Sampler creation as part of the driver

BUGS=[409099263]

.clang-format

@@ -0,0 +1,68 @@
+BasedOnStyle: LLVM
+AccessModifierOffset: -4
+AlignAfterOpenBracket: DontAlign
+AlignConsecutiveAssignments: None
+AlignOperands: Align
+AllowAllArgumentsOnNextLine: false
+AllowAllConstructorInitializersOnNextLine: false
+AllowAllParametersOfDeclarationOnNextLine: true
+AllowShortBlocksOnASingleLine: false
+AllowShortCaseLabelsOnASingleLine: false
+AllowShortFunctionsOnASingleLine: All
+AllowShortIfStatementsOnASingleLine: Always
+AllowShortLambdasOnASingleLine: All
+AllowShortLoopsOnASingleLine: true
+AlwaysBreakAfterReturnType: None
+AlwaysBreakTemplateDeclarations: Yes
+BreakBeforeBraces: Custom
+BracedInitializerIndentWidth: 4
+BraceWrapping:
+  AfterCaseLabel: false
+  AfterClass: false
+  AfterControlStatement: Never
+  AfterEnum: false
+  AfterFunction: false
+  AfterNamespace: false
+  AfterUnion: false
+  BeforeCatch: false
+  BeforeElse: false
+  IndentBraces: false
+  SplitEmptyFunction: false
+  SplitEmptyRecord: true
+BreakBeforeBinaryOperators: None
+BreakBeforeTernaryOperators: true
+BreakConstructorInitializers: BeforeColon
+BreakInheritanceList: BeforeColon
+ColumnLimit: 100
+CompactNamespaces: false
+ContinuationIndentWidth: 8
+Cpp11BracedListStyle: false
+IndentCaseLabels: true
+IndentPPDirectives: None
+IndentWidth: 4
+KeepEmptyLinesAtTheStartOfBlocks: true
+MaxEmptyLinesToKeep: 2
+NamespaceIndentation: None
+ObjCSpaceAfterProperty: false
+ObjCSpaceBeforeProtocolList: true
+PointerAlignment: Left
+ReflowComments: true
+SpaceAfterCStyleCast: true
+SpaceAfterLogicalNot: false
+SpaceAfterTemplateKeyword: false
+SpaceBeforeAssignmentOperators: true
+SpaceBeforeCpp11BracedList: false
+SpaceBeforeCtorInitializerColon: true
+SpaceBeforeInheritanceColon: true
+SpaceBeforeParens: ControlStatements
+SpaceBeforeRangeBasedForLoopColon: false
+SpaceInEmptyParentheses: false
+SpacesBeforeTrailingComments: 0
+SpacesInAngles: false
+SpacesInCStyleCastParentheses: false
+SpacesInContainerLiterals: false
+SpacesInParentheses: false
+SpacesInSquareBrackets: false
+TabWidth: 4
+UseTab: Never
+PackConstructorInitializers: Never

.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

@@ -12,10 +12,13 @@ jobs:
     name: build-android
     # We intentially use a larger runner here to enable larger disk space
     # (standard linux runner will fail on disk space and faster build time).
-    runs-on: ubuntu-22.04-32core
+    runs-on: 'ubuntu-24.04-32core'
 
     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

@@ -10,10 +10,13 @@ on:
 jobs:
   build-linux:
     name: build-linux
-    runs-on: ubuntu-22.04-16core
+    runs-on: 'ubuntu-24.04-16core'
 
     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 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

@@ -9,20 +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
@@ -33,6 +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
@@ -40,10 +54,13 @@ jobs:
 
   build-android:
     name: build-android
-    runs-on: ubuntu-22.04-16core
+    runs-on: 'ubuntu-24.04-16core'
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
       - uses: actions/setup-java@v3
         with:
           distribution: 'temurin'
@@ -60,6 +77,9 @@ 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 presubmit
@@ -69,17 +89,20 @@ jobs:
 
   build-web:
     name: build-web
-    runs-on: ubuntu-22.04-16core
+    runs-on: 'ubuntu-24.04-16core'
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
       - name: Run build script
         run: |
           cd build/web && printf "y" | ./build.sh presubmit
 
   validate-docs:
     name: validate-docs
-    runs-on: ubuntu-22.04-4core
+    runs-on: 'ubuntu-24.04-4core'
     steps:
       - uses: actions/checkout@v4.1.6
         with:
@@ -92,15 +115,58 @@ jobs:
 
   test-renderdiff:
     name: test-renderdiff
-    runs-on: ubuntu-22.04-16core
-
+    runs-on: macos-14-xlarge
     steps:
       - uses: actions/checkout@v4.1.6
-      - uses: ./.github/actions/ubuntu-apt-add-src
-      - name: Run script
-        run: |
-          source ./build/linux/ci-common.sh && bash test/renderdiff/test.sh
+        with:
+          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
           path: ./out/renderdiff_tests
+
+  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: ./build.sh -W debug test_filamat filament
+      - name: Run test
+        run: ./out/cmake-debug/libs/filamat/test_filamat --gtest_filter=MaterialCompiler.Wgsl*
+
+  code-correcteness:
+    name: code-correctness
+    runs-on: 'macos-14-xlarge'
+    steps:
+      - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/mac-prereq
+      - name: Install clang-tidy and deps
+        run: |
+          pip install pyyaml
+          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

.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 }}
@@ -70,7 +103,7 @@ jobs:
 
   build-web:
     name: build-web
-    runs-on: ubuntu-22.04-16core
+    runs-on: 'ubuntu-24.04-16core'
     if: github.event_name == 'release' || github.event.inputs.platform == 'web'
 
     steps:
@@ -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 }}
@@ -103,7 +137,7 @@ jobs:
 
   build-android:
     name: build-android
-    runs-on: ubuntu-22.04-16core
+    runs-on: 'ubuntu-24.04-16core'
     if: github.event_name == 'release' || github.event.inputs.platform == 'android'
 
     steps:
@@ -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

@@ -10,10 +10,13 @@ on:
 jobs:
   build-web:
     name: build-web
-    runs-on: ubuntu-22.04-16core
+    runs-on: 'ubuntu-24.04-16core'
 
     steps:
       - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
       - name: Run build script
         run: |
           cd build/web && printf "y" | ./build.sh continuous

CMakeLists.txt

@@ -139,14 +139,14 @@ else()
     set(LINUX FALSE)
 endif()
 
-if (LINUX)
-    if (NOT FILAMENT_OSMESA_PATH STREQUAL "")
-        if (NOT EXISTS ${FILAMENT_OSMESA_PATH}/)
-            message(FATAL_ERROR "Cannot find specified OSMesa build directory: ${FILAMENT_OSMESA_PATH}")
-        endif()
-        set(FILAMENT_SUPPORTS_OSMESA TRUE)
+if (NOT FILAMENT_OSMESA_PATH STREQUAL "")
+    if (NOT EXISTS ${FILAMENT_OSMESA_PATH}/)
+        message(FATAL_ERROR "Cannot find specified OSMesa build directory: ${FILAMENT_OSMESA_PATH}")
     endif()
+    set(FILAMENT_SUPPORTS_OSMESA TRUE)
+endif()
 
+if (LINUX)
     if (FILAMENT_SUPPORTS_WAYLAND)
         add_definitions(-DFILAMENT_SUPPORTS_WAYLAND)
         set(FILAMENT_SUPPORTS_X11 FALSE)
@@ -184,6 +184,12 @@ if (NOT ANDROID AND NOT WEBGL AND NOT IOS AND NOT FILAMENT_LINUX_IS_MOBILE)
     set(IS_HOST_PLATFORM TRUE)
 endif()
 
+if (APPLE)
+    if (FILAMENT_SUPPORTS_OSMESA)
+        add_definitions(-DFILAMENT_SUPPORTS_OSMESA)
+    endif()
+endif()
+
 if (WIN32)
     # Link statically against c/c++ lib to avoid missing redistriburable such as
     # "VCRUNTIME140.dll not found. Try reinstalling the app.", but give users
@@ -328,6 +334,9 @@ if (MSVC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${CXX_STANDARD} /W0 /Zc:__cplusplus")
 else()
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${CXX_STANDARD} -fstrict-aliasing -Wno-unknown-pragmas -Wno-unused-function -Wno-deprecated-declarations")
+    if (APPLE)
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-nullability-extension")
+    endif()
 endif()
 
 if (FILAMENT_USE_EXTERNAL_GLES3)
@@ -601,6 +610,14 @@ if (FILAMENT_SUPPORTS_METAL)
     set(MATC_API_FLAGS ${MATC_API_FLAGS} -a metal)
 endif()
 
+# With WebGPU, push constants are not supported. Skinning uses them.
+# WebGPU has a proposal to add push constants at https://github.com/gpuweb/gpuweb/blob/main/proposals/push-constants.md
+# With WebGPU, Tint does not support ClipDistance which is used in Stereo.  Mentioned in comment
+# https://github.com/google/dawn/blob/855d17b08abdf02f9142bf5a8f14d0ea088810a4/src/tint/lang/spirv/reader/ast_parser/function.cc#L4434
+if (FILAMENT_SUPPORTS_WEBGPU)
+    set(MATC_API_FLAGS ${MATC_API_FLAGS} -a webgpu --variant-filter=skinning,stereo)
+endif()
+
 # Disable ESSL 1.0 code generation.
 if (NOT FILAMENT_ENABLE_FEATURE_LEVEL_0)
     set(MATC_API_FLAGS ${MATC_API_FLAGS} -1)
@@ -784,6 +801,7 @@ add_subdirectory(${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)

