Include webgpu matc arg in android builds when appropriate.

- Refactor VulkanTextureState and VulkanTexture so that most
  state fields are consts.
- Add a field for VkYcbcrConversion.
- Clear the image view cache when YcbcrConversion changes
- Make sure that getAttachmentView and getView are better defined
  and called.

.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/workflows/android-continuous.yml

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

.github/workflows/linux-continuous.yml

@@ -10,7 +10,7 @@ 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

.github/workflows/presubmit.yml

@@ -40,7 +40,7 @@ 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
@@ -69,7 +69,7 @@ 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
@@ -79,7 +79,7 @@ jobs:
 
   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 +92,33 @@ 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: Set up Homebrew
+        id: set-up-homebrew
+        uses: Homebrew/actions/setup-homebrew@master
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.x'
+      - name: Install python prereqs
+        run: pip install mako setuptools pyyaml
       - name: Run script
         run: |
-          source ./build/linux/ci-common.sh && bash test/renderdiff/test.sh
+          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
+      - name: Run build script
+        run: source ./build/linux/ci-common.sh && ./build.sh -W debug test_filamat filament
+      - name: Run test
+        run: ./out/cmake-debug/libs/filamat/test_filamat --gtest_filter=MaterialCompiler.Wgsl*

.github/workflows/release.yml

@@ -9,6 +9,11 @@ name: Release
 env:
   RELEASE_TAG: ${{ github.event.inputs.release_tag }}
 
+# The default GITHUB_TOKEN does not have write permissions, which is needed to upload release
+# assets.
+permissions:
+  contents: write
+
 on:
   workflow_dispatch:
     inputs:
@@ -65,7 +70,7 @@ 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:
@@ -98,7 +103,7 @@ 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:

.github/workflows/web-continuous.yml

@@ -10,7 +10,7 @@ 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

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)

NEW_RELEASE_NOTES.md

@@ -7,4 +7,5 @@ for next branch cut* header.
 appropriate header in [RELEASE_NOTES.md](./RELEASE_NOTES.md).
 
 ## Release notes for next branch cut
-- materials: add an optional `transformName` field to sampler material parameters. [⚠️ **New Material Version**]
+
+- Fix build/compile errors when upgrading to MacOS 15.4

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.58.0'
+    implementation 'com.google.android.filament:filament-android:1.59.0'
 }
 ```
 
@@ -51,7 +51,7 @@ Here are all the libraries available in the group `com.google.android.filament`:
 iOS projects can use CocoaPods to install the latest release:
 
