::backend::BufferObjectBinding::UNIFORM:
+            return wgpu::BufferUsage::Uniform;
+        case filament::backend::BufferObjectBinding::SHADER_STORAGE:
+            return wgpu::BufferUsage::Storage;
+    }
+}
+
+wgpu::Buffer createBuffer(wgpu::Device const& device, wgpu::BufferUsage usage, uint32_t size,
+        char const* label) {
+    wgpu::BufferDescriptor descriptor{ .label = label,
+        .usage = usage,
+        .size = size,
+        .mappedAtCreation = false };
+    return device.CreateBuffer(&descriptor);
+}
+
+wgpu::VertexFormat getVertexFormat(filament::backend::ElementType type, bool normalized, bool integer) {
+    using ElementType = filament::backend::ElementType;
+    using VertexFormat = wgpu::VertexFormat;
+    if (normalized) {
+        switch (type) {
+            // Single Component Types
+            case ElementType::BYTE: return VertexFormat::Snorm8;
+            case ElementType::UBYTE: return VertexFormat::Unorm8;
+            case ElementType::SHORT: return VertexFormat::Snorm16;
+            case ElementType::USHORT: return VertexFormat::Unorm16;
+            // Two Component Types
+            case ElementType::BYTE2: return VertexFormat::Snorm8x2;
+            case ElementType::UBYTE2: return VertexFormat::Unorm8x2;
+            case ElementType::SHORT2: return VertexFormat::Snorm16x2;
+            case ElementType::USHORT2: return VertexFormat::Unorm16x2;
+            // Three Component Types
+            // There is no vertex format type for 3 byte data in webgpu. Use
+            // 4 byte signed normalized type and ignore the last byte.
+            // TODO: This is to be verified.
+            case ElementType::BYTE3: return VertexFormat::Snorm8x4;    // NOT MINSPEC
+            case ElementType::UBYTE3: return VertexFormat::Unorm8x4;   // NOT MINSPEC
+            case ElementType::SHORT3: return VertexFormat::Snorm16x4;  // NOT MINSPEC
+            case ElementType::USHORT3: return VertexFormat::Unorm16x4; // NOT MINSPEC
+            // Four Component Types
+            case ElementType::BYTE4: return VertexFormat::Snorm8x4;
+            case ElementType::UBYTE4: return VertexFormat::Unorm8x4;
+            case ElementType::SHORT4: return VertexFormat::Snorm16x4;
+            case ElementType::USHORT4: return VertexFormat::Unorm8x4;
+            default:
+                FILAMENT_CHECK_POSTCONDITION(false) << "Normalized format does not exist.";
+                return VertexFormat::Float32x3;
+        }
+    }
+    switch (type) {
+        // Single Component Types
+        // There is no direct alternative for SSCALED in webgpu. Convert them to Float32 directly.
+        // This will result in increased memory on the cpu side.
+        // TODO: Is Float16 acceptable instead with some potential accuracy errors?
+        case ElementType::BYTE: return integer ? VertexFormat::Sint8 : VertexFormat::Float32;
+        case ElementType::UBYTE: return integer ? VertexFormat::Uint8 : VertexFormat::Float32;
+        case ElementType::SHORT: return integer ? VertexFormat::Sint16 : VertexFormat::Float32;
+        case ElementType::USHORT: return integer ? VertexFormat::Uint16 : VertexFormat::Float32;
+        case ElementType::HALF: return VertexFormat::Float16;
+        case ElementType::INT: return VertexFormat::Sint32;
+        case ElementType::UINT: return VertexFormat::Uint32;
+        case ElementType::FLOAT: return VertexFormat::Float32;
+        // Two Component Types
+        case ElementType::BYTE2: return integer ? VertexFormat::Sint8x2 : VertexFormat::Float32x2;
+        case ElementType::UBYTE2: return integer ? VertexFormat::Uint8x2 : VertexFormat::Float32x2;
+        case ElementType::SHORT2: return integer ? VertexFormat::Sint16x2 : VertexFormat::Float32x2;
+        case ElementType::USHORT2: return integer ? VertexFormat::Uint16x2 : VertexFormat::Float32x2;
+        case ElementType::HALF2: return VertexFormat::Float16x2;
+        case ElementType::FLOAT2: return VertexFormat::Float32x2;
+        // Three Component Types
+        case ElementType::BYTE3: return VertexFormat::Sint8x4;    // NOT MINSPEC
+        case ElementType::UBYTE3: return VertexFormat::Uint8x4;   // NOT MINSPEC
+        case ElementType::SHORT3: return VertexFormat::Sint16x4;  // NOT MINSPEC
+        case ElementType::USHORT3: return VertexFormat::Uint16x4; // NOT MINSPEC
+        case ElementType::HALF3: return VertexFormat::Float16x4;  // NOT MINSPEC
+        case ElementType::FLOAT3: return VertexFormat::Float32x3;
+        // Four Component Types
+        case ElementType::BYTE4: return integer ? VertexFormat::Sint8x4 : VertexFormat::Float32x4;
+        case ElementType::UBYTE4: return integer ? VertexFormat::Uint8x4 : VertexFormat::Float32x4;
+        case ElementType::SHORT4: return integer ? VertexFormat::Sint16x4 : VertexFormat::Float32x4;
+        case ElementType::USHORT4: return integer ? VertexFormat::Uint16x4 : VertexFormat::Float32x4;
+        case ElementType::HALF4: return VertexFormat::Float16x4;
+        case ElementType::FLOAT4: return VertexFormat::Float32x4;
+    }
+}
+
+}// namespace
+
+namespace filament::backend {
+
+WGPUVertexBufferInfo::WGPUVertexBufferInfo(uint8_t bufferCount, uint8_t attributeCount,
+        AttributeArray const& attributes)
+    : HwVertexBufferInfo(bufferCount, attributeCount),
+      mVertexBufferLayout(bufferCount),
+      mAttributes(bufferCount) {
+    assert_invariant(attributeCount > 0);
+    assert_invariant(bufferCount > 0);
+    for (uint32_t attribIndex = 0; attribIndex < attributes.size(); attribIndex++) {
+        Attribute const& attrib = attributes[attribIndex];
+        // Ignore the attributes which are not bind to vertex buffers.
+        if (attrib.buffer == Attribute::BUFFER_UNUSED) {
+            continue;
+        }
+
+        assert_invariant(attrib.buffer < bufferCount);
+        bool const isInteger = attrib.flags & Attribute::FLAG_INTEGER_TARGET;
+        bool const isNormalized = attrib.flags & Attribute::FLAG_NORMALIZED;
+        wgpu::VertexFormat vertexFormat = getVertexFormat(attrib.type, isNormalized, isInteger);
+
+        // Attributes are sequential per buffer
+        mAttributes[attrib.buffer].push_back({
+            .format = vertexFormat,
+            .offset = attrib.offset,
+            .shaderLocation = static_cast<uint32_t>(mAttributes[attrib.buffer].size()),
+        });
+
+        mVertexBufferLayout[attrib.buffer].stepMode = wgpu::VertexStepMode::Vertex;
+        if (mVertexBufferLayout[attrib.buffer].arrayStride == 0) {
+            mVertexBufferLayout[attrib.buffer].arrayStride = attrib.stride;
+        } else {
+            assert_invariant(mVertexBufferLayout[attrib.buffer].arrayStride == attrib.stride);
+        }
+    }
+
+    for (uint32_t bufferIndex = 0; bufferIndex < bufferCount; bufferIndex++) {
+        mVertexBufferLayout[bufferIndex].attributeCount = mAttributes[bufferIndex].size();
