wip

This commit adds `getConstant()` and `setConstant()` methods to
`MaterialInstance`.

When a material instance is created, it doesn't create or manage any spec
constants on its own, but instead relies on `Material`. The client can change
constants at any time, after which the necessary resources are copied and
maintained by `MaterialInstance`.

As part of this change, most internal rendering methods which depended on
`Material` have been refactored to instead depend on `MaterialInstance`.

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

@@ -27,17 +27,24 @@ runs:
         echo "$HASH" > /tmp/commit_hash.txt
     - name: Find commit message (PR)
       shell: bash
-      id: checkout_code
       if: github.event_name == 'pull_request'
       run: |
-        BEFORE_HASH=$(git rev-parse HEAD)
-        echo "hash=$BEFORE_HASH" >> "$GITHUB_OUTPUT"
-        # Next we will checkout the actual head (not the merge commits) of the PR
-        AFTER_HASH="${{ github.event.pull_request.head.sha }}"
-        git checkout $AFTER_HASH
-        COMMIT_MESSAGE=$(git log -1 --no-merges)
-        echo "$COMMIT_MESSAGE" > /tmp/commit_msg.txt
-        echo "$AFTER_HASH" > /tmp/commit_hash.txt
+        PR_NUMBER="${{ github.event.pull_request.number }}"
+        # Fetch the head of the PR explicitly to handle forks. Depth 50 ensures we can traverse past recent merge commits.
+        git fetch --depth=50 origin "pull/$PR_NUMBER/head:pr-head"
+
+        COMMIT_HASH=$(git log -1 --no-merges pr-head --format=%H)
+        AUTHOR_NAME=$(git log -1 --no-merges pr-head --format=%an)
+        AUTHOR_EMAIL=$(git log -1 --no-merges pr-head --format=%ae)
+        TSTAMP=$(git log -1 --no-merges pr-head --format=%aI)
+        COMMIT_MESSAGE=$(git log -1 --no-merges pr-head --format=%B)
+
+        echo "commit $COMMIT_HASH" > /tmp/commit_msg.txt
+        echo "Author: ${AUTHOR_NAME}<${AUTHOR_EMAIL}>" >> /tmp/commit_msg.txt
+        echo "Date: ${TSTAMP}" >> /tmp/commit_msg.txt
+        echo "" >> /tmp/commit_msg.txt
+        echo "$COMMIT_MESSAGE" >> /tmp/commit_msg.txt
+        echo "$COMMIT_HASH" > /tmp/commit_hash.txt
     - shell: bash
       id: action_output
       run: |
@@ -47,9 +54,4 @@ runs:
         cat /tmp/commit_msg.txt >> "$GITHUB_OUTPUT"
         echo "$DELIMITER" >> "$GITHUB_OUTPUT"
         # Get the commit hash
-        echo "hash=$(cat /tmp/commit_hash.txt)" >> "$GITHUB_OUTPUT"
-    - name: Cleanup Find commit message (PR)
-      shell: bash
-      if: github.event_name == 'pull_request'
-      run: |
-        git checkout ${{ steps.checkout_code.outputs.hash }}
+        echo "hash=$(cat /tmp/commit_hash.txt)" >> "$GITHUB_OUTPUT"

.github/actions/renderdiff-generate/action.yml

@@ -0,0 +1,26 @@
+name: 'Renderdiff Generate'
+description: 'Sets up prerequisites and runs the generate script for renderdiff'
+runs:
+  using: "composite"
+  steps:
+    - uses: ./.github/actions/mac-prereq
+    - uses: ./.github/actions/get-gltf-assets
+    - uses: ./.github/actions/get-mesa
+    - uses: ./.github/actions/get-vulkan-sdk
+    - name: Prerequisites
+      run: |
+        # Must have at least clang-16 for a webgpu/dawn build.
+        sudo xcode-select -s /Applications/Xcode_16.2.app/Contents/Developer
+      shell: bash
+    - name: Generate images
+      run: |
+        TEST_DIR=test/renderdiff
+        source ${TEST_DIR}/src/preamble.sh
+        set -eux
+        bash ${TEST_DIR}/generate.sh
+        set +eux
+      shell: bash
+    - name: Build diffimg tool
+      run: |
+        ./build.sh release diffimg
+      shell: bash