 ```shell
-pod 'Filament', '~> 1.58.0'
+pod 'Filament', '~> 1.59.0'
 ```
 
 ## Documentation

RELEASE_NOTES.md

@@ -7,6 +7,24 @@ 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).
 
+## v1.59.1
+
+
+## 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: breaking changes to API KTX1Loader::createIndirectLight and KTX1Loader::createSkybox

android/build.gradle

@@ -71,10 +71,9 @@ buildscript {
             .gradleProperty("com.google.android.filament.include-webgpu")
             .isPresent()
 
-    // TODO: Uncomment below when fgviewer is ready
-    // def fgviewer = providers
-    //         .gradleProperty("com.google.android.filament.fgviewer")
-    //         .isPresent()
+    def fgviewer = providers
+                 .gradleProperty("com.google.android.filament.fgviewer")
+                 .isPresent()
 
     def matdbg = providers
             .gradleProperty("com.google.android.filament.matdbg")
@@ -131,8 +130,7 @@ buildscript {
         "-DFILAMENT_DIST_DIR=${filamentPath}".toString(),
         "-DFILAMENT_SUPPORTS_VULKAN=${excludeVulkan ? 'OFF' : 'ON'}".toString(),
         "-DFILAMENT_SUPPORTS_WEBGPU=${includeWebGPU ? 'ON' : 'OFF'}".toString(),
-        // TODO: Uncomment below when fgviewer is ready
-        // "-DFILAMENT_ENABLE_FGVIEWER=${fgviewer ? 'ON' : 'OFF'}".toString(),
+        "-DFILAMENT_ENABLE_FGVIEWER=${fgviewer ? 'ON' : 'OFF'}".toString(),
         "-DFILAMENT_ENABLE_MATDBG=${matdbg ? 'ON' : 'OFF'}".toString(),
         "-DFILAMENT_DISABLE_MATOPT=${matnopt ? 'ON' : 'OFF'}".toString()
     ]

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

@@ -143,6 +143,15 @@ abstract class MaterialCompiler extends TaskWithBinary {
                     matcArgs += ['-a', 'vulkan']
                 }
 
+                def include_webgpu = providers
+                        .gradleProperty("com.google.android.filament.include-webgpu")
+                        .forUseAtConfigurationTime().present
+
+                if (!include_webgpu) {
+                    matcArgs += ['-a', 'webgpu']
+                }
+
+
                 def mat_no_opt = providers
                         .gradleProperty("com.google.android.filament.matnopt")
                         .forUseAtConfigurationTime().present

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)
@@ -50,4 +57,5 @@ target_link_libraries(filamat-jni
         utils
         log
         smol-v
+        $<$<STREQUAL:${FILAMENT_SUPPORTS_WEBGPU},ON>:tint>
 )

android/filament-android/CMakeLists.txt

@@ -2,8 +2,7 @@ cmake_minimum_required(VERSION 3.19)
 project(filament-android)
 
 option(FILAMENT_SUPPORTS_VULKAN "Enables Vulkan on Android" OFF)
-# TODO: Uncomment below when fgviewer is ready
-# option(FILAMENT_ENABLE_FGVIEWER "Enables Frame Graph Viewer" OFF)
+option(FILAMENT_ENABLE_FGVIEWER "Enables Frame Graph Viewer" OFF)
 option(FILAMENT_ENABLE_MATDBG "Enables Material debugger" OFF)
 option(FILAMENT_DISABLE_MATOPT "Disables material optimizations" OFF)
 option(FILAMENT_SUPPORTS_WEBGPU "Enables WebGPU on Android" OFF)
@@ -30,10 +29,6 @@ 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)
@@ -60,12 +55,11 @@ add_library(smol-v STATIC IMPORTED)
 set_target_properties(smol-v PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libsmol-v.a)
 
-# TODO: Uncomment below when fgviewer is ready
-# 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)
@@ -133,8 +127,7 @@ target_link_libraries(filament-jni
     # libgeometry is PUBLIC because gltfio uses it.
     PUBLIC geometry
 
-    # TODO: Uncomment below when fgviewer is ready
-    # $<$<STREQUAL:${FILAMENT_ENABLE_FGVIEWER},ON>:fgviewer>
+    $<$<STREQUAL:${FILAMENT_ENABLE_FGVIEWER},ON>:fgviewer>
     $<$<STREQUAL:${FILAMENT_ENABLE_MATDBG},ON>:matdbg>
     $<$<STREQUAL:${FILAMENT_ENABLE_MATDBG},ON>:filamat>
     $<$<STREQUAL:${FILAMENT_SUPPORTS_VULKAN},ON>:bluevk>

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/gradle.properties

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.58.0
+VERSION_NAME=1.59.0
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 
@@ -30,4 +30,5 @@ com.google.android.filament.abis=all
 #com.google.android.filament.include-webgpu
 
 #com.google.android.filament.matdbg
+#com.google.android.filament.fgviewer
 #com.google.android.filament.matnopt

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/linux/ci-common.sh

@@ -30,6 +30,9 @@ if [[ "$GITHUB_WORKFLOW" ]]; then
     sudo apt-get install clang-$GITHUB_CLANG_VERSION libc++-$GITHUB_CLANG_VERSION-dev libc++abi-$GITHUB_CLANG_VERSION-dev
     sudo apt-get install mesa-common-dev libxi-dev libxxf86vm-dev
 
+    # For dawn
+    sudo apt-get install libxrandr-dev libxinerama-dev libxcursor-dev libxi-dev libx11-xcb-dev
+
     sudo update-alternatives --install /usr/bin/cc cc /usr/bin/clang-${GITHUB_CLANG_VERSION} 100
     sudo update-alternatives --install /usr/bin/c++ c++ /usr/bin/clang++-${GITHUB_CLANG_VERSION} 100
 fi

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)
@@ -173,10 +177,12 @@ 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/VulkanDescriptorSetCache.cpp
+            src/vulkan/VulkanDescriptorSetCache.h
+            src/vulkan/VulkanDescriptorSetLayoutCache.cpp
+            src/vulkan/VulkanDescriptorSetLayoutCache.h
+            src/vulkan/VulkanPipelineLayoutCache.cpp
+            src/vulkan/VulkanPipelineLayoutCache.h
             src/vulkan/memory/ResourceManager.cpp
             src/vulkan/memory/ResourceManager.h
             src/vulkan/memory/ResourcePointer.h
@@ -227,23 +233,56 @@ if (FILAMENT_SUPPORTS_VULKAN)
             src/vulkan/VulkanReadPixels.h
             src/vulkan/VulkanTexture.cpp
             src/vulkan/VulkanTexture.h
+            src/vulkan/VulkanYcbcrConversionCache.cpp
+            src/vulkan/VulkanYcbcrConversionCache.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/WebGPUSwapChain.cpp
+            src/webgpu/WebGPUSwapChain.h
+            src/webgpu/WebGPUHandles.cpp
+            src/webgpu/WebGPUHandles.h
     )
+    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)
+    list(APPEND SRCS src/BackendUtilsAndroid.cpp)
 endif()
 
 # ==================================================================================================
@@ -269,6 +308,11 @@ 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.
 # ==================================================================================================
@@ -389,9 +433,12 @@ 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)
@@ -426,6 +473,10 @@ 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}>
@@ -454,6 +505,9 @@ if (APPLE OR LINUX)
         test/ShaderGenerator.cpp
         test/TrianglePrimitive.cpp
         test/Arguments.cpp
+        test/ImageExpectations.cpp
+        test/Lifetimes.cpp
+        test/Shader.cpp
         test/test_FeedbackLoops.cpp
         test/test_Blit.cpp
         test/test_MissingRequiredAttributes.cpp
@@ -469,6 +523,7 @@ if (APPLE OR LINUX)
         test/test_Handles.cpp
     )
     set(BACKEND_TEST_LIBS
+        absl::str_format
         backend
         getopt
         gtest
@@ -484,7 +539,7 @@ 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/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
  */
@@ -807,6 +814,33 @@ static constexpr bool isStencilFormat(TextureFormat format) noexcept {
     }
 }
 
+inline constexpr bool isColorFormat(TextureFormat format) noexcept {
+    switch (format) {
+        // Standard color formats
+        case TextureFormat::R8:
+        case TextureFormat::RG8:
+        case TextureFormat::RGBA8:
+        case TextureFormat::R16F:
+        case TextureFormat::RG16F:
+        case TextureFormat::RGBA16F:
+        case TextureFormat::R32F:
+        case TextureFormat::RG32F:
+        case TextureFormat::RGBA32F:
+        case TextureFormat::RGB10_A2:
+        case TextureFormat::R11F_G11F_B10F:
+        case TextureFormat::SRGB8:
+        case TextureFormat::SRGB8_A8:
+        case TextureFormat::RGB8:
+        case TextureFormat::RGB565:
+        case TextureFormat::RGB5_A1:
+        case TextureFormat::RGBA4:
+            return true;
+        default:
+            break;
+    }
+    return false;
+}
+
 static constexpr bool isUnsignedIntFormat(TextureFormat format) {
     switch (format) {
         case TextureFormat::R8UI:
@@ -938,8 +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)
@@ -994,6 +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
@@ -1001,6 +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>
 
@@ -317,6 +318,11 @@ public:
          */
         uint32_t layers;
 
+        /**
+         * The numbers of samples per texel
+         */
+        VkSampleCountFlagBits samples;
+
         /**
          * The format of the external image
          */
@@ -347,22 +353,47 @@ public:
          */
         uint32_t memoryTypeBits;
     };
-    virtual ExternalImageMetadata getExternalImageMetadata(void* externalImage);
+    virtual ExternalImageMetadata getExternalImageMetadata(ExternalImageHandleRef externalImage);
 
     using ImageData = std::pair<VkImage, VkDeviceMemory>;
-    virtual ImageData createExternalImage(void* externalImage,
-            const ExternalImageMetadata& metadata);
+    virtual ImageData createExternalImageData(ExternalImageHandleRef externalImage,
+            const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex,
+            VkImageUsageFlags usage);
+
+    virtual VkSampler createExternalSampler(SamplerYcbcrConversion chroma,
+            SamplerParams sampler, uint32_t internalFormat);
+
+    virtual VkImageView createExternalImageView(SamplerYcbcrConversion chroma,
+            uint32_t internalFormat, VkImage image, VkImageSubresourceRange range,
+            VkImageViewType viewType, VkComponentMapping swizzle);
+
+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,
+    // Platform dependent helper methods
+    static ExtensionSet getSwapchainInstanceExtensionsImpl();
+
+    static ExternalImageMetadata getExternalImageMetadataImpl(ExternalImageHandleRef externalImage,
             VkDevice device);
-    static ImageData createExternalImageImpl(void* externalImage, VkDevice device,
-            const VkAllocationCallbacks* allocator, const ExternalImageMetadata& metadata);
+
+    static ImageData createExternalImageDataImpl(ExternalImageHandleRef externalImage,
+            VkDevice device, const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex,
+            VkImageUsageFlags usage);
+    static VkSampler createExternalSamplerImpl(VkDevice device,
+            SamplerYcbcrConversion chroma, SamplerParams sampler,
+            uint32_t internalFormat);
+    static VkImageView createExternalImageViewImpl(VkDevice device,
+            SamplerYcbcrConversion chroma, uint32_t internalFormat, VkImage image,
+            VkImageSubresourceRange range, VkImageViewType viewType,
+            VkComponentMapping swizzle);
 
     // 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 +1,71 @@
+/*
+ * 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:
+    Platform::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;
+
+protected:
+    struct ExternalImageVulkanAndroid : public Platform::ExternalImage {
+        AHardwareBuffer* aHardwareBuffer = nullptr;
+        bool sRGB = false;
+        unsigned int width;  // Texture width
+        unsigned int height; // Texture height
+        TextureFormat format;// Texture format
+        TextureUsage usage;  // Texture usage flags
+
+    protected:
+        ~ExternalImageVulkanAndroid() override;
+    };
+
+    virtual ExternalImageMetadata getExternalImageMetadata(ExternalImageHandleRef externalImage);
+
+    using ImageData = VulkanPlatform::ImageData;
+    virtual ImageData createExternalImageData(ExternalImageHandleRef externalImage,
+            const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex,
+            VkImageUsageFlags usage);
+
+    virtual ExtensionSet getSwapchainInstanceExtensions() const;
+
+    using SurfaceBundle = VulkanPlatform::SurfaceBundle;
+    virtual SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
+            uint64_t flags) const noexcept;
+};
+
+}// namespace filament::backend
+
+#endif// TNT_FILAMENT_BACKEND_PLATFORMS_VULKAN_PLATFORM_ANDROID_H

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

@@ -0,0 +1,59 @@
+/*
+ * 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; }
+
+    // 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/BackendUtilsAndroid.h

@@ -0,0 +1,36 @@
+/*
+ * 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_PRIVATE_BACKENDUTILSANDROID_H
+#define TNT_FILAMENT_BACKEND_PRIVATE_BACKENDUTILSANDROID_H
+
+#include <backend/DriverEnums.h>
+
+namespace filament::backend {
+
+/**
+ * Maps AHardwareBuffer format to TextureFormat.
+ */
+TextureFormat mapToFilamentFormat(unsigned int format, bool isSrgbTransfer) noexcept;
+
+/**
+ * Maps AHardwareBuffer usage to TextureUsage.
+ */
+TextureUsage mapToFilamentUsage(unsigned int usage, TextureFormat format) noexcept;
+
+} // namespace filament
+
+#endif // TNT_FILAMENT_BACKEND_PRIVATE_BACKENDUTILSANDROID_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/BackendUtilsAndroid.cpp

@@ -0,0 +1,98 @@
+/*
+ * 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 "private/backend/BackendUtilsAndroid.h"
+
+#include <android/hardware_buffer.h>
+
+namespace filament::backend {
+
+TextureFormat mapToFilamentFormat(unsigned int format, bool isSrgbTransfer) noexcept {
+    if (isSrgbTransfer) {
+        switch (format) {
+            case AHARDWAREBUFFER_FORMAT_R8G8B8_UNORM:
+                return TextureFormat::SRGB8;
+            default:
+                return TextureFormat::SRGB8_A8;
+        }
+    }
+    switch (format) {
+        case AHARDWAREBUFFER_FORMAT_R8G8B8A8_UNORM:
+        case AHARDWAREBUFFER_FORMAT_R8G8B8X8_UNORM:
+            return TextureFormat::RGBA8;
+        case AHARDWAREBUFFER_FORMAT_R8G8B8_UNORM:
+        case AHARDWAREBUFFER_FORMAT_Y8Cb8Cr8_420:
+            return TextureFormat::RGB8;
+        case AHARDWAREBUFFER_FORMAT_R5G6B5_UNORM:
+            return TextureFormat::RGB565;
+        case AHARDWAREBUFFER_FORMAT_R16G16B16A16_FLOAT:
+            return TextureFormat::RGBA16F;
+        case AHARDWAREBUFFER_FORMAT_R10G10B10A2_UNORM:
+            return TextureFormat::RGB10_A2;
+        case AHARDWAREBUFFER_FORMAT_D16_UNORM:
+            return TextureFormat::DEPTH16;
+        case AHARDWAREBUFFER_FORMAT_D24_UNORM:
+            return TextureFormat::DEPTH24;
+        case AHARDWAREBUFFER_FORMAT_D24_UNORM_S8_UINT:
+            return TextureFormat::DEPTH24_STENCIL8;
+        case AHARDWAREBUFFER_FORMAT_D32_FLOAT:
+            return TextureFormat::DEPTH32F;
+        case AHARDWAREBUFFER_FORMAT_D32_FLOAT_S8_UINT:
+            return TextureFormat::DEPTH32F_STENCIL8;
+        case AHARDWAREBUFFER_FORMAT_S8_UINT:
+            return TextureFormat::STENCIL8;
+        case AHARDWAREBUFFER_FORMAT_R8_UNORM:
+            return TextureFormat::R8;
+        default:
+            return TextureFormat::UNUSED;
+    }
+}
+
+TextureUsage mapToFilamentUsage(unsigned int usage, TextureFormat format) noexcept {
+    TextureUsage usageFlags = TextureUsage::NONE;
+
+    if (usage & AHARDWAREBUFFER_USAGE_GPU_SAMPLED_IMAGE) {
+        usageFlags |= TextureUsage::SAMPLEABLE;
+    }
+
+    if (usage & AHARDWAREBUFFER_USAGE_GPU_FRAMEBUFFER) {
+        if (isDepthFormat(format)) {
+            usageFlags |= TextureUsage::DEPTH_ATTACHMENT;
+        }
+        if (isStencilFormat(format)) {
+            usageFlags |= TextureUsage::STENCIL_ATTACHMENT;
+        }
+        if (isColorFormat(format)) {
+            usageFlags |= TextureUsage::COLOR_ATTACHMENT;
+        }
+    }
+
+    if (usage & AHARDWAREBUFFER_USAGE_GPU_DATA_BUFFER) {
+        usageFlags |= TextureUsage::UPLOADABLE;
+    }
+
+    if (usage & AHARDWAREBUFFER_USAGE_PROTECTED_CONTENT) {
+        usageFlags |= TextureUsage::PROTECTED;
+    }
+
+    if (usageFlags == TextureUsage::NONE) {
+        usageFlags = TextureUsage::COLOR_ATTACHMENT | TextureUsage::SAMPLEABLE;
+    }
+
+    return usageFlags;
+}
+
+} // namespace backend::filament

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)

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/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;
 }
 
@@ -2460,6 +2460,26 @@ 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/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/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/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)
 

filament/backend/src/vulkan/VulkanDescriptorSetCache.cpp

@@ -0,0 +1,399 @@
+/*
+ * 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 DescriptorCount = VulkanDescriptorSetLayout::Count;
+using DescriptorSetLayoutArray = VulkanDescriptorSetCache::DescriptorSetLayoutArray;
+
+// 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(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
+        auto const vklayout = layout->getVkLayout();
+        DescriptorPool* sameTypePool = nullptr;
+        for (auto& pool: mPools) {
+            if (!pool->canAllocate(layout->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,
+                DescriptorCount::fromLayoutBitmask(layout->bitmask), 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& setMask) {
+    // setMask indicates the set of descriptor sets the driver wants to bind, curMask is the
+    // actual set of sets that *needs* to be bound.
+    fvkutils::DescriptorSetMask curMask = setMask;
+
+    auto& updateSets = mStashedSets;
+    auto& lastBoundSets = mLastBoundInfo.boundSets;
+
+    setMask.forEachSetBit([&](size_t index) {
+        if (!updateSets[index] || updateSets[index] == lastBoundSets[index]) {
+            curMask.unset(index);
+        }
+    });
+
+    if (curMask.none() &&
+            (mLastBoundInfo.pipelineLayout == pipelineLayout && mLastBoundInfo.setMask == setMask &&
+                    mLastBoundInfo.boundSets == updateSets)) {
+        return;
+    }
+
+    curMask.forEachSetBit([&updateSets, commands, pipelineLayout](size_t index) {
+        // This code actually binds the descriptor sets.
+        auto set = updateSets[index];
+        VkCommandBuffer const cmdbuffer = commands->buffer();
+        vkCmdBindDescriptorSets(cmdbuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, index,
+                1, &set->vkSet, set->uniqueDynamicUboCount, set->getOffsets()->data());
+        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 const descriptorWrite = {
+        .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
+        .dstSet = set->vkSet,
+        .dstBinding = binding,
+        .descriptorCount = 1,
+        .descriptorType = type,
+        .pBufferInfo = &info,
+    };
+    vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
+    set->acquire(bufferObject);
+}
+
+void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+        uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
+        VkSampler sampler) noexcept {
+    VkDescriptorImageInfo info{
+        .sampler = sampler,
+        .imageLayout = fvkutils::getVkLayout(texture->getDefaultLayout()),
+    };
+    VkImageSubresourceRange range = texture->getPrimaryViewRange();
+    VkImageViewType const expectedType = texture->getViewType();
+    if (any(texture->usage & TextureUsage::DEPTH_ATTACHMENT) &&
+            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;
+    }
+    info.imageView = texture->getView(range);
+
+    VkWriteDescriptorSet const descriptorWrite = {
+        .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
+        .pNext = nullptr,
+        .dstSet = set->vkSet,
+        .dstBinding = binding,
+        .descriptorCount = 1,
+        .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+        .pImageInfo = &info,
+    };
+    vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
+    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);
+    auto const& count = layout->count;
+    auto const vklayout = layout->getVkLayout();
+    return fvkmemory::resource_ptr<VulkanDescriptorSet>::make(mResourceManager, handle, vkSet,
+            layout->bitmask.dynamicUbo, layout->count.dynamicUbo,
+            [vkSet, count, vklayout, this](VulkanDescriptorSet*) {
+                // Note that mDescriptorPool could be gone due to terminate (when the backend shuts
+                // down).
+                if (mDescriptorPool) {
+                    mDescriptorPool->recycle(count, vklayout, vkSet);
+                }
+            });
+}
+
+void VulkanDescriptorSetCache::gc() {
+    mStashedSets = {};
+}
+
+} // namespace filament::backend

filament/backend/src/vulkan/VulkanDescriptorSetCache.h

@@ -0,0 +1,94 @@
+/*
+ * 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;
+
+    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 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& setMask);
+
+    fvkmemory::resource_ptr<VulkanDescriptorSet> createSet(Handle<HwDescriptorSet> handle,
+            fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
+
+    void gc();
+
+private:
+    class DescriptorInfinitePool;
+
+    using DescriptorSetArray =
+            std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>, UNIQUE_DESCRIPTOR_SET_COUNT>;
+
+    VkDevice mDevice;
+    fvkmemory::ResourceManager* mResourceManager;
+    std::unique_ptr<DescriptorInfinitePool> mDescriptorPool;
+    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 +1,110 @@
+/*
+ * 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"
+
+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;
+}
+
+} // 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);
+    }
+}
+
+fvkmemory::resource_ptr<VulkanDescriptorSetLayout> VulkanDescriptorSetLayoutCache::createLayout(
+        Handle<HwDescriptorSetLayout> handle, backend::DescriptorSetLayout&& info) {
+    auto layout = fvkmemory::resource_ptr<VulkanDescriptorSetLayout>::make(mResourceManager, handle,
+            info);
+    VkDescriptorSetLayout vklayout = VK_NULL_HANDLE;
+    auto const& bitmasks = layout->bitmask;
+    if (auto itr = mVkLayouts.find(bitmasks); itr != mVkLayouts.end()) {
+        vklayout = itr->second;
+    } else {
+        VkDescriptorSetLayoutBinding toBind[VulkanDescriptorSetLayout::MAX_BINDINGS];
+        uint32_t count = 0;
+        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC,
+                bitmasks.dynamicUbo);
+        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, bitmasks.ubo);
+        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+                bitmasks.sampler);
+        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
+                bitmasks.inputAttachment);
+
+        assert_invariant(count != 0 && "Need at least one binding for descriptor set layout.");
+        VkDescriptorSetLayoutCreateInfo dlinfo = {
+            .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
+            .pNext = nullptr,
+            .bindingCount = count,
+            .pBindings = toBind,
+        };
+
+        vkCreateDescriptorSetLayout(mDevice, &dlinfo, VKALLOC, &vklayout);
+        mVkLayouts[bitmasks] = vklayout;
+    }
+    layout->setVkLayout(vklayout);
+    return layout;
+}
+
+} // namespace filament::backend

filament/backend/src/vulkan/VulkanDescriptorSetLayoutCache.h

@@ -0,0 +1,63 @@
+/*
+ * 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 <bluevk/BlueVK.h>
+#include <tsl/robin_map.h>
+
+#include <memory>
+
+namespace filament::backend {
+
+class VulkanDescriptorSetLayoutCache {
+public:
+    VulkanDescriptorSetLayoutCache(VkDevice device, fvkmemory::ResourceManager* resourceManager);
+    ~VulkanDescriptorSetLayoutCache();
+
+    void terminate() noexcept;
+
+    fvkmemory::resource_ptr<VulkanDescriptorSetLayout> createLayout(
+            Handle<HwDescriptorSetLayout> handle, backend::DescriptorSetLayout&& info);
+
+private:
+    VkDevice mDevice;
+    fvkmemory::ResourceManager* mResourceManager;
+
+    using BitmaskGroup = VulkanDescriptorSetLayout::Bitmask;
+    using BitmaskGroupHashFn = utils::hash::MurmurHashFn<BitmaskGroup>;
+    struct BitmaskGroupEqual {
+        bool operator()(BitmaskGroup const& k1, BitmaskGroup const& k2) const { return k1 == k2; }
+    };
+
+    tsl::robin_map<BitmaskGroup, VkDescriptorSetLayout, BitmaskGroupHashFn, BitmaskGroupEqual>
+            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"
@@ -26,16 +25,17 @@
 #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"
 #include "vulkan/utils/Definitions.h"
+#include "vulkan/vulkan_core.h"
 
 #include <backend/DriverEnums.h>
 #include <backend/platforms/VulkanPlatform.h>
 
 #include <utils/CString.h>
-#include <utils/FixedCapacityVector.h>
 #include <utils/Panic.h>
 
 #ifndef NDEBUG
@@ -44,8 +44,6 @@
 
 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.
@@ -216,10 +214,12 @@ 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()),
       mIsSRGBSwapChainSupported(mPlatform->getCustomization().isSRGBSwapChainSupported),
       mStereoscopicType(driverConfig.stereoscopicType) {
@@ -329,7 +329,8 @@ void VulkanDriver::terminate() {
     mPipelineCache.terminate();
     mFramebufferCache.reset();
     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 +366,7 @@ 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();
@@ -408,7 +409,7 @@ 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 +421,12 @@ 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);
+    // TODO: YcbcrConversion?
+    VulkanSamplerCache::Params cacheParams = {
+        .sampler = params,
+    };
+    VkSampler const vksampler = mSamplerCache.getSampler(cacheParams);
+    mDescriptorSetCache.updateSampler(set, binding, texture, vksampler);
 }
 
 void VulkanDriver::flush(int) {
@@ -554,41 +559,61 @@ 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();
-
-    // FIXME: implement createTextureExternalImage2R
-}
-
-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);
+    auto const& metadata = mPlatform->getExternalImageMetadata(externalImage);
     if (metadata.isProtected) {
         usage |= backend::TextureUsage::PROTECTED;
     }
 
+    VkImageUsageFlags vkUsage = metadata.usage;
+    if (any(usage & TextureUsage::BLIT_SRC)) {
+        vkUsage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
+    }
+
+    if (any(usage & (TextureUsage::BLIT_DST | TextureUsage::UPLOADABLE))) {
+        vkUsage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT;
+    }
+
     assert_invariant(width == metadata.width);
     assert_invariant(height == metadata.height);
     assert_invariant(fvkutils::getVkFormat(format) == metadata.format);
 
-    const auto& data = mPlatform->createExternalImage(externalImage, metadata);
+    VkMemoryPropertyFlags const requiredMemoryFlags = any(usage & TextureUsage::UPLOADABLE)
+                                                              ? VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
+                                                              : VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+    uint32_t const memoryTypeIndex =
+            mContext.selectMemoryType(metadata.memoryTypeBits, requiredMemoryFlags);
+    FILAMENT_CHECK_POSTCONDITION(memoryTypeIndex != VK_MAX_MEMORY_TYPES)
+            << "failed to find a valid memory type for external image memory.";
 
-    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);
+    VkImage vkimg;
+    VkDeviceMemory deviceMemory;
+    std::tie(vkimg, deviceMemory) =
+            mPlatform->createExternalImageData(externalImage, metadata, memoryTypeIndex, vkUsage);
+
+    auto texture = resource_ptr<VulkanTexture>::make(&mResourceManager, th, mContext,
+            mPlatform->getDevice(), mAllocator, &mResourceManager, &mCommands, vkimg, deviceMemory,
+            metadata.format, VK_NULL_HANDLE, metadata.samples, metadata.width, metadata.height,
+            metadata.layerCount, usage, mStagePool);
 
     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();
+    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 +621,7 @@ 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) {
@@ -761,8 +787,7 @@ 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);
+    auto layout = mDescriptorSetLayoutCache.createLayout(dslh, std::move(info));
     layout.inc();
 }
 
@@ -771,7 +796,7 @@ 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();
 }
 