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.58.1'
+    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', '~> 1.58.1'
+pod 'Filament', '~> 1.59.3'
 ```
 
 ## Documentation

RELEASE_NOTES.md

@@ -7,8 +7,32 @@ 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
+
+
+## v1.59.0
+
+- materials: five custom variables (varyings) are now available on the condition that the `color` attribute is not requested (b/404930099). [⚠️ **New Material Version**]
+
+## v1.58.2
+
+- engine: Generate 1D instead of 3D LUTs for color grading whenever possible.
+
 ## v1.58.1
 
+
+## v1.58.0
+
 - materials: add an optional `transformName` field to sampler material parameters. [⚠️ **New Material Version**]
 
 ## v1.57.3

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

@@ -142,6 +142,12 @@ abstract class MaterialCompiler extends TaskWithBinary {
                 if (!exclude_vulkan) {
                     matcArgs += ['-a', 'vulkan']
                 }
+                def include_webgpu = providers
+                        .gradleProperty("com.google.android.filament.include-webgpu")
+                        .forUseAtConfigurationTime().present
+                if (include_webgpu) {
+                    matcArgs += ['-a', 'webgpu', '--variant-filter=skinning,stereo']
+                }
 
                 def mat_no_opt = providers
                         .gradleProperty("com.google.android.filament.matnopt")

android/filamat-android/CMakeLists.txt

@@ -11,6 +11,13 @@ if (FILAMENT_SUPPORTS_VULKAN)
     message("Library filamat ignores Vulkan settings")
 endif()
 
+if (FILAMENT_SUPPORTS_WEBGPU)
+    # See third_party/dawn/tnt/CMakeLists.txt and libs/filamat/CMakeLists.txt
+    add_library(tint STATIC IMPORTED)
+    set_target_properties(tint PROPERTIES IMPORTED_LOCATION
+            ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libtint_combined.a)
+endif()
+
 add_library(${FILAMAT_FLAVOR} STATIC IMPORTED)
 set_target_properties(${FILAMAT_FLAVOR} PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/lib${FILAMAT_FLAVOR}.a)
@@ -19,6 +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)
@@ -33,6 +44,7 @@ set_target_properties(shaders PROPERTIES IMPORTED_LOCATION
 
 set(FILAMAT_INCLUDE_DIRS
         ../../libs/utils/include
+        ../../third_party/perfetto
         )
 
 include_directories(${FILAMENT_DIR}/include)
@@ -48,6 +60,8 @@ 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)
@@ -29,10 +33,6 @@ add_library(filaflat STATIC IMPORTED)
 set_target_properties(filaflat PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libfilaflat.a)
 
-add_library(filamat STATIC IMPORTED)
-set_target_properties(filamat PROPERTIES IMPORTED_LOCATION
-        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libfilamat.a)
-
 add_library(geometry STATIC IMPORTED)
 set_target_properties(geometry PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libgeometry.a)
@@ -59,11 +59,11 @@ add_library(smol-v STATIC IMPORTED)
 set_target_properties(smol-v PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libsmol-v.a)
 
- if (FILAMENT_ENABLE_FGVIEWER)
-     add_library(fgviewer STATIC IMPORTED)
-     set_target_properties(fgviewer PROPERTIES IMPORTED_LOCATION
-             ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libfgviewer.a)
- endif()
+if (FILAMENT_ENABLE_FGVIEWER)
+    add_library(fgviewer STATIC IMPORTED)
+    set_target_properties(fgviewer PROPERTIES IMPORTED_LOCATION
+            ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libfgviewer.a)
+endif()
 
 if (FILAMENT_ENABLE_MATDBG)
     add_library(matdbg STATIC IMPORTED)
@@ -127,6 +127,7 @@ target_link_libraries(filament-jni
     PRIVATE android
     PRIVATE jnigraphics
     PRIVATE utils
+    PRIVATE perfetto
 
     # libgeometry is PUBLIC because gltfio uses it.
     PUBLIC geometry
@@ -145,6 +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/filament-android/src/main/cpp/Texture.cpp

@@ -68,6 +68,21 @@ Java_com_google_android_filament_Texture_nIsTextureSwizzleSupported(JNIEnv*, jcl
     return (jboolean) Texture::isTextureSwizzleSupported(*engine);
 }
 
+
+extern "C" JNIEXPORT jint JNICALL
+Java_com_google_android_filament_Texture_nGetMaxTextureSize(JNIEnv *, jclass,
+        jlong nativeEngine, jint sampler) {
+    Engine *engine = (Engine *) nativeEngine;
+    return Texture::getMaxTextureSize(*engine, (Texture::Sampler)sampler);
+}
+
+extern "C" JNIEXPORT jint JNICALL
+Java_com_google_android_filament_Texture_nGetMaxArrayTextureLayers(JNIEnv *, jclass,
+        jlong nativeEngine) {
+    Engine *engine = (Engine *) nativeEngine;
+    return Texture::getMaxArrayTextureLayers(*engine);
+}
+
 extern "C" JNIEXPORT jboolean JNICALL
 Java_com_google_android_filament_Texture_nValidatePixelFormatAndType(JNIEnv*, jclass,
         jint internalFormat, jint pixelDataFormat, jint pixelDataType) {

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

@@ -46,9 +46,8 @@ import static com.google.android.filament.Asserts.assertFloat4In;
  *
  * <h1>Performance</h1>
  *
- * Creating a new ColorGrading object may be more expensive than other Filament objects as a
- * 3D LUT may need to be generated. The generation of a 3D LUT, if necessary, may happen on
- * the CPU.
+ * Creating a new ColorGrading object may be more expensive than other Filament objects as a LUT may
+ * need to be generated. The generation of this LUT, if necessary, may happen on the CPU.
  *
  * <h1>Ordering</h1>
  *
@@ -160,6 +159,10 @@ public class ColorGrading {
          * quality level will use a 32x32x32 10 bit LUT, a high quality will use a 32x32x32 16 bit
          * LUT, and a ultra quality will use a 64x64x64 16 bit LUT.
          *
+         * This setting has no effect if generating a 1D LUT.
+         *
+         * This overrides the values set by format() and dimensions().
+         *
          * The default quality is {@link QualityLevel#MEDIUM}.
          *
          * @param qualityLevel The desired quality of the color grading process
@@ -175,6 +178,8 @@ public class ColorGrading {
          * When color grading is implemented using a 3D LUT, this sets the texture format of
          * of the LUT. This overrides the value set by quality().
          *
+         * This setting has no effect if generating a 1D LUT.
+         *
          * The default is INTEGER
          *
          * @param format The desired format of the 3D LUT.
@@ -190,6 +195,8 @@ public class ColorGrading {
          * When color grading is implemented using a 3D LUT, this sets the dimension of the LUT.
          * This overrides the value set by quality().
          *
+         * This setting has no effect if generating a 1D LUT.
+         *
          * The default is 32
          *
          * @param dim The desired dimension of the LUT. Between 16 and 64.
@@ -616,4 +623,3 @@ public class ColorGrading {
 
     private static native long nBuilderBuild(long nativeBuilder, long nativeEngine);
 }
-

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

@@ -691,6 +691,24 @@ public class Texture {
                 pixelDataType.ordinal());
     }
 
+    /**
+     * @param engine {@link Engine}
+     * @param type Texture sampler type
+     * @return The maximum size in texels of a texture of type \p type. At least 2048 for
+     *         2D textures, 256 for 3D textures
+     */
+    public static int getMaxTextureSize(@NonNull Engine engine, Sampler type) {
+        return nGetMaxTextureSize(engine.getNativeObject(), type.ordinal());
+    }
+
+    /**
+     * @param engine {@link Engine}
+     * @return The maximum number of layers supported by texture arrays. At least 256.
+     */
+    public static int getMaxArrayTextureLayers(@NonNull Engine engine) {
+        return nGetMaxArrayTextureLayers(engine.getNativeObject());
+    }
+
     /**
      * Use <code>Builder</code> to construct a <code>Texture</code> object instance.
      */
@@ -1289,6 +1307,8 @@ public class Texture {
     private static native boolean nIsTextureFormatSupported(long nativeEngine, int internalFormat);
     private static native boolean nIsTextureFormatMipmappable(long nativeEngine, int internalFormat);
     private static native boolean nIsTextureSwizzleSupported(long nativeEngine);
+    private static native int nGetMaxTextureSize(long nativeObject, int ordinal);
+    private static native int nGetMaxArrayTextureLayers(long nativeObject);
 
     private static native boolean nValidatePixelFormatAndType(int internalFormat, int pixelDataFormat,
             int pixelDataType);

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

@@ -31,8 +31,6 @@ package com.google.android.filament;
  *   <li>DisplayRangeToneMapper</li>
  * </ul>
  * </ul>
- *
- * You can create custom tone mapping operators by subclassing ToneMapper.
  */
 public class ToneMapper {
     private final long mNativeObject;

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=1.58.1
+VERSION_NAME=1.59.3
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 

build.sh

@@ -200,8 +200,9 @@ WEBGPU_ANDROID_GRADLE_OPTION=""
 EGL_ON_LINUX_OPTION="-DFILAMENT_SUPPORTS_EGL_ON_LINUX=OFF"
 
 MATDBG_OPTION="-DFILAMENT_ENABLE_MATDBG=OFF"
-FGVIEWER_OPTION="-DFILAMENT_ENABLE_FGVIEWER=OFF"
 MATDBG_GRADLE_OPTION=""
+FGVIEWER_OPTION="-DFILAMENT_ENABLE_FGVIEWER=OFF"
+FGVIEWER_GRADLE_OPTION=""
 
 MATOPT_OPTION=""
 MATOPT_GRADLE_OPTION=""
@@ -533,6 +534,7 @@ function build_android {
             ${VULKAN_ANDROID_GRADLE_OPTION} \
             ${WEBGPU_ANDROID_GRADLE_OPTION} \
             ${MATDBG_GRADLE_OPTION} \
+            ${FGVIEWER_GRADLE_OPTION} \
             ${MATOPT_GRADLE_OPTION} \
             :filament-android:assembleDebug \
             :gltfio-android:assembleDebug \
@@ -541,6 +543,7 @@ function build_android {
         ./gradlew \
             -Pcom.google.android.filament.dist-dir=../out/android-debug/filament \
             -Pcom.google.android.filament.abis=${ABI_GRADLE_OPTION} \
+            ${WEBGPU_ANDROID_GRADLE_OPTION} \
             :filamat-android:assembleDebug
 
         if [[ "${BUILD_ANDROID_SAMPLES}" == "true" ]]; then
@@ -583,6 +586,7 @@ function build_android {
             ${VULKAN_ANDROID_GRADLE_OPTION} \
             ${WEBGPU_ANDROID_GRADLE_OPTION} \
             ${MATDBG_GRADLE_OPTION} \
+            ${FGVIEWER_GRADLE_OPTION} \
             ${MATOPT_GRADLE_OPTION} \
             :filament-android:assembleRelease \
             :gltfio-android:assembleRelease \
@@ -591,6 +595,7 @@ function build_android {
         ./gradlew \
             -Pcom.google.android.filament.dist-dir=../out/android-release/filament \
             -Pcom.google.android.filament.abis=${ABI_GRADLE_OPTION} \
+            ${WEBGPU_ANDROID_GRADLE_OPTION} \
             :filamat-android:assembleRelease
 
         if [[ "${BUILD_ANDROID_SAMPLES}" == "true" ]]; then

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

@@ -1,35 +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
-
-    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

docs/Materials.md.html

@@ -13,7 +13,7 @@ This document is part of the [Filament project](https://github.com/google/filame
 ## Authors
 
 - [Romain Guy](https://github.com/romainguy), [@romainguy](https://twitter.com/romainguy)
-- [Mathias Agopian](https://github.com/pixelflinger), [@darthmoosious](https://twitter.com/darthmoosious)
+- [Mathias Agopian](https://github.com/pixelflinger), [@pixelflinger](https://bsky.app/profile/pixelflinger.bsky.social)
 
 # Overview
 
@@ -78,26 +78,27 @@ in table [standardProperties].
         Property        |      Definition
 -----------------------:|:---------------------
 **baseColor**           | Diffuse albedo for non-metallic surfaces, and specular color for metallic surfaces
-**metallic**            | Whether a surface appears to be dielectric (0.0) or conductor (1.0). Often used as a binary value (0 or 1)
 **roughness**           | Perceived smoothness (1.0) or roughness (0.0) of a surface. Smooth surfaces exhibit sharp reflections
+**metallic**            | Whether a surface appears to be dielectric (0.0) or conductor (1.0). Often used as a binary value (0 or 1)
 **reflectance**         | Fresnel reflectance at normal incidence for dielectric surfaces. This directly controls the strength of the reflections
-**sheenColor**          | Strength of the sheen layer
-**sheenRoughness**      | Perceived smoothness or roughness of the sheen layer
+**ambientOcclusion**    | Defines how much of the ambient light is accessible to a surface point. It is a per-pixel shadowing factor between 0.0 and 1.0
 **clearCoat**           | Strength of the clear coat layer
 **clearCoatRoughness**  | Perceived smoothness or roughness of the clear coat layer
+**clearCoatNormal**     | A detail normal used to perturb the clear coat layer using _bump mapping_ (_normal mapping_)
 **anisotropy**          | Amount of anisotropy in either the tangent or bitangent direction
 **anisotropyDirection** | Local surface direction in tangent space
-**ambientOcclusion**    | Defines how much of the ambient light is accessible to a surface point. It is a per-pixel shadowing factor between 0.0 and 1.0
-**normal**              | A detail normal used to perturb the surface using _bump mapping_ (_normal mapping_)
-**bentNormal**          | A normal pointing in the average unoccluded direction. Can be used to improve indirect lighting quality
-**clearCoatNormal**     | A detail normal used to perturb the clear coat layer using _bump mapping_ (_normal mapping_)
-**emissive**            | Additional diffuse albedo to simulate emissive surfaces (such as neons, etc.) This property is mostly useful in an HDR pipeline with a bloom pass
-**postLightingColor**   | Additional color that can be blended with the result of the lighting computations. See `postLightingBlending`
-**ior**                 | Index of refraction, either for refractive objects or as an alternative to reflectance
-**transmission**        | Defines how much of the diffuse light of a dielectric is transmitted through the object, in other words this defines how transparent an object is
-**absorption**          | Absorption factor for refractive objects
-**microThickness**      | Thickness of the thin layer of refractive objects
 **thickness**           | Thickness of the solid volume of refractive objects
+**sheenColor**          | Strength of the sheen layer
+**sheenRoughness**      | Perceived smoothness or roughness of the sheen layer
+**emissive**            | Additional diffuse albedo to simulate emissive surfaces (such as neons, etc.) This property is mostly useful in an HDR pipeline with a bloom pass
+**normal**              | A detail normal used to perturb the surface using _bump mapping_ (_normal mapping_)
+**postLightingColor**   | Additional color that can be blended with the result of the lighting computations. See `postLightingBlending`
+**absorption**          | Absorption factor for refractive objects
+**transmission**        | Defines how much of the diffuse light of a dielectric is transmitted through the object, in other words this defines how transparent an object is
+**ior**                 | Index of refraction, either for refractive objects or as an alternative to reflectance
+**microThickness**      | Thickness of the thin layer of refractive objects
+**bentNormal**          | A normal pointing in the average unoccluded direction. Can be used to improve indirect lighting quality
+**shadowStrength**      | Strength factor between 0 and 1 for all shadows received by this material
 [Table [standardProperties]: Properties of the standard model]
 
 The type and range of each property is described in table [standardPropertiesTypes].
@@ -1272,6 +1273,9 @@ Description
       when selecting any shading model that is not `unlit`. See the shader sections of this document
       for more information on how to access these attributes from the shaders.
 
+!!! Note: Interaction with custom variables
+    When the `color` attribute is specified, only four custom variables are available instead of five.
+
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ JSON
 material {
     parameters : [
@@ -1302,7 +1306,7 @@ Type
 :    array of `string`
 
 Value
-:     Up to 4 strings, each must be a valid GLSL identifier.
+:     Up to 5 strings, each must be a valid GLSL identifier.
 
 Description
 :     Defines custom interpolants (or variables) that are output by the material's vertex shader.
@@ -1318,6 +1322,10 @@ Description
       particular if `default` is specified the default precision is used is the fragment shader
       (`mediump`) and in the vertex shader (`highp`).
 
+!!! Warning: Interaction with required attributes
+    If the `color` attribute is specified in the `required` list, then only four variables can be used
+    instead of five.
+
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ JSON
 material {
     name : Skybox,

filament/backend/CMakeLists.txt

@@ -108,8 +108,12 @@ if (FILAMENT_SUPPORTS_OPENGL AND NOT FILAMENT_USE_EXTERNAL_GLES3)
         list(APPEND SRCS src/opengl/platforms/PlatformCocoaTouchGL.mm)
         list(APPEND SRCS src/opengl/platforms/CocoaTouchExternalImage.mm)
     elseif (APPLE)
-        list(APPEND SRCS src/opengl/platforms/PlatformCocoaGL.mm)
-        list(APPEND SRCS src/opengl/platforms/CocoaExternalImage.mm)
+        if (FILAMENT_SUPPORTS_OSMESA)
+            list(APPEND SRCS src/opengl/platforms/PlatformOSMesa.cpp)
+        else()
+            list(APPEND SRCS src/opengl/platforms/PlatformCocoaGL.mm)
+            list(APPEND SRCS src/opengl/platforms/CocoaExternalImage.mm)
+        endif()
     elseif (WEBGL)
         list(APPEND SRCS src/opengl/platforms/PlatformWebGL.cpp)
     elseif (LINUX)
@@ -173,27 +177,6 @@ endif()
 if (FILAMENT_SUPPORTS_VULKAN)
     list(APPEND SRCS
             include/backend/platforms/VulkanPlatform.h
-            src/vulkan/caching/VulkanDescriptorSetManager.cpp
-            src/vulkan/caching/VulkanDescriptorSetManager.h
-            src/vulkan/caching/VulkanPipelineLayoutCache.cpp
-            src/vulkan/caching/VulkanPipelineLayoutCache.h
-            src/vulkan/memory/ResourceManager.cpp
-            src/vulkan/memory/ResourceManager.h
-            src/vulkan/memory/ResourcePointer.h
-            src/vulkan/memory/Resource.cpp
-            src/vulkan/memory/Resource.h
-            src/vulkan/platform/VulkanPlatform.cpp
-            src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp
-            src/vulkan/platform/VulkanPlatformSwapChainImpl.h
-            src/vulkan/utils/Conversion.cpp
-            src/vulkan/utils/Conversion.h
-            src/vulkan/utils/Definitions.h
-            src/vulkan/utils/Helper.h
-            src/vulkan/utils/Image.h
-            src/vulkan/utils/Image.cpp
-            src/vulkan/utils/Spirv.h