.github/workflows/postsubmit-main.yml

@@ -10,16 +10,27 @@ jobs:
   # a branch on filament-assets.
   update-renderdiff-goldens:
     name: update-renderdiff-goldens
-    runs-on: 'ubuntu-24.04-4core'
+    runs-on: 'macos-14-xlarge'
     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: Build diffimg
-        run: ./build.sh release diffimg
+      - name: Check if accepting new goldens
+        id: check_accept
+        env:
+          COMMIT_MESSAGE: ${{ steps.get_commit_msg.outputs.msg }}
+        run: |
+          if echo "${COMMIT_MESSAGE}" | python3 test/renderdiff/src/commit_msg.py --mode=accept_new_goldens; then
+            echo "accept=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "accept=false" >> "$GITHUB_OUTPUT"
+          fi
+        shell: bash
+      - name: Renderdiff Generate for new goldens
+        if: steps.check_accept.outputs.accept == 'true'
+        uses: ./.github/actions/renderdiff-generate
       - name: Run update script
         env:
           GH_TOKEN: ${{ secrets.FILAMENTBOT_TOKEN }}
@@ -27,10 +38,24 @@ jobs:
         run: |
           GOLDEN_BRANCH=$(echo "${COMMIT_MESSAGE}" | python3 test/renderdiff/src/commit_msg.py)
           COMMIT_HASH="${{ steps.get_commit_msg.outputs.hash }}"
+
+          git config --global user.email "filament.bot@gmail.com"
+          git config --global user.name "Filament Bot"
+          git config --global credential.helper cache
+
+          if [[ "${{ steps.check_accept.outputs.accept }}" == "true" ]]; then
+              SHORT_HASH="${COMMIT_HASH:0:8}"
+              GOLDEN_BRANCH="accept-goldens-${SHORT_HASH}"
+              echo "Generating new goldens for branch ${GOLDEN_BRANCH}"
+              python3 test/renderdiff/src/update_golden.py \
+                  --branch=${GOLDEN_BRANCH} \
+                  --source=$(pwd)/out/renderdiff/renders \
+                  --commit-msg="Auto-update goldens from ${COMMIT_HASH}" \
+                  --push-to-remote \
+                  --golden-repo-token=${GH_TOKEN}
+          fi
+
           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} \

.github/workflows/presubmit.yml

@@ -126,18 +126,24 @@ jobs:
       - uses: actions/checkout@v4.1.6
         with:
           fetch-depth: 0
-      - uses: ./.github/actions/mac-prereq
-      - uses: ./.github/actions/get-gltf-assets
-      - uses: ./.github/actions/get-mesa
-      - uses: ./.github/actions/get-vulkan-sdk
       - id: get_commit_msg
         uses: ./.github/actions/get-commit-msg
-      - name: Prerequisites
+      - name: Check if accepting new goldens
+        id: check_accept
+        env:
+          COMMIT_MESSAGE: ${{ steps.get_commit_msg.outputs.msg }}
         run: |
-          # Must have at least clang-16 for a webgpu/dawn build.
-          sudo xcode-select -s /Applications/Xcode_16.2.app/Contents/Developer
+          if echo "${COMMIT_MESSAGE}" | python3 test/renderdiff/src/commit_msg.py --mode=accept_new_goldens; then
+            echo "accept=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "accept=false" >> "$GITHUB_OUTPUT"
+          fi
         shell: bash
+      - name: Renderdiff Generate
+        if: steps.check_accept.outputs.accept != 'true'
+        uses: ./.github/actions/renderdiff-generate
       - name: Render and compare
+        if: steps.check_accept.outputs.accept != 'true'
         id: render_compare
         env:
           COMMIT_MESSAGE: ${{ steps.get_commit_msg.outputs.msg }}
@@ -146,9 +152,6 @@ jobs:
           source ${TEST_DIR}/src/preamble.sh
           set -eux
           GOLDEN_BRANCH=$(echo "${COMMIT_MESSAGE}" | python3 ${TEST_DIR}/src/commit_msg.py)
-          bash ${TEST_DIR}/generate.sh
-          # Build diffimg tool
-          ./build.sh release diffimg
 
           python3 ${TEST_DIR}/src/golden_manager.py \
                   --branch=${GOLDEN_BRANCH} \
@@ -172,10 +175,12 @@ jobs:
           fi
         shell: bash
       - uses: actions/upload-artifact@v4
+        if: steps.check_accept.outputs.accept != 'true'
         with:
           name: presubmit-renderdiff-result
           path: ./out/renderdiff
       - name: Compare result
+        if: steps.check_accept.outputs.accept != 'true'
         run: |
           ERROR_STR="${{ steps.render_compare.outputs.err }}"
           if [ -n "${ERROR_STR}" ]; then