+        mVertexBufferLayout[bufferIndex].attributes = mAttributes[bufferIndex].data();
+    }
+}
+
+WGPUIndexBuffer::WGPUIndexBuffer(wgpu::Device const& device, uint8_t elementSize,
+        uint32_t indexCount)
+    : buffer(createBuffer(device, wgpu::BufferUsage::CopyDst | wgpu::BufferUsage::Index,
+              elementSize * indexCount, "index_buffer")),
+      indexFormat(elementSize == 2 ? wgpu::IndexFormat::Uint16 : wgpu::IndexFormat::Uint32) {}
+
+
+WGPUVertexBuffer::WGPUVertexBuffer(wgpu::Device const& device, uint32_t vertexCount,
+        uint32_t bufferCount, Handle<HwVertexBufferInfo> vbih)
+    : HwVertexBuffer(vertexCount),
+      vbih(vbih),
+      buffers(bufferCount) {}
+
+WGPUBufferObject::WGPUBufferObject(wgpu::Device const& device, BufferObjectBinding bindingType,
+        uint32_t byteCount)
+    : HwBufferObject(byteCount),
+      buffer(createBuffer(device, wgpu::BufferUsage::CopyDst | getBufferObjectUsage(bindingType),
+              byteCount, "buffer_object")),
+      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??
+    mLayoutSize = wEntries.size();
+    mLayout = device.CreateBindGroupLayout(&layoutDescriptor);
+}
+
+WebGPUDescriptorSetLayout::~WebGPUDescriptorSetLayout() {}
+
+WebGPUDescriptorSet::WebGPUDescriptorSet(const wgpu::BindGroupLayout& layout, uint layoutSize)
+    : mLayout(layout),
+      entries(layoutSize, wgpu::BindGroupEntry{}) {
+    // Establish the size of entries based on the layout. This should be reliable and efficient.
+}
+WebGPUDescriptorSet::~WebGPUDescriptorSet() {}
+
+wgpu::BindGroup WebGPUDescriptorSet::lockAndReturn(const wgpu::Device& device) {
+    if (mBindGroup) {
+        return mBindGroup;
+    }
+    // TODO label? Should we just copy layout label?
+    wgpu::BindGroupDescriptor desc{ .layout = mLayout,
+        .entryCount = entries.size(),
+        .entries = entries.data() };
+    mBindGroup = device.CreateBindGroup(&desc);
+    return mBindGroup;
+}
+
+void WebGPUDescriptorSet::addEntry(uint index, wgpu::BindGroupEntry&& entry) {
+    if (mBindGroup) {
+        // We will keep getting hits from future updates, but shouldn't do anything
+        // Filament guarantees this won't change after things have locked.
+        return;
+    }
+    // TODO: Putting some level of trust that Filament is not going to reuse indexes or go past the
+    // layout index for efficiency. Add guards if wrong.
+    entries[index] = std::move(entry);
+}
+// From createTextureR
+WGPUTexture::WGPUTexture(SamplerType target, uint8_t levels, TextureFormat format, uint8_t samples,
+        uint32_t width, uint32_t height, uint32_t depth, TextureUsage usage,
+        wgpu::Device device) noexcept {
+
+    // First the texture aspect
+    wgpu::TextureDescriptor desc;
+
+    switch (target) {
+        case SamplerType::SAMPLER_CUBEMAP:
+        case SamplerType::SAMPLER_CUBEMAP_ARRAY:
+        case SamplerType::SAMPLER_2D:
+        case SamplerType::SAMPLER_2D_ARRAY:
+        // Should be safe to assume external is 2d
+        case SamplerType::SAMPLER_EXTERNAL: {
+            desc.dimension = wgpu::TextureDimension::e2D;
+            break;
+        }
+        case SamplerType::SAMPLER_3D: {
+            desc.dimension = wgpu::TextureDimension::e3D;
+            break;
+        }
+    }
+    desc.size = { .width = width, .height = height, .depthOrArrayLayers = depth };
+    desc.format = fToWGPUTextureFormat(format);
+    assert_invariant(desc.format != wgpu::TextureFormat::Undefined);
+
+    // WGPU requires this to be true. Filament should comply
+    assert(samples == 1 || samples || 4);
+
+    desc.sampleCount = samples;
+    desc.usage = fToWGPUTextureUsage(usage);
+    desc.mipLevelCount = levels;
+
+    // TODO Is this fine? Could do all-the-things, a naive mapping or get something from Filament
+    desc.viewFormats = nullptr;
+
+    texture = device.CreateTexture(&desc);
+
+    // TODO should a default levelCount be something other than 0? Sample count?
+    texView = makeTextureView(0, 1);
+}
+// From createTextureViewR
+WGPUTexture::WGPUTexture(WGPUTexture* src, uint8_t baseLevel, uint8_t levelCount) noexcept {
+    texture = src->texture;
+    texView = makeTextureView(baseLevel, levelCount);
+}
+
+wgpu::TextureUsage WGPUTexture::fToWGPUTextureUsage(const TextureUsage& fUsage) {
+    wgpu::TextureUsage retUsage = wgpu::TextureUsage::None;
+
+    // Basing this mapping off of VulkanTexture.cpp's getUsage func and suggestions from Gemini
+    // TODO Validate assumptions, revisit if issues.
+    if (any(TextureUsage::BLIT_SRC & fUsage)) {
+        retUsage |= wgpu::TextureUsage::CopySrc;
+    }
+    if (any((TextureUsage::BLIT_DST | TextureUsage::UPLOADABLE) & fUsage)) {
+        retUsage |= wgpu::TextureUsage::CopyDst;
+    }
+    if (any(TextureUsage::SAMPLEABLE & fUsage)) {
+        retUsage |= wgpu::TextureUsage::TextureBinding;
+    }
+    // WGPU Render attachment covers either color or stencil situation dependant
+    // NOTE: Depth attachment isn't used this way in Vulkan but logically maps to WGPU docs. If
+    // issues, investigate here.
+    if (any((TextureUsage::COLOR_ATTACHMENT | TextureUsage::STENCIL_ATTACHMENT |
+                    TextureUsage::DEPTH_ATTACHMENT) &
+                fUsage)) {
+        retUsage |= wgpu::TextureUsage::RenderAttachment;
+    }
+
+    // This is from Vulkan logic- if there are any issues try disabling this first, allows perf
+    // benefit though
+    const bool useTransientAttachment =
+            // Usage consists of attachment flags only.
+            none(fUsage & ~TextureUsage::ALL_ATTACHMENTS) &&
+            // Usage contains at least one attachment flag.
+            any(fUsage & TextureUsage::ALL_ATTACHMENTS) &&
+            // Depth resolve cannot use transient attachment because it uses a custom shader.
+            // TODO: see VulkanDriver::isDepthStencilResolveSupported() to know when to remove this
+            // restriction.
+            // Note that the custom shader does not resolve stencil. We do need to move to vk 1.2
+            // and above to be able to support stencil resolve (along with depth).
+            !(any(fUsage & TextureUsage::DEPTH_ATTACHMENT) && samples > 1);
+    if (useTransientAttachment) {
+        retUsage |= wgpu::TextureUsage::TransientAttachment;
+    }
+    // NOTE: Unused wgpu flags:
+    //  StorageBinding
+    //  StorageAttachment
+
+    // NOTE: Unused Filament flags:
+    //  SUBPASS_INPUT VK goes to input attachment which we don't support right now
+    //  PROTECTED
+    return retUsage;
+}
+wgpu::TextureFormat WGPUTexture::fToWGPUTextureFormat(const TextureFormat& fUsage) {
+    switch (fUsage) {
+        case filament::backend::TextureFormat::R8:
+            return wgpu::TextureFormat::R8Unorm;
+        case filament::backend::TextureFormat::R8_SNORM:
+            return wgpu::TextureFormat::R8Snorm;
+        case filament::backend::TextureFormat::R8UI:
+            return wgpu::TextureFormat::R8Uint;
+        case filament::backend::TextureFormat::R8I:
+            return wgpu::TextureFormat::R8Sint;
+        case filament::backend::TextureFormat::STENCIL8:
+            return wgpu::TextureFormat::Stencil8;
+        case filament::backend::TextureFormat::R16F:
+            return wgpu::TextureFormat::R16Float;
+        case filament::backend::TextureFormat::R16UI:
+            return wgpu::TextureFormat::R16Uint;
+        case filament::backend::TextureFormat::R16I:
+            return wgpu::TextureFormat::R16Sint;
+        case filament::backend::TextureFormat::RG8:
+            return wgpu::TextureFormat::RG8Unorm;
+        case filament::backend::TextureFormat::RG8_SNORM:
+            return wgpu::TextureFormat::RG8Snorm;
+        case filament::backend::TextureFormat::RG8UI:
+            return wgpu::TextureFormat::RG8Uint;
+        case filament::backend::TextureFormat::RG8I:
+            return wgpu::TextureFormat::RG8Sint;
+        case filament::backend::TextureFormat::R32F:
+            return wgpu::TextureFormat::R32Float;
+        case filament::backend::TextureFormat::R32UI:
+            return wgpu::TextureFormat::R32Uint;
+        case filament::backend::TextureFormat::R32I:
+            return wgpu::TextureFormat::R32Sint;
+        case filament::backend::TextureFormat::RG16F:
+            return wgpu::TextureFormat::RG16Float;
+        case filament::backend::TextureFormat::RG16UI:
+            return wgpu::TextureFormat::RG16Uint;
+        case filament::backend::TextureFormat::RG16I:
+            return wgpu::TextureFormat::RG16Sint;
+        case filament::backend::TextureFormat::RGBA8:
+            return wgpu::TextureFormat::RGBA8Unorm;
+        case filament::backend::TextureFormat::SRGB8_A8:
+            return wgpu::TextureFormat::RGBA8UnormSrgb;
+        case filament::backend::TextureFormat::RGBA8_SNORM:
+            return wgpu::TextureFormat::RGBA8Snorm;
+        case filament::backend::TextureFormat::RGBA8UI:
+            return wgpu::TextureFormat::RGBA8Uint;
+        case filament::backend::TextureFormat::RGBA8I:
+            return wgpu::TextureFormat::RGBA8Sint;
+        case filament::backend::TextureFormat::DEPTH16:
+            return wgpu::TextureFormat::Depth16Unorm;
+        case filament::backend::TextureFormat::DEPTH24:
+            return wgpu::TextureFormat::Depth24Plus;
+        case filament::backend::TextureFormat::DEPTH32F:
+            return wgpu::TextureFormat::Depth32Float;
+        case filament::backend::TextureFormat::DEPTH24_STENCIL8:
+            return wgpu::TextureFormat::Depth24PlusStencil8;
+        case filament::backend::TextureFormat::DEPTH32F_STENCIL8:
+            return wgpu::TextureFormat::Depth32FloatStencil8;
+        case filament::backend::TextureFormat::RG32F:
+            return wgpu::TextureFormat::RG32Float;
+        case filament::backend::TextureFormat::RG32UI:
+            return wgpu::TextureFormat::RG32Uint;
+        case filament::backend::TextureFormat::RG32I:
+            return wgpu::TextureFormat::RG32Sint;
+        case filament::backend::TextureFormat::RGBA16F:
+            return wgpu::TextureFormat::RGBA16Float;
+        case filament::backend::TextureFormat::RGBA16UI:
+            return wgpu::TextureFormat::RGBA16Uint;
+        case filament::backend::TextureFormat::RGBA16I:
+            return wgpu::TextureFormat::RGBA16Sint;
+        case filament::backend::TextureFormat::RGBA32F:
+            return wgpu::TextureFormat::RGBA32Float;
+        case filament::backend::TextureFormat::RGBA32UI:
+            return wgpu::TextureFormat::RGBA32Uint;
+        case filament::backend::TextureFormat::RGBA32I:
+            return wgpu::TextureFormat::RGBA32Sint;
+        case filament::backend::TextureFormat::EAC_R11:
+            return wgpu::TextureFormat::EACR11Unorm;
+        case filament::backend::TextureFormat::EAC_R11_SIGNED:
+            return wgpu::TextureFormat::EACR11Snorm;
+        case filament::backend::TextureFormat::EAC_RG11:
+            return wgpu::TextureFormat::EACRG11Unorm;
+        case filament::backend::TextureFormat::EAC_RG11_SIGNED:
+            return wgpu::TextureFormat::EACRG11Snorm;
+        case filament::backend::TextureFormat::ETC2_RGB8:
+            return wgpu::TextureFormat::ETC2RGB8Unorm;
+        case filament::backend::TextureFormat::ETC2_SRGB8:
+            return wgpu::TextureFormat::ETC2RGB8UnormSrgb;
+        case filament::backend::TextureFormat::ETC2_RGB8_A1:
+            return wgpu::TextureFormat::ETC2RGB8A1Unorm;
+        case filament::backend::TextureFormat::ETC2_SRGB8_A1:
+            return wgpu::TextureFormat::ETC2RGB8A1UnormSrgb;
+        case filament::backend::TextureFormat::ETC2_EAC_RGBA8:
+            return wgpu::TextureFormat::ETC2RGBA8Unorm;
+        case filament::backend::TextureFormat::ETC2_EAC_SRGBA8:
+            return wgpu::TextureFormat::ETC2RGBA8UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_4x4:
+            return wgpu::TextureFormat::ASTC4x4Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_4x4:
+            return wgpu::TextureFormat::ASTC4x4UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_5x4:
+            return wgpu::TextureFormat::ASTC5x4Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_5x4:
+            return wgpu::TextureFormat::ASTC5x4UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_5x5:
+            return wgpu::TextureFormat::ASTC5x5Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_5x5:
+            return wgpu::TextureFormat::ASTC5x5UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_6x5:
+            return wgpu::TextureFormat::ASTC6x5Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_6x5:
+            return wgpu::TextureFormat::ASTC6x5UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_6x6:
+            return wgpu::TextureFormat::ASTC6x6Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_6x6:
+            return wgpu::TextureFormat::ASTC6x6UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_8x5:
+            return wgpu::TextureFormat::ASTC8x5Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_8x5:
+            return wgpu::TextureFormat::ASTC8x5UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_8x6:
+            return wgpu::TextureFormat::ASTC8x6Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_8x6:
+            return wgpu::TextureFormat::ASTC8x6UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_8x8:
+            return wgpu::TextureFormat::ASTC8x8Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_8x8:
+            return wgpu::TextureFormat::ASTC8x8UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_10x5:
+            return wgpu::TextureFormat::ASTC10x5Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_10x5:
+            return wgpu::TextureFormat::ASTC10x5UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_10x6:
+            return wgpu::TextureFormat::ASTC10x6Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_10x6:
+            return wgpu::TextureFormat::ASTC10x6UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_10x8:
+            return wgpu::TextureFormat::ASTC10x8Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_10x8:
+            return wgpu::TextureFormat::ASTC10x8UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_10x10:
+            return wgpu::TextureFormat::ASTC10x10Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_10x10:
+            return wgpu::TextureFormat::ASTC10x10UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_12x10:
+            return wgpu::TextureFormat::ASTC12x10Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_12x10:
+            return wgpu::TextureFormat::ASTC12x10UnormSrgb;
+        case filament::backend::TextureFormat::RGBA_ASTC_12x12:
+            return wgpu::TextureFormat::ASTC12x12Unorm;
+        case filament::backend::TextureFormat::SRGB8_ALPHA8_ASTC_12x12:
+            return wgpu::TextureFormat::ASTC12x12UnormSrgb;
+        case filament::backend::TextureFormat::RED_RGTC1:
+            return wgpu::TextureFormat::BC4RUnorm;
+        case filament::backend::TextureFormat::SIGNED_RED_RGTC1:
+            return wgpu::TextureFormat::BC4RSnorm;
+        case filament::backend::TextureFormat::RED_GREEN_RGTC2:
+            return wgpu::TextureFormat::BC5RGUnorm;
+        case filament::backend::TextureFormat::SIGNED_RED_GREEN_RGTC2:
+            return wgpu::TextureFormat::BC5RGSnorm;
+        case filament::backend::TextureFormat::RGB_BPTC_UNSIGNED_FLOAT:
+            return wgpu::TextureFormat::BC6HRGBUfloat;
+        case filament::backend::TextureFormat::RGB_BPTC_SIGNED_FLOAT:
+            return wgpu::TextureFormat::BC6HRGBFloat;
+        case filament::backend::TextureFormat::RGBA_BPTC_UNORM:
+            return wgpu::TextureFormat::BC7RGBAUnorm;
+        case filament::backend::TextureFormat::SRGB_ALPHA_BPTC_UNORM:
+            return wgpu::TextureFormat::BC7RGBAUnormSrgb;
+        case filament::backend::TextureFormat::RGB565:
+            // No direct mapping in wgpu. Could potentially map to RGBA8Unorm
+            // and discard the alpha and lower precision.
+            return wgpu::TextureFormat::Undefined;
+        case filament::backend::TextureFormat::RGB9_E5:
+            return wgpu::TextureFormat::RGB9E5Ufloat;
+        case filament::backend::TextureFormat::RGB5_A1:
+            // No direct mapping in wgpu. Could potentially map to RGBA8Unorm
+            // and handle the packing/unpacking in shaders.
+            return wgpu::TextureFormat::Undefined;
+        case filament::backend::TextureFormat::RGBA4:
+            // No direct mapping in wgpu. Could potentially map to RGBA8Unorm
+            // and handle the packing/unpacking in shaders.
+            return wgpu::TextureFormat::Undefined;
+        case filament::backend::TextureFormat::RGB8:
+            // No direct sRGB equivalent in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA8Unorm;
+        case filament::backend::TextureFormat::SRGB8:
+            // No direct sRGB equivalent in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA8UnormSrgb;
+        case filament::backend::TextureFormat::RGB8_SNORM:
+            // No direct mapping in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA8Snorm;
+        case filament::backend::TextureFormat::RGB8UI:
+            // No direct mapping in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA8Uint;
+        case filament::backend::TextureFormat::RGB8I:
+            // No direct mapping in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA8Sint;
+        case filament::backend::TextureFormat::R11F_G11F_B10F:
+            return wgpu::TextureFormat::RG11B10Ufloat;
+        case filament::backend::TextureFormat::UNUSED:
+            return wgpu::TextureFormat::Undefined;
+        case filament::backend::TextureFormat::RGB10_A2:
+            return wgpu::TextureFormat::RGB10A2Unorm;
+        case filament::backend::TextureFormat::RGB16F:
+            // No direct mapping in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA16Float;
+        case filament::backend::TextureFormat::RGB16UI:
+            // No direct mapping in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA16Uint;
+        case filament::backend::TextureFormat::RGB16I:
+            // No direct mapping in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA16Sint;
+        case filament::backend::TextureFormat::RGB32F:
+            // No direct mapping in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA32Float;
+        case filament::backend::TextureFormat::RGB32UI:
+            // No direct mapping in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA32Uint;
+        case filament::backend::TextureFormat::RGB32I:
+            // No direct mapping in wgpu without alpha.
+            return wgpu::TextureFormat::RGBA32Sint;
+        case filament::backend::TextureFormat::DXT1_RGB:
+            return wgpu::TextureFormat::BC1RGBAUnorm;
+        case filament::backend::TextureFormat::DXT1_RGBA:
+            return wgpu::TextureFormat::BC1RGBAUnorm;
+        case filament::backend::TextureFormat::DXT3_RGBA:
+            return wgpu::TextureFormat::BC2RGBAUnorm;
+        case filament::backend::TextureFormat::DXT5_RGBA:
+            return wgpu::TextureFormat::BC3RGBAUnorm;
+        case filament::backend::TextureFormat::DXT1_SRGB:
+            return wgpu::TextureFormat::BC1RGBAUnormSrgb;
+        case filament::backend::TextureFormat::DXT1_SRGBA:
+            return wgpu::TextureFormat::BC1RGBAUnormSrgb;
+        case filament::backend::TextureFormat::DXT3_SRGBA:
+            return wgpu::TextureFormat::BC2RGBAUnormSrgb;
+        case filament::backend::TextureFormat::DXT5_SRGBA:
+            return wgpu::TextureFormat::BC3RGBAUnormSrgb;
+    }
+}
+
+wgpu::TextureView WGPUTexture::makeTextureView(const uint8_t& baseLevel,
+        const uint8_t& levelCount) {
+    wgpu::TextureViewDescriptor desc;
+    desc.baseMipLevel = baseLevel;
+    desc.mipLevelCount = levelCount;
+
+    // baseArrayLayer is required, making a guess
+    desc.baseArrayLayer = 0;
+    // Have not found an analouge to aspect in other drivers, but ALL should be unrestrictive.
+    // TODO Can we make this better?
+    desc.aspect = wgpu::TextureAspect::All;
+
+    // The rest of the properties should be fine to leave as default, using the texture params.
+    desc.label = "TODO";
+
+    desc.format = wgpu::TextureFormat::Undefined;
+    desc.dimension = wgpu::TextureViewDimension::Undefined;
+    desc.usage = wgpu::TextureUsage::None;
+
+    return texture.CreateView(&desc);
+}
+
+}// namespace filament::backend

