Update timer Query call back

Moved them out of DriverEnums because only the vk backend needs
them.

.clang-format

@@ -57,7 +57,7 @@ SpaceBeforeInheritanceColon: true
 SpaceBeforeParens: ControlStatements
 SpaceBeforeRangeBasedForLoopColon: false
 SpaceInEmptyParentheses: false
-SpacesBeforeTrailingComments: 0
+SpacesBeforeTrailingComments: 1
 SpacesInAngles: false
 SpacesInCStyleCastParentheses: false
 SpacesInContainerLiterals: false

.github/actions/get-commit-msg/action.yml

@@ -0,0 +1,45 @@
+name: 'Get commit message'
+outputs:
+  msg:
+    value: ${{ steps.action_output.outputs.msg }}
+runs:
+  using: "composite"
+  steps:
+    - name: Find commit message (on push)
+      if: github.event_name == 'push'
+      shell: bash
+      run: |
+        AUTHOR_NAME="${{ github.event.head_commit.author.name }}"
+        AUTHOR_EMAIL="${{ github.event.head_commit.author.email }}"
+        TSTAMP="${{ github.event.head_commit.timestamp }}"
+        echo "commit ${{ github.event.head_commit.id }}" >> /tmp/commit_msg.txt
+        echo "Author: ${AUTHOR_NAME}<${AUTHOR_EMAIL}>" >> /tmp/commit_msg.txt
+        echo "Date: ${TSTAMP}" >> /tmp/commit_msg.txt
+        echo "" >> /tmp/commit_msg.txt
+        echo "${{ github.event.head_commit.message }}" >> /tmp/commit_msg.txt
+    - name: Find commit message (PR)
+      shell: bash
+      id: checkout_code
+      if: github.event_name == 'pull_request'
+      run: |
+        echo "+++++ head commit message +++++"
+        echo "$(git log -1 --no-merges)"
+        echo "+++++++++++++++++++++++++++++++"
+        echo "hash=$(git rev-parse HEAD)" >> "$GITHUB_OUTPUT"
+        git checkout ${{ github.event.pull_request.head.sha }}
+        echo "$(git log -1 --no-merges)" >> /tmp/commit_msg.txt
+    - shell: bash
+      id: action_output
+      run: |
+        DELIMITER="EOF_FILE_CONTENT_$(date +%s)" # Using timestamp to make it more unique
+        echo "msg<<$DELIMITER" >> "$GITHUB_OUTPUT"
+        cat /tmp/commit_msg.txt >> "$GITHUB_OUTPUT"
+        echo "$DELIMITER" >> "$GITHUB_OUTPUT"
+        echo "----- got commit message ---"
+        cat /tmp/commit_msg.txt
+        echo "----------------------------"
+    - name: Cleanup Find commit message (PR)
+      shell: bash
+      if: github.event_name == 'pull_request'
+      run: |
+        git checkout ${{ steps.checkout_code.outputs.hash }}

.github/workflows/postsubmit.yml

@@ -0,0 +1,35 @@
+name: 'Post-submit tasks'
+
+on:
+  push:
+    branches:
+      - main
+
+jobs:
+  update-renderdiff-goldens:
+    name: update-renderdiff-goldens
+    runs-on: 'ubuntu-24.04-4core'
+    steps:
+      - uses: actions/checkout@v4.1.6
+        with:
+          fetch-depth: 0
+      - uses: ./.github/actions/linux-prereq
+      - id: get_commit_msg
+        uses: ./.github/actions/get-commit-msg
+      - name: Prerequisites
+        run: pip install tifffile numpy
+      - name: Run update script
+        env:
+          GH_TOKEN: ${{ secrets.FILAMENTBOT_TOKEN }}
+        run: |
+          GOLDEN_BRANCH=$(echo "${{ steps.get_commit_msg.outputs.msg }}" | python3 test/renderdiff/src/commit_msg.py)
+          COMMIT_HASH=$(echo "${{ steps.get_commit_msg.outputs.msg }}" | head -n 1 | tr -d 'commit ')
+          if [[ "${GOLDEN_BRANCH}" != "main" ]]; then
+              git config --global user.email "filament.bot@gmail.com"
+              git config --global user.name "Filament Bot"
+              git config --global credential.helper cache
+              echo "branch==${GOLDEN_BRANCH}"
+              echo "hash==${COMMIT_HASH}"
+              python3 test/renderdiff/src/update_golden.py --branch=${GOLDEN_BRANCH} \
+                  --merge-to-main --filament-tag=${COMMIT_HASH} --golden-repo-token=${GH_TOKEN}
+          fi

.github/workflows/presubmit.yml

@@ -3,10 +3,10 @@ name: Presubmit
 on:
   push:
     branches:
-    - main
+      - main
   pull_request:
     branches:
-    - main
+      - main
 
 jobs:
   build-desktop-mac:
@@ -41,8 +41,7 @@ jobs:
 
   build-windows:
     name: build-windows
-    runs-on: win-2019-16core
+    runs-on: windows-2022-32core
-
     steps:
       - uses: actions/checkout@v4.1.6
         with:
@@ -110,9 +109,9 @@ jobs:
           fetch-depth: 0
       - name: Check for manual edits to /docs
         run: |
-           echo "${{ github.event.pull_request.head.sha }} -- ${{ github.event.pull_request.head.sha }}"
+          echo "${{ github.event.pull_request.head.sha }} -- ${{ github.event.pull_request.head.sha }}"
-           # disable for now
+          # disable for now
-           # bash docs_src/build/presubmit_check.sh ${{ github.event.pull_request.head.sha }}
+          # bash docs_src/build/presubmit_check.sh ${{ github.event.pull_request.head.sha }}
 
   test-renderdiff:
     name: test-renderdiff
@@ -121,20 +120,21 @@ jobs:
       - uses: actions/checkout@v4.1.6
         with:
           fetch-depth: 0
+      - id: get_commit_msg
+        uses: ./.github/actions/get-commit-msg
       - uses: ./.github/actions/mac-prereq
       - name: Cache Mesa and deps
-        id: mesa-cache
         uses: actions/cache@v4
         with:
           path: mesa
           key: ${{ runner.os }}-mesa-deps-2-${{ vars.MESA_VERSION }}
       - name: Prerequisites
-        id: prereqs
         run: |
           bash build/common/get-mesa.sh
           pip install tifffile numpy
       - name: Run Test
-        run: bash test/renderdiff/test.sh
+        run: |
+          echo "${{ steps.get_commit_msg.outputs.msg }}" | bash test/renderdiff/test.sh
       - uses: actions/upload-artifact@v4
         with:
           name: presubmit-renderdiff-result

.github/workflows/release.yml

@@ -226,7 +226,7 @@ jobs:
 
   build-windows:
     name: build-windows
-    runs-on: windows-2019-32core
+    runs-on: windows-2022-32core
     if: github.event_name == 'release' || github.event.inputs.platform == 'windows'
 
     steps:

.github/workflows/windows-continuous.yml

@@ -10,7 +10,7 @@ on:
 jobs:
   build-windows:
     name: build-windows
-    runs-on: windows-2019-32core
+    runs-on: windows-2022-32core
 
     steps:
       - uses: actions/checkout@v4.1.6

BUILDING.md

@@ -97,6 +97,10 @@ Make sure you've installed the following dependencies:
 - `libxcomposite-dev` (`libXcomposite-devel` on Fedora)
 - `libxxf86vm-dev` (`libXxf86vm-devel` on Fedora)
 
+```shell
+sudo apt install clang-14 libglu1-mesa-dev libc++-14-dev libc++abi-14-dev ninja-build libxi-dev libxcomposite-dev libxxf86vm-dev -y
+```
+
 After dependencies have been installed, we highly recommend using the [easy build](#easy-build)
 script.
 

CMakeLists.txt

@@ -49,6 +49,12 @@ option(FILAMENT_SUPPORTS_OSMESA "Enable OSMesa (headless GL context) for Filamen
 
 option(FILAMENT_ENABLE_FGVIEWER "Enable the frame graph viewer" OFF)
 
+option(FILAMENT_USE_ABSEIL_LOGGING "Use Abseil to log, may increase binary size" OFF)
+
+# This is to disable GTAO for the short-term while we investigate a way to better manage size increases.
+# On the regular filament build (where size is of less concern), we enable GTAO by default.
+option(FILAMENT_DISABLE_GTAO "Disable GTAO" OFF)
+
 set(FILAMENT_NDK_VERSION "" CACHE STRING
     "Android NDK version or version prefix to be used when building for Android."
 )
@@ -587,6 +593,10 @@ if (CMAKE_BUILD_TYPE STREQUAL "Debug" AND NOT FILAMENT_BACKEND_DEBUG_FLAG STREQU
     unset(FILAMENT_BACKEND_DEBUG_FLAG)
 endif()
 
+if (FILAMENT_USE_ABSEIL_LOGGING)
+    add_definitions(-DFILAMENT_USE_ABSEIL_LOGGING)
+endif()
+
 # ==================================================================================================
 # Material compilation flags
 # ==================================================================================================

NEW_RELEASE_NOTES.md

@@ -1,9 +1,9 @@
 # Filament Release Notes log
 
 **If you are merging a PR into main**: please add the release note below, under the *Release notes
-for next branch cut* header.
+We are chaning the way Vulkan buffers are handled. We need to switch over to a managed (or view-based) model where the data stored inside the object is a proxy to a Vulkan object that can dynamically be swapped around.
 
 **If you are cherry-picking a commit into an rc/ branch**: add the release note under the
 appropriate header in [RELEASE_NOTES.md](./RELEASE_NOTES.md).
 
-## Release notes for next branch cut
+## Release notes for next branch cut

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.60.1'
+    implementation 'com.google.android.filament:filament-android:1.61.1'
 }
 ```
 
@@ -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', '~> 1.60.1'
+pod 'Filament', '~> 1.61.1'
 ```
 
 ## Documentation

RELEASE_NOTES.md

@@ -7,6 +7,13 @@ 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).
 
+## v1.61.2
+
+- samples: samples now have a CLI to select backend api
+
+## v1.61.1
+
+
 ## v1.61.0
 
 - materials: sampler now export their type in the material binary [⚠️ **New Material Version**]

android/filament-android/CMakeLists.txt

@@ -59,6 +59,10 @@ add_library(smol-v STATIC IMPORTED)
 set_target_properties(smol-v PROPERTIES IMPORTED_LOCATION
         ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libsmol-v.a)
 
+add_library(abseil STATIC IMPORTED)
+set_target_properties(abseil PROPERTIES IMPORTED_LOCATION
+        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libabseil.a)
+
 if (FILAMENT_ENABLE_FGVIEWER)
     add_library(fgviewer STATIC IMPORTED)
     set_target_properties(fgviewer PROPERTIES IMPORTED_LOCATION
@@ -128,6 +132,7 @@ target_link_libraries(filament-jni
     PRIVATE jnigraphics
     PRIVATE utils
     PRIVATE perfetto
+    PRIVATE abseil
 
     # libgeometry is PUBLIC because gltfio uses it.
     PUBLIC geometry

android/gradle.properties

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.60.1
+VERSION_NAME=1.61.1
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 

build/common/get-mesa.sh

@@ -41,6 +41,8 @@ for cmd in "${NEEDED_PYTHON_DEPS[@]}"; do
 done
 deactivate
 
+LOCAL_PKG_CONFIG_PATH=
+
 # Install system deps
 if [[ "$OS_NAME" == "Linux" ]]; then
     if [[ "$GITHUB_WORKFLOW" ]]; then
@@ -82,6 +84,9 @@ elif [[ "$OS_NAME" == "Darwin" ]]; then
         fi
     fi
     HOMEBREW_NO_INSTALLED_DEPENDENTS_CHECK=true brew install autoconf automake libx11 libxext libxrandr llvm@${LLVM_VERSION} ninja meson pkg-config libxshmfence
+
+    # For reasons unknown, this is necessary for pkg-config to find homebrew's packages
+    LOCAL_PKG_CONFIG_PATH="/opt/homebrew/lib/pkgconfig:$PKG_CONFIG_PATH"
 fi # [[ "$OS_NAME" == x ]]
 
 LOCAL_LDFLAGS=${LDFLAGS}
@@ -134,9 +139,11 @@ fi
 # -Dgallium-drivers=swrast  => builds GL software rasterizer
 # -Dvulkan-drivers=swrast   => builds VK software rasterizer
 # -Dgallium-drivers=llvmpipe is needed for GL >= 4.1 pipe-screen (see src/gallium/auxiliary/target-helpers/inline_sw_helper.h)
-CXX=${LOCAL_CXX} CC=${LOCAL_CC} PATH=${LOCAL_PATH} LDFLAGS=${LOCAL_LDFLAGS} CPPFLAGS=${LOCAL_CPPFLAGS} \
+PKG_CONFIG_PATH=${LOCAL_PKG_CONFIG_PATH} PATH=${LOCAL_PATH} \
+CXX=${LOCAL_CXX} CC=${LOCAL_CC} LDFLAGS=${LOCAL_LDFLAGS} CPPFLAGS=${LOCAL_CPPFLAGS} \
    meson setup --wipe builddir/ -Dprefix="${MESA_DIR}/out" -Dglx=xlib -Dosmesa=true -Dgallium-drivers=llvmpipe,swrast -Dvulkan-drivers=swrast
-CXX=${LOCAL_CXX} CC=${LOCAL_CC} PATH=${LOCAL_PATH} LDFLAGS=${LOCAL_LDFLAGS} CPPFLAGS=${LOCAL_CPPFLAGS} \
+PKG_CONFIG_PATH=${LOCAL_PKG_CONFIG_PATH} PATH=${LOCAL_PATH} \
+CXX=${LOCAL_CXX} CC=${LOCAL_CC} LDFLAGS=${LOCAL_LDFLAGS} CPPFLAGS=${LOCAL_CPPFLAGS} \
    meson install -C builddir/
 
 # Disable python venv

build/common/upload-release-assets/package-lock.json

@@ -204,9 +204,10 @@
       "license": "Apache-2.0"
     },
     "node_modules/brace-expansion": {
-      "version": "1.1.11",
+      "version": "1.1.12",
-      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.11.tgz",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.12.tgz",
-      "integrity": "sha512-iCuPHDFgrHX7H2vEI/5xpz07zSHB00TpugqhmYtVmMO6518mCuRMoOYFldEBl0g187ufozdaHgWKcYFb61qGiA==",
+      "integrity": "sha512-9T9UjW3r0UW5c1Q7GTwllptXwhvYmEzFhzMfZ9H7FQWt+uZePjZPjBP/W1ZEyZ1twGWom5/56TF4lPcqjnDHcg==",
+      "license": "MIT",
       "dependencies": {
         "balanced-match": "^1.0.0",
         "concat-map": "0.0.1"
@@ -409,9 +410,9 @@
       "integrity": "sha512-Nik3Sc0ncrMK4UUdXQmAnRtzmNQTAAXmXIopizwZ1W1t8QmfJj+zL4OA2I7XPTPW5z5TDqv4hRo/JzouDJnX3A=="
     },
     "brace-expansion": {
-      "version": "1.1.11",
+      "version": "1.1.12",
-      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.11.tgz",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.12.tgz",
-      "integrity": "sha512-iCuPHDFgrHX7H2vEI/5xpz07zSHB00TpugqhmYtVmMO6518mCuRMoOYFldEBl0g187ufozdaHgWKcYFb61qGiA==",
+      "integrity": "sha512-9T9UjW3r0UW5c1Q7GTwllptXwhvYmEzFhzMfZ9H7FQWt+uZePjZPjBP/W1ZEyZ1twGWom5/56TF4lPcqjnDHcg==",
       "requires": {
         "balanced-match": "^1.0.0",
         "concat-map": "0.0.1"

build/windows/build-github.bat

@@ -47,7 +47,11 @@ if "%RUNNING_LOCALLY%" == "1" (
     set "PATH=%PATH%;C:\Program Files\7-Zip"
 )
 
-call "C:\Program Files (x86)\Microsoft Visual Studio\2019\%VISUAL_STUDIO_VERSION%\VC\Auxiliary\Build\vcvars64.bat"
+:: Outdated windows-2019 pattern
+:: call "C:\Program Files (x86)\Microsoft Visual Studio\2019\%VISUAL_STUDIO_VERSION%\VC\Auxiliary\Build\vcvars64.bat"
+
+call "C:\Program Files\Microsoft Visual Studio\2022\%VISUAL_STUDIO_VERSION%\VC\Auxiliary\Build\vcvars64.bat"
+echo Passed vcvars64.bat
 if errorlevel 1 exit /b %errorlevel%
 
 msbuild /version
@@ -107,7 +111,7 @@ cd out\cmake-%variant%
 if errorlevel 1 exit /b %errorlevel%
 
 cmake ..\.. ^
-    -G "Visual Studio 16 2019" ^
+    -G "Visual Studio 17 2022" ^
     -A x64 ^
     %flag% ^
     -DCMAKE_INSTALL_PREFIX=..\%variant% ^

filament/CMakeLists.txt

@@ -262,11 +262,16 @@ set(MATERIAL_SRCS
         src/materials/ssao/mipmapDepth.mat
         src/materials/ssao/sao.mat
         src/materials/ssao/saoBentNormals.mat
-        src/materials/ssao/gtao.mat
-        src/materials/ssao/gtaoBentNormals.mat
         src/materials/vsmMipmap.mat
 )
 
+if (NOT FILAMENT_DISABLE_GTAO)
+    list(APPEND MATERIAL_SRCS
+        src/materials/ssao/gtao.mat
+        src/materials/ssao/gtaoBentNormals.mat
+    )
+endif()
+
 set(MATERIAL_FL0_SRCS
         src/materials/defaultMaterial.mat
         src/materials/skybox.mat
@@ -316,6 +321,10 @@ if (FILAMENT_FORCE_PROFILING_MODE)
     add_definitions(-DFILAMENT_FORCE_PROFILING_MODE)
 endif()
 
+if (FILAMENT_DISABLE_GTAO)
+    add_definitions(-DFILAMENT_DISABLE_GTAO)
+endif()
+
 # ==================================================================================================
 # Definitions
 # ==================================================================================================
@@ -463,27 +472,29 @@ add_custom_command(
         APPEND
 )
 
-add_custom_command(
+if (NOT FILAMENT_DISABLE_GTAO)
-        OUTPUT "${MATERIAL_DIR}/gtao.filamat"
+    add_custom_command(
-        DEPENDS src/materials/ssao/ssaoUtils.fs
+            OUTPUT "${MATERIAL_DIR}/gtao.filamat"
-        DEPENDS src/materials/ssao/ssct.fs
+            DEPENDS src/materials/ssao/ssaoUtils.fs
-        DEPENDS src/materials/utils/depthUtils.fs
+            DEPENDS src/materials/ssao/ssct.fs
-        DEPENDS src/materials/utils/geometry.fs
+            DEPENDS src/materials/utils/depthUtils.fs
-        DEPENDS src/materials/ssao/gtaoImpl.fs
+            DEPENDS src/materials/utils/geometry.fs
-        DEPENDS src/materials/ssao/ssctImpl.fs
+            DEPENDS src/materials/ssao/gtaoImpl.fs
-        APPEND
+            DEPENDS src/materials/ssao/ssctImpl.fs
-)
+            APPEND
+    )
 
-add_custom_command(
+    add_custom_command(
-        OUTPUT "${MATERIAL_DIR}/gtaoBentNormals.filamat"
+            OUTPUT "${MATERIAL_DIR}/gtaoBentNormals.filamat"
-        DEPENDS src/materials/ssao/ssaoUtils.fs
+            DEPENDS src/materials/ssao/ssaoUtils.fs
-        DEPENDS src/materials/ssao/ssct.fs
+            DEPENDS src/materials/ssao/ssct.fs
-        DEPENDS src/materials/utils/depthUtils.fs
+            DEPENDS src/materials/utils/depthUtils.fs
-        DEPENDS src/materials/utils/geometry.fs
+            DEPENDS src/materials/utils/geometry.fs
-        DEPENDS src/materials/ssao/gtaoImpl.fs
+            DEPENDS src/materials/ssao/gtaoImpl.fs
-        DEPENDS src/materials/ssao/ssctImpl.fs
+            DEPENDS src/materials/ssao/ssctImpl.fs
-        APPEND
+            APPEND
-)
+    )
+endif()
 
 add_custom_command(
         OUTPUT "${MATERIAL_DIR}/bilateralBlur.filamat"
@@ -590,6 +601,10 @@ target_link_libraries(${TARGET} PUBLIC filaflat)
 target_link_libraries(${TARGET} PUBLIC filabridge)
 target_link_libraries(${TARGET} PUBLIC ibl-lite)
 
+if (FILAMENT_USE_ABSEIL_LOGGING)
+    target_link_libraries(${TARGET} PUBLIC absl::log)
+endif()
+
 if (FILAMENT_ENABLE_FGVIEWER)
     target_link_libraries(${TARGET} PUBLIC fgviewer)
     add_definitions(-DFILAMENT_ENABLE_FGVIEWER=1)

filament/README.md

@@ -92,7 +92,7 @@ Copy your platform's Makefile below into a `Makefile` inside the same directory.
 ### Linux
 
 ```make
-FILAMENT_LIBS=-lfilament -lbackend -lbluegl -lbluevk -lfilabridge -lfilaflat -lutils -lgeometry -lsmol-v -lvkshaders -libl
+FILAMENT_LIBS=-lfilament -lbackend -lbluegl -lbluevk -lfilabridge -lfilaflat -lutils -lgeometry -lsmol-v -lvkshaders -libl -labseil
 CC=clang++
 
 main: main.o
@@ -110,12 +110,13 @@ clean:
 ### macOS
 
 ```make
-FILAMENT_LIBS=-lfilament -lbackend -lbluegl -lbluevk -lfilabridge -lfilaflat -lutils -lgeometry -lsmol-v -lvkshaders -libl
+FILAMENT_LIBS=-lfilament -lbackend -lbluegl -lbluevk -lfilabridge -lfilaflat -lutils -lgeometry -lsmol-v -lvkshaders -libl -labseil
 FRAMEWORKS=-framework Cocoa -framework Metal -framework CoreVideo
 CC=clang++
+ARCH ?= $(shell uname -m)
 
 main: main.o
-	$(CC) -Llib/x86_64/ main.o $(FILAMENT_LIBS) $(FRAMEWORKS) -o main
+	$(CC) -Llib/$(ARCH)/ main.o $(FILAMENT_LIBS) $(FRAMEWORKS) -o main
 