.github/workflows/release.yml

@@ -188,6 +188,54 @@ jobs:
             const globber = await glob.create(['out/*.aar', 'out/*.apk', 'out/*.tgz'].join('\n'));
             await upload({ github, context }, await globber.glob(), TAG);
 
+  sonatype-publish:
+    name: sonatype-publish
+    runs-on: 'ubuntu-24.04-16core'
+    # Depends on the the Android build for the Android binaries.
+    # Depends on the Mac, Linux, and Windows builds for host tools.
+    needs: [build-mac, build-linux, build-windows, build-android]
+    if: github.event_name == 'release' || github.event.inputs.platform == 'android'
+
+    steps:
+      - name: Decide Git ref
+        id: git_ref
+        run: |
+          REF=${RELEASE_TAG:-${GITHUB_REF}}
+          TAG=${REF##*/}
+          echo "ref=${REF}" >> $GITHUB_OUTPUT
+          echo "tag=${TAG}" >> $GITHUB_OUTPUT
+      - uses: actions/checkout@v4.1.6
+        with:
+          ref: ${{ steps.git_ref.outputs.ref }}
+      - uses: actions/setup-java@v3
+        with:
+          distribution: 'temurin'
+          java-version: '17'
+      - uses: ./.github/actions/linux-prereq
+      - name: Download Android Release
+        run: |
+          gh release download ${TAG} \
+            --repo ${{ github.repository }} \
+            --pattern 'filament-*-android-native.tgz'
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          TAG: ${{ steps.git_ref.outputs.tag }}
+      - name: Unzip Android Release
+        run: |
+          mkdir -p out/android-release
+          tar -xzvf filament-${TAG}-android-native.tgz -C out/android-release/
+        env:
+          TAG: ${{ steps.git_ref.outputs.tag }}
+      - name: Publish To Sonatype
+        run: |
+          cd android
+          ./gradlew publishToSonatype closeSonatypeStagingRepository
+        env:
+          ORG_GRADLE_PROJECT_sonatypeUsername: ${{ secrets.SONATYPE_USERNAME }}
+          ORG_GRADLE_PROJECT_sonatypePassword: ${{ secrets.SONATYPE_PASSWORD }}
+          ORG_GRADLE_PROJECT_signingKey: ${{ secrets.MAVEN_SIGNING_KEY }}
+          ORG_GRADLE_PROJECT_signingPassword: ${{ secrets.MAVEN_SIGNING_PASSWORD }}
+
   build-ios:
     name: build-ios
     runs-on: macos-14-xlarge

NEW_RELEASE_NOTES.md

@@ -7,4 +7,4 @@ appropriate header in [RELEASE_NOTES.md](./RELEASE_NOTES.md).
 
 ## Release notes for next branch cut
 
-- engine: fix crash when using variance shadow maps
+- engine: add `MaterialInstance::setConstant()` and `MaterialInstance::getConstant()` methods. These allow for per-material instance specialization constant overrides.

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.69.4'
+    implementation 'com.google.android.filament:filament-android:1.69.5'
 }
 ```
 
@@ -50,7 +50,7 @@ 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', '~> 1.69.4'
+pod 'Filament', '~> 1.69.5'
 ```
 
 ## Documentation

RELEASE_NOTES.md

@@ -7,6 +7,15 @@ 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).
 
+## v1.70.0
+
+- engine: fix crash when using variance shadow maps
+- materials: better shadow normal-bias calculations [⚠️ **New Material Version**]
+
+## v1.69.5
+
+- engine: fix crash when using variance shadow maps
+
 ## v1.69.4
 