-            src/vulkan/utils/Spirv.cpp
-            src/vulkan/utils/StaticVector.h
             src/vulkan/VulkanAsyncHandles.h
             src/vulkan/VulkanBlitter.cpp
             src/vulkan/VulkanBlitter.h
@@ -204,46 +187,103 @@ if (FILAMENT_SUPPORTS_VULKAN)
             src/vulkan/VulkanConstants.h
             src/vulkan/VulkanContext.cpp
             src/vulkan/VulkanContext.h
+            src/vulkan/VulkanDescriptorSetCache.cpp
+            src/vulkan/VulkanDescriptorSetCache.h
+            src/vulkan/VulkanDescriptorSetLayoutCache.cpp
+            src/vulkan/VulkanDescriptorSetLayoutCache.h
             src/vulkan/VulkanDriver.cpp
             src/vulkan/VulkanDriver.h
             src/vulkan/VulkanDriverFactory.h
+            src/vulkan/VulkanExternalImageManager.cpp
+            src/vulkan/VulkanExternalImageManager.h
             src/vulkan/VulkanFboCache.cpp
             src/vulkan/VulkanFboCache.h
             src/vulkan/VulkanHandles.cpp
             src/vulkan/VulkanHandles.h
-            src/vulkan/VulkanMemory.h
             src/vulkan/VulkanMemory.cpp
+            src/vulkan/VulkanMemory.h
             src/vulkan/VulkanPipelineCache.cpp
             src/vulkan/VulkanPipelineCache.h
+            src/vulkan/VulkanPipelineLayoutCache.cpp
+            src/vulkan/VulkanPipelineLayoutCache.h
             src/vulkan/VulkanQueryManager.cpp
             src/vulkan/VulkanQueryManager.h
+            src/vulkan/VulkanReadPixels.cpp
+            src/vulkan/VulkanReadPixels.h
             src/vulkan/VulkanSamplerCache.cpp
             src/vulkan/VulkanSamplerCache.h
             src/vulkan/VulkanStagePool.cpp
             src/vulkan/VulkanStagePool.h
             src/vulkan/VulkanSwapChain.cpp
             src/vulkan/VulkanSwapChain.h
-            src/vulkan/VulkanReadPixels.cpp
-            src/vulkan/VulkanReadPixels.h
             src/vulkan/VulkanTexture.cpp
             src/vulkan/VulkanTexture.h
+            src/vulkan/VulkanYcbcrConversionCache.cpp
+            src/vulkan/VulkanYcbcrConversionCache.h
+            src/vulkan/memory/Resource.cpp
+            src/vulkan/memory/Resource.h
+            src/vulkan/memory/ResourceManager.cpp
+            src/vulkan/memory/ResourceManager.h
+            src/vulkan/memory/ResourcePointer.h
+            src/vulkan/platform/VulkanPlatform.cpp
+            src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp
+            src/vulkan/platform/VulkanPlatformSwapChainImpl.h
+            src/vulkan/utils/Conversion.cpp
+            src/vulkan/utils/Conversion.h
+            src/vulkan/utils/Definitions.h
+            src/vulkan/utils/Helper.h
+            src/vulkan/utils/Image.cpp
+            src/vulkan/utils/Image.h
+            src/vulkan/utils/Spirv.cpp
+            src/vulkan/utils/Spirv.h
+            src/vulkan/utils/StaticVector.h
     )
     if (LINUX OR WIN32)
         list(APPEND SRCS src/vulkan/platform/VulkanPlatformLinuxWindows.cpp)
     elseif (APPLE OR IOS)
         list(APPEND SRCS src/vulkan/platform/VulkanPlatformApple.mm)
     elseif (ANDROID)
-        list(APPEND SRCS src/vulkan/platform/VulkanPlatformAndroid.cpp)
+        list(APPEND SRCS
+            include/backend/platforms/VulkanPlatformAndroid.h
+            src/vulkan/platform/VulkanPlatformAndroid.cpp
+        )
     endif()
 endif()
 
 if (FILAMENT_SUPPORTS_WEBGPU)
     list(APPEND SRCS
+            include/backend/platforms/WebGPUPlatform.h
+            src/webgpu/platform/WebGPUPlatform.cpp
+            src/webgpu/WebGPUConstants.h
             src/webgpu/WebGPUDriver.cpp
             src/webgpu/WebGPUDriver.h
-            src/webgpu/WebGPUPlatform.cpp
-            src/webgpu/WebGPUPlatform.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)
+    elseif (LINUX)
+        list(APPEND SRCS src/webgpu/platform/WebGPUPlatformLinux.cpp)
+    elseif (APPLE OR IOS)
+        list(APPEND SRCS src/webgpu/platform/WebGPUPlatformApple.mm)
+    elseif (ANDROID)
+        list(APPEND SRCS src/webgpu/platform/WebGPUPlatformAndroid.cpp)
+    endif()
+
+    if (TNT_DEV)
+        set(FILAMENT_WEBGPU_IMMEDIATE_ERROR_HANDLING_DEFAULT ON)
+    else()
+        set(FILAMENT_WEBGPU_IMMEDIATE_ERROR_HANDLING_DEFAULT OFF)
+    endif()
+
+    option(FILAMENT_WEBGPU_IMMEDIATE_ERROR_HANDLING
+            "Enable immediate_error_handling for the WebGPU backend Dawn implementation"
+            ${FILAMENT_WEBGPU_IMMEDIATE_ERROR_HANDLING_DEFAULT})
 endif()
 
 if (ANDROID)
@@ -273,6 +313,11 @@ target_include_directories(${TARGET} PUBLIC ${PUBLIC_HDR_DIR})
 # add this subproject to the Filament folder
 set_target_properties(${TARGET} PROPERTIES FOLDER Filament)
 
+# we need to export the headers properly for backend.lib on windows
+if (WIN32 AND FILAMENT_SUPPORTS_WEBGPU)
+    target_compile_definitions(${TARGET} PRIVATE "WGPU_IMPLEMENTATION")
+endif()
+
 # ==================================================================================================
 # Expose a header-only target to minimize dependencies.
 # ==================================================================================================
@@ -393,9 +438,12 @@ set(LINUX_LINKER_OPTIMIZATION_FLAGS
         -Wl,--exclude-libs,bluegl
 )
 
-if (LINUX AND FILAMENT_SUPPORTS_OSMESA)
-    set(OSMESA_COMPILE_FLAGS
-        -I${FILAMENT_OSMESA_PATH}/include/GL)
+if (FILAMENT_SUPPORTS_OSMESA)
+    if (LINUX)
+        set(OSMESA_COMPILE_FLAGS -I${FILAMENT_OSMESA_PATH}/include/GL)
+    elseif (APPLE)
+        set(OSMESA_COMPILE_FLAGS -I${FILAMENT_OSMESA_PATH}/include)
+    endif()
 endif()
 
 if (MSVC)
@@ -430,6 +478,10 @@ if (FILAMENT_SUPPORTS_METAL)
     target_compile_definitions(${TARGET} PRIVATE $<$<BOOL:${FILAMENT_METAL_PROFILING}>:FILAMENT_METAL_PROFILING>)
 endif()
 
+if (FILAMENT_SUPPORTS_WEBGPU)
+    target_compile_definitions(${TARGET} PRIVATE $<$<BOOL:${FILAMENT_WEBGPU_IMMEDIATE_ERROR_HANDLING}>:FILAMENT_WEBGPU_IMMEDIATE_ERROR_HANDLING>)
+endif()
+
 target_link_libraries(${TARGET} PRIVATE
         ${OSMESA_LINKER_FLAGS}
         $<$<AND:$<PLATFORM_ID:Linux>,$<CONFIG:Release>>:${LINUX_LINKER_OPTIMIZATION_FLAGS}>
@@ -458,6 +510,12 @@ if (APPLE OR LINUX)
         test/ShaderGenerator.cpp
         test/TrianglePrimitive.cpp
         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
@@ -473,6 +531,7 @@ if (APPLE OR LINUX)
         test/test_Handles.cpp
     )
     set(BACKEND_TEST_LIBS
+        absl::str_format
         backend
         getopt
         gtest
@@ -480,6 +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
@@ -488,7 +550,7 @@ if (APPLE AND NOT IOS)
     list(APPEND BACKEND_TEST_SRC
          test/test_RenderExternalImage.cpp)
     add_library(backend_test STATIC ${BACKEND_TEST_SRC})
-    target_link_libraries(backend_test PRIVATE ${BACKEND_TEST_LIBS})
+    target_link_libraries(backend_test PUBLIC ${BACKEND_TEST_LIBS})
 
     set(BACKEND_TEST_DEPS
             OSDependent

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/DriverEnums.h

@@ -126,6 +126,10 @@ static_assert(MAX_VERTEX_BUFFER_COUNT <= MAX_VERTEX_ATTRIBUTE_COUNT,
 static constexpr size_t CONFIG_UNIFORM_BINDING_COUNT = 9;   // This is guaranteed by OpenGL ES.
 static constexpr size_t CONFIG_SAMPLER_BINDING_COUNT = 4;   // This is guaranteed by OpenGL ES.
 
+static constexpr uint8_t EXTERNAL_SAMPLER_DATA_INDEX_UNUSED =
+        uint8_t(-1);// Case where the descriptor set binding isnt using any external sampler state
+                     // and therefore doesn't have a valid entry.
+
 /**
  * Defines the backend's feature levels.
  */
@@ -182,6 +186,7 @@ enum class ShaderLanguage {
     SPIRV = 2,
     MSL = 3,
     METAL_LIBRARY = 4,
+    WGSL = 5,
 };
 
 static constexpr const char* shaderLanguageToString(ShaderLanguage shaderLanguage) {
@@ -196,6 +201,8 @@ static constexpr const char* shaderLanguageToString(ShaderLanguage shaderLanguag
             return "MSL";
         case ShaderLanguage::METAL_LIBRARY:
             return "Metal precompiled library";
+        case ShaderLanguage::WGSL:
+            return "WGSL";
     }
 }
 
@@ -249,21 +256,21 @@ struct DescriptorSetLayoutBinding {
     DescriptorFlags flags = DescriptorFlags::NONE;
     uint16_t count = 0;
 
-    friend inline bool operator==(
-            DescriptorSetLayoutBinding const& lhs,
+    //  TODO: uncomment when needed.  Note that this class is used as hash key.  We need to ensure
+    //  no uninitialized padding bytes.
+    //    uint8_t externalSamplerDataIndex = EXTERNAL_SAMPLER_DATA_INDEX_UNUSED;
+
+    friend inline bool operator==(DescriptorSetLayoutBinding const& lhs,
             DescriptorSetLayoutBinding const& rhs) noexcept {
         return lhs.type == rhs.type &&
                lhs.flags == rhs.flags &&
                lhs.count == rhs.count &&
                lhs.stageFlags == rhs.stageFlags;
+//               lhs.stageFlags == rhs.stageFlags &&
+//               lhs.externalSamplerDataIndex == rhs.externalSamplerDataIndex;
     }
 };
 
-struct DescriptorSetLayout {
-    utils::FixedCapacityVector<DescriptorSetLayoutBinding> bindings;
-};
-
-
 /**
  * Bitmask for selecting render buffers
  */
@@ -965,8 +972,28 @@ enum class SamplerCompareFunc : uint8_t {
     N           //!< Never. The depth / stencil test always fails.
 };
 
+//! this API is copied from (and only applies to) the Vulkan spec.
+//! These specify YUV to RGB conversions.
+enum class SamplerYcbcrModelConversion : uint8_t {
+    RGB_IDENTITY = 0,
+    YCBCR_IDENTITY = 1,
+    YCBCR_709 = 2,
+    YCBCR_601 = 3,
+    YCBCR_2020 = 4,
+};
+
+enum class SamplerYcbcrRange : uint8_t {
+    ITU_FULL = 0,
+    ITU_NARROW = 1,
+};
+
+enum class ChromaLocation : uint8_t {
+    COSITED_EVEN = 0,
+    MIDPOINT = 1,
+};
+
 //! Sampler parameters
-struct SamplerParams { // NOLINT
+struct SamplerParams {             // NOLINT
     SamplerMagFilter filterMag      : 1;    //!< magnification filter (NEAREST)
     SamplerMinFilter filterMin      : 3;    //!< minification filter  (NEAREST)
     SamplerWrapMode wrapS           : 2;    //!< s-coordinate wrap mode (CLAMP_TO_EDGE)
@@ -1021,6 +1048,7 @@ private:
         return SamplerParams::LessThan{}(lhs, rhs);
     }
 };
+
 static_assert(sizeof(SamplerParams) == 4);
 
 // The limitation to 64-bits max comes from how we store a SamplerParams in our JNI code
@@ -1028,6 +1056,95 @@ static_assert(sizeof(SamplerParams) == 4);
 static_assert(sizeof(SamplerParams) <= sizeof(uint64_t),
         "SamplerParams must be no more than 64 bits");
 
+//! Sampler parameters
+struct SamplerYcbcrConversion {// NOLINT
+    SamplerYcbcrModelConversion ycbcrModel : 4;
+    TextureSwizzle r : 4;
+    TextureSwizzle g : 4;
+    TextureSwizzle b : 4;
+    TextureSwizzle a : 4;
+    SamplerYcbcrRange ycbcrRange : 1;
+    ChromaLocation xChromaOffset : 1;
+    ChromaLocation yChromaOffset : 1;
+    SamplerMagFilter chromaFilter : 1;
+    uint8_t padding;
+
+    struct Hasher {
+        size_t operator()(const SamplerYcbcrConversion p) const noexcept {
+            // we don't use std::hash<> here, so we don't have to include <functional>
+            return *reinterpret_cast<uint32_t const*>(reinterpret_cast<char const*>(&p));
+        }
+    };
+
+    struct EqualTo {
+        bool operator()(SamplerYcbcrConversion lhs, SamplerYcbcrConversion rhs) const noexcept {
+            assert_invariant(lhs.padding == 0);
+            auto* pLhs = reinterpret_cast<uint32_t const*>(reinterpret_cast<char const*>(&lhs));
+            auto* pRhs = reinterpret_cast<uint32_t const*>(reinterpret_cast<char const*>(&rhs));
+            return *pLhs == *pRhs;
+        }
+    };
+
+    struct LessThan {
+        bool operator()(SamplerYcbcrConversion lhs, SamplerYcbcrConversion rhs) const noexcept {
+            assert_invariant(lhs.padding == 0);
+            auto* pLhs = reinterpret_cast<uint32_t const*>(reinterpret_cast<char const*>(&lhs));
+            auto* pRhs = reinterpret_cast<uint32_t const*>(reinterpret_cast<char const*>(&rhs));
+            return *pLhs < *pRhs;
+        }
+    };
+
+private:
+    friend inline bool operator == (SamplerYcbcrConversion lhs, SamplerYcbcrConversion rhs)
+            noexcept {
+        return SamplerYcbcrConversion::EqualTo{}(lhs, rhs);
+    }
+    friend inline bool operator != (SamplerYcbcrConversion lhs, SamplerYcbcrConversion rhs)
+            noexcept {
+        return  !SamplerYcbcrConversion::EqualTo{}(lhs, rhs);
+    }
+    friend inline bool operator < (SamplerYcbcrConversion lhs, SamplerYcbcrConversion rhs)
+            noexcept {
+        return SamplerYcbcrConversion::LessThan{}(lhs, rhs);
+    }
+};
+
+static_assert(sizeof(SamplerYcbcrConversion) == 4);
+
+static_assert(sizeof(SamplerYcbcrConversion) <= sizeof(uint64_t),
+        "SamplerYcbcrConversion must be no more than 64 bits");
+
+struct ExternalSamplerDatum {
+    ExternalSamplerDatum(SamplerYcbcrConversion ycbcr, SamplerParams spm, uint32_t extFmt)
+        : YcbcrConversion(ycbcr),
+          samplerParams(spm),
+          externalFormat(extFmt) {}
+    bool operator==(ExternalSamplerDatum const& rhs) const {
+        return (YcbcrConversion == rhs.YcbcrConversion && samplerParams == rhs.samplerParams &&
+                externalFormat == rhs.externalFormat);
+    }
+    struct EqualTo {
+        bool operator()(const ExternalSamplerDatum& lhs,
+                const ExternalSamplerDatum& rhs) const noexcept {
+            return (lhs.YcbcrConversion == rhs.YcbcrConversion &&
+                lhs.samplerParams == rhs.samplerParams &&
+                lhs.externalFormat == rhs.externalFormat);
+        }
+    };
+    SamplerYcbcrConversion YcbcrConversion;
+    SamplerParams samplerParams;
+    uint32_t externalFormat;
+};
+// No implicit padding allowed due to it being a hash key.
+static_assert(sizeof(ExternalSamplerDatum) == 12);
+
+struct DescriptorSetLayout {
+    utils::FixedCapacityVector<DescriptorSetLayoutBinding> bindings;
+
+//  TODO: uncomment when needed
+//    utils::FixedCapacityVector<ExternalSamplerDatum> externalSamplerData;
+};
+
 //! blending equation function
 enum class BlendEquation : uint8_t {
     ADD,                    //!< the fragment is added to the color buffer

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

@@ -33,7 +33,7 @@ public:
     PlatformCocoaGL();
     ~PlatformCocoaGL() noexcept override;
 
-    ExternalImageHandle createExternalImage(void* cvPixelBuffer) noexcept;
+    ExternalImageHandle UTILS_PUBLIC createExternalImage(void* cvPixelBuffer) noexcept;
 
 protected:
     // --------------------------------------------------------------------------------------------

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

@@ -32,7 +32,7 @@ public:
     PlatformCocoaTouchGL();
     ~PlatformCocoaTouchGL() noexcept override;
 
-    ExternalImageHandle createExternalImage(void* cvPixelBuffer) noexcept;
+    ExternalImageHandle UTILS_PUBLIC createExternalImage(void* cvPixelBuffer) noexcept;
 
     // --------------------------------------------------------------------------------------------
     // Platform Interface

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

@@ -50,7 +50,7 @@ public:
     /**
      * Creates an ExternalImage from a EGLImageKHR
      */
-    ExternalImageHandle createExternalImage(EGLImageKHR eglImage) noexcept;
+    ExternalImageHandle UTILS_PUBLIC createExternalImage(EGLImageKHR eglImage) noexcept;
 
 protected:
     // --------------------------------------------------------------------------------------------

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

@@ -23,6 +23,7 @@
 #include <backend/platforms/PlatformEGL.h>
 
 #include <utils/android/PerformanceHintManager.h>
+#include <utils/compiler.h>
 
 #include <chrono>
 
@@ -43,8 +44,21 @@ public:
     PlatformEGLAndroid() noexcept;
     ~PlatformEGLAndroid() noexcept override;
 
-protected:
+    /**
+     * Creates an ExternalImage from a EGLImageKHR
+     */
+    ExternalImageHandle UTILS_PUBLIC createExternalImage(AHardwareBuffer const* buffer, bool sRGB) noexcept;
 
+    struct UTILS_PUBLIC ExternalImageDescAndroid {
+        uint32_t width;      // Texture width
+        uint32_t height;     // Texture height
+        TextureFormat format;// Texture format
+        TextureUsage usage;  // Texture usage flags
+    };
+
+    ExternalImageDescAndroid UTILS_PUBLIC getExternalImageDesc(ExternalImageHandle externalImage) noexcept;
+
+protected:
     // --------------------------------------------------------------------------------------------
     // Platform Interface
 
@@ -57,11 +71,6 @@ protected:
     Driver* createDriver(void* sharedContext,
             const Platform::DriverConfig& driverConfig) noexcept override;
 
-    /**
-     * Creates an ExternalImage from a EGLImageKHR
-     */
-    ExternalImageHandle createExternalImage(AHardwareBuffer const *buffer, bool sRGB) noexcept;
-
     // --------------------------------------------------------------------------------------------
     // OpenGLPlatform Interface
 
@@ -95,15 +104,26 @@ protected:
      */
     AcquiredImage transformAcquiredImage(AcquiredImage source) noexcept override;
 
-    bool setExternalImage(ExternalImageHandleRef externalImage, ExternalTexture* texture) noexcept override;
+    OpenGLPlatform::ExternalTexture* createExternalImageTexture() noexcept override;
+    void destroyExternalImageTexture(ExternalTexture* texture) noexcept override;
 
     struct ExternalImageEGLAndroid : public ExternalImageEGL {
         AHardwareBuffer* aHardwareBuffer = nullptr;
+        uint32_t width;      // Texture width
+        uint32_t height;     // Texture height
+        TextureFormat format;// Texture format
+        TextureUsage usage;  // Texture usage flags
         bool sRGB = false;
+
     protected:
         ~ExternalImageEGLAndroid() override;
     };
 
+    bool setExternalImage(ExternalImageHandleRef externalImage,
+            ExternalTexture* texture) noexcept override;
+    bool setImage(ExternalImageEGLAndroid const* eglExternalImage,
+            ExternalTexture* texture) noexcept;
+
 protected:
     bool makeCurrent(ContextType type,
             SwapChain* drawSwapChain,

filament/backend/include/backend/platforms/PlatformOSMesa.h

@@ -21,7 +21,12 @@
 
 #include "bluegl/BlueGL.h"
 
+#if defined(__linux__)
 #include <osmesa.h>
+#elif defined(__APPLE__)
+#undef GLAPI
+#include <GL/osmesa.h>
+#endif
 
 #include <backend/platforms/OpenGLPlatform.h>
 #include <backend/DriverEnums.h>

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

@@ -18,6 +18,7 @@
 #define TNT_FILAMENT_BACKEND_PLATFORMS_VULKANPLATFORM_H
 
 #include <backend/Platform.h>
+#include <backend/DriverEnums.h>
 
 #include <bluevk/BlueVK.h>
 
@@ -103,10 +104,6 @@ public:
 
         // Semaphore to be signaled once the image is available.
         VkSemaphore imageReadySemaphore = VK_NULL_HANDLE;