@@ -1043,7 +1068,7 @@ 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 +1145,28 @@ 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 +1305,9 @@ 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,
@@ -1467,7 +1505,7 @@ void VulkanDriver::nextSubpass(int) {
 
     if (mCurrentRenderPass.params.subpassMask & 0x1) {
         VulkanAttachment& subpassInput = renderTarget->getColor0();
-        mDescriptorSetManager.updateInputAttachment({}, subpassInput);
+        mDescriptorSetCache.updateInputAttachment({}, subpassInput);
     }
 }
 
@@ -1794,9 +1832,9 @@ 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);
     }
 }
 
@@ -1804,7 +1842,7 @@ void VulkanDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t ins
     FVK_SYSTRACE_SCOPE();
     VkCommandBuffer cmdbuffer = mCurrentRenderPass.commandBuffer->buffer();
 
-    mDescriptorSetManager.commit(mCurrentRenderPass.commandBuffer,
+    mDescriptorSetCache.commit(mCurrentRenderPass.commandBuffer,
             mBoundPipeline.pipelineLayout,
             mBoundPipeline.descriptorSetMask);
 

filament/backend/src/vulkan/VulkanDriver.h

@@ -27,8 +27,10 @@
 #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/VulkanPipelineLayoutCache.h"
 #include "vulkan/memory/ResourceManager.h"
 #include "vulkan/memory/ResourcePointer.h"
 #include "vulkan/utils/Definitions.h"
@@ -134,10 +136,12 @@ private:
     VulkanPipelineCache mPipelineCache;
     VulkanStagePool mStagePool;
     VulkanFboCache mFramebufferCache;
+    VulkanYcbcrConversionCache mYcbcrConversionCache;
     VulkanSamplerCache mSamplerCache;
     VulkanBlitter mBlitter;
     VulkanReadPixels mReadPixels;
-    VulkanDescriptorSetManager mDescriptorSetManager;
+    VulkanDescriptorSetLayoutCache mDescriptorSetLayoutCache;
+    VulkanDescriptorSetCache mDescriptorSetCache;
     VulkanQueryManager mQueryManager;
 
     // This is necessary for us to write to push constants after binding a pipeline.

filament/backend/src/vulkan/VulkanPipelineCache.cpp

@@ -15,15 +15,12 @@
  */
 
 #include "VulkanPipelineCache.h"
-#include "VulkanMemory.h"
-#include "caching/VulkanDescriptorSetManager.h"
 
 #include <utils/Log.h>
 #include <utils/Panic.h>
 
 #include "VulkanConstants.h"
 #include "VulkanHandles.h"
-#include "VulkanTexture.h"
 #include "vulkan/utils/Conversion.h"
 
 #if defined(__clang__)

filament/backend/src/vulkan/VulkanPipelineCache.h

@@ -18,24 +18,18 @@
 #define TNT_FILAMENT_BACKEND_VULKANPIPELINECACHE_H
 
 #include "VulkanCommands.h"
-#include "VulkanMemory.h"
 
 #include <backend/DriverEnums.h>
 #include <backend/TargetBufferInfo.h>
 
-#include "backend/Program.h"
-
 #include <bluevk/BlueVK.h>
 
 #include <utils/bitset.h>
 #include <utils/compiler.h>
 #include <utils/Hash.h>
 
-#include <list>
 #include <tsl/robin_map.h>
 #include <type_traits>
-#include <vector>
-#include <unordered_map>
 
 namespace filament::backend {
 

filament/backend/src/vulkan/VulkanPipelineLayoutCache.h

@@ -17,7 +17,8 @@
 #ifndef TNT_FILAMENT_BACKEND_VULKANPIPELINELAYOUTCACHE_H
 #define TNT_FILAMENT_BACKEND_VULKANPIPELINELAYOUTCACHE_H
 
-#include <vulkan/VulkanHandles.h>
+#include "VulkanHandles.h"
+
 #include <bluevk/BlueVK.h>
 
 #include <utils/Hash.h>

filament/backend/src/vulkan/VulkanSamplerCache.cpp

@@ -14,7 +14,9 @@
  * limitations under the License.
  */
 
-#include "vulkan/VulkanSamplerCache.h"
+#include "VulkanSamplerCache.h"
+
+#include "VulkanConstants.h"
 #include "vulkan/utils/Conversion.h"
 
 #include <utils/Panic.h>
@@ -23,117 +25,43 @@ using namespace bluevk;
 
 namespace filament::backend {
 
-constexpr inline VkSamplerAddressMode getWrapMode(SamplerWrapMode mode) noexcept {
-    switch (mode) {
-        case SamplerWrapMode::REPEAT:
-            return VK_SAMPLER_ADDRESS_MODE_REPEAT;
-        case SamplerWrapMode::CLAMP_TO_EDGE:
-            return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
-        case SamplerWrapMode::MIRRORED_REPEAT:
-            return VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT;
-    }
-}
-
-constexpr inline VkFilter getFilter(SamplerMinFilter filter) noexcept {
-    switch (filter) {
-        case SamplerMinFilter::NEAREST:
-            return VK_FILTER_NEAREST;
-        case SamplerMinFilter::LINEAR:
-            return VK_FILTER_LINEAR;
-        case SamplerMinFilter::NEAREST_MIPMAP_NEAREST:
-            return VK_FILTER_NEAREST;
-        case SamplerMinFilter::LINEAR_MIPMAP_NEAREST:
-            return VK_FILTER_LINEAR;
-        case SamplerMinFilter::NEAREST_MIPMAP_LINEAR:
-            return VK_FILTER_NEAREST;
-        case SamplerMinFilter::LINEAR_MIPMAP_LINEAR:
-            return VK_FILTER_LINEAR;
-    }
-}
-
-constexpr inline VkFilter getFilter(SamplerMagFilter filter) noexcept {
-    switch (filter) {
-        case SamplerMagFilter::NEAREST:
-            return VK_FILTER_NEAREST;
-        case SamplerMagFilter::LINEAR:
-            return VK_FILTER_LINEAR;
-    }
-}
-
-constexpr inline VkSamplerMipmapMode getMipmapMode(SamplerMinFilter filter) noexcept {
-    switch (filter) {
-        case SamplerMinFilter::NEAREST:
-            return VK_SAMPLER_MIPMAP_MODE_NEAREST;
-        case SamplerMinFilter::LINEAR:
-            return VK_SAMPLER_MIPMAP_MODE_NEAREST;
-        case SamplerMinFilter::NEAREST_MIPMAP_NEAREST:
-            return VK_SAMPLER_MIPMAP_MODE_NEAREST;
-        case SamplerMinFilter::LINEAR_MIPMAP_NEAREST:
-            return VK_SAMPLER_MIPMAP_MODE_NEAREST;
-        case SamplerMinFilter::NEAREST_MIPMAP_LINEAR:
-            return VK_SAMPLER_MIPMAP_MODE_LINEAR;
-        case SamplerMinFilter::LINEAR_MIPMAP_LINEAR:
-            return VK_SAMPLER_MIPMAP_MODE_LINEAR;
-    }
-}
-
-constexpr inline float getMaxLod(SamplerMinFilter filter) noexcept {
-    switch (filter) {
-        case SamplerMinFilter::NEAREST:
-        case SamplerMinFilter::LINEAR:
-            // The Vulkan spec recommends a max LOD of 0.25 to "disable" mipmapping.
-            // See "Mapping of OpenGL to Vulkan filter modes" in the VK Spec.
-            return 0.25f;
-        case SamplerMinFilter::NEAREST_MIPMAP_NEAREST:
-        case SamplerMinFilter::LINEAR_MIPMAP_NEAREST:
-        case SamplerMinFilter::NEAREST_MIPMAP_LINEAR:
-        case SamplerMinFilter::LINEAR_MIPMAP_LINEAR:
-            return VK_LOD_CLAMP_NONE;
-    }
-}
-
-constexpr inline VkBool32 getCompareEnable(SamplerCompareMode mode) noexcept {
-    return mode == SamplerCompareMode::NONE ? VK_FALSE : VK_TRUE;
-}
-
 VulkanSamplerCache::VulkanSamplerCache(VkDevice device)
     : mDevice(device) {}
 
-VkSampler VulkanSamplerCache::getSampler(SamplerParams params) noexcept {
+VkSampler VulkanSamplerCache::getSampler(Params params) noexcept {
     auto iter = mCache.find(params);
     if (UTILS_LIKELY(iter != mCache.end())) {
         return iter->second;
     }
-    VkSamplerCreateInfo samplerInfo {
-        .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
-        .magFilter = getFilter(params.filterMag),
-        .minFilter = getFilter(params.filterMin),
-        .mipmapMode = getMipmapMode(params.filterMin),
-        .addressModeU = getWrapMode(params.wrapS),
-        .addressModeV = getWrapMode(params.wrapT),
-        .addressModeW = getWrapMode(params.wrapR),
-        .anisotropyEnable = params.anisotropyLog2 == 0 ? 0u : 1u,
-        .maxAnisotropy = (float)(1u << params.anisotropyLog2),
-        .compareEnable = getCompareEnable(params.compareMode),
-        .compareOp = fvkutils::getCompareOp(params.compareFunc),
+    auto const& samplerParams = params.sampler;
+    VkSamplerCreateInfo samplerInfo{ .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
+        .magFilter = fvkutils::getFilter(samplerParams.filterMag),
+        .minFilter = fvkutils::getFilter(samplerParams.filterMin),
+        .mipmapMode = fvkutils::getMipmapMode(samplerParams.filterMin),
+        .addressModeU = fvkutils::getWrapMode(samplerParams.wrapS),
+        .addressModeV = fvkutils::getWrapMode(samplerParams.wrapT),
+        .addressModeW = fvkutils::getWrapMode(samplerParams.wrapR),
+        .anisotropyEnable = samplerParams.anisotropyLog2 == 0 ? VK_FALSE : VK_TRUE,
+        .maxAnisotropy = (float) (1u << samplerParams.anisotropyLog2),
+        .compareEnable = fvkutils::getCompareEnable(samplerParams.compareMode),
+        .compareOp = fvkutils::getCompareOp(samplerParams.compareFunc),
         .minLod = 0.0f,
-        .maxLod = getMaxLod(params.filterMin),
+        .maxLod = fvkutils::getMaxLod(samplerParams.filterMin),
         .borderColor = VK_BORDER_COLOR_INT_OPAQUE_BLACK,
-        .unnormalizedCoordinates = VK_FALSE
-    };
+        .unnormalizedCoordinates = VK_FALSE };
     VkSampler sampler;
     VkResult result = vkCreateSampler(mDevice, &samplerInfo, VKALLOC, &sampler);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "Unable to create sampler."
                                                        << " error=" << static_cast<int32_t>(result);
-    mCache.insert({params, sampler});
+    mCache.insert({ params, sampler });
     return sampler;
 }
 
 void VulkanSamplerCache::terminate() noexcept {
-    for (auto pair : mCache) {
+    for (auto pair: mCache) {
         vkDestroySampler(mDevice, pair.second, VKALLOC);
     }
     mCache.clear();
 }
 
-} // namespace filament::backend
+}// namespace filament::backend

filament/backend/src/vulkan/VulkanSamplerCache.h

@@ -17,8 +17,11 @@
 #ifndef TNT_FILAMENT_BACKEND_VULKANSAMPLERCACHE_H
 #define TNT_FILAMENT_BACKEND_VULKANSAMPLERCACHE_H
 
-#include "VulkanContext.h"
+#include <backend/DriverEnums.h>
 
+#include <utils/Hash.h>
+
+#include <bluevk/BlueVK.h>
 #include <tsl/robin_map.h>
 
 namespace filament::backend {
@@ -26,12 +29,28 @@ namespace filament::backend {
 // Simple manager for VkSampler objects.
 class VulkanSamplerCache {
 public:
+    struct Params {
+        SamplerParams sampler = {};
+        uint32_t padding = 0;
+        VkSamplerYcbcrConversion conversion = VK_NULL_HANDLE;
+    };
+
+    static_assert(sizeof(Params) == 16);
+
     explicit VulkanSamplerCache(VkDevice device);
-    VkSampler getSampler(SamplerParams params) noexcept;
+    VkSampler getSampler(Params params) noexcept;
     void terminate() noexcept;
 private:
     VkDevice mDevice;
-    tsl::robin_map<SamplerParams, VkSampler, SamplerParams::Hasher, SamplerParams::EqualTo> mCache;
+
+    struct SamplerEqualTo {
+        bool operator()(Params lhs, Params rhs) const noexcept {
+            SamplerParams::EqualTo equal;
+            return equal(lhs.sampler, rhs.sampler) && lhs.conversion == rhs.conversion;
+        }
+    };
+    using SamplerHashFn = utils::hash::MurmurHashFn<Params>;
+    tsl::robin_map<Params, VkSampler, SamplerHashFn, SamplerEqualTo> mCache;
 };
 
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanSwapChain.cpp

@@ -32,6 +32,7 @@ VulkanSwapChain::VulkanSwapChain(VulkanPlatform* platform, VulkanContext const&
         VulkanCommands* commands, VulkanStagePool& stagePool, void* nativeWindow, uint64_t flags,
         VkExtent2D extent)
     : mPlatform(platform),
+      mContext(context),
       mResourceManager(resourceManager),
       mCommands(commands),
       mAllocator(allocator),
@@ -77,16 +78,16 @@ void VulkanSwapChain::update() {
     }
     for (auto const color: bundle.colors) {
         auto colorTexture = fvkmemory::resource_ptr<VulkanTexture>::construct(mResourceManager,
-                device, mAllocator, mResourceManager, mCommands, color, VK_NULL_HANDLE,
-                bundle.colorFormat, 1, bundle.extent.width, bundle.extent.height, bundle.layerCount, colorUsage,
-                mStagePool);
+                mContext, device, mAllocator, mResourceManager, mCommands, color, VK_NULL_HANDLE,
+                bundle.colorFormat, VK_NULL_HANDLE /*ycrcb */, 1, bundle.extent.width,
+                bundle.extent.height, bundle.layerCount, colorUsage, mStagePool);
         mColors.push_back(colorTexture);
     }
 
-    mDepth = fvkmemory::resource_ptr<VulkanTexture>::construct(mResourceManager, device,
-        mAllocator, mResourceManager, mCommands, bundle.depth, VK_NULL_HANDLE,
-        bundle.depthFormat, 1, bundle.extent.width, bundle.extent.height, bundle.layerCount, depthUsage,
-        mStagePool);
+    mDepth = fvkmemory::resource_ptr<VulkanTexture>::construct(mResourceManager, mContext, device,
+            mAllocator, mResourceManager, mCommands, bundle.depth, VK_NULL_HANDLE,
+            bundle.depthFormat, VK_NULL_HANDLE /*ycrcb */, 1, bundle.extent.width,
+            bundle.extent.height, bundle.layerCount, depthUsage, mStagePool);
 
     mExtent = bundle.extent;
     mLayerCount = bundle.layerCount;

filament/backend/src/vulkan/VulkanSwapChain.h

@@ -19,6 +19,7 @@
 
 #include "DriverBase.h"
 
+#include "VulkanConstants.h"
 #include "VulkanContext.h"
 #include "VulkanTexture.h"
 #include "vulkan/memory/Resource.h"
@@ -28,7 +29,6 @@
 #include <bluevk/BlueVK.h>
 #include <utils/FixedCapacityVector.h>
 
-#include <memory>
 
 using namespace bluevk;
 
@@ -83,6 +83,7 @@ private:
     void update();
 
     VulkanPlatform* mPlatform;
+    VulkanContext const& mContext;
     fvkmemory::ResourceManager* mResourceManager;
     VulkanCommands* mCommands;
     VmaAllocator mAllocator;

filament/backend/src/vulkan/VulkanTexture.cpp

@@ -15,6 +15,7 @@
  */
 
 #include "VulkanCommands.h"
+#include "VulkanContext.h"
 #include "VulkanMemory.h"
 #include "VulkanTexture.h"
 #include "vulkan/memory/ResourcePointer.h"
@@ -54,64 +55,61 @@ VkComponentMapping composeSwizzle(VkComponentMapping const& prev, VkComponentMap
         VK_COMPONENT_SWIZZLE_B,
         VK_COMPONENT_SWIZZLE_A,
     };
-
-    auto const compose = [](VkComponentSwizzle out, VkComponentMapping const& prev,
-                                 uint8_t channelIndex) {
-        // We need to first change all identities to its equivalent channel.
-        if (out == VK_COMPONENT_SWIZZLE_IDENTITY) {
-            out = IDENTITY[channelIndex];
+    auto const compose = [](VkComponentMapping const& prev,
+                                 VkComponentMapping const& next) -> VkComponentMapping {
+        VkComponentSwizzle vals[4] = { next.r, next.g, next.b, next.a };
+        for (auto& out: vals) {
+            switch (out) {
+                case VK_COMPONENT_SWIZZLE_R:
+                    out = prev.r;
+                    break;
+                case VK_COMPONENT_SWIZZLE_G:
+                    out = prev.g;
+                    break;
+                case VK_COMPONENT_SWIZZLE_B:
+                    out = prev.b;
+                    break;
+                case VK_COMPONENT_SWIZZLE_A:
+                    out = prev.a;
+                    break;
+                // Do not modify the value
+                case VK_COMPONENT_SWIZZLE_IDENTITY:
+                case VK_COMPONENT_SWIZZLE_ZERO:
+                case VK_COMPONENT_SWIZZLE_ONE:
+                // Below is not exposed in Vulkan's API, but needs to be there for compilation.
+                case VK_COMPONENT_SWIZZLE_MAX_ENUM:
+                    break;
+            }
         }
-        switch (out) {
-            case VK_COMPONENT_SWIZZLE_R:
-                out = prev.r;
-                break;
-            case VK_COMPONENT_SWIZZLE_G:
-                out = prev.g;
-                break;
-            case VK_COMPONENT_SWIZZLE_B:
-                out = prev.b;
-                break;
-            case VK_COMPONENT_SWIZZLE_A:
-                out = prev.a;
-                break;
-            case VK_COMPONENT_SWIZZLE_IDENTITY:
-            case VK_COMPONENT_SWIZZLE_ZERO:
-            case VK_COMPONENT_SWIZZLE_ONE:
-                return out;
-            // Below is not exposed in Vulkan's API, but needs to be there for compilation.
-            case VK_COMPONENT_SWIZZLE_MAX_ENUM:
-                break;
-        }
-        // If the result correctly corresponds to the identity, just return identity.
-        if (IDENTITY[channelIndex] == out) {
-            return VK_COMPONENT_SWIZZLE_IDENTITY;
-        }
-        return out;
+        return { vals[0], vals[1], vals[2], vals[3] };
     };
 
-    auto const identityToChannel = [](VkComponentSwizzle val, uint8_t channelIndex) {
-        if (val != VK_COMPONENT_SWIZZLE_IDENTITY) {
-            return val;
+    auto const identityToChannel = [](VkComponentMapping const& mapping) -> VkComponentMapping {
+        VkComponentSwizzle vals[4] = { mapping.r, mapping.g, mapping.b, mapping.a };
+        for (uint8_t i = 0; i < 4; i++) {
+            if (vals[i] != VK_COMPONENT_SWIZZLE_IDENTITY) {
+                continue;
+            }
+            vals[i] = IDENTITY[i];
         }
-        return IDENTITY[channelIndex];
+        return { vals[0], vals[1], vals[2], vals[3] };
+    };
+    auto const channelToIdentity = [](VkComponentMapping const& mapping) -> VkComponentMapping {
+        VkComponentSwizzle vals[4] = { mapping.r, mapping.g, mapping.b, mapping.a };
+        for (uint8_t i = 0; i < 4; i++) {
+            if (IDENTITY[i] != vals[i]) {
+                continue;
+            }
+            vals[i] = VK_COMPONENT_SWIZZLE_IDENTITY;
+        }
+        return { vals[0], vals[1], vals[2], vals[3] };
     };
 
     // We make sure all all identities are mapped into respective channels so that actual channel
     // mapping will be passed onto the output.
-    VkComponentMapping const prevExplicit = {
-            identityToChannel(prev.r, 0),
-            identityToChannel(prev.g, 1),
-            identityToChannel(prev.b, 2),
-            identityToChannel(prev.a, 3),
-    };
-
-    // Note that the channel index corresponds to the VkComponentMapping struct layout.
-    return {
-        compose(next.r, prevExplicit, 0),
-        compose(next.g, prevExplicit, 1),
-        compose(next.b, prevExplicit, 2),
-        compose(next.a, prevExplicit, 3),
-    };
+    VkComponentMapping const prevExplicit = identityToChannel(prev);
+    VkComponentMapping const nextExplicit = identityToChannel(next);
+    return channelToIdentity(compose(prevExplicit, nextExplicit));
 }
 
 inline VulkanLayout getDefaultLayoutImpl(TextureUsage usage) {
@@ -153,36 +151,152 @@ uint8_t getLayerCountFromDepth(uint32_t const depth) {
     return getLayerCount(getSamplerTypeFromDepth(depth), depth);
 }
 
+VkImageUsageFlags getUsage(VulkanContext const& context, uint8_t samples,
+        VkPhysicalDevice physicalDevice, VkFormat vkFormat, TextureUsage tusage) {
+    VkImageUsageFlags usage = {};
+    if (any(tusage & TextureUsage::BLIT_SRC)) {
+        usage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
+    }
+    if (any(tusage & TextureUsage::BLIT_DST)) {
+        usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT;
+    }
+
+    // Determine if we can use the transient usage flag combined with lazily allocated memory.
+    const bool useTransientAttachment =
+            // Lazily allocated memory is available.
+            context.isLazilyAllocatedMemorySupported() &&
+            // Usage consists of attachment flags only.
+            none(tusage & ~TextureUsage::ALL_ATTACHMENTS) &&
+            // Usage contains at least one attachment flag.
+            any(tusage & TextureUsage::ALL_ATTACHMENTS) &&
+            // Depth resolve cannot use transient attachment because it uses a custom shader.
+            // TODO: see VulkanDriver::isDepthStencilResolveSupported() to know when to remove this
+            // restriction.
+            // Note that the custom shader does not resolve stencil. We do need to move to vk 1.2
+            // and above to be able to support stencil resolve (along with depth).
+            !(any(tusage & TextureUsage::DEPTH_ATTACHMENT) && samples > 1);
+
+    const VkImageUsageFlags transientFlag =
+       useTransientAttachment ? VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT : 0U;
+
+    if (any(tusage & TextureUsage::SAMPLEABLE)) {
+
+#if FVK_ENABLED(FVK_DEBUG_TEXTURE)
+        if (physicalDevice != VK_NULL_HANDLE) {
+            // Validate that the format is actually sampleable.
+            VkFormatProperties props;
+            vkGetPhysicalDeviceFormatProperties(physicalDevice, vkFormat, &props);
+            if (!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT)) {
+                FVK_LOGW << "Texture usage is SAMPLEABLE but format " << mState->mVkFormat << " is not "
+                        "sampleable with optimal tiling." << utils::io::endl;
+            }
+        }
+#endif
+
+        usage |= VK_IMAGE_USAGE_SAMPLED_BIT;
+    }
+    if (any(tusage & TextureUsage::COLOR_ATTACHMENT)) {
+        usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | transientFlag;
+        if (any(tusage & TextureUsage::SUBPASS_INPUT)) {
+            usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT;
+        }
+    }
+    if (any(tusage & TextureUsage::STENCIL_ATTACHMENT)) {
+        usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | transientFlag;
+    }
+    if (any(tusage & TextureUsage::UPLOADABLE)) {
+        usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT;
+    }
+    if (any(tusage & TextureUsage::DEPTH_ATTACHMENT)) {
+        usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | transientFlag;
+
+        // Depth resolves uses a custom shader and therefore needs to be sampleable.
+        if (samples > 1) {
+            usage |= VK_IMAGE_USAGE_SAMPLED_BIT;
+        }
+    }
+    return usage;
+}
 
 } // anonymous namespace
 
-VulkanTextureState::VulkanTextureState(VkDevice device, VmaAllocator allocator,
-        VulkanCommands* commands, VulkanStagePool& stagePool, VkFormat format,
-        VkImageViewType viewType, uint8_t levels, uint8_t layerCount, VulkanLayout defaultLayout,
+VulkanTextureState::VulkanTextureState(VulkanStagePool& stagePool, VulkanCommands* commands,
+        VmaAllocator allocator, VkDevice device, VkImage image, VkDeviceMemory deviceMemory,
+        VkFormat format, VkImageViewType viewType, uint8_t levels, uint8_t layerCount,
+        VkSamplerYcbcrConversion ycbcrConversion, bool isExternalFormat, VkImageUsageFlags usage,
         bool isProtected)
-    : mVkFormat(format),
-      mViewType(viewType),
-      mFullViewRange{fvkutils::getImageAspect(format), 0, levels, 0, layerCount},
-      mDefaultLayout(defaultLayout),
-      mIsProtected(isProtected),
-      mStagePool(stagePool),
-      mDevice(device),
-      mAllocator(allocator),
+    : mStagePool(stagePool),
       mCommands(commands),
-      mIsTransientAttachment(false) {}
+      mAllocator(allocator),
+      mDevice(device),
+      mTextureImage(image),
+      mTextureImageMemory(deviceMemory),
+      mVkFormat(format),
+      mViewType(viewType),
+      mFullViewRange{ fvkutils::getImageAspect(format), 0, levels, 0, layerCount },
+      mYcbcr{ ycbcrConversion, isExternalFormat },
+      mDefaultLayout(getDefaultLayoutImpl(usage)),
+      mUsage(usage),
+      mIsProtected(isProtected) {}
 
-// Constructor for internally passed VkImage
-VulkanTexture::VulkanTexture(VkDevice device, VmaAllocator allocator,
+VulkanTextureState::~VulkanTextureState() {
+    if (mTextureImageMemory != VK_NULL_HANDLE) {
+        vkDestroyImage(mDevice, mTextureImage, VKALLOC);
+        vkFreeMemory(mDevice, mTextureImageMemory, VKALLOC);
+    }
+    clearCachedImageViews();
+}
+
+void VulkanTextureState::clearCachedImageViews() noexcept {
+    for (auto entry: mCachedImageViews) {
+        vkDestroyImageView(mDevice, entry.second, VKALLOC);
+    }
+    mCachedImageViews.clear();
+}
+
+VkImageView VulkanTextureState::getImageView(VkImageSubresourceRange range, VkImageViewType viewType,
+        VkComponentMapping swizzle) {
+    ImageViewKey const key{ range, viewType, swizzle };
+    if (auto iter = mCachedImageViews.find(key); iter != mCachedImageViews.end()) {
+        return iter->second;
+    }
+
+    VkSamplerYcbcrConversionInfo conversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO,
+        .conversion = mYcbcr.conversion,
+    };
+
+    VkImageViewCreateInfo viewInfo = {
+        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+        .pNext = mYcbcr.conversion != VK_NULL_HANDLE ? &conversionInfo : nullptr,
+        .flags = 0,
+        .image = mTextureImage,
+        .viewType = viewType,
+        .format = mYcbcr.isExternalFormat ? VK_FORMAT_UNDEFINED : mVkFormat,
+        .components = swizzle,
+        .subresourceRange = range,
+    };
+    VkImageView imageView;
+    vkCreateImageView(mDevice, &viewInfo, VKALLOC, &imageView);
+    mCachedImageViews.emplace(key, imageView);
+    return imageView;
+}
+
+// Constructor for internally passed VkImage - including swapchain images and external images.
+VulkanTexture::VulkanTexture(VulkanContext const& context, VkDevice device, VmaAllocator allocator,
         fvkmemory::ResourceManager* resourceManager, VulkanCommands* commands, VkImage image,
-        VkDeviceMemory memory, VkFormat format, uint8_t samples, uint32_t width,
-        uint32_t height, uint32_t depth, TextureUsage tusage, VulkanStagePool& stagePool)
-    : HwTexture(getSamplerTypeFromDepth(depth), 1, samples, width, height, depth, TextureFormat::UNUSED,
-              tusage),
-      mState(fvkmemory::resource_ptr<VulkanTextureState>::construct(resourceManager, device,
-              allocator, commands, stagePool, format, fvkutils::getViewType(SamplerType::SAMPLER_2D),
-              /*mipLevels=*/1, getLayerCountFromDepth(depth), getDefaultLayoutImpl(tusage), any(usage & TextureUsage::PROTECTED))) {
-    mState->mTextureImage = image;
-    mState->mTextureImageMemory = memory;
+        VkDeviceMemory memory, VkFormat format, VkSamplerYcbcrConversion conversion,
+        uint8_t samples, uint32_t width, uint32_t height, uint32_t depth, TextureUsage tusage,
+        VulkanStagePool& stagePool)
+    : HwTexture(getSamplerTypeFromDepth(depth), 1, samples, width, height, depth,
+              TextureFormat::UNUSED, tusage),
+      mState(fvkmemory::resource_ptr<VulkanTextureState>::construct(resourceManager, stagePool,
+              commands, allocator, device, image, memory, format,
+              fvkutils::getViewType(SamplerType::SAMPLER_2D),
+              /*mipLevels=*/1, getLayerCountFromDepth(depth), conversion,
+              /*isExternalFormat=*/false,
+              getUsage(context, samples, VK_NULL_HANDLE, format, tusage),
+              any(usage & TextureUsage::PROTECTED))) {
     mPrimaryViewRange = mState->mFullViewRange;
 }
 
@@ -192,21 +306,19 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
         fvkmemory::ResourceManager* resourceManager, VulkanCommands* commands, SamplerType target,
         uint8_t levels, TextureFormat tformat, uint8_t samples, uint32_t w, uint32_t h,
         uint32_t depth, TextureUsage tusage, VulkanStagePool& stagePool)
-    : HwTexture(target, levels, samples, w, h, depth, tformat, tusage),
-      mState(fvkmemory::resource_ptr<VulkanTextureState>::construct(resourceManager, device,
-              allocator, commands, stagePool, fvkutils::getVkFormat(tformat),
-              fvkutils::getViewType(target), levels, getLayerCount(target, depth),
-              VulkanLayout::UNDEFINED, any(usage & TextureUsage::PROTECTED))) {
+     : HwTexture(target, levels, samples, w, h, depth, tformat, tusage) {
     // Create an appropriately-sized device-only VkImage, but do not fill it yet.
+    VkFormat const vkFormat = fvkutils::getVkFormat(tformat);
+    bool const isProtected = any(tusage & TextureUsage::PROTECTED);
     VkImageCreateInfo imageInfo{
         .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
         .imageType = target == SamplerType::SAMPLER_3D ? VK_IMAGE_TYPE_3D : VK_IMAGE_TYPE_2D,
-        .format = mState->mVkFormat,
+        .format = vkFormat,
         .extent = {w, h, depth},
         .mipLevels = levels,
         .arrayLayers = 1,
         .tiling = VK_IMAGE_TILING_OPTIMAL,
-        .usage = 0,
+        .usage = getUsage(context, samples, physicalDevice, vkFormat, tusage),
     };
     if (target == SamplerType::SAMPLER_3D && any(tusage & TextureUsage::ALL_ATTACHMENTS)) {
         if (context.isImageView2DOn3DImageSupported()) {
@@ -236,72 +348,17 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
         imageInfo.arrayLayers = depth * 6;
         imageInfo.extent.depth = 1;
     }
-    if (any(usage & TextureUsage::PROTECTED)) {
+    if (isProtected) {
         imageInfo.flags |= VK_IMAGE_CREATE_PROTECTED_BIT;