 main.o: main.cpp
 	$(CC) -Iinclude/ -std=c++17 -c main.cpp
@@ -139,7 +140,7 @@ used to change the run-time library version.
 
 ```make
 FILAMENT_LIBS=filament.lib backend.lib bluegl.lib bluevk.lib filabridge.lib filaflat.lib \
-              utils.lib geometry.lib smol-v.lib ibl.lib vkshaders.lib
+              utils.lib geometry.lib smol-v.lib ibl.lib vkshaders.lib abseil.lib
 CC=cl.exe
 
 main.exe: main.obj

filament/backend/CMakeLists.txt

@@ -180,8 +180,11 @@ if (FILAMENT_SUPPORTS_VULKAN)
             src/vulkan/VulkanAsyncHandles.h
             src/vulkan/VulkanBlitter.cpp
             src/vulkan/VulkanBlitter.h
-            src/vulkan/VulkanBuffer.cpp
             src/vulkan/VulkanBuffer.h
+            src/vulkan/VulkanBufferCache.h
+            src/vulkan/VulkanBufferCache.cpp
+            src/vulkan/VulkanBufferProxy.h
+            src/vulkan/VulkanBufferProxy.cpp
             src/vulkan/VulkanCommands.cpp
             src/vulkan/VulkanCommands.h
             src/vulkan/VulkanConstants.h
@@ -254,16 +257,38 @@ if (FILAMENT_SUPPORTS_WEBGPU)
     list(APPEND SRCS
             include/backend/platforms/WebGPUPlatform.h
             src/webgpu/platform/WebGPUPlatform.cpp
+            src/webgpu/SpdMipmapGenerator/SpdMipmapGenerator.cpp
+            src/webgpu/WebGPUBufferBase.cpp
+            src/webgpu/WebGPUBufferBase.h
+            src/webgpu/WebGPUBufferObject.cpp
+            src/webgpu/WebGPUBufferObject.h
             src/webgpu/WebGPUConstants.h
+            src/webgpu/WebGPUDescriptorSet.cpp
+            src/webgpu/WebGPUDescriptorSet.h
+            src/webgpu/WebGPUDescriptorSetLayout.cpp
+            src/webgpu/WebGPUDescriptorSetLayout.h
             src/webgpu/WebGPUDriver.cpp
             src/webgpu/WebGPUDriver.h
-            src/webgpu/WebGPUHandles.cpp
+            src/webgpu/WebGPUIndexBuffer.cpp
-            src/webgpu/WebGPUHandles.h
+            src/webgpu/WebGPUIndexBuffer.h
             src/webgpu/WebGPUPipelineCreation.cpp
             src/webgpu/WebGPUPipelineCreation.h
+            src/webgpu/WebGPUProgram.cpp
+            src/webgpu/WebGPUProgram.h
+            src/webgpu/WebGPURenderPrimitive.h
+            src/webgpu/WebGPURenderTarget.cpp
+            src/webgpu/WebGPURenderTarget.h
+            src/webgpu/WebGPUStrings.h
             src/webgpu/WebGPUSwapChain.cpp
             src/webgpu/WebGPUSwapChain.h
-            src/webgpu/WGPUProgram.cpp
+            src/webgpu/WebGPUTexture.cpp
+            src/webgpu/WebGPUTexture.h
+            src/webgpu/WebGPUTimerQuery.cpp
+            src/webgpu/WebGPUTimerQuery.h
+            src/webgpu/WebGPUVertexBuffer.cpp
+            src/webgpu/WebGPUVertexBuffer.h
+            src/webgpu/WebGPUVertexBufferInfo.cpp
+            src/webgpu/WebGPUVertexBufferInfo.h
     )
     if (WIN32)
         list(APPEND SRCS src/webgpu/platform/WebGPUPlatformWindows.cpp)
@@ -381,6 +406,10 @@ endif()
 target_link_libraries(${TARGET} PUBLIC math)
 target_link_libraries(${TARGET} PUBLIC utils)
 
+if (FILAMENT_USE_ABSEIL_LOGGING)
+    target_link_libraries(${TARGET} PRIVATE absl::log)
+endif()
+
 # Android, iOS, and WebGL do not use bluegl.
 if(FILAMENT_SUPPORTS_OPENGL AND NOT IOS AND NOT ANDROID AND NOT WEBGL)
     target_link_libraries(${TARGET} PRIVATE bluegl)

filament/backend/include/backend/DriverEnums.h

@@ -495,18 +495,12 @@ struct DescriptorSetLayoutBinding {
     DescriptorFlags flags = DescriptorFlags::NONE;
     uint16_t count = 0;
 
-    //  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 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;
     }
 };
 
@@ -1254,26 +1248,6 @@ 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
     SamplerMagFilter filterMag      : 1;    //!< magnification filter (NEAREST)
@@ -1342,94 +1316,9 @@ 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 bool operator == (SamplerYcbcrConversion lhs, SamplerYcbcrConversion rhs)
-            noexcept {
-        return SamplerYcbcrConversion::EqualTo{}(lhs, rhs);
-    }
-    friend bool operator != (SamplerYcbcrConversion lhs, SamplerYcbcrConversion rhs)
-            noexcept {
-        return  !SamplerYcbcrConversion::EqualTo{}(lhs, rhs);
-    }
-    friend 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 {
     std::variant<utils::StaticString, utils::CString, std::monostate> label;
     utils::FixedCapacityVector<DescriptorSetLayoutBinding> bindings;
-
-//  TODO: uncomment when needed
-//    utils::FixedCapacityVector<ExternalSamplerDatum> externalSamplerData;
 };
 
 //! blending equation function

filament/backend/include/backend/Platform.h

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

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

@@ -199,7 +199,7 @@ public:
      */
     virtual bool makeCurrent(ContextType type,
             SwapChain* UTILS_NONNULL drawSwapChain,
-            SwapChain* UTILS_NONNULL readSwapChain) noexcept = 0;
+            SwapChain* UTILS_NONNULL readSwapChain) = 0;
 
     /**
      * Called by the driver to make the OpenGL context active on the calling thread and bind
@@ -219,7 +219,7 @@ public:
             SwapChain* UTILS_NONNULL drawSwapChain,
             SwapChain* UTILS_NONNULL readSwapChain,
             utils::Invocable<void()> preContextChange,
-            utils::Invocable<void(size_t index)> postContextChange) noexcept;
+            utils::Invocable<void(size_t index)> postContextChange);
 
     /**
      * Called by the backend just before calling commit()

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

@@ -58,7 +58,7 @@ protected:
     SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
     SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
     void destroySwapChain(SwapChain* swapChain) noexcept override;
-    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
+    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) override;
     void commit(SwapChain* swapChain) noexcept override;
     ExternalTexture* createExternalImageTexture() noexcept override;
     void destroyExternalImageTexture(ExternalTexture* texture) noexcept override;

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

@@ -55,7 +55,7 @@ public:
     SwapChain* createSwapChain(void* nativewindow, uint64_t flags) noexcept override;
     SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
     void destroySwapChain(SwapChain* swapChain) noexcept override;
-    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept override;
+    bool makeCurrent(ContextType type, SwapChain* drawSwapChain, SwapChain* readSwapChain) override;
     void commit(SwapChain* swapChain) noexcept override;
 
     ExternalTexture* createExternalImageTexture() noexcept override;

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

@@ -109,11 +109,11 @@ protected:
 
     bool makeCurrent(ContextType type,
             SwapChain* drawSwapChain,
-            SwapChain* readSwapChain) noexcept override;
+            SwapChain* readSwapChain) override;
 
     void makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain,
             utils::Invocable<void()> preContextChange,
-            utils::Invocable<void(size_t index)> postContextChange) noexcept override;
+            utils::Invocable<void(size_t index)> postContextChange) override;
 
     void commit(SwapChain* swapChain) noexcept override;
 
@@ -128,7 +128,7 @@ protected:
     bool setExternalImage(ExternalImageHandleRef externalImage, ExternalTexture* texture) noexcept override;
 
     /**
-     * Logs glGetError() to slog.e
+     * Logs glGetError() to LOG(ERROR)
      * @param name a string giving some context on the error. Typically __func__.
      */
     static void logEglError(const char* name) noexcept;
@@ -148,12 +148,12 @@ protected:
     EGLContext getContextForType(ContextType type) const noexcept;
 
     // makes the draw and read surface current without changing the current context
-    EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) noexcept {
+    EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) {
         return egl.makeCurrent(drawSurface, readSurface);
     }
 
     // makes context current and set draw and read surfaces to EGL_NO_SURFACE
-    EGLBoolean makeCurrent(EGLContext context) noexcept {
+    EGLBoolean makeCurrent(EGLContext context) {
         return egl.makeCurrent(context, mEGLDummySurface, mEGLDummySurface);
     }
 
@@ -211,9 +211,9 @@ private:
     public:
         explicit EGL(EGLDisplay& dpy) : mEGLDisplay(dpy) {}
         EGLBoolean makeCurrent(EGLContext context,
-                EGLSurface drawSurface, EGLSurface readSurface) noexcept;
+                EGLSurface drawSurface, EGLSurface readSurface);
 
-        EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) noexcept {
+        EGLBoolean makeCurrent(EGLSurface drawSurface, EGLSurface readSurface) {
             return makeCurrent(mCurrentContext, drawSurface, readSurface);
         }
     } egl{ mEGLDisplay };

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

@@ -127,7 +127,7 @@ protected:
 protected:
     bool makeCurrent(ContextType type,
             SwapChain* drawSwapChain,
-            SwapChain* readSwapChain) noexcept override;
+            SwapChain* readSwapChain) override;
 
 private:
     struct InitializeJvmForPerformanceManagerIfNeeded {

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

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

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

@@ -56,7 +56,7 @@ protected:
     SwapChain* createSwapChain(uint32_t width, uint32_t height, uint64_t flags) noexcept override;
     void destroySwapChain(SwapChain* swapChain) noexcept override;
     bool makeCurrent(ContextType type, SwapChain* drawSwapChain,
-            SwapChain* readSwapChain) noexcept override;
+            SwapChain* readSwapChain) override;
     void commit(SwapChain* swapChain) noexcept override;
 
 private:

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

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

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

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

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

@@ -22,6 +22,7 @@
 #include <webgpu/webgpu_cpp.h>
 
 #include <cstdint>
+#include <vector>
 
 namespace filament::backend {
 
@@ -56,6 +57,10 @@ protected:
             const Platform::DriverConfig& driverConfig) noexcept override;
 
 private:
+    // returns adapter request option variations applicable for the particular
+    // platform
+    [[nodiscard]] static std::vector<wgpu::RequestAdapterOptions> getAdapterOptions();
+
     // we may consider having the driver own this in the future
     wgpu::Instance mInstance;
 };

filament/backend/include/private/backend/CircularBuffer.h

@@ -72,7 +72,7 @@ public:
     Range getBuffer() noexcept;
 
 private:
-    void* alloc(size_t size) noexcept;
+    void* alloc(size_t size);
     void dealloc() noexcept;
 
     // pointer to the beginning of the circular buffer (constant)

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

@@ -76,7 +76,7 @@ public:
 
     // all commands buffers (Slices) written to this point are returned by waitForCommand(). This
     // call blocks until the CircularBuffer has at least mRequiredSize bytes available.
-    void flush() noexcept;
+    void flush();
 
     // returns from waitForCommands() immediately.
     void requestExit();

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

@@ -39,7 +39,7 @@
 #define FILAMENT_DEBUG_COMMANDS_NONE         0x0
 // Command debugging enabled. No logging by default.
 #define FILAMENT_DEBUG_COMMANDS_ENABLE       0x1
-// Command debugging enabled. Every command logged to slog.d
+// Command debugging enabled. Every command logged to DLOG(INFO)
 #define FILAMENT_DEBUG_COMMANDS_LOG          0x2
 // Command debugging enabled. Every command logged to systrace
 #define FILAMENT_DEBUG_COMMANDS_SYSTRACE     0x4

filament/backend/include/private/backend/VirtualMachineEnv.h

@@ -27,13 +27,13 @@ namespace filament {
 class VirtualMachineEnv {
 public:
     // must be called before VirtualMachineEnv::get() from a thread that is attached to the JavaVM
-    static jint JNI_OnLoad(JavaVM* vm) noexcept;
+    static jint JNI_OnLoad(JavaVM* vm);
 
     // must be called on backend thread
     static VirtualMachineEnv& get() noexcept;
 
     // can be called from any thread that already has a JniEnv
-    static JNIEnv* getThreadEnvironment() noexcept;
+    static JNIEnv* getThreadEnvironment();
 
     // must be called from the backend thread
     JNIEnv* getEnvironment() noexcept {
@@ -49,7 +49,7 @@ public:
 private:
     explicit VirtualMachineEnv(JavaVM* vm) noexcept;
     ~VirtualMachineEnv() noexcept;
-    JNIEnv* getEnvironmentSlow() noexcept;
+    JNIEnv* getEnvironmentSlow();
 
     static utils::Mutex sLock;
     static JavaVM* sVirtualMachine;

filament/backend/src/CircularBuffer.cpp

@@ -16,7 +16,7 @@
 
 #include "private/backend/CircularBuffer.h"
 
-#include <utils/Log.h>
+#include <utils/Logger.h>
 #include <utils/Panic.h>
 #include <utils/architecture.h>
 #include <utils/ashmem.h>
@@ -65,7 +65,7 @@ CircularBuffer::~CircularBuffer() noexcept {
 // to each others and a special case in circularize()
 
 UTILS_NOINLINE
-void* CircularBuffer::alloc(size_t size) noexcept {
+void* CircularBuffer::alloc(size_t size) {
 #if HAS_MMAP
     void* data = nullptr;
     void* vaddr = MAP_FAILED;
@@ -127,7 +127,7 @@ void* CircularBuffer::alloc(size_t size) noexcept {
                 "couldn't allocate " << (size * 2 / 1024) <<
                 " KiB of virtual address space for the command buffer";
 
-        slog.w << "Using 'soft' CircularBuffer (" << (size * 2 / 1024) << " KiB)" << io::endl;
+        LOG(WARNING) << "Using 'soft' CircularBuffer (" << (size * 2 / 1024) << " KiB)";
 
         // guard page at the end
         void* guard = (void*)(uintptr_t(data) + size * 2);

filament/backend/src/CommandBufferQueue.cpp

@@ -20,12 +20,12 @@
 
 #include <private/utils/Tracing.h>
 
-#include <utils/compiler.h>
+#include <utils/Logger.h>
-#include <utils/Log.h>
 #include <utils/Mutex.h>
-#include <utils/ostream.h>
 #include <utils/Panic.h>
+#include <utils/compiler.h>
 #include <utils/debug.h>
+#include <utils/ostream.h>
 
 #include <algorithm>
 #include <mutex>
@@ -53,18 +53,18 @@ CommandBufferQueue::~CommandBufferQueue() {
 }
 
 void CommandBufferQueue::requestExit() {
-    std::lock_guard<utils::Mutex> const lock(mLock);
+    std::lock_guard const lock(mLock);
     mExitRequested = EXIT_REQUESTED;
     mCondition.notify_one();
 }
 
 bool CommandBufferQueue::isPaused() const noexcept {
-    std::lock_guard<utils::Mutex> const lock(mLock);
+    std::lock_guard const lock(mLock);
     return mPaused;
 }
 
 void CommandBufferQueue::setPaused(bool paused) {
-    std::lock_guard<utils::Mutex> const lock(mLock);
+    std::lock_guard const lock(mLock);
     if (paused) {
         mPaused = true;
     } else {
@@ -74,12 +74,12 @@ void CommandBufferQueue::setPaused(bool paused) {
 }
 
 bool CommandBufferQueue::isExitRequested() const {
-    std::lock_guard<utils::Mutex> const lock(mLock);
+    std::lock_guard const lock(mLock);
-    return (bool)mExitRequested;
+    return bool(mExitRequested);
 }
 
 
-void CommandBufferQueue::flush() noexcept {
+void CommandBufferQueue::flush() {
     FILAMENT_TRACING_CALL(FILAMENT_TRACING_CATEGORY_FILAMENT);
 
     CircularBuffer& circularBuffer = mCircularBuffer;
@@ -103,7 +103,7 @@ void CommandBufferQueue::flush() noexcept {
             static_cast<char const*>(begin), static_cast<char const*>(end));
 
 
-    std::unique_lock<utils::Mutex> lock(mLock);
+    std::unique_lock lock(mLock);
 
     // circular buffer is too small, we corrupted the stream
     FILAMENT_CHECK_POSTCONDITION(used <= mFreeSpace) <<
@@ -121,11 +121,10 @@ void CommandBufferQueue::flush() noexcept {
 
 #ifndef NDEBUG
         size_t const totalUsed = circularBuffer.size() - mFreeSpace;
-        slog.d << "CommandStream used too much space (will block): "
+        DLOG(INFO) << "CommandStream used too much space (will block): "
-                << "needed space " << requiredSize << " out of " << mFreeSpace
+                   << "needed space " << requiredSize << " out of " << mFreeSpace
-                << ", totalUsed=" << totalUsed << ", current=" << used
+                   << ", totalUsed=" << totalUsed << ", current=" << used
-                << ", queue size=" << mCommandBuffersToExecute.size() << " buffers"
+                   << ", queue size=" << mCommandBuffersToExecute.size() << " buffers";
-                << io::endl;
 
         mHighWatermark = std::max(mHighWatermark, totalUsed);
 #endif
@@ -147,7 +146,7 @@ std::vector<CommandBufferQueue::Range> CommandBufferQueue::waitForCommands() con
     if (!UTILS_HAS_THREADING) {
         return std::move(mCommandBuffersToExecute);
     }
-    std::unique_lock<utils::Mutex> lock(mLock);
+    std::unique_lock lock(mLock);
     while ((mCommandBuffersToExecute.empty() || mPaused) && !mExitRequested) {
         mCondition.wait(lock);
     }
@@ -157,7 +156,7 @@ std::vector<CommandBufferQueue::Range> CommandBufferQueue::waitForCommands() con
 void CommandBufferQueue::releaseBuffer(CommandBufferQueue::Range const& buffer) {
     size_t const used = std::distance(
             static_cast<char const*>(buffer.begin), static_cast<char const*>(buffer.end));
-    std::lock_guard<utils::Mutex> const lock(mLock);
+    std::lock_guard const lock(mLock);
     mFreeSpace += used;
     mCondition.notify_one();
 }

filament/backend/src/CommandStream.cpp

@@ -22,10 +22,13 @@
 #include <utils/CallStack.h>
 #endif
 
-#include <utils/compiler.h>
+#include <private/utils/Tracing.h>
-#include <utils/Log.h>
+
-#include <utils/ostream.h>
+#include <utils/Logger.h>
 #include <utils/Profiler.h>
+#include <utils/compiler.h>
+#include <utils/ostream.h>
+#include <utils/sstream.h>
 
 #include <cstddef>
 #include <functional>
@@ -129,9 +132,10 @@ void CommandType<void (Driver::*)(ARGS...)>::Command<METHOD>::log(std::index_seq
 #if DEBUG_COMMAND_STREAM
     static_assert(UTILS_HAS_RTTI, "DEBUG_COMMAND_STREAM can only be used with RTTI");
     std::string command = utils::CallStack::demangleTypeName(typeid(Command).name()).c_str();
-    slog.d << extractMethodName(command) << " : size=" << sizeof(Command) << "\n\t";
+    DLOG(INFO) << extractMethodName(command) << " : size=" << sizeof(Command);
-    printParameterPack(slog.d, std::get<I>(mArgs)...);
+    utils::io::sstream parameterPack;
-    slog.d << io::endl;
+    printParameterPack(parameterPack, std::get<I>(mArgs)...);
+    DLOG(INFO) << "\t" << parameterPack.c_str();
 #endif
 }
 

filament/backend/src/Driver.cpp

@@ -25,9 +25,9 @@
 
 #include <private/utils/Tracing.h>
 
+#include <utils/Logger.h>
 #include <utils/compiler.h>
 #include <utils/debug.h>
-#include <utils/Log.h>
 #include <utils/ostream.h>
 
 #include <math/half.h>
@@ -149,7 +149,7 @@ void DriverBase::scheduleRelease(AcquiredImage const& image) noexcept {
 void DriverBase::debugCommandBegin(CommandStream* cmds, bool synchronous, const char* methodName) noexcept {
     if constexpr (bool(FILAMENT_DEBUG_COMMANDS > FILAMENT_DEBUG_COMMANDS_NONE)) {
         if constexpr (bool(FILAMENT_DEBUG_COMMANDS & FILAMENT_DEBUG_COMMANDS_LOG)) {
-            utils::slog.d << methodName << utils::io::endl;
+            DLOG(INFO) << methodName;
         }
         if constexpr (bool(FILAMENT_DEBUG_COMMANDS & FILAMENT_DEBUG_COMMANDS_SYSTRACE)) {
             FILAMENT_TRACING_CONTEXT(FILAMENT_TRACING_CATEGORY_FILAMENT);

filament/backend/src/HandleAllocator.cpp

@@ -20,7 +20,7 @@
 
 #include <utils/Allocator.h>
 #include <utils/CString.h>
-#include <utils/Log.h>
+#include <utils/Logger.h>
 #include <utils/Panic.h>
 #include <utils/compiler.h>
 #include <utils/debug.h>
@@ -57,8 +57,8 @@ HandleAllocator<P0, P1, P2>::Allocator::Allocator(AreaPolicy::HeapArea const& ar
     size_t const maxHeapSize = std::min(area.size(), HANDLE_INDEX_MASK * getAlignment());
 
     if (UTILS_UNLIKELY(maxHeapSize != area.size())) {
-        slog.w << "HandleAllocator heap size reduced to "
+        LOG(WARNING) << "HandleAllocator heap size reduced to " << maxHeapSize << " from "
-               << maxHeapSize << " from " << area.size() << io::endl;
+                     << area.size();
     }
 
     // make sure we start with a clean arena. This is needed to ensure that all blocks start

filament/backend/src/VirtualMachineEnv.cpp

@@ -50,7 +50,7 @@ JavaVM* VirtualMachineEnv::getVirtualMachine() {
  */
 UTILS_PUBLIC
 UTILS_NOINLINE
-jint VirtualMachineEnv::JNI_OnLoad(JavaVM* vm) noexcept {
+jint VirtualMachineEnv::JNI_OnLoad(JavaVM* vm) {
     std::lock_guard const lock(sLock);
     if (sVirtualMachine) {
         // It doesn't make sense for JNI_OnLoad() to be called more than once
@@ -77,7 +77,7 @@ VirtualMachineEnv& VirtualMachineEnv::get() noexcept {
 }
 
 UTILS_NOINLINE
-JNIEnv* VirtualMachineEnv::getThreadEnvironment() noexcept {
+JNIEnv* VirtualMachineEnv::getThreadEnvironment() {
     JavaVM* const vm = getVirtualMachine();
     JNIEnv* env = nullptr;
     jint const result = vm->GetEnv(reinterpret_cast<void**>(&env), JNI_VERSION_1_6);
@@ -101,7 +101,7 @@ VirtualMachineEnv::~VirtualMachineEnv() noexcept {
 }
 
 UTILS_NOINLINE
-JNIEnv* VirtualMachineEnv::getEnvironmentSlow() noexcept {
+JNIEnv* VirtualMachineEnv::getEnvironmentSlow() {
     FILAMENT_CHECK_PRECONDITION(mVirtualMachine)
             << "JNI_OnLoad() has not been called";
 

filament/backend/src/metal/MetalBlitter.mm

@@ -19,8 +19,8 @@
 #include "MetalContext.h"
 #include "MetalUtils.h"
 
+#include <utils/Logger.h>
 #include <utils/Panic.h>
-#include <utils/Log.h>
 
 namespace filament::backend {
 
@@ -323,7 +323,7 @@ id<MTLFunction> MetalBlitter::compileFragmentFunction(BlitFunctionKey key) const
     if (!library || !function) {
         if (error) {
             auto description = [error.localizedDescription cStringUsingEncoding:NSUTF8StringEncoding];
-            utils::slog.e << description << utils::io::endl;
+            LOG(ERROR) << description;
         }
     }
     FILAMENT_CHECK_POSTCONDITION(library && function)
@@ -349,7 +349,7 @@ id<MTLFunction> MetalBlitter::getBlitVertexFunction() {
     if (!library || !function) {
         if (error) {
             auto description = [error.localizedDescription cStringUsingEncoding:NSUTF8StringEncoding];
-            utils::slog.e << description << utils::io::endl;
+            LOG(ERROR) << description;
         }
     }
     FILAMENT_CHECK_POSTCONDITION(library && function)

filament/backend/src/metal/MetalBufferPool.mm

@@ -18,8 +18,8 @@
 
 #include "MetalContext.h"
 
+#include <utils/Logger.h>
 #include <utils/Panic.h>
-#include <utils/Log.h>
 #include <utils/trap.h>
 
 #include <thread>
@@ -77,8 +77,7 @@ void MetalBufferPool::releaseBuffer(MetalBufferPoolEntry const *stage) noexcept
 
     auto iter = mUsedStages.find(stage);
     if (iter == mUsedStages.end()) {
-        utils::slog.e << "Unknown Metal buffer: " << stage->capacity << " bytes"
+        LOG(ERROR) << "Unknown Metal buffer: " << stage->capacity << " bytes";
-                << utils::io::endl;
         return;
     }
     stage->lastAccessed = mCurrentFrame;

filament/backend/src/metal/MetalContext.mm

@@ -18,8 +18,9 @@
 
 #include "MetalHandles.h"
 
-#include <utils/debug.h>
 #include <utils/FixedCapacityVector.h>
+#include <utils/Logger.h>
+#include <utils/debug.h>
 
 #include <utility>
 
@@ -112,8 +113,8 @@ id<MTLCommandBuffer> getPendingCommandBuffer(MetalContext* context) {
         auto errorCode = (MTLCommandBufferError)buffer.error.code;
         if (@available(macOS 11.0, *)) {
             if (errorCode == MTLCommandBufferErrorMemoryless) {
-                utils::slog.w << "Metal: memoryless geometry limit reached. "
+                LOG(WARNING) << "Metal: memoryless geometry limit reached. Continuing with private "
-                        "Continuing with private storage mode." << utils::io::endl;
+                                "storage mode.";
                 context->memorylessLimitsReached = true;
             }
         }

filament/backend/src/metal/MetalDriver.mm

@@ -38,10 +38,10 @@
 #include <Metal/Metal.h>
 #include <QuartzCore/QuartzCore.h>
 
-#include <utils/Log.h>
+#include <utils/Invocable.h>
+#include <utils/Logger.h>
 #include <utils/Panic.h>
 #include <utils/sstream.h>
-#include <utils/Invocable.h>
 
 #include <algorithm>
 
@@ -78,20 +78,18 @@ Driver* MetalDriverFactory::create(PlatformMetal* const platform, const Platform
     //    MetalVertexBufferInfo        : 552       moderate
     // -- less than or equal to 552 bytes
 
-    utils::slog.d
+    DLOG(INFO) << "MetalSwapChain: " << sizeof(MetalSwapChain);
-           << "\nMetalSwapChain: " << sizeof(MetalSwapChain)
+    DLOG(INFO) << "MetalBufferObject: " << sizeof(MetalBufferObject);
-           << "\nMetalBufferObject: " << sizeof(MetalBufferObject)
+    DLOG(INFO) << "MetalVertexBuffer: " << sizeof(MetalVertexBuffer);
-           << "\nMetalVertexBuffer: " << sizeof(MetalVertexBuffer)
+    DLOG(INFO) << "MetalVertexBufferInfo: " << sizeof(MetalVertexBufferInfo);
-           << "\nMetalVertexBufferInfo: " << sizeof(MetalVertexBufferInfo)
+    DLOG(INFO) << "MetalIndexBuffer: " << sizeof(MetalIndexBuffer);
-           << "\nMetalIndexBuffer: " << sizeof(MetalIndexBuffer)
+    DLOG(INFO) << "MetalRenderPrimitive: " << sizeof(MetalRenderPrimitive);
-           << "\nMetalRenderPrimitive: " << sizeof(MetalRenderPrimitive)
+    DLOG(INFO) << "MetalTexture: " << sizeof(MetalTexture);
-           << "\nMetalTexture: " << sizeof(MetalTexture)
+    DLOG(INFO) << "MetalTimerQuery: " << sizeof(MetalTimerQuery);
-           << "\nMetalTimerQuery: " << sizeof(MetalTimerQuery)
+    DLOG(INFO) << "HwStream: " << sizeof(HwStream);
-           << "\nHwStream: " << sizeof(HwStream)
+    DLOG(INFO) << "MetalRenderTarget: " << sizeof(MetalRenderTarget);
-           << "\nMetalRenderTarget: " << sizeof(MetalRenderTarget)
+    DLOG(INFO) << "MetalFence: " << sizeof(MetalFence);
-           << "\nMetalFence: " << sizeof(MetalFence)
+    DLOG(INFO) << "MetalProgram: " << sizeof(MetalProgram);
-           << "\nMetalProgram: " << sizeof(MetalProgram)
-           << utils::io::endl;
 #endif
     return MetalDriver::create(platform, driverConfig);
 }
@@ -135,19 +133,18 @@ MetalDriver::MetalDriver(
 
     initializeSupportedGpuFamilies(mContext);
 
-    utils::slog.v << "Supported GPU families: " << utils::io::endl;
+    LOG(INFO) << "Supported GPU families: ";
     if (mContext->highestSupportedGpuFamily.common > 0) {
-        utils::slog.v << "  MTLGPUFamilyCommon" << (int) mContext->highestSupportedGpuFamily.common << utils::io::endl;
+        LOG(INFO) << "  MTLGPUFamilyCommon" << (int) mContext->highestSupportedGpuFamily.common;
     }
     if (mContext->highestSupportedGpuFamily.apple > 0) {
-        utils::slog.v << "  MTLGPUFamilyApple" << (int) mContext->highestSupportedGpuFamily.apple << utils::io::endl;
+        LOG(INFO) << "  MTLGPUFamilyApple" << (int) mContext->highestSupportedGpuFamily.apple;
     }
     if (mContext->highestSupportedGpuFamily.mac > 0) {
-        utils::slog.v << "  MTLGPUFamilyMac" << (int) mContext->highestSupportedGpuFamily.mac << utils::io::endl;
+        LOG(INFO) << "  MTLGPUFamilyMac" << (int) mContext->highestSupportedGpuFamily.mac;
     }
-    utils::slog.v << "Features:" << utils::io::endl;
+    LOG(INFO) << "Features:";
-    utils::slog.v << "  readWriteTextureSupport: " <<
+    LOG(INFO) << "  readWriteTextureSupport: " << (bool) mContext->device.readWriteTextureSupport;
-            (bool) mContext->device.readWriteTextureSupport << utils::io::endl;
 
     // In order to support texture swizzling, the GPU needs to support it and the system be running
     // iOS 13+.
@@ -588,7 +585,7 @@ void MetalDriver::createProgramR(Handle<HwProgram> rph, Program&& program) {
 #if FILAMENT_METAL_DEBUG_LOG
     auto handleId = rph.getId();
     DEBUG_LOG("createProgramR(rph = %d, program = ", handleId);
-    utils::slog.d << program << utils::io::endl;
+    DLOG(INFO) << program;
 #endif
     construct_handle<MetalProgram>(rph, *mContext, std::move(program));
 }
@@ -2077,7 +2074,7 @@ void MetalDriver::dispatchCompute(Handle<HwProgram> program, math::uint3 workGro
                                                             error:&error];
     if (error) {
         auto description = [error.localizedDescription cStringUsingEncoding:NSUTF8StringEncoding];
-        utils::slog.e << description << utils::io::endl;
+        LOG(ERROR) << description;