-
-        // A function called right before vkQueueSubmit. After this call, the image must be 
-        // available. This pointer can be null if imageReadySemaphore is not VK_NULL_HANDLE.
-        std::function<void(SwapChainPtr handle)> explicitImageReadyWait = nullptr;
     };
 
     VulkanPlatform();
@@ -297,6 +294,16 @@ public:
     VkQueue getProtectedGraphicsQueue() const noexcept;
 
     struct ExternalImageMetadata {
+        /**
+         * The Filament texture format.
+         */
+        TextureFormat filamentFormat;
+
+        /**
+         * The Filament texture usage.
+         */
+        TextureUsage filamentUsage;
+
         /**
          * The width of the external image
          */
@@ -307,26 +314,21 @@ public:
          */
         uint32_t height;
 
-        /**
-         * The layerCount of the external image
-         */
-        uint32_t layerCount;
-
         /**
          * The layer count of the external image
          */
         uint32_t layers;
 
+        /**
+         * The numbers of samples per texel
+         */
+        VkSampleCountFlagBits samples;
+
         /**
          * The format of the external image
          */
         VkFormat format;
 
-        /**
-         * An external buffer can be protected. This tells you if it is.
-         */
-        bool isProtected;
-
         /**
          * The type of external format (opaque int) if used.
          */
@@ -346,23 +348,75 @@ public:
          * Heap information
          */
         uint32_t memoryTypeBits;
-    };
-    virtual ExternalImageMetadata getExternalImageMetadata(void* externalImage);
 
-    using ImageData = std::pair<VkImage, VkDeviceMemory>;
-    virtual ImageData createExternalImage(void* externalImage,
-            const ExternalImageMetadata& metadata);
+        /**
+         * Ycbcr conversion components
+         */
+        VkComponentMapping ycbcrConversionComponents;
+
+        /**
+         * Ycbcr model
+         */
+        VkSamplerYcbcrModelConversion ycbcrModel;
+
+        /**
+         * Ycbcr range
+         */
+        VkSamplerYcbcrRange ycbcrRange;
+
+        /**
+         * Ycbcr x chroma offset
+         */
+        VkChromaLocation xChromaOffset;
+
+        /**
+         * Ycbcr y chroma offset
+         */
+        VkChromaLocation yChromaOffset;
+    };
+
+
+    // Note that the image metadata might change per-frame, hence we need a method for extracting
+    // it.
+    virtual ExternalImageMetadata extractExternalImageMetadata(ExternalImageHandleRef image) const {
+        return {};
+    }
+
+    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 {};
+    }
+
+protected:
+    virtual ExtensionSet getSwapchainInstanceExtensions() const;
+
+    using SurfaceBundle = std::tuple<VkSurfaceKHR, VkExtent2D>;
+    virtual SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
+            uint64_t flags) const noexcept;
 
 private:
-    static ExtensionSet getSwapchainInstanceExtensions();
-    static ExternalImageMetadata getExternalImageMetadataImpl(void* externalImage,
-            VkDevice device);
-    static ImageData createExternalImageImpl(void* externalImage, VkDevice device,
-            const VkAllocationCallbacks* allocator, const ExternalImageMetadata& metadata);
+    // Platform dependent helper methods
+    static ExtensionSet getSwapchainInstanceExtensionsImpl();
 
     // Platform dependent helper methods
-    using SurfaceBundle = std::tuple<VkSurfaceKHR, VkExtent2D>;
-    static SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
+    static SurfaceBundle createVkSurfaceKHRImpl(void* nativeWindow, VkInstance instance,
             uint64_t flags) noexcept;
 
     friend struct VulkanPlatformPrivate;

filament/backend/include/backend/platforms/VulkanPlatformAndroid.h

@@ -0,0 +1,66 @@
+/*
+ * 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_PLATFORMS_VULKAN_PLATFORM_ANDROID_H
+#define TNT_FILAMENT_BACKEND_PLATFORMS_VULKAN_PLATFORM_ANDROID_H
+
+#include <backend/DriverEnums.h>
+#include <backend/platforms/VulkanPlatform.h>
+
+#include <android/hardware_buffer.h>
+
+namespace filament::backend {
+
+class VulkanPlatformAndroid : public VulkanPlatform {
+public:
+    ExternalImageHandle UTILS_PUBLIC createExternalImage(AHardwareBuffer const* buffer,
+            bool sRGB) noexcept;
+
+    struct UTILS_PUBLIC ExternalImageDescAndroid {
+        uint32_t width;      // Texture width
+        uint32_t height;     // Texture height
+        TextureFormat format;// Texture format
+        TextureUsage usage;  // Texture usage flags
+    };
+
+    ExternalImageDescAndroid UTILS_PUBLIC getExternalImageDesc(
+            ExternalImageHandleRef externalImage) const noexcept;
+
+    virtual ExternalImageMetadata extractExternalImageMetadata(
+            ExternalImageHandleRef image) const override;
+
+    virtual ImageData createVkImageFromExternal(ExternalImageHandleRef image) const override;
+
+protected:
+    virtual ExtensionSet getSwapchainInstanceExtensions() const override;
+
+    using SurfaceBundle = VulkanPlatform::SurfaceBundle;
+    virtual SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
+            uint64_t flags) const noexcept override;
+
+private:
+    struct ExternalImageVulkanAndroid : public Platform::ExternalImage {
+        AHardwareBuffer* aHardwareBuffer = nullptr;
+        bool sRGB = false;
+
+    protected:
+        ~ExternalImageVulkanAndroid() override;
+    };
+};
+
+}// namespace filament::backend
+
+#endif// TNT_FILAMENT_BACKEND_PLATFORMS_VULKAN_PLATFORM_ANDROID_H

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

@@ -0,0 +1,65 @@
+/*
+ * 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_PLATFORMS_WEBGPUPLATFORM_H
+#define TNT_FILAMENT_BACKEND_PLATFORMS_WEBGPUPLATFORM_H
+
+#include <backend/Platform.h>
+
+#include <webgpu/webgpu_cpp.h>
+
+#include <cstdint>
+
+namespace filament::backend {
+
+/**
+ * A Platform interface, handling the environment-specific concerns, e.g. OS, for creating a WebGPU
+ * driver (backend).
+ */
+class WebGPUPlatform final : public Platform {
+public:
+    WebGPUPlatform();
+    ~WebGPUPlatform() override = default;
+
+    [[nodiscard]] int getOSVersion() const noexcept final { return 0; }
+
+    [[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
+    [[nodiscard]] wgpu::Adapter requestAdapter(wgpu::Surface const& surface);
+    // either returns a valid device or panics
+    [[nodiscard]] wgpu::Device requestDevice(wgpu::Adapter const& adapter);
+
+protected:
+    [[nodiscard]] Driver* createDriver(void* sharedContext,
+            const Platform::DriverConfig& driverConfig) noexcept override;
+
+private:
+    // we may consider having the driver own this in the future
+    wgpu::Instance mInstance;
+};
+
+}// namespace filament::backend
+
+#endif// TNT_FILAMENT_BACKEND_PLATFORMS_WEBGPUPLATFORM_H

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

@@ -362,6 +362,8 @@ DECL_DRIVER_API_SYNCHRONOUS_0(bool, isProtectedTexturesSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isDepthClampSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(uint8_t, getMaxDrawBuffers)
 DECL_DRIVER_API_SYNCHRONOUS_0(size_t, getMaxUniformBufferSize)
+DECL_DRIVER_API_SYNCHRONOUS_N(size_t, getMaxTextureSize, backend::SamplerType, target)
+DECL_DRIVER_API_SYNCHRONOUS_0(size_t, getMaxArrayTextureLayers)
 DECL_DRIVER_API_SYNCHRONOUS_0(math::float2, getClipSpaceParams)
 DECL_DRIVER_API_SYNCHRONOUS_N(void, setupExternalImage2, backend::Platform::ExternalImageHandleRef, image)
 DECL_DRIVER_API_SYNCHRONOUS_N(void, setupExternalImage, void*, image)

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

@@ -40,9 +40,11 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#define HandleAllocatorGL  HandleAllocator<32,  96, 184>    // ~4520 / pool / MiB
-#define HandleAllocatorVK  HandleAllocator<64, 160, 312>    // ~1820 / pool / MiB
-#define HandleAllocatorMTL HandleAllocator<32,  64, 552>    // ~1660 / pool / MiB
+#define HandleAllocatorGL   HandleAllocator<32,  96, 184>    // ~4520 / pool / MiB
+#define HandleAllocatorVK   HandleAllocator<64, 160, 312>    // ~1820 / pool / MiB
+#define HandleAllocatorMTL  HandleAllocator<32,  64, 552>    // ~1660 / pool / MiB
+// TODO WebGPU examine right size of handles
+#define HandleAllocatorWGPU HandleAllocator<64, 160, 552>    // ~1820 / pool / MiB
 
 namespace filament::backend {
 

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/HandleAllocator.cpp

@@ -211,4 +211,8 @@ template class HandleAllocatorVK;
 template class HandleAllocatorMTL;
 #endif
 
+#if defined (FILAMENT_SUPPORTS_WEBGPU)
+template class HandleAllocatorWGPU;
+#endif
+
 } // namespace filament::backend

filament/backend/src/PlatformFactory.cpp

@@ -19,6 +19,11 @@
 #include <utils/Systrace.h>
 #include <utils/debug.h>
 
+// We need to keep this up top for the linux (X11) name collisions.
+#if defined(FILAMENT_SUPPORTS_WEBGPU)
+    #include "backend/platforms/WebGPUPlatform.h"
+#endif
+
 #if defined(__ANDROID__)
     #include <sys/system_properties.h>
     #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3)
@@ -30,7 +35,11 @@
     #endif
 #elif defined(__APPLE__)
     #if defined(FILAMENT_SUPPORTS_OPENGL) && !defined(FILAMENT_USE_EXTERNAL_GLES3)
-        #include <backend/platforms/PlatformCocoaGL.h>
+        #if defined(FILAMENT_SUPPORTS_OSMESA)
+            #include <backend/platforms/PlatformOSMesa.h>
+        #else
+            #include <backend/platforms/PlatformCocoaGL.h>
+        #endif
     #endif
 #elif defined(__linux__)
     #if defined(FILAMENT_SUPPORTS_X11)
@@ -55,7 +64,11 @@
 #endif
 
 #if defined(FILAMENT_DRIVER_SUPPORTS_VULKAN)
-    #include "backend/platforms/VulkanPlatform.h"
+    #if defined(__ANDROID__)
+        #include "backend/platforms/VulkanPlatformAndroid.h"
+    #else
+        #include "backend/platforms/VulkanPlatform.h"
+    #endif
 #endif
 
 #if defined (FILAMENT_SUPPORTS_METAL)
@@ -65,9 +78,6 @@ filament::backend::Platform* createDefaultMetalPlatform();
 #endif
 
 #include "noop/PlatformNoop.h"
-#if defined(FILAMENT_SUPPORTS_WEBGPU)
-    #include "webgpu/WebGPUPlatform.h"
-#endif
 
 namespace filament::backend {
 
@@ -104,7 +114,11 @@ Platform* PlatformFactory::create(Backend* backend) noexcept {
     }
     if (*backend == Backend::VULKAN) {
         #if defined(FILAMENT_DRIVER_SUPPORTS_VULKAN)
-            return new VulkanPlatform();
+            #if defined(__ANDROID__)
+                return new VulkanPlatformAndroid();
+            #else
+                return new VulkanPlatform();
+            #endif
         #else
             return nullptr;
         #endif
@@ -132,7 +146,11 @@ Platform* PlatformFactory::create(Backend* backend) noexcept {
         #elif defined(FILAMENT_IOS)
             return new PlatformCocoaTouchGL();
         #elif defined(__APPLE__)
-            return new PlatformCocoaGL();
+            #if defined(FILAMENT_SUPPORTS_OSMESA)
+                return new PlatformOSMesa();
+            #else
+                return new PlatformCocoaGL();
+            #endif
         #elif defined(__linux__)
             #if defined(FILAMENT_SUPPORTS_X11)
                 return new PlatformGLX();

filament/backend/src/metal/MetalDriver.mm

@@ -1177,6 +1177,24 @@ size_t MetalDriver::getMaxUniformBufferSize() {
     return 256 * 1024 * 1024;   // TODO: return the actual size instead of hardcoding the minspec
 }
 
+size_t MetalDriver::getMaxTextureSize(SamplerType type) {
+    if (type == SamplerType::SAMPLER_3D) {
+        return 2048;
+    }
+    if (mContext->highestSupportedGpuFamily.apple >= 3 ||
+            mContext->highestSupportedGpuFamily.mac >= 2) {
+        return 16384;
+    } else if (mContext->highestSupportedGpuFamily.apple >= 2) {
+        return 8192;
+    }
+    return 4096;
+}
+
+size_t MetalDriver::getMaxArrayTextureLayers() {
+    // TODO: return the actual size instead of hardcoding the minspec
+    return 256;
+}
+
 void MetalDriver::updateIndexBuffer(Handle<HwIndexBuffer> ibh, BufferDescriptor&& data,
         uint32_t byteOffset) {
     FILAMENT_CHECK_PRECONDITION(data.buffer)
@@ -1390,8 +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/MetalShaderCompiler.mm

@@ -138,6 +138,7 @@ bool MetalShaderCompiler::isParallelShaderCompileSupported() const noexcept {
             case ShaderLanguage::ESSL1:
             case ShaderLanguage::ESSL3:
             case ShaderLanguage::SPIRV:
+            case ShaderLanguage::WGSL:
                 break;
         }
 

filament/backend/src/noop/NoopDriver.cpp

@@ -240,6 +240,16 @@ size_t NoopDriver::getMaxUniformBufferSize() {
     return 16384u;
 }
 
+size_t NoopDriver::getMaxTextureSize(SamplerType target) {
+    // NoopDriver is being actively used for other purposes.  This needs to be resolved before we
+    // can change it to 2048. b/406832484
+    return 16384u;
+}
+
+size_t NoopDriver::getMaxArrayTextureLayers() {
+    return 256u;
+}
+
 void NoopDriver::updateIndexBuffer(Handle<HwIndexBuffer> ibh, BufferDescriptor&& p,
         uint32_t byteOffset) {
     scheduleDestroy(std::move(p));

filament/backend/src/noop/PlatformNoop.cpp

@@ -20,7 +20,7 @@
 
 namespace filament::backend {
 
-Driver* PlatformNoop::createDriver(void* const sharedGLContext, const Platform::DriverConfig& driverConfig) noexcept {
+Driver* PlatformNoop::createDriver(void* sharedGLContext, const Platform::DriverConfig& driverConfig) noexcept {
     return NoopDriver::create();
 }
 

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/OpenGLContext.cpp

@@ -94,18 +94,22 @@ OpenGLContext::OpenGLContext(OpenGLPlatform& platform,
     }
     #endif
 
-    OpenGLContext::initExtensions(&ext, state.major, state.minor);
+    initExtensions(&ext, state.major, state.minor);
 
-    OpenGLContext::initProcs(&procs, ext, state.major, state.minor);
+    initProcs(&procs, ext, state.major, state.minor);
 
-    OpenGLContext::initBugs(&bugs, ext, state.major, state.minor,
+    initBugs(&bugs, ext, state.major, state.minor,
             state.vendor, state.renderer, state.version, state.shader);
 
     glGetIntegerv(GL_MAX_RENDERBUFFER_SIZE,             &gets.max_renderbuffer_size);
     glGetIntegerv(GL_MAX_TEXTURE_IMAGE_UNITS,           &gets.max_texture_image_units);
     glGetIntegerv(GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS,  &gets.max_combined_texture_image_units);
+    glGetIntegerv(GL_MAX_TEXTURE_SIZE,                  &gets.max_texture_size);
+    glGetIntegerv(GL_MAX_CUBE_MAP_TEXTURE_SIZE,         &gets.max_cubemap_texture_size);
+    glGetIntegerv(GL_MAX_3D_TEXTURE_SIZE,               &gets.max_3d_texture_size);
+    glGetIntegerv(GL_MAX_ARRAY_TEXTURE_LAYERS,          &gets.max_array_texture_layers);
 
-    mFeatureLevel = OpenGLContext::resolveFeatureLevel(state.major, state.minor, ext, gets, bugs);
+    mFeatureLevel = resolveFeatureLevel(state.major, state.minor, ext, gets, bugs);
 
 #ifdef BACKEND_OPENGL_VERSION_GLES
     mShaderModel = ShaderModel::MOBILE;
@@ -177,6 +181,14 @@ OpenGLContext::OpenGLContext(OpenGLPlatform& platform,
                     << gets.max_anisotropy << '\n'
             << "GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS = "
                     << gets.max_combined_texture_image_units << '\n'
+            << "GL_MAX_TEXTURE_SIZE = "
+                    << gets.max_texture_size << '\n'
+            << "GL_MAX_CUBE_MAP_TEXTURE_SIZE = "
+                    << gets.max_cubemap_texture_size << '\n'
+            << "GL_MAX_3D_TEXTURE_SIZE = "
+                    << gets.max_3d_texture_size << '\n'
+            << "GL_MAX_ARRAY_TEXTURE_LAYERS = "
+                    << gets.max_array_texture_layers << '\n'
             << "GL_MAX_DRAW_BUFFERS = "
                     << gets.max_draw_buffers << '\n'
             << "GL_MAX_RENDERBUFFER_SIZE = "
@@ -554,7 +566,11 @@ void OpenGLContext::initBugs(Bugs* bugs, Extensions const& exts,
         } else if (strstr(renderer, "AMD") ||
                    strstr(renderer, "ATI")) {
             // AMD/ATI GPU
-        } else if (strstr(vendor, "Mesa")) {
+        } else if (strstr(renderer, "Mozilla")) {
+            bugs->disable_invalidate_framebuffer = true;
+        }
+
+        if (strstr(vendor, "Mesa")) {
             // Seen on
             //  [Mesa],
             //  [llvmpipe (LLVM 17.0.6, 256 bits)],
@@ -562,8 +578,6 @@ void OpenGLContext::initBugs(Bugs* bugs, Extensions const& exts,
             //  [4.50]
             // not known which version are affected
             bugs->rebind_buffer_after_deletion = true;
-        } else if (strstr(renderer, "Mozilla")) {
-            bugs->disable_invalidate_framebuffer = true;
         }
     } else {
         // When running under ANGLE, it's a different set of workaround that we need.
@@ -575,6 +589,19 @@ void OpenGLContext::initBugs(Bugs* bugs, Extensions const& exts,
         }
     }
 
+    if (strstr(vendor, "Mozilla")) {
+        // Seen on
+        //  [Mozilla],
+        //  [GeForce GTX 980, or similar]
+        //    or [ANGLE (NVIDIA, NVIDIA GeForce GTX 980 Direct3D11 vs_5_0 ps_5_0), or similar]
+        //    or anything else,
+        //  [OpenGL ES 3.0 (WebGL 2.0)],
+        //  [OpenGL ES GLSL ES 3.00 (WebGL GLSL ES 3.00)]
+        // For Mozilla, the issue appears to be observed regardless of whether the renderer is
+        // ANGLE or not. (b/376125497)
+        bugs->rebind_buffer_after_deletion = true;
+    }
+
 #ifdef BACKEND_OPENGL_VERSION_GLES
 #   ifndef FILAMENT_IOS // FILAMENT_IOS is guaranteed to have ES3.x
     if (UTILS_UNLIKELY(major == 2)) {

filament/backend/src/opengl/OpenGLContext.h

@@ -203,6 +203,10 @@ public:
         GLint max_renderbuffer_size;
         GLint max_samples;
         GLint max_texture_image_units;
+        GLint max_texture_size;
+        GLint max_cubemap_texture_size;
+        GLint max_3d_texture_size;
+        GLint max_array_texture_layers;
         GLint max_transform_feedback_separate_attribs;
         GLint max_uniform_block_size;
         GLint max_uniform_buffer_bindings;
@@ -316,8 +320,8 @@ public:
         // a glFinish. So we must delay the destruction until we know the GPU is finished.
         bool delay_fbo_destruction;
 
-        // Mesa sometimes clears the generic buffer binding when *another* buffer is destroyed,
-        // if that other buffer is bound on an *indexed* buffer binding.
+        // Mesa and Mozilla(web) sometimes clear the generic buffer binding when *another* buffer
+        // is destroyed, if that other buffer is bound on an *indexed* buffer binding.
         bool rebind_buffer_after_deletion;
 
         // Force feature level 0. Typically used for low end ES3 devices with significant driver

filament/backend/src/opengl/OpenGLDriver.cpp

@@ -148,8 +148,8 @@ using namespace utils;
 namespace filament::backend {
 