     }
 
-    if (any(usage & TextureUsage::BLIT_SRC)) {
-        imageInfo.usage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
-    }
-    if (any(usage & TextureUsage::BLIT_DST)) {
-        imageInfo.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT;
-    }
-
-    // Determine if we can use the transient usage flag combined with lazily allocated memory.
-    const bool useTransientAttachment =
-        // Lazily allocated memory is available.
-        context.isLazilyAllocatedMemorySupported() &&
-        // Usage consists of attachment flags only.
-        none(tusage & ~TextureUsage::ALL_ATTACHMENTS) &&
-        // Usage contains at least one attachment flag.
-        any(tusage & TextureUsage::ALL_ATTACHMENTS);
-    mState->mIsTransientAttachment = useTransientAttachment;
-
-    const VkImageUsageFlags transientFlag =
-       useTransientAttachment ? VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT : 0U;
-
-    if (any(usage & TextureUsage::SAMPLEABLE)) {
-
-#if FVK_ENABLED(FVK_DEBUG_TEXTURE)
-        // Validate that the format is actually sampleable.
-        VkFormatProperties props;
-        vkGetPhysicalDeviceFormatProperties(physicalDevice, mState->mVkFormat, &props);
-        if (!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT)) {
-            FVK_LOGW << "Texture usage is SAMPLEABLE but format " << mState->mVkFormat << " is not "
-                    "sampleable with optimal tiling." << utils::io::endl;
-        }
-#endif
-
-        imageInfo.usage |= VK_IMAGE_USAGE_SAMPLED_BIT;
-    }
-    if (any(usage & TextureUsage::COLOR_ATTACHMENT)) {
-        imageInfo.usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | transientFlag;
-        if (any(usage & TextureUsage::SUBPASS_INPUT)) {
-            imageInfo.usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT;
-        }
-    }
-    if (any(usage & TextureUsage::STENCIL_ATTACHMENT)) {
-        imageInfo.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | transientFlag;
-    }
-    if (any(usage & TextureUsage::UPLOADABLE)) {
-        imageInfo.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT;
-    }
-    if (any(usage & TextureUsage::DEPTH_ATTACHMENT)) {
-        imageInfo.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | transientFlag;
-
-        // Depth resolves uses a custom shader and therefore needs to be sampleable.
-        if (samples > 1) {
-            imageInfo.usage |= VK_IMAGE_USAGE_SAMPLED_BIT;
-        }
-    }
-
     // Constrain the sample count according to the sample count masks in VkPhysicalDeviceProperties.
     // Note that VulkanRenderTarget holds a single MSAA count, so we play it safe if this is used as
     // any kind of attachment (color or depth).
-    const auto& limits = context.getPhysicalDeviceLimits();
+    auto const& limits = context.getPhysicalDeviceLimits();
     if (imageInfo.usage & VK_IMAGE_USAGE_SAMPLED_BIT) {
         samples = fvkutils::reduceSampleCount(samples,
-                fvkutils::isVkDepthFormat(mState->mVkFormat)
+                fvkutils::isVkDepthFormat(vkFormat)
                         ? limits.sampledImageDepthSampleCounts
                         : limits.sampledImageColorSampleCounts);
     }
@@ -315,33 +372,36 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
     this->samples = samples;
     imageInfo.samples = (VkSampleCountFlagBits) samples;
 
-    VkResult result = vkCreateImage(mState->mDevice, &imageInfo, VKALLOC, &mState->mTextureImage);
+    VkImage textureImage;
+    VkResult result = vkCreateImage(device, &imageInfo, VKALLOC, &textureImage);
     if (result != VK_SUCCESS || FVK_ENABLED(FVK_DEBUG_TEXTURE)) {
         FVK_LOGD << "vkCreateImage: "
-            << "image = " << mState->mTextureImage << ", "
+            << "image = " << textureImage << ", "
             << "result = " << result << ", "
-            << "handle = " << utils::io::hex << mState->mTextureImage << utils::io::dec << ", "
+            << "handle = " << utils::io::hex << textureImage << utils::io::dec << ", "
             << "extent = " << w << "x" << h << "x"<< depth << ", "
             << "mipLevels = " << int(levels) << ", "
-            << "TextureUsage = " << static_cast<int>(usage) << ", "
+            << "TextureUsage = " << static_cast<int>(tusage) << ", "
             << "usage = " << imageInfo.usage << ", "
             << "samples = " << imageInfo.samples << ", "
             << "type = " << imageInfo.imageType << ", "
             << "flags = " << imageInfo.flags << ", "
             << "target = " << static_cast<int>(target) <<", "
-            << "format = " << mState->mVkFormat << utils::io::endl;
+            << "format = " << vkFormat << utils::io::endl;
     }
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "Unable to create image."
                                                        << " error=" << static_cast<int32_t>(result);
 
     // Allocate memory for the VkImage and bind it.
-    VkMemoryRequirements memReqs = {};
-    vkGetImageMemoryRequirements(mState->mDevice, mState->mTextureImage, &memReqs);
+    VkMemoryRequirements memReqs;
+    vkGetImageMemoryRequirements(device, textureImage, &memReqs);
 
-    const VkFlags requiredMemoryFlags =
+    bool const useTransientAttachment = imageInfo.usage & VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT;
+
+    VkFlags const requiredMemoryFlags =
         VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
         (useTransientAttachment ? VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT : 0U) |
-        (mState->mIsProtected ? VK_MEMORY_PROPERTY_PROTECTED_BIT : 0U);
+        (isProtected ? VK_MEMORY_PROPERTY_PROTECTED_BIT : 0U);
     uint32_t memoryTypeIndex
             = context.selectMemoryType(memReqs.memoryTypeBits, requiredMemoryFlags);
 
@@ -353,21 +413,22 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
         .allocationSize = memReqs.size,
         .memoryTypeIndex = memoryTypeIndex,
     };
-    result = vkAllocateMemory(mState->mDevice, &allocInfo, nullptr, &mState->mTextureImageMemory);
+    VkDeviceMemory textureImageMemory;
+    result = vkAllocateMemory(device, &allocInfo, nullptr, &textureImageMemory);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "Unable to allocate image memory."
                                                        << " error=" << static_cast<int32_t>(result);
-    result = vkBindImageMemory(mState->mDevice, mState->mTextureImage, mState->mTextureImageMemory,
-            0);
+    result = vkBindImageMemory(device, textureImage, textureImageMemory, 0);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "Unable to bind image."
                                                        << " error=" << static_cast<int32_t>(result);
 
+    mState = fvkmemory::resource_ptr<VulkanTextureState>::construct(resourceManager, stagePool,
+            commands, allocator, device, textureImage, textureImageMemory, vkFormat,
+            fvkutils::getViewType(target), levels, getLayerCount(target, depth),
+            VK_NULL_HANDLE /* ycbcrConversion */, false /*isExternalFormat*/, imageInfo.usage,
+            isProtected);
+
     // Spec out the "primary" VkImageView that shaders use to sample from the image.
     mPrimaryViewRange = mState->mFullViewRange;
-
-    // Go ahead and create the primary image view.
-    getImageView(mPrimaryViewRange, mState->mViewType, mSwizzle);
-
-    mState->mDefaultLayout = getDefaultLayoutImpl(imageInfo.usage);
 }
 
 // Constructor for creating a texture view
@@ -376,8 +437,8 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
         fvkmemory::resource_ptr<VulkanTexture> src, uint8_t baseLevel,
         uint8_t levelCount)
     : HwTexture(src->target, src->levels, src->samples, src->width, src->height, src->depth,
-            src->format, src->usage) {
-    mState = src->mState;
+            src->format, src->usage),
+      mState(src->mState) {
     mPrimaryViewRange = src->mPrimaryViewRange;
     mPrimaryViewRange.baseMipLevel = src->mPrimaryViewRange.baseMipLevel + baseLevel;
     mPrimaryViewRange.levelCount = levelCount;
@@ -388,21 +449,10 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
         VulkanContext const& context, VmaAllocator allocator, VulkanCommands* commands,
         fvkmemory::resource_ptr<VulkanTexture> src, VkComponentMapping swizzle)
     : HwTexture(src->target, src->levels, src->samples, src->width, src->height, src->depth,
-              src->format, src->usage) {
-    mState = src->mState;
-    mPrimaryViewRange = src->mPrimaryViewRange;
-    mSwizzle = composeSwizzle(src->mSwizzle, swizzle);
-}
-
-VulkanTextureState::~VulkanTextureState() {
-    if (mTextureImageMemory != VK_NULL_HANDLE) {
-        vkDestroyImage(mDevice, mTextureImage, VKALLOC);
-        vkFreeMemory(mDevice, mTextureImageMemory, VKALLOC);
-    }
-    for (auto entry: mCachedImageViews) {
-        vkDestroyImageView(mDevice, entry.second, VKALLOC);
-    }
-}
+              src->format, src->usage),
+      mState(src->mState),
+      mPrimaryViewRange(src->mPrimaryViewRange),
+      mSwizzle(composeSwizzle(src->mSwizzle, swizzle)) {}
 
 void VulkanTexture::updateImage(const PixelBufferDescriptor& data, uint32_t width, uint32_t height,
         uint32_t depth, uint32_t xoffset, uint32_t yoffset, uint32_t zoffset, uint32_t miplevel) {
@@ -509,20 +559,20 @@ void VulkanTexture::updateImageWithBlit(const PixelBufferDescriptor& hostData, u
     commands.acquire(fvkmemory::resource_ptr<VulkanTexture>::cast(this));
 
     // TODO: support blit-based format conversion for 3D images and cubemaps.
-    const int layer = 0;
+    constexpr int layer = 0;
 
-    const VkOffset3D rect[2] { {0, 0, 0}, {int32_t(width), int32_t(height), 1} };
+    VkOffset3D const rect[2] { {0, 0, 0}, {int32_t(width), int32_t(height), 1} };
 
-    const VkImageAspectFlags aspect = getImageAspect();
+    VkImageAspectFlags const aspect = getImageAspect();
 
-    const VkImageBlit blitRegions[1] = {{
+    VkImageBlit const blitRegions[1] = {{
         .srcSubresource = { aspect, 0, 0, 1 },
         .srcOffsets = { rect[0], rect[1] },
         .dstSubresource = { aspect, uint32_t(miplevel), layer, 1 },
         .dstOffsets = { rect[0], rect[1] }
     }};
 
-    const VkImageSubresourceRange range = { aspect, miplevel, 1, layer, 1 };
+    VkImageSubresourceRange const range = { aspect, miplevel, 1, layer, 1 };
 
     VulkanLayout const newLayout = VulkanLayout::TRANSFER_DST;
     VulkanLayout const oldLayout = getLayout(layer, miplevel);
@@ -538,41 +588,21 @@ VulkanLayout VulkanTexture::getDefaultLayout() const {
     return mState->mDefaultLayout;
 }
 
-VkImageView VulkanTexture::getAttachmentView(VkImageSubresourceRange range) {
-    range.levelCount = 1;
-    range.layerCount = 1;
-    return getImageView(range, VK_IMAGE_VIEW_TYPE_2D, {});
-}
-
-VkImageView VulkanTexture::getMultiviewAttachmentView(VkImageSubresourceRange range) {
-    return getImageView(range, VK_IMAGE_VIEW_TYPE_2D_ARRAY, {});
-}
-
-VkImageView VulkanTexture::getViewForType(VkImageSubresourceRange const& range, VkImageViewType type) {
-    return getImageView(range, type, mSwizzle);
-}
-
-VkImageView VulkanTexture::getImageView(VkImageSubresourceRange range, VkImageViewType viewType,
-        VkComponentMapping swizzle) {
-    VulkanTextureState::ImageViewKey const key{ range, viewType, swizzle };
-    auto iter = mState->mCachedImageViews.find(key);
-    if (iter != mState->mCachedImageViews.end()) {
-        return iter->second;
+VkImageView VulkanTexture::getAttachmentView(VkImageSubresourceRange const& range, VkImageViewType type) {
+    assert_invariant(mState->mYcbcr.conversion == VK_NULL_HANDLE &&
+                     "We are not yet supporting external image as attachments.");
+    if (type == VK_IMAGE_VIEW_TYPE_2D) {
+        VkImageSubresourceRange copy = range;
+        copy.levelCount = 1;
+        copy.layerCount = 1;
+        return mState->getImageView(copy, type, {});
+    } else {
+        return mState->getImageView(range, type, {});
     }
-    VkImageViewCreateInfo viewInfo = {
-        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .image = mState->mTextureImage,
-        .viewType = viewType,
-        .format = mState->mVkFormat,
-        .components = swizzle,
-        .subresourceRange = range,
-    };
-    VkImageView imageView;
-    vkCreateImageView(mState->mDevice, &viewInfo, VKALLOC, &imageView);
-    mState->mCachedImageViews.emplace(key, imageView);
-    return imageView;
+}
+
+VkImageView VulkanTexture::getView(VkImageSubresourceRange const& range) {
+    return mState->getImageView(range, mState->mViewType, mSwizzle);
 }
 
 VkImageAspectFlags VulkanTexture::getImageAspect() const {
@@ -734,6 +764,21 @@ void VulkanTexture::setLayout(VkImageSubresourceRange const& range, VulkanLayout
     }
 }
 
+void VulkanTexture::setYcbcrConversion(VkSamplerYcbcrConversion conversion, bool isExternalFormat) {
+    // Note that this comparison is valid because we only ever create VkSamplerYcbcrConversion from
+    // a cache.  So for each set of parameters, there is exactly one conversion (similar to
+    // samplers).
+    VulkanTextureState::Ycbcr ycbcr = {
+        .conversion = conversion,
+        .isExternalFormat = isExternalFormat,
+    };
+
+    if (mState->mYcbcr != ycbcr) {
+        mState->mYcbcr = ycbcr;
+        mState->clearCachedImageViews();
+    }
+}
+
 VulkanLayout VulkanTexture::getLayout(uint32_t layer, uint32_t level) const {
     assert_invariant(level <= 0xffff && layer <= 0xffff);
     const uint32_t key = (layer << 16) | level;

filament/backend/src/vulkan/VulkanTexture.h

@@ -19,7 +19,10 @@
 
 #include "DriverBase.h"
 
-#include "VulkanBuffer.h"
+#include "VulkanCommands.h"
+#include "VulkanConstants.h"
+#include "VulkanMemory.h"
+#include "VulkanStagePool.h"
 #include "vulkan/memory/Resource.h"
 #include "vulkan/memory/ResourcePointer.h"
 #include "vulkan/utils/Image.h"
@@ -31,10 +34,14 @@
 
 namespace filament::backend {
 
+struct VulkanTexture;
+
 struct VulkanTextureState : public fvkmemory::Resource {
-    VulkanTextureState(VkDevice device, VmaAllocator allocator, VulkanCommands* commands,
-            VulkanStagePool& stagePool, VkFormat format, VkImageViewType viewType, uint8_t levels,
-            uint8_t layerCount, VulkanLayout defaultLayout, bool isProtected);
+    VulkanTextureState(VulkanStagePool& stagePool, VulkanCommands* commands, VmaAllocator allocator,
+            VkDevice device, VkImage image, VkDeviceMemory deviceMemory, VkFormat format,
+            VkImageViewType viewType, uint8_t levels, uint8_t layerCount,
+            VkSamplerYcbcrConversion ycbcrConversion, bool isExternalFormat,
+            VkImageUsageFlags usage, bool isProtected);
 
     ~VulkanTextureState();
 
@@ -57,29 +64,50 @@ struct VulkanTextureState : public fvkmemory::Resource {
     // No implicit padding allowed due to it being a hash key.
     static_assert(sizeof(ImageViewKey) == 40);
 
-    using ImageViewHash = utils::hash::MurmurHashFn<ImageViewKey>;
+    VkImageView getImageView(VkImageSubresourceRange range, VkImageViewType viewType,
+            VkComponentMapping swizzle);
+private:
+    void clearCachedImageViews() noexcept;
+    VulkanStagePool& mStagePool;
+    VulkanCommands* const mCommands;
+    VmaAllocator const mAllocator;
+    VkDevice const mDevice;
 
     // The texture with the sidecar owns the sidecar.
     fvkmemory::resource_ptr<VulkanTexture> mSidecarMSAA;
-    VkDeviceMemory mTextureImageMemory = VK_NULL_HANDLE;
 
+    VkImage const mTextureImage;
+    VkDeviceMemory const mTextureImageMemory;
     VkFormat const mVkFormat;
     VkImageViewType const mViewType;
     VkImageSubresourceRange const mFullViewRange;
-    VkImage mTextureImage = VK_NULL_HANDLE;
-    VulkanLayout mDefaultLayout;
 
-    bool mIsProtected = false;
+    // Note that this parameter is not constant due to the fact that AHB can force a change in the
+    // conversion matrix per-frame.
+    struct Ycbcr {
+        VkSamplerYcbcrConversion conversion;
+        bool isExternalFormat;
+
+        bool operator==(Ycbcr const& other) const {
+            return conversion == other.conversion && isExternalFormat == other.isExternalFormat;
+        }
+
+        bool operator!=(Ycbcr const& other) const {
+            return !((*this) == other);
+        }
+
+    } mYcbcr;
+
+    VulkanLayout const mDefaultLayout;
+    VkImageUsageFlags const mUsage;
+    bool const mIsProtected;
 
     // Track the image layout of each subresource using a sparse range map.
     utils::RangeMap<uint32_t, VulkanLayout> mSubresourceLayouts;
-
+    using ImageViewHash = utils::hash::MurmurHashFn<ImageViewKey>;
     std::unordered_map<ImageViewKey, VkImageView, ImageViewHash> mCachedImageViews;
-    VulkanStagePool& mStagePool;
-    VkDevice mDevice;
-    VmaAllocator mAllocator;
-    VulkanCommands* mCommands;
-    bool mIsTransientAttachment;
+
+    friend struct VulkanTexture;
 };
 
 struct VulkanTexture : public HwTexture, fvkmemory::Resource {
@@ -91,10 +119,10 @@ struct VulkanTexture : public HwTexture, fvkmemory::Resource {
             VulkanStagePool& stagePool);
 
     // Specialized constructor for internally created textures (e.g. from a swap chain)
-    // The texture will never destroy the given VkImage, but it does manages its subresources.
-    VulkanTexture(VkDevice device, VmaAllocator allocator,
+    VulkanTexture(VulkanContext const& context, VkDevice device, VmaAllocator allocator,
             fvkmemory::ResourceManager* resourceManager, VulkanCommands* commands, VkImage image,
-            VkDeviceMemory memory, VkFormat format, uint8_t samples, uint32_t width, uint32_t height, uint32_t depth,
+            VkDeviceMemory memory, VkFormat format, VkSamplerYcbcrConversion conversion,
+            uint8_t samples, uint32_t width, uint32_t height, uint32_t depth,
             TextureUsage tusage, VulkanStagePool& stagePool);
 
     // Constructor for creating a texture view for wrt specific mip range
@@ -113,11 +141,6 @@ struct VulkanTexture : public HwTexture, fvkmemory::Resource {
     void updateImage(const PixelBufferDescriptor& data, uint32_t width, uint32_t height,
             uint32_t depth, uint32_t xoffset, uint32_t yoffset, uint32_t zoffset, uint32_t miplevel);
 
-    // Returns the primary image view, which is used for shader sampling.
-    VkImageView getPrimaryImageView() {
-        return getImageView(mPrimaryViewRange, mState->mViewType, mSwizzle);
-    }
-
     VkImageViewType getViewType() const {
         return mState->mViewType;
     }
@@ -132,20 +155,11 @@ struct VulkanTexture : public HwTexture, fvkmemory::Resource {
     // time of the call.
     VulkanLayout getDefaultLayout() const;
 
-    // Gets or creates a cached VkImageView for a single subresource that can be used as a render
-    // target attachment.  Unlike the primary image view, this always has type VK_IMAGE_VIEW_TYPE_2D
-    // and the identity swizzle.
-    VkImageView getAttachmentView(VkImageSubresourceRange range);
+    // Gets or creates a cached VkImageView for a subresource that can be used as a render
+    // target attachment.  Unlike the primary image view, this always the identity swizzle.
+    VkImageView getAttachmentView(VkImageSubresourceRange const& range, VkImageViewType type);
 
-    // Gets or creates a cached VkImageView for a single subresource that can be used as a render
-    // target attachment when rendering with multiview.
-    VkImageView getMultiviewAttachmentView(VkImageSubresourceRange range);
-
-    // This is a workaround for the first few frames where we're waiting for the texture to actually
-    // be uploaded.  In that case, we bind the sampler to an empty texture, but the corresponding
-    // imageView needs to be of the right type. Hence, we provide an option to indicate the
-    // view type. Swizzle option does not matter in this case.
-    VkImageView getViewForType(VkImageSubresourceRange const& range, VkImageViewType type);
+    VkImageView getView(VkImageSubresourceRange const& range);
 
     VkFormat getVkFormat() const {
         return mState->mVkFormat;
@@ -165,7 +179,7 @@ struct VulkanTexture : public HwTexture, fvkmemory::Resource {
     }
 
     bool isTransientAttachment() const {
-        return mState->mIsTransientAttachment;
+        return mState->mUsage & VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT;
     }
 
     bool getIsProtected() const {
@@ -192,6 +206,11 @@ struct VulkanTexture : public HwTexture, fvkmemory::Resource {
     // manually (outside of calls to transitionLayout).
     void setLayout(VkImageSubresourceRange const& range, VulkanLayout newLayout);
 
+    // This is used in the case of external images and external samplers. AHB might update the
+    // conversion per-frame. This implies that we need to invalidate the view cache when that
+    // happens.
+    void setYcbcrConversion(VkSamplerYcbcrConversion conversion, bool isExternal);
+
 #if FVK_ENABLED(FVK_DEBUG_TEXTURE)
     void print() const;
 #endif

filament/backend/src/vulkan/VulkanYcbcrConversionCache.cpp

@@ -0,0 +1,82 @@
+/*
+ * 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 "VulkanYcbcrConversionCache.h"
+
+#include "vulkan/VulkanConstants.h"
+#include "vulkan/utils/Conversion.h"
+
+#include <utils/Panic.h>
+
+#if defined(__ANDROID__)
+#include <vulkan/vulkan_android.h>  // for VkExternalFormatANDROID
+#endif
+
+using namespace bluevk;
+
+namespace filament::backend {
+
+VulkanYcbcrConversionCache::VulkanYcbcrConversionCache(VkDevice device)
+    : mDevice(device) {}
+
+VkSamplerYcbcrConversion VulkanYcbcrConversionCache::getConversion(
+        VulkanYcbcrConversionCache::Params params) noexcept {
+    auto iter = mCache.find(params);
+    if (UTILS_LIKELY(iter != mCache.end())) {
+        return iter->second;
+    }
+
+    auto const& chroma = params.conversion;
+    TextureSwizzle const swizzleArray[] = { chroma.r, chroma.g, chroma.b, chroma.a };
+    VkSamplerYcbcrConversionCreateInfo conversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_CREATE_INFO,
+        .format = VK_FORMAT_UNDEFINED,
+        .ycbcrModel = fvkutils::getYcbcrModelConversion(chroma.ycbcrModel),
+        .ycbcrRange = fvkutils::getYcbcrRange(chroma.ycbcrRange),
+        .components = fvkutils::getSwizzleMap(swizzleArray),
+        .xChromaOffset = fvkutils::getChromaLocation(chroma.xChromaOffset),
+        .yChromaOffset = fvkutils::getChromaLocation(chroma.yChromaOffset),
+        .chromaFilter = fvkutils::getFilter(chroma.chromaFilter),
+    };
+
+#if defined(__ANDROID__)
+    VkExternalFormatANDROID externalFormat = {
+        .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
+        .externalFormat = params.externalFormat,
+    };
+    if (params.externalFormat) {
+        conversionInfo.pNext = &externalFormat;
+    }
+#endif
+
+    VkSamplerYcbcrConversion conversion;
+    VkResult result =
+            vkCreateSamplerYcbcrConversion(mDevice, &conversionInfo, nullptr, &conversion);
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "Unable to create Ycbcr Conversion."
+                                                       << " error=" << static_cast<int32_t>(result);
+
+    mCache.insert({ params, conversion });
+    return conversion;
+}
+
+void VulkanYcbcrConversionCache::terminate() noexcept {
+    for (auto& [param, conv]: mCache) {
+        vkDestroySamplerYcbcrConversion(mDevice, conv, VKALLOC);
+    }
+    mCache.clear();
+}
+
+} // namespace filament::backend

filament/backend/src/vulkan/VulkanYcbcrConversionCache.h

@@ -0,0 +1,59 @@
+/*
+ * 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_VULKANYCBCRCONVERSIONCACHE_H
+#define TNT_FILAMENT_BACKEND_VULKANYCBCRCONVERSIONCACHE_H
+
+#include <backend/DriverEnums.h>
+
+#include <utils/Hash.h>
+
+#include <bluevk/BlueVK.h>
+#include <tsl/robin_map.h>
+
+namespace filament::backend {
+
+// Simple manager for VkSamplerYcbcrConversion objects.
+class VulkanYcbcrConversionCache {
+public:
+    struct Params {
+        SamplerYcbcrConversion conversion = {};
+        uint32_t padding = 0;
+        uint64_t externalFormat = 0;
+    };
+    static_assert(sizeof(Params) == 16);
+
+    explicit VulkanYcbcrConversionCache(VkDevice device);
+    VkSamplerYcbcrConversion getConversion(Params params) noexcept;
+    void terminate() noexcept;
+
+private:
+    VkDevice mDevice;
+
+    struct ConversionEqualTo {
+        bool operator()(Params lhs, Params rhs) const noexcept {
+            SamplerYcbcrConversion::EqualTo equal;
+            return equal(lhs.conversion, rhs.conversion) &&
+                   lhs.externalFormat == rhs.externalFormat;
+        }
+    };
+    using ConversionHashFn = utils::hash::MurmurHashFn<Params>;
+    tsl::robin_map<Params, VkSamplerYcbcrConversion, ConversionHashFn, ConversionEqualTo> mCache;
+};
+
+}// namespace filament::backend
+
+#endif// TNT_FILAMENT_BACKEND_VULKANYCBCRCONVERSIONCACHE_H

filament/backend/src/vulkan/caching/VulkanDescriptorSetManager.cpp

@@ -1,504 +0,0 @@
-/*
- * 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 "VulkanDescriptorSetManager.h"
-
-#include "vulkan/VulkanCommands.h"
-#include "vulkan/VulkanHandles.h"
-#include "vulkan/VulkanConstants.h"
-
-#include <utils/FixedCapacityVector.h>
-#include <utils/Panic.h>
-
-#include <math.h>
-
-#include <algorithm>
-#include <memory>
-#include <type_traits>
-#include <vector>
-
-namespace filament::backend {
-
-namespace {
-
-using BitmaskGroup = VulkanDescriptorSetLayout::Bitmask;
-using DescriptorCount = VulkanDescriptorSetLayout::Count;
-using DescriptorSetLayoutArray = VulkanDescriptorSetManager::DescriptorSetLayoutArray;
-using BitmaskGroupHashFn = utils::hash::MurmurHashFn<BitmaskGroup>;
-struct BitmaskGroupEqual {
-    bool operator()(BitmaskGroup const& k1, BitmaskGroup const& k2) const {
-        return k1 == k2;
-    }
-};
-
-// 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));
-    }
-};
-
-template<typename Bitmask>
-uint32_t createBindings(VkDescriptorSetLayoutBinding* toBind, uint32_t count, VkDescriptorType type,
-        Bitmask const& mask) {
-    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;
-}
-
-inline VkDescriptorSetLayout createLayout(VkDevice device, BitmaskGroup const& bitmaskGroup) {
-    // Note that the following *needs* to be static so that VkDescriptorSetLayoutCreateInfo will not
-    // refer to stack memory.
-    VkDescriptorSetLayoutBinding toBind[VulkanDescriptorSetLayout::MAX_BINDINGS];
-    uint32_t count = 0;
-
-    count = createBindings(toBind, count, VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC,
-            bitmaskGroup.dynamicUbo);
-    count = createBindings(toBind, count, VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, bitmaskGroup.ubo);
-    count = createBindings(toBind, count, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
-            bitmaskGroup.sampler);
-    count = createBindings(toBind, count, VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
-            bitmaskGroup.inputAttachment);
-
-    assert_invariant(count != 0 && "Need at least one binding for descriptor set layout.");
-    VkDescriptorSetLayoutCreateInfo dlinfo = {
-            .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
-            .pNext = nullptr,
-            .bindingCount = count,
-            .pBindings = toBind,
-    };
-
-    VkDescriptorSetLayout layout;
-    vkCreateDescriptorSetLayout(device, &dlinfo, VKALLOC, &layout);
-    return layout;
-}
-
-} // 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 VulkanDescriptorSetManager::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(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-        auto const vklayout = layout->getVkLayout();
-        DescriptorPool* sameTypePool = nullptr;
-        for (auto& pool: mPools) {
-            if (!pool->canAllocate(layout->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,
-                DescriptorCount::fromLayoutBitmask(layout->bitmask), 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;
-};
-
-class VulkanDescriptorSetManager::DescriptorSetLayoutManager {
-public:
-    DescriptorSetLayoutManager(VkDevice device)
-        : mDevice(device) {}
-
-    VkDescriptorSetLayout getVkLayout(VulkanDescriptorSetLayout::Bitmask const& bitmasks) {