 

android/build.gradle

@@ -40,15 +40,21 @@
 //   - Build and upload artifacts with ./gradlew publish
 //   - Close and release staging repo on Nexus with ./gradlew closeAndReleaseStagingRepository
 //
-// The following is needed in ~/gradle/gradle.properties:
+// The following properties need to be set (either in ~/gradle/gradle.properties, on the command
+// line, or as environment variables, e.g.: ORG_GRADLE_PROJECT_property=value):
 //
 //     sonatypeUsername=nexus_user
 //     sonatypePassword=nexus_password
 //
+//     To sign with a key ring file:
 //     signing.keyId=pgp_key_id
 //     signing.password=pgp_key_password
 //     signing.secretKeyRingFile=/Users/user/.gnupg/maven_signing.key
 //
+//     To sign with in-memory keys (useful for CI):,
+//     signingKey=ASCII armored key (begins with -----BEGIN PGP PRIVATE KEY BLOCK-----)
+//     signingPassword=key password
+//
 
 buildscript {
     def path = providers

android/filament-android/src/main/cpp/MaterialInstance.cpp

@@ -20,6 +20,8 @@
 #include <filament/Texture.h>
 #include <filament/TextureSampler.h>
 
+#include "common/CallbackUtils.h"
+
 #include <math/mat3.h>
 #include <math/mat4.h>
 #include <math/vec2.h>
@@ -246,6 +248,69 @@ Java_com_google_android_filament_MaterialInstance_nSetFloatParameterArray(JNIEnv
     env->ReleaseStringUTFChars(name_, name);
 }
 
+extern "C"
+JNIEXPORT jboolean JNICALL
+Java_com_google_android_filament_MaterialInstance_nGetConstantBool(JNIEnv *env, jclass,
+        jlong nativeMaterialInstance, jstring name_) {
+    MaterialInstance* instance = (MaterialInstance*) nativeMaterialInstance;
+    const char *name = env->GetStringUTFChars(name_, 0);
+    jboolean result = instance->getConstant<bool>(name);
+    env->ReleaseStringUTFChars(name_, name);
+    return result;
+}
+
+extern "C"
+JNIEXPORT jfloat JNICALL
+Java_com_google_android_filament_MaterialInstance_nGetConstantFloat(JNIEnv *env, jclass,
+        jlong nativeMaterialInstance, jstring name_) {
+    MaterialInstance* instance = (MaterialInstance*) nativeMaterialInstance;
+    const char *name = env->GetStringUTFChars(name_, 0);
+    jfloat result = instance->getConstant<float>(name);
+    env->ReleaseStringUTFChars(name_, name);
+    return result;
+}
+
+extern "C"
+JNIEXPORT jint JNICALL
+Java_com_google_android_filament_MaterialInstance_nGetConstantInt(JNIEnv *env, jclass,
+        jlong nativeMaterialInstance, jstring name_) {
+    MaterialInstance* instance = (MaterialInstance*) nativeMaterialInstance;
+    const char *name = env->GetStringUTFChars(name_, 0);
+    jint result = instance->getConstant<int32_t>(name);
+    env->ReleaseStringUTFChars(name_, name);
+    return result;
+}
+
+extern "C"
+JNIEXPORT void JNICALL
+Java_com_google_android_filament_MaterialInstance_nSetConstantBool(JNIEnv *env, jclass,
+        jlong nativeMaterialInstance, jstring name_, jboolean x) {
+    MaterialInstance* instance = (MaterialInstance*) nativeMaterialInstance;
+    const char *name = env->GetStringUTFChars(name_, 0);
+    instance->setConstant<bool>(name, x);
+    env->ReleaseStringUTFChars(name_, name);
+}
+
+extern "C"
+JNIEXPORT void JNICALL
+Java_com_google_android_filament_MaterialInstance_nSetConstantFloat(JNIEnv *env, jclass,
+        jlong nativeMaterialInstance, jstring name_, jfloat x) {
+    MaterialInstance* instance = (MaterialInstance*) nativeMaterialInstance;
+    const char *name = env->GetStringUTFChars(name_, 0);
+    instance->setConstant<float>(name, x);
+    env->ReleaseStringUTFChars(name_, name);
+}
+
+extern "C"
+JNIEXPORT void JNICALL
+Java_com_google_android_filament_MaterialInstance_nSetConstantInt(JNIEnv *env, jclass,
+        jlong nativeMaterialInstance, jstring name_, jint x) {
+    MaterialInstance* instance = (MaterialInstance*) nativeMaterialInstance;
+    const char *name = env->GetStringUTFChars(name_, 0);
+    instance->setConstant<int32_t>(name, x);
+    env->ReleaseStringUTFChars(name_, name);
+}
+
 // defined in TextureSampler.cpp
 namespace filament::JniUtils {
     TextureSampler from_long(jlong params) noexcept;
@@ -580,3 +645,17 @@ Java_com_google_android_filament_MaterialInstance_nGetTransparencyMode(JNIEnv*,
     MaterialInstance* instance = (MaterialInstance*) nativeMaterialInstance;
     return (jint) instance->getTransparencyMode();
 }
+
+extern "C"
+JNIEXPORT void JNICALL
+Java_com_google_android_filament_MaterialInstance_nCompile(JNIEnv *env, jclass clazz,
+        jlong nativeMaterialInstance, jint priority, jint variants, jobject handler, jobject runnable) {
+    MaterialInstance* materialInstance = (MaterialInstance*) nativeMaterialInstance;
+    JniCallback* jniCallback = JniCallback::make(env, handler, runnable);
+    materialInstance->compile(
+            (MaterialInstance::CompilerPriorityQueue) priority,
+            (UserVariantFilterBit) variants,
+            jniCallback->getHandler(), [jniCallback](MaterialInstance*){
+                JniCallback::postToJavaAndDestroy(jniCallback);
+            });
+}