 Driver* OpenGLDriverFactory::create(
-        OpenGLPlatform* const platform,
-        void* const sharedGLContext,
+        OpenGLPlatform* platform,
+        void* sharedGLContext,
         const Platform::DriverConfig& driverConfig) noexcept {
     return OpenGLDriver::create(platform, sharedGLContext, driverConfig);
 }
@@ -159,10 +159,10 @@ using namespace GLUtils;
 // ------------------------------------------------------------------------------------------------
 
 UTILS_NOINLINE
-OpenGLDriver* OpenGLDriver::create(OpenGLPlatform* const platform,
-        void* const /*sharedGLContext*/, const Platform::DriverConfig& driverConfig) noexcept {
+OpenGLDriver* OpenGLDriver::create(OpenGLPlatform* platform,
+        void* /*sharedGLContext*/, const Platform::DriverConfig& driverConfig) noexcept {
     assert_invariant(platform);
-    OpenGLPlatform* const ec = platform;
+    OpenGLPlatform* ec = platform;
 
 #if 0
     // this is useful for development, but too verbose even for debug builds
@@ -230,7 +230,7 @@ OpenGLDriver* OpenGLDriver::create(OpenGLPlatform* const platform,
     constexpr size_t defaultSize = FILAMENT_OPENGL_HANDLE_ARENA_SIZE_IN_MB * 1024U * 1024U;
     Platform::DriverConfig validConfig{ driverConfig };
     validConfig.handleArenaSize = std::max(driverConfig.handleArenaSize, defaultSize);
-    OpenGLDriver* const driver = new(std::nothrow) OpenGLDriver(ec, validConfig);
+    OpenGLDriver* driver = new(std::nothrow) OpenGLDriver(ec, validConfig);
     return driver;
 }
 
@@ -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
@@ -2460,6 +2461,26 @@ size_t OpenGLDriver::getMaxUniformBufferSize() {
     return mContext.gets.max_uniform_block_size;
 }
 
+size_t OpenGLDriver::getMaxTextureSize(SamplerType target) {
+    switch (target) {
+        case SamplerType::SAMPLER_2D:
+        case SamplerType::SAMPLER_2D_ARRAY:
+        case SamplerType::SAMPLER_EXTERNAL:
+            return mContext.gets.max_texture_size;
+        case SamplerType::SAMPLER_CUBEMAP:
+            return mContext.gets.max_cubemap_texture_size;
+        case SamplerType::SAMPLER_3D:
+            return mContext.gets.max_3d_texture_size;
+        case SamplerType::SAMPLER_CUBEMAP_ARRAY:
+            return mContext.gets.max_cubemap_texture_size;
+    }
+    return 0;
+}
+
+size_t OpenGLDriver::getMaxArrayTextureLayers() {
+    return mContext.gets.max_array_texture_layers;
+}
+
 // ------------------------------------------------------------------------------------------------
 // Swap chains
 // ------------------------------------------------------------------------------------------------

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/opengl/platforms/PlatformCocoaTouchGL.mm

@@ -64,7 +64,7 @@ PlatformCocoaTouchGL::~PlatformCocoaTouchGL() noexcept {
     delete pImpl;
 }
 
-Driver* PlatformCocoaTouchGL::createDriver(void* const sharedGLContext, const Platform::DriverConfig& driverConfig) noexcept {
+Driver* PlatformCocoaTouchGL::createDriver(void* sharedGLContext, const Platform::DriverConfig& driverConfig) noexcept {
     EAGLSharegroup* sharegroup = (__bridge EAGLSharegroup*) sharedGLContext;
 
     EAGLContext *context = [[EAGLContext alloc] initWithAPI:kEAGLRenderingAPIOpenGLES3 sharegroup:sharegroup];

filament/backend/src/opengl/platforms/PlatformEGLAndroid.cpp

@@ -19,6 +19,7 @@
 #include <backend/platforms/PlatformEGL.h>
 #include <backend/platforms/PlatformEGLAndroid.h>
 
+#include <private/backend/BackendUtilsAndroid.h>
 #include <private/backend/VirtualMachineEnv.h>
 
 #include "opengl/GLUtils.h"
@@ -34,6 +35,8 @@
 #include <utils/ostream.h>
 #include <utils/Panic.h>
 #include <utils/Log.h>
+#include <utils/compiler.h>
+#include <utils/ostream.h>
 
 #include <EGL/egl.h>
 #include <EGL/eglext.h>
@@ -231,26 +234,133 @@ Driver* PlatformEGLAndroid::createDriver(void* sharedContext,
     return driver;
 }
 
-PlatformEGLAndroid::ExternalImageEGLAndroid::~ExternalImageEGLAndroid() = default;
+PlatformEGLAndroid::ExternalImageEGLAndroid::~ExternalImageEGLAndroid() {
+    if (__builtin_available(android 26, *)) {
+        if (aHardwareBuffer) {
+            AHardwareBuffer_release(aHardwareBuffer);
+        }
+    }
+}
 
-Platform::ExternalImageHandle PlatformEGLAndroid::createExternalImage(AHardwareBuffer const* buffer, bool sRGB) noexcept {
-    auto* const p = new(std::nothrow) ExternalImageEGLAndroid;
-    p->aHardwareBuffer = const_cast<AHardwareBuffer*>(buffer);
-    p->sRGB = sRGB;
-    return ExternalImageHandle{ p };
+Platform::ExternalImageHandle PlatformEGLAndroid::createExternalImage(AHardwareBuffer const* buffer,
+        bool sRGB) noexcept {
+    if (__builtin_available(android 26, *)) {
+        auto* const p = new (std::nothrow) ExternalImageEGLAndroid;
+        auto hardwareBuffer = const_cast<AHardwareBuffer*>(buffer);
+        AHardwareBuffer_acquire(hardwareBuffer);
+        p->aHardwareBuffer = hardwareBuffer;
+        p->sRGB = sRGB;
+        AHardwareBuffer_Desc hardwareBufferDescription = {};
+        AHardwareBuffer_describe(hardwareBuffer, &hardwareBufferDescription);
+        p->height = hardwareBufferDescription.height;
+        p->width = hardwareBufferDescription.width;
+        auto textureFormat = mapToFilamentFormat(hardwareBufferDescription.format, sRGB);
+        p->usage = mapToFilamentUsage(hardwareBufferDescription.usage, textureFormat);
+        return ExternalImageHandle{ p };
+    }
+
+    return Platform::ExternalImageHandle{};
+}
+
+PlatformEGLAndroid::ExternalImageDescAndroid PlatformEGLAndroid::getExternalImageDesc(
+        ExternalImageHandle externalImage) noexcept {
+    auto const* const eglExternalImage =
+            static_cast<ExternalImageEGLAndroid const*>(externalImage.get());
+    ExternalImageDescAndroid metadata = {};
+    if (!eglExternalImage) {
+        return metadata;
+    }
+    metadata.height = eglExternalImage->height;
+    metadata.width = eglExternalImage->width;
+    metadata.format = eglExternalImage->format;
+    metadata.usage = eglExternalImage->usage;
+    return metadata;
 }
 
 bool PlatformEGLAndroid::setExternalImage(ExternalImageHandleRef externalImage,
         UTILS_UNUSED_IN_RELEASE ExternalTexture* texture) noexcept {
-    auto const* const eglExternalImage = static_cast<ExternalImageEGLAndroid const*>(externalImage.get());
+    auto const* const eglExternalImage =
+            static_cast<ExternalImageEGLAndroid const*>(externalImage.get());
     if (eglExternalImage->aHardwareBuffer) {
-        // TODO: implement PlatformEGLAndroid::setExternalImage w/ AHardwareBuffer
-        return true;
+        return PlatformEGLAndroid::setImage(eglExternalImage, texture);
     }
     // not a AHardwareBuffer, fallback to the inherited version
     return PlatformEGL::setExternalImage(externalImage, texture);
 }
 
+OpenGLPlatform::ExternalTexture* PlatformEGLAndroid::createExternalImageTexture() noexcept {
+    ExternalTexture* outTexture = new (std::nothrow) ExternalTexture{};
+    glGenTextures(1, &outTexture->id);
+    return outTexture;
+}
+
+void PlatformEGLAndroid::destroyExternalImageTexture(ExternalTexture* texture) noexcept {
+    glDeleteTextures(1, &texture->id);
+    delete texture;
+}
+
+bool PlatformEGLAndroid::setImage(ExternalImageEGLAndroid const* eglExternalImage,
+        UTILS_UNUSED_IN_RELEASE ExternalTexture* texture) noexcept {
+    AHardwareBuffer* hardwareBuffer = eglExternalImage->aHardwareBuffer;
+
+    // Get the EGL client buffer from AHardwareBuffer
+    EGLClientBuffer clientBuffer = eglGetNativeClientBufferANDROID(hardwareBuffer);
+    EGLint imageAttrs[] = {
+        EGL_IMAGE_PRESERVED_KHR, EGL_TRUE,
+        EGL_NONE, EGL_NONE,  // Reserve space
+        EGL_NONE, EGL_NONE,  // Reserve space
+        EGL_NONE             // Ensure the list always ends with EGL_NONE
+    };
+    int attrIndex = 2;
+    if (eglExternalImage->sRGB) {
+        imageAttrs[attrIndex++] = EGL_GL_COLORSPACE;
+        imageAttrs[attrIndex++] = EGL_GL_COLORSPACE_SRGB;
+    }
+
+    if (static_cast<bool>(eglExternalImage->usage & TextureUsage::PROTECTED)) {
+        imageAttrs[attrIndex++] = EGL_PROTECTED_CONTENT_EXT;
+        imageAttrs[attrIndex++] = EGL_TRUE;
+    }
+    // Create an EGLImage from the client buffer
+    EGLImageKHR eglImage = eglCreateImageKHR(eglGetCurrentDisplay(), EGL_NO_CONTEXT,
+            EGL_NATIVE_BUFFER_ANDROID, clientBuffer, imageAttrs);
+    if (eglImage == EGL_NO_IMAGE_KHR) {
+        // Handle error
+        slog.e << "Failed to create EGL image" << io::endl;
+        glDeleteTextures(1, &texture->id);
+        return false;
+    }
+
+    // Create and bind the OpenGL texture
+    GLint prevActiveTexture, prevTexture;
+    glGetIntegerv(GL_ACTIVE_TEXTURE, &prevActiveTexture);
+    glGetIntegerv(GL_TEXTURE_BINDING_2D, &prevTexture);
+    glActiveTexture(GL_TEXTURE0);
+    glBindTexture(texture->target, texture->id);
+    GLenum error = glGetError();
+    if (UTILS_UNLIKELY(error != GL_NO_ERROR)) {
+        slog.e << "Error after glBindTexture: " << error << io::endl;
+        glDeleteTextures(1, &texture->id);
+        eglDestroyImageKHR(eglGetCurrentDisplay(), eglImage);
+        glActiveTexture(prevActiveTexture);
+        glBindTexture(GL_TEXTURE_2D, prevTexture);
+        return false;
+    }
+    glEGLImageTargetTexture2DOES(texture->target, static_cast<GLeglImageOES>(eglImage));
+    error = glGetError();
+    if (UTILS_UNLIKELY(error != GL_NO_ERROR)) {
+        slog.e << "Error after glEGLImageTargetTexture2DOES: " << error << io::endl;
+        glDeleteTextures(1, &texture->id);
+        eglDestroyImageKHR(eglGetCurrentDisplay(), eglImage);
+        glActiveTexture(prevActiveTexture);
+        glBindTexture(GL_TEXTURE_2D, prevTexture);
+        return false;
+    }
+    glActiveTexture(prevActiveTexture);
+    glBindTexture(GL_TEXTURE_2D, prevTexture);
+    return true;
+}
+
 void PlatformEGLAndroid::setPresentationTime(int64_t presentationTimeInNanosecond) noexcept {
     EGLSurface currentDrawSurface = eglGetCurrentSurface(EGL_DRAW);
     if (currentDrawSurface != EGL_NO_SURFACE) {

filament/backend/src/opengl/platforms/PlatformGLX.cpp

@@ -128,7 +128,7 @@ namespace filament::backend {
 
 using namespace backend;
 
-Driver* PlatformGLX::createDriver(void* const sharedGLContext,
+Driver* PlatformGLX::createDriver(void* sharedGLContext,
         const DriverConfig& driverConfig) noexcept {
     loadLibraries();
     // Get the display device

filament/backend/src/opengl/platforms/PlatformOSMesa.cpp

@@ -22,6 +22,12 @@
 #include <dlfcn.h>
 #include <memory>
 
+#if defined(__linux__)
+// This is to ensure that linking during compilation will not fail even if
+// OSMesaGetProcAddress is not linked.
+__attribute__((weak)) OSMESAproc OSMesaGetProcAddress(char const*);
+#endif
+
 namespace filament::backend {
 
 using namespace backend;
@@ -44,20 +50,27 @@ struct OSMesaSwapchain {
 
 struct OSMesaAPI {
 private:
-    using CreateContextFunc = OSMesaContext (*)(GLenum format, OSMesaContext);
+    using CreateContextAttribsFunc = OSMesaContext (*)(const int *, OSMesaContext);
     using DestroyContextFunc = GLboolean (*)(OSMesaContext);
     using MakeCurrentFunc = GLboolean (*)(OSMesaContext ctx, void* buffer, GLenum type,
             GLsizei width, GLsizei height);
     using GetProcAddressFunc = OSMESAproc (*)(const char* funcName);
 
 public:
-    CreateContextFunc OSMesaCreateContext;
-    DestroyContextFunc OSMesaDestroyContext;
-    MakeCurrentFunc OSMesaMakeCurrent;
-    GetProcAddressFunc OSMesaGetProcAddress;
+    CreateContextAttribsFunc fOSMesaCreateContextAttribs;
+    DestroyContextFunc fOSMesaDestroyContext;
+    MakeCurrentFunc fOSMesaMakeCurrent;
+    GetProcAddressFunc fOSMesaGetProcAddress;
 
     OSMesaAPI() {
-        constexpr char const* libraryNames[] = {"libOSMesa.so", "libosmesa.so"};
+        static constexpr char const* libraryNames[] = {
+#if defined(__linux__)
+            "libOSMesa.so",
+            "libosmesa.so",
+#elif defined(__APPLE__)
+            "libOSMesa.dylib",
+#endif
+        };
         for (char const* libName: libraryNames) {
             mLib = dlopen(libName, RTLD_GLOBAL | RTLD_NOW);
             if (mLib) {
@@ -65,18 +78,27 @@ public:
             }
         }
         if (mLib) {
-            OSMesaGetProcAddress = (GetProcAddressFunc) dlsym(mLib, "OSMesaGetProcAddress");
-        } else {
-            OSMesaGetProcAddress = (GetProcAddressFunc) dlsym(RTLD_LOCAL, "OSMesaGetProcAddress");
+            // Loading from a libosmesa.os
+            fOSMesaGetProcAddress = (GetProcAddressFunc) dlsym(mLib, "OSMesaGetProcAddress");
         }
+#if defined(__linux__)
+        else {
+            // Filament is built into a .so
+            fOSMesaGetProcAddress = (GetProcAddressFunc) dlsym(RTLD_LOCAL, "OSMesaGetProcAddress");
+        }
+        if (!fOSMesaGetProcAddress) {
+            // Statically linking osmesa
+            fOSMesaGetProcAddress = OSMesaGetProcAddress;
+        }
+#endif // __linux__
 
-        FILAMENT_CHECK_PRECONDITION(OSMesaGetProcAddress)
-                << "Unable to against libOSMesa to create a software GL context";
+        FILAMENT_CHECK_PRECONDITION(fOSMesaGetProcAddress)
+                << "Unable to link against libOSMesa to create a software GL context";
 
-        OSMesaCreateContext = (CreateContextFunc) OSMesaGetProcAddress("OSMesaCreateContext");
-        OSMesaDestroyContext =
-                (DestroyContextFunc) OSMesaGetProcAddress("OSMesaDestroyContext");
-        OSMesaMakeCurrent = (MakeCurrentFunc) OSMesaGetProcAddress("OSMesaMakeCurrent");
+        fOSMesaCreateContextAttribs =
+                (CreateContextAttribsFunc) fOSMesaGetProcAddress("OSMesaCreateContextAttribs");
+        fOSMesaDestroyContext = (DestroyContextFunc) fOSMesaGetProcAddress("OSMesaDestroyContext");
+        fOSMesaMakeCurrent = (MakeCurrentFunc) fOSMesaGetProcAddress("OSMesaMakeCurrent");
     }
 
     ~OSMesaAPI() {
@@ -90,14 +112,24 @@ private:
 
 }// anonymous namespace
 
-Driver* PlatformOSMesa::createDriver(void* const sharedGLContext,
+Driver* PlatformOSMesa::createDriver(void* sharedGLContext,
         const DriverConfig& driverConfig) noexcept {
+
     OSMesaAPI* api = new OSMesaAPI();
     mOsMesaApi = api;
 
+    static constexpr int attribs[] = {
+        OSMESA_FORMAT, GL_RGBA,
+        OSMESA_DEPTH_BITS, 24,
+        OSMESA_STENCIL_BITS, 8,
+        OSMESA_ACCUM_BITS, 0,
+        OSMESA_PROFILE, OSMESA_CORE_PROFILE,
+        0,
+    };
+
     FILAMENT_CHECK_PRECONDITION(sharedGLContext == nullptr)
             << "shared GL context is not supported with PlatformOSMesa";
-    mContext = api->OSMesaCreateContext(GL_RGBA, NULL);
+    mContext = api->fOSMesaCreateContextAttribs(attribs, NULL);
 
     // We need to do a no-op makecurrent here so that the context will be in a correct state before
     // any GL calls.
@@ -113,7 +145,7 @@ Driver* PlatformOSMesa::createDriver(void* const sharedGLContext,
 
 void PlatformOSMesa::terminate() noexcept {
     OSMesaAPI* api = (OSMesaAPI*) mOsMesaApi;
-    api->OSMesaDestroyContext(mContext);
+    api->fOSMesaDestroyContext(mContext);
     delete api;
     mOsMesaApi = nullptr;
 
@@ -141,7 +173,7 @@ bool PlatformOSMesa::makeCurrent(ContextType type, SwapChain* drawSwapChain,
     OSMesaAPI* api = (OSMesaAPI*) mOsMesaApi;
     OSMesaSwapchain* impl = (OSMesaSwapchain*) drawSwapChain;
 
-    auto result = api->OSMesaMakeCurrent(mContext, (BackingType*) impl->buffer.get(),
+    auto result = api->fOSMesaMakeCurrent(mContext, (BackingType*) impl->buffer.get(),
             BACKING_GL_TYPE, impl->width, impl->height);
     FILAMENT_CHECK_POSTCONDITION(result == GL_TRUE) << "OSMesaMakeCurrent failed!";
 

filament/backend/src/opengl/platforms/PlatformWGL.cpp

@@ -75,7 +75,7 @@ struct WGLSwapChain {
 
 static PFNWGLCREATECONTEXTATTRIBSARBPROC wglCreateContextAttribs = nullptr;
 
-Driver* PlatformWGL::createDriver(void* const sharedGLContext,
+Driver* PlatformWGL::createDriver(void* sharedGLContext,
         const Platform::DriverConfig& driverConfig) noexcept {
     int result = 0;
     int pixelFormat = 0;

filament/backend/src/opengl/platforms/PlatformWebGL.cpp

@@ -20,7 +20,7 @@ namespace filament::backend {
 
 using namespace backend;
 
-Driver* PlatformWebGL::createDriver(void* const sharedGLContext,
+Driver* PlatformWebGL::createDriver(void* sharedGLContext,
         const Platform::DriverConfig& driverConfig) noexcept {
     return OpenGLPlatform::createDefaultDriver(this, sharedGLContext, driverConfig);
 }

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/VulkanBlitter.cpp

@@ -17,9 +17,6 @@
 #include "VulkanBlitter.h"
 #include "VulkanCommands.h"
 #include "VulkanContext.h"
-#include "VulkanFboCache.h"
-#include "VulkanHandles.h"
-#include "VulkanSamplerCache.h"
 #include "VulkanTexture.h"
 #include "vulkan/utils/Image.h"
 

filament/backend/src/vulkan/VulkanBlitter.h

@@ -24,13 +24,6 @@
 
 namespace filament::backend {
 
-class VulkanBuffer;
-class VulkanFboCache;
-class VulkanPipelineCache;
-class VulkanSamplerCache;
-
-struct VulkanProgram;
-
 class VulkanBlitter {
 public:
     VulkanBlitter(VkPhysicalDevice physicalDevice, VulkanCommands* commands) noexcept;

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

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

filament/backend/src/vulkan/VulkanContext.cpp

@@ -56,9 +56,9 @@ VkImageView VulkanAttachment::getImageView() {
     assert_invariant(texture);
     VkImageSubresourceRange range = getSubresourceRange();
     if (range.layerCount > 1) {
-        return texture->getMultiviewAttachmentView(range);
+        return texture->getAttachmentView(range, VK_IMAGE_VIEW_TYPE_2D_ARRAY);
     }
-    return texture->getAttachmentView(range);
+    return texture->getAttachmentView(range, VK_IMAGE_VIEW_TYPE_2D);
 }
 
 bool VulkanAttachment::isDepth() const {
@@ -68,11 +68,11 @@ bool VulkanAttachment::isDepth() const {
 VkImageSubresourceRange VulkanAttachment::getSubresourceRange() const {
     assert_invariant(texture);
     return {
-            .aspectMask = texture->getImageAspect(),
-            .baseMipLevel = uint32_t(level),
-            .levelCount = 1,
-            .baseArrayLayer = uint32_t(layer),
-            .layerCount = layerCount,
+        .aspectMask = texture->getImageAspect(),
+        .baseMipLevel = uint32_t(level),
+        .levelCount = 1,
+        .baseArrayLayer = uint32_t(layer),
+        .layerCount = layerCount,
     };
 }
 

filament/backend/src/vulkan/VulkanContext.h

@@ -17,7 +17,6 @@
 #ifndef TNT_FILAMENT_BACKEND_VULKANCONTEXT_H
 #define TNT_FILAMENT_BACKEND_VULKANCONTEXT_H
 
-#include "VulkanConstants.h"
 #include "vulkan/utils/Image.h"
 #include "vulkan/utils/Definitions.h"
 
@@ -30,8 +29,6 @@
 
 #include <bluevk/BlueVK.h>
 