-        if (auto itr = mVkLayouts.find(bitmasks); itr != mVkLayouts.end()) {
-            return itr->second;
-        }
-        auto vklayout = createLayout(mDevice, bitmasks);
-        mVkLayouts[bitmasks] = vklayout;
-        return vklayout;
-    }
-
-    ~DescriptorSetLayoutManager() {
-        for (auto& itr: mVkLayouts) {
-            vkDestroyDescriptorSetLayout(mDevice, itr.second, VKALLOC);
-        }
-    }
-
-private:
-    VkDevice mDevice;
-    tsl::robin_map<BitmaskGroup, VkDescriptorSetLayout, BitmaskGroupHashFn, BitmaskGroupEqual>
-            mVkLayouts;
-};
-
-VulkanDescriptorSetManager::VulkanDescriptorSetManager(VkDevice device,
-        fvkmemory::ResourceManager* resourceManager)
-    : mDevice(device),
-      mResourceManager(resourceManager),
-      mLayoutManager(std::make_unique<DescriptorSetLayoutManager>(device)),
-      mDescriptorPool(std::make_unique<DescriptorInfinitePool>(device)) {}
-
-VulkanDescriptorSetManager::~VulkanDescriptorSetManager() = default;
-
-void VulkanDescriptorSetManager::terminate() noexcept{
-    mLayoutManager.reset();
-    mDescriptorPool.reset();
-    clearHistory();
-}
-
-// bind() is not really binding the set but just stashing until we have all the info
-// (pipelinelayout).
-void VulkanDescriptorSetManager::bind(uint8_t setIndex,
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        backend::DescriptorSetOffsetArray&& offsets) {
-    set->setOffsets(std::move(offsets));
-    mStashedSets[setIndex] = set;
-}
-
-void VulkanDescriptorSetManager::unbind(uint8_t setIndex) {
-    mStashedSets[setIndex] = {};
-}
-
-void VulkanDescriptorSetManager::commit(VulkanCommandBuffer* commands,
-        VkPipelineLayout pipelineLayout, fvkutils::DescriptorSetMask const& setMask) {
-    // setMask indicates the set of descriptor sets the driver wants to bind, curMask is the
-    // actual set of sets that *needs* to be bound.
-    fvkutils::DescriptorSetMask curMask = setMask;
-
-    auto& updateSets = mStashedSets;
-    auto& lastBoundSets = mLastBoundInfo.boundSets;
-
-    setMask.forEachSetBit([&](size_t index) {
-        if (!updateSets[index] || updateSets[index] == lastBoundSets[index]) {
-            curMask.unset(index);
-        }
-    });
-
-    if (curMask.none() &&
-            (mLastBoundInfo.pipelineLayout == pipelineLayout && mLastBoundInfo.setMask == setMask &&
-                    mLastBoundInfo.boundSets == updateSets)) {
-        return;
-    }
-
-    curMask.forEachSetBit([&updateSets, commands, pipelineLayout](size_t index) {
-        // This code actually binds the descriptor sets.
-        auto set = updateSets[index];
-        VkCommandBuffer const cmdbuffer = commands->buffer();
-        vkCmdBindDescriptorSets(cmdbuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, index,
-                1, &set->vkSet, set->uniqueDynamicUboCount, set->getOffsets()->data());
-        commands->acquire(set);
-    });
-
-    mStashedSets = {};
-
-    mLastBoundInfo = {
-        pipelineLayout,
-        setMask,
-        updateSets,
-    };
-}
-
-void VulkanDescriptorSetManager::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 const descriptorWrite = {
-        .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
-        .pNext = nullptr,
-        .dstSet = set->vkSet,
-        .dstBinding = binding,
-        .descriptorCount = 1,
-        .descriptorType = type,
-        .pBufferInfo = &info,
-    };
-    vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
-    set->acquire(bufferObject);
-}
-
-void VulkanDescriptorSetManager::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
-        VkSampler sampler) noexcept {
-    VkDescriptorImageInfo info{
-        .sampler = sampler,
-    };
-    VkImageSubresourceRange const 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 sampler for this texture has the same view properties as a
-        // view for an attachment. Therefore, we can use getAttachmentView to get a
-        // corresponding VkImageView.
-        info.imageView = texture->getAttachmentView(range);
-    } else {
-        info.imageView = texture->getViewForType(range, expectedType);
-    }
-    info.imageLayout = fvkutils::getVkLayout(texture->getDefaultLayout());
-    VkWriteDescriptorSet const descriptorWrite = {
-        .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
-        .pNext = nullptr,
-        .dstSet = set->vkSet,
-        .dstBinding = binding,
-        .descriptorCount = 1,
-        .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
-        .pImageInfo = &info,
-    };
-    vkUpdateDescriptorSets(mDevice, 1, &descriptorWrite, 0, nullptr);
-    set->acquire(texture);
-}
-
-void VulkanDescriptorSetManager::updateInputAttachment(
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        VulkanAttachment const& attachment) noexcept {
-    // TOOD: fill-in this region
-}
-
-fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetManager::createSet(
-        Handle<HwDescriptorSet> handle, fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-    auto const vkSet = mDescriptorPool->obtainSet(layout);
-    auto const& count = layout->count;
-    auto const vklayout = layout->getVkLayout();
-    return fvkmemory::resource_ptr<VulkanDescriptorSet>::make(mResourceManager, handle, vkSet,
-            layout->bitmask.dynamicUbo, layout->count.dynamicUbo,
-            [vkSet, count, vklayout, this](VulkanDescriptorSet*) {
-                // Note that mDescriptorPool could be gone due to terminate (when the backend shuts
-                // down).
-                if (mDescriptorPool) {
-                    mDescriptorPool->recycle(count, vklayout, vkSet);
-                }
-            });
-}
-
-void VulkanDescriptorSetManager::initVkLayout(
-        fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-    layout->setVkLayout(mLayoutManager->getVkLayout(layout->bitmask));
-}
-
-void VulkanDescriptorSetManager::clearHistory() {
-    mStashedSets = {};
-}
-
-} // namespace filament::backend

filament/backend/src/vulkan/caching/VulkanDescriptorSetManager.h

@@ -1,102 +0,0 @@
-/*
- * 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_VULKANDESCRIPTORSETMANAGER_H
-#define TNT_FILAMENT_BACKEND_CACHING_VULKANDESCRIPTORSETMANAGER_H
-
-#include "vulkan/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 {
-
-// [GDSR]: Great-Descriptor-Set-Refactor: As of 03/20/24, the Filament frontend is planning to
-// introduce descriptor set. This PR will arrive before that change is complete. As such, some of
-// the methods introduced here will be obsolete, and certain logic will be generalized.
-
-// Abstraction over the pool and the layout cache.
-class VulkanDescriptorSetManager {
-public:
-    static constexpr uint8_t UNIQUE_DESCRIPTOR_SET_COUNT =
-            VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT;
-
-    using DescriptorSetLayoutArray = VulkanDescriptorSetLayout::DescriptorSetLayoutArray;
-
-    VulkanDescriptorSetManager(VkDevice device, fvkmemory::ResourceManager* resourceManager);
-    ~VulkanDescriptorSetManager();
-
-    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 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& setMask);
-
-    fvkmemory::resource_ptr<VulkanDescriptorSet> createSet(Handle<HwDescriptorSet> handle,
-            fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
-
-    void initVkLayout(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
-
-    void clearHistory();
-
-private:
-    class DescriptorSetLayoutManager;
-    class DescriptorInfinitePool;
-
-    using DescriptorSetArray =
-            std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>, UNIQUE_DESCRIPTOR_SET_COUNT>;
-
-    VkDevice mDevice;
-    fvkmemory::ResourceManager* mResourceManager;
-    std::unique_ptr<DescriptorSetLayoutManager> mLayoutManager;
-    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_VULKANDESCRIPTORSETMANAGER_H

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

@@ -736,10 +736,6 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
     chainStruct(&context.mPhysicalDeviceFeatures, &queryProtectedMemoryFeatures);
     chainStruct(&context.mPhysicalDeviceProperties, &protectedMemoryProperties);
 
-    // We know we need to allocate the protected version of the VK objects
-    context.mProtectedMemorySupported =
-            static_cast<bool>(queryProtectedMemoryFeatures.protectedMemory);
-
     // Initialize the following fields: physicalDeviceProperties, memoryProperties,
     // physicalDeviceFeatures, graphicsQueueFamilyIndex.
     vkGetPhysicalDeviceProperties2(mImpl->mPhysicalDevice, &context.mPhysicalDeviceProperties);
@@ -759,6 +755,10 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
         mImpl->mGraphicsQueueIndex = 0;
     }
 
+    // We know we need to allocate the protected version of the VK objects
+    context.mProtectedMemorySupported =
+            static_cast<bool>(queryProtectedMemoryFeatures.protectedMemory);
+
     if (context.mProtectedMemorySupported) {
         mImpl->mProtectedGraphicsQueueFamilyIndex = identifyGraphicsQueueFamilyIndex(
                 mImpl->mPhysicalDevice, (VK_QUEUE_GRAPHICS_BIT | VK_QUEUE_PROTECTED_BIT));
@@ -978,15 +978,37 @@ VkQueue VulkanPlatform::getProtectedGraphicsQueue() const noexcept {
 }
 
 VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadata(
-        void* externalImage) {
+        ExternalImageHandleRef externalImage) {
     return getExternalImageMetadataImpl(externalImage, mImpl->mDevice);
 }
 
-VulkanPlatform::ImageData VulkanPlatform::createExternalImage(void* externalImage,
-            const ExternalImageMetadata& metadata) {
-    return createExternalImageImpl(externalImage, mImpl->mDevice, nullptr, metadata);
+VulkanPlatform::ImageData VulkanPlatform::createExternalImageData(
+        ExternalImageHandleRef externalImage, const ExternalImageMetadata& metadata,
+        uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
+    return createExternalImageDataImpl(externalImage, mImpl->mDevice, metadata, memoryTypeIndex,
+            usage);
 }
 
+VkSampler VulkanPlatform::createExternalSampler(SamplerYcbcrConversion chroma,
+        SamplerParams sampler, uint32_t internalFormat) {
+    return createExternalSamplerImpl(mImpl->mDevice, chroma, sampler, internalFormat);
+}
+
+VkImageView VulkanPlatform::createExternalImageView(SamplerYcbcrConversion chroma,
+        uint32_t internalFormat, VkImage image, VkImageSubresourceRange range,
+        VkImageViewType viewType, VkComponentMapping swizzle) {
+    return createExternalImageViewImpl(mImpl->mDevice, chroma, internalFormat, image, range,
+            viewType, swizzle);
+}
+
+ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() const {
+    return getSwapchainInstanceExtensionsImpl();
+}
+
+VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHR(void* nativeWindow,
+        VkInstance instance, uint64_t flags) const noexcept {
+    return createVkSurfaceKHRImpl(nativeWindow, instance, flags);
+}
 #undef SWAPCHAIN_RET_FUNC
 
 }// namespace filament::backend

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

@@ -13,28 +13,60 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#include <backend/platforms/VulkanPlatform.h>
+#include <backend/platforms/VulkanPlatformAndroid.h>
+
+#include <backend/DriverEnums.h>
+#include <private/backend/BackendUtilsAndroid.h>
 
 #include "vulkan/VulkanConstants.h"
 
 #include <utils/Panic.h>
+#include "vulkan/utils/Image.h"
+#include "vulkan/utils/Conversion.h"
 
 #include <bluevk/BlueVK.h>
 
 #include <android/hardware_buffer.h>
 #include <android/native_window.h>
 
+#include <utility>
+
 using namespace bluevk;
 
 namespace filament::backend {
 
 namespace {
-void getVKFormatAndUsage(const AHardwareBuffer_Desc& desc, VkFormat& format,
-        VkImageUsageFlags& usage, bool& isProtected) {
+
+VkFormat transformVkFormat(VkFormat format, bool sRGB) {
+    if (!sRGB) {
+        return format;
+    }
+
+    switch (format) {
+        case VK_FORMAT_R8G8B8A8_UNORM:
+            format = VK_FORMAT_R8G8B8A8_SRGB;
+            break;
+        case VK_FORMAT_R8G8B8_UNORM:
+            format = VK_FORMAT_R8G8B8_SRGB;
+            break;
+        default:
+            break;
+    }
+
+    return format;
+}
+
+bool isProtectedFromUsage(uint64_t usage) {
+    return (usage & AHARDWAREBUFFER_USAGE_PROTECTED_CONTENT) ? true : false;
+}
+
+std::pair<VkFormat, VkImageUsageFlags> getVKFormatAndUsage(const AHardwareBuffer_Desc& desc,
+        bool sRGB) {
+    VkFormat format = VK_FORMAT_UNDEFINED;
+    VkImageUsageFlags usage = 0;
     // Refer to "11.2.17. External Memory Handle Types" in the spec, and
     // Tables 13/14 for how the following derivation works.
     bool isDepthFormat = false;
-    isProtected = false;
     switch (desc.format) {
         case AHARDWAREBUFFER_FORMAT_R8G8B8A8_UNORM:
             format = VK_FORMAT_R8G8B8A8_UNORM;
@@ -82,6 +114,8 @@ void getVKFormatAndUsage(const AHardwareBuffer_Desc& desc, VkFormat& format,
             format = VK_FORMAT_UNDEFINED;
     }
 
+    format = transformVkFormat(format, sRGB);
+
     // The following only concern usage flags derived from Table 14.
     usage = 0;
     if (desc.usage & AHARDWAREBUFFER_USAGE_GPU_SAMPLED_IMAGE) {
@@ -98,16 +132,14 @@ void getVKFormatAndUsage(const AHardwareBuffer_Desc& desc, VkFormat& format,
     if (desc.usage & AHARDWAREBUFFER_USAGE_GPU_DATA_BUFFER) {
         usage = VK_IMAGE_USAGE_STORAGE_BIT;
     }
-    if (desc.usage & AHARDWAREBUFFER_USAGE_PROTECTED_CONTENT) {
-        isProtected = true;
-    }
+
+    return { format, usage };
 }
 
-VulkanPlatform::ImageData allocateExternalImage(void* externalBuffer,
-        VkDevice device, const VkAllocationCallbacks* allocator,
-        VulkanPlatform::ExternalImageMetadata const& metadata) {
+VulkanPlatform::ImageData allocateExternalImage(AHardwareBuffer* buffer, VkDevice device,
+        VulkanPlatform::ExternalImageMetadata const& metadata, uint32_t memoryTypeIndex,
+        VkImageUsageFlags usage) {
     VulkanPlatform::ImageData data;
-    AHardwareBuffer* buffer = static_cast<AHardwareBuffer*>(externalBuffer);
 
     // if external format we need to specifiy it in the allocation
     const bool useExternalFormat = metadata.format == VK_FORMAT_UNDEFINED;
@@ -115,8 +147,8 @@ VulkanPlatform::ImageData allocateExternalImage(void* externalBuffer,
     const VkExternalFormatANDROID externalFormat = {
         .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
         .pNext = nullptr,
-        .externalFormat = metadata
-                .externalFormat,// pass down the format (external means we don't have it VK defined)
+        // pass down the format (external means we don't have it VK defined)
+        .externalFormat = metadata.externalFormat,
     };
     const VkExternalMemoryImageCreateInfo externalCreateInfo = {
         .sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO,
@@ -124,7 +156,7 @@ VulkanPlatform::ImageData allocateExternalImage(void* externalBuffer,
         .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID,
     };
 
-    VkImageCreateInfo imageInfo{.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO};
+    VkImageCreateInfo imageInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO };
     imageInfo.pNext = &externalCreateInfo;
     imageInfo.format = metadata.format;
     imageInfo.imageType = VK_IMAGE_TYPE_2D;
@@ -135,12 +167,10 @@ VulkanPlatform::ImageData allocateExternalImage(void* externalBuffer,
     };
     imageInfo.mipLevels = 1;
     imageInfo.arrayLayers = metadata.layers;
-    imageInfo.usage = metadata.usage;
-    // In the unprotected case add R/W capabilities
-    if (metadata.isProtected == false)
-        imageInfo.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
+    imageInfo.samples = metadata.samples;
+    imageInfo.usage = usage;
 
-    VkResult result = vkCreateImage(device, &imageInfo, allocator, &data.first);
+    VkResult result = vkCreateImage(device, &imageInfo, VKALLOC, &data.first);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
             << "vkCreateImage failed with error=" << static_cast<int32_t>(result);
 
@@ -156,11 +186,13 @@ VulkanPlatform::ImageData allocateExternalImage(void* externalBuffer,
         .image = data.first,
         .buffer = VK_NULL_HANDLE,
     };
-    VkMemoryAllocateInfo allocInfo = {.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
+    VkMemoryAllocateInfo allocInfo = {
+        .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
         .pNext = &memoryDedicatedAllocateInfo,
         .allocationSize = metadata.allocationSize,
-        .memoryTypeIndex = metadata.memoryTypeBits};
-    result = vkAllocateMemory(device, &allocInfo, allocator, &data.second);
+        .memoryTypeIndex = memoryTypeIndex,
+    };
+    result = vkAllocateMemory(device, &allocInfo, VKALLOC, &data.second);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
             << "vkAllocateMemory failed with error=" << static_cast<int32_t>(result);
 
@@ -169,23 +201,60 @@ VulkanPlatform::ImageData allocateExternalImage(void* externalBuffer,
 
 }// namespace
 
-VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadataImpl(void* externalImage,
-        VkDevice device) {
+VulkanPlatformAndroid::ExternalImageVulkanAndroid::~ExternalImageVulkanAndroid() = default;
+
+Platform::ExternalImageHandle VulkanPlatformAndroid::createExternalImage(
+        AHardwareBuffer const* buffer, bool sRGB) noexcept {
+    if (__builtin_available(android 26, *)) {
+        AHardwareBuffer_Desc hardwareBufferDescription = {};
+        AHardwareBuffer_describe(buffer, &hardwareBufferDescription);
+
+        auto* const p = new (std::nothrow) ExternalImageVulkanAndroid;
+        p->aHardwareBuffer = const_cast<AHardwareBuffer*>(buffer);
+        p->sRGB = sRGB;
+        p->height = hardwareBufferDescription.height;
+        p->width = hardwareBufferDescription.width;
+        TextureFormat textureFormat = mapToFilamentFormat(hardwareBufferDescription.format, sRGB);
+        p->format = textureFormat;
+        p->usage = mapToFilamentUsage(hardwareBufferDescription.usage, textureFormat);
+        return Platform::ExternalImageHandle{ p };
+    }
+
+    return Platform::ExternalImageHandle{};
+}
+
+VulkanPlatformAndroid::ExternalImageDescAndroid VulkanPlatformAndroid::getExternalImageDesc(
+        ExternalImageHandleRef externalImage) const noexcept {
+    auto const* fvkExternalImage =
+            static_cast<ExternalImageVulkanAndroid const*>(externalImage.get());
+
+    return {
+        .width = fvkExternalImage->width,
+        .height = fvkExternalImage->height,
+        .format = fvkExternalImage->format,
+        .usage = fvkExternalImage->usage,
+    };
+}
+
+VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::getExternalImageMetadata(
+        ExternalImageHandleRef externalImage) {
+    auto const* fvkExternalImage =
+            static_cast<ExternalImageVulkanAndroid const*>(externalImage.get());
+
     ExternalImageMetadata metadata;
-    AHardwareBuffer* buffer = static_cast<AHardwareBuffer*>(externalImage);
+    AHardwareBuffer* buffer = fvkExternalImage->aHardwareBuffer;
     if (__builtin_available(android 26, *)) {
         AHardwareBuffer_Desc bufferDesc;
         AHardwareBuffer_describe(buffer, &bufferDesc);
         metadata.width = bufferDesc.width;
         metadata.height = bufferDesc.height;
         metadata.layers = bufferDesc.layers;
+        metadata.isProtected = isProtectedFromUsage(bufferDesc.usage);
+        std::tie(metadata.format, metadata.usage) =
+                getVKFormatAndUsage(bufferDesc, fvkExternalImage->sRGB);
+    }
 
-        getVKFormatAndUsage(bufferDesc, metadata.format, metadata.usage, metadata.isProtected);
-    }
-    // In the unprotected case add R/W capabilities
-    if (metadata.isProtected == false) {
-        metadata.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
-    }
+    metadata.samples = VK_SAMPLE_COUNT_1_BIT;
 
     VkAndroidHardwareBufferFormatPropertiesANDROID formatInfo = {
         .sType = VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_FORMAT_PROPERTIES_ANDROID,
@@ -195,36 +264,152 @@ VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadataIm
         .sType = VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_PROPERTIES_ANDROID,
         .pNext = &formatInfo,
     };
-    VkResult result = vkGetAndroidHardwareBufferPropertiesANDROID(device, buffer, &properties);
+    VkResult result = vkGetAndroidHardwareBufferPropertiesANDROID(getDevice(), buffer, &properties);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
             << "vkGetAndroidHardwareBufferProperties failed with error="
             << static_cast<int32_t>(result);
-    FILAMENT_CHECK_POSTCONDITION(metadata.format == formatInfo.format)
-            << "mismatched image format for external image (AHB)";
+
+    VkFormat bufferPropertiesFormat = transformVkFormat(formatInfo.format, fvkExternalImage->sRGB);
+    FILAMENT_CHECK_POSTCONDITION(metadata.format == bufferPropertiesFormat)
+            << "mismatched image format( " << metadata.format << ") and queried format("
+            << bufferPropertiesFormat << ") for external image (AHB)";
     metadata.externalFormat = formatInfo.externalFormat;
     metadata.allocationSize = properties.allocationSize;
     metadata.memoryTypeBits = properties.memoryTypeBits;
     return metadata;
 }
 
-VulkanPlatform::ImageData VulkanPlatform::createExternalImageImpl(void* externalImage,
-        VkDevice device, const VkAllocationCallbacks* allocator,
-        const ExternalImageMetadata& metadata) {
-    ImageData data = allocateExternalImage(externalImage, device, allocator, metadata);
-    VkResult result = vkBindImageMemory(device, data.first, data.second, 0);
+VulkanPlatformAndroid::ImageData VulkanPlatformAndroid::createExternalImageData(
+        ExternalImageHandleRef externalImage, const ExternalImageMetadata& metadata,
+        uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
+    auto const* fvkExternalImage =
+            static_cast<ExternalImageVulkanAndroid const*>(externalImage.get());
+    ImageData data = allocateExternalImage(fvkExternalImage->aHardwareBuffer, getDevice(), metadata,
+            memoryTypeIndex, usage);
+    VkResult result = vkBindImageMemory(getDevice(), data.first, data.second, 0);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-            << "vkBindImageMemory error=" << static_cast<int32_t>(result);
+        << "vkBindImageMemory error=" << static_cast<int32_t>(result);
     return data;
 }
 
-VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() {
+VkImageView VulkanPlatform::createExternalImageViewImpl(VkDevice device, SamplerYcbcrConversion chroma,
+            uint32_t internalFormat, VkImage image, VkImageSubresourceRange range,
+            VkImageViewType viewType, VkComponentMapping swizzle){
+    VkExternalFormatANDROID externalFormat = {
+        .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
+        .externalFormat = internalFormat,
+    };
+
+    TextureSwizzle const swizzleArray[] = {chroma.r, chroma.g, chroma.b, chroma.a};
+    VkSamplerYcbcrConversionCreateInfo conversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_CREATE_INFO,
+        .pNext = &externalFormat,
+        .format = VK_FORMAT_UNDEFINED,
+        .ycbcrModel = fvkutils::getYcbcrModelConversion(chroma.ycbcrModel),
+        .ycbcrRange = fvkutils::getYcbcrRange(chroma.ycbcrRange),
+        .components = fvkutils::getSwizzleMap(swizzleArray),
+        .xChromaOffset = fvkutils::getChromaLocation(chroma.xChromaOffset),
+        .yChromaOffset = fvkutils::getChromaLocation(chroma.yChromaOffset),
+        .chromaFilter = fvkutils::getFilter(chroma.chromaFilter),
+    };
+    VkSamplerYcbcrConversion conversion = VK_NULL_HANDLE;
+    VkResult result = vkCreateSamplerYcbcrConversion(device, &conversionInfo,
+                                                     nullptr, &conversion);
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+        << "Unable to create Ycbcr Conversion."
+        << " error=" << static_cast<int32_t>(result);
+
+    VkSamplerYcbcrConversionInfo samplerYcbcrConversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO,
+        .conversion = conversion,
+    };
+
+    VkImageViewCreateInfo viewInfo = {
+        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+        .pNext = &samplerYcbcrConversionInfo,
+        .flags = 0,
+        .image = image,
+        .viewType = viewType,
+        .format = VK_FORMAT_UNDEFINED,
+        .components = swizzle,
+        .subresourceRange = range,
+    };
+    VkImageView imageView;
+    result = vkCreateImageView(device, &viewInfo, VKALLOC, &imageView);
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+        << "Unable to create VkImageView."
+        << " error=" << static_cast<int32_t>(result);
+
+    return imageView;
+}
+
+VkSampler VulkanPlatform::createExternalSamplerImpl(
+        VkDevice device, SamplerYcbcrConversion chroma, SamplerParams params,
+        uint32_t internalFormat) {
+    VkExternalFormatANDROID externalFormat = {
+        .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
+        .externalFormat = internalFormat,
+    };
+
+    TextureSwizzle const swizzleArray[] = {chroma.r, chroma.g, chroma.b, chroma.a};
+    VkSamplerYcbcrConversionCreateInfo conversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_CREATE_INFO,
+        .pNext = &externalFormat,
+        .format = VK_FORMAT_UNDEFINED,
+        .ycbcrModel = fvkutils::getYcbcrModelConversion(chroma.ycbcrModel),
+        .ycbcrRange = fvkutils::getYcbcrRange(chroma.ycbcrRange),
+        .components = fvkutils::getSwizzleMap(swizzleArray),
+        .xChromaOffset = fvkutils::getChromaLocation(chroma.xChromaOffset),
+        .yChromaOffset = fvkutils::getChromaLocation(chroma.yChromaOffset),
+        .chromaFilter = fvkutils::getFilter(chroma.chromaFilter),
+    };
+    VkSamplerYcbcrConversion conversion = VK_NULL_HANDLE;
+    VkResult result = vkCreateSamplerYcbcrConversion(device, &conversionInfo,
+                                                     nullptr, &conversion);
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+        << "Unable to create Ycbcr Conversion."
+        << " error=" << static_cast<int32_t>(result);
+
+    VkSamplerYcbcrConversionInfo samplerYcbcrConversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO,
+        .pNext = nullptr,
+        .conversion = conversion,
+    };
+
+    VkSamplerCreateInfo samplerInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
+        .pNext = &samplerYcbcrConversionInfo,
+        .magFilter = fvkutils::getFilter(params.filterMag),
+        .minFilter = fvkutils::getFilter(params.filterMin),
+        .mipmapMode = fvkutils::getMipmapMode(params.filterMin),
+        .addressModeU = fvkutils::getWrapMode(params.wrapS),
+        .addressModeV = fvkutils::getWrapMode(params.wrapT),
+        .addressModeW = fvkutils::getWrapMode(params.wrapR),
+        .anisotropyEnable = params.anisotropyLog2 == 0 ? VK_FALSE : VK_TRUE,
+        .maxAnisotropy = (float)(1u << params.anisotropyLog2),
+        .compareEnable = fvkutils::getCompareEnable(params.compareMode),
+        .compareOp = fvkutils::getCompareOp(params.compareFunc),
+        .minLod = 0.0f,
+        .maxLod = fvkutils::getMaxLod(params.filterMin),
+        .borderColor = VK_BORDER_COLOR_INT_OPAQUE_BLACK,
+        .unnormalizedCoordinates = VK_FALSE,
+    };
+    VkSampler sampler;
+    result = vkCreateSampler(device, &samplerInfo, VKALLOC, &sampler);
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+        << "Unable to create sampler."
+        << " error=" << static_cast<int32_t>(result);
+    return sampler;
+}
+
+VulkanPlatform::ExtensionSet VulkanPlatformAndroid::getSwapchainInstanceExtensions() const {
     return {
         VK_KHR_ANDROID_SURFACE_EXTENSION_NAME,
     };
 }
 
-VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHR(void* nativeWindow,
-    VkInstance instance, uint64_t flags) noexcept {
+VulkanPlatform::SurfaceBundle VulkanPlatformAndroid::createVkSurfaceKHR(void* nativeWindow,
+        VkInstance instance, uint64_t flags) const noexcept {
     VkSurfaceKHR surface;
     VkExtent2D extent;
 
@@ -236,6 +421,26 @@ VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHR(void* nativeWin