     }
     assert_invariant(!error);
 

filament/backend/src/metal/MetalExternalImage.mm

@@ -20,14 +20,14 @@
 #include "MetalEnums.h"
 #include "MetalUtils.h"
 
+#include <utils/Logger.h>
 #include <utils/Panic.h>
-#include <utils/Log.h>
 #include <utils/trap.h>
 
 #define NSERROR_CHECK(message)                                                                     \
     if (error) {                                                                                   \
         auto description = [error.localizedDescription cStringUsingEncoding:NSUTF8StringEncoding]; \
-        utils::slog.e << description << utils::io::endl;                                           \
+        LOG(ERROR) << description;                                                                 \
     }                                                                                              \
     FILAMENT_CHECK_POSTCONDITION(error == nil) << message;
 

filament/backend/src/metal/MetalHandles.mm

@@ -27,10 +27,11 @@
 
 #include "private/backend/BackendUtils.h"
 
-#include <utils/compiler.h>
+#include <utils/Logger.h>
 #include <utils/Panic.h>
-#include <utils/trap.h>
+#include <utils/compiler.h>
 #include <utils/debug.h>
+#include <utils/trap.h>
 
 #include <math/scalar.h>
 
@@ -115,14 +116,14 @@ MetalSwapChain::MetalSwapChain(
       type(SwapChainType::CAMETALLAYER) {
 
     if (!(flags & SwapChain::CONFIG_TRANSPARENT) && !nativeWindow.opaque) {
-        utils::slog.w << "Warning: Filament SwapChain has no CONFIG_TRANSPARENT flag, "
+        LOG(WARNING) << "Warning: Filament SwapChain has no CONFIG_TRANSPARENT flag, but the "
-                         "but the CAMetaLayer(" << (__bridge void*) nativeWindow << ")"
+                        "CAMetaLayer("
-                         " has .opaque set to NO." << utils::io::endl;
+                     << (__bridge void*) nativeWindow << ") has .opaque set to NO.";
     }
     if ((flags & SwapChain::CONFIG_TRANSPARENT) && nativeWindow.opaque) {
-        utils::slog.w << "Warning: Filament SwapChain has the CONFIG_TRANSPARENT flag, "
+        LOG(WARNING) << "Warning: Filament SwapChain has the CONFIG_TRANSPARENT flag, but the "
-                         "but the CAMetaLayer(" << (__bridge void*) nativeWindow << ")"
+                        "CAMetaLayer("
-                         " has .opaque set to YES." << utils::io::endl;
+                     << (__bridge void*) nativeWindow << ") has .opaque set to YES.";
     }
 
     // Needed so we can use the SwapChain as a blit source.

filament/backend/src/metal/MetalShaderCompiler.mm

@@ -21,6 +21,7 @@
 #include <backend/Program.h>
 
 #include <utils/JobSystem.h>
+#include <utils/Logger.h>
 #include <utils/Mutex.h>
 
 #include <chrono>
@@ -147,7 +148,7 @@ bool MetalShaderCompiler::isParallelShaderCompileSupported() const noexcept {
             if (error) {
                 auto description =
                         [error.localizedDescription cStringUsingEncoding:NSUTF8StringEncoding];
-                utils::slog.w << description << utils::io::endl;
+                LOG(WARNING) << description;
                 errorMessage = error.localizedDescription;
             }
             PANIC_LOG("Failed to compile Metal program.");
@@ -178,7 +179,7 @@ bool MetalShaderCompiler::isParallelShaderCompileSupported() const noexcept {
             if (error) {
                 auto description =
                         [error.localizedDescription cStringUsingEncoding:NSUTF8StringEncoding];
-                utils::slog.w << description << utils::io::endl;
+                LOG(WARNING) << description;
                 errorMessage = error.localizedDescription;
             }
             PANIC_LOG("Failed to load main0 in Metal program.");

filament/backend/src/metal/MetalState.mm

@@ -18,7 +18,7 @@
 
 #include "MetalEnums.h"
 
-#include <utils/Log.h>
+#include <utils/Logger.h>
 
 namespace filament {
 namespace backend {
@@ -95,7 +95,7 @@ id<MTLRenderPipelineState> PipelineStateCreator::operator()(id<MTLDevice> device
             [NSString stringWithFormat:@"Could not create Metal pipeline state: %@",
                 error ? error.localizedDescription : @"unknown error"];
         auto description = [errorMessage cStringUsingEncoding:NSUTF8StringEncoding];
-        utils::slog.e << description << utils::io::endl;
+        LOG(ERROR) << description;
         [[NSException exceptionWithName:@"MetalRenderPipelineFailure"
                                  reason:errorMessage
                                userInfo:nil] raise];
@@ -155,7 +155,7 @@ id<MTLSamplerState> SamplerStateCreator::operator()(id<MTLDevice> device,
     // MTLSamplerDescriptor.
     // In practice, this means shadows are not supported when running in the simulator.
     if (![device supportsFeatureSet:MTLFeatureSet_iOS_GPUFamily3_v1]) {
-        utils::slog.w << "Warning: sample comparison not supported by this GPU" << utils::io::endl;
+        LOG(WARNING) << "Warning: sample comparison not supported by this GPU";
         samplerDescriptor.compareFunction = MTLCompareFunctionNever;
     }
 #endif

filament/backend/src/metal/PlatformMetal.mm

@@ -19,7 +19,7 @@
 
 #include "MetalDriverFactory.h"
 
-#include <utils/Log.h>
+#include <utils/Logger.h>
 
 #import <Foundation/Foundation.h>
 
@@ -53,6 +53,9 @@ PlatformMetal::~PlatformMetal() noexcept {
 }
 
 Driver* PlatformMetal::createDriver(void* /*sharedContext*/, const Platform::DriverConfig& driverConfig) noexcept {
+    pImpl->mDrawableFailureBehavior = driverConfig.metalDisablePanicOnDrawableFailure
+            ? DrawableFailureBehavior::ABORT_FRAME
+            : DrawableFailureBehavior::PANIC;
     return MetalDriverFactory::create(this, driverConfig);
 }
 
@@ -129,9 +132,8 @@ void PlatformMetalImpl::createDeviceImpl(MetalDevice& outDevice) {
         result = MTLCreateSystemDefaultDevice();
     }
 
-    utils::slog.i << "Selected physical device '"
+    LOG(INFO) << "Selected physical device '"
-                  << [result.name cStringUsingEncoding:NSUTF8StringEncoding] << "'"
+              << [result.name cStringUsingEncoding:NSUTF8StringEncoding] << "'";
-                  << utils::io::endl;
 
     outDevice.device = result;
     mDevice = result;

filament/backend/src/opengl/GLDescriptorSet.cpp

@@ -162,8 +162,11 @@ void GLDescriptorSet::update(OpenGLContext&,
     }, descriptors[binding].desc);
 }
 
-void GLDescriptorSet::update(OpenGLContext& gl,
+void GLDescriptorSet::update(OpenGLContext& gl, HandleAllocatorGL& handleAllocator,
-        descriptor_binding_t binding, GLTexture* t, SamplerParams params) noexcept {
+        descriptor_binding_t binding, TextureHandle th, SamplerParams params) noexcept {
+
+    GLTexture* t = th ? handleAllocator.handle_cast<GLTexture*>(th) : nullptr;
+
     assert_invariant(binding < descriptors.size());
     std::visit([=, &gl](auto&& arg) mutable {
         using T = std::decay_t<decltype(arg)>;
@@ -196,20 +199,12 @@ void GLDescriptorSet::update(OpenGLContext& gl,
                 }
             }
 
-            arg.target = t ? t->gl.target : 0;
+            arg.handle = th;
-            arg.id = t ? t->gl.id : 0;
-            arg.external = t ? t->gl.external :  false;
             if constexpr (std::is_same_v<T, Sampler> ||
                           std::is_same_v<T, SamplerWithAnisotropyWorkaround>) {
                 if constexpr (std::is_same_v<T, SamplerWithAnisotropyWorkaround>) {
                     arg.anisotropy = float(1u << params.anisotropyLog2);
                 }
-                if (t) {
-                    arg.ref = t->ref;
-                    arg.baseLevel = t->gl.baseLevel;
-                    arg.maxLevel = t->gl.maxLevel;
-                    arg.swizzle = t->gl.swizzle;
-                }
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
                 arg.sampler = gl.getSampler(params);
 #else
@@ -225,39 +220,39 @@ void GLDescriptorSet::update(OpenGLContext& gl,
     }, descriptors[binding].desc);
 }
 
-template<typename T>
 void GLDescriptorSet::updateTextureView(OpenGLContext& gl,
-        HandleAllocatorGL& handleAllocator, GLuint unit, T const& desc) noexcept {
+        HandleAllocatorGL& handleAllocator, GLuint unit, GLTexture const* t) noexcept {
     // The common case is that we don't have a ref handle (we only have one if
     // the texture ever had a View on it).
-    assert_invariant(desc.ref);
+    assert_invariant(t);
-    GLTextureRef* const ref = handleAllocator.handle_cast<GLTextureRef*>(desc.ref);
+    assert_invariant(t->ref);
-    if (UTILS_UNLIKELY((desc.baseLevel != ref->baseLevel || desc.maxLevel != ref->maxLevel))) {
+    GLTextureRef* const ref = handleAllocator.handle_cast<GLTextureRef*>(t->ref);
+    if (UTILS_UNLIKELY((t->gl.baseLevel != ref->baseLevel || t->gl.maxLevel != ref->maxLevel))) {
         // If we have views, then it's still uncommon that we'll switch often
         // handle the case where we reset to the original texture
-        GLint baseLevel = GLint(desc.baseLevel); // NOLINT(*-signed-char-misuse)
+        GLint baseLevel = GLint(t->gl.baseLevel); // NOLINT(*-signed-char-misuse)
-        GLint maxLevel = GLint(desc.maxLevel); // NOLINT(*-signed-char-misuse)
+        GLint maxLevel = GLint(t->gl.maxLevel); // NOLINT(*-signed-char-misuse)
         if (baseLevel > maxLevel) {
             baseLevel = 0;
             maxLevel = 1000; // per OpenGL spec
         }
         // that is very unfortunate that we have to call activeTexture here
         gl.activeTexture(unit);
-        glTexParameteri(desc.target, GL_TEXTURE_BASE_LEVEL, baseLevel);
+        glTexParameteri(t->gl.target, GL_TEXTURE_BASE_LEVEL, baseLevel);
-        glTexParameteri(desc.target, GL_TEXTURE_MAX_LEVEL,  maxLevel);
+        glTexParameteri(t->gl.target, GL_TEXTURE_MAX_LEVEL,  maxLevel);
-        ref->baseLevel = desc.baseLevel;
+        ref->baseLevel = t->gl.baseLevel;
-        ref->maxLevel = desc.maxLevel;
+        ref->maxLevel = t->gl.maxLevel;
     }
-    if (UTILS_UNLIKELY(desc.swizzle != ref->swizzle)) {
+    if (UTILS_UNLIKELY(t->gl.swizzle != ref->swizzle)) {
         using namespace GLUtils;
         gl.activeTexture(unit);
 #if !defined(__EMSCRIPTEN__)  && !defined(FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2)
-        glTexParameteri(desc.target, GL_TEXTURE_SWIZZLE_R, (GLint)getSwizzleChannel(desc.swizzle[0]));
+        glTexParameteri(t->gl.target, GL_TEXTURE_SWIZZLE_R, (GLint)getSwizzleChannel(t->gl.swizzle[0]));
-        glTexParameteri(desc.target, GL_TEXTURE_SWIZZLE_G, (GLint)getSwizzleChannel(desc.swizzle[1]));
+        glTexParameteri(t->gl.target, GL_TEXTURE_SWIZZLE_G, (GLint)getSwizzleChannel(t->gl.swizzle[1]));
-        glTexParameteri(desc.target, GL_TEXTURE_SWIZZLE_B, (GLint)getSwizzleChannel(desc.swizzle[2]));
+        glTexParameteri(t->gl.target, GL_TEXTURE_SWIZZLE_B, (GLint)getSwizzleChannel(t->gl.swizzle[2]));
-        glTexParameteri(desc.target, GL_TEXTURE_SWIZZLE_A, (GLint)getSwizzleChannel(desc.swizzle[3]));
+        glTexParameteri(t->gl.target, GL_TEXTURE_SWIZZLE_A, (GLint)getSwizzleChannel(t->gl.swizzle[3]));
 #endif
-        ref->swizzle = desc.swizzle;
+        ref->swizzle = t->gl.swizzle;
     }
 }
 
@@ -310,27 +305,31 @@ void GLDescriptorSet::bind(
                 }
             } else if constexpr (std::is_same_v<T, Sampler>) {
                 GLuint const unit = p.getTextureUnit(set, binding);
-                if (arg.target) {
+
-                    gl.bindTexture(unit, arg.target, arg.id, arg.external);
+
+                if (arg.handle) {
+                    GLTexture const* const t = handleAllocator.handle_cast<GLTexture*>(arg.handle);
+                    gl.bindTexture(unit, t->gl.target, t->gl.id, t->gl.external);
                     gl.bindSampler(unit, arg.sampler);
-                    if (UTILS_UNLIKELY(arg.ref)) {
+                    if (UTILS_UNLIKELY(t->ref)) {
-                        updateTextureView(gl, handleAllocator, unit, arg);
+                        updateTextureView(gl, handleAllocator, unit, t);
                     }
                 } else {
                     gl.unbindTextureUnit(unit);
                 }
             } else if constexpr (std::is_same_v<T, SamplerWithAnisotropyWorkaround>) {
                 GLuint const unit = p.getTextureUnit(set, binding);
-                if (arg.target) {
+                if (arg.handle) {
-                    gl.bindTexture(unit, arg.target, arg.id, arg.external);
+                    GLTexture const* const t = handleAllocator.handle_cast<GLTexture*>(arg.handle);
+                    gl.bindTexture(unit, t->gl.target, t->gl.id, t->gl.external);
                     gl.bindSampler(unit, arg.sampler);
-                    if (UTILS_UNLIKELY(arg.ref)) {
+                    if (UTILS_UNLIKELY(t->ref)) {
-                        updateTextureView(gl, handleAllocator, unit, arg);
+                        updateTextureView(gl, handleAllocator, unit, t);
                     }
 #if defined(GL_EXT_texture_filter_anisotropic)
                     // Driver claims to support anisotropic filtering, but it fails when set on
                     // the sampler, we have to set it on the texture instead.
-                    glTexParameterf(arg.target, GL_TEXTURE_MAX_ANISOTROPY_EXT,
+                    glTexParameterf(t->gl.target, GL_TEXTURE_MAX_ANISOTROPY_EXT,
                             std::min(gl.gets.max_anisotropy, float(arg.anisotropy)));
 #endif
                 } else {
@@ -339,19 +338,20 @@ void GLDescriptorSet::bind(
             } else if constexpr (std::is_same_v<T, SamplerGLES2>) {
                 // in ES2 the sampler parameters need to be set on the texture itself
                 GLuint const unit = p.getTextureUnit(set, binding);
-                if (arg.target) {
+                if (arg.handle) {
-                    gl.bindTexture(unit, arg.target, arg.id, arg.external);
+                    GLTexture const* const t = handleAllocator.handle_cast<GLTexture*>(arg.handle);
+                    gl.bindTexture(unit, t->gl.target, t->gl.id, t->gl.external);
                     SamplerParams const params = arg.params;
-                    glTexParameteri(arg.target, GL_TEXTURE_MIN_FILTER,
+                    glTexParameteri(t->gl.target, GL_TEXTURE_MIN_FILTER,
                             (GLint)GLUtils::getTextureFilter(params.filterMin));
-                    glTexParameteri(arg.target, GL_TEXTURE_MAG_FILTER,
+                    glTexParameteri(t->gl.target, GL_TEXTURE_MAG_FILTER,
                             (GLint)GLUtils::getTextureFilter(params.filterMag));
-                    glTexParameteri(arg.target, GL_TEXTURE_WRAP_S,
+                    glTexParameteri(t->gl.target, GL_TEXTURE_WRAP_S,
                             (GLint)GLUtils::getWrapMode(params.wrapS));
-                    glTexParameteri(arg.target, GL_TEXTURE_WRAP_T,
+                    glTexParameteri(t->gl.target, GL_TEXTURE_WRAP_T,
                             (GLint)GLUtils::getWrapMode(params.wrapT));
 #if defined(GL_EXT_texture_filter_anisotropic)
-                    glTexParameterf(arg.target, GL_TEXTURE_MAX_ANISOTROPY_EXT,
+                    glTexParameterf(t->gl.target, GL_TEXTURE_MAX_ANISOTROPY_EXT,
                             std::min(gl.gets.max_anisotropy, arg.anisotropy));
 #endif
                 } else {
@@ -360,7 +360,7 @@ void GLDescriptorSet::bind(
             }
         }, entry.desc);
     });
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void GLDescriptorSet::validate(HandleAllocatorGL& allocator,

filament/backend/src/opengl/GLDescriptorSet.h

@@ -59,8 +59,8 @@ struct GLDescriptorSet : public HwDescriptorSet {
             descriptor_binding_t binding, GLBufferObject* bo, size_t offset, size_t size) noexcept;
 
     // update a sampler descriptor in the set
-    void update(OpenGLContext& gl,
+    void update(OpenGLContext& gl, HandleAllocatorGL& handleAllocator,
-            descriptor_binding_t binding, GLTexture* t, SamplerParams params) noexcept;
+            descriptor_binding_t binding, TextureHandle th, SamplerParams params) noexcept;
 
     // conceptually bind the set to the command buffer
     void bind(
@@ -111,46 +111,19 @@ private:
 
     // A sampler descriptor
     struct Sampler {
-        uint16_t target;                        // 2 (GLenum)
+        TextureHandle handle;                   // 4
-        bool external = false;                  // 1
-        bool reserved = false;                  // 1
-        GLuint id = 0;                          // 4
         GLuint sampler = 0;                     // 4
-        Handle<GLTextureRef> ref;               // 4
-        int8_t baseLevel = 0x7f;                // 1
-        int8_t maxLevel = -1;                   // 1
-        std::array<TextureSwizzle, 4> swizzle{  // 4
-                TextureSwizzle::CHANNEL_0,
-                TextureSwizzle::CHANNEL_1,
-                TextureSwizzle::CHANNEL_2,
-                TextureSwizzle::CHANNEL_3
-        };
     };
 
     struct SamplerWithAnisotropyWorkaround {
-        uint16_t target;                        // 2 (GLenum)
+        TextureHandle handle;                   // 4
-        bool external = false;                  // 1
-        bool reserved = false;                  // 1
-        GLuint id = 0;                          // 4
         GLuint sampler = 0;                     // 4
-        Handle<GLTextureRef> ref;               // 4
         math::half anisotropy = 1.0f;           // 2
-        int8_t baseLevel = 0x7f;                // 1
-        int8_t maxLevel = -1;                   // 1
-        std::array<TextureSwizzle, 4> swizzle{  // 4
-                TextureSwizzle::CHANNEL_0,
-                TextureSwizzle::CHANNEL_1,
-                TextureSwizzle::CHANNEL_2,
-                TextureSwizzle::CHANNEL_3
-        };
     };
 
     // A sampler descriptor for ES2
     struct SamplerGLES2 {
-        uint16_t target;                        // 2 (GLenum)
+        TextureHandle handle;                   // 4
-        bool external = false;                  // 1
-        bool reserved = false;                  // 1
-        GLuint id = 0;                          // 4
         SamplerParams params{};                 // 4
         float anisotropy = 1.0f;                // 4
     };
@@ -165,9 +138,8 @@ private:
     };
     static_assert(sizeof(Descriptor) <= 32);
 
-    template<typename T>
     static void updateTextureView(OpenGLContext& gl,
-            HandleAllocatorGL& handleAllocator, GLuint unit, T const& desc) noexcept;
+            HandleAllocatorGL& handleAllocator, GLuint unit, GLTexture const* t) noexcept;
 
     utils::FixedCapacityVector<Descriptor> descriptors;     // 16
     utils::bitset64 dynamicBuffers;                         // 8

filament/backend/src/opengl/GLUtils.cpp

@@ -18,6 +18,7 @@
 
 #include "private/backend/Driver.h"
 
+#include <utils/Logger.h>
 #include <utils/compiler.h>
 #include <utils/ostream.h>
 #include <utils/trap.h>
@@ -25,6 +26,7 @@
 #include <string_view>
 
 #include <stddef.h>
+#include <cstdio>
 
 namespace filament::backend {
 
@@ -55,19 +57,21 @@ std::string_view getGLErrorString(GLenum error) noexcept {
 }
 
 UTILS_NOINLINE
-GLenum checkGLError(io::ostream& out, const char* function, size_t line) noexcept {
+GLenum checkGLError(const char* function, size_t line) noexcept {
     GLenum const error = glGetError();
     if (UTILS_VERY_UNLIKELY(error != GL_NO_ERROR)) {
         auto const string = getGLErrorString(error);
-        out << "OpenGL error " << io::hex << error << " (" << string << ") in \""
+        char hexError[16];
-            << function << "\" at line " << io::dec << line << io::endl;
+        snprintf(hexError, sizeof(hexError), "%#x", error);
+        LOG(ERROR) << "OpenGL error " << hexError << " (" << string << ") in \"" << function
+                   << "\" at line " << line;
     }
     return error;
 }
 
 UTILS_NOINLINE
-void assertGLError(io::ostream& out, const char* function, size_t line) noexcept {
+void assertGLError(const char* function, size_t line) noexcept {
-    GLenum const err = checkGLError(out, function, line);
+    GLenum const err = checkGLError(function, line);
     if (UTILS_VERY_UNLIKELY(err != GL_NO_ERROR)) {
         debug_trap();
     }
@@ -97,19 +101,21 @@ std::string_view getFramebufferStatusString(GLenum status) noexcept {
 }
 
 UTILS_NOINLINE
-GLenum checkFramebufferStatus(io::ostream& out, GLenum target, const char* function, size_t line) noexcept {
+GLenum checkFramebufferStatus(GLenum target, const char* function, size_t line) noexcept {
     GLenum const status = glCheckFramebufferStatus(target);
     if (UTILS_VERY_UNLIKELY(status != GL_FRAMEBUFFER_COMPLETE)) {
         auto const string = getFramebufferStatusString(status);
-        out << "OpenGL framebuffer error " << io::hex << status << " (" << string << ") in \""
+        char hexStatus[16];
-            << function << "\" at line " << io::dec << line << io::endl;
+        snprintf(hexStatus, sizeof(hexStatus), "%#x", status);
+        LOG(ERROR) << "OpenGL framebuffer error " << hexStatus << " (" << string << ") in \""
+                   << function << "\" at line " << line;
     }
     return status;
 }
 
 UTILS_NOINLINE
-void assertFramebufferStatus(io::ostream& out, GLenum target, const char* function, size_t line) noexcept {
+void assertFramebufferStatus(GLenum target, const char* function, size_t line) noexcept {
-    GLenum const status = checkFramebufferStatus(out, target, function, line);
+    GLenum const status = checkFramebufferStatus(target, function, line);
     if (UTILS_VERY_UNLIKELY(status != GL_FRAMEBUFFER_COMPLETE)) {
         debug_trap();
     }

filament/backend/src/opengl/GLUtils.h

@@ -33,21 +33,21 @@
 namespace filament::backend::GLUtils {
 
 std::string_view getGLErrorString(GLenum error) noexcept;
-GLenum checkGLError(utils::io::ostream& out, const char* function, size_t line) noexcept;
+GLenum checkGLError(const char* function, size_t line) noexcept;
-void assertGLError(utils::io::ostream& out, const char* function, size_t line) noexcept;
+void assertGLError(const char* function, size_t line) noexcept;
 
 std::string_view getFramebufferStatusString(GLenum err) noexcept;
-GLenum checkFramebufferStatus(utils::io::ostream& out, GLenum target, const char* function, size_t line) noexcept;
+GLenum checkFramebufferStatus(GLenum target, const char* function, size_t line) noexcept;
-void assertFramebufferStatus(utils::io::ostream& out, GLenum target, const char* function, size_t line) noexcept;
+void assertFramebufferStatus(GLenum target, const char* function, size_t line) noexcept;
 
 #ifdef NDEBUG
-#   define CHECK_GL_ERROR(out)
+#   define CHECK_GL_ERROR()
-#   define CHECK_GL_ERROR_NON_FATAL(out)
+#   define CHECK_GL_ERROR_NON_FATAL()
-#   define CHECK_GL_FRAMEBUFFER_STATUS(out, target)
+#   define CHECK_GL_FRAMEBUFFER_STATUS(target)
 #else
-#   define CHECK_GL_ERROR(out) { GLUtils::assertGLError(out, __func__, __LINE__); }
+#   define CHECK_GL_ERROR() { GLUtils::assertGLError(__func__, __LINE__); }
-#   define CHECK_GL_ERROR_NON_FATAL(out) { GLUtils::checkGLError(out, __func__, __LINE__); }
+#   define CHECK_GL_ERROR_NON_FATAL() { GLUtils::checkGLError(__func__, __LINE__); }
-#   define CHECK_GL_FRAMEBUFFER_STATUS(out, target) { GLUtils::checkFramebufferStatus(out, target, __func__, __LINE__); }
+#   define CHECK_GL_FRAMEBUFFER_STATUS(target) { GLUtils::checkFramebufferStatus( target, __func__, __LINE__); }
 #endif
 
 constexpr GLuint getComponentCount(ElementType const type) noexcept {

filament/backend/src/opengl/OpenGLContext.cpp

@@ -22,9 +22,9 @@
 #include <backend/platforms/OpenGLPlatform.h>
 #include <backend/DriverEnums.h>
 
+#include <utils/Logger.h>
 #include <utils/compiler.h>
 #include <utils/debug.h>
-#include <utils/Log.h>
 #include <utils/ostream.h>
 
 #include <functional>
@@ -77,8 +77,8 @@ OpenGLContext::OpenGLContext(OpenGLPlatform& platform,
     state.version  = (char const*)glGetString(GL_VERSION);
     state.shader   = (char const*)glGetString(GL_SHADING_LANGUAGE_VERSION);
 
-    slog.v << "[" << state.vendor << "], [" << state.renderer << "], "
+    LOG(INFO) << "[" << state.vendor << "], [" << state.renderer << "], "
-              "[" << state.version << "], [" << state.shader << "]" << io::endl;
+                 "[" << state.version << "], [" << state.shader << "]";
 
     /*
      * Figure out GL / GLES version, extensions and capabilities we need to
@@ -164,51 +164,33 @@ OpenGLContext::OpenGLContext(OpenGLPlatform& platform,
     }
 #endif
 
-    slog.v << "Feature level: " << +mFeatureLevel << '\n';
+    LOG(INFO) << "Feature level: " << +mFeatureLevel;
-    slog.v << "Active workarounds: " << '\n';
+    LOG(INFO) << "Active workarounds: ";
     UTILS_NOUNROLL
     for (auto [enabled, name, _] : mBugDatabase) {
         if (enabled) {
-            slog.v << name << '\n';
+            LOG(INFO) << name;
         }
     }
-    flush(slog.v);
 
 #ifndef NDEBUG
     // this is useful for development
-    slog.v
+    LOG(INFO) << "GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT = " << gets.max_anisotropy;
-            << "GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT = "
+    LOG(INFO) << "GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS = " << gets.max_combined_texture_image_units;
-                    << gets.max_anisotropy << '\n'
+    LOG(INFO) << "GL_MAX_TEXTURE_SIZE = " << gets.max_texture_size;
-            << "GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS = "
+    LOG(INFO) << "GL_MAX_CUBE_MAP_TEXTURE_SIZE = " << gets.max_cubemap_texture_size;
-                    << gets.max_combined_texture_image_units << '\n'
+    LOG(INFO) << "GL_MAX_3D_TEXTURE_SIZE = " << gets.max_3d_texture_size;
-            << "GL_MAX_TEXTURE_SIZE = "
+    LOG(INFO) << "GL_MAX_ARRAY_TEXTURE_LAYERS = " << gets.max_array_texture_layers;
-                    << gets.max_texture_size << '\n'
+    LOG(INFO) << "GL_MAX_DRAW_BUFFERS = " << gets.max_draw_buffers;
-            << "GL_MAX_CUBE_MAP_TEXTURE_SIZE = "
+    LOG(INFO) << "GL_MAX_RENDERBUFFER_SIZE = " << gets.max_renderbuffer_size;
-                    << gets.max_cubemap_texture_size << '\n'
+    LOG(INFO) << "GL_MAX_SAMPLES = " << gets.max_samples;
-            << "GL_MAX_3D_TEXTURE_SIZE = "
+    LOG(INFO) << "GL_MAX_TEXTURE_IMAGE_UNITS = " << gets.max_texture_image_units;
-                    << gets.max_3d_texture_size << '\n'
+    LOG(INFO) << "GL_MAX_TRANSFORM_FEEDBACK_SEPARATE_ATTRIBS = "
-            << "GL_MAX_ARRAY_TEXTURE_LAYERS = "
+              << gets.max_transform_feedback_separate_attribs;
-                    << gets.max_array_texture_layers << '\n'
+    LOG(INFO) << "GL_MAX_UNIFORM_BLOCK_SIZE = " << gets.max_uniform_block_size;
-            << "GL_MAX_DRAW_BUFFERS = "
+    LOG(INFO) << "GL_MAX_UNIFORM_BUFFER_BINDINGS = " << gets.max_uniform_buffer_bindings;
-                    << gets.max_draw_buffers << '\n'
+    LOG(INFO) << "GL_NUM_PROGRAM_BINARY_FORMATS = " << gets.num_program_binary_formats;
-            << "GL_MAX_RENDERBUFFER_SIZE = "
+    LOG(INFO) << "GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT = " << gets.uniform_buffer_offset_alignment;
-                    << gets.max_renderbuffer_size << '\n'
-            << "GL_MAX_SAMPLES = "
-                    << gets.max_samples << '\n'
-            << "GL_MAX_TEXTURE_IMAGE_UNITS = "
-                    << gets.max_texture_image_units << '\n'
-            << "GL_MAX_TRANSFORM_FEEDBACK_SEPARATE_ATTRIBS = "
-                    << gets.max_transform_feedback_separate_attribs << '\n'
-            << "GL_MAX_UNIFORM_BLOCK_SIZE = "
-                    << gets.max_uniform_block_size << '\n'
-            << "GL_MAX_UNIFORM_BUFFER_BINDINGS = "
-                    << gets.max_uniform_buffer_bindings << '\n'
-            << "GL_NUM_PROGRAM_BINARY_FORMATS = "
-                    << gets.num_program_binary_formats << '\n'
-            << "GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT = "
-                    << gets.uniform_buffer_offset_alignment << '\n'
-            ;
-    flush(slog.v);
 #endif
 
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
@@ -242,15 +224,14 @@ OpenGLContext::OpenGLContext(OpenGLPlatform& platform,
     if (ext.KHR_debug) {
         auto cb = +[](GLenum, GLenum type, GLuint, GLenum severity, GLsizei length,
                 const GLchar* message, const void *) {
-            io::ostream* stream = &slog.i;
+            auto logSeverity = utils::LogSeverity::kInfo;