-#include <memory>
-
 VK_DEFINE_HANDLE(VmaAllocator)
 VK_DEFINE_HANDLE(VmaPool)
 
@@ -73,13 +70,11 @@ struct VulkanRenderPass {
 // context are stored in VulkanPlatform.
 struct VulkanContext {
 public:
-    inline uint32_t selectMemoryType(uint32_t flags, VkFlags reqs) const {
-        if ((reqs & VK_MEMORY_PROPERTY_PROTECTED_BIT) != 0) {
-            assert_invariant(isProtectedMemorySupported() == true);
-        }
+    static uint32_t selectMemoryType(VkPhysicalDeviceMemoryProperties const& memoryProperties,
+            uint32_t flags, VkFlags reqs) {
         for (uint32_t i = 0; i < VK_MAX_MEMORY_TYPES; i++) {
             if (flags & 1) {
-                if ((mMemoryProperties.memoryTypes[i].propertyFlags & reqs) == reqs) {
+                if ((memoryProperties.memoryTypes[i].propertyFlags & reqs) == reqs) {
                     return i;
                 }
             }
@@ -88,6 +83,13 @@ public:
         return (uint32_t) VK_MAX_MEMORY_TYPES;
     }
 
+    inline uint32_t selectMemoryType(uint32_t flags, VkFlags reqs) const {
+        if ((reqs & VK_MEMORY_PROPERTY_PROTECTED_BIT) != 0) {
+            assert_invariant(isProtectedMemorySupported());
+        }
+        return selectMemoryType(mMemoryProperties, flags, reqs);
+    }
+
     inline fvkutils::VkFormatList const& getAttachmentDepthStencilFormats() const {
         return mDepthStencilFormats;
     }
@@ -121,7 +123,7 @@ public:
     }
 
     inline bool isMultiviewEnabled() const noexcept {
-        return mMultiviewEnabled;
+        return mPhysicalDeviceVk11Features.multiview == VK_TRUE;
     }
 
     inline bool isClipDistanceSupported() const noexcept {
@@ -145,6 +147,9 @@ private:
     VkPhysicalDeviceProperties2 mPhysicalDeviceProperties = {
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2,
     };
+    VkPhysicalDeviceVulkan11Features mPhysicalDeviceVk11Features = {
+        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES,
+    };
     VkPhysicalDeviceFeatures2 mPhysicalDeviceFeatures = {
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2,
     };
@@ -157,7 +162,6 @@ private:
     };
     bool mDebugMarkersSupported = false;
     bool mDebugUtilsSupported = false;
-    bool mMultiviewEnabled = false;
     bool mLazilyAllocatedMemorySupported = false;
     bool mProtectedMemorySupported = false;
 

filament/backend/src/vulkan/VulkanDescriptorSetCache.cpp

@@ -0,0 +1,429 @@
+/*
+ * Copyright (C) 2024 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "VulkanDescriptorSetCache.h"
+
+#include "VulkanCommands.h"
+#include "VulkanHandles.h"
+#include "VulkanConstants.h"
+
+#include <utils/FixedCapacityVector.h>
+#include <utils/Panic.h>
+
+#include <algorithm>
+#include <memory>
+#include <vector>
+
+namespace filament::backend {
+
+namespace {
+
+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
+// 'A' =>
+//   layout(binding = 0, set = 1) uniform {};
+//   layout(binding = 1, set = 1) sampler1;
+//   layout(binding = 2, set = 1) sampler2;
+//
+// would be equivalent to
+// 'B' =>
+//   layout(binding = 1, set = 2) uniform {};
+//   layout(binding = 2, set = 2) sampler2;
+//   layout(binding = 3, set = 2) sampler3;
+//
+// TODO: we might do better if we understand the types of unique layouts and can combine them in a
+// single pool without too much waste.
+class DescriptorPool {
+public:
+    DescriptorPool(VkDevice device,  DescriptorCount const& count, uint16_t capacity)
+        : mDevice(device),
+          mCount(count),
+          mCapacity(capacity),
+          mSize(0),
+          mUnusedCount(0) {
+        DescriptorCount const actual = mCount * capacity;
+        VkDescriptorPoolSize sizes[4];
+        uint8_t npools = 0;
+        if (actual.ubo) {
+            sizes[npools++] = {
+                .type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
+                .descriptorCount = actual.ubo,
+            };
+        }
+        if (actual.dynamicUbo) {
+            sizes[npools++] = {
+                .type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC,
+                .descriptorCount = actual.dynamicUbo,
+            };
+        }
+        if (actual.sampler) {
+            sizes[npools++] = {
+              .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+              .descriptorCount = actual.sampler,
+            };
+        }
+        if (actual.inputAttachment) {
+            sizes[npools++] = {
+                .type = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
+                .descriptorCount = actual.inputAttachment,
+            };
+        }
+        VkDescriptorPoolCreateInfo info{
+            .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+            .maxSets = capacity,
+            .poolSizeCount = npools,
+            .pPoolSizes = sizes,
+        };
+        vkCreateDescriptorPool(mDevice, &info, VKALLOC, &mPool);
+    }
+
+    DescriptorPool(DescriptorPool const&) = delete;
+    DescriptorPool& operator=(DescriptorPool const&) = delete;
+
+    ~DescriptorPool() {
+        vkDestroyDescriptorPool(mDevice, mPool, VKALLOC);
+    }
+
+    uint16_t const& capacity() {
+        return mCapacity;
+    }
+
+    // A convenience method for checking if this pool can allocate sets for a given layout.
+    inline bool canAllocate(DescriptorCount const& count) {
+        return count == mCount;
+    }
+
+    VkDescriptorSet obtainSet(VkDescriptorSetLayout vklayout) {
+        auto itr = findSets(vklayout);
+        if (itr != mUnused.end()) {
+            // If we don't have any unused, then just return an empty handle.
+            if (itr->second.empty()) {
+                return VK_NULL_HANDLE;
+            }
+            std::vector<VkDescriptorSet>& sets = itr->second;
+            auto set = sets.back();
+            sets.pop_back();
+            mUnusedCount--;
+            return set;
+        }
+        if (mSize + 1 > mCapacity) {
+            return VK_NULL_HANDLE;
+        }
+        // Creating a new set
+        VkDescriptorSetLayout layouts[1] = {vklayout};
+        VkDescriptorSetAllocateInfo allocInfo = {
+            .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
+            .pNext = nullptr,
+            .descriptorPool = mPool,
+            .descriptorSetCount = 1,
+            .pSetLayouts = layouts,
+        };
+        VkDescriptorSet vkSet;
+        UTILS_UNUSED VkResult result = vkAllocateDescriptorSets(mDevice, &allocInfo, &vkSet);
+        FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+                << "Failed to allocate descriptor set code=" << result << " size=" << mSize
+                << " capacity=" << mCapacity << " count=" << mUnusedCount;
+        mSize++;
+        return vkSet;
+    }
+
+    void recycle(VkDescriptorSetLayout vklayout, VkDescriptorSet vkSet) {
+        // We are recycling - release the set back into the pool. Note that the
+        // vk handle has not changed, but we need to change the backend handle to allow
+        // for proper refcounting of resources referenced in this set.
+        auto itr = findSets(vklayout);
+        if (itr != mUnused.end()) {
+            itr->second.push_back(vkSet);
+        } else {
+            mUnused.push_back(std::make_pair(vklayout, std::vector<VkDescriptorSet> {vkSet}));
+        }
+        mUnusedCount++;
+    }
+
+private:
+    using UnusedSets = std::pair<VkDescriptorSetLayout, std::vector<VkDescriptorSet>>;
+    using UnusedSetMap = std::vector<UnusedSets>;
+
+    inline UnusedSetMap::iterator findSets(VkDescriptorSetLayout vklayout) {
+        return std::find_if(mUnused.begin(), mUnused.end(), [vklayout](auto const& value) {
+            return value.first == vklayout;
+        });
+    }
+
+    VkDevice mDevice;
+    VkDescriptorPool mPool;
+    DescriptorCount const mCount;
+    uint16_t const mCapacity;
+
+    // Tracks the number of allocated descriptor sets.
+    uint16_t mSize;
+    // Tracks  the number of in-use descriptor sets.
+    uint16_t mUnusedCount;
+
+    // This maps a layout to a list of descriptor sets allocated for that layout.
+    UnusedSetMap mUnused;
+};
+
+template<typename Key>
+struct Equal {
+    bool operator()(Key const& k1, Key const& k2) const {
+        return 0 == memcmp((const void*) &k1, (const void*) &k2, sizeof(Key));
+    }
+};
+
+} // anonymous namespace
+
+// This is an ever-expanding pool of sets where it
+//    1. Keeps a list of smaller pools of different layout-dimensions.
+//    2. Will add a pool if existing pool are not compatible with the requested layout o runs out.
+class VulkanDescriptorSetCache::DescriptorInfinitePool {
+private:
+    static constexpr uint16_t EXPECTED_SET_COUNT = 10;
+    static constexpr float SET_COUNT_GROWTH_FACTOR = 1.5;
+
+public:
+    DescriptorInfinitePool(VkDevice device)
+        : mDevice(device) {}
+
+    VkDescriptorSet obtainSet(DescriptorCount const& count, VkDescriptorSetLayout vklayout) {
+        DescriptorPool* sameTypePool = nullptr;
+        for (auto& pool: mPools) {
+            if (!pool->canAllocate(count)) {
+                continue;
+            }
+            if (auto set = pool->obtainSet(vklayout); set != VK_NULL_HANDLE) {
+                return set;
+            }
+            if (!sameTypePool || sameTypePool->capacity() < pool->capacity()) {
+                sameTypePool = pool.get();
+            }
+        }
+
+        uint16_t capacity = EXPECTED_SET_COUNT;
+        if (sameTypePool) {
+            // Exponentially increase the size of the pool  to ensure we don't hit this too often.
+            capacity = std::ceil(sameTypePool->capacity() * SET_COUNT_GROWTH_FACTOR);
+        }
+
+        // We need to increase the set of pools by one.
+        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?");
+        return ret;
+    }
+
+    void recycle(DescriptorCount const& count, VkDescriptorSetLayout vklayout,
+            VkDescriptorSet vkSet) {
+        for (auto& pool: mPools) {
+            if (!pool->canAllocate(count)) {
+                continue;
+            }
+            pool->recycle(vklayout, vkSet);
+            break;
+        }
+    }
+
+private:
+    VkDevice mDevice;
+    std::vector<std::unique_ptr<DescriptorPool>> mPools;
+};
+
+
+VulkanDescriptorSetCache::VulkanDescriptorSetCache(VkDevice device,
+        fvkmemory::ResourceManager* resourceManager)
+    : mDevice(device),
+      mResourceManager(resourceManager),
+      mDescriptorPool(std::make_unique<DescriptorInfinitePool>(device)) {}
+
+VulkanDescriptorSetCache::~VulkanDescriptorSetCache() = default;
+
+void VulkanDescriptorSetCache::terminate() noexcept{
+    mDescriptorPool.reset();
+}
+
+// bind() is not really binding the set but just stashing until we have all the info
+// (pipelinelayout).
+void VulkanDescriptorSetCache::bind(uint8_t setIndex,
+        fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+        backend::DescriptorSetOffsetArray&& offsets) {
+    set->setOffsets(std::move(offsets));
+    mStashedSets[setIndex] = set;
+}
+
+void VulkanDescriptorSetCache::unbind(uint8_t setIndex) {
+    mStashedSets[setIndex] = {};
+}
+
+void VulkanDescriptorSetCache::commit(VulkanCommandBuffer* commands,
+        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 const& updateSets = mStashedSets;
+    curMask.forEachSetBit([&](size_t index) {
+        if (!updateSets[index]) {
+            curMask.unset(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([&](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, &vkset, set->uniqueDynamicUboCount, set->getOffsets()->data());
+        commands->acquire(set);
+    });
+
+    mStashedSets = {};
+
+    mLastBoundInfo = {
+        pipelineLayout,
+        setMask,
+        updateSets,
+    };
+}
+
+void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+        uint8_t binding, fvkmemory::resource_ptr<VulkanBufferObject> bufferObject,
+        VkDeviceSize offset, VkDeviceSize size) noexcept {
+    VkDescriptorBufferInfo const info = {
+        .buffer = bufferObject->buffer.getGpuBuffer(),
+        .offset = offset,
+        .range = size,
+    };
+    VkDescriptorType type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
+
+    if (set->dynamicUboMask.test(binding)) {
+        type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC;
+    }
+    VkWriteDescriptorSet descriptorWrite = {
+        .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
+        .dstSet = set->getVkSet(),
+        .dstBinding = binding,
+        .descriptorCount = 1,
+        .descriptorType = type,
+        .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::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) &&
+            expectedType == VK_IMAGE_VIEW_TYPE_2D) {
+        // If the sampler is part of a mipmapped depth texture, where one of the level *can* be
+        // an attachment, then the range for this view has exactly one level and one layer.
+        range.levelCount = 1;
+        range.layerCount = 1;
+    }
+    VkDescriptorImageInfo info{
+        .sampler = sampler,
+        .imageView = texture->getView(range),
+        .imageLayout = fvkutils::getVkLayout(texture->getDefaultLayout()),
+    };
+
+    VkWriteDescriptorSet descriptorWrite = {
+        .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
+        .pNext = nullptr,
+        .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);
+}
+
+void VulkanDescriptorSetCache::updateInputAttachment(
+        fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+        VulkanAttachment const& attachment) noexcept {
+    // TOOD: fill this in.
+}
+
+
+fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetCache::createSet(
+        Handle<HwDescriptorSet> handle, fvkmemory::resource_ptr<VulkanDescriptorSetLayout> 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*) { this->manualRecycle(count, vklayout, vkSet); },
+            vkSet);
+    return set;
+}
+
+VkDescriptorSet VulkanDescriptorSetCache::getVkSet(DescriptorCount const& count,
+        VkDescriptorSetLayout vklayout) {
+    return mDescriptorPool->obtainSet(count, vklayout);
+}
+
+void VulkanDescriptorSetCache::manualRecycle(VulkanDescriptorSetLayout::Count const& count,
+        VkDescriptorSetLayout vklayout, VkDescriptorSet 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 = {}; }
+
+} // namespace filament::backend

filament/backend/src/vulkan/VulkanDescriptorSetCache.h

@@ -0,0 +1,111 @@
+/*
+ * Copyright (C) 2024 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef TNT_FILAMENT_BACKEND_CACHING_VULKANDESCRIPTORSETCACHE_H
+#define TNT_FILAMENT_BACKEND_CACHING_VULKANDESCRIPTORSETCACHE_H
+
+#include "VulkanHandles.h"
+#include "vulkan/memory/ResourcePointer.h"
+#include "vulkan/utils/Definitions.h"  // For DescriptorSetMask
+
+#include <backend/DriverEnums.h>
+#include <backend/Program.h>
+#include <backend/TargetBufferInfo.h>
+
+#include <utils/bitset.h>
+
+#include <bluevk/BlueVK.h>
+#include <tsl/robin_map.h>
+
+#include <memory>
+
+namespace filament::backend {
+
+// Abstraction over the descriptor set pool.
+class VulkanDescriptorSetCache {
+public:
+    static constexpr uint8_t UNIQUE_DESCRIPTOR_SET_COUNT =
+            VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT;
+
+    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();
+
+    void terminate() noexcept;
+
+    void updateBuffer(fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t binding,
+            fvkmemory::resource_ptr<VulkanBufferObject> bufferObject, VkDeviceSize offset,
+            VkDeviceSize size) noexcept;
+
+    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;
+
+    void bind(uint8_t setIndex, fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+            backend::DescriptorSetOffsetArray&& offsets);
+
+    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 meant to be used with external samplers
+    VkDescriptorSet getVkSet(DescriptorCount 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; }
+
+    void gc();
+
+private:
+    void updateSamplerImpl(VkDescriptorSet set, uint8_t binding,
+            fvkmemory::resource_ptr<VulkanTexture> texture, VkSampler sampler) noexcept;
+
+    class DescriptorInfinitePool;
+
+    VkDevice mDevice;
+    fvkmemory::ResourceManager* mResourceManager;
+    std::unique_ptr<DescriptorInfinitePool> mDescriptorPool;
+    std::pair<VulkanAttachment, VkDescriptorImageInfo> mInputAttachment;
+    DescriptorSetArray mStashedSets = {};
+
+    struct {
+        VkPipelineLayout pipelineLayout = VK_NULL_HANDLE;
+        fvkutils::DescriptorSetMask setMask;
+        DescriptorSetArray boundSets = {};
+    } mLastBoundInfo;
+};
+
+}// namespace filament::backend
+
+#endif// TNT_FILAMENT_BACKEND_CACHING_VULKANDESCRIPTORSETCACHE_H

filament/backend/src/vulkan/VulkanDescriptorSetLayoutCache.cpp

@@ -0,0 +1,170 @@
+/*
+ * 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 "VulkanDescriptorSetLayoutCache.h"
+
+#include "VulkanHandles.h"
+
+#include <utils/Hash.h>
+
+namespace filament::backend {
+
+namespace {
+
+using BitmaskGroup = VulkanDescriptorSetLayout::Bitmask;
+
+template<typename Bitmask>
+uint32_t appendBindings(VkDescriptorSetLayoutBinding* toBind, VkDescriptorType type,
+        Bitmask const& mask) {
+    uint32_t count = 0;
+    Bitmask alreadySeen;
+    mask.forEachSetBit([&](size_t index) {
+        VkShaderStageFlags stages = 0;
+        uint32_t binding = 0;
+        if (index < fvkutils::getFragmentStageShift<Bitmask>()) {
+            binding = (uint32_t) index;
+            stages |= VK_SHADER_STAGE_VERTEX_BIT;
+            auto fragIndex = index + fvkutils::getFragmentStageShift<Bitmask>();
+            if (mask.test(fragIndex)) {
+                stages |= VK_SHADER_STAGE_FRAGMENT_BIT;
+                alreadySeen.set(fragIndex);
+            }
+        } else if (!alreadySeen.test(index)) {
+            // We are in fragment stage bits
+            binding = (uint32_t) (index - fvkutils::getFragmentStageShift<Bitmask>());
+            stages |= VK_SHADER_STAGE_FRAGMENT_BIT;
+        }
+
+        if (stages) {
+            toBind[count++] = {
+                .binding = binding,
+                .descriptorType = type,
+                .descriptorCount = 1,
+                .stageFlags = stages,
+            };
+        }
+    });
+    return count;
+}
+
+uint32_t appendSamplerBindings(VkDescriptorSetLayoutBinding* toBind,
+        fvkutils::SamplerBitmask const& mask, fvkutils::SamplerBitmask const& external,
+        utils::FixedCapacityVector<VkSampler> const& immutableSamplers) {
+    using Bitmask = fvkutils::SamplerBitmask;
+    uint32_t count = 0;
+    Bitmask alreadySeen;
+    uint8_t immutableIndex = 0;
+    size_t const immutableSamplerCount = immutableSamplers.size();
+    mask.forEachSetBit([&](size_t index) {
+        VkShaderStageFlags stages = 0;