             vkCreateAndroidSurfaceKHR(instance, &createInfo, VKALLOC, (VkSurfaceKHR*) &surface);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
             << "vkCreateAndroidSurfaceKHR with error=" << static_cast<int32_t>(result);
-    return {surface, extent};
+    return { surface, extent };
 }
+
+// Deprecated platform dependent helper methods
+VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl() { return {}; }
+
+VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device) {
+    return ExternalImageMetadata{};
 }
+
+VulkanPlatform::ImageData VulkanPlatform::createExternalImageDataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device,
+        const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
+    return ImageData{};
+}
+
+VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHRImpl(void* nativeWindow,
+        VkInstance instance, uint64_t flags) noexcept {
+    return SurfaceBundle{};
+}
+
+}// namespace filament::backend

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

@@ -52,7 +52,7 @@ using namespace bluevk;
 
 namespace filament::backend {
 
-VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() {
+VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl() {
     ExtensionSet const ret = {
     #if defined(__APPLE__)
         VK_MVK_MACOS_SURFACE_EXTENSION_NAME,  // TODO: replace with VK_EXT_metal_surface
@@ -64,17 +64,30 @@ VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() {
 }
 
 VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadataImpl(
-        void* externalImage, VkDevice device) {
+        ExternalImageHandleRef externalImage, VkDevice device) {
     return {};
 }
 
-VulkanPlatform::ImageData VulkanPlatform::createExternalImageImpl(void* externalImage,
-        VkDevice device, const VkAllocationCallbacks* allocator,
-        const ExternalImageMetadata& metadata) {
+VulkanPlatform::ImageData VulkanPlatform::createExternalImageDataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device,
+        const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
     return {};
 }
 
-VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHR(void* nativeWindow,
+VkSampler VulkanPlatform::createExternalSamplerImpl(VkDevice device,
+    SamplerYcbcrConversion chroma,
+    SamplerParams sampler,
+    uint32_t internalFormat) {
+    return VK_NULL_HANDLE;
+}
+
+VkImageView VulkanPlatform::createExternalImageViewImpl(VkDevice device,
+        SamplerYcbcrConversion chroma, uint32_t internalFormat, VkImage image,
+        VkImageSubresourceRange range, VkImageViewType viewType, VkComponentMapping swizzle) {
+    return VK_NULL_HANDLE;
+}
+
+VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHRImpl(void* nativeWindow,
         VkInstance instance, uint64_t flags) noexcept {
     VkSurfaceKHR surface;
     #if defined(__APPLE__)

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

@@ -85,17 +85,30 @@ using namespace bluevk;
 namespace filament::backend {
 
 VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadataImpl(
-        void* externalImage, VkDevice device) {
+        ExternalImageHandleRef externalImage, VkDevice device) {
     return {};
 }
 
-VulkanPlatform::ImageData VulkanPlatform::createExternalImageImpl(void* externalImage,
-        VkDevice device, const VkAllocationCallbacks* allocator,
-        const ExternalImageMetadata& metadata) {
+VulkanPlatform::ImageData VulkanPlatform::createExternalImageDataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device,
+        const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
     return {};
 }
 
-VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() {
+VkSampler VulkanPlatform::createExternalSamplerImpl(VkDevice device,
+    SamplerYcbcrConversion chroma,
+    SamplerParams sampler,
+    uint32_t internalFormat) {
+    return VK_NULL_HANDLE;
+}
+
+VkImageView VulkanPlatform::createExternalImageViewImpl(VkDevice device,
+        SamplerYcbcrConversion chroma, uint32_t internalFormat, VkImage image,
+        VkImageSubresourceRange range, VkImageViewType viewType, VkComponentMapping swizzle) {
+    return VK_NULL_HANDLE;
+}
+
+VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl() {
     VulkanPlatform::ExtensionSet const ret = {
 #if defined(__linux__) && defined(FILAMENT_SUPPORTS_WAYLAND)
         VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME,
@@ -113,7 +126,7 @@ VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() {
     return ret;
 }
 
-VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHR(void* nativeWindow,
+VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHRImpl(void* nativeWindow,
         VkInstance instance, uint64_t flags) noexcept {
     VkSurfaceKHR surface;
 

filament/backend/src/vulkan/platform/VulkanPlatformSwapChainImpl.h

@@ -18,12 +18,12 @@
 #define TNT_FILAMENT_BACKEND_VULKANSWAPCHAIN_IMPL_H
 
 #include "vulkan/VulkanContext.h"
+#include "vulkan/VulkanConstants.h"
 
 #include <backend/platforms/VulkanPlatform.h>
 
 #include <bluevk/BlueVK.h>
 
-#include <tuple>
 #include <unordered_map>
 
 using namespace bluevk;

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

@@ -588,6 +588,79 @@ VkComponentMapping getSwizzleMap(TextureSwizzle const swizzle[4]) {
     return map;
 }
 
+VkFilter getFilter(SamplerMinFilter filter) {
+    switch (filter) {
+    case SamplerMinFilter::NEAREST:
+        return VK_FILTER_NEAREST;
+    case SamplerMinFilter::LINEAR:
+        return VK_FILTER_LINEAR;
+    case SamplerMinFilter::NEAREST_MIPMAP_NEAREST:
+        return VK_FILTER_NEAREST;
+    case SamplerMinFilter::LINEAR_MIPMAP_NEAREST:
+        return VK_FILTER_LINEAR;
+    case SamplerMinFilter::NEAREST_MIPMAP_LINEAR:
+        return VK_FILTER_NEAREST;
+    case SamplerMinFilter::LINEAR_MIPMAP_LINEAR:
+        return VK_FILTER_LINEAR;
+    }
+}
+
+VkFilter getFilter(SamplerMagFilter filter) {
+    switch (filter) {
+    case SamplerMagFilter::NEAREST:
+        return VK_FILTER_NEAREST;
+    case SamplerMagFilter::LINEAR:
+        return VK_FILTER_LINEAR;
+    }
+}
+
+VkSamplerMipmapMode getMipmapMode(SamplerMinFilter filter) {
+    switch (filter) {
+    case SamplerMinFilter::NEAREST:
+        return VK_SAMPLER_MIPMAP_MODE_NEAREST;
+    case SamplerMinFilter::LINEAR:
+        return VK_SAMPLER_MIPMAP_MODE_NEAREST;
+    case SamplerMinFilter::NEAREST_MIPMAP_NEAREST:
+        return VK_SAMPLER_MIPMAP_MODE_NEAREST;
+    case SamplerMinFilter::LINEAR_MIPMAP_NEAREST:
+        return VK_SAMPLER_MIPMAP_MODE_NEAREST;
+    case SamplerMinFilter::NEAREST_MIPMAP_LINEAR:
+        return VK_SAMPLER_MIPMAP_MODE_LINEAR;
+    case SamplerMinFilter::LINEAR_MIPMAP_LINEAR:
+        return VK_SAMPLER_MIPMAP_MODE_LINEAR;
+    }
+}
+
+VkSamplerAddressMode getWrapMode(SamplerWrapMode mode) {
+    switch (mode) {
+    case SamplerWrapMode::REPEAT:
+        return VK_SAMPLER_ADDRESS_MODE_REPEAT;
+    case SamplerWrapMode::CLAMP_TO_EDGE:
+        return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
+    case SamplerWrapMode::MIRRORED_REPEAT:
+        return VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT;
+    }
+}
+
+VkBool32 getCompareEnable(SamplerCompareMode mode) {
+    return mode == SamplerCompareMode::NONE ? VK_FALSE : VK_TRUE;
+}
+
+float getMaxLod(SamplerMinFilter filter) {
+    switch (filter) {
+    case SamplerMinFilter::NEAREST:
+    case SamplerMinFilter::LINEAR:
+        // The Vulkan spec recommends a max LOD of 0.25 to "disable" mipmapping.
+        // See "Mapping of OpenGL to Vulkan filter modes" in the VK Spec.
+        return 0.25f;
+    case SamplerMinFilter::NEAREST_MIPMAP_NEAREST:
+    case SamplerMinFilter::LINEAR_MIPMAP_NEAREST:
+    case SamplerMinFilter::NEAREST_MIPMAP_LINEAR:
+    case SamplerMinFilter::LINEAR_MIPMAP_LINEAR:
+        return VK_LOD_CLAMP_NONE;
+    }
+}
+
 VkShaderStageFlags getShaderStageFlags(ShaderStageFlags stageFlags) {
     VkShaderStageFlags flags = 0x0;
     if (any(stageFlags & ShaderStageFlags::VERTEX))     flags |= VK_SHADER_STAGE_VERTEX_BIT;

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

@@ -56,6 +56,19 @@ uint32_t getComponentCount(VkFormat format);
 VkComponentMapping getSwizzleMap(TextureSwizzle const swizzle[4]);
 VkShaderStageFlags getShaderStageFlags(ShaderStageFlags stageFlags);
 
+// Needed by the Platform for external sampler creation
+VkFilter getFilter(SamplerMinFilter filter);
+VkFilter getFilter(SamplerMagFilter filter);
+VkSamplerMipmapMode getMipmapMode(SamplerMinFilter filter);
+VkSamplerAddressMode getWrapMode(SamplerWrapMode mode);
+VkBool32 getCompareEnable(SamplerCompareMode mode);
+float getMaxLod(SamplerMinFilter filter);
+
+// Ycbcr related functions
+VkSamplerYcbcrModelConversion getYcbcrModelConversion(SamplerYcbcrModelConversion model);
+VkSamplerYcbcrRange getYcbcrRange(SamplerYcbcrRange range);
+VkChromaLocation getChromaLocation(ChromaLocation loc);
+
 inline VkImageViewType getViewType(SamplerType target) {
     switch (target) {
         case SamplerType::SAMPLER_CUBEMAP:

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

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

filament/backend/src/webgpu/WebGPUConstants.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_WEBGPUCONSTANTS_H
+#define TNT_FILAMENT_BACKEND_WEBGPUCONSTANTS_H
+
+#include <utils/Log.h>
+
+#include <cstdint>
+
+// FWGPU is short for Filament WebGPU
+
+// turn on runtime validation, namely for debugging, that would normally not run (for release)
+#define FWGPU_DEBUG_VALIDATION     0x00000001
+// print/log system component details to the console, e.g. about the
+// instance, surface, adapter, device, etc.
+#define FWGPU_PRINT_SYSTEM         0x00000002
+
+// Set this to enable logging "only" to one output stream. This is useful in the case where we want
+// to debug with print statements and want ordered logging (e.g slog.i and slog.e will not appear in
+// order of calls).
+#define FWGPU_DEBUG_FORCE_LOG_TO_I 0x00000004
+
+// Useful default combinations
+#define FWGPU_DEBUG_EVERYTHING     0xFFFFFFFF
+
+#if defined(FILAMENT_BACKEND_DEBUG_FLAG)
+    #define FWGPU_DEBUG_FORWARDED_FLAG (FILAMENT_BACKEND_DEBUG_FLAG & FWGPU_DEBUG_EVERYTHING)
+#else
+    #define FWGPU_DEBUG_FORWARDED_FLAG 0
+#endif
+
+#ifndef NDEBUG
+    #define FWGPU_DEBUG_FLAGS FWGPU_DEBUG_FORWARDED_FLAG
+#else
+    #define FWGPU_DEBUG_FLAGS 0
+#endif
+
+#define FWGPU_ENABLED(flags) (((FWGPU_DEBUG_FLAGS) & (flags)) == (flags))
+
+#if FWGPU_ENABLED(FWGPU_DEBUG_FORCE_LOG_TO_I)
+    #define FWGPU_LOGI (utils::slog.i)
+    #define FWGPU_LOGD FWGPU_LOGI
+    #define FWGPU_LOGE FWGPU_LOGI
+    #define FWGPU_LOGW FWGPU_LOGI
+#else
+    #define FWGPU_LOGE (utils::slog.e)
+    #define FWGPU_LOGW (utils::slog.w)
+    #define FWGPU_LOGD (utils::slog.d)
+    #define FWGPU_LOGI (utils::slog.i)
+#endif
+
+#endif// TNT_FILAMENT_BACKEND_WEBGPUCONSTANTS_H

filament/backend/src/webgpu/WebGPUDriver.cpp

@@ -14,21 +14,222 @@
  * limitations under the License.
  */
 
+#include "webgpu/WebGPUDriver.h"
+
+#include "webgpu/WebGPUConstants.h"
+#include <backend/platforms/WebGPUPlatform.h>
+
+#include "CommandStreamDispatcher.h"
+#include "DriverBase.h"
+#include "private/backend/Dispatcher.h"
+
 #include <backend/DriverEnums.h>
 #include <backend/Handle.h>
 
-#include "webgpu/WebGPUDriver.h"
-#include "CommandStreamDispatcher.h"
+#include <math/mat3.h>
+#include <utils/CString.h>
+#include <utils/ostream.h>
 
-#include <stdint.h>
+#include <dawn/webgpu_cpp_print.h>
+#include <webgpu/webgpu_cpp.h>
+
+#include <algorithm>
+#include <cstddef>
+#include <cstdint>
+#include <memory>
+#include <sstream>
+#include <string_view>
+#include <utility>
+#include <variant>
 
 namespace filament::backend {
 
-Driver* WebGPUDriver::create() {
-    return new WebGPUDriver();
+namespace {
+
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+void printInstanceDetails(wgpu::Instance const& instance) {
+    wgpu::SupportedWGSLLanguageFeatures supportedWGSLLanguageFeatures{};
+    if (!instance.GetWGSLLanguageFeatures(&supportedWGSLLanguageFeatures)) {
+        FWGPU_LOGW << "Failed to get WebGPU instance supported WGSL language features"
+                   << utils::io::endl;
+    } else {
+        FWGPU_LOGI << "WebGPU instance supported WGSL language features ("
+                   << supportedWGSLLanguageFeatures.featureCount << "):" << utils::io::endl;
+        if (supportedWGSLLanguageFeatures.featureCount > 0 &&
+                supportedWGSLLanguageFeatures.features != nullptr) {
+            std::for_each(supportedWGSLLanguageFeatures.features,
+                    supportedWGSLLanguageFeatures.features +
+                            supportedWGSLLanguageFeatures.featureCount,
+                    [](wgpu::WGSLLanguageFeatureName const featureName) {
+                        std::stringstream nameStream{};
+                        nameStream << featureName;
+                        FWGPU_LOGI << "  " << nameStream.str() << utils::io::endl;
+                    });
+        }
+    }
+}
+#endif
+
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+void printLimit(std::string_view name, const std::variant<uint32_t, uint64_t> value) {
+    FWGPU_LOGI << "  " << name.data() << ": ";
+    bool undefined = true;
+    if (std::holds_alternative<uint32_t>(value)) {
+        if (std::get<uint32_t>(value) != WGPU_LIMIT_U32_UNDEFINED) {
+            undefined = false;
+            FWGPU_LOGI << std::get<uint32_t>(value);
+        }
+    } else if (std::holds_alternative<uint64_t>(value)) {
+        if (std::get<uint64_t>(value) != WGPU_LIMIT_U64_UNDEFINED) {
+            undefined = false;
+            FWGPU_LOGI << std::get<uint64_t>(value);
+        }
+    }
+    if (undefined) {
+        FWGPU_LOGI << "UNDEFINED";
+    }
+    FWGPU_LOGI << utils::io::endl;
+}
+#endif
+
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+void printLimits(wgpu::Limits const& limits) {
+    printLimit("maxTextureDimension1D", limits.maxTextureDimension1D);
+    printLimit("maxTextureDimension2D", limits.maxTextureDimension2D);
+    printLimit("maxTextureDimension3D", limits.maxTextureDimension3D);
+    printLimit("maxTextureArrayLayers", limits.maxTextureArrayLayers);
+    printLimit("maxBindGroups", limits.maxBindGroups);
+    printLimit("maxBindGroupsPlusVertexBuffers", limits.maxBindGroupsPlusVertexBuffers);
+    printLimit("maxBindingsPerBindGroup", limits.maxBindingsPerBindGroup);
+    printLimit("maxDynamicUniformBuffersPerPipelineLayout",
+            limits.maxDynamicUniformBuffersPerPipelineLayout);
+    printLimit("maxDynamicStorageBuffersPerPipelineLayout",
+            limits.maxDynamicStorageBuffersPerPipelineLayout);
+    printLimit("maxSampledTexturesPerShaderStage", limits.maxSampledTexturesPerShaderStage);
+    printLimit("maxSamplersPerShaderStage", limits.maxSamplersPerShaderStage);
+    printLimit("maxStorageBuffersPerShaderStage", limits.maxStorageBuffersPerShaderStage);
+    printLimit("maxStorageTexturesPerShaderStage", limits.maxStorageTexturesPerShaderStage);
+    printLimit("maxUniformBuffersPerShaderStage", limits.maxUniformBuffersPerShaderStage);
+    printLimit("maxUniformBufferBindingSize", limits.maxUniformBufferBindingSize);
+    printLimit("maxStorageBufferBindingSize", limits.maxStorageBufferBindingSize);
+    printLimit("minUniformBufferOffsetAlignment", limits.minUniformBufferOffsetAlignment);
+    printLimit("minStorageBufferOffsetAlignment", limits.minStorageBufferOffsetAlignment);
+    printLimit("maxVertexBuffers", limits.maxVertexBuffers);
+    printLimit("maxBufferSize", limits.maxBufferSize);
+    printLimit("maxVertexAttributes", limits.maxVertexAttributes);
+    printLimit("maxVertexBufferArrayStride", limits.maxVertexBufferArrayStride);
+    printLimit("maxInterStageShaderVariables", limits.maxInterStageShaderVariables);
+    printLimit("maxColorAttachments", limits.maxColorAttachments);
+    printLimit("maxColorAttachmentBytesPerSample", limits.maxColorAttachmentBytesPerSample);
+    printLimit("maxComputeWorkgroupStorageSize", limits.maxComputeWorkgroupStorageSize);
+    printLimit("maxComputeInvocationsPerWorkgroup", limits.maxComputeInvocationsPerWorkgroup);
+    printLimit("maxComputeWorkgroupSizeX", limits.maxComputeWorkgroupSizeX);
+    printLimit("maxComputeWorkgroupSizeY", limits.maxComputeWorkgroupSizeY);
+    printLimit("maxComputeWorkgroupSizeZ", limits.maxComputeWorkgroupSizeZ);
+    printLimit("maxComputeWorkgroupsPerDimension", limits.maxComputeWorkgroupsPerDimension);
+    printLimit("maxStorageBuffersInVertexStage", limits.maxStorageBuffersInVertexStage);
+    printLimit("maxStorageTexturesInVertexStage", limits.maxStorageTexturesInVertexStage);
+    printLimit("maxStorageBuffersInFragmentStage", limits.maxStorageBuffersInFragmentStage);
+    printLimit("maxStorageTexturesInFragmentStage", limits.maxStorageTexturesInFragmentStage);
+}
+#endif
+
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+void printAdapterDetails(wgpu::Adapter const& adapter) {
+    wgpu::DawnAdapterPropertiesPowerPreference powerPreferenceProperties{};
+    wgpu::AdapterInfo adapterInfo{};
+    adapterInfo.nextInChain = &powerPreferenceProperties;
+    if (!adapter.GetInfo(&adapterInfo)) {
+        FWGPU_LOGW << "Failed to get WebGPU adapter info" << utils::io::endl;
+    } else {
+        std::stringstream backendTypeStream{};
+        backendTypeStream << adapterInfo.backendType;
+        std::stringstream adapterTypeStream{};
+        adapterTypeStream << adapterInfo.adapterType;
+        std::stringstream powerPreferenceStream{};
+        powerPreferenceStream << powerPreferenceProperties.powerPreference;
+        FWGPU_LOGI << "WebGPU adapter info:" << utils::io::endl;
+        FWGPU_LOGI << "  vendor: " << adapterInfo.vendor.data << utils::io::endl;
+        FWGPU_LOGI << "  architecture: " << adapterInfo.architecture.data << utils::io::endl;
+        FWGPU_LOGI << "  device: " << adapterInfo.device.data << utils::io::endl;
+        FWGPU_LOGI << "  description: " << adapterInfo.description.data << utils::io::endl;
+        FWGPU_LOGI << "  backend type: " << backendTypeStream.str().data() << utils::io::endl;
+        FWGPU_LOGI << "  adapter type: " << adapterTypeStream.str().data() << utils::io::endl;
+        FWGPU_LOGI << "  device ID: " << adapterInfo.deviceID << utils::io::endl;
+        FWGPU_LOGI << "  vendor ID: " << adapterInfo.vendorID << utils::io::endl;
+        FWGPU_LOGI << "  subgroup min size: " << adapterInfo.subgroupMinSize << utils::io::endl;
+        FWGPU_LOGI << "  subgroup max size: " << adapterInfo.subgroupMaxSize << utils::io::endl;
+        FWGPU_LOGI << "  power preference: " << powerPreferenceStream.str() << utils::io::endl;
+    }
+    wgpu::SupportedFeatures supportedFeatures{};
+    adapter.GetFeatures(&supportedFeatures);
+    FWGPU_LOGI << "WebGPU adapter supported features (" << supportedFeatures.featureCount
+               << "):" << utils::io::endl;
+    if (supportedFeatures.featureCount > 0 && supportedFeatures.features != nullptr) {
+        std::for_each(supportedFeatures.features,
+                supportedFeatures.features + supportedFeatures.featureCount,
+                [](wgpu::FeatureName const name) {
+                    std::stringstream nameStream{};
+                    nameStream << name;
+                    FWGPU_LOGI << "  " << nameStream.str().data() << utils::io::endl;
+                });
+    }
+    wgpu::Limits supportedLimits{};
+    if (!adapter.GetLimits(&supportedLimits)) {
+        FWGPU_LOGW << "Failed to get WebGPU adapter supported limits" << utils::io::endl;
+    } else {
+        FWGPU_LOGI << "WebGPU adapter supported limits:" << utils::io::endl;
+        printLimits(supportedLimits);
+    }
+}
+#endif
+
+
+
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+void printDeviceDetails(wgpu::Device const& device) {
+    wgpu::SupportedFeatures supportedFeatures{};
+    device.GetFeatures(&supportedFeatures);
+    FWGPU_LOGI << "WebGPU device supported features (" << supportedFeatures.featureCount
+               << "):" << utils::io::endl;
+    if (supportedFeatures.featureCount > 0 && supportedFeatures.features != nullptr) {
+        std::for_each(supportedFeatures.features,
+                supportedFeatures.features + supportedFeatures.featureCount,
+                [](wgpu::FeatureName const name) {
+                    std::stringstream nameStream{};
+                    nameStream << name;
+                    FWGPU_LOGI << "  " << nameStream.str().data() << utils::io::endl;
+                });
+    }
+    wgpu::Limits supportedLimits{};
+    if (!device.GetLimits(&supportedLimits)) {
+        FWGPU_LOGW << "Failed to get WebGPU supported device limits" << utils::io::endl;
+    } else {
+        FWGPU_LOGI << "WebGPU device supported limits:" << utils::io::endl;
+        printLimits(supportedLimits);
+    }
+}
+#endif
+
+}// namespace
+
+Driver* WebGPUDriver::create(WebGPUPlatform& platform, const Platform::DriverConfig& driverConfig) noexcept {
+    constexpr size_t defaultSize = FILAMENT_WEBGPU_HANDLE_ARENA_SIZE_IN_MB * 1024U * 1024U;
+    Platform::DriverConfig validConfig {driverConfig};
+    validConfig.handleArenaSize = std::max(driverConfig.handleArenaSize, defaultSize);
+    return new WebGPUDriver(platform, validConfig);
 }
 
-WebGPUDriver::WebGPUDriver() noexcept = default;
+WebGPUDriver::WebGPUDriver(WebGPUPlatform& platform, const Platform::DriverConfig& driverConfig) noexcept
+    : mPlatform(platform),
+      mHandleAllocator("Handles",
+              driverConfig.handleArenaSize,
+              driverConfig.disableHandleUseAfterFreeCheck,
+              driverConfig.disableHeapHandleTags) {
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+    printInstanceDetails(mPlatform.getInstance());
+#endif
+}
 
 WebGPUDriver::~WebGPUDriver() noexcept = default;
 
@@ -45,7 +246,7 @@ ShaderModel WebGPUDriver::getShaderModel() const noexcept {
 }
 
 ShaderLanguage WebGPUDriver::getShaderLanguage() const noexcept {
-    return ShaderLanguage::ESSL3;
+    return ShaderLanguage::WGSL;
 }
 
 // explicit instantiation of the Dispatcher
@@ -56,10 +257,14 @@ void WebGPUDriver::terminate() {
 }
 
 void WebGPUDriver::tick(int) {
+    mDevice.Tick();
 }
 
 void WebGPUDriver::beginFrame(int64_t monotonic_clock_ns,
         int64_t refreshIntervalNs, uint32_t frameId) {
+    wgpu::CommandEncoderDescriptor commandEncoderDescriptor{};
+    commandEncoderDescriptor.nextInChain = nullptr;
+    mCommandEncoder = mDevice.CreateCommandEncoder(&commandEncoderDescriptor);
 }
 
 void WebGPUDriver::setFrameScheduledCallback(Handle<HwSwapChain> sch,
@@ -76,6 +281,12 @@ void WebGPUDriver::setPresentationTime(int64_t monotonic_clock_ns) {
 }
 
 void WebGPUDriver::endFrame(uint32_t frameId) {
+    // FWGPU_LOGW << __FUNCTION__<< "\n";
+    mQueue.Submit(1, &mCommandBuffer);
+    mCommandEncoder = nullptr;
+    mCommandBuffer = nullptr;
+    mTextureView = nullptr;
+    mSwapChain->Present();
 }
 
 void WebGPUDriver::flush(int) {
@@ -109,6 +320,13 @@ void WebGPUDriver::destroyRenderTarget(Handle<HwRenderTarget> rth) {
 }
 
 void WebGPUDriver::destroySwapChain(Handle<HwSwapChain> sch) {
+    mSwapChain = nullptr;
+    // TODO:  use webgpu handle allocator from
+    //        https://github.com/google/filament/pull/8566
+//    if (sch) {
+//        HwSwapChain* hwSwapChain = handleCast<HwSwapChain*>(sch);
+//        destruct(sch, hwSwapChain);
+//    }
 }
 
 void WebGPUDriver::destroyStream(Handle<HwStream> sh) {
@@ -123,6 +341,222 @@ void WebGPUDriver::destroyDescriptorSetLayout(Handle<HwDescriptorSetLayout> tqh)
 void WebGPUDriver::destroyDescriptorSet(Handle<HwDescriptorSet> tqh) {
 }
 
+Handle<HwSwapChain> WebGPUDriver::createSwapChainS() noexcept {
+    // TODO:  use webgpu handle allocator from.
+    //        https://github.com/google/filament/pull/8566
+    // return allocAndConstructHandle<HwSwapChain>();
+    return Handle<HwSwapChain>((Handle<HwSwapChain>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwSwapChain> WebGPUDriver::createSwapChainHeadlessS() noexcept {
+    return Handle<HwSwapChain>((Handle<HwSwapChain>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwTexture> WebGPUDriver::createTextureS() noexcept {
+    return allocHandle<WGPUTexture>();
+}
+
+Handle<HwTexture> WebGPUDriver::importTextureS() noexcept {
+    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwProgram> WebGPUDriver::createProgramS() noexcept {
+    return Handle<HwProgram>((Handle<HwProgram>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwFence> WebGPUDriver::createFenceS() noexcept {
+    return Handle<HwFence>((Handle<HwFence>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwTimerQuery> WebGPUDriver::createTimerQueryS() noexcept {
+    return Handle<HwTimerQuery>((Handle<HwTimerQuery>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwIndexBuffer> WebGPUDriver::createIndexBufferS() noexcept {
+    return allocHandle<HwIndexBuffer>();
+}
+
+Handle<HwTexture> WebGPUDriver::createTextureViewS() noexcept {
+    return allocHandle<WGPUTexture>();
+}
+
+Handle<HwBufferObject> WebGPUDriver::createBufferObjectS() noexcept {
+    return allocHandle<WGPUBufferObject>();
+}
+
+Handle<HwRenderTarget> WebGPUDriver::createRenderTargetS() noexcept {
+    return allocHandle<WGPURenderTarget>();
+}
+
+Handle<HwVertexBuffer> WebGPUDriver::createVertexBufferS() noexcept {
+    return allocHandle<WGPUVertexBuffer>();
+}
+
+Handle<HwDescriptorSet> WebGPUDriver::createDescriptorSetS() noexcept {
+    return Handle<HwDescriptorSet>((Handle<HwDescriptorSet>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwRenderPrimitive> WebGPUDriver::createRenderPrimitiveS() noexcept {
+    return Handle<HwRenderPrimitive>((Handle<HwRenderPrimitive>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwVertexBufferInfo> WebGPUDriver::createVertexBufferInfoS() noexcept {
+    return allocHandle<WGPUVertexBufferInfo>();
+}
+
+Handle<HwTexture> WebGPUDriver::createTextureViewSwizzleS() noexcept {
+    return allocHandle<WGPUTexture>();
+}
+
+Handle<HwRenderTarget> WebGPUDriver::createDefaultRenderTargetS() noexcept {
+    return allocHandle<WGPURenderTarget>();
+}
+
+Handle<HwDescriptorSetLayout> WebGPUDriver::createDescriptorSetLayoutS() noexcept {
+    return Handle<HwDescriptorSetLayout>(
+            (Handle<HwDescriptorSetLayout>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwTexture> WebGPUDriver::createTextureExternalImageS() noexcept {
+    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwTexture> WebGPUDriver::createTextureExternalImage2S() noexcept {
+    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+}
+
+Handle<HwTexture> WebGPUDriver::createTextureExternalImagePlaneS() noexcept {
+    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+}
+
+void WebGPUDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow, uint64_t flags) {
+    // TODO:  use webgpu handle allocator from.
+    //        https://github.com/google/filament/pull/8566
+    // HwSwapChain* hwSwapChain = handleCast<HwSwapChain*>(sch);
+    assert_invariant(!mSwapChain);
+    wgpu::Surface surface = mPlatform.createSurface(nativeWindow, flags);
+    mAdapter = mPlatform.requestAdapter(surface);
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+    printAdapterDetails(mAdapter);
+#endif
+    mDevice = mPlatform.requestDevice(mAdapter);
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+    printDeviceDetails(mDevice);
+#endif
+    mQueue = mDevice.GetQueue();
+    mSwapChain = std::make_unique<WebGPUSwapChain>(std::move(surface), mAdapter, mDevice, flags);
+    // TODO configure the surface (maybe before or after creating the swapchain?
+    //                             how do we get the surface extent?)
+    // TODO actually create the swapchain
+    // auto onQueueWorkDone = [](wgpu::QueueWorkDoneStatus status, void* /* pUserData */) {
+        // FWGPU_LOGW << "Queued work finished with status: " << status << std::endl;
+    // };
+    // mQueue.OnSubmittedWorkDone(wgpu::CallbackMode::WaitAnyOnly, [](wgpu::QueueWorkDoneStatus status, void* pUserdata) {
+    //     FWGPU_LOGW << "Queued work finished with status:\n";
+    // }, nullptr /* pUserData */);
+    void* userDataPtr = nullptr;
+    mQueue.OnSubmittedWorkDone(
+        wgpu::CallbackMode::AllowProcessEvents,
+        [](wgpu::QueueWorkDoneStatus status, void* pUserData) {