             switch (severity) {
-                case GL_DEBUG_SEVERITY_HIGH:    stream = &slog.e;   break;
+                case GL_DEBUG_SEVERITY_HIGH:    logSeverity = utils::LogSeverity::kError;   break;
-                case GL_DEBUG_SEVERITY_MEDIUM:  stream = &slog.w;   break;
+                case GL_DEBUG_SEVERITY_MEDIUM:  logSeverity = utils::LogSeverity::kWarning; break;
-                case GL_DEBUG_SEVERITY_LOW:     stream = &slog.d;   break;
+                case GL_DEBUG_SEVERITY_LOW:     logSeverity = utils::LogSeverity::kInfo;    break;
                 case GL_DEBUG_SEVERITY_NOTIFICATION:
                 default: break;
             }
-            io::ostream& out = *stream;
             const char* level = ": ";
             switch (type) {
                 case GL_DEBUG_TYPE_ERROR:               level = "ERROR: ";               break;
@@ -262,7 +243,7 @@ OpenGLContext::OpenGLContext(OpenGLPlatform& platform,
                 case GL_DEBUG_TYPE_MARKER:              level = "MARKER: ";              break;
                 default: break;
             }
-            out << "KHR_debug " << level << std::string_view{ message, size_t(length) } << io::endl;
+            LOG(LEVEL(logSeverity)) << "KHR_debug " << level << std::string_view{ message, size_t(length) };
         };
         glEnable(GL_DEBUG_OUTPUT);
         glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS);
@@ -707,9 +688,8 @@ void OpenGLContext::initExtensionsGLES(Extensions* ext, GLint major, GLint minor
     GLUtils::unordered_string_set const exts = GLUtils::split(extensions);
     if constexpr (DEBUG_PRINT_EXTENSIONS) {
         for (auto extension: exts) {
-            slog.d << "\"" << std::string_view(extension) << "\"\n";
+            DLOG(INFO) << "\"" << std::string_view(extension) << "\"";
         }
-        flush(slog.d);
     }
 
     // figure out and initialize the extensions we need
@@ -783,9 +763,8 @@ void OpenGLContext::initExtensionsGL(Extensions* ext, GLint major, GLint minor)
     }
     if constexpr (DEBUG_PRINT_EXTENSIONS) {
         for (auto extension: exts) {
-            slog.d << "\"" << std::string_view(extension) << "\"\n";
+            DLOG(INFO) << "\"" << std::string_view(extension) << "\"";
         }
-        flush(slog.d);
     }
 
     using namespace std::literals;
@@ -1045,7 +1024,7 @@ GLuint OpenGLContext::getSamplerSlow(SamplerParams params) const noexcept {
                 std::min(gets.max_anisotropy, anisotropy));
     }
 #endif
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
     mSamplerMap[params] = s;
     return s;
 }

filament/backend/src/opengl/OpenGLDriver.cpp

@@ -48,12 +48,13 @@
 
 #include <type_traits>
 #include <utils/BitmaskEnum.h>
-#include <utils/FixedCapacityVector.h>
 #include <utils/CString.h>
+#include <utils/FixedCapacityVector.h>
 #include <utils/Invocable.h>
-#include <utils/Log.h>
+#include <utils/Logger.h>
 #include <utils/Panic.h>
 #include <utils/Slice.h>
+#include <utils/Systrace.h>
 #include <utils/compiler.h>
 #include <utils/debug.h>
 #include <utils/ostream.h>
@@ -184,20 +185,18 @@ OpenGLDriver* OpenGLDriver::create(OpenGLPlatform* platform,
     //    GLVertexBufferInfo        : 132       moderate
     // -- less than or equal to 136 bytes
 
-    slog.d
+    DLOG(INFO) << "GLSwapChain: " << sizeof(GLSwapChain);
-           << "\nGLSwapChain: " << sizeof(GLSwapChain)
+    DLOG(INFO) << "GLBufferObject: " << sizeof(GLBufferObject);
-           << "\nGLBufferObject: " << sizeof(GLBufferObject)
+    DLOG(INFO) << "GLVertexBuffer: " << sizeof(GLVertexBuffer);
-           << "\nGLVertexBuffer: " << sizeof(GLVertexBuffer)
+    DLOG(INFO) << "GLVertexBufferInfo: " << sizeof(GLVertexBufferInfo);
-           << "\nGLVertexBufferInfo: " << sizeof(GLVertexBufferInfo)
+    DLOG(INFO) << "GLIndexBuffer: " << sizeof(GLIndexBuffer);
-           << "\nGLIndexBuffer: " << sizeof(GLIndexBuffer)
+    DLOG(INFO) << "GLRenderPrimitive: " << sizeof(GLRenderPrimitive);
-           << "\nGLRenderPrimitive: " << sizeof(GLRenderPrimitive)
+    DLOG(INFO) << "GLTexture: " << sizeof(GLTexture);
-           << "\nGLTexture: " << sizeof(GLTexture)
+    DLOG(INFO) << "GLTimerQuery: " << sizeof(GLTimerQuery);
-           << "\nGLTimerQuery: " << sizeof(GLTimerQuery)
+    DLOG(INFO) << "GLStream: " << sizeof(GLStream);
-           << "\nGLStream: " << sizeof(GLStream)
+    DLOG(INFO) << "GLRenderTarget: " << sizeof(GLRenderTarget);
-           << "\nGLRenderTarget: " << sizeof(GLRenderTarget)
+    DLOG(INFO) << "GLFence: " << sizeof(GLFence);
-           << "\nGLFence: " << sizeof(GLFence)
+    DLOG(INFO) << "OpenGLProgram: " << sizeof(OpenGLProgram);
-           << "\nOpenGLProgram: " << sizeof(OpenGLProgram)
-           << io::endl;
 #endif
 
     // here we check we're on a supported version of GL before initializing the driver
@@ -289,7 +288,7 @@ OpenGLDriver::OpenGLDriver(OpenGLPlatform* platform, const Platform::DriverConfi
     mStreamsWithPendingAcquiredImage.reserve(8);
 
 #ifndef NDEBUG
-    slog.i << "OS version: " << mPlatform.getOSVersion() << io::endl;
+    LOG(INFO) << "OS version: " << mPlatform.getOSVersion();
 #endif
 
     // Timer queries are core in GL 3.3, otherwise we need EXT_disjoint_timer_query
@@ -668,7 +667,7 @@ void OpenGLDriver::createIndexBufferR(
     gl.bindVertexArray(nullptr);
     gl.bindBuffer(GL_ELEMENT_ARRAY_BUFFER, ib->gl.buffer);
     glBufferData(GL_ELEMENT_ARRAY_BUFFER, size, nullptr, getBufferUsage(usage));
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::createBufferObjectR(Handle<HwBufferObject> boh,
@@ -693,7 +692,7 @@ void OpenGLDriver::createBufferObjectR(Handle<HwBufferObject> boh,
         glBufferData(bo->gl.binding, byteCount, nullptr, getBufferUsage(usage));
     }
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::createRenderPrimitiveR(Handle<HwRenderPrimitive> rph,
@@ -730,7 +729,7 @@ void OpenGLDriver::createRenderPrimitiveR(Handle<HwRenderPrimitive> rph,
     // this records the index buffer into the currently bound VAO
     gl.bindBuffer(GL_ELEMENT_ARRAY_BUFFER, ib->gl.buffer);
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {
@@ -760,7 +759,7 @@ void OpenGLDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {
     }
 
     construct<OpenGLProgram>(ph, *this, std::move(program));
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 UTILS_NOINLINE
@@ -932,7 +931,12 @@ void OpenGLDriver::createTextureR(Handle<HwTexture> th, SamplerType target, uint
 #if defined(BACKEND_OPENGL_LEVEL_GLES31)
                 if (gl.features.multisample_texture) {
                     // multi-sample texture on GL 3.2 / GLES 3.1 and above
-                    t->gl.target = GL_TEXTURE_2D_MULTISAMPLE;
+                    if (depth <= 1) {
+                        // We forcibly change the target to 2D-multisample only for flat texture.
+                        // A depth value greater than 1 may indicate multiview usage, which requires
+                        // GL_TEXTURE_2D_ARRAY. Also 2D MSAA won't work with non-flat texture anyway.
+                        t->gl.target = GL_TEXTURE_2D_MULTISAMPLE;
+                    }
                 } else {
                     // Turn off multi-sampling for that texture. It's just not supported.
                 }
@@ -948,7 +952,7 @@ void OpenGLDriver::createTextureR(Handle<HwTexture> th, SamplerType target, uint
         renderBufferStorage(t->gl.id, internalFormat, w, h, samples);
     }
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::createTextureViewR(Handle<HwTexture> th,
@@ -994,7 +998,7 @@ void OpenGLDriver::createTextureViewR(Handle<HwTexture> th,
     assert_invariant(ref);
     ref->count++;
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::createTextureViewSwizzleR(Handle<HwTexture> th, Handle<HwTexture> srch,
@@ -1057,7 +1061,7 @@ void OpenGLDriver::createTextureViewSwizzleR(Handle<HwTexture> th, Handle<HwText
     assert_invariant(ref);
     ref->count++;
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::createTextureExternalImage2R(Handle<HwTexture> th, SamplerType target,
@@ -1208,14 +1212,19 @@ void OpenGLDriver::importTextureR(Handle<HwTexture> th, intptr_t id,
 #if defined(BACKEND_OPENGL_LEVEL_GLES31)
         if (gl.features.multisample_texture) {
             // multi-sample texture on GL 3.2 / GLES 3.1 and above
-            t->gl.target = GL_TEXTURE_2D_MULTISAMPLE;
+            if (depth <= 1) {
+                // We forcibly change the target to 2D-multisample only for flat texture.
+                // A depth value greater than 1 may indicate multiview usage, which requires
+                // GL_TEXTURE_2D_ARRAY. Also 2D MSAA won't work with non-flat texture anyway.
+                t->gl.target = GL_TEXTURE_2D_MULTISAMPLE;
+            }
         } else {
             // Turn off multi-sampling for that texture. It's just not supported.
         }
 #endif
     }
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::updateVertexArrayObject(GLRenderPrimitive* rp, GLVertexBuffer const* vb) {
@@ -1453,8 +1462,15 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
 #if !defined(__EMSCRIPTEN__) && !defined(FILAMENT_IOS)
                 if (layerCount > 1) {
                     // if layerCount > 1, it means we use the multiview extension.
-                    glFramebufferTextureMultiviewOVR(GL_FRAMEBUFFER, attachment,
+                    if (rt->gl.samples > 1) {
-                        t->gl.id, 0, binfo.layer, layerCount);
+                        // For MSAA
+                        glFramebufferTextureMultisampleMultiviewOVR(GL_FRAMEBUFFER, attachment,
+                                t->gl.id, 0, rt->gl.samples, binfo.layer, layerCount);
+                    }
+                    else {
+                        glFramebufferTextureMultiviewOVR(GL_FRAMEBUFFER, attachment, t->gl.id, 0,
+                                binfo.layer, layerCount);
+                    }
                 } else
 #endif // !defined(__EMSCRIPTEN__) && !defined(FILAMENT_IOS)
                 {
@@ -1468,7 +1484,7 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
                 // we shouldn't be here
                 break;
         }
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
     } else
 #ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_multisampled_render_to_texture
@@ -1489,7 +1505,7 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
             glFramebufferRenderbuffer(GL_FRAMEBUFFER, attachment,
                     GL_RENDERBUFFER, t->gl.id);
         }
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
     } else
 #endif // GL_EXT_multisampled_render_to_texture
 #endif // __EMSCRIPTEN__
@@ -1571,13 +1587,13 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
                 break;
         }
 
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
     }
 
     rt->gl.resolve |= resolveFlags;
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_FRAMEBUFFER)
+    CHECK_GL_FRAMEBUFFER_STATUS(GL_FRAMEBUFFER)
 }
 
 void OpenGLDriver::renderBufferStorage(GLuint rbo, GLenum internalformat, uint32_t width, // NOLINT(readability-convert-member-functions-to-static)
@@ -1606,7 +1622,7 @@ void OpenGLDriver::renderBufferStorage(GLuint rbo, GLenum internalformat, uint32
     // unbind the renderbuffer, to avoid any later confusion
     glBindRenderbuffer(GL_RENDERBUFFER, 0);
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::createDefaultRenderTargetR(
@@ -1703,7 +1719,7 @@ void OpenGLDriver::createRenderTargetR(Handle<HwRenderTarget> rth,
         if (UTILS_LIKELY(!getContext().isES2())) {
             glDrawBuffers((GLsizei)maxDrawBuffers, bufs);
         }
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
     }
 #endif
 
@@ -1743,7 +1759,7 @@ void OpenGLDriver::createRenderTargetR(Handle<HwRenderTarget> rth,
     assert_invariant(any(targets & TargetBufferFlags::ALL));
     assert_invariant(tmin == tmax);
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::createFenceR(Handle<HwFence> fh, int) {
@@ -2094,7 +2110,7 @@ void OpenGLDriver::setAcquiredImage(Handle<HwStream> sh, void* hwbuffer, const m
 
     if (UTILS_UNLIKELY(glstream->user_thread.pending.image)) {
         scheduleRelease(glstream->user_thread.pending);
-        slog.w << "Acquired image is set more than once per frame." << io::endl;
+        LOG(WARNING) << "Acquired image is set more than once per frame.";
     }
 
     glstream->user_thread.pending = mPlatform.transformAcquiredImage({
@@ -2514,13 +2530,36 @@ void OpenGLDriver::makeCurrent(Handle<HwSwapChain> schDraw, Handle<HwSwapChain>
 
     mPlatform.makeCurrent(scDraw->swapChain, scRead->swapChain,
             [this]() {
+                for (auto t: mTexturesWithStreamsAttached) {
+                    if (t->hwStream->streamType == StreamType::NATIVE) {
+                        mPlatform.detach(t->hwStream->stream);
+                    }
+                }
                 // OpenGL context is about to change, unbind everything
                 mContext.unbindEverything();
             },
             [this](size_t index) {
+                for (auto t: mTexturesWithStreamsAttached) {
+                    if (t->hwStream->streamType == StreamType::NATIVE) {
+                        if (t->externalTexture) {
+                            glGenTextures(1, &t->externalTexture->id);
+                            t->gl.id = t->externalTexture->id;
+                        } else {
+                            glGenTextures(1, &t->gl.id);
+                        }
+                        mPlatform.attach(t->hwStream->stream, t->gl.id);
+                        mContext.updateTexImage(GL_TEXTURE_EXTERNAL_OES, t->gl.id);
+                    }
+                }
+
+                // force invalidation of all bound descriptor sets
+                decltype(mInvalidDescriptorSetBindings) changed;
+                changed.setValue((1 << MAX_DESCRIPTOR_SET_COUNT) - 1);
+                mInvalidDescriptorSetBindings |= changed;
+
                 // OpenGL context has changed, resynchronize the state with the cache
                 mContext.synchronizeStateAndCache(index);
-                slog.d << "*** OpenGL context change : " << (index ? "protected" : "default") << io::endl;
+                DLOG(INFO) << "*** OpenGL context change : " << (index ? "protected" : "default");
             });
 
     mCurrentDrawSwapChain = scDraw;
@@ -2561,7 +2600,7 @@ void OpenGLDriver::setVertexBufferObject(Handle<HwVertexBuffer> vbh,
         vb->bufferObjectsVersion = (version + 1) % kMaxVersion;
     }
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::updateIndexBuffer(
@@ -2578,7 +2617,7 @@ void OpenGLDriver::updateIndexBuffer(
 
     scheduleDestroy(std::move(p));
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::registerBufferObjectStreams(Handle<HwBufferObject> boh, BufferObjectStreamDescriptor&& streams) {
@@ -2658,7 +2697,7 @@ void OpenGLDriver::updateBufferObject(
 
     scheduleDestroy(std::move(bd));
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::updateBufferObjectUnsynchronized(
@@ -2705,7 +2744,7 @@ retry:
             scheduleDestroy(std::move(bd));
         }
     }
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 #endif
 }
 
@@ -2755,7 +2794,7 @@ void OpenGLDriver::generateMipmaps(Handle<HwTexture> th) {
 
     glGenerateMipmap(t->gl.target);
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::setTextureData(GLTexture const* t, uint32_t level,
@@ -2864,7 +2903,7 @@ void OpenGLDriver::setTextureData(GLTexture const* t, uint32_t level,
 
     scheduleDestroy(std::move(p));
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::setCompressedTextureData(GLTexture const* t, uint32_t level,
@@ -2950,7 +2989,7 @@ void OpenGLDriver::setCompressedTextureData(GLTexture const* t, uint32_t level,
 
     scheduleDestroy(std::move(p));
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::setupExternalImage2(Platform::ExternalImageHandleRef image) {
@@ -2992,6 +3031,7 @@ void OpenGLDriver::attachStream(GLTexture* t, GLStream* hwStream) noexcept {
     switch (hwStream->streamType) {
         case StreamType::NATIVE:
             mPlatform.attach(hwStream->stream, t->gl.id);
+            mContext.updateTexImage(GL_TEXTURE_EXTERNAL_OES, t->gl.id);
             break;
         case StreamType::ACQUIRED:
             break;
@@ -3020,7 +3060,12 @@ void OpenGLDriver::detachStream(GLTexture* t) noexcept {
             break;
     }
 
-    glGenTextures(1, &t->gl.id);
+    if (t->externalTexture) {
+        glGenTextures(1, &t->externalTexture->id);
+        t->gl.id = t->externalTexture->id;
+    } else {
+        glGenTextures(1, &t->gl.id);
+    }
 
     t->hwStream = nullptr;
 }
@@ -3044,8 +3089,14 @@ void OpenGLDriver::replaceStream(GLTexture* texture, GLStream* newStream) noexce
 
     switch (newStream->streamType) {
         case StreamType::NATIVE:
-            glGenTextures(1, &texture->gl.id);
+            if (texture->externalTexture) {
+                glGenTextures(1, &texture->externalTexture->id);
+                texture->gl.id = texture->externalTexture->id;
+            } else {
+                glGenTextures(1, &texture->gl.id);
+            }
             mPlatform.attach(newStream->stream, texture->gl.id);
+            mContext.updateTexImage(GL_TEXTURE_EXTERNAL_OES, texture->gl.id);
             break;
         case StreamType::ACQUIRED:
             // Just re-use the old texture id.
@@ -3101,7 +3152,7 @@ void OpenGLDriver::beginRenderPass(Handle<HwRenderTarget> rth,
     TargetBufferFlags discardFlags = params.flags.discardStart & rt->targets;
 
     GLuint const fbo = gl.bindFramebuffer(GL_FRAMEBUFFER, rt->gl.fbo);
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_FRAMEBUFFER)
+    CHECK_GL_FRAMEBUFFER_STATUS(GL_FRAMEBUFFER)
 
     // each render-pass starts with a disabled scissor
     gl.disable(GL_SCISSOR_TEST);
@@ -3113,7 +3164,7 @@ void OpenGLDriver::beginRenderPass(Handle<HwRenderTarget> rth,
         if (attachmentCount) {
             gl.procs.invalidateFramebuffer(GL_FRAMEBUFFER, attachmentCount, attachments.data());
         }
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
     } else {
         // It's important to clear the framebuffer before drawing, as it resets
         // the fb to a known state (resets fb compression and possibly other things).
@@ -3200,7 +3251,7 @@ void OpenGLDriver::endRenderPass(int) {
             if (attachmentCount) {
                 gl.procs.invalidateFramebuffer(GL_FRAMEBUFFER, attachmentCount, attachments.data());
             }
-           CHECK_GL_ERROR(utils::slog.e)
+           CHECK_GL_ERROR()
         }
     }
 
@@ -3246,13 +3297,13 @@ void OpenGLDriver::resolvePass(ResolveAction action, GLRenderTarget const* rt,
         gl.bindFramebuffer(GL_READ_FRAMEBUFFER, read);
         gl.bindFramebuffer(GL_DRAW_FRAMEBUFFER, draw);
 
-        CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_READ_FRAMEBUFFER)
+        CHECK_GL_FRAMEBUFFER_STATUS(GL_READ_FRAMEBUFFER)
-        CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_DRAW_FRAMEBUFFER)
+        CHECK_GL_FRAMEBUFFER_STATUS(GL_DRAW_FRAMEBUFFER)
 
         gl.disable(GL_SCISSOR_TEST);
         glBlitFramebuffer(0, 0, (GLint)rt->width, (GLint)rt->height,
                 0, 0, (GLint)rt->width, (GLint)rt->height, mask, GL_NEAREST);
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
     }
 #endif
 }
@@ -3431,7 +3482,7 @@ void OpenGLDriver::readPixels(Handle<HwRenderTarget> src,
         if (buffer) {
             gl.bindFramebuffer(GL_FRAMEBUFFER, s->gl.fbo_read ? s->gl.fbo_read : s->gl.fbo);
             glReadPixels(GLint(x), GLint(y), GLint(width), GLint(height), glFormat, glType, buffer);
-            CHECK_GL_ERROR(utils::slog.e)
+            CHECK_GL_ERROR()
 
             // now we need to flip the buffer vertically to match our API
             size_t const stride = p.stride ? p.stride : width;
@@ -3463,7 +3514,7 @@ void OpenGLDriver::readPixels(Handle<HwRenderTarget> src,
     glBufferData(GL_PIXEL_PACK_BUFFER, pboSize, nullptr, GL_STATIC_DRAW);
     glReadPixels(GLint(x), GLint(y), GLint(width), GLint(height), glFormat, glType, nullptr);
     gl.bindBuffer(GL_PIXEL_PACK_BUFFER, 0);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     // we're forced to make a copy on the heap because otherwise it deletes std::function<> copy
     // constructor.
@@ -3503,7 +3554,7 @@ void OpenGLDriver::readPixels(Handle<HwRenderTarget> src,
         glDeleteBuffers(1, &pbo);
         scheduleDestroy(std::move(p));
         delete pUserBuffer;
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
     });
 #endif
 }
@@ -3528,7 +3579,7 @@ void OpenGLDriver::readBufferSubData(BufferObjectHandle boh,
         glCopyBufferSubData(bo->gl.binding, GL_PIXEL_PACK_BUFFER, offset, 0, size);
         gl.bindBuffer(bo->gl.binding, 0);
         gl.bindBuffer(GL_PIXEL_PACK_BUFFER, 0);
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
 
         // then, we schedule a mapBuffer of the PBO later, once the fence has signaled
         auto* pUserBuffer = new BufferDescriptor(std::move(p));
@@ -3545,7 +3596,7 @@ void OpenGLDriver::readBufferSubData(BufferObjectHandle boh,
             glDeleteBuffers(1, &pbo);
             scheduleDestroy(std::move(p));
             delete pUserBuffer;
-            CHECK_GL_ERROR(utils::slog.e)
+            CHECK_GL_ERROR()
         });
     } else {
         gl.bindBuffer(bo->gl.binding, bo->gl.id);
@@ -3558,7 +3609,7 @@ void OpenGLDriver::readBufferSubData(BufferObjectHandle boh,
         }
         gl.bindBuffer(bo->gl.binding, 0);
         scheduleDestroy(std::move(p));
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
     }
 #endif
 }
@@ -3588,7 +3639,7 @@ void OpenGLDriver::whenFrameComplete(const std::function<void()>& fn) noexcept {
 void OpenGLDriver::whenGpuCommandsComplete(const std::function<void()>& fn) noexcept {
     GLsync sync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
     mGpuCommandCompleteOps.emplace_back(sync, fn);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::executeGpuCommandsCompleteOps() noexcept {
@@ -3708,8 +3759,7 @@ void OpenGLDriver::updateDescriptorSetTexture(
         TextureHandle th,
         SamplerParams params) {
     GLDescriptorSet* ds = handle_cast<GLDescriptorSet*>(dsh);
-    GLTexture* t = th ? handle_cast<GLTexture*>(th) : nullptr;
+    ds->update(mContext, mHandleAllocator, binding, th, params);
-    ds->update(mContext, binding, t, params);
 }
 
 void OpenGLDriver::flush(int) {
@@ -3807,7 +3857,7 @@ void OpenGLDriver::clearWithRasterPipe(TargetBufferFlags clearFlags,
         }
     }
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void OpenGLDriver::resolve(
@@ -3918,7 +3968,7 @@ void OpenGLDriver::blit(
         case SamplerType::SAMPLER_EXTERNAL:
             break;
     }
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_DRAW_FRAMEBUFFER)
+    CHECK_GL_FRAMEBUFFER_STATUS(GL_DRAW_FRAMEBUFFER)
 
     gl.bindFramebuffer(GL_READ_FRAMEBUFFER, fbo[1]);
     switch (s->target) {
@@ -3944,14 +3994,14 @@ void OpenGLDriver::blit(
         case SamplerType::SAMPLER_EXTERNAL:
             break;
     }
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_READ_FRAMEBUFFER)
+    CHECK_GL_FRAMEBUFFER_STATUS(GL_READ_FRAMEBUFFER)
 
     gl.disable(GL_SCISSOR_TEST);
     glBlitFramebuffer(
             srcOrigin.x, srcOrigin.y, srcOrigin.x + size.x, srcOrigin.y + size.y,
             dstOrigin.x, dstOrigin.y, dstOrigin.x + size.x, dstOrigin.y + size.y,
             mask, GL_NEAREST);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     gl.unbindFramebuffer(GL_DRAW_FRAMEBUFFER);
     gl.unbindFramebuffer(GL_READ_FRAMEBUFFER);
@@ -4014,15 +4064,15 @@ void OpenGLDriver::blitDEPRECATED(TargetBufferFlags buffers,
     gl.bindFramebuffer(GL_READ_FRAMEBUFFER, s->gl.fbo);
     gl.bindFramebuffer(GL_DRAW_FRAMEBUFFER, d->gl.fbo);
 
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_READ_FRAMEBUFFER)
+    CHECK_GL_FRAMEBUFFER_STATUS(GL_READ_FRAMEBUFFER)
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_DRAW_FRAMEBUFFER)
+    CHECK_GL_FRAMEBUFFER_STATUS(GL_DRAW_FRAMEBUFFER)
 
     gl.disable(GL_SCISSOR_TEST);
     glBlitFramebuffer(
             srcRect.left, srcRect.bottom, srcRect.right(), srcRect.top(),
             dstRect.left, dstRect.bottom, dstRect.right(), dstRect.top(),
             GL_COLOR_BUFFER_BIT, glFilterMode);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 #endif
 }
 
@@ -4149,9 +4199,9 @@ void OpenGLDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t ins
 #endif
 
 #if FILAMENT_ENABLE_MATDBG
-    CHECK_GL_ERROR_NON_FATAL(utils::slog.e)
+    CHECK_GL_ERROR_NON_FATAL()
 #else
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 #endif
 }
 
@@ -4181,9 +4231,9 @@ void OpenGLDriver::draw2GLES2(uint32_t indexOffset, uint32_t indexCount, uint32_
             reinterpret_cast<const void*>(indexOffset << rp->gl.indicesShift));
 
 #if FILAMENT_ENABLE_MATDBG
-    CHECK_GL_ERROR_NON_FATAL(utils::slog.e)
+    CHECK_GL_ERROR_NON_FATAL()
 #else
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 #endif
 }
 
@@ -4233,9 +4283,9 @@ void OpenGLDriver::dispatchCompute(Handle<HwProgram> program, uint3 workGroupCou
 #endif // BACKEND_OPENGL_LEVEL_GLES31
 
 #if FILAMENT_ENABLE_MATDBG
-    CHECK_GL_ERROR_NON_FATAL(utils::slog.e)
+    CHECK_GL_ERROR_NON_FATAL()
 #else
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 #endif
 }
 

filament/backend/src/opengl/OpenGLPlatform.cpp

@@ -64,7 +64,7 @@ utils::CString OpenGLPlatform::getRendererString(Driver const* driver) {
 }
 
 void OpenGLPlatform::makeCurrent(SwapChain* drawSwapChain, SwapChain* readSwapChain,
-        utils::Invocable<void()>, utils::Invocable<void(size_t)>) noexcept {
+        utils::Invocable<void()>, utils::Invocable<void(size_t)>) {
     makeCurrent(getCurrentContextType(), drawSwapChain, readSwapChain);
 }
 

filament/backend/src/opengl/OpenGLProgram.cpp

@@ -214,7 +214,7 @@ void OpenGLProgram::initializeProgramState(OpenGLContext& context, GLuint progra
                 case DescriptorType::INPUT_ATTACHMENT:
                     break;
             }
-            CHECK_GL_ERROR(utils::slog.e)
+            CHECK_GL_ERROR()
         }
     }
 

filament/backend/src/opengl/OpenGLTimerQuery.cpp

@@ -114,14 +114,14 @@ void TimerQueryNativeFactory::createTimerQuery(GLTimerQuery* tq) {
 
     tq->state = std::make_shared<GLTimerQuery::State>();
     mContext.procs.genQueries(1u, &tq->state->gl.query);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void TimerQueryNativeFactory::destroyTimerQuery(GLTimerQuery* tq) {
     assert_invariant(tq->state);
 
     mContext.procs.deleteQueries(1u, &tq->state->gl.query);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     tq->state.reset();
 }
@@ -131,14 +131,14 @@ void TimerQueryNativeFactory::beginTimeElapsedQuery(GLTimerQuery* tq) {
 
     tq->state->elapsed.store(int64_t(TimerQueryResult::NOT_READY), std::memory_order_relaxed);
     mContext.procs.beginQuery(GL_TIME_ELAPSED, tq->state->gl.query);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 void TimerQueryNativeFactory::endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQuery* tq) {
     assert_invariant(tq->state);
 
     mContext.procs.endQuery(GL_TIME_ELAPSED);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     std::weak_ptr<GLTimerQuery::State> const weak = tq->state;
 
@@ -153,7 +153,7 @@ void TimerQueryNativeFactory::endTimeElapsedQuery(OpenGLDriver& driver, GLTimerQ
 
         GLuint available = 0;
         context.procs.getQueryObjectuiv(state->gl.query, GL_QUERY_RESULT_AVAILABLE, &available);
-        CHECK_GL_ERROR(utils::slog.e)
+        CHECK_GL_ERROR()
         if (!available) {
             // we need to try this one again later
             return false;

filament/backend/src/opengl/ShaderCompilerService.cpp

@@ -23,6 +23,7 @@
 #include "OpenGLDriver.h"
 
 #include <iterator>
+#include <optional>
 #include <private/backend/BackendUtils.h>
 
 #include <backend/DriverEnums.h>
@@ -30,14 +31,14 @@
 
 #include <private/utils/Tracing.h>
 
-#include <utils/compiler.h>
 #include <utils/CString.h>
-#include <utils/debug.h>
 #include <utils/FixedCapacityVector.h>
 #include <utils/JobSystem.h>
-#include <utils/Log.h>
+#include <utils/Logger.h>
-#include <utils/ostream.h>
 #include <utils/Panic.h>
+#include <utils/compiler.h>
+#include <utils/debug.h>
+#include <utils/ostream.h>
 
 #include <algorithm>
 #include <array>
@@ -64,9 +65,9 @@ static std::string to_string(bool const b) { return b ? "true" : "false"; }
 static std::string to_string(int const i) { return std::to_string(i); }