+        uint32_t binding = 0;
+        if (index < fvkutils::getFragmentStageShift<Bitmask>()) {
+            binding = (uint32_t) index;
+            stages |= VK_SHADER_STAGE_VERTEX_BIT;
+            auto fragIndex = index + fvkutils::getFragmentStageShift<Bitmask>();
+            if (mask.test(fragIndex)) {
+                stages |= VK_SHADER_STAGE_FRAGMENT_BIT;
+                alreadySeen.set(fragIndex);
+            }
+        } else if (!alreadySeen.test(index)) {
+            // We are in fragment stage bits
+            binding = (uint32_t) (index - fvkutils::getFragmentStageShift<Bitmask>());
+            stages |= VK_SHADER_STAGE_FRAGMENT_BIT;
+        }
+
+        if (stages) {
+            toBind[count++] = {
+                .binding = binding,
+                .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+                .descriptorCount = 1,
+                .stageFlags = stages,
+                .pImmutableSamplers = external[index] && immutableSamplerCount > immutableIndex
+                                              ? &immutableSamplers[immutableIndex++]
+                                              : nullptr,
+            };
+        }
+    });
+    return count;
+}
+
+uint64_t computeImmutableSamplerHash(utils::FixedCapacityVector<VkSampler> const& samplers) {
+    size_t const size = samplers.size();
+    if (size == 0) {
+        return 0;
+    } else if (size == 1) {
+        return (uint64_t) samplers[0];
+    }
+    return utils::hash::murmur3((uint32_t*) samplers.data(), samplers.size() * 2, 0);
+}
+
+} // anonymous namespace
+
+VulkanDescriptorSetLayoutCache::VulkanDescriptorSetLayoutCache(VkDevice device,
+        fvkmemory::ResourceManager* resourceManager)
+    : mDevice(device),
+      mResourceManager(resourceManager) {}
+
+VulkanDescriptorSetLayoutCache::~VulkanDescriptorSetLayoutCache() = default;
+
+void VulkanDescriptorSetLayoutCache::terminate() noexcept {
+    for (auto& itr: mVkLayouts) {
+        vkDestroyDescriptorSetLayout(mDevice, itr.second, VKALLOC);
+    }
+}
+
+VkDescriptorSetLayout VulkanDescriptorSetLayoutCache::getVkLayout(
+        VulkanDescriptorSetLayout::Bitmask const& bitmasks,
+        fvkutils::SamplerBitmask externalSamplers,
+        utils::FixedCapacityVector<VkSampler> immutableSamplers) {
+    LayoutKey key = {
+        .bitmask = bitmasks,
+        .immutableSamplerHash = computeImmutableSamplerHash(immutableSamplers),
+    };
+    if (auto itr = mVkLayouts.find(key); itr != mVkLayouts.end()) {
+        return itr->second;
+    }
+
+    VkDescriptorSetLayoutBinding toBind[VulkanDescriptorSetLayout::MAX_BINDINGS];
+    uint32_t count = 0;
+    count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC,
+            bitmasks.dynamicUbo);
+    count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, bitmasks.ubo);
+    count += appendSamplerBindings(&toBind[count], bitmasks.sampler, externalSamplers,
+            immutableSamplers);
+    count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
+            bitmasks.inputAttachment);
+
+    assert_invariant(count != 0 && "Need at least one binding for descriptor set layout.");
+    VkDescriptorSetLayoutCreateInfo dlinfo = {
+        .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
+        .bindingCount = count,
+        .pBindings = toBind,
+    };
+    VkDescriptorSetLayout vklayout;
+    vkCreateDescriptorSetLayout(mDevice, &dlinfo, VKALLOC, &vklayout);
+    mVkLayouts[key] = vklayout;
+    return vklayout;
+}
+
+fvkmemory::resource_ptr<VulkanDescriptorSetLayout> VulkanDescriptorSetLayoutCache::createLayout(
+        Handle<HwDescriptorSetLayout> handle, backend::DescriptorSetLayout&& info) {
+    BitmaskGroup maskGroup = VulkanDescriptorSetLayout::Bitmask::fromLayoutDescription(info);
+    auto layout = fvkmemory::resource_ptr<VulkanDescriptorSetLayout>::make(mResourceManager, handle,
+            std::move(info), getVkLayout(maskGroup, maskGroup.externalSampler));
+    return layout;
+}
+
+} // namespace filament::backend

filament/backend/src/vulkan/VulkanDescriptorSetLayoutCache.h

@@ -0,0 +1,76 @@
+/*
+ * 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_CACHING_VULKANDESCRIPTORSETLAYOUTCACHE_H
+#define TNT_FILAMENT_BACKEND_CACHING_VULKANDESCRIPTORSETLAYOUTCACHE_H
+
+#include "VulkanHandles.h"
+
+#include "vulkan/memory/ResourcePointer.h"
+
+#include <backend/DriverEnums.h>
+#include <backend/Program.h>
+#include <backend/TargetBufferInfo.h>
+
+#include <utils/bitset.h>
+#include <utils/FixedCapacityVector.h>
+
+#include <bluevk/BlueVK.h>
+#include <tsl/robin_map.h>
+
+namespace filament::backend {
+
+class VulkanDescriptorSetLayoutCache {
+public:
+    VulkanDescriptorSetLayoutCache(VkDevice device, fvkmemory::ResourceManager* resourceManager);
+    ~VulkanDescriptorSetLayoutCache();
+
+    void terminate() noexcept;
+
+    // Just a wrapper around getVkLayout()
+    fvkmemory::resource_ptr<VulkanDescriptorSetLayout> createLayout(
+            Handle<HwDescriptorSetLayout> handle, backend::DescriptorSetLayout&& info);
+
+    // This method is meant to be used with external samplers
+    VkDescriptorSetLayout getVkLayout(VulkanDescriptorSetLayout::Bitmask const& bitmasks,
+            fvkutils::SamplerBitmask externalSamplers,
+            utils::FixedCapacityVector<VkSampler> immutableSamplers = {});
+
+private:
+    VkDevice mDevice;
+    fvkmemory::ResourceManager* mResourceManager;
+
+    struct LayoutKey {
+        // this describes the layout using bitset.
+        VulkanDescriptorSetLayout::Bitmask bitmask = {};
+        // number of immutable samplers can be arbitrary; so we hash them into 64-bit.
+        uint64_t immutableSamplerHash = 0;
+    };
+    static_assert(sizeof(LayoutKey) == 48);
+
+    using LayoutKeyHashFn = utils::hash::MurmurHashFn<LayoutKey>;
+    struct LayoutKeyEqual {
+        bool operator()(LayoutKey const& k1, LayoutKey const& k2) const {
+            return k1.bitmask == k2.bitmask && k1.immutableSamplerHash == k2.immutableSamplerHash;
+        }
+    };
+
+    tsl::robin_map<LayoutKey, VkDescriptorSetLayout, LayoutKeyHashFn, LayoutKeyEqual> mVkLayouts;
+};
+
+} // namespace filament::backend
+
+#endif// TNT_FILAMENT_BACKEND_CACHING_VULKANDESCRIPTORSETLAYOUTCACHE_H

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -17,7 +17,6 @@
 #include "VulkanDriver.h"
 
 #include "CommandStreamDispatcher.h"
-#include "DataReshaper.h"
 #include "SystraceProfile.h"
 #include "VulkanAsyncHandles.h"
 #include "VulkanBuffer.h"
@@ -26,6 +25,7 @@
 #include "VulkanHandles.h"
 #include "VulkanMemory.h"
 #include "VulkanTexture.h"
+#include "vulkan/VulkanSamplerCache.h"
 #include "vulkan/memory/ResourceManager.h"
 #include "vulkan/memory/ResourcePointer.h"
 #include "vulkan/utils/Conversion.h"
@@ -35,7 +35,6 @@
 #include <backend/platforms/VulkanPlatform.h>
 
 #include <utils/CString.h>
-#include <utils/FixedCapacityVector.h>
 #include <utils/Panic.h>
 
 #ifndef NDEBUG
@@ -44,8 +43,6 @@
 
 using namespace bluevk;
 
-using utils::FixedCapacityVector;
-
 #if defined(__clang__)
 // Vulkan functions often immediately dereference pointers, so it's fine to pass in a pointer
 // to a stack-allocated variable.
@@ -200,11 +197,9 @@ 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,
+      mResourceManager(driverConfig.handleArenaSize, driverConfig.disableHandleUseAfterFreeCheck,
               driverConfig.disableHeapHandleTags),
       mAllocator(createAllocator(mPlatform->getInstance(), mPlatform->getPhysicalDevice(),
               mPlatform->getDevice())),
@@ -216,11 +211,15 @@ VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext const& contex
       mPipelineCache(mPlatform->getDevice()),
       mStagePool(mAllocator, &mCommands),
       mFramebufferCache(mPlatform->getDevice()),
+      mYcbcrConversionCache(mPlatform->getDevice()),
       mSamplerCache(mPlatform->getDevice()),
       mBlitter(mPlatform->getPhysicalDevice(), &mCommands),
       mReadPixels(mPlatform->getDevice()),
-      mDescriptorSetManager(mPlatform->getDevice(), &mResourceManager),
+      mDescriptorSetLayoutCache(mPlatform->getDevice(), &mResourceManager),
+      mDescriptorSetCache(mPlatform->getDevice(), &mResourceManager),
       mQueryManager(mPlatform->getDevice()),
+      mExternalImageManager(platform, &mSamplerCache, &mYcbcrConversionCache, &mDescriptorSetCache,
+              &mDescriptorSetLayoutCache),
       mIsSRGBSwapChainSupported(mPlatform->getCustomization().isSRGBSwapChainSupported),
       mStereoscopicType(driverConfig.stereoscopicType) {
 
@@ -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:
@@ -313,7 +312,7 @@ void VulkanDriver::terminate() {
 
     mCurrentSwapChain = {};
     mDefaultRenderTarget = {};
-    mBoundPipeline = {};
+    mPipelineState = {};
 
     mQueryManager.terminate();
 
@@ -325,11 +324,16 @@ void VulkanDriver::terminate() {
 
     mCommands.terminate();
 
+    // Must come before samplerCache, ycbcrConversionCache, descriptorSetCache,
+    // descriptorSetLayoutCache
+    mExternalImageManager.terminate();
+
     mStagePool.terminate();
     mPipelineCache.terminate();
-    mFramebufferCache.reset();
+    mFramebufferCache.terminate();
     mSamplerCache.terminate();
-    mDescriptorSetManager.terminate();
+    mDescriptorSetLayoutCache.terminate();
+    mDescriptorSetCache.terminate();
     mPipelineLayoutCache.terminate();
 
     // Before terminating ResourceManager, we must make sure all of the resource_ptrs have been unset.
@@ -365,7 +369,7 @@ void VulkanDriver::collectGarbage() {
     FVK_SYSTRACE_SCOPE();
     // Command buffers need to be submitted and completed before other resources can be gc'd.
     mCommands.gc();
-    mDescriptorSetManager.clearHistory();
+    mDescriptorSetCache.gc();
     mStagePool.gc();
     mFramebufferCache.gc();
     mPipelineCache.gc();
@@ -380,6 +384,10 @@ void VulkanDriver::beginFrame(int64_t monotonic_clock_ns,
         int64_t refreshIntervalNs, uint32_t frameId) {
     FVK_PROFILE_MARKER(PROFILE_NAME_BEGINFRAME);
     // Do nothing.
+
+    if (mAppState.hasExternalSamplers()) {
+        mExternalImageManager.onBeginFrame();
+    }
 }
 
 void VulkanDriver::setFrameScheduledCallback(Handle<HwSwapChain> sch, CallbackHandler* handler,
@@ -408,7 +416,7 @@ void VulkanDriver::updateDescriptorSetBuffer(
     FVK_SYSTRACE_SCOPE();
     auto set = resource_ptr<VulkanDescriptorSet>::cast(&mResourceManager, dsh);
     auto buffer = resource_ptr<VulkanBufferObject>::cast(&mResourceManager, boh);
-    mDescriptorSetManager.updateBuffer(set, binding, buffer, offset, size);
+    mDescriptorSetCache.updateBuffer(set, binding, buffer, offset, size);
 }
 
 void VulkanDriver::updateDescriptorSetTexture(
@@ -420,8 +428,17 @@ void VulkanDriver::updateDescriptorSetTexture(
     auto set = resource_ptr<VulkanDescriptorSet>::cast(&mResourceManager, dsh);
     auto texture = resource_ptr<VulkanTexture>::cast(&mResourceManager, th);
 
-    VkSampler const vksampler = mSamplerCache.getSampler(params);
-    mDescriptorSetManager.updateSampler(set, binding, texture, vksampler);
+    if (mExternalImageManager.isExternallySampledTexture(texture)) {
+        mExternalImageManager.bindExternallySampledTexture(set, binding, texture, params);
+        mAppState.hasBoundExternalImages = true;
+    } else {
+        VulkanSamplerCache::Params cacheParams = {
+            .sampler = params,
+        };
+        VkSampler const vksampler = mSamplerCache.getSampler(cacheParams);
+        mDescriptorSetCache.updateSampler(set, binding, texture, vksampler);
+        mExternalImageManager.clearTextureBinding(set, binding);
+    }
 }
 
 void VulkanDriver::flush(int) {
@@ -441,6 +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);
@@ -451,12 +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();
@@ -466,12 +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);
@@ -482,12 +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);
@@ -498,12 +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();
@@ -513,6 +539,7 @@ void VulkanDriver::destroyBufferObject(Handle<HwBufferObject> boh) {
     if (!boh) {
         return;
     }
+    FVK_SYSTRACE_SCOPE();
     auto bo = resource_ptr<VulkanBufferObject>::cast(&mResourceManager, boh);
     bo.dec();
 }
@@ -554,41 +581,68 @@ void VulkanDriver::createTextureViewSwizzleR(Handle<HwTexture> th, Handle<HwText
    texture.inc();
 }
 
-void VulkanDriver::createTextureExternalImage2R(Handle<HwTexture> th,
-        backend::SamplerType target,  backend::TextureFormat format,
-        uint32_t width, uint32_t height, backend::TextureUsage usage,
+void VulkanDriver::createTextureExternalImage2R(Handle<HwTexture> th, backend::SamplerType target,
+        backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
         Platform::ExternalImageHandleRef externalImage) {
     FVK_SYSTRACE_SCOPE();
+    auto metadata = mPlatform->extractExternalImageMetadata(externalImage);
 
-    // FIXME: implement createTextureExternalImage2R
+    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);
+
+    auto imgData = mPlatform->createVkImageFromExternal(externalImage);
+
+    assert_invariant(imgData.internal.valid() || imgData.external.valid());
+
+    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.
+    texture->transitionLayout(&commands, texture->getPrimaryViewRange(), VulkanLayout::READ_ONLY);
+
+    if (imgData.external.valid()) {
+        mExternalImageManager.addExternallySampledTexture(texture, externalImage);
+    }
+
+    texture.inc();
 }
 
 void VulkanDriver::createTextureExternalImageR(Handle<HwTexture> th, backend::SamplerType target,
         backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
         void* externalImage) {
     FVK_SYSTRACE_SCOPE();
-
-    const auto& metadata = mPlatform->getExternalImageMetadata(externalImage);
-    if (metadata.isProtected) {
-        usage |= backend::TextureUsage::PROTECTED;
-    }
-
-    assert_invariant(width == metadata.width);
-    assert_invariant(height == metadata.height);
-    assert_invariant(fvkutils::getVkFormat(format) == metadata.format);
-
-    const auto& data = mPlatform->createExternalImage(externalImage, metadata);
-
-    auto texture = resource_ptr<VulkanTexture>::make(&mResourceManager, th, mPlatform->getDevice(),
-        mAllocator, &mResourceManager, &mCommands, data.first, data.second, metadata.format,
-        1, metadata.width, metadata.height, /*depth=*/1, usage, mStagePool);
-
-    texture.inc();
+    assert_invariant(false && "Not supported in Vulkan backend");
+    // not supported in this backend
 }
 
 void VulkanDriver::createTextureExternalImagePlaneR(Handle<HwTexture> th,
         backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
         void* image, uint32_t plane) {
+    assert_invariant(false && "Not supported in Vulkan backend");
 }
 
 void VulkanDriver::importTextureR(Handle<HwTexture> th, intptr_t id,
@@ -596,6 +650,7 @@ void VulkanDriver::importTextureR(Handle<HwTexture> th, intptr_t id,
         TextureFormat format, uint8_t samples, uint32_t w, uint32_t h, uint32_t depth,
         TextureUsage usage) {
     // not supported in this backend
+    assert_invariant(false && "Not supported in Vulkan backend");
 }
 
 void VulkanDriver::destroyTexture(Handle<HwTexture> th) {
@@ -604,6 +659,8 @@ void VulkanDriver::destroyTexture(Handle<HwTexture> th) {
     }
     auto texture = resource_ptr<VulkanTexture>::cast(&mResourceManager, th);
     texture.dec();
+
+    mExternalImageManager.removeExternallySampledTexture(texture);
 }
 
 void VulkanDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {
@@ -720,6 +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.
@@ -761,8 +819,8 @@ void VulkanDriver::createTimerQueryR(Handle<HwTimerQuery> tqh, int) {
 
 void VulkanDriver::createDescriptorSetLayoutR(Handle<HwDescriptorSetLayout> dslh,
         backend::DescriptorSetLayout&& info) {
-    auto layout = resource_ptr<VulkanDescriptorSetLayout>::make(&mResourceManager, dslh, info);
-    mDescriptorSetManager.initVkLayout(layout);
+    FVK_SYSTRACE_SCOPE();
+    auto layout = mDescriptorSetLayoutCache.createLayout(dslh, std::move(info));
     layout.inc();
 }
 
@@ -771,8 +829,12 @@ void VulkanDriver::createDescriptorSetR(Handle<HwDescriptorSet> dsh,
     FVK_SYSTRACE_SCOPE();
     fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout =
             fvkmemory::resource_ptr<VulkanDescriptorSetLayout>::cast(&mResourceManager, dslh);
-    auto set = mDescriptorSetManager.createSet(dsh, layout);
+    auto set = mDescriptorSetCache.createSet(dsh, layout);
     set.inc();
+
+    if (layout->hasExternalSamplers()) {
+        mAppState.hasExternalSamplerLayouts = true;
+    }
 }
 
 Handle<HwVertexBufferInfo> VulkanDriver::createVertexBufferInfoS() noexcept {
@@ -897,6 +959,10 @@ void VulkanDriver::destroyDescriptorSetLayout(Handle<HwDescriptorSetLayout> dslh
 void VulkanDriver::destroyDescriptorSet(Handle<HwDescriptorSet> dsh) {
     auto set = resource_ptr<VulkanDescriptorSet>::cast(&mResourceManager, dsh);
     set.dec();
+
+    if (mAppState.hasExternalSamplers() && set->getExternalSamplerVkSet() != VK_NULL_HANDLE) {
+        mExternalImageManager.removeDescriptorSet(set);
+    }
 }
 
 Handle<HwStream> VulkanDriver::createStreamNative(void* nativeStream) {
@@ -1043,7 +1109,7 @@ bool VulkanDriver::isDepthStencilBlitSupported(TextureFormat format) {
            formats.end();
 }
 
-bool VulkanDriver::isProtectedTexturesSupported() { return false; }
+bool VulkanDriver::isProtectedTexturesSupported() { return isProtectedContentSupported(); }
 
 bool VulkanDriver::isDepthClampSupported() {
     return mContext.isDepthClampSupported();
@@ -1120,14 +1186,28 @@ math::float2 VulkanDriver::getClipSpaceParams() {
 }
 
 uint8_t VulkanDriver::getMaxDrawBuffers() {
-    return MRT::MIN_SUPPORTED_RENDER_TARGET_COUNT; // TODO: query real value
+    return mContext.getPhysicalDeviceLimits().maxColorAttachments;
 }
 
 size_t VulkanDriver::getMaxUniformBufferSize() {
-    // TODO: return the actual size instead of hardcoded value
-    // TODO: devices that return less than 32768 should be rejected. This represents only 3%
-    //       of android devices.
-    return 32768;
+    return mContext.getPhysicalDeviceLimits().maxUniformBufferRange;
+}
+
+size_t VulkanDriver::getMaxTextureSize(SamplerType type) {
+    switch (type) {
+        case SamplerType::SAMPLER_2D:
+            return mContext.getPhysicalDeviceLimits().maxImageDimension2D;
+        case SamplerType::SAMPLER_3D:
+            return mContext.getPhysicalDeviceLimits().maxImageDimension3D;
+        case SamplerType::SAMPLER_CUBEMAP:
+            return mContext.getPhysicalDeviceLimits().maxImageDimensionCube;
+        default:
+            return mContext.getPhysicalDeviceLimits().maxImageDimension1D;