+            FWGPU_LOGW << "Queued work finished with status: " << static_cast<int>(status) << "\n";
+            if (pUserData == nullptr) {
+                // Expected case
+            } else {
+                FWGPU_LOGW << "Unexpected non-null pUserData received.\n";
+            }
+        },
+        userDataPtr /* pUserData */
+    );
+    FWGPU_LOGW << "WebGPU support is still essentially a no-op at this point in development (only "
+                  "background components have been instantiated/selected, such as surface/screen, "
+                  "graphics device/GPU, etc.), thus nothing is being drawn to the screen."
+               << utils::io::endl;
+#if !FWGPU_ENABLED(FWGPU_PRINT_SYSTEM) && !defined(NDEBUG)
+    FWGPU_LOGI << "If the FILAMENT_BACKEND_DEBUG_FLAG variable were set with the " << utils::io::hex
+               << FWGPU_PRINT_SYSTEM << utils::io::dec
+               << " bit flag on during build time the application would print system details "
+                  "about the selected graphics device, surface, etc. To see this try "
+                  "rebuilding Filament with that flag, e.g. ./build.sh -x "
+               << FWGPU_PRINT_SYSTEM << " ..." << utils::io::endl;
+#endif
+    // TODO:  use webgpu handle allocator from.
+    //        https://github.com/google/filament/pull/8566
+    // hwSwapChain->swapChain = mSwapChain.get();
+}
+
+void WebGPUDriver::createSwapChainHeadlessR(Handle<HwSwapChain> sch, uint32_t width,
+        uint32_t height, uint64_t flags) {}
+
+void WebGPUDriver::createVertexBufferInfoR(Handle<HwVertexBufferInfo> vbih, uint8_t bufferCount,
+        uint8_t attributeCount, AttributeArray attributes) {}
+
+void WebGPUDriver::createVertexBufferR(Handle<HwVertexBuffer> vbh, uint32_t vertexCount,
+        Handle<HwVertexBufferInfo> vbih) {}
+
+void WebGPUDriver::createIndexBufferR(Handle<HwIndexBuffer> ibh, ElementType elementType,
+        uint32_t indexCount, BufferUsage usage) {}
+
+void WebGPUDriver::createBufferObjectR(Handle<HwBufferObject> boh, uint32_t byteCount,
+        BufferObjectBinding bindingType, BufferUsage usage) {}
+
+void WebGPUDriver::createTextureR(Handle<HwTexture> th, SamplerType target, uint8_t levels,
+        TextureFormat format, uint8_t samples, uint32_t w, uint32_t h, uint32_t depth,
+        TextureUsage usage) {}
+
+void WebGPUDriver::createTextureViewR(Handle<HwTexture> th, Handle<HwTexture> srch,
+        uint8_t baseLevel, uint8_t levelCount) {
+    // FWGPU_LOGW << __FUNCTION__<< "\n";
+    WGPUTexture const* src = handleCast<WGPUTexture>(srch);
+    (void) src;
+    // textures.insert(
+    //         constructHandle<WGPUTexture>(th, src, baseLevel, levelCount));
+}
+
+void WebGPUDriver::createTextureViewSwizzleR(Handle<HwTexture> th, Handle<HwTexture> srch,
+        backend::TextureSwizzle r, backend::TextureSwizzle g, backend::TextureSwizzle b,
+        backend::TextureSwizzle a) {}
+
+void WebGPUDriver::createTextureExternalImage2R(Handle<HwTexture> th, backend::SamplerType target,
+        backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
+        Platform::ExternalImageHandleRef externalImage) {}
+
+void WebGPUDriver::createTextureExternalImageR(Handle<HwTexture> th, backend::SamplerType target,
+        backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
+        void* externalImage) {}
+
+void WebGPUDriver::createTextureExternalImagePlaneR(Handle<HwTexture> th,
+        backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
+        void* image, uint32_t plane) {}
+
+void WebGPUDriver::importTextureR(Handle<HwTexture> th, intptr_t id, SamplerType target,
+        uint8_t levels, TextureFormat format, uint8_t samples, uint32_t w, uint32_t h,
+        uint32_t depth, TextureUsage usage) {}
+
+void WebGPUDriver::createRenderPrimitiveR(Handle<HwRenderPrimitive> rph, Handle<HwVertexBuffer> vbh,
+        Handle<HwIndexBuffer> ibh, PrimitiveType pt) {}
+
+void WebGPUDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {}
+
+void WebGPUDriver::createDefaultRenderTargetR(Handle<HwRenderTarget> rth, int) {
+    assert_invariant(!mDefaultRenderTarget);
+    mDefaultRenderTarget = constructHandle<WGPURenderTarget>(rth);
+    assert_invariant(mDefaultRenderTarget);
+}
+
+void WebGPUDriver::createRenderTargetR(Handle<HwRenderTarget> rth, TargetBufferFlags targets,
+        uint32_t width, uint32_t height, uint8_t samples, uint8_t layerCount, MRT color,
+        TargetBufferInfo depth, TargetBufferInfo stencil) {}
+
+void WebGPUDriver::createFenceR(Handle<HwFence> fh, int) {}
+
+void WebGPUDriver::createTimerQueryR(Handle<HwTimerQuery> tqh, int) {}
+
+void WebGPUDriver::createDescriptorSetLayoutR(Handle<HwDescriptorSetLayout> dslh,
+        backend::DescriptorSetLayout&& info) {}
+
+void WebGPUDriver::createDescriptorSetR(Handle<HwDescriptorSet> dsh,
+        Handle<HwDescriptorSetLayout> dslh) {}
+
 Handle<HwStream> WebGPUDriver::createStreamNative(void* nativeStream) {
     return {};
 }
@@ -131,9 +565,11 @@ Handle<HwStream> WebGPUDriver::createStreamAcquired() {
     return {};
 }
 
-void WebGPUDriver::setAcquiredImage(Handle<HwStream> sh, void* image,
-        CallbackHandler* handler, StreamCallback cb, void* userData) {
-}
+void WebGPUDriver::setAcquiredImage(Handle<HwStream> sh, void* image, const math::mat3f& transform,
+        CallbackHandler* handler, StreamCallback cb, void* userData) {}
+
+void WebGPUDriver::registerBufferObjectStreams(Handle<HwBufferObject> boh,
+        BufferObjectStreamDescriptor&& streams) {}
 
 void WebGPUDriver::setStreamDimensions(Handle<HwStream> sh, uint32_t width, uint32_t height) {
 }
@@ -238,6 +674,14 @@ size_t WebGPUDriver::getMaxUniformBufferSize() {
     return 16384u;
 }
 
+size_t WebGPUDriver::getMaxTextureSize(SamplerType target) {
+    return 2048u;
+}
+
+size_t WebGPUDriver::getMaxArrayTextureLayers() {
+    return 256u;
+}
+
 void WebGPUDriver::updateIndexBuffer(Handle<HwIndexBuffer> ibh, BufferDescriptor&& p,
         uint32_t byteOffset) {
     scheduleDestroy(std::move(p));
@@ -258,6 +702,11 @@ void WebGPUDriver::resetBufferObject(Handle<HwBufferObject> boh) {
 
 void WebGPUDriver::setVertexBufferObject(Handle<HwVertexBuffer> vbh, uint32_t index,
         Handle<HwBufferObject> boh) {
+    auto* vertexBuffer = handleCast<WGPUVertexBuffer>(vbh);
+    auto* bufferObject = handleCast<WGPUBufferObject>(boh);
+    assert_invariant(index < vertexBuffer->mBuffers.size());
+    vertexBuffer->setBuffer(bufferObject, index);
+
 }
 
 void WebGPUDriver::update3DImage(Handle<HwTexture> th,
@@ -290,9 +739,33 @@ void WebGPUDriver::compilePrograms(CompilerPriorityQueue priority,
 }
 
 void WebGPUDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassParams& params) {
+    // FWGPU_LOGW << __FUNCTION__<< "\n";
+    mTextureView = mSwapChain->GetNextSurfaceTextureView(params.viewport.width, params.viewport.height);
+    wgpu::RenderPassColorAttachment renderPassColorAttachment = {};
+    renderPassColorAttachment.view = mTextureView;
+    renderPassColorAttachment.resolveTarget = nullptr;
+    renderPassColorAttachment.loadOp = wgpu::LoadOp::Clear;
+    renderPassColorAttachment.storeOp = wgpu::StoreOp::Store;
+    renderPassColorAttachment.clearValue = wgpu::Color{1, 0 , 0 , 1};
+    renderPassColorAttachment.depthSlice = WGPU_DEPTH_SLICE_UNDEFINED;
+
+    wgpu::RenderPassDescriptor renderPassDescriptor = {};
+    renderPassDescriptor.nextInChain = nullptr;
+    renderPassDescriptor.colorAttachmentCount = 1;
+    renderPassDescriptor.colorAttachments = &renderPassColorAttachment;
+    renderPassDescriptor.depthStencilAttachment = nullptr;
+    renderPassDescriptor.timestampWrites = nullptr;
+
+    mRenderPassEncoder = mCommandEncoder.BeginRenderPass(&renderPassDescriptor);
+    mRenderPassEncoder.SetViewport((float)params.viewport.left, (float)params.viewport.bottom,
+            (float) params.viewport.width, (float) params.viewport.height, params.depthRange.near, params.depthRange.far);
+            // (float) 1024/*params.viewport.width*/, (float) 640/*params.viewport.height*/, params.depthRange.near, params.depthRange.far);
 }
 
 void WebGPUDriver::endRenderPass(int) {
+    // FWGPU_LOGW << __FUNCTION__<< "\n";
+    mRenderPassEncoder.End();
+    mRenderPassEncoder = nullptr;
 }
 
 void WebGPUDriver::nextSubpass(int) {
@@ -302,6 +775,12 @@ void WebGPUDriver::makeCurrent(Handle<HwSwapChain> drawSch, Handle<HwSwapChain>
 }
 
 void WebGPUDriver::commit(Handle<HwSwapChain> sch) {
+    wgpu::CommandBufferDescriptor commandBufferDescriptor = {};
+    commandBufferDescriptor.nextInChain = nullptr;
+    mCommandBuffer = mCommandEncoder.Finish(&commandBufferDescriptor);
+
+    // mQueue.Submit(1, &mCommandBuffer);
+    // mCommandBuffer = nullptr;
 }
 
 void WebGPUDriver::setPushConstant(backend::ShaderStage stage, uint8_t index,
@@ -358,6 +837,8 @@ void WebGPUDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
 }
 
 void WebGPUDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
+    // mRenderPassEncoder.DrawIndexed(indexCount, instanceCount, indexOffset, 0, 0);
+
 }
 
 void WebGPUDriver::draw(PipelineState pipelineState, Handle<HwRenderPrimitive> rph,

filament/backend/src/webgpu/WebGPUDriver.h

@@ -17,48 +17,99 @@
 #ifndef TNT_FILAMENT_BACKEND_WEBGPUDRIVER_H
 #define TNT_FILAMENT_BACKEND_WEBGPUDRIVER_H
 
-#include "private/backend/Driver.h"
+#include "webgpu/WebGPUSwapChain.h"
+#include <backend/platforms/WebGPUPlatform.h>
+
 #include "DriverBase.h"
+#include "private/backend/Dispatcher.h"
+#include "private/backend/Driver.h"
+#include "private/backend/HandleAllocator.h"
+#include <backend/DriverEnums.h>
 
 #include <utils/compiler.h>
 
+#include <webgpu/webgpu_cpp.h>
+
+#include "WebGPUHandles.h"
+
+#include <cstdint>
+#include <memory>
+
+#ifndef FILAMENT_WEBGPU_HANDLE_ARENA_SIZE_IN_MB
+#    define FILAMENT_WEBGPU_HANDLE_ARENA_SIZE_IN_MB 8
+#endif
+
 namespace filament::backend {
 
+/**
+ * WebGPU backend (driver) implementation
+ */
 class WebGPUDriver final : public DriverBase {
-    WebGPUDriver() noexcept;
-    ~WebGPUDriver() noexcept override;
-    Dispatcher getDispatcher() const noexcept final;
-
 public:
-    static Driver* create();
+    ~WebGPUDriver() noexcept override;
+
+    [[nodiscard]] Dispatcher getDispatcher() const noexcept final;
+    [[nodiscard]] static Driver* create(WebGPUPlatform& platform, const Platform::DriverConfig& driverConfig) noexcept;
 
 private:
-    ShaderModel getShaderModel() const noexcept final;
-    ShaderLanguage getShaderLanguage() const noexcept final;
-
-    uint64_t nextFakeHandle = 1;
+    explicit WebGPUDriver(WebGPUPlatform& platform, const Platform::DriverConfig& driverConfig) noexcept;
+    [[nodiscard]] ShaderModel getShaderModel() const noexcept final;
+    [[nodiscard]] ShaderLanguage getShaderLanguage() const noexcept final;
 
+    // the platform (e.g. OS) specific aspects of the WebGPU backend are strictly only
+    // handled in the WebGPUPlatform
+    WebGPUPlatform& mPlatform;
+    wgpu::Adapter mAdapter = nullptr;
+    wgpu::Device mDevice = nullptr;
+    wgpu::Queue mQueue = nullptr;
+    // TODO consider moving to handle allocator when ready
+    std::unique_ptr<WebGPUSwapChain> mSwapChain = nullptr;
+    uint64_t mNextFakeHandle = 1;
+    wgpu::CommandEncoder mCommandEncoder = nullptr;
+    wgpu::TextureView mTextureView = nullptr;
+    wgpu::RenderPassEncoder mRenderPassEncoder = nullptr;
+    wgpu::CommandBuffer mCommandBuffer = nullptr;
+    WGPURenderTarget* mDefaultRenderTarget = nullptr;
     /*
      * Driver interface
      */
-
     template<typename T>
     friend class ConcreteDispatcher;
 
-#define DECL_DRIVER_API(methodName, paramsDecl, params) \
+#define DECL_DRIVER_API(methodName, paramsDecl, params)                                            \
     UTILS_ALWAYS_INLINE inline void methodName(paramsDecl);
 
-#define DECL_DRIVER_API_SYNCHRONOUS(RetType, methodName, paramsDecl, params) \
+#define DECL_DRIVER_API_SYNCHRONOUS(RetType, methodName, paramsDecl, params)                       \
     RetType methodName(paramsDecl) override;
 
-#define DECL_DRIVER_API_RETURN(RetType, methodName, paramsDecl, params) \
-    RetType methodName##S() noexcept override { \
-        return RetType((RetType::HandleId)nextFakeHandle++); } \
-    UTILS_ALWAYS_INLINE inline void methodName##R(RetType, paramsDecl) { }
+#define DECL_DRIVER_API_RETURN(RetType, methodName, paramsDecl, params)                            \
+    RetType methodName##S() noexcept override;                                                     \
+    UTILS_ALWAYS_INLINE inline void methodName##R(RetType, paramsDecl);
 
 #include "private/backend/DriverAPI.inc"
+
+    /*
+     * Memory management
+     */
+
+    HandleAllocatorWGPU mHandleAllocator;
+
+    template<typename D>
+    Handle<D> allocHandle() {
+        return mHandleAllocator.allocate<D>();
+    }
+
+    template<typename D, typename B, typename ... ARGS>
+    D* constructHandle(Handle<B>& handle, ARGS&& ... args) noexcept {
+        return mHandleAllocator.construct<D>(handle, std::forward<ARGS>(args)...);
+    }
+    template<typename D, typename B>
+    D* handleCast(Handle<B> handle) noexcept {
+        return mHandleAllocator.handle_cast<D*>(handle);
+    }
+
 };
 
-} // namespace filament
+}// namespace filament::backend
 
 #endif // TNT_FILAMENT_BACKEND_WEBGPUDRIVER_H

filament/backend/src/webgpu/WebGPUHandles.cpp

@@ -0,0 +1,19 @@
+//
+// Created by Idris Idris Shah on 3/21/25.
+//
+
+#include "WebGPUHandles.h"
+
+namespace filament::backend {
+
+WGPUVertexBuffer::WGPUVertexBuffer(uint32_t vextexCount, uint32_t bufferCount,
+        Handle<WGPUVertexBufferInfo> vbih)
+    : HwVertexBuffer(vextexCount),
+      vbih(vbih),
+      mBuffers(MAX_VERTEX_BUFFER_COUNT) {}
+void WGPUVertexBuffer::setBuffer(WGPUBufferObject* bufferObject, uint32_t index) {}
+
+WGPUBufferObject::WGPUBufferObject(BufferObjectBinding bindingType, uint32_t byteCount)
+    : HwBufferObject(byteCount),
+      mBindingType(bindingType) {}
+}// namespace filament::backend

filament/backend/src/webgpu/WebGPUHandles.h

@@ -0,0 +1,109 @@
+//
+// Created by Idris Idris Shah on 3/21/25.
+//
+
+#ifndef TNT_FILAMENT_BACKEND_WEBGPUHANDLES_H
+#define TNT_FILAMENT_BACKEND_WEBGPUHANDLES_H
+
+#include "DriverBase.h"
+
+#include <backend/DriverEnums.h>
+#include <backend/Handle.h>
+#include <utils/FixedCapacityVector.h>
+
+#include <webgpu/webgpu_cpp.h>
+
+namespace filament::backend {
+
+struct WGPUBufferObject;
+struct WGPUVertexBufferInfo : public HwVertexBufferInfo {
+    WGPUVertexBufferInfo(uint8_t bufferCount, uint8_t attributeCount,
+            AttributeArray const& attributes)
+        : HwVertexBufferInfo(bufferCount, attributeCount),
+          attributes(attributes) {}
+    AttributeArray attributes;
+};
+
+struct WGPUVertexBuffer : public HwVertexBuffer {
+    WGPUVertexBuffer(uint32_t vextexCount, uint32_t bufferCount, Handle<WGPUVertexBufferInfo> vbih);
+    void setBuffer(WGPUBufferObject* bufferObject, uint32_t index);
+
+    Handle<WGPUVertexBufferInfo> vbih;
+    utils::FixedCapacityVector<wgpu::Buffer> mBuffers;
+};
+
+struct WGPUIndexBuffer : public HwIndexBuffer {
+    WGPUIndexBuffer(BufferUsage usage, uint8_t elementSize, uint32_t indexCount);
+
+    wgpu::Buffer buffer;
+};
+
+struct WGPUBufferObject : HwBufferObject {
+    WGPUBufferObject(BufferObjectBinding bindingType, uint32_t byteCount);
+
+    wgpu::Buffer mBuffer;
+    const BufferObjectBinding mBindingType;
+};
+
+class WGPUTexture : public HwTexture {
+public:
+    WGPUTexture(SamplerType target, uint8_t levels, TextureFormat format, uint8_t samples,
+            uint32_t width, uint32_t height, uint32_t depth, TextureUsage usage) noexcept;
+
+    // constructors for creating texture views
+    WGPUTexture(WGPUTexture const* src, uint8_t baseLevel, uint8_t levelCount) noexcept;
+
+    wgpu::Texture texture = nullptr;
+};
+
+class WGPURenderTarget : public HwRenderTarget {
+public:
+    class Attachment {
+    public:
+        friend class WGPURenderTarget;
+
+        Attachment() = default;
+        Attachment(WGPUTexture* gpuTexture, uint8_t level = 0, uint16_t layer = 0)
+            : level(level),
+              layer(layer),
+              texture(gpuTexture->texture),
+              mWGPUTexture(gpuTexture) {}
+
+        uint8_t level = 0;
+        uint16_t layer = 0;
+
+    private:
+        wgpu::Texture texture = nullptr;
+        WGPUTexture* mWGPUTexture = nullptr;
+    };
+
+    WGPURenderTarget(uint32_t width, uint32_t height, uint8_t samples,
+            Attachment colorAttachments[MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT]);
+    WGPURenderTarget()
+        : HwRenderTarget(0, 0),
+          defaultRenderTarget(true) {}
+
+    void setUpRenderPassAttachments(wgpu::RenderPassDescriptor* descriptor,
+            const RenderPassParams& params);
+
+    math::uint2 getAttachmentSize() noexcept;
+
+    bool isDefaultRenderTarget() const { return defaultRenderTarget; }
+    uint8_t getSamples() const { return samples; }
+
+    Attachment getDrawColorAttachment(size_t index);
+    Attachment getReadColorAttachment(size_t index);
+
+private:
+    static wgpu::LoadOp getLoadAction(const RenderPassParams& params, TargetBufferFlags buffer);
+    static wgpu::LoadOp getStoreAction(const RenderPassParams& params, TargetBufferFlags buffer);
+
+    bool defaultRenderTarget = false;
+    uint8_t samples = 1;
+
+    Attachment color[MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT] = {};
+    math::uint2 attachmentSize = {};
+};
+
+}// namespace filament::backend
+#endif// TNT_FILAMENT_BACKEND_WEBGPUHANDLES_H

filament/backend/src/webgpu/WebGPUPlatform.cpp

@@ -1,36 +0,0 @@
-/*
- * 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 "webgpu/WebGPUPlatform.h"
-#include "webgpu/WebGPUDriver.h"
-
-#include "utils/Log.h"
-
-namespace filament::backend {
-
-Driver* WebGPUPlatform::createDriver(void* const sharedContext , const Platform::DriverConfig& driverConfig) noexcept {
-    wgpu::InstanceDescriptor instance_descriptor {};
-    wgpu::Instance instance = wgpu::CreateInstance(&instance_descriptor);
-    if (instance) {
-        utils::slog.i << " WebGPU instance created\n";
-    } else {
-        utils::slog.e << " WebGPU instance failed to create\n";
-        return nullptr;
-    }
-    return WebGPUDriver::create();
-}
-
-} // namespace filament

filament/backend/src/webgpu/WebGPUPlatform.h

@@ -1,42 +0,0 @@
-/*
- * 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_WEBGPUPLATFORM_H
-#define TNT_FILAMENT_BACKEND_WEBGPUPLATFORM_H
-
-#include <backend/DriverEnums.h>
-#include <backend/Platform.h>
-
-#include "webgpu/webgpu_cpp.h"
-
-namespace filament::backend {
-
-class WebGPUPlatform
-    final : public Platform {
-public:
-
-    int getOSVersion() const noexcept final { return 0; }
-
-    ~WebGPUPlatform() noexcept override = default;
-
-protected:
-
-    Driver* createDriver(void* sharedContext, const Platform::DriverConfig& driverConfig) noexcept override;
-};
-
-} // namespace filament
-
-#endif // TNT_FILAMENT_BACKEND_WEBGPUPLATFORM_H

filament/backend/src/webgpu/WebGPUSwapChain.cpp

@@ -0,0 +1,277 @@
+/*
+ * 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 "webgpu/WebGPUSwapChain.h"
+
+#include "webgpu/WebGPUConstants.h"
+
+#include "backend/DriverEnums.h"
+
+#include <utils/Panic.h>
+#include <utils/ostream.h>
+
+#include <dawn/webgpu_cpp_print.h>
+#include <webgpu/webgpu_cpp.h>
+
+#include <algorithm>
+#include <cstdint>
+#include <sstream>
+
+namespace {
+
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+void printSurfaceCapabilitiesDetails(wgpu::SurfaceCapabilities const& capabilities) {
+    std::stringstream usagesStream{};
+    usagesStream << capabilities.usages;
+    FWGPU_LOGI << "WebGPU surface capabilities:" << utils::io::endl;
+    FWGPU_LOGI << "  surface usages: " << usagesStream.str().data() << utils::io::endl;
+    FWGPU_LOGI << "  surface formats (" << capabilities.formatCount << "):" << utils::io::endl;
+    if (capabilities.formatCount > 0 && capabilities.formats != nullptr) {
+        std::for_each(capabilities.formats, capabilities.formats + capabilities.formatCount,
+                [](wgpu::TextureFormat const format) {
+                    std::stringstream formatStream{};
+                    formatStream << format;
+                    FWGPU_LOGI << "    " << formatStream.str().data() << utils::io::endl;
+                });
+    }
+    FWGPU_LOGI << "  surface present modes (" << capabilities.presentModeCount
+               << "):" << utils::io::endl;
+    if (capabilities.presentModeCount > 0 && capabilities.presentModes != nullptr) {
+        std::for_each(capabilities.presentModes,
+                capabilities.presentModes + capabilities.presentModeCount,
+                [](wgpu::PresentMode const presentMode) {
+                    std::stringstream presentModeStream{};
+                    presentModeStream << presentMode;
+                    FWGPU_LOGI << "    " << presentModeStream.str().data() << utils::io::endl;
+                });
+    }
+    FWGPU_LOGI << "  surface alpha modes (" << capabilities.alphaModeCount
+               << "):" << utils::io::endl;
+    if (capabilities.alphaModeCount > 0 && capabilities.alphaModes != nullptr) {
+        std::for_each(capabilities.alphaModes,
+                capabilities.alphaModes + capabilities.alphaModeCount,
+                [](wgpu::CompositeAlphaMode const alphaMode) {
+                    std::stringstream alphaModeStream{};
+                    alphaModeStream << alphaMode;
+                    FWGPU_LOGI << "    " << alphaModeStream.str().data() << utils::io::endl;
+                });
+    }
+}
+#endif
+
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+void printSurfaceConfiguration(wgpu::SurfaceConfiguration const& config) {
+    std::stringstream formatStream{};
+    formatStream << config.format;
+    std::stringstream usageStream{};
+    usageStream << config.usage;
+    std::stringstream alphaModeStream{};
+    alphaModeStream << config.alphaMode;
+    std::stringstream presentModeStream{};
+    presentModeStream << config.presentMode;
+    FWGPU_LOGI << "WebGPU surface configuration:" << utils::io::endl;
+    FWGPU_LOGI << "  surface format: " << formatStream.str() << utils::io::endl;
+    FWGPU_LOGI << "  surface usage: " << usageStream.str() << utils::io::endl;
+    FWGPU_LOGI << "  surface view formats (" << config.viewFormatCount << "):" << utils::io::endl;
+    if (config.viewFormatCount > 0 && config.viewFormats != nullptr) {
+        std::for_each(config.viewFormats, config.viewFormats + config.viewFormatCount,
+                [](wgpu::TextureFormat const viewFormat) {
+                    std::stringstream viewFormatStream{};
+                    viewFormatStream << viewFormat;
+                    FWGPU_LOGI << "    " << viewFormatStream.str().data() << utils::io::endl;
+                });
+    }
+    FWGPU_LOGI << "  surface alpha mode: " << alphaModeStream.str() << utils::io::endl;
+    FWGPU_LOGI << "  surface width: " << config.width << utils::io::endl;
+    FWGPU_LOGI << "  surface height: " << config.height << utils::io::endl;
+    FWGPU_LOGI << "  surface present mode: " << presentModeStream.str() << utils::io::endl;
+}
+#endif
+
+wgpu::TextureFormat selectColorFormat(size_t availableFormatsCount,
+        wgpu::TextureFormat const* availableFormats, bool useSRGBColorSpace) {
+    const std::array expectedColorFormats =
+            useSRGBColorSpace ?
+                              std::array{
+                                  wgpu::TextureFormat::RGBA8UnormSrgb,
+                                  wgpu::TextureFormat::BGRA8UnormSrgb }
+                              : std::array{
+                                    wgpu::TextureFormat::RGBA8Unorm,
+                                    wgpu::TextureFormat::BGRA8Unorm };
+    auto const firstFoundColorFormat = std::find_first_of(expectedColorFormats.begin(),
+            expectedColorFormats.end(), availableFormats, availableFormats + availableFormatsCount);
+    FILAMENT_CHECK_POSTCONDITION(firstFoundColorFormat != expectedColorFormats.end())
+            << "Cannot find a suitable WebGPU swapchain "
+            << (useSRGBColorSpace ? "sRGB" : "non-standard (e.g. linear) RGB") << " color format";
+    return *firstFoundColorFormat;
+}
+
+wgpu::PresentMode selectPresentMode(size_t availablePresentModesCount,
+        wgpu::PresentMode const* availablePresentModes) {
+    // Verify that our chosen present mode is supported. In practice all devices support the FIFO
+    // mode, but we check for it anyway for completeness.  (and to avoid validation warnings)
+    const wgpu::PresentMode desiredPresentMode = wgpu::PresentMode::Fifo;
+    FILAMENT_CHECK_POSTCONDITION(
+            std::any_of(availablePresentModes, availablePresentModes + availablePresentModesCount,
+                    [](const wgpu::PresentMode availablePresentMode) {
+                        return availablePresentMode == desiredPresentMode;
+                    }))
+            << "Cannot find a suitable WebGPU swapchain present mode";
+    return desiredPresentMode;
+}
+
+wgpu::CompositeAlphaMode selectAlphaMode(size_t availableAlphaModesCount,
+        wgpu::CompositeAlphaMode const* availableAlphaModes) {
+    bool autoAvailable = false;
+    bool inheritAvailable = false;
+    bool opaqueAvailable = false;
+    bool premultipliedAvailable = false;
+    bool unpremultipliedAvailable = false;
+    std::for_each(availableAlphaModes, availableAlphaModes + availableAlphaModesCount,
+            [&](const wgpu::CompositeAlphaMode alphMode) {
+                switch (alphMode) {
+                    // in practice, the surface capabilities would not list Auto,
+                    // but for completeness and defensive programming we can leverage it
+                    // if it explicitly comes back as available/supported
+                    case wgpu::CompositeAlphaMode::Auto:
+                        autoAvailable = true;
+                        break;
+                    case wgpu::CompositeAlphaMode::Opaque:
+                        opaqueAvailable = true;
+                        break;
+                    case wgpu::CompositeAlphaMode::Premultiplied:
+                        premultipliedAvailable = true;
+                        break;
+                    case wgpu::CompositeAlphaMode::Unpremultiplied:
+                        unpremultipliedAvailable = true;
+                        break;
+                    case wgpu::CompositeAlphaMode::Inherit:
+                        inheritAvailable = true;
+                        break;
+                }
+            });
+    if (autoAvailable || inheritAvailable || opaqueAvailable) {
+        return wgpu::CompositeAlphaMode::Auto;
+    } else if (premultipliedAvailable) {
+        // In practice, we do not expect this to possibly happen, as opaque should be supported,
+        // which we select first. However, if the underlying system actually does not support
+        // opaque we allow it to be tested, but warn that this may not likely work as they expect
+        // (untested territory).
+        // We prefer premultiplied to unpremuliplied until that assumption should be adjusted.
+        FWGPU_LOGW << "Auto, Inherit, & Opaque composite alpha modes not supported. Filament has "
+                      "historically used these. Premultiplied alpha composite mode for "
+                      "transparency is being selected as a fallback, but may not work as expected."
+                   << utils::io::endl;
+        return wgpu::CompositeAlphaMode::Premultiplied;
+    } else {
+        FILAMENT_CHECK_POSTCONDITION(unpremultipliedAvailable)
+                << "No available composite alpha modes? Unknown/unhandled composite alpha mode?";
+        // Again, we don't expect this in practice, but allow if for the same reason as premultiplied.
+        // We prefer premultiplied to unpremuliplied until that assumption should be adjusted.
+        FWGPU_LOGW << "Auto, Inherit, & Opaque composite alpha modes not supported. Filament has "
+                      "historically used these. Unpremultiplied alpha composite mode for "
+                      "transparency is being selected as a fallback "
+                      "(premulitipled is not available either), but may not work as expected."