 static std::string to_string(float const f) { return "float(" + std::to_string(f) + ")"; }
 
-static void logCompilationError(io::ostream& out, ShaderStage shaderType, const char* name,
+static void logCompilationError(ShaderStage shaderType, const char* name, GLuint shaderId,
-        GLuint shaderId, CString const& sourceCode) noexcept;
+        CString const& sourceCode) noexcept;
-static void logProgramLinkError(io::ostream& out, char const* name, GLuint program) noexcept;
+static void logProgramLinkError(char const* name, GLuint program) noexcept;
 
 static void process_GOOGLE_cpp_style_line_directive(OpenGLContext const& context, char* source,
         size_t len) noexcept;
@@ -110,17 +111,33 @@ struct ShaderCompilerService::OpenGLProgramToken : ProgramToken {
         cond.wait(l, [this] { return signaled; });
     }
 
-    CallbackManager::Handle handle{};
+    // This is invoked upon token completion, which occurs after a successful `gl.program`
+    // population or upon cancellation. In either scenario, the callback handle must be submitted
+    // to notify the caller that resource loading has concluded.
+    void trySubmittingCallback() noexcept {
+        if (handle) {
+            compiler.submitCallbackHandle(*handle);
+            handle = std::nullopt;
+        }
+    }
+
+    std::optional<CallbackManager::Handle> handle{};
+
+    // Only valid when the blob functions are provided by users. The validity of this variable
+    // doesn't guarantee that the program was created from the cache blob.
     BlobCacheKey key;
 
     // Used for the `THREAD_POOL` mode.
     mutable Mutex lock;
     mutable Condition cond;
     bool signaled = false;
+
+    // Indicate this program was created from the cache blob.
+    bool retrievedFromBlobCache = false;
 };
 
 ShaderCompilerService::OpenGLProgramToken::~OpenGLProgramToken() {
-    compiler.submitCallbackHandle(handle);
+    trySubmittingCallback();
 }
 
 /* static */ void ShaderCompilerService::setUserData(const program_token_t& token,
@@ -248,6 +265,7 @@ ShaderCompilerService::program_token_t ShaderCompilerService::createProgram(
     // Try retrieving the cached program blob if available.
     token->gl.program = mBlobCache.retrieve(&token->key, mDriver.mPlatform, program);
     if (token->gl.program) {
+        token->retrievedFromBlobCache = true;
         return token;
     }
 
@@ -326,12 +344,15 @@ GLuint ShaderCompilerService::getProgram(program_token_t& token) {
 
     assert_invariant(token);// This function should be called when the token is still alive.
 
-    if (token->compiler.mMode == Mode::THREAD_POOL) {
+    // Finalize any pending shader compilation tasks only when the token was created without cache.
-        auto const job = token->compiler.mCompilerThreadPool.dequeue(token);
+    if (!token->retrievedFromBlobCache) {
-        if (!job) {
+        if (token->compiler.mMode == Mode::THREAD_POOL) {
-            // It's likely that the job was already completed. But it may be still being
+            auto const job = token->compiler.mCompilerThreadPool.dequeue(token);
-            // executed at this moment. Just try waiting for it to avoid a race.
+            if (!job) {
-            token->wait();
+                // It's likely that the job was already completed. But it may be still being
+                // executed at this moment. Just try waiting for it to avoid a race.
+                token->wait();
+            }
         }
     }
 
@@ -339,7 +360,7 @@ GLuint ShaderCompilerService::getProgram(program_token_t& token) {
 
     // Cleanup the token.
     token->compiler.cancelTickOp(token);
-    token = nullptr;// This will submit a callback condition (handle) to the callback manager.
+    token = nullptr; // This will try submitting a callback handle to the callback manager.
 }
 
 void ShaderCompilerService::tick() {
@@ -392,7 +413,7 @@ GLuint ShaderCompilerService::initialize(program_token_t& token) {
 
     // Cleanup the token.
     token->compiler.cancelTickOp(token);
-    token = nullptr;// This will submit a callback condition (handle) to the callback manager.
+    token = nullptr;
 
     return program;
 }
@@ -422,8 +443,9 @@ void ShaderCompilerService::ensureTokenIsReady(program_token_t const& token) {
             // just log warnings here instead of repeatedly checking compile status. If this turns
             // out to be a real issue later, we would need to consider doing the canonical way.
             if (!isCompileCompleted(token)) {
-                slog.w << "Shader compilation for OpenGL program " << token->name.c_str_safe()
+                LOG(WARNING)
-                       << " is not completed yet. The following program link may not succeed.";
+                        << "Shader compilation for OpenGL program " << token->name.c_str_safe()
+                        << " is not completed yet. The following program link may not succeed.";
             }
 
             linkProgram(mDriver.getContext(), token);
@@ -633,8 +655,7 @@ void ShaderCompilerService::executeTickOps() noexcept {
         }
         // Something went wrong. Log the error message.
         const ShaderStage type = static_cast<ShaderStage>(i);
-        logCompilationError(slog.e, type, token->name.c_str_safe(), shader,
+        logCompilationError(type, token->name.c_str_safe(), shader, token->shaderSourceCode[i]);
-                token->shaderSourceCode[i]);
     }
 }
 
@@ -659,6 +680,7 @@ void ShaderCompilerService::executeTickOps() noexcept {
     }
     glLinkProgram(program);
     token->gl.program = program;
+    token->trySubmittingCallback();
 }
 
 /* static */ bool ShaderCompilerService::isLinkCompleted(program_token_t const& token) noexcept {
@@ -685,7 +707,7 @@ void ShaderCompilerService::executeTickOps() noexcept {
     glGetProgramiv(token->gl.program, GL_LINK_STATUS, &status);
     if (UTILS_UNLIKELY(status != GL_TRUE)) {
         // Something went wrong. Log the error message.
-        logProgramLinkError(slog.e, token->name.c_str_safe(), token->gl.program);
+        logProgramLinkError(token->name.c_str_safe(), token->gl.program);
         linked = false;
     }
     // No need to keep the shaders around regardless of the result of the program linking.
@@ -735,7 +757,7 @@ void ShaderCompilerService::executeTickOps() noexcept {
 // ------------------------------------------------------------------------------------------------
 
 UTILS_NOINLINE
-/* static */ void logCompilationError(io::ostream& out, ShaderStage shaderType, const char* name,
+/* static */ void logCompilationError(ShaderStage shaderType, const char* name,
         GLuint const shaderId, UTILS_UNUSED_IN_RELEASE CString const& sourceCode) noexcept {
 
     { // scope for the temporary string storage
@@ -757,8 +779,9 @@ UTILS_NOINLINE
         CString infoLog(length);
         glGetShaderInfoLog(shaderId, length, nullptr, infoLog.data());
 
-        out << "Compilation error in " << to_string(shaderType) << " shader \"" << name << "\":\n"
+        LOG(ERROR) << "Compilation error in " << to_string(shaderType) << " shader \"" << name
-            << "\"" << infoLog.c_str() << "\"" << io::endl;
+                   << "\":";
+        LOG(ERROR) << "\"" << infoLog.c_str() << "\"";
     }
 
 #ifndef NDEBUG
@@ -773,26 +796,25 @@ UTILS_NOINLINE
         } else {
             line = shader.substr(start, end - start);
         }
-        out << lc++ << ":   " << line.c_str() << '\n';
+        LOG(ERROR) << lc++ << ":   " << line.c_str();
         if (end == std::string::npos) {
             break;
         }
         start = end + 1;
     }
-    out << io::endl;
+    LOG(ERROR) << "";
 #endif
 }
 
 UTILS_NOINLINE
-/* static */ void logProgramLinkError(io::ostream& out, char const* name, GLuint program) noexcept {
+/* static */ void logProgramLinkError(char const* name, GLuint program) noexcept {
     GLint length = 0;
     glGetProgramiv(program, GL_INFO_LOG_LENGTH, &length);
 
     CString infoLog(length);
     glGetProgramInfoLog(program, length, nullptr, infoLog.data());
 
-    out << "Link error in \"" << name << "\":\n"
+    LOG(ERROR) << "Link error in \"" << name << "\":\n" << "\"" << infoLog.c_str() << "\"";
-        << "\"" << infoLog.c_str() << "\"" << io::endl;
 }
 
 // If usages of the Google-style line directive are present, remove them, as some

filament/backend/src/opengl/gl_headers.cpp

@@ -70,6 +70,9 @@ PFNGLMAXSHADERCOMPILERTHREADSKHRPROC glMaxShaderCompilerThreadsKHR;
 #ifdef GL_OVR_multiview
 PFNGLFRAMEBUFFERTEXTUREMULTIVIEWOVRPROC glFramebufferTextureMultiviewOVR;
 #endif
+#ifdef GL_OVR_multiview_multisampled_render_to_texture
+PFNGLFRAMEBUFFERTEXTUREMULTISAMPLEMULTIVIEWOVRPROC glFramebufferTextureMultisampleMultiviewOVR;
+#endif
 
 #if defined(__ANDROID__) && !defined(FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2)
 // On Android, If we want to support a build system less than ANDROID_API 21, we need to
@@ -123,6 +126,9 @@ void importGLESExtensionsEntryPoints() {
 #ifdef GL_OVR_multiview
         getProcAddress(glFramebufferTextureMultiviewOVR, "glFramebufferTextureMultiviewOVR");
 #endif
+#ifdef GL_OVR_multiview_multisampled_render_to_texture
+        getProcAddress(glFramebufferTextureMultisampleMultiviewOVR, "glFramebufferTextureMultisampleMultiviewOVR");
+#endif
 #if defined(__ANDROID__) && !defined(FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2)
         getProcAddress(glDispatchCompute, "glDispatchCompute");
 #endif

filament/backend/src/opengl/gl_headers.h

@@ -154,6 +154,9 @@ extern PFNGLMAXSHADERCOMPILERTHREADSKHRPROC glMaxShaderCompilerThreadsKHR;
 #ifdef GL_OVR_multiview
 extern PFNGLFRAMEBUFFERTEXTUREMULTIVIEWOVRPROC glFramebufferTextureMultiviewOVR;
 #endif
+#ifdef GL_OVR_multiview_multisampled_render_to_texture
+extern PFNGLFRAMEBUFFERTEXTUREMULTISAMPLEMULTIVIEWOVRPROC glFramebufferTextureMultisampleMultiviewOVR;
+#endif
 #if defined(__ANDROID__) && !defined(FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2)
 extern PFNGLDISPATCHCOMPUTEPROC glDispatchCompute;
 #endif

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

@@ -96,7 +96,7 @@ CocoaExternalImage::SharedGl::~SharedGl() noexcept {
 CocoaExternalImage::CocoaExternalImage(const CVOpenGLTextureCacheRef textureCache,
         const SharedGl &sharedGl) noexcept : mSharedGl(sharedGl), mTextureCache(textureCache) {
     glGenFramebuffers(1, &mFBO);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 CocoaExternalImage::~CocoaExternalImage() noexcept {
@@ -125,7 +125,7 @@ bool CocoaExternalImage::set(CVPixelBufferRef image) noexcept {
     mTexture = createTextureFromImage(image);
     mRgbaTexture = encodeCopyRectangleToTexture2D(CVOpenGLTextureGetName(mTexture),
             CVPixelBufferGetWidth(image), CVPixelBufferGetHeight(image));
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     return true;
 }
@@ -182,33 +182,33 @@ GLuint CocoaExternalImage::encodeCopyRectangleToTexture2D(GLuint rectangle,
     // Create a texture to hold the result of the blit image.
     glBindTexture(GL_TEXTURE_2D, texture);
     glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, width, height);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     // source textures
     glBindSampler(0, mSharedGl.sampler);
     glActiveTexture(GL_TEXTURE0);
     glBindTexture(GL_TEXTURE_RECTANGLE, rectangle);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     // destination texture
     glBindFramebuffer(GL_FRAMEBUFFER, mFBO);
     glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, texture, 0);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_FRAMEBUFFER)
+    CHECK_GL_FRAMEBUFFER_STATUS(GL_FRAMEBUFFER)
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     // draw
     glViewport(0, 0, width, height);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
     glUseProgram(mSharedGl.program);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
     glDisableVertexAttribArray(0);
     glDrawArrays(GL_TRIANGLES, 0, 3);
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     mState.restore();
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     return texture;
 }

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

@@ -119,7 +119,7 @@ CocoaTouchExternalImage::CocoaTouchExternalImage(const CVOpenGLESTextureCacheRef
 
     glGenFramebuffers(1, &mFBO);
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 }
 
 CocoaTouchExternalImage::~CocoaTouchExternalImage() noexcept {
@@ -247,7 +247,7 @@ GLuint CocoaTouchExternalImage::encodeColorConversionPass(GLuint yPlaneTexture,
     glBindTexture(GL_TEXTURE_2D, texture);
     glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA8, width, height);
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     // source textures
     glBindSampler(0, mSharedGl.sampler);
@@ -261,8 +261,8 @@ GLuint CocoaTouchExternalImage::encodeColorConversionPass(GLuint yPlaneTexture,
     glBindFramebuffer(GL_FRAMEBUFFER, mFBO);
     glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, texture, 0);
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_FRAMEBUFFER)
+    CHECK_GL_FRAMEBUFFER_STATUS(GL_FRAMEBUFFER)
 
     // geometry
     glBindVertexArray(0);
@@ -275,7 +275,7 @@ GLuint CocoaTouchExternalImage::encodeColorConversionPass(GLuint yPlaneTexture,
     glUseProgram(mSharedGl.program);
     glDrawArrays(GL_TRIANGLES, 0, 3);
 
-    CHECK_GL_ERROR(utils::slog.e)
+    CHECK_GL_ERROR()
 
     mState.restore();
 

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

@@ -20,9 +20,9 @@
 
 #include <backend/Platform.h>
 
+#include <utils/Logger.h>
 #include <utils/compiler.h>
 #include <utils/debug.h>
-#include <utils/Log.h>
 #include <utils/ostream.h>
 
 #if __has_include(<android/surface_texture.h>)
@@ -59,7 +59,7 @@ void ExternalStreamManagerAndroid::destroy(ExternalStreamManagerAndroid* pExtern
 ExternalStreamManagerAndroid::ExternalStreamManagerAndroid() noexcept
         : mVm(VirtualMachineEnv::get()) {
     if (__builtin_available(android 28, *)) {
-        slog.d << "Using ASurfaceTexture" << io::endl;
+        DLOG(INFO) << "Using ASurfaceTexture";
     }
 }
 

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

@@ -255,7 +255,7 @@ void PlatformCocoaGL::destroySwapChain(Platform::SwapChain* swapChain) noexcept
 }
 
 bool PlatformCocoaGL::makeCurrent(ContextType type, SwapChain* drawSwapChain,
-        SwapChain* readSwapChain) noexcept {
+        SwapChain* readSwapChain) {
     ASSERT_PRECONDITION_NON_FATAL(drawSwapChain == readSwapChain,
             "ContextManagerCocoa does not support using distinct draw/read swap chains.");
     CocoaGLSwapChain* swapChain = (CocoaGLSwapChain*)drawSwapChain;

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

@@ -155,7 +155,7 @@ uint32_t PlatformCocoaTouchGL::getDefaultFramebufferObject() noexcept {
 }
 
 bool PlatformCocoaTouchGL::makeCurrent(ContextType type, SwapChain* drawSwapChain,
-        SwapChain* readSwapChain) noexcept {
+        SwapChain* readSwapChain) {
     ASSERT_PRECONDITION_NON_FATAL(drawSwapChain == readSwapChain,
             "PlatformCocoaTouchGL does not support using distinct draw/read swap chains.");
     CAEAGLLayer* const glLayer = (__bridge CAEAGLLayer*) drawSwapChain;

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

@@ -32,9 +32,9 @@
 #endif
 #include <utils/compiler.h>
 
-#include <utils/debug.h>
 #include <utils/Invocable.h>
-#include <utils/Log.h>
+#include <utils/Logger.h>
+#include <utils/debug.h>
 #include <utils/ostream.h>
 
 #include <algorithm>
@@ -74,7 +74,7 @@ void PlatformEGL::logEglError(const char* name) noexcept {
 }
 
 void PlatformEGL::logEglError(const char* name, EGLint error) noexcept {
-    slog.e << name << " failed with " << getEglErrorName(error) << io::endl;
+    LOG(ERROR) << name << " failed with " << getEglErrorName(error);
 }
 
 const char* PlatformEGL::getEglErrorName(EGLint error) noexcept {
@@ -101,7 +101,7 @@ void PlatformEGL::clearGlError() noexcept {
     // clear GL error that may have been set by previous calls
     GLenum const error = glGetError();
     if (error != GL_NO_ERROR) {
-        slog.w << "Ignoring pending GL error " << io::hex << error << io::endl;
+        LOG(WARNING) << "Ignoring pending GL error " << io::hex << error;
     }
 }
 
@@ -142,7 +142,7 @@ Driver* PlatformEGL::createDriver(void* sharedContext, const DriverConfig& drive
     }
 
     if (UTILS_UNLIKELY(!initialized)) {
-        slog.e << "eglInitialize failed" << io::endl;
+        LOG(ERROR) << "eglInitialize failed";
         return nullptr;
     }
 
@@ -577,18 +577,18 @@ OpenGLPlatform::ContextType PlatformEGL::getCurrentContextType() const noexcept
 }
 
 bool PlatformEGL::makeCurrent(ContextType type,
-        SwapChain* drawSwapChain, SwapChain* readSwapChain) noexcept {
+        SwapChain* drawSwapChain, SwapChain* readSwapChain) {
     SwapChainEGL const* const dsc = static_cast<SwapChainEGL const*>(drawSwapChain);
     SwapChainEGL const* const rsc = static_cast<SwapChainEGL const*>(readSwapChain);
     EGLContext context = getContextForType(type);
     EGLBoolean const success = egl.makeCurrent(context, dsc->sur, rsc->sur);
-    return success == EGL_TRUE ? true : false;
+    return success == EGL_TRUE;
 }
 
 void PlatformEGL::makeCurrent(SwapChain* drawSwapChain,
         SwapChain* readSwapChain,
         Invocable<void()> preContextChange,
-        Invocable<void(size_t index)> postContextChange) noexcept {
+        Invocable<void(size_t index)> postContextChange) {
 
     assert_invariant(drawSwapChain);
     assert_invariant(readSwapChain);
@@ -796,7 +796,7 @@ void PlatformEGL::Config::erase(EGLint name) noexcept {
 // ------------------------------------------------------------------------------------------------
 
 EGLBoolean PlatformEGL::EGL::makeCurrent(EGLContext context, EGLSurface drawSurface,
-        EGLSurface readSurface) noexcept {
+        EGLSurface readSurface) {
     if (UTILS_UNLIKELY((
             mCurrentContext != context ||
             drawSurface != mCurrentDrawSurface || readSurface != mCurrentReadSurface))) {

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

@@ -29,13 +29,12 @@
 #include <android/native_window.h>
 #include <android/hardware_buffer.h>
 
+#include <utils/Logger.h>
+#include <utils/Panic.h>
 #include <utils/android/PerformanceHintManager.h>
-
+#include <utils/compiler.h>
 #include <utils/compiler.h>
 #include <utils/ostream.h>
-#include <utils/Panic.h>
-#include <utils/Log.h>
-#include <utils/compiler.h>
 #include <utils/ostream.h>
 
 #include <EGL/egl.h>
@@ -135,7 +134,7 @@ static constexpr const std::string_view kNativeWindowInvalidMsg =
 
 bool PlatformEGLAndroid::makeCurrent(ContextType type,
         SwapChain* drawSwapChain,
-        SwapChain* readSwapChain) noexcept {
+        SwapChain* readSwapChain) {
 
     // fast & safe path
     if (UTILS_LIKELY(!mAssertNativeWindowIsValid)) {
@@ -327,7 +326,7 @@ bool PlatformEGLAndroid::setImage(ExternalImageEGLAndroid const* eglExternalImag
             EGL_NATIVE_BUFFER_ANDROID, clientBuffer, imageAttrs);
     if (eglImage == EGL_NO_IMAGE_KHR) {
         // Handle error
-        slog.e << "Failed to create EGL image" << io::endl;
+        LOG(ERROR) << "Failed to create EGL image";
         glDeleteTextures(1, &texture->id);
         return false;
     }
@@ -340,7 +339,7 @@ bool PlatformEGLAndroid::setImage(ExternalImageEGLAndroid const* eglExternalImag
     glBindTexture(texture->target, texture->id);
     GLenum error = glGetError();
     if (UTILS_UNLIKELY(error != GL_NO_ERROR)) {
-        slog.e << "Error after glBindTexture: " << error << io::endl;
+        LOG(ERROR) << "Error after glBindTexture: " << error;
         glDeleteTextures(1, &texture->id);
         eglDestroyImageKHR(eglGetCurrentDisplay(), eglImage);
         glActiveTexture(prevActiveTexture);
@@ -350,7 +349,7 @@ bool PlatformEGLAndroid::setImage(ExternalImageEGLAndroid const* eglExternalImag
     glEGLImageTargetTexture2DOES(texture->target, static_cast<GLeglImageOES>(eglImage));
     error = glGetError();
     if (UTILS_UNLIKELY(error != GL_NO_ERROR)) {
-        slog.e << "Error after glEGLImageTargetTexture2DOES: " << error << io::endl;
+        LOG(ERROR) << "Error after glEGLImageTargetTexture2DOES: " << error;
         glDeleteTextures(1, &texture->id);
         eglDestroyImageKHR(eglGetCurrentDisplay(), eglImage);
         glActiveTexture(prevActiveTexture);
@@ -408,7 +407,7 @@ AcquiredImage PlatformEGLAndroid::transformAcquiredImage(AcquiredImage source) n
 
     EGLClientBuffer clientBuffer = eglGetNativeClientBufferANDROID(pHardwareBuffer);
     if (!clientBuffer) {
-        slog.e << "Unable to get EGLClientBuffer from AHardwareBuffer." << io::endl;
+        LOG(ERROR) << "Unable to get EGLClientBuffer from AHardwareBuffer.";
         return {};
     }
 
@@ -427,7 +426,7 @@ AcquiredImage PlatformEGLAndroid::transformAcquiredImage(AcquiredImage source) n
     EGLImageKHR eglImage = eglCreateImageKHR(mEGLDisplay,
             EGL_NO_CONTEXT, EGL_NATIVE_BUFFER_ANDROID, clientBuffer, attributes.data());
     if (eglImage == EGL_NO_IMAGE_KHR) {
-        slog.e << "eglCreateImageKHR returned no image." << io::endl;
+        LOG(ERROR) << "eglCreateImageKHR returned no image.";
         return {};
     }
 
@@ -442,7 +441,7 @@ AcquiredImage PlatformEGLAndroid::transformAcquiredImage(AcquiredImage source) n
     auto patchedCallback = [](void* image, void* userdata) {
         Closure* closure = (Closure*)userdata;
         if (eglDestroyImageKHR(closure->display, (EGLImageKHR) image) == EGL_FALSE) {
-            slog.e << "eglDestroyImageKHR failed." << io::endl;
+            LOG(ERROR) << "eglDestroyImageKHR failed.";
         }
         closure->acquiredImage.callback(closure->acquiredImage.image, closure->acquiredImage.userData);
         delete closure;

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

@@ -21,9 +21,9 @@
 #include <EGL/egl.h>
 #include <EGL/eglext.h>
 
-#include <utils/compiler.h>
+#include <utils/Logger.h>
-#include <utils/Log.h>
 #include <utils/Panic.h>
+#include <utils/compiler.h>
 
 using namespace utils;
 
@@ -42,12 +42,12 @@ backend::Driver* PlatformEGLHeadless::createDriver(void* sharedContext,
         const Platform::DriverConfig& driverConfig) noexcept {
     EGLBoolean bindAPI = eglBindAPI(EGL_OPENGL_API);
     if (UTILS_UNLIKELY(!bindAPI)) {
-        slog.e << "eglBindAPI EGL_OPENGL_API failed" << io::endl;
+        LOG(ERROR) << "eglBindAPI EGL_OPENGL_API failed";
         return nullptr;
     }
     int bindBlueGL = bluegl::bind();
     if (UTILS_UNLIKELY(bindBlueGL != 0)) {
-        slog.e << "bluegl bind failed" << io::endl;
+        LOG(ERROR) << "bluegl bind failed";
         return nullptr;
     }
 

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

@@ -16,7 +16,7 @@
 
 #include <backend/platforms/PlatformGLX.h>
 
-#include <utils/Log.h>
+#include <utils/Logger.h>
 #include <utils/Panic.h>
 
 #include <X11/Xlib.h>
@@ -84,7 +84,7 @@ static PFNGLXGETPROCADDRESSPROC getProcAddress;
 static bool loadLibraries() {
     g_glx.library = dlopen(LIBRARY_GLX, RTLD_LOCAL | RTLD_NOW);
     if (!g_glx.library) {
-        utils::slog.e << "Could not find library " << LIBRARY_GLX << utils::io::endl;
+        LOG(ERROR) << "Could not find library " << LIBRARY_GLX;
         return false;
     }
 
@@ -115,7 +115,7 @@ static bool loadLibraries() {
 
     g_x11.library = dlopen(LIBRARY_X11, RTLD_LOCAL | RTLD_NOW);
     if (!g_x11.library) {
-        utils::slog.e << "Could not find library " << LIBRARY_X11 << utils::io::endl;
+        LOG(ERROR) << "Could not find library " << LIBRARY_X11;
         return false;
     }
 
@@ -134,7 +134,7 @@ Driver* PlatformGLX::createDriver(void* sharedGLContext,
     // Get the display device
     mGLXDisplay = g_x11.openDisplay(NULL);
     if (mGLXDisplay == nullptr) {
-        utils::slog.e << "Failed to open X display. (exiting)." << utils::io::endl;
+        LOG(ERROR) << "Failed to open X display. (exiting).";
         exit(1);
     }
 
@@ -145,8 +145,7 @@ Driver* PlatformGLX::createDriver(void* sharedGLContext,
 
         r = g_glx.queryContext(mGLXDisplay, sharedCtx, GLX_FBCONFIG_ID, &usedFbId);
         if (r != 0) {
-            utils::slog.e << "Failed to get GLX_FBCONFIG_ID from shared GL context."
+            LOG(ERROR) << "Failed to get GLX_FBCONFIG_ID from shared GL context.";
-                          << utils::io::endl;
             return nullptr;
         }
 
@@ -154,7 +153,7 @@ Driver* PlatformGLX::createDriver(void* sharedGLContext,
         GLXFBConfig* fbConfigs = g_glx.getFbConfigs(mGLXDisplay, 0, &numConfigs);
 
         if (fbConfigs == nullptr) {
-            utils::slog.e << "Failed to get the available GLXFBConfigs." << utils::io::endl;
+            LOG(ERROR) << "Failed to get the available GLXFBConfigs.";
             return nullptr;
         }
 
@@ -164,8 +163,7 @@ Driver* PlatformGLX::createDriver(void* sharedGLContext,
         for (int i = 0; i < numConfigs; ++i) {
             r = g_glx.getFbConfigAttrib(mGLXDisplay, fbConfigs[i], GLX_FBCONFIG_ID, &fbId);
             if (r != 0) {
-                utils::slog.e << "Failed to get GLX_FBCONFIG_ID for entry " << i << "."
+                LOG(ERROR) << "Failed to get GLX_FBCONFIG_ID for entry " << i << ".";
-                              << utils::io::endl;
                 continue;
             }
 
@@ -176,8 +174,7 @@ Driver* PlatformGLX::createDriver(void* sharedGLContext,
         }
 
         if (fbIndex < 0) {
-            utils::slog.e << "Failed to find an `GLXFBConfig` with the requested ID."
+            LOG(ERROR) << "Failed to find an `GLXFBConfig` with the requested ID.";
-                          << utils::io::endl;
             return nullptr;
         }
 
@@ -202,8 +199,7 @@ Driver* PlatformGLX::createDriver(void* sharedGLContext,
             getProcAddress((GLubyte*)"glXCreateContextAttribsARB");
 
     if (glXCreateContextAttribs == nullptr) {
-        utils::slog.i << "Unable to retrieve function pointer for `glXCreateContextAttribs()`."
+        LOG(INFO) << "Unable to retrieve function pointer for `glXCreateContextAttribs()`.";
-                      << utils::io::endl;
         return nullptr;
     }
 
@@ -266,7 +262,7 @@ void PlatformGLX::destroySwapChain(Platform::SwapChain* swapChain) noexcept {
 }
 
 bool PlatformGLX::makeCurrent(ContextType type, SwapChain* drawSwapChain,
-        SwapChain* readSwapChain) noexcept {
+        SwapChain* readSwapChain) {
     g_glx.setCurrentContext(mGLXDisplay,
             (GLXDrawable)drawSwapChain, (GLXDrawable)readSwapChain, mGLXContext);
     return true;

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

@@ -169,7 +169,7 @@ void PlatformOSMesa::destroySwapChain(Platform::SwapChain* swapChain) noexcept {
 }
 
 bool PlatformOSMesa::makeCurrent(ContextType type, SwapChain* drawSwapChain,
-        SwapChain* readSwapChain) noexcept {
+        SwapChain* readSwapChain) {
     OSMesaAPI* api = (OSMesaAPI*) mOsMesaApi;
     OSMesaSwapchain* impl = (OSMesaSwapchain*) drawSwapChain;
 

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

@@ -32,7 +32,7 @@
 #include "GL/glext.h"
 #include "GL/wglext.h"
 
-#include <utils/Log.h>
+#include <utils/Logger.h>
 #include <utils/Panic.h>
 
 namespace {
@@ -55,8 +55,7 @@ void reportWindowsError(DWORD dwError) {
         0, nullptr
 	);
 
-    utils::slog.e << "Windows error code: " << dwError << ". " << lpMessageBuffer
+    LOG(ERROR) << "Windows error code: " << dwError << ". " << lpMessageBuffer;
-            << utils::io::endl;
 
     LocalFree(lpMessageBuffer);
 }
@@ -106,7 +105,7 @@ Driver* PlatformWGL::createDriver(void* sharedGLContext,
     HDC whdc = mWhdc = GetDC(mHWnd);
     if (whdc == NULL) {
         dwError = GetLastError();
-        utils::slog.e << "CreateWindowA() failed" << utils::io::endl;
+        LOG(ERROR) << "CreateWindowA() failed";
         goto error;
     }
 
@@ -117,8 +116,7 @@ Driver* PlatformWGL::createDriver(void* sharedGLContext,
     tempContext = wglCreateContext(whdc);
     if (!wglMakeCurrent(whdc, tempContext)) {
         dwError = GetLastError();
-        utils::slog.e << "wglMakeCurrent() failed, whdc=" << whdc << ", tempContext=" <<
+        LOG(ERROR) << "wglMakeCurrent() failed, whdc=" << whdc << ", tempContext=" << tempContext;
-                tempContext << utils::io::endl;
         goto error;
     }
 
@@ -142,7 +140,7 @@ Driver* PlatformWGL::createDriver(void* sharedGLContext,
     }
 
     if (!mContext) {
-        utils::slog.e << "wglCreateContextAttribs() failed, whdc=" << whdc << utils::io::endl;