filament/backend/src/webgpu/WebGPUHandles.h

@@ -0,0 +1,224 @@
+/*
+* 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
+
+#include "DriverBase.h"
+
+#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;
+
+// VertexBufferInfo contains layout info for Vertex Buffer based on WebGPU structs. In WebGPU each
+// VertexBufferLayout is associated with a single vertex buffer. So number of mVertexBufferLayout
+// is equal to bufferCount. Each VertexBufferLayout can contain multiple VertexAttribute. Bind index
+// of vertex buffer is implicitly calculated by the position of VertexBufferLayout in an array.
+class WGPUVertexBufferInfo : public HwVertexBufferInfo {
+public:
+    WGPUVertexBufferInfo(uint8_t bufferCount, uint8_t attributeCount,
+            AttributeArray const& attributes);
+    inline  wgpu::VertexBufferLayout const* getVertexBufferLayout() const {
+        return mVertexBufferLayout.data();
+    }
+
+    inline uint32_t getVertexBufferLayoutSize() const {
+        return mVertexBufferLayout.size();
+    }
+
+    inline wgpu::VertexAttribute const* getVertexAttributeForIndex(uint32_t index) const {
+        assert_invariant(index < mAttributes.size());
+        return mAttributes[index].data();
+    }
+
+    inline uint32_t getVertexAttributeSize(uint32_t index) const {
+        assert_invariant(index < mAttributes.size());
+        return mAttributes[index].size();
+    }
+
+private:
+    // TODO: can we do better in terms on heap management.
+    std::vector<wgpu::VertexBufferLayout> mVertexBufferLayout {};
+    std::vector<std::vector<wgpu::VertexAttribute>> mAttributes {};
+};
+
+struct WGPUVertexBuffer : public HwVertexBuffer {
+    WGPUVertexBuffer(wgpu::Device const &device, uint32_t vertexCount, uint32_t bufferCount,
+                     Handle<HwVertexBufferInfo> vbih);
+
+    Handle<HwVertexBufferInfo> vbih;
+    utils::FixedCapacityVector<wgpu::Buffer> buffers;
+};
+
+struct WGPUIndexBuffer : public HwIndexBuffer {
+    WGPUIndexBuffer(wgpu::Device const &device, uint8_t elementSize,
+                    uint32_t indexCount);
+
+    wgpu::Buffer buffer;
+    wgpu::IndexFormat indexFormat;
+};
+
+struct WGPUBufferObject : HwBufferObject {
+    WGPUBufferObject(wgpu::Device const &device, BufferObjectBinding bindingType, uint32_t byteCount);
+
+    wgpu::Buffer buffer = nullptr;
+    const BufferObjectBinding bufferObjectBinding;
+};
+class WebGPUDescriptorSetLayout final : public HwDescriptorSetLayout {
+public:
+    WebGPUDescriptorSetLayout(DescriptorSetLayout const& layout, wgpu::Device const& device);
+    ~WebGPUDescriptorSetLayout();
+    [[nodiscard]] const wgpu::BindGroupLayout& getLayout() const { return mLayout; }
+    [[nodiscard]] uint getLayoutSize() const { return mLayoutSize; }
+
+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);
+    uint mLayoutSize;
+    wgpu::BindGroupLayout mLayout;
+};
+
+class WebGPUDescriptorSet final : public HwDescriptorSet {
+public:
+    WebGPUDescriptorSet(const wgpu::BindGroupLayout& layout, uint layoutSize);
+    ~WebGPUDescriptorSet();
+
+    wgpu::BindGroup lockAndReturn(wgpu::Device const& device);
+    void addEntry(uint index, wgpu::BindGroupEntry&& entry);
+    [[nodiscard]] bool getIsLocked() const { return mBindGroup != nullptr; }
+
+private:
+    // TODO: Consider storing what we used to make the layout. However we need to essentially
+    // Recreate some of the info (Sampler in slot X with the actual sampler) so letting Dawn confirm
+    // there isn't a mismatch may be easiest.
+    // Also storing the wgpu ObjectBase takes care of ownership challenges in theory
+    wgpu::BindGroupLayout mLayout;
+    std::vector<wgpu::BindGroupEntry> entries;
+    wgpu::BindGroup mBindGroup;
+};
+
+class WGPUTexture : public HwTexture {
+public:
+    WGPUTexture(SamplerType target, uint8_t levels, TextureFormat format, uint8_t samples,
+            uint32_t width, uint32_t height, uint32_t depth, TextureUsage usage,
+            wgpu::Device device) noexcept;
+
+    WGPUTexture(WGPUTexture* src, uint8_t baseLevel, uint8_t levelCount) noexcept;
+
+    const wgpu::Texture& getTexture() const { return texture; }
+    const wgpu::Sampler& getSampler() const { return sampler; }
+    const wgpu::TextureView& getTexView() const { return texView; }
+
+    static wgpu::TextureFormat fToWGPUTextureFormat(const filament::backend::TextureFormat& fUsage);
+
+private:
+    wgpu::TextureView makeTextureView(const uint8_t& baseLevel, const uint8_t& levelCount);
+    // CreateTextureR has info for a texture and sampler. Texture Views are needed for binding,
+    // along with a sampler Current plan: Inherit the sampler and Texture to always exist (It is a
+    // ref counted pointer) when making views. View is optional
+    wgpu::Texture texture = nullptr;
+    // TODO: Adding this but not yet setting it up. Filament "Textures" are combined image samplers,
+    // rep both.
+    wgpu::Sampler sampler = nullptr;
+    // TODO: Not sure all the ways HwTexture is used. Overloading like this might be entirely wrong.
+    wgpu::TextureView texView = nullptr;
+    wgpu::TextureUsage fToWGPUTextureUsage(const filament::backend::TextureUsage& fUsage);
+};
+
+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 struct WGPURenderTarget;
+
+        Attachment() = default;
+        Attachment(WGPUTexture* gpuTexture, uint8_t level = 0, uint16_t layer = 0)
+            : level(level),
+              layer(layer),
+              texture(gpuTexture->getTexture()),
+              mWGPUTexture(gpuTexture) {}
+
+        uint8_t level = 0;
+        uint16_t layer = 0;
+
+    private:
+        wgpu::Texture texture = nullptr;
+        WGPUTexture* mWGPUTexture = nullptr;
+    };
+
+    WGPURenderTarget(uint32_t width, uint32_t height, uint8_t samples,
+            Attachment colorAttachments[MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT]);
+    WGPURenderTarget()
+        : HwRenderTarget(0, 0),
+          defaultRenderTarget(true) {}
+
+    void setUpRenderPassAttachments(wgpu::RenderPassDescriptor* descriptor,
+            const RenderPassParams& params);
+
+    math::uint2 getAttachmentSize() noexcept;
+
+    bool isDefaultRenderTarget() const { return defaultRenderTarget; }
+    uint8_t getSamples() const { return samples; }
+
+    Attachment getDrawColorAttachment(size_t index);
+    Attachment getReadColorAttachment(size_t index);
+
+private:
+    static wgpu::LoadOp getLoadAction(const RenderPassParams& params, TargetBufferFlags buffer);
+    static wgpu::LoadOp getStoreAction(const RenderPassParams& params, TargetBufferFlags buffer);
+
+    bool defaultRenderTarget = false;
+    uint8_t samples = 1;
+
+    Attachment color[MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT] = {};
+    math::uint2 attachmentSize = {};
+};
+
+}// namespace filament::backend
+#endif// TNT_FILAMENT_BACKEND_WEBGPUHANDLES_H