+                   << utils::io::endl;
+        return wgpu::CompositeAlphaMode::Unpremultiplied;
+    }
+}
+
+void initConfig(wgpu::SurfaceConfiguration& config, wgpu::Device& device,
+        wgpu::SurfaceCapabilities const& capabilities, bool useSRGBColorSpace) {
+    config.device = device;
+    config.usage = wgpu::TextureUsage::RenderAttachment;
+    config.format =
+            selectColorFormat(capabilities.formatCount, capabilities.formats, useSRGBColorSpace);
+    config.presentMode =
+            selectPresentMode(capabilities.presentModeCount, capabilities.presentModes);
+    config.alphaMode = selectAlphaMode(capabilities.alphaModeCount, capabilities.alphaModes);
+}
+
+}// namespace
+
+namespace filament::backend {
+
+WebGPUSwapChain::WebGPUSwapChain(wgpu::Surface&& surface, wgpu::Adapter& adapter,
+        wgpu::Device& device, uint64_t flags)
+    : mSurface(surface) {
+    wgpu::SurfaceCapabilities capabilities = {};
+    if (!mSurface.GetCapabilities(adapter, &capabilities)) {
+        FWGPU_LOGW << "Failed to get WebGPU surface capabilities" << utils::io::endl;
+    } else {
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+        printSurfaceCapabilitiesDetails(capabilities);
+#endif
+    }
+    const bool useSRGBColorSpace = (flags & SWAP_CHAIN_CONFIG_SRGB_COLORSPACE) != 0;
+    initConfig(mConfig, device, capabilities, useSRGBColorSpace);
+}
+
+WebGPUSwapChain::~WebGPUSwapChain() {
+    if (mConfigured) {
+        mSurface.Unconfigure();
+        mConfigured = false;
+    }
+}
+
+void WebGPUSwapChain::GetCurrentTexture(uint32_t width, uint32_t height, wgpu::SurfaceTexture* texture) {
+    if (width < 1 || height < 1) {
+        PANIC_LOG("WebGPUSwapChain::GetCurrentTexture: Invalid width and/or height requested.");
+        return;
+    }
+    if (mConfig.width != width || mConfig.height != height || !mConfigured) {
+        mConfig.width = width;
+        mConfig.height = height;
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+        printSurfaceConfiguration(mConfig);
+#endif
+        mSurface.Configure(&mConfig);
+        mConfigured = true;
+        return;
+    }
+
+    mSurface.GetCurrentTexture(texture);
+}
+
+wgpu::TextureView WebGPUSwapChain::GetNextSurfaceTextureView(uint32_t width, uint32_t height) {
+    wgpu::SurfaceTexture surfaceTexture;
+    GetCurrentTexture(width, height, &surfaceTexture);
+    if (surfaceTexture.status != wgpu::SurfaceGetCurrentTextureStatus::Success) {
+        return nullptr;
+    }
+
+    // Create a view for this surface texture
+    wgpu::TextureViewDescriptor textureViewDescriptor;
+    textureViewDescriptor.nextInChain = nullptr;
+    textureViewDescriptor.label = "Surface texture view";
+    textureViewDescriptor.format = surfaceTexture.texture.GetFormat();
+    textureViewDescriptor.dimension = wgpu::TextureViewDimension::e2D;
+    textureViewDescriptor.baseMipLevel = 0;
+    textureViewDescriptor.mipLevelCount = 1;
+    textureViewDescriptor.baseArrayLayer = 0;
+    textureViewDescriptor.arrayLayerCount = 1;
+    textureViewDescriptor.aspect = wgpu::TextureAspect::All;
+    wgpu::TextureView textureView = surfaceTexture.texture.CreateView(&textureViewDescriptor);
+
+    return textureView;
+}
+
+void WebGPUSwapChain::Present() {
+    assert_invariant(mSurface);
+    mSurface.Present();
+}
+
+}// namespace filament::backend

filament/backend/src/webgpu/WebGPUSwapChain.h

@@ -0,0 +1,46 @@
+/*
+ * Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef TNT_FILAMENT_BACKEND_WEBGPUSWAPCHAIN_H
+#define TNT_FILAMENT_BACKEND_WEBGPUSWAPCHAIN_H
+
+#include <webgpu/webgpu_cpp.h>
+
+#include <backend/Platform.h>
+
+#include <cstdint>
+
+namespace filament::backend {
+
+class WebGPUSwapChain : public Platform::SwapChain {
+public:
+    WebGPUSwapChain(wgpu::Surface&& surface, wgpu::Adapter& adapter, wgpu::Device& device,
+            uint64_t flags);
+    ~WebGPUSwapChain();
+
+    void GetCurrentTexture(uint32_t width, uint32_t height, wgpu::SurfaceTexture*);
+    wgpu::TextureView GetNextSurfaceTextureView(uint32_t width, uint32_t height);
+    void Present();
+
+private:
+    wgpu::Surface mSurface = {};
+    wgpu::SurfaceConfiguration mConfig = {};
+    bool mConfigured = false;
+};
+
+} // namespace filament::backend
+
+#endif //TNT_FILAMENT_BACKEND_WEBGPUSWAPCHAIN_H

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

@@ -0,0 +1,159 @@
+/*
+ * Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <backend/platforms/WebGPUPlatform.h>
+
+#include "webgpu/WebGPUConstants.h"
+#include "webgpu/WebGPUDriver.h"
+
+#include <backend/Platform.h>
+#include <utils/Panic.h>
+#include <utils/ostream.h>
+
+#include <dawn/webgpu_cpp_print.h>
+#include <webgpu/webgpu_cpp.h>
+
+#include <cstdint>
+#include <sstream>
+
+/**
+ * WebGPU Backend implementation common across platforms or operating systems (at least for now).
+ * Some of these functions may likely be refactored to platform/OS-specific implementations
+ * over time as needed. The caller of the WebGPUPlatform doesn't need to care which is the case.
+ */
+
+namespace filament::backend {
+
+namespace {
+
+//either returns a valid instance or panics
+[[nodiscard]] wgpu::Instance createInstance() {
+    wgpu::DawnTogglesDescriptor dawnTogglesDescriptor{};
+#if defined(FILAMENT_WEBGPU_IMMEDIATE_ERROR_HANDLING)
+#if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
+    FWGPU_LOGI << "setting on toggle enable_immediate_error_handling" << utils::io::endl;
+#endif
+    /**
+     * Have the un-captured error callback invoked immediately when an error occurs, rather than
+     * waiting for the next Tick. This enables using the stack trace in which the un-captured error
+     * occurred when breaking into the un-captured error callback.
+     * https://crbug.com/dawn/1789
+     */
+    static const char* toggleName = "enable_immediate_error_handling";
+    dawnTogglesDescriptor.enabledToggleCount = 1;
+    dawnTogglesDescriptor.enabledToggles = &toggleName;
+#endif
+    wgpu::InstanceDescriptor instanceDescriptor{
+        .nextInChain = &dawnTogglesDescriptor,
+        .capabilities = {
+            .timedWaitAnyEnable = true// TODO consider using pure async instead
+        }
+    };
+    wgpu::Instance instance = wgpu::CreateInstance(&instanceDescriptor);
+    FILAMENT_CHECK_POSTCONDITION(instance != nullptr) << "Unable to create webgpu instance.";
+    return instance;
+}
+
+}// namespace
+
+wgpu::Adapter WebGPUPlatform::requestAdapter(wgpu::Surface const& surface) {
+    // TODO consider power preference etc. (can be custom preferences passed to the platform or
+    //                                      based on whether this is a Mobile or Desktop system,
+    //                                      etc...)
+    wgpu::RequestAdapterOptions adaptorOptions{ .compatibleSurface = surface };
+    // note this just gets the first adapter
+    wgpu::Adapter adapter = nullptr;
+    wgpu::WaitStatus status = mInstance.WaitAny(
+            mInstance.RequestAdapter(&adaptorOptions, wgpu::CallbackMode::WaitAnyOnly,
+                    [&adapter](wgpu::RequestAdapterStatus const status,
+                            wgpu::Adapter const& readyAdapter, wgpu::StringView const message) {
+                        // TODO consider more robust error handling
+                        FILAMENT_CHECK_POSTCONDITION(status == wgpu::RequestAdapterStatus::Success)
+                                << "Unable to request a WebGPU adapter. Status "
+                                << static_cast<uint32_t>(status)
+                                << " with message: " << message.data;
+                        adapter = readyAdapter;
+                    }),
+            UINT16_MAX);// TODO define reasonable timeout (or do this asynchronously)
+    FILAMENT_CHECK_POSTCONDITION(status == wgpu::WaitStatus::Success)
+            << "Non-successful wait status requesting a WebGPU adapter "
+            << static_cast<uint32_t>(status);
+    FILAMENT_CHECK_POSTCONDITION(adapter != nullptr)
+            << "Failed to get a WebGPU adapter for the platform";
+    // TODO consider validating adapter has required features and/or limits
+    return adapter;
+}
+
+wgpu::Device WebGPUPlatform::requestDevice(wgpu::Adapter const& adapter) {
+    // TODO consider passing required features and/or limits
+    wgpu::DeviceDescriptor deviceDescriptor{};
+    deviceDescriptor.label = "graphics_device";
+    deviceDescriptor.defaultQueue.label = "default_queue";
+    deviceDescriptor.SetDeviceLostCallback(wgpu::CallbackMode::AllowSpontaneous,
+            [](wgpu::Device const&, wgpu::DeviceLostReason const& reason,
+                    wgpu::StringView message) {
+                if (reason == wgpu::DeviceLostReason::Destroyed) {
+                    FWGPU_LOGD << "WebGPU device lost due to being destroyed (expected)"
+                               << utils::io::endl;
+                    return;
+                }
+                // TODO try recreating the device instead of just panicking
+                std::stringstream reasonStream{};
+                reasonStream << reason;
+                FILAMENT_CHECK_POSTCONDITION(reason != wgpu::DeviceLostReason::Destroyed)
+                        << "WebGPU device lost: " << reasonStream.str() << " " << message.data;
+            });
+    deviceDescriptor.SetUncapturedErrorCallback(
+            [](wgpu::Device const&, wgpu::ErrorType errorType, wgpu::StringView message) {
+                std::stringstream typeStream{};
+                typeStream << errorType;
+                FWGPU_LOGE << "WebGPU device error: " << typeStream.str() << " " << message.data
+                           << utils::io::endl;
+            });
+    wgpu::Device device = nullptr;
+    wgpu::WaitStatus status = mInstance.WaitAny(
+            adapter.RequestDevice(&deviceDescriptor, wgpu::CallbackMode::WaitAnyOnly,
+                    [&device](wgpu::RequestDeviceStatus const status,
+                            wgpu::Device const& readyDevice, wgpu::StringView const message) {
+                        FILAMENT_CHECK_POSTCONDITION(status == wgpu::RequestDeviceStatus::Success)
+                                << "Unable to request a WebGPU device. Status: "
+                                << static_cast<uint32_t>(status)
+                                << " with message: " << message.data;
+                        device = readyDevice;
+                    }),
+            UINT64_MAX);// TODO define reasonable timeout (or do this asynchronously)
+    FILAMENT_CHECK_POSTCONDITION(status == wgpu::WaitStatus::Success)
+            << "Non-successful wait status requesting a WebGPU device "
+            << static_cast<uint32_t>(status);
+    FILAMENT_CHECK_POSTCONDITION(device != nullptr)
+            << "Failed to get a WebGPU device for the platform.";
+    return device;
+}
+
+Driver* WebGPUPlatform::createDriver(void* sharedContext,
+        const Platform::DriverConfig& driverConfig) noexcept {
+    if (sharedContext) {
+        FWGPU_LOGW << "sharedContext is ignored/unused in the WebGPU backend. A non-null "
+                      "sharedContext was provided, but it will be ignored."
+                   << utils::io::endl;
+    }
+    return WebGPUDriver::create(*this, driverConfig);
+}
+
+WebGPUPlatform::WebGPUPlatform()
+    : mInstance(createInstance()) {}
+
+}// namespace filament::backend

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

@@ -0,0 +1,43 @@
+/*
+ * 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 <backend/platforms/WebGPUPlatform.h>
+
+#include <utils/Panic.h>
+
+#include <webgpu/webgpu_cpp.h>
+
+#include <cstdint>
+
+/**
+ * Android OS specific implementation aspects of the WebGPU backend
+ */
+
+namespace filament::backend {
+
+wgpu::Surface WebGPUPlatform::createSurface(void* nativeWindow, uint64_t /*flags*/) {
+    wgpu::SurfaceSourceAndroidNativeWindow surfaceSourceAndroidWindow{};
+    surfaceSourceAndroidWindow.window = nativeWindow;
+    wgpu::SurfaceDescriptor surfaceDescriptor{
+        .nextInChain = &surfaceSourceAndroidWindow,
+        .label = "android_surface"
+    };
+    wgpu::Surface surface = mInstance.CreateSurface(&surfaceDescriptor);
+    FILAMENT_CHECK_POSTCONDITION(surface != nullptr) << "Unable to create Android-backed surface.";
+    return surface;
+}
+
+}// namespace filament::backend

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

@@ -0,0 +1,81 @@
+/*
+ * 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 <backend/platforms/WebGPUPlatform.h>
+
+#include <utils/Panic.h>
+#include <utils/ostream.h>
+
+#include <webgpu/webgpu_cpp.h>
+
+#include <cstdint>
+
+// Platform specific includes and defines
+#if defined(__APPLE__)
+    #include <Cocoa/Cocoa.h>
+    #import <QuartzCore/CAMetalLayer.h>
+#elif defined(FILAMENT_IOS)
+    // Metal is not available when building for the iOS simulator on Desktop.
+    #define METAL_AVAILABLE __has_include(<QuartzCore/CAMetalLayer.h>)
+    #if METAL_AVAILABLE
+        #import <Metal/Metal.h>
+        #import <QuartzCore/CAMetalLayer.h>
+    #endif
+    // is this needed?
+    #define METALVIEW_TAG 255
+#else
+    #error Not a supported Apple + WebGPU platform
+#endif
+
+/**
+ * Apple (Mac OS and IOS) specific implementation aspects of the WebGPU backend
+ */
+
+namespace filament::backend {
+
+wgpu::Surface WebGPUPlatform::createSurface(void* nativeWindow, uint64_t /*flags*/) {
+    wgpu::Surface surface = nullptr;
+#if defined(__APPLE__)
+    auto nsView = (__bridge NSView*) nativeWindow;
+    FILAMENT_CHECK_POSTCONDITION(nsView) << "Unable to obtain Metal-backed NSView.";
+    [nsView setWantsLayer:YES];
+    id metalLayer = [CAMetalLayer layer];
+    [nsView setLayer:metalLayer];
+    wgpu::SurfaceSourceMetalLayer surfaceSourceMetalLayer{};
+    surfaceSourceMetalLayer.layer = (__bridge void*) metalLayer;
+    wgpu::SurfaceDescriptor surfaceDescriptor = {
+        .nextInChain = &surfaceSourceMetalLayer,
+        .label = "metal_surface",
+    };
+    surface = mInstance.CreateSurface(&surfaceDescriptor);
+    FILAMENT_CHECK_POSTCONDITION(surface != nullptr) << "Unable to create Metal-backed surface.";
+#elif defined(FILAMENT_IOS)
+    CAMetalLayer* metalLayer = (CAMetalLayer*) nativeWindow;
+    wgpu::SurfaceSourceMetalLayer surfaceSourceMetalLayer{};
+    surfaceSourceMetalLayer.layer = (__bridge void*) metalLayer;
+    wgpu::SurfaceDescriptor surfaceDescriptor = {
+        .nextInChain = &surfaceSourceMetalLayer,
+        .label = "metal_surface",
+    };
+    surface = mInstance.CreateSurface(&surfaceDescriptor);
+    FILAMENT_CHECK_POSTCONDITION(surface != nullptr) << "Unable to create Metal-backed surface.";
+#else
+    #error Not a supported Apple + WebGPU platform
+#endif
+    return surface;
+}
+
+}// namespace filament::backend

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

@@ -0,0 +1,163 @@
+/*
+ * 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 <backend/platforms/WebGPUPlatform.h>
+
+#include <backend/DriverEnums.h>
+
+#include <utils/Panic.h>
+
+#include <webgpu/webgpu_cpp.h>
+
+#include <cstddef>
+#include <cstdint>
+
+#if defined(__linux__) || defined(__FreeBSD__)
+    #define LINUX_OR_FREEBSD 1
+#endif
+
+// Platform specific includes and defines
+#if defined(__linux__) && defined(FILAMENT_SUPPORTS_WAYLAND)
+    #include <dlfcn.h>
+    namespace {
+    typedef struct _wl {
+        struct wl_display* display;
+        struct wl_surface* surface;
+        uint32_t width;
+        uint32_t height;
+    } wl;
+    }// namespace
+#elif defined(LINUX_OR_FREEBSD) && defined(FILAMENT_SUPPORTS_X11)
+    // TODO: we should allow for headless on Linux explicitly. Right now this is the headless path
+    // (with no FILAMENT_SUPPORTS_XCB or FILAMENT_SUPPORTS_XLIB).
+    #include <dlfcn.h>
+    #if defined(FILAMENT_SUPPORTS_XCB)
+        #include <xcb/xcb.h>
+        namespace {
+        typedef xcb_connection_t* (*XCB_CONNECT)(const char* displayname, int* screenp);
+        }// namespace
+    #endif
+    #if defined(FILAMENT_SUPPORTS_XLIB)
+        #include <X11/Xlib.h>
+        namespace {
+        typedef Display* (*X11_OPEN_DISPLAY)(const char*);
+        }// namespace
+    #endif
+    static constexpr const char* LIBRARY_X11 = "libX11.so.6";
+    namespace {
+    struct XEnv {
+    #if defined(FILAMENT_SUPPORTS_XCB)
+        XCB_CONNECT xcbConnect;
+        xcb_connection_t* connection;
+    #endif
+    #if defined(FILAMENT_SUPPORTS_XLIB)
+        X11_OPEN_DISPLAY openDisplay;
+        Display* display;
+    #endif
+        void* library = nullptr;
+    } g_x11;
+    }// namespace
+#else
+    #error Not a supported Linux or FeeBSD + WebGPU platform
+#endif
+
+/**
+ * Linux OS specific implementation aspects of the WebGPU Backend
+ */
+
+namespace filament::backend {
+
+wgpu::Surface WebGPUPlatform::createSurface(void* nativeWindow, uint64_t flags) {
+    wgpu::Surface surface = nullptr;
+#if defined(__linux__) && defined(FILAMENT_SUPPORTS_WAYLAND)
+    wl* ptrval = reinterpret_cast<wl*>(nativeWindow);
+    wgpu::SurfaceSourceWaylandSurface surfaceSourceWayland{};
+    surfaceSourceWayland.display = ptrval->display;
+    surfaceSourceWayland.surface = ptrval->surface;
+    wgpu::SurfaceDescriptor surfaceDescriptor{
+        .nextInChain = &surfaceSourceWayland,
+        .label = "linux_wayland_surface"
+    };
+    surface = mInstance.CreateSurface(&surfaceDescriptor);
+    FILAMENT_CHECK_POSTCONDITION(surface != nullptr)
+            << "Unable to create Linux Wayland-backed surface.";
+#elif defined(LINUX_OR_FREEBSD) && defined(FILAMENT_SUPPORTS_X11)
+    if (g_x11.library == nullptr) {
+        g_x11.library = dlopen(LIBRARY_X11, RTLD_LOCAL | RTLD_NOW);
+        FILAMENT_CHECK_PRECONDITION(g_x11.library) << "Unable to open X11 library.";
+        #if defined(FILAMENT_SUPPORTS_XCB)
+            g_x11.xcbConnect = (XCB_CONNECT) dlsym(g_x11.library, "xcb_connect");
+            int screen = 0;
+            g_x11.connection = g_x11.xcbConnect(nullptr, &screen);
+        #endif
+        #if defined(FILAMENT_SUPPORTS_XLIB)
+            g_x11.openDisplay = (X11_OPEN_DISPLAY) dlsym(g_x11.library, "XOpenDisplay");
+            g_x11.display = g_x11.openDisplay(NULL);
+            FILAMENT_CHECK_PRECONDITION(g_x11.display) << "Unable to open X11 display.";
+        #endif
+    }
+    #if defined(FILAMENT_SUPPORTS_XCB) || defined(FILAMENT_SUPPORTS_XLIB)
+        bool useXcb = false;
+    #endif
+    #if defined(FILAMENT_SUPPORTS_XCB)
+        #if defined(FILAMENT_SUPPORTS_XLIB)
+            useXcb = (flags & SWAP_CHAIN_CONFIG_ENABLE_XCB) != 0;
+        #else
+            useXcb = true;
+        #endif
+        if (useXcb) {
+            wgpu::SurfaceSourceXCBWindow surfaceSourceXcb{};
+            surfaceSourceXcb.connection = g_x11.connection;
+
+            // TODO: this looks really wrong, please fix!!
+            surfaceSourceXcb.window = *((uint32_t*) nativeWindow);
+            wgpu::SurfaceDescriptor surfaceDescriptor{
+                .nextInChain = &surfaceSourceXcb,
+                .label = "linux_xcb_surface"
+            };
+            surface = mInstance.CreateSurface(&surfaceDescriptor);
+            FILAMENT_CHECK_POSTCONDITION(surface != nullptr)
+                    << "Unable to create Linux (or FreeBSD) XCB-backed surface.";
+        }
+    #endif
+    #if defined(FILAMENT_SUPPORTS_XLIB)
+        if (!useXcb) {
+            wgpu::SurfaceSourceXlibWindow surfaceSourceXlib{};
+            surfaceSourceXlib.display = g_x11.display;
+            surfaceSourceXlib.window = reinterpret_cast<uint64_t>(nativeWindow);
+            wgpu::SurfaceDescriptor surfaceDescriptor{
+                .nextInChain = &surfaceSourceXlib,
+                .label = "linux_xlib_surface"
+            };
+            surface = mInstance.CreateSurface(&surfaceDescriptor);
+            FILAMENT_CHECK_POSTCONDITION(surface != nullptr)
+                    << "Unable to create Linux (or FreeBSD) XLib-backed surface.";
+        }
+    #endif
+    FILAMENT_CHECK_POSTCONDITION(surface != nullptr)
+            << "Cannot create WebGPU X11 surface for Linux (or FreeBSD) OS "
+               "(not built with support for XCB or XLIB?)";
+#elif defined(__linux__)
+    FILAMENT_CHECK_POSTCONDITION(surface != nullptr)
+            << "Cannot create WebGPU surface for Linux (or FreeBSD) OS "
+               "(not built with support for Wayland or X11?)";
+#else
+    FILAMENT_CHECK_POSTCONDITION(surface != nullptr)
+            << "Not a supported (Linux) OS + WebGPU platform";
+#endif
+    return surface;
+}
+
+}// namespace filament::backend

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

@@ -0,0 +1,55 @@
+/*
+ * Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <backend/platforms/WebGPUPlatform.h>
+
+#include <utils/Panic.h>
+
+#include <webgpu/webgpu_cpp.h>
+
+#include <cstdint>
+
+#include <Windows.h>
+
+/**
+ * Windows OS specific implementation aspects of the WebGPU backend
+ */
+
+namespace filament::backend {
+
+wgpu::Surface WebGPUPlatform::createSurface(void* nativeWindow, uint64_t /*flags*/) {
+    // TODO verify this is necessary for Dawn implementation as well:
+    // On (at least) NVIDIA drivers, the Vulkan implementation (specifically the call to
+    // vkGetPhysicalDeviceSurfaceCapabilitiesKHR()) does not correctly handle the fact that
+    // each native window has its own DPI_AWARENESS_CONTEXT, and erroneously uses the context
+    // of the calling thread. As a workaround, we set the current thread's DPI_AWARENESS_CONTEXT
+    // to that of the native window we've been given. This isn't a perfect solution, because an
+    // application could create swap chains on multiple native windows with varying DPI-awareness,
+    // but even then, at least one of the windows would be guaranteed to work correctly.
+    SetThreadDpiAwarenessContext(GetWindowDpiAwarenessContext((HWND) nativeWindow));
+    wgpu::SurfaceSourceWindowsHWND surfaceSourceWin{};
+    surfaceSourceWin.hinstance = GetModuleHandle(nullptr);
+    surfaceSourceWin.hwnd = nativeWindow;
+    const wgpu::SurfaceDescriptor surfaceDescriptor{
+        .nextInChain = &surfaceSourceWin,
+        .label = "windows_surface"
+    };
+    wgpu::Surface surface = mInstance.CreateSurface(&surfaceDescriptor);
+    FILAMENT_CHECK_POSTCONDITION(surface.Get() != nullptr) << "Unable to create Windows-backed surface.";
+    return surface;
+}
+
+}// namespace filament::backend

filament/backend/test/ImageExpectations.cpp

@@ -0,0 +1,172 @@
+/*
+ * 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 "ImageExpectations.h"
+
+#include "gtest/gtest.h"
+#include "gmock/gmock.h"
+#include "absl/strings/str_format.h"
+#include "utils/Hash.h"
+#include <fstream>
+
+#include "backend/PixelBufferDescriptor.h"
+#include "private/backend/DriverApi.h"
+
+#ifndef FILAMENT_IOS
+
+#include <imageio/ImageEncoder.h>
+#include <image/ColorTransform.h>
+
+#endif
+
+ScreenshotParams::ScreenshotParams(int width, int height, std::string fileName,
+        uint32_t expectedPixelHash)
+        : mWidth(width), mHeight(height), mFileName(std::move(fileName)),
+          mExpectedPixelHash(expectedPixelHash) {}
+
+int ScreenshotParams::width() const {
+    return mWidth;
+}
+
+int ScreenshotParams::height() const {
+    return mHeight;
+}
+
+uint32_t ScreenshotParams::expectedHash() const {
+    return mExpectedPixelHash;
+}
+
+std::string ScreenshotParams::outputDirectoryPath() const {
+    return ".";
+}
+
+std::string ScreenshotParams::generatedActualFileName() const {
+    return absl::StrFormat("%s_actual.png", mFileName);
+}
+
+std::string ScreenshotParams::generatedActualFilePath() const {
+    return absl::StrFormat("%s/%s", outputDirectoryPath(), generatedActualFileName());
+}
+
+std::string ScreenshotParams::goldenFileName() const {
+    return absl::StrFormat("%s_golden.png", mFileName);
+}
+
+std::string ScreenshotParams::goldenFilePath() const {
+    return absl::StrFormat("%s/%s", outputDirectoryPath(), goldenFileName());
+}
+
+ImageExpectation::ImageExpectation(const char* fileName, int lineNumber,
+        filament::backend::DriverApi& api, ScreenshotParams params,
+        filament::backend::RenderTargetHandle renderTarget)
+        : mFileName(fileName), mLineNumber(lineNumber), mParams(std::move(params)),
+          mResult(api, renderTarget, mParams) {}
+
+void ImageExpectation::evaluate() {
+    // Ensure this is only evaluated once.
+    if (mEvaluated) {
+        return;
+    }
+    mEvaluated = true;
+
+    // Do the actual image comparison inside a scoped trace with the stored file and line.
+    {
+        testing::ScopedTrace trace(mFileName, mLineNumber, "");
+        compareImage();
+    }
+}
+
+void ImageExpectation::compareImage() const {
+    bool bytesFilled = mResult.bytesFilled();
+    EXPECT_THAT(bytesFilled, testing::IsTrue())
+                        << "Render target wasn't copied to the buffer for " << mFileName;
+    if (bytesFilled) {
+        uint32_t actualHash = mResult.hash();
+        EXPECT_THAT(actualHash, testing::Eq(mParams.expectedHash()));
+    }
+}
+
+ImageExpectations::ImageExpectations(filament::backend::DriverApi& api) : mApi(api) {}
+
+ImageExpectations::~ImageExpectations() {
+    // Guarantee that all expectations are evaluated when this leaves scope even if the caller
+    // forgot to manually evaluate them.
+    evaluate();
+}
+
+void ImageExpectations::addExpectation(const char* fileName, int lineNumber,
+        filament::backend::RenderTargetHandle renderTarget, ScreenshotParams params) {
+    mExpectations.emplace_back(fileName, lineNumber, mApi, std::move(params), renderTarget);
+}
+
+void ImageExpectations::evaluate() {
+    for (auto& expectation: mExpectations) {
+        expectation.evaluate();
+    }
+    mExpectations.clear();
+}
+
+RenderTargetDump::RenderTargetDump(filament::backend::DriverApi& api,
+        filament::backend::RenderTargetHandle renderTarget, const ScreenshotParams& params)
+        : mInternal(std::make_unique<RenderTargetDump::Internal>(params)) {
+#ifdef FILAMENT_IOS
+    bytesFilled_ = true;
+    bytes_.resize(size);
+    std::fill(bytes_.begin(), bytes_.end(), 0);
+#else
+    const size_t size = mInternal->params.width() * mInternal->params.height() * 4;
+    mInternal->bytes.resize(size);
+
+    auto cb = [](void* buffer, size_t size, void* user) {
+        auto* internal = static_cast<RenderTargetDump::Internal*>(user);
+        image::LinearImage image(internal->params.width(), internal->params.width(), 4);
+        image = image::toLinearWithAlpha<uint8_t>(internal->params.width(),
+                internal->params.height(),
+                internal->params.width() * 4, (uint8_t*)buffer);
+        std::string filePath = internal->params.generatedActualFilePath();
+        std::ofstream pngStream(filePath, std::ios::binary | std::ios::trunc);
+        image::ImageEncoder::encode(pngStream, image::ImageEncoder::Format::PNG, image, "",
+                filePath);
+        internal->bytesFilled = true;
+    };
+    filament::backend::PixelBufferDescriptor pb(mInternal->bytes.data(), size,
+            filament::backend::PixelDataFormat::RGBA, filament::backend::PixelDataType::UBYTE, cb,
+            (void*)mInternal.get());
+    api.readPixels(renderTarget, 0, 0, mInternal->params.width(), mInternal->params.height(),
+            std::move(pb));
+#endif
+}
+
+RenderTargetDump::~RenderTargetDump() {
+    // If the GPU callback hasn't been made yet then there's a callback elsewhere that has a copy of
+    // the internal pointer. But there's no guarantee that the callback will be ever made if the GPU
+    // pipeline wasn't run for some reason. So it is necessary to leak the memory.
+    // It would be possible to try to coordinate with the callback to have it clean up the memory,
+    // but if this condition happens there's already an issue with the test case so there's no need.
+    if (!bytesFilled()) {
+        mInternal.release();
+    }
+}
+
+uint32_t RenderTargetDump::hash() const {
+    return utils::hash::murmur3((uint32_t*)mInternal->bytes.data(), mInternal->bytes.size() / 4, 0);
+}
+
+bool RenderTargetDump::bytesFilled() const {
+    return mInternal->bytesFilled;
+}
+
+RenderTargetDump::Internal::Internal(const ScreenshotParams& params) : params(params) {}