+        LOG(ERROR) << "wglCreateContextAttribs() failed, whdc=" << whdc;
         goto error;
     }
 
@@ -152,8 +150,7 @@ Driver* PlatformWGL::createDriver(void* sharedGLContext,
 
     if (!wglMakeCurrent(whdc, mContext)) {
         dwError = GetLastError();
-        utils::slog.e << "wglMakeCurrent() failed, whdc=" << whdc << ", mContext=" <<
+        LOG(ERROR) << "wglMakeCurrent() failed, whdc=" << whdc << ", mContext=" << mContext;
-                mContext << utils::io::endl;
         goto error;
     }
 
@@ -262,7 +259,7 @@ void PlatformWGL::destroySwapChain(Platform::SwapChain* swapChain) noexcept {
 }
 
 bool PlatformWGL::makeCurrent(ContextType type, SwapChain* drawSwapChain,
-        SwapChain* readSwapChain) noexcept {
+        SwapChain* readSwapChain) {
     ASSERT_PRECONDITION_NON_FATAL(drawSwapChain == readSwapChain,
                                   "PlatformWGL does not support distinct draw/read swap chains.");
 

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

@@ -47,7 +47,7 @@ void PlatformWebGL::destroySwapChain(Platform::SwapChain* swapChain) noexcept {
 }
 
 bool PlatformWebGL::makeCurrent(ContextType type, SwapChain* drawSwapChain,
-        SwapChain* readSwapChain) noexcept {
+        SwapChain* readSwapChain) {
     return true;
 }
 

filament/backend/src/vulkan/VulkanBlitter.cpp

@@ -40,7 +40,7 @@ inline void blitFast(VulkanCommandBuffer* commands, VkImageAspectFlags aspect, V
         FVK_LOGD << "Fast blit from=" << src.texture->getVkImage() << ",level=" << (int) src.level
                       << " layout=" << src.getLayout()
                       << " to=" << dst.texture->getVkImage() << ",level=" << (int) dst.level
-                      << " layout=" << dst.getLayout() << utils::io::endl;
+                      << " layout=" << dst.getLayout();
     }
 
     VkImageSubresourceRange const srcRange = src.getSubresourceRange();
@@ -80,7 +80,7 @@ inline void resolveFast(VulkanCommandBuffer* commands, VkImageAspectFlags aspect
         FVK_LOGD << "Fast blit from=" << src.texture->getVkImage() << ",level=" << (int) src.level
                       << " layout=" << src.getLayout()
                       << " to=" << dst.texture->getVkImage() << ",level=" << (int) dst.level
-                      << " layout=" << dst.getLayout() << utils::io::endl;
+                      << " layout=" << dst.getLayout();
     }
 
     VkImageSubresourceRange const srcRange = src.getSubresourceRange();

filament/backend/src/vulkan/VulkanBuffer.h

@@ -17,35 +17,36 @@
 #ifndef TNT_FILAMENT_BACKEND_VULKANBUFFER_H
 #define TNT_FILAMENT_BACKEND_VULKANBUFFER_H
 
-#include "VulkanContext.h"
-#include "VulkanStagePool.h"
 #include "VulkanMemory.h"
+#include "memory/Resource.h"
+
+#include <functional>
 
 namespace filament::backend {
 
-// Encapsulates a Vulkan buffer, its attached DeviceMemory and a staging area.
+class VulkanBuffer : public fvkmemory::Resource {
-class VulkanBuffer {
 public:
-    VulkanBuffer(VmaAllocator allocator, VulkanStagePool& stagePool, VkBufferUsageFlags usage,
+    // Because we need to recycle the unused `VulkanGpuBuffer`, we allow for a callback that the
-            uint32_t numBytes);
+    // "Pool" can use to acquire the buffer back.
-    ~VulkanBuffer();
+    using OnRecycle = std::function<void(VulkanGpuBuffer const*)>;
-    void loadFromCpu(VkCommandBuffer cmdbuf, const void* cpuData, uint32_t byteOffset,
+
-            uint32_t numBytes);
+    VulkanBuffer(VulkanGpuBuffer const* gpuBuffer, OnRecycle&& onRecycleFn)
-    VkBuffer getGpuBuffer() const {
+        : mGpuBuffer(gpuBuffer),
-        return mGpuBuffer;
+          mOnRecycleFn(onRecycleFn) {}
+
+    ~VulkanBuffer() {
+        if (mOnRecycleFn) {
+            mOnRecycleFn(mGpuBuffer);
+        }
     }
 
-private:
+    VulkanGpuBuffer const* getGpuBuffer() const { return mGpuBuffer; }
-    VmaAllocator mAllocator;
-    VulkanStagePool& mStagePool;
 
-    VmaAllocation mGpuMemory = VK_NULL_HANDLE;
+private:
-    VkBuffer mGpuBuffer = VK_NULL_HANDLE;
+    VulkanGpuBuffer const* mGpuBuffer;
-    VkBufferUsageFlags mUsage = {};
+    OnRecycle mOnRecycleFn;
-	uint32_t mUpdatedOffset = 0;
-    uint32_t mUpdatedBytes = 0;
 };
 
-} // namespace filament::backend
+}// namespace filament::backend
 
-#endif // TNT_FILAMENT_BACKEND_VULKANBUFFER_H
+#endif// TNT_FILAMENT_BACKEND_VULKANBUFFER_H

filament/backend/src/vulkan/VulkanBufferCache.cpp

@@ -0,0 +1,207 @@
+/*
+ * 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 "VulkanBufferCache.h"
+
+#include "VulkanBuffer.h"
+#include "VulkanConstants.h"
+#include "VulkanMemory.h"
+#include "memory/Resource.h"
+#include "memory/ResourceManager.h"
+
+#include <utility>
+
+namespace filament::backend {
+
+namespace {
+
+VkBufferUsageFlags getVkBufferUsage(VulkanBufferUsage usage) {
+    switch (usage) {
+        case VulkanBufferUsage::VERTEX:
+            return VK_BUFFER_USAGE_VERTEX_BUFFER_BIT;
+        case VulkanBufferUsage::INDEX:
+            return VK_BUFFER_USAGE_INDEX_BUFFER_BIT;
+        case VulkanBufferUsage::UNIFORM:
+            return VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT;
+        case VulkanBufferUsage::SHADER_STORAGE:
+            return VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
+        case VulkanBufferUsage::UNKNOWN:
+            return 0;
+    }
+
+    return 0;
+}
+
+}// namespace
+
+VulkanBufferCache::VulkanBufferCache(VulkanContext const& context,
+        fvkmemory::ResourceManager& resourceManager, VmaAllocator allocator)
+    : mContext(context),
+      mResourceManager(resourceManager),
+      mAllocator(allocator) {}
+
+fvkmemory::resource_ptr<VulkanBuffer> VulkanBufferCache::acquire(VulkanBufferUsage usage,
+        uint32_t numBytes) noexcept {
+    assert_invariant(usage != VulkanBufferUsage::UNKNOWN);
+
+    BufferPool& bufferPool = getPool(usage);
+
+    // First check if an allocation exists whose capacity is greater than or equal to the requested
+    // size.
+    auto iter = bufferPool.lower_bound(numBytes);
+    if (iter != bufferPool.end()) {
+        VulkanGpuBuffer const* gpuBuffer = iter->second.gpuBuffer;
+        bufferPool.erase(iter);
+        return fvkmemory::resource_ptr<VulkanBuffer>::construct(&mResourceManager, gpuBuffer,
+                [this](VulkanGpuBuffer const* gpuBuffer) { this->release(gpuBuffer); });
+    }
+
+    // We were not able to find a sufficiently large allocation, so create a new one that is
+    // recycled after being yielded.
+    VulkanGpuBuffer const* gpuBuffer = allocate(usage, numBytes);
+    return fvkmemory::resource_ptr<VulkanBuffer>::construct(&mResourceManager, gpuBuffer,
+            [this](VulkanGpuBuffer const* gpuBuffer) { this->release(gpuBuffer); });
+}
+
+void VulkanBufferCache::gc() noexcept {
+    FVK_SYSTRACE_CONTEXT();
+    FVK_SYSTRACE_START("VulkanBufferCache::gc");
+
+    // If this is one of the first few frames, return early to avoid wrapping unsigned integers.
+    constexpr uint32_t TIME_BEFORE_EVICTION = 3;
+    if (++mCurrentFrame <= TIME_BEFORE_EVICTION) {
+        return;
+    }
+    const uint64_t evictionTime = mCurrentFrame - TIME_BEFORE_EVICTION;
+
+    // Destroy buffers that have not been used for several frames.
+    for (auto& bufferPool: mGpuBufferPools) {
+        for (auto poolIter = bufferPool.begin(); poolIter != bufferPool.end();) {
+            if (poolIter->second.lastAccessed < evictionTime) {
+#if FVK_ENABLED(FVK_DEBUG_VULKAN_BUFFER_CACHE)
+                FVK_LOGD << "VulkanBufferCache - Destroyed vkBuffer "
+                         << poolIter->second.gpuBuffer->vkbuffer << " with usage "
+                         << static_cast<int>(poolIter->second.gpuBuffer->usage) << utils::io::endl;
+#endif// FVK_DEBUG_VULKAN_BUFFER_CACHE
+
+                destroy(poolIter->second.gpuBuffer);
+                poolIter = bufferPool.erase(poolIter);
+            } else {
+                ++poolIter;
+            }
+        }
+    }
+
+    FVK_SYSTRACE_END();
+}
+
+void VulkanBufferCache::terminate() noexcept {
+    for (auto& bufferPool: mGpuBufferPools) {
+        for (auto& poolEntry: bufferPool) {
+            destroy(poolEntry.second.gpuBuffer);
+        }
+        bufferPool.clear();
+    }
+}
+
+void VulkanBufferCache::release(VulkanGpuBuffer const* gpuBuffer) noexcept {
+    assert_invariant(gpuBuffer != nullptr);
+
+    BufferPool& bufferPool = getPool(gpuBuffer->usage);
+    bufferPool.insert(std::make_pair(gpuBuffer->numBytes, UnusedGpuBuffer{
+                                                              .lastAccessed = mCurrentFrame,
+                                                              .gpuBuffer = gpuBuffer,
+                                                          }));
+}
+
+VulkanGpuBuffer const* VulkanBufferCache::allocate(VulkanBufferUsage usage,
+        uint32_t numBytes) noexcept {
+    VkBufferCreateInfo const bufferInfo{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+        .size = numBytes,
+        // `VK_BUFFER_USAGE_TRANSFER_DST_BIT` is needed to allow updating the buffer through
+        // a staging using `vkCmdCopyBuffer`.
+        .usage = getVkBufferUsage(usage) | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+    };
+
+    VmaAllocationCreateFlags vmaFlags = 0;
+    if (usage == VulkanBufferUsage::UNIFORM) {
+        // In the case of UMA, the uniform buffers will always be mappable
+        if (mContext.isUnifiedMemoryArchitecture()) {
+            vmaFlags |= VMA_ALLOCATION_CREATE_MAPPED_BIT |
+                        VMA_ALLOCATION_CREATE_HOST_ACCESS_SEQUENTIAL_WRITE_BIT;
+        }
+    }
+
+    VulkanGpuBuffer* gpuBuffer = new VulkanGpuBuffer{
+        .numBytes = numBytes,
+        .usage = usage,
+    };
+    VmaAllocationCreateInfo const allocInfo{
+        .flags = vmaFlags,
+        .usage = VMA_MEMORY_USAGE_AUTO,
+        .requiredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
+    };
+    UTILS_UNUSED_IN_RELEASE VkResult result = vmaCreateBuffer(mAllocator, &bufferInfo, &allocInfo,
+            &gpuBuffer->vkbuffer, &gpuBuffer->vmaAllocation, &gpuBuffer->allocationInfo);
+
+#if FVK_ENABLED(FVK_DEBUG_VULKAN_BUFFER_CACHE)
+    if (result != VK_SUCCESS) {
+        FVK_LOGE << "VulkanBufferCache - failed to allocate a new vkBuffer of size " << numBytes
+                 << " and usage " << static_cast<int>(usage) << ", error: " << result
+                 << utils::io::endl;
+    } else {
+        FVK_LOGD << "VulkanBufferCache - allocated a vkBuffer " << gpuBuffer->vkbuffer
+                 << " of size " << numBytes << " and usage = " << static_cast<int>(usage)
+                 << "  successfully" << utils::io::endl;
+    }
+#endif// FVK_DEBUG_VULKAN_BUFFER_CACHE
+
+    return gpuBuffer;
+}
+
+void VulkanBufferCache::destroy(VulkanGpuBuffer const* gpuBuffer) noexcept {
+    vmaDestroyBuffer(mAllocator, gpuBuffer->vkbuffer, gpuBuffer->vmaAllocation);
+    delete gpuBuffer;
+    gpuBuffer = nullptr;
+}
+
+VulkanBufferCache::BufferPool& VulkanBufferCache::getPool(VulkanBufferUsage usage) noexcept {
+
+    int poolIndex = -1;
+    switch (usage) {
+        case VulkanBufferUsage::VERTEX:
+            poolIndex = 0;
+            break;
+        case VulkanBufferUsage::INDEX:
+            poolIndex = 1;
+            break;
+        case VulkanBufferUsage::UNIFORM:
+            poolIndex = 2;
+            break;
+        case VulkanBufferUsage::SHADER_STORAGE:
+            poolIndex = 3;
+            break;
+        case VulkanBufferUsage::UNKNOWN:
+            PANIC_LOG("There's no pool for buffers with unkown usage.");
+            break;
+    }
+
+    assert_invariant(poolIndex >= 0 && poolIndex < MAX_POOL_COUNT);
+    return mGpuBufferPools[poolIndex];
+}
+
+}// namespace filament::backend

filament/backend/src/vulkan/VulkanBufferCache.h

@@ -0,0 +1,84 @@
+/*
+ * 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_VULKANBUFFERCACHE_H
+#define TNT_FILAMENT_BACKEND_VULKANBUFFERCACHE_H
+
+#include "VulkanBuffer.h"
+#include "VulkanContext.h"
+#include "VulkanMemory.h"
+#include "memory/Resource.h"
+#include "memory/ResourceManager.h"
+
+#include <map>
+
+namespace filament::backend {
+
+class VulkanBufferCache {
+public:
+    VulkanBufferCache(VulkanContext const& context, fvkmemory::ResourceManager& resourceManager,
+            VmaAllocator allocator);
+
+    // `VulkanBufferCache` is not copyable.
+    VulkanBufferCache(const VulkanBufferCache&) = delete;
+    VulkanBufferCache& operator=(const VulkanBufferCache&) = delete;
+
+    // Allocates or reuse a new VkBuffer that is device local.
+    // In the case of Unified memory architecture, uniform buffers are also host visible.
+    fvkmemory::resource_ptr<VulkanBuffer> acquire(VulkanBufferUsage usage,
+            uint32_t numBytes) noexcept;
+
+    // Evicts old unused `VulkanGpuBuffer` and bumps the current frame number
+    void gc() noexcept;
+
+    // Destroys all unused `VulkanGpuBuffer`.
+    // This should be called while the context's VkDevice is still alive.
+    void terminate() noexcept;
+
+private:
+    struct UnusedGpuBuffer {
+        uint64_t lastAccessed;
+        VulkanGpuBuffer const* gpuBuffer;
+    };
+
+    using BufferPool = std::multimap<uint32_t, UnusedGpuBuffer>;
+
+    // Return a `VulkanGpuBuffer` back to its corresponding pool
+    void release(VulkanGpuBuffer const* gpuBuffer) noexcept;
+
+    // Allocate a new VkBuffer from the VMA pool of the corresponding `numBytes` and `usage`.
+    VulkanGpuBuffer const* allocate(VulkanBufferUsage usage, uint32_t numBytes) noexcept;
+
+    // Destroy the corresponding VkBuffer and return the VkDeviceMemory to the VMA pool.
+    void destroy(VulkanGpuBuffer const* gpuBuffer) noexcept;
+
+    BufferPool& getPool(VulkanBufferUsage usage) noexcept;
+
+    VulkanContext const& mContext;
+    fvkmemory::ResourceManager& mResourceManager;
+    VmaAllocator mAllocator;
+
+    // Buffers can be recycled, after they are released. Each type of buffer have its own pool
+    static constexpr int MAX_POOL_COUNT = 4;
+    BufferPool mGpuBufferPools[MAX_POOL_COUNT];
+
+    // Store the current "time" (really just a frame count) and LRU eviction parameters.
+    uint64_t mCurrentFrame = 0;
+};
+
+}// namespace filament::backend
+
+#endif// TNT_FILAMENT_BACKEND_VULKANBUFFERCACHE_H

filament/backend/src/vulkan/VulkanBufferProxy.cpp

@@ -14,49 +14,35 @@
  * limitations under the License.
  */
 
-#include "VulkanBuffer.h"
+#include "VulkanBufferProxy.h"
+#include "VulkanCommands.h"
 #include "VulkanMemory.h"
 
-#include <utils/Panic.h>
+#include "VulkanBufferCache.h"
+#include "VulkanMemory.h"
 
 using namespace bluevk;
 
 namespace filament::backend {
 
-VulkanBuffer::VulkanBuffer(VmaAllocator allocator, VulkanStagePool& stagePool,
+VulkanBufferProxy::VulkanBufferProxy(VmaAllocator allocator, VulkanStagePool& stagePool,
-        VkBufferUsageFlags usage, uint32_t numBytes)
+        VulkanBufferCache& bufferCache, VulkanBufferUsage usage, uint32_t numBytes)
     : mAllocator(allocator),
       mStagePool(stagePool),
-      mUsage(usage),
+      mBufferCache(bufferCache),
+      mBuffer(mBufferCache.acquire(usage, numBytes)),
       mUpdatedOffset(0),
-      mUpdatedBytes(0) {
+      mUpdatedBytes(0) {}
-    // for now make sure that only 1 bit is set in usage
-    // (because loadFromCpu() assumes that somewhat)
-    assert_invariant(usage && !(usage & (usage - 1)));
 
-    // Create the VkBuffer.
+void VulkanBufferProxy::loadFromCpu(VulkanCommandBuffer& commands, const void* cpuData,
-    VkBufferCreateInfo bufferInfo {
+        uint32_t byteOffset, uint32_t numBytes) {
-        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+    // Note: this should be stored within the command buffer before going out of
-        .size = numBytes,
+    // scope, so that the command buffer can manage its lifecycle.
-        .usage = usage | VK_BUFFER_USAGE_TRANSFER_DST_BIT
+    fvkmemory::resource_ptr<VulkanStage::Segment> stage = mStagePool.acquireStage(numBytes);
-    };
+    assert_invariant(stage->memory());
-
+    commands.acquire(stage);
-    VmaAllocationCreateInfo allocInfo { .usage = VMA_MEMORY_USAGE_GPU_ONLY };
+    memcpy(stage->mapping(), cpuData, numBytes);
-    vmaCreateBuffer(mAllocator, &bufferInfo, &allocInfo, &mGpuBuffer, &mGpuMemory, nullptr);
+    vmaFlushAllocation(mAllocator, stage->memory(), stage->offset(), numBytes);
-}
-
-VulkanBuffer::~VulkanBuffer() {
-    vmaDestroyBuffer(mAllocator, mGpuBuffer, mGpuMemory);
-}
-
-void VulkanBuffer::loadFromCpu(VkCommandBuffer cmdbuf, const void* cpuData, uint32_t byteOffset,
-        uint32_t numBytes) {
-    VulkanStage const* stage = mStagePool.acquireStage(numBytes);
-    void* mapped;
-    vmaMapMemory(mAllocator, stage->memory, &mapped);
-    memcpy(mapped, cpuData, numBytes);
-    vmaUnmapMemory(mAllocator, stage->memory);
-    vmaFlushAllocation(mAllocator, stage->memory, 0, numBytes);
 
     // If there was a previous update, then we need to make sure the following write is properly
     // synced with the previous read.
@@ -64,13 +50,13 @@ void VulkanBuffer::loadFromCpu(VkCommandBuffer cmdbuf, const void* cpuData, uint
             (byteOffset >= mUpdatedOffset && byteOffset <= (mUpdatedOffset + mUpdatedBytes))) {
         VkAccessFlags srcAccess = 0;
         VkPipelineStageFlags srcStage = 0;
-        if (mUsage & VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT) {
+        if (getUsage() == VulkanBufferUsage::UNIFORM) {
             srcAccess = VK_ACCESS_SHADER_READ_BIT;
             srcStage = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT;
-        } else if (mUsage & VK_BUFFER_USAGE_VERTEX_BUFFER_BIT) {
+        } else if (getUsage() == VulkanBufferUsage::VERTEX) {
             srcAccess = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
             srcStage = VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
-        } else if (mUsage & VK_BUFFER_USAGE_INDEX_BUFFER_BIT) {
+        } else if (getUsage() == VulkanBufferUsage::INDEX) {
             srcAccess = VK_ACCESS_INDEX_READ_BIT;
             srcStage = VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
         }
@@ -81,20 +67,20 @@ void VulkanBuffer::loadFromCpu(VkCommandBuffer cmdbuf, const void* cpuData, uint
             .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
             .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
             .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-            .buffer = mGpuBuffer,
+            .buffer = getVkBuffer(),
             .offset = byteOffset,
             .size = numBytes,
         };
-        vkCmdPipelineBarrier(cmdbuf, srcStage, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 1,
+        vkCmdPipelineBarrier(commands.buffer(), srcStage, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0,
-                &barrier, 0, nullptr);
+                nullptr, 1, &barrier, 0, nullptr);
     }
 
     VkBufferCopy region = {
-        .srcOffset = 0,
+        .srcOffset = stage->offset(),
         .dstOffset = byteOffset,
         .size = numBytes,
     };
-    vkCmdCopyBuffer(cmdbuf, stage->buffer, mGpuBuffer, 1, &region);
+    vkCmdCopyBuffer(commands.buffer(), stage->buffer(), getVkBuffer(), 1, &region);
 
     mUpdatedOffset = byteOffset;
     mUpdatedBytes = numBytes;
@@ -106,16 +92,16 @@ void VulkanBuffer::loadFromCpu(VkCommandBuffer cmdbuf, const void* cpuData, uint
     VkAccessFlags dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
     VkPipelineStageFlags dstStageMask = VK_PIPELINE_STAGE_TRANSFER_BIT;
 
-    if (mUsage & VK_BUFFER_USAGE_VERTEX_BUFFER_BIT) {
+    if (getUsage() == VulkanBufferUsage::VERTEX) {
         dstAccessMask |= VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
         dstStageMask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
-    } else if (mUsage & VK_BUFFER_USAGE_INDEX_BUFFER_BIT) {
+    } else if (getUsage() == VulkanBufferUsage::INDEX) {
         dstAccessMask |= VK_ACCESS_INDEX_READ_BIT;
         dstStageMask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
-    } else if (mUsage & VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT) {
+    } else if (getUsage() == VulkanBufferUsage::UNIFORM) {
         dstAccessMask |= VK_ACCESS_SHADER_READ_BIT;
         dstStageMask |= VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT;
-    } else if (mUsage & VK_BUFFER_USAGE_STORAGE_BUFFER_BIT) {
+    } else if (getUsage() == VulkanBufferUsage::SHADER_STORAGE) {
         // TODO: implement me
     }
 
@@ -125,13 +111,21 @@ void VulkanBuffer::loadFromCpu(VkCommandBuffer cmdbuf, const void* cpuData, uint
         .dstAccessMask = dstAccessMask,
         .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
         .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
-        .buffer = mGpuBuffer,
+        .buffer = getVkBuffer(),
         .offset = byteOffset,
         .size = numBytes,
     };
 
-    vkCmdPipelineBarrier(cmdbuf, VK_PIPELINE_STAGE_TRANSFER_BIT, dstStageMask, 0, 0, nullptr, 1,
+    vkCmdPipelineBarrier(commands.buffer(), VK_PIPELINE_STAGE_TRANSFER_BIT, dstStageMask, 0, 0,
-            &barrier, 0, nullptr);
+            nullptr, 1, &barrier, 0, nullptr);
 }
 
-} // namespace filament::backend
+VkBuffer VulkanBufferProxy::getVkBuffer() const noexcept {
+    return mBuffer->getGpuBuffer()->vkbuffer;
+}
+
+VulkanBufferUsage VulkanBufferProxy::getUsage() const noexcept {
+    return mBuffer->getGpuBuffer()->usage;
+}
+
+}// namespace filament::backend

filament/backend/src/vulkan/VulkanBufferProxy.h

@@ -0,0 +1,54 @@
+/*
+ * Copyright (C) 2018 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_VULKANBUFFERPROXY_H
+#define TNT_FILAMENT_BACKEND_VULKANBUFFERPROXY_H
+
+#include "VulkanBufferCache.h"
+#include "VulkanCommands.h"
+#include "VulkanContext.h"
+#include "VulkanMemory.h"
+#include "VulkanStagePool.h"
+
+namespace filament::backend {
+
+// This class acts as a dynamic wrapper for a `VulkanBuffer`. It allows you to modify the
+// `VulkanBuffer` it references at runtime, wihtout affecting any external objects.
+class VulkanBufferProxy {
+public:
+    VulkanBufferProxy(VmaAllocator allocator, VulkanStagePool& stagePool,
+            VulkanBufferCache& bufferCache, VulkanBufferUsage usage, uint32_t numBytes);
+
+    void loadFromCpu(VulkanCommandBuffer& commands, const void* cpuData, uint32_t byteOffset,
+            uint32_t numBytes);
+
+    VkBuffer getVkBuffer() const noexcept;
+
+    VulkanBufferUsage getUsage() const noexcept;
+
+private:
+    VmaAllocator mAllocator;
+    VulkanStagePool& mStagePool;
+    VulkanBufferCache& mBufferCache;
+
+    fvkmemory::resource_ptr<VulkanBuffer> mBuffer;
+    uint32_t mUpdatedOffset = 0;
+    uint32_t mUpdatedBytes = 0;
+};
+
+}// namespace filament::backend
+
+#endif// TNT_FILAMENT_BACKEND_VULKANBUFFERPROXY_H

filament/backend/src/vulkan/VulkanCommands.cpp

@@ -215,7 +215,7 @@ VkSemaphore VulkanCommandBuffer::submit() {
     }
     FVK_LOGI << ") "
              << " signal=" << mSubmission
-             << " fence=" << mFence << utils::io::endl;
+             << " fence=" << mFence;
 #endif
 
     mFenceStatus->setStatus(VK_NOT_READY);
@@ -224,7 +224,7 @@ VkSemaphore VulkanCommandBuffer::submit() {
 
 #if FVK_ENABLED(FVK_DEBUG_COMMAND_BUFFER)
     if (result != VK_SUCCESS) {
-        FVK_LOGD << "Failed command buffer submission result: " << result << utils::io::endl;
+        FVK_LOGD << "Failed command buffer submission result: " << result;
     }
 #endif
     assert_invariant(result == VK_SUCCESS);
@@ -490,7 +490,7 @@ void VulkanCommands::pushGroupMarker(char const* str, VulkanGroupMarkers::Timest
         mProtectedPool->pushMarker(str, timestamp);
     }
 #if FVK_ENABLED(FVK_DEBUG_PRINT_GROUP_MARKERS)
-    FVK_LOGD << "----> " << str << utils::io::endl;
+    FVK_LOGD << "----> " << str;
 #endif
 }
 
@@ -502,8 +502,7 @@ void VulkanCommands::popGroupMarker() {
     auto const& startTime = ret.second;
     auto const endTime = std::chrono::high_resolution_clock::now();
     std::chrono::duration<double> diff = endTime - startTime;
-    FVK_LOGD << "<---- " << marker << " elapsed: " << (diff.count() * 1000) << " ms"
+    FVK_LOGD << "<---- " << marker << " elapsed: " << (diff.count() * 1000) << " ms";
-             << utils::io::endl;
 #else
     mPool->popMarker();
 #endif // FVK_DEBUG_PRINT_GROUP_MARKERS

filament/backend/src/vulkan/VulkanConstants.h

@@ -17,7 +17,7 @@
 #ifndef TNT_FILAMENT_BACKEND_VULKANCONSTANTS_H
 #define TNT_FILAMENT_BACKEND_VULKANCONSTANTS_H
 
-#include <utils/Log.h>
+#include <utils/Logger.h>
 
 #include <stdint.h>
 
@@ -73,14 +73,16 @@
 #define FVK_DEBUG_RESOURCE_LEAK           0x00010000
 
 // 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
+// to debug with print statements and want ordered logging (e.g LOG(INFO) and LOG(ERROR) will not
-// order of calls).
+// appear in order of calls).
 #define FVK_DEBUG_FORCE_LOG_TO_I          0x00020000
 
 // Enable a minimal set of traces to assess the performance of the backend.
 // All other debug features must be disabled.
 #define FVK_DEBUG_PROFILING               0x00040000
 
+#define FVK_DEBUG_VULKAN_BUFFER_CACHE     0x00080000
+
 // Useful default combinations
 #define FVK_DEBUG_EVERYTHING              (0xFFFFFFFF & ~FVK_DEBUG_PROFILING)
 #define FVK_DEBUG_PERFORMANCE     \
@@ -172,15 +174,15 @@ static_assert(FVK_ENABLED(FVK_DEBUG_VALIDATION));
 #endif
 
 #if FVK_ENABLED(FVK_DEBUG_FORCE_LOG_TO_I)
-    #define FVK_LOGI (utils::slog.i)
+    #define FVK_LOGI LOG(INFO)
     #define FVK_LOGD FVK_LOGI
     #define FVK_LOGE FVK_LOGI
     #define FVK_LOGW FVK_LOGI
 #else
-    #define FVK_LOGE (utils::slog.e)
+    #define FVK_LOGE LOG(ERROR)
-    #define FVK_LOGW (utils::slog.w)
+    #define FVK_LOGW LOG(WARNING)
-    #define FVK_LOGD (utils::slog.d)
+    #define FVK_LOGD DLOG(INFO)
-    #define FVK_LOGI (utils::slog.i)
+    #define FVK_LOGI LOG(INFO)
 #endif
 
 // All vkCreate* functions take an optional allocator. For now we select the default allocator by

filament/backend/src/vulkan/VulkanContext.h

@@ -71,23 +71,23 @@ struct VulkanRenderPass {
 struct VulkanContext {
 public:
     static uint32_t selectMemoryType(VkPhysicalDeviceMemoryProperties const& memoryProperties,
-            uint32_t flags, VkFlags reqs) {
+            uint32_t types, VkFlags reqs) {
         for (uint32_t i = 0; i < VK_MAX_MEMORY_TYPES; i++) {
-            if (flags & 1) {
+            if (types & 1) {
                 if ((memoryProperties.memoryTypes[i].propertyFlags & reqs) == reqs) {
                     return i;
                 }
             }
-            flags >>= 1;
+            types >>= 1;
         }
         return (uint32_t) VK_MAX_MEMORY_TYPES;
     }
 
-    inline uint32_t selectMemoryType(uint32_t flags, VkFlags reqs) const {
+    inline uint32_t selectMemoryType(uint32_t types, VkFlags reqs) const {
         if ((reqs & VK_MEMORY_PROPERTY_PROTECTED_BIT) != 0) {
             assert_invariant(isProtectedMemorySupported());
         }
-        return selectMemoryType(mMemoryProperties, flags, reqs);
+        return selectMemoryType(mMemoryProperties, types, reqs);
     }
 
     inline fvkutils::VkFormatList const& getAttachmentDepthStencilFormats() const {

filament/backend/src/vulkan/VulkanDescriptorSetCache.cpp

@@ -320,7 +320,7 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
         uint8_t binding, fvkmemory::resource_ptr<VulkanBufferObject> bufferObject,
         VkDeviceSize offset, VkDeviceSize size) noexcept {
     VkDescriptorBufferInfo const info = {
-        .buffer = bufferObject->buffer.getGpuBuffer(),
+        .buffer = bufferObject->buffer.getVkBuffer(),
         .offset = offset,
         .range = size,
     };

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -19,7 +19,8 @@
 #include "CommandStreamDispatcher.h"
 #include "SystraceProfile.h"
 #include "VulkanAsyncHandles.h"
-#include "VulkanBuffer.h"
+#include "VulkanBufferCache.h"
+#include "VulkanBufferProxy.h"
 #include "VulkanCommands.h"
 #include "VulkanDriverFactory.h"
 #include "VulkanHandles.h"
@@ -84,7 +85,11 @@ VmaAllocator createAllocator(VkInstance instance, VkPhysicalDevice physicalDevic
         .vkGetImageMemoryRequirements2KHR = vkGetImageMemoryRequirements2KHR
 #endif
     };
-    VmaAllocatorCreateInfo const allocatorInfo {
+    VmaAllocatorCreateInfo const allocatorInfo{
+        // Disable the internal VMA synchronization because the backend is singled threaded.
+        // Improve CPU performance when using VMA functions. The backend will guarantee that all
+        // access to VMA is done in a thread safe way.