+    }
+}
+
+size_t VulkanDriver::getMaxArrayTextureLayers() {
+    return mContext.getPhysicalDeviceLimits().maxImageArrayLayers;
 }
 
 void VulkanDriver::setVertexBufferObject(Handle<HwVertexBuffer> vbh, uint32_t index,
@@ -1266,10 +1346,9 @@ void VulkanDriver::generateMipmaps(Handle<HwTexture> th) {
             mBlitter.blit(VK_FILTER_LINEAR, dst, dstOffsets, src, srcOffsets);
         }
 
-        level++;
         srcw = dstw;
         srch = dsth;
-    } while ((srcw > 1 || srch > 1) && level < t->levels);
+    } while ((srcw > 1 || srch > 1) && ++level < t->levels - 1);
 }
 
 void VulkanDriver::compilePrograms(CompilerPriorityQueue priority,
@@ -1444,7 +1523,6 @@ void VulkanDriver::endRenderPass(int) {
     // pipeline barrier between framebuffer writes and shader reads.
     rt->emitBarriersEndRenderPass(*mCurrentRenderPass.commandBuffer);
 
-    mRenderPassFboInfo = {};
     mCurrentRenderPass.renderTarget = {};
     mCurrentRenderPass.renderPass = VK_NULL_HANDLE;
 
@@ -1467,7 +1545,7 @@ void VulkanDriver::nextSubpass(int) {
 
     if (mCurrentRenderPass.params.subpassMask & 0x1) {
         VulkanAttachment& subpassInput = renderTarget->getColor0();
-        mDescriptorSetManager.updateInputAttachment({}, subpassInput);
+        mDescriptorSetCache.updateInputAttachment({}, subpassInput);
     }
 }
 
@@ -1485,7 +1563,7 @@ void VulkanDriver::makeCurrent(Handle<HwSwapChain> drawSch, Handle<HwSwapChain>
     swapChain->acquire(resized);
 
     if (resized) {
-        mFramebufferCache.reset();
+        mFramebufferCache.resetFramebuffers();
     }
 
     if (UTILS_LIKELY(mDefaultRenderTarget)) {
@@ -1504,10 +1582,10 @@ void VulkanDriver::commit(Handle<HwSwapChain> sch) {
 
 void VulkanDriver::setPushConstant(backend::ShaderStage stage, uint8_t index,
         backend::PushConstantVariant value) {
-    assert_invariant(mBoundPipeline.program && "Expect a program when writing to push constants");
+    assert_invariant(mPipelineState.program && "Expect a program when writing to push constants");
     assert_invariant(mCurrentRenderPass.commandBuffer && "Should be called within a renderpass");
-    mBoundPipeline.program->writePushConstant(mCurrentRenderPass.commandBuffer->buffer(),
-            mBoundPipeline.pipelineLayout, stage, index, value);
+    mPipelineState.program->writePushConstant(mCurrentRenderPass.commandBuffer->buffer(),
+            mPipelineState.pipelineLayout, stage, index, value);
 }
 
 void VulkanDriver::insertEventMarker(char const* string) {
@@ -1686,16 +1764,69 @@ void VulkanDriver::blitDEPRECATED(TargetBufferFlags buffers,
 }
 
 void VulkanDriver::bindPipeline(PipelineState const& pipelineState) {
+    // 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 const haveExternalSamplers = [&](auto layoutHandle) {
+            if (!layoutHandle) {
+                return false;
+            }
+            return layoutHandle->hasExternalSamplers();
+        };
+        if (std::any_of(layoutHandles.begin(), layoutHandles.end(), haveExternalSamplers)) {
+            BindInDrawBundle bundle = {
+                .pipelineState = pipelineState,
+                .dsLayoutHandles = layoutHandles,
+                .descriptorSetMask = descriptorSetMask,
+            };
+            mPipelineState.bindInDraw = { true, bundle };
+            return;
+        }
+    }
+
+    // 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,
+        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.
@@ -1737,28 +1868,11 @@ void VulkanDriver::bindPipeline(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};
-
-    mBoundPipeline = {
-        .program = program,
-        .pipelineLayout = pipelineLayout,
-        .descriptorSetMask = fvkutils::DescriptorSetMask(descriptorSetMaskTable[layoutCount]),
-    };
+    // 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);
@@ -1794,24 +1908,50 @@ void VulkanDriver::bindDescriptorSet(
         backend::DescriptorSetOffsetArray&& offsets) {
     if (dsh) {
         auto set = resource_ptr<VulkanDescriptorSet>::cast(&mResourceManager, dsh);
-        mDescriptorSetManager.bind(setIndex, set, std::move(offsets));
+        mDescriptorSetCache.bind(setIndex, set, std::move(offsets));
     } else {
-        mDescriptorSetManager.unbind(setIndex);
+        mDescriptorSetCache.unbind(setIndex);
     }
 }
 
 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;
 
-    mDescriptorSetManager.commit(mCurrentRenderPass.commandBuffer,
-            mBoundPipeline.pipelineLayout,
-            mBoundPipeline.descriptorSetMask);
+    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,
+            setsWithExternalSamplers, mPipelineState.descriptorSetMask);
 
     // Finally, make the actual draw call. TODO: support subranges
-    const uint32_t firstIndex = indexOffset;
-    const int32_t vertexOffset = 0;
-    const uint32_t firstInstId = 0;
+    uint32_t const firstIndex = indexOffset;
+    constexpr int32_t vertexOffset = 0;
+    constexpr uint32_t firstInstId = 0;
 
     vkCmdDrawIndexed(cmdbuffer, indexCount, instanceCount, firstIndex, vertexOffset, firstInstId);
 }

filament/backend/src/vulkan/VulkanDriver.h

@@ -27,8 +27,11 @@
 #include "VulkanSamplerCache.h"
 #include "VulkanStagePool.h"
 #include "VulkanQueryManager.h"
-#include "vulkan/caching/VulkanDescriptorSetManager.h"
-#include "vulkan/caching/VulkanPipelineLayoutCache.h"
+#include "VulkanYcbcrConversionCache.h"
+#include "vulkan/VulkanDescriptorSetCache.h"
+#include "vulkan/VulkanDescriptorSetLayoutCache.h"
+#include "vulkan/VulkanExternalImageManager.h"
+#include "vulkan/VulkanPipelineLayoutCache.h"
 #include "vulkan/memory/ResourceManager.h"
 #include "vulkan/memory/ResourcePointer.h"
 #include "vulkan/utils/Definitions.h"
@@ -52,7 +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
@@ -87,8 +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;
 
@@ -117,6 +120,8 @@ private:
 
 private:
     void collectGarbage();
+    void bindPipelineImpl(PipelineState const& pipelineState, VkPipelineLayout pipelineLayout,
+            fvkutils::DescriptorSetMask descriptorSetMask);
 
     VulkanPlatform* mPlatform = nullptr;
     fvkmemory::ResourceManager mResourceManager;
@@ -134,26 +139,46 @@ private:
     VulkanPipelineCache mPipelineCache;
     VulkanStagePool mStagePool;
     VulkanFboCache mFramebufferCache;
+    VulkanYcbcrConversionCache mYcbcrConversionCache;
     VulkanSamplerCache mSamplerCache;
     VulkanBlitter mBlitter;
     VulkanReadPixels mReadPixels;
-    VulkanDescriptorSetManager mDescriptorSetManager;
+    VulkanDescriptorSetLayoutCache mDescriptorSetLayoutCache;
+    VulkanDescriptorSetCache mDescriptorSetCache;
     VulkanQueryManager mQueryManager;
+    VulkanExternalImageManager mExternalImageManager;
 
     // This is necessary for us to write to push constants after binding a pipeline.
-    struct {
-        resource_ptr<VulkanProgram> program;
-        VkPipelineLayout pipelineLayout;
-        fvkutils::DescriptorSetMask descriptorSetMask;
-    } mBoundPipeline = {};
+    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 = {};
+    };
 
-    // We need to store information about a render pass to enable better barriers at the end of a
-    // renderpass.
     struct {
-        using AttachmentArray =
-                fvkutils::StaticVector<VulkanAttachment, MAX_RENDERTARGET_ATTACHMENT_TEXTURES>;
-        AttachmentArray attachments;
-    } mRenderPassFboInfo = {};
+        // For push constant
+        resource_ptr<VulkanProgram> program = {};
+        // For push commiting dynamic ubos in draw()
+        VkPipelineLayout pipelineLayout = VK_NULL_HANDLE;
+        fvkutils::DescriptorSetMask descriptorSetMask = {};
+
+        std::pair<bool, BindInDrawBundle> bindInDraw = {false, {}};
+    } mPipelineState = {};
+
+    struct {
+        // This tracks whether the app has seen external samplers bound to a the descriptor set.
+        // This will force bindPipeline to take a slow path.
+        bool hasExternalSamplerLayouts = false;
+        bool hasBoundExternalImages = false;
+
+        bool hasExternalSamplers() const noexcept {
+            return hasExternalSamplerLayouts && hasBoundExternalImages;
+        }
+    } mAppState;
 
     bool const mIsSRGBSwapChainSupported;
     backend::StereoscopicType const mStereoscopicType;

filament/backend/src/vulkan/VulkanExternalImageManager.cpp

@@ -0,0 +1,294 @@
+/*
+ * 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 "VulkanExternalImageManager.h"
+
+#include "VulkanDescriptorSetCache.h"
+#include "VulkanDescriptorSetLayoutCache.h"
+#include "VulkanSamplerCache.h"
+#include "VulkanYcbcrConversionCache.h"
+#include "vulkan/memory/ResourcePointer.h"
+#include "vulkan/utils/Conversion.h"
+
+#include <backend/platforms/VulkanPlatform.h>
+
+#include <algorithm>
+
+namespace filament::backend {
+
+namespace {
+
+using Bitmask = fvkutils::UniformBufferBitmask;
+static_assert(sizeof(Bitmask) * 8 == fvkutils::MAX_DESCRIPTOR_SET_BITMASK_BITS);
+
+template<typename T>
+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());
+}
+
+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,
+        VulkanDescriptorSetCache* setCache, VulkanDescriptorSetLayoutCache* layoutCache)
+    : mPlatform(platform),
+      mSamplerCache(samplerCache),
+      mYcbcrConversionCache(ycbcrConversionCache),
+      mDescriptorSetCache(setCache),
+      mDescriptorSetLayoutCache(layoutCache) {
+}
+
+VulkanExternalImageManager::~VulkanExternalImageManager() = default;
+
+void VulkanExternalImageManager::terminate() {
+    mSetBindings.clear();
+    mImages.clear();
+}
+
+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;
+    });
+}
+
+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 (hasExternalSampler(set)) {
+            updateSetAndLayout(set, layout);
+            shouldUseExternalSampler.set(i);
+        }
+    }
+    return shouldUseExternalSampler;
+}
+
+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) {
+    utils::FixedCapacityVector<
+            std::tuple<uint8_t, VkSampler, fvkmemory::resource_ptr<VulkanTexture>>>
+            samplerAndBindings;
+    samplerAndBindings.reserve(MAX_SAMPLER_COUNT);
+
+    fvkutils::SamplerBitmask actualExternalSamplers;
+    for (auto& bindingInfo : mSetBindings) {
+        if (bindingInfo.set != set || bindingInfo.bound) {
+            continue;
+        }
+        auto& imageData = findImage(mImages, bindingInfo.image);
+        updateImage(&imageData);
+
+        auto samplerParams = bindingInfo.samplerParams;
+        // according to spec, these must match chromaFilter
+        // https://registry.khronos.org/vulkan/specs/latest/man/html/VkSamplerCreateInfo.html#VUID-VkSamplerCreateInfo-minFilter-01645
+        samplerParams.filterMag = SamplerMagFilter::NEAREST;
+        samplerParams.filterMin = SamplerMinFilter::NEAREST;
+
+        auto sampler = mSamplerCache->getSampler({
+            .sampler = samplerParams,
+            .conversion = imageData.conversion,
+        });
+        actualExternalSamplers.set(bindingInfo.binding);
+        samplerAndBindings.push_back({ bindingInfo.binding, sampler, bindingInfo.image });
+        bindingInfo.bound = true;
+    }
+
+    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(std::get<1>(b)); });
+
+    VkDescriptorSetLayout const oldLayout = layout->getExternalSamplerVkLayout();
+    VkDescriptorSetLayout const newLayout = mDescriptorSetLayoutCache->getVkLayout(layout->bitmask,
+            actualExternalSamplers, outSamplers);
+
+    // 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);
+
+        // 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);
+
+        set->setExternalSamplerVkSet(newSet, [&](VulkanDescriptorSet*) {
+            mDescriptorSetCache->manualRecycle(layout->count, newLayout, newSet);
+        });
+        if (oldLayout != newLayout) {
+            layout->setExternalSamplerVkLayout(newLayout);
+        }
+    }
+
+    // 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 && !fvkutils::isVKYcbcrConversionFormat(metadata.format)) {
+        return VK_NULL_HANDLE;
+    }
+    VulkanYcbcrConversionCache::Params ycbcrParams = {
+        .conversion = {
+            .ycbcrModel = fvkutils::getYcbcrModelConversionFilament(metadata.ycbcrModel),
+            .r = fvkutils::getSwizzleFilament(metadata.ycbcrConversionComponents.r, 0),
+            .g = fvkutils::getSwizzleFilament(metadata.ycbcrConversionComponents.g, 1),
+            .b = fvkutils::getSwizzleFilament(metadata.ycbcrConversionComponents.b, 2),
+            .a = fvkutils::getSwizzleFilament(metadata.ycbcrConversionComponents.a, 3),
+            .ycbcrRange = fvkutils::getYcbcrRangeFilament(metadata.ycbcrRange),
+            .xChromaOffset = fvkutils::getChromaLocationFilament(metadata.xChromaOffset),
+            .yChromaOffset = fvkutils::getChromaLocationFilament(metadata.yChromaOffset),
+
+            // Unclear where to get the chromaFilter, we just assume it's nearest.
+            .chromaFilter = SamplerMagFilter::NEAREST,
+        },
+        .format = metadata.format,
+        .externalFormat = metadata.externalFormat,
+    };
+    return mYcbcrConversionCache->getConversion(ycbcrParams);
+}
+
+void VulkanExternalImageManager::updateImage(ImageData* image) {
+    if (image->hasBeenValidated) {
+        return;
+    }
+    image->hasBeenValidated = true;
+
+    auto metadata = mPlatform->extractExternalImageMetadata(image->platformHandle);
+    auto vkYcbcr = getVkSamplerYcbcrConversion(metadata);
+    if (vkYcbcr == image->conversion) {
+        return;
+    }
+
+    image->image->setYcbcrConversion(vkYcbcr);
+    image->conversion = vkYcbcr;
+    return;
+}
+
+void VulkanExternalImageManager::removeDescriptorSet(
+        fvkmemory::resource_ptr<VulkanDescriptorSet> inSet) {
+    erasep<SetBindingInfo>(mSetBindings,
+            [&](auto const& bindingInfo) { return (bindingInfo.set == inSet); });
+}
+
+void VulkanExternalImageManager::bindExternallySampledTexture(
+        fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t bindingPoint,
+        fvkmemory::resource_ptr<VulkanTexture> image, SamplerParams samplerParams) {
+    // Should we do duplicate validation here?
+    auto& imageData = findImage(mImages, image);
+    mSetBindings.push_back({ bindingPoint, imageData.image, set, samplerParams });
+}
+
+void VulkanExternalImageManager::addExternallySampledTexture(
+       fvkmemory::resource_ptr<VulkanTexture> image,
+        Platform::ExternalImageHandleRef platformHandleRef) {
+    mImages.push_back({ image, platformHandleRef, false });
+}
+
+void VulkanExternalImageManager::removeExternallySampledTexture(
+        fvkmemory::resource_ptr<VulkanTexture> image) {
+    erasep<SetBindingInfo>(mSetBindings,
+            [&](auto const& bindingInfo) { return (bindingInfo.image == 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.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

@@ -0,0 +1,121 @@
+/*
+ * 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_CACHING_VULKANEXTERNALIMAGEMANAGER_H
+#define TNT_FILAMENT_BACKEND_CACHING_VULKANEXTERNALIMAGEMANAGER_H
+
+#include "VulkanHandles.h"
+
+#include <backend/DriverEnums.h>
+
+#include <array>
+#include <vector>
+
+namespace filament::backend {
+
+class VulkanYcbcrConversionCache;
+class VulkanSamplerCache;
+class VulkanDescriptorSetLayoutCache;
+class VulkanDescriptorSetCache;
+
+// Manages the logic of external images and their quirks wrt Vulikan.
+class VulkanExternalImageManager {
+public:
+
+    VulkanExternalImageManager(
+            VulkanPlatform* platform,
+            VulkanSamplerCache* samplerCache,
+            VulkanYcbcrConversionCache* ycbcrConversionCache,
+            VulkanDescriptorSetCache* setCache,
+            VulkanDescriptorSetLayoutCache* layoutCache);
+
+    ~VulkanExternalImageManager();
+
+    void terminate();
+
+    void onBeginFrame();
+
+    using SetArray = std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>,
+            VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT>;
+
+    using LayoutArray = std::array<fvkmemory::resource_ptr<VulkanDescriptorSetLayout>,
+            VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT>;
+
+    using VkLayoutArray = VulkanDescriptorSetLayout::DescriptorSetLayoutArray;
+
+    // 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);
+
+    void bindExternallySampledTexture(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+            uint8_t bindingPoint, fvkmemory::resource_ptr<VulkanTexture> image,
+            SamplerParams samplerParams);
+
+    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);
+
+    bool isExternallySampledTexture(fvkmemory::resource_ptr<VulkanTexture> image) const;
+
+    VkSamplerYcbcrConversion getVkSamplerYcbcrConversion(
+            VulkanPlatform::ExternalImageMetadata const& metadata);
+
+    struct ImageData {
+        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;
+    };
+
+private:
+    bool hasExternalSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set);
+
+    void updateSetAndLayout(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+            fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
+
+    void updateImage(ImageData* imageData);
+
+    VulkanPlatform* mPlatform;
+    VulkanSamplerCache* mSamplerCache;
+    VulkanYcbcrConversionCache* mYcbcrConversionCache;
+    VulkanDescriptorSetCache* mDescriptorSetCache;
+    VulkanDescriptorSetLayoutCache* mDescriptorSetLayoutCache;
+
+    using SetAndLayout = std::pair<fvkmemory::resource_ptr<VulkanDescriptorSet>,
+            fvkmemory::resource_ptr<VulkanDescriptorSetLayout>>;
+
+    struct SetBindingInfo {
+        uint8_t binding = 0;
+        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<SetBindingInfo> mSetBindings;
+    std::vector<ImageData> mImages;
+};
+
+} // filament::backend
+
+#endif // TNT_FILAMENT_BACKEND_CACHING_VULKANEXTERNALIMAGEMANAGER_H

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

@@ -30,6 +30,7 @@
 
 #include <backend/platforms/VulkanPlatform.h>
 
+#include <utils/compiler.h> // UTILS_FALLTHROUGH
 #include <utils/Panic.h>    // ASSERT_POSTCONDITION
 
 using namespace bluevk;
@@ -89,8 +90,9 @@ BitmaskGroup fromBackendLayout(DescriptorSetLayout const& layout) {
                 }
                 break;
             }
-            // TODO: properly handle external sampler
             case DescriptorType::SAMPLER_EXTERNAL:
+                fromStageFlags(binding.stageFlags, binding.binding, mask.externalSampler);
+                UTILS_FALLTHROUGH;
             case DescriptorType::SAMPLER: {
                 fromStageFlags(binding.stageFlags, binding.binding, mask.sampler);
                 break;
@@ -147,11 +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);