filament/backend/src/webgpu/WebGPUPipelineCreation.cpp

@@ -0,0 +1,254 @@
+/*
+ * 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 "WebGPUPipelineCreation.h"
+
+#include "WebGPUHandles.h"
+
+#include <backend/DriverEnums.h>
+#include <backend/TargetBufferInfo.h>
+
+#include <utils/Panic.h>
+#include <utils/debug.h>
+
+#include <webgpu/webgpu_cpp.h>
+
+#include <array>
+#include <cstdint>
+#include <sstream>
+
+namespace filament::backend {
+
+namespace {
+
+constexpr wgpu::PrimitiveTopology toWebGPU(PrimitiveType primitiveType) {
+    switch (primitiveType) {
+        case PrimitiveType::POINTS:
+            return wgpu::PrimitiveTopology::PointList;
+        case PrimitiveType::LINES:
+            return wgpu::PrimitiveTopology::LineList;
+        case PrimitiveType::LINE_STRIP:
+            return wgpu::PrimitiveTopology::LineStrip;
+        case PrimitiveType::TRIANGLES:
+            return wgpu::PrimitiveTopology::TriangleList;
+        case PrimitiveType::TRIANGLE_STRIP:
+            return wgpu::PrimitiveTopology::TriangleStrip;
+    }
+}
+
+constexpr wgpu::CullMode toWebGPU(CullingMode cullMode) {
+    switch (cullMode) {
+        case CullingMode::NONE:
+            return wgpu::CullMode::None;
+        case CullingMode::FRONT:
+            return wgpu::CullMode::Front;
+        case CullingMode::BACK:
+            return wgpu::CullMode::Back;
+        case CullingMode::FRONT_AND_BACK:
+            // no WegGPU equivalent of front and back
+            FILAMENT_CHECK_POSTCONDITION(false)
+                    << "WebGPU does not support CullingMode::FRONT_AND_BACK";
+            return wgpu::CullMode::Undefined;
+    }
+}
+
+constexpr wgpu::CompareFunction toWebGPU(SamplerCompareFunc compareFunction) {
+    switch (compareFunction) {
+        case SamplerCompareFunc::LE:
+            return wgpu::CompareFunction::LessEqual;
+        case SamplerCompareFunc::GE:
+            return wgpu::CompareFunction::GreaterEqual;
+        case SamplerCompareFunc::L:
+            return wgpu::CompareFunction::Less;
+        case SamplerCompareFunc::G:
+            return wgpu::CompareFunction::Greater;
+        case SamplerCompareFunc::E:
+            return wgpu::CompareFunction::Equal;
+        case SamplerCompareFunc::NE:
+            return wgpu::CompareFunction::NotEqual;
+        case SamplerCompareFunc::A:
+            return wgpu::CompareFunction::Always;
+        case SamplerCompareFunc::N:
+            return wgpu::CompareFunction::Never;
+    }
+}
+
+constexpr wgpu::StencilOperation toWebGPU(StencilOperation stencilOp) {
+    switch (stencilOp) {
+        case StencilOperation::KEEP:
+            return wgpu::StencilOperation::Keep;
+        case StencilOperation::ZERO:
+            return wgpu::StencilOperation::Zero;
+        case StencilOperation::REPLACE:
+            return wgpu::StencilOperation::Replace;
+        case StencilOperation::INCR:
+            return wgpu::StencilOperation::IncrementClamp;
+        case StencilOperation::INCR_WRAP:
+            return wgpu::StencilOperation::IncrementWrap;
+        case StencilOperation::DECR:
+            return wgpu::StencilOperation::DecrementClamp;
+        case StencilOperation::DECR_WRAP:
+            return wgpu::StencilOperation::DecrementWrap;
+        case StencilOperation::INVERT:
+            return wgpu::StencilOperation::Invert;
+    }
+}
+
+constexpr wgpu::BlendOperation toWebGPU(BlendEquation blendOp) {
+    switch (blendOp) {
+        case BlendEquation::ADD:
+            return wgpu::BlendOperation::Add;
+        case BlendEquation::SUBTRACT:
+            return wgpu::BlendOperation::Subtract;
+        case BlendEquation::REVERSE_SUBTRACT:
+            return wgpu::BlendOperation::ReverseSubtract;
+        case BlendEquation::MIN:
+            return wgpu::BlendOperation::Min;
+        case BlendEquation::MAX:
+            return wgpu::BlendOperation::Max;
+    }
+}
+
+constexpr wgpu::BlendFactor toWebGPU(BlendFunction blendFunction) {
+    switch (blendFunction) {
+        case BlendFunction::ZERO:
+            return wgpu::BlendFactor::Zero;
+        case BlendFunction::ONE:
+            return wgpu::BlendFactor::One;
+        case BlendFunction::SRC_COLOR:
+            return wgpu::BlendFactor::Src;
+        case BlendFunction::ONE_MINUS_SRC_COLOR:
+            return wgpu::BlendFactor::OneMinusSrc;
+        case BlendFunction::DST_COLOR:
+            return wgpu::BlendFactor::Dst;
+        case BlendFunction::ONE_MINUS_DST_COLOR:
+            return wgpu::BlendFactor::OneMinusDst;
+        case BlendFunction::SRC_ALPHA:
+            return wgpu::BlendFactor::SrcAlpha;
+        case BlendFunction::ONE_MINUS_SRC_ALPHA:
+            return wgpu::BlendFactor::OneMinusSrcAlpha;
+        case BlendFunction::DST_ALPHA:
+            return wgpu::BlendFactor::DstAlpha;
+        case BlendFunction::ONE_MINUS_DST_ALPHA:
+            return wgpu::BlendFactor::OneMinusDstAlpha;
+        case BlendFunction::SRC_ALPHA_SATURATE:
+            return wgpu::BlendFactor::SrcAlphaSaturated;
+    }
+}
+
+}// namespace
+
+wgpu::RenderPipeline createWebGPURenderPipeline(wgpu::Device const& device,
+        WGPUProgram const& program, WGPUVertexBufferInfo const& vertexBufferInfo,
+        wgpu::PipelineLayout const& layout, RasterState const& rasterState,
+        StencilState const& stencilState, PolygonOffset const& polygonOffset,
+        PrimitiveType primitiveType, wgpu::TextureFormat colorFormat,
+        wgpu::TextureFormat depthFormat) {
+    assert_invariant(program.vertexShaderModule);
+    const wgpu::DepthStencilState depthStencilState {
+        .format = depthFormat,
+        .depthWriteEnabled = rasterState.depthWrite,
+        .depthCompare = toWebGPU(rasterState.depthFunc),
+        .stencilFront = {
+            .compare = toWebGPU(stencilState.front.stencilFunc),
+            .failOp = toWebGPU(stencilState.front.stencilOpStencilFail),
+            .depthFailOp = toWebGPU(stencilState.front.stencilOpDepthFail),
+            .passOp = toWebGPU(stencilState.front.stencilOpDepthStencilPass),
+        },
+        .stencilBack = {
+            .compare = toWebGPU(stencilState.back.stencilFunc),
+            .failOp = toWebGPU(stencilState.back.stencilOpStencilFail),
+            .depthFailOp = toWebGPU(stencilState.back.stencilOpDepthFail),
+            .passOp = toWebGPU(stencilState.back.stencilOpDepthStencilPass),
+        },
+        .stencilReadMask = 0,
+        .stencilWriteMask = stencilState.stencilWrite ? 0xFFFFFFFF : 0,
+        .depthBias = static_cast<int32_t>(polygonOffset.constant),
+        .depthBiasSlopeScale = polygonOffset.slope,
+        .depthBiasClamp = 0.0f
+    };
+    std::stringstream pipelineLabelStream;
+    pipelineLabelStream << program.name.c_str() << " pipeline";
+    const auto pipelineLabel = pipelineLabelStream.str();
+    wgpu::RenderPipelineDescriptor pipelineDescriptor{
+        .label = wgpu::StringView(pipelineLabel),
+        .layout = layout,
+        .vertex = {
+            .module = program.vertexShaderModule,
+            .entryPoint = "main",
+            .constantCount = program.constants.size(),
+            .constants = program.constants.data(),
+            .bufferCount = vertexBufferInfo.getVertexBufferLayoutSize(),
+            .buffers = vertexBufferInfo.getVertexBufferLayout()
+        },
+        .primitive = {
+            .topology = toWebGPU(primitiveType),
+            // TODO should we assume some constant format here or is there a way to get
+            //      this from PipelineState somehow or elsewhere?
+            //      Perhaps, cache/assert format from index buffers as they are requested?
+            .stripIndexFormat = wgpu::IndexFormat::Undefined,
+            .frontFace = rasterState.inverseFrontFaces ? wgpu::FrontFace::CW : wgpu::FrontFace::CCW,
+            .cullMode = toWebGPU(rasterState.culling),
+            // TODO no depth clamp in WebGPU supported directly. unclippedDepth is close, so we are
+            //      starting there
+            .unclippedDepth = !rasterState.depthClamp &&
+                              device.HasFeature(wgpu::FeatureName::DepthClipControl)
+        },
+        .depthStencil = &depthStencilState,
+        .multisample = {
+            .count = 1, // TODO need to get this from the render target
+            .mask = 0xFFFFFFFF,
+            .alphaToCoverageEnabled = rasterState.alphaToCoverage
+        },
+        .fragment = nullptr // will add below if fragment module is included
+    };
+    wgpu::FragmentState fragmentState = {};
+    std::array<wgpu::ColorTargetState, MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT> colorTargets {};
+    const wgpu::BlendState blendState {
+        .color = {
+            .operation = toWebGPU(rasterState.blendEquationRGB),
+            .srcFactor = toWebGPU(rasterState.blendFunctionSrcRGB),
+            .dstFactor = toWebGPU(rasterState.blendFunctionDstRGB)
+        },
+        .alpha = {
+            .operation = toWebGPU(rasterState.blendEquationAlpha),
+            .srcFactor = toWebGPU(rasterState.blendFunctionSrcAlpha),
+            .dstFactor = toWebGPU(rasterState.blendFunctionDstAlpha)
+        }
+    };
+    if (program.fragmentShaderModule != nullptr) {
+        fragmentState.module = program.fragmentShaderModule;
+        fragmentState.entryPoint = "main";
+        fragmentState.constantCount = program.constants.size(),
+        fragmentState.constants = program.constants.data(),
+        fragmentState.targetCount = 1; // TODO need to get this from the render target
+        assert_invariant(fragmentState.targetCount <= MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT);
+        for (size_t targetIndex = 0; targetIndex < fragmentState.targetCount; targetIndex++) {
+            auto& colorTarget = colorTargets[targetIndex];
+            colorTarget.format = colorFormat;
+            colorTarget.blend = rasterState.hasBlending() ? &blendState : nullptr;
+            colorTarget.writeMask =
+                    rasterState.colorWrite ? wgpu::ColorWriteMask::All : wgpu::ColorWriteMask::None;
+        }
+        pipelineDescriptor.fragment = &fragmentState;
+    }
+    const wgpu::RenderPipeline pipeline = device.CreateRenderPipeline(&pipelineDescriptor);
+    FILAMENT_CHECK_POSTCONDITION(pipeline)
+            << "Failed to create render pipeline for " << pipelineDescriptor.label;
+    return pipeline;
+}
+
+}// namespace filament::backend