+        .flags = VMA_ALLOCATOR_CREATE_EXTERNALLY_SYNCHRONIZED_BIT,
         .physicalDevice = physicalDevice,
         .device = device,
         .pVulkanFunctions = &funcs,
@@ -99,15 +104,15 @@ VKAPI_ATTR VkBool32 VKAPI_CALL debugReportCallback(VkDebugReportFlagsEXT flags,
         VkDebugReportObjectTypeEXT objectType, uint64_t object, size_t location,
         int32_t messageCode, const char* pLayerPrefix, const char* pMessage, void* pUserData) {
     if (flags & VK_DEBUG_REPORT_ERROR_BIT_EXT) {
-        FVK_LOGE << "VULKAN ERROR: (" << pLayerPrefix << ") " << pMessage << utils::io::endl;
+        FVK_LOGE << "VULKAN ERROR: (" << pLayerPrefix << ") " << pMessage;
     } else {
         // TODO: emit best practices warnings about aggressive pipeline barriers.
         if (strstr(pMessage, "ALL_GRAPHICS_BIT") || strstr(pMessage, "ALL_COMMANDS_BIT")) {
             return VK_FALSE;
         }
-        FVK_LOGW << "VULKAN WARNING: (" << pLayerPrefix << ") " << pMessage << utils::io::endl;
+        FVK_LOGW << "VULKAN WARNING: (" << pLayerPrefix << ") " << pMessage;
     }
-    FVK_LOGE << utils::io::endl;
+    FVK_LOGE;
     return VK_FALSE;
 }
 #endif // FVK_ENABLED(FVK_DEBUG_VALIDATION)
@@ -117,18 +122,16 @@ VKAPI_ATTR VkBool32 VKAPI_CALL debugUtilsCallback(VkDebugUtilsMessageSeverityFla
         VkDebugUtilsMessageTypeFlagsEXT types, const VkDebugUtilsMessengerCallbackDataEXT* cbdata,
         void* pUserData) {
     if (severity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT) {
-        FVK_LOGE << "VULKAN ERROR: (" << cbdata->pMessageIdName << ") " << cbdata->pMessage
+        FVK_LOGE << "VULKAN ERROR: (" << cbdata->pMessageIdName << ") " << cbdata->pMessage;
-                 << utils::io::endl;
     } else {
         // TODO: emit best practices warnings about aggressive pipeline barriers.
         if (strstr(cbdata->pMessage, "ALL_GRAPHICS_BIT")
                 || strstr(cbdata->pMessage, "ALL_COMMANDS_BIT")) {
             return VK_FALSE;
         }
-        FVK_LOGW << "VULKAN WARNING: (" << cbdata->pMessageIdName << ") " << cbdata->pMessage
+        FVK_LOGW << "VULKAN WARNING: (" << cbdata->pMessageIdName << ") " << cbdata->pMessage;
-                 << utils::io::endl;
     }
-    FVK_LOGE << utils::io::endl;
+    FVK_LOGE << "";
     return VK_FALSE;
 }
 #endif // FVK_ENABLED(FVK_DEBUG_DEBUG_UTILS)
@@ -209,7 +212,8 @@ VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext const& contex
               mPlatform->getProtectedGraphicsQueueFamilyIndex(), &mContext),
       mPipelineLayoutCache(mPlatform->getDevice()),
       mPipelineCache(mPlatform->getDevice()),
-      mStagePool(mAllocator, &mCommands),
+      mStagePool(mAllocator, &mResourceManager, &mCommands, &mContext.getPhysicalDeviceLimits()),
+      mBufferCache(context, mResourceManager, mAllocator),
       mFramebufferCache(mPlatform->getDevice()),
       mYcbcrConversionCache(mPlatform->getDevice()),
       mSamplerCache(mPlatform->getDevice()),
@@ -270,20 +274,18 @@ Driver* VulkanDriver::create(VulkanPlatform* platform, VulkanContext const& cont
     //    VulkanRenderTarget            : 312       few
     // -- less than or equal to 312 bytes
 
-    FVK_LOGD
+    FVK_LOGD << "VulkanSwapChain: " << sizeof(VulkanSwapChain);
-           << "\nVulkanSwapChain: " << sizeof(VulkanSwapChain)
+    FVK_LOGD << "VulkanBufferObject: " << sizeof(VulkanBufferObject);
-           << "\nVulkanBufferObject: " << sizeof(VulkanBufferObject)
+    FVK_LOGD << "VulkanVertexBuffer: " << sizeof(VulkanVertexBuffer);
-           << "\nVulkanVertexBuffer: " << sizeof(VulkanVertexBuffer)
+    FVK_LOGD << "VulkanVertexBufferInfo: " << sizeof(VulkanVertexBufferInfo);
-           << "\nVulkanVertexBufferInfo: " << sizeof(VulkanVertexBufferInfo)
+    FVK_LOGD << "VulkanIndexBuffer: " << sizeof(VulkanIndexBuffer);
-           << "\nVulkanIndexBuffer: " << sizeof(VulkanIndexBuffer)
+    FVK_LOGD << "VulkanRenderPrimitive: " << sizeof(VulkanRenderPrimitive);
-           << "\nVulkanRenderPrimitive: " << sizeof(VulkanRenderPrimitive)
+    FVK_LOGD << "VulkanTexture: " << sizeof(VulkanTexture);
-           << "\nVulkanTexture: " << sizeof(VulkanTexture)
+    FVK_LOGD << "VulkanTimerQuery: " << sizeof(VulkanTimerQuery);
-           << "\nVulkanTimerQuery: " << sizeof(VulkanTimerQuery)
+    FVK_LOGD << "HwStream: " << sizeof(HwStream);
-           << "\nHwStream: " << sizeof(HwStream)
+    FVK_LOGD << "VulkanRenderTarget: " << sizeof(VulkanRenderTarget);
-           << "\nVulkanRenderTarget: " << sizeof(VulkanRenderTarget)
+    FVK_LOGD << "VulkanFence: " << sizeof(VulkanFence);
-           << "\nVulkanFence: " << sizeof(VulkanFence)
+    FVK_LOGD << "VulkanProgram: " << sizeof(VulkanProgram);
-           << "\nVulkanProgram: " << sizeof(VulkanProgram)
-           << utils::io::endl;
 #endif
 
     assert_invariant(platform);
@@ -328,7 +330,6 @@ void VulkanDriver::terminate() {
     // descriptorSetLayoutCache
     mExternalImageManager.terminate();
 
-    mStagePool.terminate();
     mPipelineCache.terminate();
     mFramebufferCache.terminate();
     mSamplerCache.terminate();
@@ -339,6 +340,15 @@ void VulkanDriver::terminate() {
     // Before terminating ResourceManager, we must make sure all of the resource_ptrs have been unset.
     mResourceManager.terminate();
 
+    // Must come after `mResourceManager`.
+    // Before terminating the memory pool, we must make sure all the VulkanBufferMemory are yielded
+    // back to the pool.
+    mBufferCache.terminate();
+
+    // Before terminating stagePool, we need all resources to have been
+    // reclaimed, as they perform cleanup within the stage pool.
+    mStagePool.terminate();
+
 #if FVK_ENABLED(FVK_DEBUG_RESOURCE_LEAK)
     mResourceManager.print();
 #endif
@@ -371,6 +381,7 @@ void VulkanDriver::collectGarbage() {
     mCommands.gc();
     mDescriptorSetCache.gc();
     mStagePool.gc();
+    mBufferCache.gc();
     mFramebufferCache.gc();
     mPipelineCache.gc();
 
@@ -514,7 +525,7 @@ void VulkanDriver::createIndexBufferR(Handle<HwIndexBuffer> ibh, ElementType ele
     FVK_SYSTRACE_SCOPE();
     auto elementSize = (uint8_t) getElementTypeSize(elementType);
     auto ib = resource_ptr<VulkanIndexBuffer>::make(&mResourceManager, ibh, mAllocator, mStagePool,
-            elementSize, indexCount);
+            mBufferCache, elementSize, indexCount);
     ib.inc();
 }
 
@@ -531,7 +542,7 @@ void VulkanDriver::createBufferObjectR(Handle<HwBufferObject> boh, uint32_t byte
         BufferObjectBinding bindingType, BufferUsage usage) {
     FVK_SYSTRACE_SCOPE();
     auto bo = resource_ptr<VulkanBufferObject>::make(&mResourceManager, boh, mAllocator, mStagePool,
-            byteCount, bindingType);
+            mBufferCache, byteCount, bindingType);
     bo.inc();
 }
 
@@ -784,14 +795,12 @@ void VulkanDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow,
     mResourceManager.gc();
 
     if ((flags & backend::SWAP_CHAIN_CONFIG_SRGB_COLORSPACE) != 0 && !isSRGBSwapChainSupported()) {
-        FVK_LOGW << "sRGB swapchain requested, but Platform does not support it"
+        FVK_LOGW << "sRGB swapchain requested, but Platform does not support it";
-                 << utils::io::endl;
         flags = flags | ~(backend::SWAP_CHAIN_CONFIG_SRGB_COLORSPACE);
     }
     if (flags & backend::SWAP_CHAIN_CONFIG_PROTECTED_CONTENT) {
         if (!isProtectedContentSupported()) {
-            FVK_LOGW << "protected swapchain requested, but Platform does not support it"
+            FVK_LOGW << "protected swapchain requested, but Platform does not support it";
-                << utils::io::endl;
         }
     }
     auto swapChain = resource_ptr<VulkanSwapChain>::make(&mResourceManager, sch, mPlatform,
@@ -802,8 +811,7 @@ void VulkanDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow,
 void VulkanDriver::createSwapChainHeadlessR(Handle<HwSwapChain> sch, uint32_t width,
         uint32_t height, uint64_t flags) {
     if ((flags & backend::SWAP_CHAIN_CONFIG_SRGB_COLORSPACE) != 0 && !isSRGBSwapChainSupported()) {
-        FVK_LOGW << "sRGB swapchain requested, but Platform does not support it"
+        FVK_LOGW << "sRGB swapchain requested, but Platform does not support it";
-                      << utils::io::endl;
         flags = flags | ~(backend::SWAP_CHAIN_CONFIG_SRGB_COLORSPACE);
     }
     assert_invariant(width > 0 && height > 0 && "Vulkan requires non-zero swap chain dimensions.");
@@ -1223,7 +1231,7 @@ void VulkanDriver::updateIndexBuffer(Handle<HwIndexBuffer> ibh, BufferDescriptor
     VulkanCommandBuffer& commands = mCommands.get();
     auto ib = resource_ptr<VulkanIndexBuffer>::cast(&mResourceManager, ibh);
     commands.acquire(ib);
-    ib->buffer.loadFromCpu(commands.buffer(), p.buffer, byteOffset, p.size);
+    ib->buffer.loadFromCpu(commands, p.buffer, byteOffset, p.size);
 
     scheduleDestroy(std::move(p));
 }
@@ -1238,7 +1246,7 @@ void VulkanDriver::updateBufferObject(Handle<HwBufferObject> boh, BufferDescript
 
     auto bo = resource_ptr<VulkanBufferObject>::cast(&mResourceManager, boh);
     commands.acquire(bo);
-    bo->buffer.loadFromCpu(commands.buffer(), bd.buffer, byteOffset, bd.size);
+    bo->buffer.loadFromCpu(commands, bd.buffer, byteOffset, bd.size);
 
     scheduleDestroy(std::move(bd));
 }
@@ -1249,7 +1257,7 @@ void VulkanDriver::updateBufferObjectUnsynchronized(Handle<HwBufferObject> boh,
     auto bo = resource_ptr<VulkanBufferObject>::cast(&mResourceManager, boh);
     commands.acquire(bo);
     // TODO: implement unsynchronized version
-    bo->buffer.loadFromCpu(commands.buffer(), bd.buffer, byteOffset, bd.size);
+    bo->buffer.loadFromCpu(commands, bd.buffer, byteOffset, bd.size);
     scheduleDestroy(std::move(bd));
 }
 
@@ -1291,7 +1299,7 @@ TimerQueryResult VulkanDriver::getTimerQueryValue(Handle<HwTimerQuery> tqh, uint
     uint64_t const end = results.endTime;
     if (begin >= end) {
         // TODO: queries might have ran on different command buffers.
-        FVK_LOGW << "Timestamps are not monotonically increasing. " << utils::io::endl;
+        FVK_LOGW << "Timestamps are not monotonically increasing. ";
         *elapsedTime = 0;
         return TimerQueryResult::ERROR;
     }
@@ -1622,8 +1630,8 @@ void VulkanDriver::readPixels(Handle<HwRenderTarget> src, uint32_t x, uint32_t y
     mReadPixels.run(
             srcTarget, x, y, width, height, mPlatform->getGraphicsQueueFamilyIndex(),
             std::move(pbd),
-            [&context = mContext](uint32_t reqs, VkFlags flags) {
+            [&context = mContext](uint32_t types, VkFlags reqs) {
-                return context.selectMemoryType(reqs, flags);
+                return context.selectMemoryType(types, reqs);
             },
             [this](PixelBufferDescriptor&& pbd) {
                 scheduleDestroy(std::move(pbd));
@@ -1898,7 +1906,7 @@ void VulkanDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
     // avoid rebinding these if they are already bound, but since we do not (yet) support subranges
     // it would be rare for a client to make consecutive draw calls with the same render primitive.
     vkCmdBindVertexBuffers(cmdbuffer, 0, bufferCount, buffers, offsets);
-    vkCmdBindIndexBuffer(cmdbuffer, prim->indexBuffer->buffer.getGpuBuffer(), 0,
+    vkCmdBindIndexBuffer(cmdbuffer, prim->indexBuffer->buffer.getVkBuffer(), 0,
             prim->indexBuffer->indexType);
 }
 
@@ -2029,7 +2037,7 @@ void VulkanDriver::debugCommandBegin(CommandStream* cmds, bool synchronous, cons
         assert_invariant(inRenderPass);
         inRenderPass = false;
     } else if (inRenderPass && OUTSIDE_COMMANDS.find(command) != OUTSIDE_COMMANDS.end()) {
-        FVK_LOGE << command.data() << " issued inside a render pass." << utils::io::endl;
+        FVK_LOGE << command.data() << " issued inside a render pass.";
     }
 #endif
 }

filament/backend/src/vulkan/VulkanDriver.h

@@ -18,15 +18,17 @@
 #define TNT_FILAMENT_BACKEND_VULKANDRIVER_H
 
 #include "VulkanBlitter.h"
+#include "VulkanBufferCache.h"
 #include "VulkanConstants.h"
 #include "VulkanContext.h"
 #include "VulkanFboCache.h"
 #include "VulkanHandles.h"
+#include "VulkanMemory.h"
 #include "VulkanPipelineCache.h"
+#include "VulkanQueryManager.h"
 #include "VulkanReadPixels.h"
 #include "VulkanSamplerCache.h"
 #include "VulkanStagePool.h"
-#include "VulkanQueryManager.h"
 #include "VulkanYcbcrConversionCache.h"
 #include "vulkan/VulkanDescriptorSetCache.h"
 #include "vulkan/VulkanDescriptorSetLayoutCache.h"
@@ -138,6 +140,7 @@ private:
     VulkanPipelineLayoutCache mPipelineLayoutCache;
     VulkanPipelineCache mPipelineCache;
     VulkanStagePool mStagePool;
+    VulkanBufferCache mBufferCache;
     VulkanFboCache mFramebufferCache;
     VulkanYcbcrConversionCache mYcbcrConversionCache;
     VulkanSamplerCache mSamplerCache;

filament/backend/src/vulkan/VulkanFboCache.cpp

@@ -100,8 +100,7 @@ VkFramebuffer VulkanFboCache::getFramebuffer(FboKey const& config) noexcept {
         << "for render pass " << config.renderPass << ", "
         << "samples = " << int(config.samples) << ", "
         << "depth = " << (config.depth ? 1 : 0) << ", "
-        << "attachmentCount = " << attachmentCount
+        << "attachmentCount = " << attachmentCount;
-        << utils::io::endl;
     #endif
 
     VkFramebufferCreateInfo info {
@@ -341,8 +340,7 @@ VkRenderPass VulkanFboCache::getRenderPass(RenderPassKey const& config) noexcept
              << "needsResolveMask = " << int(config.needsResolveMask) << ", "
              << "usesLazilyAllocatedMemory = " << int(config.usesLazilyAllocatedMemory) << ", "
              << "viewCount = " << int(config.viewCount) << ", "
-             << "colorAttachmentCount[0] = " << subpasses[0].colorAttachmentCount
+             << "colorAttachmentCount[0] = " << subpasses[0].colorAttachmentCount;
-             << utils::io::endl;
 #endif
 
     return renderPass;

filament/backend/src/vulkan/VulkanHandles.cpp

@@ -287,7 +287,7 @@ VulkanProgram::VulkanProgram(VkDevice device, Program const& builder) noexcept
 
 #if FVK_ENABLED(FVK_DEBUG_SHADER_MODULE)
     FVK_LOGD << "Created VulkanProgram " << builder << ", shaders = (" << modules[0]
-             << ", " << modules[1] << ")" << utils::io::endl;
+             << ", " << modules[1] << ")";
 #endif
 }
 
@@ -583,22 +583,22 @@ void VulkanVertexBuffer::setBuffer(fvkmemory::resource_ptr<VulkanBufferObject> b
     int8_t const* const attribToBuffer = vbi->getAttributeToBuffer();
     for (uint8_t attribIndex = 0; attribIndex < count; attribIndex++) {
         if (attribToBuffer[attribIndex] == static_cast<int8_t>(index)) {
-            vkbuffers[attribIndex] = bufferObject->buffer.getGpuBuffer();
+            vkbuffers[attribIndex] = bufferObject->buffer.getVkBuffer();
         }
     }
     mResources.push_back(bufferObject);
 }
 
 VulkanBufferObject::VulkanBufferObject(VmaAllocator allocator, VulkanStagePool& stagePool,
-        uint32_t byteCount, BufferObjectBinding bindingType)
+        VulkanBufferCache& bufferCache, uint32_t byteCount, BufferObjectBinding bindingType)
     : HwBufferObject(byteCount),
-      buffer(allocator, stagePool, getBufferObjectUsage(bindingType), byteCount),
+      buffer(allocator, stagePool, bufferCache, getBufferObjectUsage(bindingType), byteCount),
       bindingType(bindingType) {}
 
 VulkanRenderPrimitive::VulkanRenderPrimitive(PrimitiveType pt,
         fvkmemory::resource_ptr<VulkanVertexBuffer> vb,
         fvkmemory::resource_ptr<VulkanIndexBuffer> ib)
-    : HwRenderPrimitive{.type = pt},
+    : HwRenderPrimitive{ .type = pt },
       vertexBuffer(vb),
       indexBuffer(ib) {}
 

filament/backend/src/vulkan/VulkanHandles.h

@@ -21,20 +21,21 @@
 #include "DriverBase.h"
 
 #include "VulkanAsyncHandles.h"
-#include "VulkanBuffer.h"
+#include "VulkanBufferCache.h"
+#include "VulkanBufferProxy.h"
 #include "VulkanFboCache.h"
 #include "VulkanSwapChain.h"
 #include "VulkanTexture.h"
 #include "vulkan/memory/Resource.h"
-#include "vulkan/utils/StaticVector.h"
 #include "vulkan/utils/Definitions.h"
+#include "vulkan/utils/StaticVector.h"
 
 #include <backend/Program.h>
 
-#include <utils/bitset.h>
 #include <utils/FixedCapacityVector.h>
 #include <utils/Mutex.h>
 #include <utils/StructureOfArrays.h>
+#include <utils/bitset.h>
 
 #include <array>
 
@@ -428,21 +429,22 @@ private:
 };
 
 struct VulkanIndexBuffer : public HwIndexBuffer, fvkmemory::Resource {
-    VulkanIndexBuffer(VmaAllocator allocator, VulkanStagePool& stagePool, uint8_t elementSize,
+    VulkanIndexBuffer(VmaAllocator allocator, VulkanStagePool& stagePool,
-            uint32_t indexCount)
+            VulkanBufferCache& bufferCache, uint8_t elementSize, uint32_t indexCount)
         : HwIndexBuffer(elementSize, indexCount),
-          buffer(allocator, stagePool, VK_BUFFER_USAGE_INDEX_BUFFER_BIT, elementSize * indexCount),
+          buffer(allocator, stagePool, bufferCache, VulkanBufferUsage::INDEX,
+                  elementSize * indexCount),
           indexType(elementSize == 2 ? VK_INDEX_TYPE_UINT16 : VK_INDEX_TYPE_UINT32) {}
 
-    VulkanBuffer buffer;
+    VulkanBufferProxy buffer;
     const VkIndexType indexType;
 };
 
 struct VulkanBufferObject : public HwBufferObject, fvkmemory::Resource {
-    VulkanBufferObject(VmaAllocator allocator, VulkanStagePool& stagePool, uint32_t byteCount,
+    VulkanBufferObject(VmaAllocator allocator, VulkanStagePool& stagePool,
-            BufferObjectBinding bindingType);
+            VulkanBufferCache& bufferCache, uint32_t byteCount, BufferObjectBinding bindingType);
 
-    VulkanBuffer buffer;
+    VulkanBufferProxy buffer;
     const BufferObjectBinding bindingType;
 };
 
@@ -455,18 +457,19 @@ struct VulkanRenderPrimitive : public HwRenderPrimitive, fvkmemory::Resource {
     fvkmemory::resource_ptr<VulkanIndexBuffer> indexBuffer;
 };
 
-inline constexpr VkBufferUsageFlagBits getBufferObjectUsage(
+inline constexpr VulkanBufferUsage getBufferObjectUsage(BufferObjectBinding bindingType) noexcept {
-        BufferObjectBinding bindingType) noexcept {
+    switch (bindingType) {
-    switch(bindingType) {
         case BufferObjectBinding::VERTEX:
-            return VK_BUFFER_USAGE_VERTEX_BUFFER_BIT;
+            return VulkanBufferUsage::VERTEX;
         case BufferObjectBinding::UNIFORM:
-            return VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT;
+            return VulkanBufferUsage::UNIFORM;
         case BufferObjectBinding::SHADER_STORAGE:
-            return VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
+            return VulkanBufferUsage::SHADER_STORAGE;
-        // when adding more buffer-types here, make sure to update VulkanBuffer::loadFromCpu()
+            // when adding more buffer-types here, make sure to update VulkanBuffer::loadFromCpu()
-        // if necessary.
+            // if necessary.
     }
+
+    return VulkanBufferUsage::UNKNOWN;
 }
 
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanMemory.h

@@ -37,4 +37,24 @@ VK_DEFINE_HANDLE(VmaAllocator)
 VK_DEFINE_HANDLE(VmaAllocation)
 VK_DEFINE_HANDLE(VmaPool)
 
+namespace filament::backend {
+
+enum class VulkanBufferUsage : uint8_t {
+    UNKNOWN,
+    VERTEX,
+    INDEX,
+    UNIFORM,
+    SHADER_STORAGE,
+};
+
+struct VulkanGpuBuffer {
+    VkBuffer vkbuffer = VK_NULL_HANDLE;
+    VmaAllocation vmaAllocation = VK_NULL_HANDLE;
+    VmaAllocationInfo allocationInfo;
+    uint32_t numBytes = 0;
+    VulkanBufferUsage usage = VulkanBufferUsage::UNKNOWN;
+};
+
+} // namespace filament::backend
+
 #endif // TNT_FILAMENT_BACKEND_VULKANMEMORY_H

filament/backend/src/vulkan/VulkanPipelineCache.cpp

@@ -218,8 +218,7 @@ VulkanPipelineCache::PipelineCacheEntry* VulkanPipelineCache::createPipeline() n
 
     #if FVK_ENABLED(FVK_DEBUG_SHADER_MODULE)
         FVK_LOGD << "vkCreateGraphicsPipelines with shaders = ("
-                 << shaderStages[0].module << ", " << shaderStages[1].module << ")"
+                 << shaderStages[0].module << ", " << shaderStages[1].module << ")";
-                 << utils::io::endl;
     #endif
     PipelineCacheEntry cacheEntry = {
         .lastUsed = mCurrentTime,
@@ -228,7 +227,7 @@ VulkanPipelineCache::PipelineCacheEntry* VulkanPipelineCache::createPipeline() n
             VKALLOC, &cacheEntry.handle);
     assert_invariant(error == VK_SUCCESS);
     if (error != VK_SUCCESS) {
-        FVK_LOGE << "vkCreateGraphicsPipelines error " << error << utils::io::endl;
+        FVK_LOGE << "vkCreateGraphicsPipelines error " << error;
         return nullptr;
     }
     return &mPipelines.emplace(mPipelineRequirements, cacheEntry).first.value();
@@ -242,7 +241,7 @@ void VulkanPipelineCache::bindProgram(fvkmemory::resource_ptr<VulkanProgram> pro
 #if FVK_ENABLED(FVK_DEBUG_SHADER_MODULE)
     if (mPipelineRequirements.shaders[0] == VK_NULL_HANDLE ||
             mPipelineRequirements.shaders[1] == VK_NULL_HANDLE) {
-        FVK_LOGE << "Binding missing shader: " << program->name.c_str() << utils::io::endl;
+        FVK_LOGE << "Binding missing shader: " << program->name.c_str();
     }
 #endif
 }

filament/backend/src/vulkan/VulkanQueryManager.cpp

@@ -42,7 +42,7 @@ fvkmemory::resource_ptr<VulkanTimerQuery> VulkanQueryManager::getNextQuery(
         fvkmemory::ResourceManager* resourceManager) {
     auto unused = ~mUsed;
     if (unused.empty()) {
-        FVK_LOGE << "More than " << mUsed.size() << " timers are not supported." << utils::io::endl;
+        FVK_LOGE << "More than " << mUsed.size() << " timers are not supported.";
         return {};
     }
 

filament/backend/src/vulkan/VulkanReadPixels.cpp

@@ -143,7 +143,7 @@ void VulkanReadPixels::run(fvkmemory::resource_ptr<VulkanRenderTarget> srcTarget
         mTaskHandler = std::make_unique<TaskHandler>();
     }
 
-    VkCommandPool& cmdpool = mCommandPool;
+    VkCommandPool const cmdpool = mCommandPool;
 
     fvkmemory::resource_ptr<VulkanTexture> srcTexture = srcTarget->getColor0().texture;
     assert_invariant(srcTexture);
@@ -152,17 +152,17 @@ void VulkanReadPixels::run(fvkmemory::resource_ptr<VulkanRenderTarget> srcTarget
             = srcFormat == VK_FORMAT_B8G8R8A8_UNORM || srcFormat == VK_FORMAT_B8G8R8A8_SRGB;
 
     // Create a host visible, linearly tiled image as a staging area.
-    VkImageCreateInfo const imageInfo{
+    VkImageCreateInfo const imageInfo = {
-            .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
+        .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
-            .imageType = VK_IMAGE_TYPE_2D,
+        .imageType = VK_IMAGE_TYPE_2D,
-            .format = srcFormat,
+        .format = srcFormat,
-            .extent = {width, height, 1},
+        .extent = { width, height, 1 },
-            .mipLevels = 1,
+        .mipLevels = 1,
-            .arrayLayers = 1,
+        .arrayLayers = 1,
-            .samples = VK_SAMPLE_COUNT_1_BIT,
+        .samples = VK_SAMPLE_COUNT_1_BIT,
-            .tiling = VK_IMAGE_TILING_LINEAR,
+        .tiling = VK_IMAGE_TILING_LINEAR,
-            .usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT,
+        .usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT,
-            .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+        .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
     };
 
     VkImage stagingImage;
@@ -171,7 +171,7 @@ void VulkanReadPixels::run(fvkmemory::resource_ptr<VulkanRenderTarget> srcTarget
 #if FVK_ENABLED(FVK_DEBUG_READ_PIXELS)
     FVK_LOGD << "readPixels created image=" << stagingImage
              << " to copy from image=" << srcTexture->getVkImage()
-             << " src-layout=" << srcTexture->getLayout(0, 0) << utils::io::endl;
+             << " src-layout=" << srcTexture->getLayout(0, 0);
 #endif
 
     VkMemoryRequirements memReqs;
@@ -188,28 +188,27 @@ void VulkanReadPixels::run(fvkmemory::resource_ptr<VulkanRenderTarget> srcTarget
         memoryTypeIndex = selectMemoryFunc(memReqs.memoryTypeBits,
                 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
         FVK_LOGW
-                << "readPixels is slow because VK_MEMORY_PROPERTY_HOST_CACHED_BIT is not available"
+                << "readPixels is slow because VK_MEMORY_PROPERTY_HOST_CACHED_BIT is not available";
-                << utils::io::endl;
     }
 
     FILAMENT_CHECK_POSTCONDITION(memoryTypeIndex < VK_MAX_MEMORY_TYPES)
             << "VulkanReadPixels: unable to find a memory type that meets requirements.";
 
     VkMemoryAllocateInfo const allocInfo = {
-            .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
+        .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
-            .allocationSize = memReqs.size,
+        .allocationSize = memReqs.size,
-            .memoryTypeIndex = memoryTypeIndex,
+        .memoryTypeIndex = memoryTypeIndex,
     };
 
     vkAllocateMemory(device, &allocInfo, VKALLOC, &stagingMemory);
     vkBindImageMemory(device, stagingImage, stagingMemory, 0);
 
     VkCommandBuffer cmdbuffer;
-    VkCommandBufferAllocateInfo const allocateInfo{
+    VkCommandBufferAllocateInfo const allocateInfo = {
-            .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO,
+        .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO,
-            .commandPool = cmdpool,
+        .commandPool = cmdpool,
-            .level = VK_COMMAND_BUFFER_LEVEL_PRIMARY,
+        .level = VK_COMMAND_BUFFER_LEVEL_PRIMARY,
-            .commandBufferCount = 1,
+        .commandBufferCount = 1,
     };
     vkAllocateCommandBuffers(device, &allocateInfo, &cmdbuffer);
 
@@ -304,9 +303,8 @@ void VulkanReadPixels::run(fvkmemory::resource_ptr<VulkanRenderTarget> srcTarget
                                  cmdpool, cmdbuffer, pUserBuffer,
                                  fence = readCompleteFence]() mutable {
         VkResult status = vkWaitForFences(device, 1, &fence, VK_TRUE, UINT64_MAX);
-        // Fence hasn't been reached. Try waiting again.
         if (status != VK_SUCCESS) {
-            FVK_LOGE << "Failed to wait for readPixels fence" << utils::io::endl;
+            FVK_LOGE << "Failed to wait for readPixels fence";
             return;
         }
 
@@ -324,7 +322,7 @@ void VulkanReadPixels::run(fvkmemory::resource_ptr<VulkanRenderTarget> srcTarget
                     fvkutils::getComponentCount(srcFormat), srcPixels,
                     static_cast<int>(subResourceLayout.rowPitch), static_cast<int>(width),
                     static_cast<int>(height), swizzle)) {
-            FVK_LOGE << "Unsupported PixelDataFormat or PixelDataType" << utils::io::endl;
+            FVK_LOGE << "Unsupported PixelDataFormat or PixelDataType";
         }
 
         vkUnmapMemory(device, stagingMemory);

filament/backend/src/vulkan/VulkanStagePool.cpp

@@ -28,46 +28,111 @@ static constexpr uint32_t TIME_BEFORE_EVICTION = 3;
 
 namespace filament::backend {
 
-VulkanStagePool::VulkanStagePool(VmaAllocator allocator, VulkanCommands* commands)
+namespace {
-    : mAllocator(allocator),
-      mCommands(commands) {}
 
-VulkanStage const* VulkanStagePool::acquireStage(uint32_t numBytes) {
+// Note: these are temporary values, they will be configurable.
-    // First check if a stage exists whose capacity is greater than or equal to the requested size.
+static constexpr uint32_t MAX_EMPTY_STAGES_TO_RETAIN = 1;
-    auto iter = mFreeStages.lower_bound(numBytes);
+constexpr uint32_t STAGE_SIZE = 1048576;
-    if (iter != mFreeStages.end()) {
+
-        auto stage = iter->second;
+}// namespace
-        mFreeStages.erase(iter);
+
-        stage->lastAccessed = mCurrentFrame;
+fvkmemory::resource_ptr<VulkanStage::Segment> VulkanStage::acquireSegment(
-        mUsedStages.push_back(stage);
+        fvkmemory::ResourceManager* resManager, uint32_t numBytes) {
-        return stage;
+    auto segment = fvkmemory::resource_ptr<Segment>::construct(
-    }
+        resManager, this, numBytes, mCurrentOffset, [this](uint32_t offset) {
-    // We were not able to find a sufficiently large stage, so create a new one.
+            mSegments.erase(offset);
-    VulkanStage* stage = new VulkanStage({
-        .memory = VK_NULL_HANDLE,
-        .buffer = VK_NULL_HANDLE,
-        .capacity = numBytes,
-        .lastAccessed = mCurrentFrame,
     });
+    mSegments.insert({mCurrentOffset, segment.get()});
+    mCurrentOffset += numBytes;
+    return segment;
+}
 
-    // Create the VkBuffer.
+VulkanStagePool::VulkanStagePool(VmaAllocator allocator, fvkmemory::ResourceManager* resManager,
-    mUsedStages.push_back(stage);
+        VulkanCommands* commands, const VkPhysicalDeviceLimits* deviceLimits)
-    VkBufferCreateInfo bufferInfo {