filament/backend/src/webgpu/WebGPUPipelineCreation.h

@@ -0,0 +1,46 @@
+/*
+ * 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_WEBGPUPIPELINECREATION_H
+#define TNT_FILAMENT_BACKEND_WEBGPUPIPELINECREATION_H
+
+#include <cstdint>
+
+namespace wgpu {
+class Device;
+class PipelineLayout;
+class RenderPipeline;
+enum class TextureFormat : uint32_t;
+}// namespace wgpu
+
+namespace filament::backend {
+
+struct PolygonOffset;
+enum class PrimitiveType : uint8_t;
+struct RasterState;
+struct StencilState;
+
+class WGPUVertexBufferInfo;
+class WGPUProgram;
+
+[[nodiscard]] wgpu::RenderPipeline createWebGPURenderPipeline(wgpu::Device const&,
+        WGPUProgram const&, WGPUVertexBufferInfo const&, wgpu::PipelineLayout const&,
+        RasterState const&, StencilState const&, PolygonOffset const&, PrimitiveType,
+        wgpu::TextureFormat colorFormat, wgpu::TextureFormat depthFormat);
+
+}// namespace filament::backend
+
+#endif// TNT_FILAMENT_BACKEND_WEBGPUPIPELINECREATION_H

filament/backend/src/webgpu/WebGPUSwapChain.cpp

@@ -73,7 +73,8 @@ void printSurfaceCapabilitiesDetails(wgpu::SurfaceCapabilities const& capabiliti
 #endif
 
 #if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
-void printSurfaceConfiguration(wgpu::SurfaceConfiguration const& config) {
+void printSurfaceConfiguration(wgpu::SurfaceConfiguration const& config,
+        wgpu::TextureFormat depthFormat) {
     std::stringstream formatStream{};
     formatStream << config.format;
     std::stringstream usageStream{};
@@ -82,6 +83,8 @@ void printSurfaceConfiguration(wgpu::SurfaceConfiguration const& config) {
     alphaModeStream << config.alphaMode;
     std::stringstream presentModeStream{};
     presentModeStream << config.presentMode;
+    std::stringstream depthFormatStream;
+    depthFormatStream << depthFormat;
     FWGPU_LOGI << "WebGPU surface configuration:" << utils::io::endl;
     FWGPU_LOGI << "  surface format: " << formatStream.str() << utils::io::endl;
     FWGPU_LOGI << "  surface usage: " << usageStream.str() << utils::io::endl;
@@ -98,10 +101,11 @@ void printSurfaceConfiguration(wgpu::SurfaceConfiguration const& config) {
     FWGPU_LOGI << "  surface width: " << config.width << utils::io::endl;
     FWGPU_LOGI << "  surface height: " << config.height << utils::io::endl;
     FWGPU_LOGI << "  surface present mode: " << presentModeStream.str() << utils::io::endl;
+    FWGPU_LOGI << "WebGPU selected depth format: " << depthFormatStream.str() << utils::io::endl;
 }
 #endif
 
-wgpu::TextureFormat selectColorFormat(size_t availableFormatsCount,
+constexpr wgpu::TextureFormat selectColorFormat(size_t availableFormatsCount,
         wgpu::TextureFormat const* availableFormats, bool useSRGBColorSpace) {
     const std::array expectedColorFormats =
             useSRGBColorSpace ?
@@ -119,7 +123,21 @@ wgpu::TextureFormat selectColorFormat(size_t availableFormatsCount,
     return *firstFoundColorFormat;
 }
 