+    : mAllocator(allocator),
+      mResManager(resManager),
+      mCommands(commands),
+      mDeviceLimits(deviceLimits) {}
+
+fvkmemory::resource_ptr<VulkanStage::Segment> VulkanStagePool::acquireStage(uint32_t numBytes) {
+    // Apply alignment to the byte count to ensure that, when we later flush
+    // data written by the host, we only flush the atoms that we modified, and
+    // no adjacent atoms.
+    numBytes = alignToNonCoherentAtomSize(numBytes);
+
+    // First check if a stage segment exists whose capacity is greater than or
+    // equal to the requested size.
+    auto iter = mStages.lower_bound(numBytes);
+
+    VulkanStage* pStage;
+    if (iter != mStages.end()) {
+        pStage = iter->second;
+        mStages.erase(iter);
+    } else {
+        pStage = allocateNewStage(std::max(numBytes, STAGE_SIZE));
+    }
+
+    // Note: this allocation updates `currentOffset` and `segments` within
+    // the parent stage. When destroyed, it will update `segments`.
+    fvkmemory::resource_ptr<VulkanStage::Segment> pSegment = pStage->acquireSegment(mResManager, numBytes);
+
+    // Update the stage's metadata, and reinsert it with the remaining segment
+    // capacity.
+    uint32_t spaceRemaining = pStage->capacity() - pStage->currentOffset();
+    mStages.insert({ spaceRemaining, pStage });
+
+    return pSegment;
+}
+
+uint32_t VulkanStagePool::alignToNonCoherentAtomSize(uint32_t bytes) {
+    VkDeviceSize alignment = mDeviceLimits->nonCoherentAtomSize;
+    if (alignment == 0) {
+        return bytes;
+    }
+
+    uint32_t remainder = bytes % alignment;
+    return remainder == 0 ? bytes : bytes + (alignment - remainder);
+}
+
+VulkanStage* VulkanStagePool::allocateNewStage(uint32_t capacity) {
+    VkBufferCreateInfo bufferInfo{
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
-        .size = numBytes,
+        .size = alignToNonCoherentAtomSize(capacity),
         .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT,
     };
     VmaAllocationCreateInfo allocInfo { .usage = VMA_MEMORY_USAGE_CPU_ONLY };
-    UTILS_UNUSED_IN_RELEASE VkResult result = vmaCreateBuffer(mAllocator, &bufferInfo,
+    VkBuffer buffer;
-            &allocInfo, &stage->buffer, &stage->memory, nullptr);
+    VmaAllocation memory;
+    VkResult result =
+            vmaCreateBuffer(mAllocator, &bufferInfo, &allocInfo, &buffer, &memory, nullptr);
 
 #if FVK_ENABLED(FVK_DEBUG_STAGING_ALLOCATION)
     if (result != VK_SUCCESS) {
-        FVK_LOGE << "Allocation error: " << result << utils::io::endl;
+        FVK_LOGE << "Allocation error: " << result;
+    } else {
+        FVK_LOGD << "Allocated stage with hndl " << buffer;
     }
 #endif
 
-    return stage;
+    void* pMapping = nullptr;
+    if (result == VK_SUCCESS) {
+        result = vmaMapMemory(mAllocator, memory, &pMapping);
+
+#if FVK_ENABLED(FVK_DEBUG_STAGING_ALLOCATION)
+        if (result != VK_SUCCESS) {
+            FVK_LOGE << "Memory mapping erryr: " << result << utils::io::endl;
+        }
+#endif
+    }
+
+    return new VulkanStage(memory, buffer, capacity, pMapping);
+}
+
+void VulkanStagePool::destroyStage(VulkanStage const*&& stage) {
+    assert(stage->isSafeToReset());  // Ensure all segments have been reset already.
+    vmaUnmapMemory(mAllocator, stage->memory());
+    vmaDestroyBuffer(mAllocator, stage->buffer(), stage->memory());
+    delete stage;
 }
 
 VulkanStageImage const* VulkanStagePool::acquireImage(PixelDataFormat format, PixelDataType type,
@@ -141,27 +206,34 @@ void VulkanStagePool::gc() noexcept {
     }
     const uint64_t evictionTime = mCurrentFrame - TIME_BEFORE_EVICTION;
 
-    // Destroy buffers that have not been used for several frames.
+    decltype(mStages) freeStages;
-    decltype(mFreeStages) freeStages;
+    freeStages.swap(mStages);
-    freeStages.swap(mFreeStages);
+    uint8_t freeStageCount = 0;  // Assuming we'll never have > 255 free stages
-    for (auto pair : freeStages) {
+    for (auto& pair : freeStages) {
-        if (pair.second->lastAccessed < evictionTime) {
+        // First, find any stages that have no segments within them.
-            vmaDestroyBuffer(mAllocator, pair.second->buffer, pair.second->memory);
+        if (pair.second->isSafeToReset()) {
-            delete pair.second;
+            if (++freeStageCount > MAX_EMPTY_STAGES_TO_RETAIN) {
-        } else {
+#if FVK_ENABLED(FVK_DEBUG_STAGING_ALLOCATION)
-            mFreeStages.insert(pair);
+                FVK_LOGD << "Destroying a staging buffer with hndl " << pair.second->buffer()
-        }
+                         << utils::io::endl;
-    }
+#endif
+                destroyStage(std::move(pair.second));
+                continue;
+            }
 
-    // Reclaim buffers that are no longer being used by any command buffer.
+#if FVK_ENABLED(FVK_DEBUG_STAGING_ALLOCATION)
-    decltype(mUsedStages) usedStages;
+            if (pair.first == 0) {
-    usedStages.swap(mUsedStages);
+                FVK_LOGD << "Recycling an unused staging buffer with hndl " << pair.second->buffer()
-    for (auto stage : usedStages) {
+                         << utils::io::endl;
-        if (stage->lastAccessed < evictionTime) {
+            }
-            stage->lastAccessed = mCurrentFrame;
+#endif
-            mFreeStages.insert(std::make_pair(stage->capacity, stage));
+
+            // Note - this segment is free, make sure the structure is cleared
+            // and reinsert it into our free stage list.
+            pair.second->reset();
+            mStages.insert({ pair.second->capacity(), pair.second });
         } else {
-            mUsedStages.push_back(stage);
+            mStages.insert(pair);
         }
     }
 
@@ -192,17 +264,10 @@ void VulkanStagePool::gc() noexcept {
 }
 
 void VulkanStagePool::terminate() noexcept {
-    for (auto stage : mUsedStages) {
+    for (auto& pair : mStages) {
-        vmaDestroyBuffer(mAllocator, stage->buffer, stage->memory);
+        destroyStage(std::move(pair.second));
-        delete stage;
     }
-    mUsedStages.clear();
+    mStages.clear();
-
-    for (auto pair : mFreeStages) {
-        vmaDestroyBuffer(mAllocator, pair.second->buffer, pair.second->memory);
-        delete pair.second;
-    }
-    mFreeStages.clear();
 
     for (auto image : mUsedImages) {
         vmaDestroyImage(mAllocator, image->image, image->memory);

filament/backend/src/vulkan/VulkanStagePool.h

@@ -17,8 +17,11 @@
 #ifndef TNT_FILAMENT_BACKEND_VULKANSTAGEPOOL_H
 #define TNT_FILAMENT_BACKEND_VULKANSTAGEPOOL_H
 
-#include "backend/DriverEnums.h"
 #include "VulkanMemory.h"
+#include "backend/DriverEnums.h"
+#include "vulkan/memory/Resource.h"
+#include "vulkan/memory/ResourceManager.h"
+#include "vulkan/memory/ResourcePointer.h"
 
 #include <map>
 #include <unordered_set>
@@ -28,12 +31,96 @@ namespace filament::backend {
 
 class VulkanCommands;
 
-// Immutable POD representing a shared CPU-GPU staging area.
+// Object representing a shared CPU-GPU staging area, which can be subdivided
-struct VulkanStage {
+// into smaller buffers as needed.
-    VmaAllocation memory;
+class VulkanStage {
-    VkBuffer buffer;
+public:
-    uint32_t capacity;
+    VulkanStage(VmaAllocation memory, VkBuffer buffer, uint32_t capacity, void* mapping)
-    mutable uint64_t lastAccessed;
+        : mMemory(memory),
+          mBuffer(buffer),
+          mCapacity(capacity),
+          mMapping(mapping) {}
+
+    ~VulkanStage() = default;
+    VulkanStage(const VulkanStage& other) = delete;
+    VulkanStage(VulkanStage&& other) = delete;
+    VulkanStage& operator=(const VulkanStage& other) = delete;
+    VulkanStage& operator=(VulkanStage&& other) = delete;
+
+    class Segment : public fvkmemory::Resource {
+    public:
+        using OnRecycle = std::function<void(uint32_t offset)>;
+
+        Segment(VulkanStage* parentStage, uint32_t capacity, uint32_t offset,
+                OnRecycle&& onRecycleFn)
+            : mParentStage(parentStage),
+              mCapacity(capacity),
+              mOffset(offset),
+              mOnRecycleFn(onRecycleFn) {}
+
+        ~Segment() {
+            if (mOnRecycleFn) {
+                mOnRecycleFn(offset());
+            }
+        }
+
+        // Should not be copying this around.
+        Segment(const Segment& other) = delete;
+        Segment(Segment&& other) = delete;
+        Segment& operator=(const Segment& other) = delete;
+        Segment& operator=(Segment&& other) = delete;
+
+        inline VulkanStage* parentStage() const { return mParentStage; }
+        inline VkBuffer buffer() const { return parentStage()->buffer(); }
+        inline VmaAllocation memory() const { return parentStage()->memory(); }
+        inline uint32_t capacity() const { return mCapacity; }
+        inline uint32_t offset() const { return mOffset; }
+
+        inline void* mapping() const {
+            return reinterpret_cast<void*>(
+                    reinterpret_cast<char*>(mParentStage->mapping()) + offset());
+        }
+
+    private:
+        // Ensure parent class can access the terminate method.
+        friend class VulkanStage;
+
+        VulkanStage* const mParentStage;
+        const uint32_t mCapacity;
+        const uint32_t mOffset;
+        OnRecycle mOnRecycleFn;
+    };
+
+    inline VmaAllocation memory() const { return mMemory; }
+    inline VkBuffer buffer() const { return mBuffer; }
+    inline uint32_t capacity() const { return mCapacity; }
+    inline void* mapping() const { return mMapping; }
+
+    inline uint32_t currentOffset() { return mCurrentOffset; }
+
+    inline bool isSafeToReset() const { return mSegments.empty(); }
+
+    inline void reset() { mCurrentOffset = 0; }
+
+    // Marks a region of the block as "in-use", and provides information about
+    // the allocated region to the caller. Note: this assumes that numBytes
+    // is aligned to the physical device's nonCoherentAtomSize.
+    fvkmemory::resource_ptr<Segment> acquireSegment(fvkmemory::ResourceManager* resManager,
+            uint32_t numBytes);
+
+private:
+    const VmaAllocation mMemory;
+    const VkBuffer mBuffer;
+    const uint32_t mCapacity;
+
+    void* mMapping;
+
+    uint32_t mCurrentOffset = 0;
+
+    // Maps the start offset of a vulkan stage block to the stage block,
+    // for easy deletions later. This is managed by the blocks themselves, in an
+    // RAII pattern, during construction and destruction.
+    std::unordered_map<uint32_t, Segment*> mSegments;
 };
 
 struct VulkanStageImage {
@@ -49,11 +136,15 @@ struct VulkanStageImage {
 // This class manages two types of host-mappable staging areas: buffer stages and image stages.
 class VulkanStagePool {
 public:
-    VulkanStagePool(VmaAllocator allocator, VulkanCommands* commands);
+    VulkanStagePool(VmaAllocator allocator, fvkmemory::ResourceManager* resManager,
+            VulkanCommands* commands, const VkPhysicalDeviceLimits* deviceLimits);
 
-    // Finds or creates a stage whose capacity is at least the given number of bytes.
+    // Finds or creates a stage block whose capacity is at least the given
-    // The stage is automatically released back to the pool after TIME_BEFORE_EVICTION frames.
+    // number of bytes. Internally, creates and manages and subdivides large
-    VulkanStage const* acquireStage(uint32_t numBytes);
+    // buffers so that we have less objects around that we have to keep track
+    // of.
+    // This function is NOT thread-safe.
+    fvkmemory::resource_ptr<VulkanStage::Segment> acquireStage(uint32_t numBytes);
 
     // Images have VK_IMAGE_LAYOUT_GENERAL and must not be transitioned to any other layout
     VulkanStageImage const* acquireImage(PixelDataFormat format, PixelDataType type,
@@ -64,17 +155,37 @@ public:
 
     // Destroys all unused stages and asserts that there are no stages currently in use.
     // This should be called while the context's VkDevice is still alive.
+    // Note: it is expected that all resources have been reclaimed before this
+    // is called. It is also expected that this stage pool does not hold any
+    // resource_ptrs, as this would lead to undefined behavior.
     void terminate() noexcept;
 
 private:
     VmaAllocator mAllocator;
+    fvkmemory::ResourceManager* mResManager;
     VulkanCommands* mCommands;
+    const VkPhysicalDeviceLimits* mDeviceLimits;
+
+    // Takes a number of bytes, and aligns it to the non-coherent atom size.
+    // This allows us to ensure that when we flush buffers from the host, we
+    // never flush more atoms than we need to.
+    uint32_t alignToNonCoherentAtomSize(uint32_t numBytes);
+
+    // Allocates a new stage buffer, and optionally subdivides it into stage
+    // blocks. If subdivideBlocks is true, predefined divisions will be used.
+    // Otherwise, it's expected that capacity is defined to a value, and that
+    // is the size that will be used for the buffer (as well as the only block
+    // being created).
+    VulkanStage* allocateNewStage(uint32_t capacity);
+
+    // Performs any bookkeeping required to delete a VulkanStage object; namely,
+    // unmapping memory, freeing the allocation, and deleting the VulkanStage
+    // object. Note: takes an r-value because after this call, `stage` won't
+    // exist.
+    void destroyStage(VulkanStage const*&& stage);
 
     // Use an ordered multimap for quick (capacity => stage) lookups using lower_bound().
-    std::multimap<uint32_t, VulkanStage const*> mFreeStages;
+    std::multimap<uint32_t, VulkanStage*> mStages;
-
-    // Simple unordered set for stashing a list of in-use stages that can be reclaimed later.
-    std::vector<VulkanStage const*> mUsedStages;
 
     std::unordered_set<VulkanStageImage const*> mFreeImages;
     std::vector<VulkanStageImage const*> mUsedImages;

filament/backend/src/vulkan/VulkanTexture.cpp

@@ -25,6 +25,7 @@
 #include <backend/DriverEnums.h>
 #include <private/backend/BackendUtils.h>
 
+#include <utils/compiler.h>
 #include <utils/Panic.h>
 
 using namespace bluevk;
@@ -187,8 +188,8 @@ VkImageUsageFlags getUsage(VulkanContext const& context, uint8_t samples,
             VkFormatProperties props;
             vkGetPhysicalDeviceFormatProperties(physicalDevice, vkFormat, &props);
             if (!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT)) {
-                FVK_LOGW << "Texture usage is SAMPLEABLE but format " << vkFormat << " is not "
+                FVK_LOGW << "Texture usage is SAMPLEABLE but format " << vkFormat
-                        "sampleable with optimal tiling." << utils::io::endl;
+                         << " is not sampleable with optimal tiling.";
             }
         }
 #endif
@@ -218,6 +219,40 @@ VkImageUsageFlags getUsage(VulkanContext const& context, uint8_t samples,
     return usage;
 }
 
+void adjustedMemcpy(void* mapped, PixelBufferDescriptor const& p, size_t width, size_t height,
+        size_t depth) {
+    uint8_t* buf = (uint8_t*) p.buffer;
+    size_t const pixelSize = PixelBufferDescriptor::computeDataSize(p.format, p.type, 1, 1, 1);
+    size_t const pbdStride = p.stride ? p.stride : width;
+
+    // Slow path of copying row by row
+    assert_invariant(pbdStride >= width);
+    if (UTILS_UNLIKELY(p.left > 0 || p.top > 0 || pbdStride > width)) {
+        size_t const pbdRowSize =
+                PixelBufferDescriptor::computeDataSize(p.format, p.type, pbdStride, 1, p.alignment);
+        size_t const pbdHeight = p.size / pixelSize / pbdStride / depth;
+        size_t const pbdLayerSize = pbdRowSize * pbdHeight;
+
+        size_t const rowSize = width * pixelSize;
+        size_t const layerSize = width * height * pixelSize;
+
+        // Size of a row to write
+        size_t const writeSize = std::min(pbdStride - p.left, width) * pixelSize;
+
+        for (size_t z = 0; z < depth; z++) {
+            for (size_t y = p.top; y < pbdHeight; y++) {
+                uint8_t* buf = (uint8_t*) p.buffer +
+                               ((p.left * pixelSize) + (y * pbdRowSize) + (z * pbdLayerSize));
+                uint8_t* curMapped = (uint8_t*) mapped + ((y - p.top) * rowSize + z * layerSize);
+                memcpy(curMapped, buf, writeSize);
+            }
+        }
+    } else {
+        size_t const writeSize = pixelSize * (width * height * depth);
+        memcpy(mapped, buf, writeSize);
+    }
+}
+
 } // anonymous namespace
 
 VulkanTextureState::VulkanTextureState(VulkanStagePool& stagePool, VulkanCommands* commands,
@@ -326,7 +361,7 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
             imageInfo.flags = VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT;
         } else {
             FVK_LOGW << "Note: creating 2D views on 3D image is not available on this platform. "
-                     << "i.e. we cannot render to slices of a 3D image" << utils::io::endl;
+                     << "i.e. we cannot render to slices of a 3D image";
         }
     } else if (target == SamplerType::SAMPLER_CUBEMAP) {
         imageInfo.arrayLayers = 6;
@@ -385,7 +420,7 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
             << "type = " << imageInfo.imageType << ", "
             << "flags = " << imageInfo.flags << ", "
             << "target = " << static_cast<int>(target) <<", "
-            << "format = " << vkFormat << utils::io::endl;
+            << "format = " << vkFormat;
     }
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "Unable to create image."
                                                        << " error=" << static_cast<int32_t>(result);
@@ -480,30 +515,40 @@ void VulkanTexture::updateImage(const PixelBufferDescriptor& data, uint32_t widt
     assert_invariant(hostData->size > 0 && "Data is empty");
 
     // Otherwise, use vkCmdCopyBufferToImage.
-    void* mapped = nullptr;
+    size_t const bpp =
-    VulkanStage const* stage = mState->mStagePool.acquireStage(hostData->size);
+            PixelBufferDescriptor::computeDataSize(hostData->format, hostData->type, 1, 1, 1);
-    assert_invariant(stage->memory);
+    size_t const writeSize = width * height * depth * bpp;
-    vmaMapMemory(mState->mAllocator, stage->memory, &mapped);
+
-    memcpy(mapped, hostData->buffer, hostData->size);
+    // Note: the following stageSegment must be stored within the command buffer
-    vmaUnmapMemory(mState->mAllocator, stage->memory);
+    // before going out of scope, to ensure proper bookkeeping within the
-    vmaFlushAllocation(mState->mAllocator, stage->memory, 0, hostData->size);
+    // staging buffer pool.
+    fvkmemory::resource_ptr<VulkanStage::Segment> stageSegment =
+            mState->mStagePool.acquireStage(writeSize);
+    assert_invariant(stageSegment->memory());
+    adjustedMemcpy(stageSegment->mapping(), *hostData, width, height, depth);
+    vmaFlushAllocation(mState->mAllocator, stageSegment->memory(), stageSegment->offset(),
+            writeSize);
 
     VulkanCommandBuffer& commands = mState->mCommands->get();
     VkCommandBuffer const cmdbuf = commands.buffer();
+    commands.acquire(stageSegment);
     commands.acquire(fvkmemory::resource_ptr<VulkanTexture>::cast(this));
 
+    bool const isDepth = getImageAspect() & VK_IMAGE_ASPECT_DEPTH_BIT;
+
     VkBufferImageCopy copyRegion = {
-        .bufferOffset = {},
+        .bufferOffset = stageSegment->offset(),
         .bufferRowLength = {},
         .bufferImageHeight = {},
         .imageSubresource = {
-            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            .aspectMask = VkImageAspectFlags(
+                    isDepth ? VK_IMAGE_ASPECT_DEPTH_BIT : VK_IMAGE_ASPECT_COLOR_BIT),
             .mipLevel = miplevel,
             .baseArrayLayer = 0,
-            .layerCount = 1
+            .layerCount = 1,
         },
         .imageOffset = { int32_t(xoffset), int32_t(yoffset), int32_t(zoffset) },
-        .imageExtent = { width, height, depth }
+        .imageExtent = { width, height, depth },
     };
 
     VkImageSubresourceRange transitionRange = {
@@ -511,7 +556,7 @@ void VulkanTexture::updateImage(const PixelBufferDescriptor& data, uint32_t widt
         .baseMipLevel = miplevel,
         .levelCount = 1,
         .baseArrayLayer = 0,
-        .layerCount = 1
+        .layerCount = 1,
     };
 
     // Vulkan specifies subregions for 3D textures differently than from 2D arrays.
@@ -536,20 +581,25 @@ void VulkanTexture::updateImage(const PixelBufferDescriptor& data, uint32_t widt
 
     transitionLayout(&commands, transitionRange, newLayout);
 
-    vkCmdCopyBufferToImage(cmdbuf, stage->buffer, mState->mTextureImage, newVkLayout, 1, &copyRegion);
+    vkCmdCopyBufferToImage(cmdbuf, stageSegment->buffer(), mState->mTextureImage, newVkLayout, 1,
+            &copyRegion);
 
     transitionLayout(&commands, transitionRange, nextLayout);
 }
 
-void VulkanTexture::updateImageWithBlit(const PixelBufferDescriptor& hostData, uint32_t width,
+void VulkanTexture::updateImageWithBlit(const PixelBufferDescriptor& data, uint32_t width,
         uint32_t height, uint32_t depth, uint32_t miplevel) {
+    // Otherwise, use vkCmdCopyBufferToImage.
+    size_t const bpp = PixelBufferDescriptor::computeDataSize(data.format, data.type, 1, 1, 1);
+    size_t const writeSize = width * height * depth * bpp;
+
     void* mapped = nullptr;
     VulkanStageImage const* stage
-            = mState->mStagePool.acquireImage(hostData.format, hostData.type, width, height);
+            = mState->mStagePool.acquireImage(data.format, data.type, width, height);
     vmaMapMemory(mState->mAllocator, stage->memory, &mapped);
-    memcpy(mapped, hostData.buffer, hostData.size);
+    adjustedMemcpy(mapped, data, width, height, depth);
     vmaUnmapMemory(mState->mAllocator, stage->memory);
-    vmaFlushAllocation(mState->mAllocator, stage->memory, 0, hostData.size);
+    vmaFlushAllocation(mState->mAllocator, stage->memory, 0, writeSize);
 
     VulkanCommandBuffer& commands = mState->mCommands->get();
     VkCommandBuffer const cmdbuf = commands.buffer();
@@ -681,14 +731,14 @@ bool VulkanTexture::transitionLayout(VkCommandBuffer cmdbuf, VkImageSubresourceR
                  << range.levelCount << ")" << " from=" << oldLayout << " to=" << newLayout
                  << " format=" << mState->mVkFormat << " depth="
                  << fvkutils::isVkDepthFormat(mState->mVkFormat)
-                 << " slice-by-slice=" << transitionSliceBySlice << utils::io::endl;
+                 << " slice-by-slice=" << transitionSliceBySlice;
 #endif
     } else {
 #if FVK_ENABLED(FVK_DEBUG_LAYOUT_TRANSITION)
         FVK_LOGD << "transition texture=" << mState->mTextureImage << " (" << range.baseArrayLayer
                  << "," << range.baseMipLevel << ")" << " count=(" << range.layerCount << ","
                  << range.levelCount << ")" << " to=" << newLayout
-                 << " is skipped because of no change in layout" << utils::io::endl;
+                 << " is skipped because of no change in layout";
 #endif
     }
 
@@ -799,8 +849,7 @@ void VulkanTexture::print() const {
                 level < (mPrimaryViewRange.baseMipLevel + mPrimaryViewRange.levelCount);
             FVK_LOGD << "[" << mState->mTextureImage << "]: (" << layer << "," << level
                           << ")=" << getLayout(layer, level)
-                          << " primary=" << primary
+                          << " primary=" << primary;
-                          << utils::io::endl;
         }
     }
 
@@ -809,8 +858,7 @@ void VulkanTexture::print() const {
         FVK_LOGD << "[" << mState->mTextureImage << ", imageView=" << view.second << "]=>"
                       << " (" << range.baseArrayLayer << "," << range.baseMipLevel << ")"
                       << " count=(" << range.layerCount << "," << range.levelCount << ")"
-                      << " aspect=" << range.aspectMask << " viewType=" << view.first.type
+                      << " aspect=" << range.aspectMask << " viewType=" << view.first.type;
-                      << utils::io::endl;
     }
 }
 #endif

filament/backend/src/vulkan/VulkanYcbcrConversionCache.cpp

@@ -29,6 +29,8 @@ using namespace bluevk;
 
 namespace filament::backend {
 
+using namespace fvkutils;
+
 VulkanYcbcrConversionCache::VulkanYcbcrConversionCache(VkDevice device)
     : mDevice(device) {}
 

filament/backend/src/vulkan/VulkanYcbcrConversionCache.h

@@ -17,6 +17,8 @@
 #ifndef TNT_FILAMENT_BACKEND_VULKANYCBCRCONVERSIONCACHE_H
 #define TNT_FILAMENT_BACKEND_VULKANYCBCRCONVERSIONCACHE_H
 
+#include "utils/Definitions.h"
+
 #include <backend/DriverEnums.h>
 
 #include <utils/Hash.h>
@@ -30,7 +32,7 @@ namespace filament::backend {
 class VulkanYcbcrConversionCache {
 public:
     struct Params {
-        SamplerYcbcrConversion conversion = {}; // 4
+        fvkutils::SamplerYcbcrConversion conversion = {}; // 4
         VkFormat format;                        // 4
         uint64_t externalFormat = 0;            // 8
     };
@@ -45,16 +47,15 @@ private:
 
     struct ConversionEqualTo {
         bool operator()(Params lhs, Params rhs) const noexcept {
-            SamplerYcbcrConversion::EqualTo equal;
+            fvkutils::SamplerYcbcrConversion::EqualTo equal;
             return equal(lhs.conversion, rhs.conversion) &&
-                   lhs.externalFormat == rhs.externalFormat &&
+                   lhs.externalFormat == rhs.externalFormat && lhs.format == rhs.format;
-                   lhs.format == rhs.format;
         }
     };
     using ConversionHashFn = utils::hash::MurmurHashFn<Params>;
     tsl::robin_map<Params, VkSamplerYcbcrConversion, ConversionHashFn, ConversionEqualTo> mCache;
 };
 
-}// namespace filament::backend
+} // namespace filament::backend
 
 #endif// TNT_FILAMENT_BACKEND_VULKANYCBCRCONVERSIONCACHE_H

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

@@ -26,6 +26,7 @@ template ResourceType getTypeEnum<VulkanIndexBuffer>() noexcept;
 template ResourceType getTypeEnum<VulkanProgram>() noexcept;
 template ResourceType getTypeEnum<VulkanRenderTarget>() noexcept;
 template ResourceType getTypeEnum<VulkanSwapChain>() noexcept;
+template ResourceType getTypeEnum<VulkanStage::Segment>() noexcept;
 template ResourceType getTypeEnum<VulkanRenderPrimitive>() noexcept;
 template ResourceType getTypeEnum<VulkanTexture>() noexcept;
 template ResourceType getTypeEnum<VulkanTextureState>() noexcept;
@@ -35,6 +36,7 @@ template ResourceType getTypeEnum<VulkanVertexBufferInfo>() noexcept;
 template ResourceType getTypeEnum<VulkanDescriptorSetLayout>() noexcept;
 template ResourceType getTypeEnum<VulkanDescriptorSet>() noexcept;
 template ResourceType getTypeEnum<VulkanFence>() noexcept;
+template ResourceType getTypeEnum<VulkanBuffer>() noexcept;
 
 template<typename D>
 ResourceType getTypeEnum() noexcept {
@@ -53,6 +55,9 @@ ResourceType getTypeEnum() noexcept {
     if constexpr (std::is_same_v<D, VulkanSwapChain>) {
         return ResourceType::SWAP_CHAIN;
     }
+    if constexpr (std::is_same_v<D, VulkanStage::Segment>) {
+        return ResourceType::STAGE_SEGMENT;
+    }
     if constexpr (std::is_same_v<D, VulkanRenderPrimitive>) {
         return ResourceType::RENDER_PRIMITIVE;
     }
@@ -80,6 +85,9 @@ ResourceType getTypeEnum() noexcept {
     if constexpr (std::is_same_v<D, VulkanFence>) {
         return ResourceType::FENCE;
     }
+    if constexpr (std::is_same_v<D, VulkanBuffer>) {
+        return ResourceType::VULKAN_BUFFER;
+    }
     return ResourceType::UNDEFINED_TYPE;
 }
 
@@ -95,6 +103,8 @@ std::string getTypeStr(ResourceType type) {
             return "RenderTarget";
         case ResourceType::SWAP_CHAIN:
             return "SwapChain";
+        case ResourceType::STAGE_SEGMENT:
+            return "Stage::Segment";
         case ResourceType::RENDER_PRIMITIVE:
             return "RenderPrimitive";
         case ResourceType::TEXTURE:
@@ -113,6 +123,8 @@ std::string getTypeStr(ResourceType type) {
             return "DescriptorSet";
         case ResourceType::FENCE:
             return "Fence";
+        case ResourceType::VULKAN_BUFFER:
+            return "VulkanBuffer";
         case ResourceType::UNDEFINED_TYPE:
             return "";
     }

filament/backend/src/vulkan/memory/Resource.h

@@ -49,7 +49,9 @@ enum class ResourceType : uint8_t {
     DESCRIPTOR_SET_LAYOUT = 11,
     DESCRIPTOR_SET = 12,
     FENCE = 13,
-    UNDEFINED_TYPE = 14,    // Must be the last enum because we use it for iterating over the enums.
+    VULKAN_BUFFER = 14,
+    STAGE_SEGMENT = 15,
+    UNDEFINED_TYPE = 16,    // Must be the last enum because we use it for iterating over the enums.
 };
 
 template<typename D>