-wgpu::PresentMode selectPresentMode(size_t availablePresentModesCount,
+constexpr wgpu::TextureFormat selectDepthFormat(bool depth32FloatStencil8Enabled,
+        bool needStencil) {
+    if (needStencil) {
+        if (depth32FloatStencil8Enabled) {
+            return wgpu::TextureFormat::Depth32FloatStencil8;
+        } else {
+            return wgpu::TextureFormat::Depth24PlusStencil8;
+        }
+    } else {
+        // other options: Depth16Unorm or Depth24Plus
+        return wgpu::TextureFormat::Depth32Float;
+    }
+}
+
+constexpr wgpu::PresentMode selectPresentMode(size_t availablePresentModesCount,
         wgpu::PresentMode const* availablePresentModes) {
     // Verify that our chosen present mode is supported. In practice all devices support the FIFO
     // mode, but we check for it anyway for completeness.  (and to avoid validation warnings)
@@ -133,7 +151,7 @@ wgpu::PresentMode selectPresentMode(size_t availablePresentModesCount,
     return desiredPresentMode;
 }
 
-wgpu::CompositeAlphaMode selectAlphaMode(size_t availableAlphaModesCount,
+constexpr wgpu::CompositeAlphaMode selectAlphaMode(size_t availableAlphaModesCount,
         wgpu::CompositeAlphaMode const* availableAlphaModes) {
     bool autoAvailable = false;
     bool inheritAvailable = false;
@@ -190,10 +208,13 @@ 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 +226,8 @@ 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 const& adapter, wgpu::Device const& device, uint64_t flags)
     : mSurface(surface) {
     wgpu::SurfaceCapabilities capabilities = {};
     if (!mSurface.GetCapabilities(adapter, &capabilities)) {
@@ -217,33 +238,63 @@ WebGPUSwapChain::WebGPUSwapChain(wgpu::Surface&& surface, wgpu::Adapter& adapter
 #endif
     }
     const bool useSRGBColorSpace = (flags & SWAP_CHAIN_CONFIG_SRGB_COLORSPACE) != 0;
-    initConfig(mConfig, device, capabilities, useSRGBColorSpace);
+    const bool needStencil = (flags & SWAP_CHAIN_HAS_STENCIL_BUFFER) != 0;
+    initConfig(mConfig, device, capabilities, surfaceSize, useSRGBColorSpace);
+    mDepthFormat = selectDepthFormat(device.HasFeature(wgpu::FeatureName::Depth32FloatStencil8),
+            needStencil);
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+    printSurfaceConfiguration(mConfig, mDepthFormat);
+#endif
+    mSurface.Configure(&mConfig);
 }
 
 WebGPUSwapChain::~WebGPUSwapChain() {
-    if (mConfigured) {
-        mSurface.Unconfigure();
-        mConfigured = false;
+    mSurface.Unconfigure();
+}
+
+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;
+        FWGPU_LOGD << "Resizing to width " << mConfig.width << " height " << mConfig.height
+                   << utils::io::endl;
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+        printSurfaceConfiguration(mConfig, mDepthFormat);
+#endif
+        // TODO we may need to ensure no surface texture is in flight when we do this. some
+        //      synchronization may be necessary
+        mSurface.Configure(&mConfig);
     }
 }
 
-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;
-#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
-        printSurfaceConfiguration(mConfig);
-#endif
-        mSurface.Configure(&mConfig);
-        mConfigured = true;
-        return;
+wgpu::TextureView WebGPUSwapChain::getCurrentSurfaceTextureView(
+        wgpu::Extent2D const& currentSurfaceSize) {
+    setExtent(currentSurfaceSize);
+    wgpu::SurfaceTexture surfaceTexture;
+    mSurface.GetCurrentTexture(&surfaceTexture);
+    if (surfaceTexture.status != wgpu::SurfaceGetCurrentTextureStatus::SuccessOptimal) {
+        return nullptr;
     }
+    // Create a view for this surface texture
+    // TODO: review these initiliazations as webgpu pipeline gets mature
+    wgpu::TextureViewDescriptor textureViewDescriptor = {
+        .label = "surface_texture_view",
+        .format = surfaceTexture.texture.GetFormat(),
+        .dimension = wgpu::TextureViewDimension::e2D,
+        .baseMipLevel = 0,
+        .mipLevelCount = 1,
+        .baseArrayLayer = 0,
+        .arrayLayerCount = 1
+    };
+    return surfaceTexture.texture.CreateView(&textureViewDescriptor);
+}
 
-    mSurface.GetCurrentTexture(texture);
+void WebGPUSwapChain::present() {
+    assert_invariant(mSurface);
+    mSurface.Present();
 }
 
 }// namespace filament::backend

filament/backend/src/webgpu/WebGPUSwapChain.h

@@ -19,24 +19,33 @@
 
 #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 const& adapter, wgpu::Device const& device, uint64_t flags);
     ~WebGPUSwapChain();
 
-    void GetCurrentTexture(uint32_t width, uint32_t height, wgpu::SurfaceTexture*);
+    [[nodiscard]] wgpu::TextureFormat getColorFormat() const { return mConfig.format; }
+
+    [[nodiscard]] wgpu::TextureFormat getDepthFormat() const { return mDepthFormat; }
+
+    [[nodiscard]] wgpu::TextureView getCurrentSurfaceTextureView(wgpu::Extent2D const&);
+
+    void present();
 
 private:
+    void setExtent(wgpu::Extent2D const&);
+
     wgpu::Surface mSurface = {};
     wgpu::SurfaceConfiguration mConfig = {};
-    bool mConfigured = false;
+    wgpu::TextureFormat mDepthFormat = wgpu::TextureFormat::Undefined;
 };
 
 } // namespace filament::backend

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

@@ -98,10 +98,23 @@ wgpu::Adapter WebGPUPlatform::requestAdapter(wgpu::Surface const& surface) {
 }
 
 wgpu::Device WebGPUPlatform::requestDevice(wgpu::Adapter const& adapter) {
-    // TODO consider passing required features and/or limits
+    // TODO consider passing limits
+    constexpr std::array desiredFeatures = {
+        wgpu::FeatureName::DepthClipControl,
+        wgpu::FeatureName::Depth32FloatStencil8,
+        wgpu::FeatureName::CoreFeaturesAndLimits };
+    std::vector<wgpu::FeatureName> requiredFeatures;
+    requiredFeatures.reserve(desiredFeatures.size());
+    wgpu::SupportedFeatures supportedFeatures;
+    adapter.GetFeatures(&supportedFeatures);
+    std::set_intersection(supportedFeatures.features,
+            supportedFeatures.features + supportedFeatures.featureCount, desiredFeatures.begin(),
+            desiredFeatures.end(), std::back_inserter(requiredFeatures));
     wgpu::DeviceDescriptor deviceDescriptor{};
     deviceDescriptor.label = "graphics_device";
     deviceDescriptor.defaultQueue.label = "default_queue";
+    deviceDescriptor.requiredFeatureCount = requiredFeatures.size();
+    deviceDescriptor.requiredFeatures = requiredFeatures.data();
     deviceDescriptor.SetDeviceLostCallback(wgpu::CallbackMode::AllowSpontaneous,
             [](wgpu::Device const&, wgpu::DeviceLostReason const& reason,
                     wgpu::StringView message) {

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;
 }