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

@@ -18,6 +18,7 @@ package com.google.android.filament;
 
 import androidx.annotation.IntRange;
 import androidx.annotation.NonNull;
+import androidx.annotation.Nullable;
 import androidx.annotation.Size;
 
 import com.google.android.filament.proguard.UsedByNative;
@@ -142,6 +143,28 @@ public class MaterialInstance {
         return mMaterial;
     }
 
+    /**
+     * Asynchronously ensures that a subset of this MaterialInstance's variants are compiled.
+     *
+     * <p>This function behaves identically to {@link Material#compile}, but takes into account
+     * the specific constants overridden by {@link #setConstant}.</p>
+     *
+     * @param priority      Priority of the compile command.
+     * @param variants      Variants to include to the compile command.
+     * @param handler       An {@link java.util.concurrent.Executor Executor}. On Android this can also be a {@link android.os.Handler Handler}.
+     * @param callback      callback called on the main thread when the compilation is done on
+     *                      by backend.
+     *
+     * @see Material#compile
+     * @see #setConstant
+     */
+    public void compile(@NonNull Material.CompilerPriorityQueue priority,
+                        int variants,
+                        @Nullable Object handler,
+                        @Nullable Runnable callback) {
+        nCompile(getNativeObject(), priority.ordinal(), variants, handler, callback);
+    }
+
     /** @return the name associated with this instance */
     @NonNull
     public String getName() {
@@ -402,6 +425,69 @@ public class MaterialInstance {
         nSetParameterFloat4(getNativeObject(), name, color[0], color[1], color[2], color[3]);
     }
 
+    /**
+     * Overrides a specialization constant of this material instance.
+     *
+     * @param name  The name of the constant as defined in the material.
+     * @param value The value of the constant.
+     * @see Material.Builder#constant
+     */
+    public void setConstant(@NonNull String name, boolean value) {
+        nSetConstantBool(getNativeObject(), name, value);
+    }
+
+    /**
+     * Overrides a specialization constant of this material instance.
+     *
+     * @param name  The name of the constant as defined in the material.
+     * @param value The value of the constant.
+     * @see Material.Builder#constant
+     */
+    public void setConstant(@NonNull String name, float value) {
+        nSetConstantFloat(getNativeObject(), name, value);
+    }
+
+    /**
+     * Overrides a specialization constant of this material instance.
+     *
+     * @param name  The name of the constant as defined in the material.
+     * @param value The value of the constant.
+     * @see Material.Builder#constant
+     */
+    public void setConstant(@NonNull String name, int value) {
+        nSetConstantInt(getNativeObject(), name, value);
+    }
+
+    /**
+     * Gets the value of a specialization constant by name.
+     *
+     * @param name  The name of the constant as defined in the material.
+     * @return The value of the constant.
+     */
+    public boolean getConstantBoolean(@NonNull String name) {
+        return nGetConstantBool(getNativeObject(), name);
+    }
+
+    /**
+     * Gets the value of a specialization constant by name.
+     *
+     * @param name  The name of the constant as defined in the material.
+     * @return The value of the constant.
+     */
+    public float getConstantFloat(@NonNull String name) {
+        return nGetConstantFloat(getNativeObject(), name);
+    }
+
+    /**
+     * Gets the value of a specialization constant by name.
+     *
+     * @param name  The name of the constant as defined in the material.
+     * @return The value of the constant.
+     */
+    public int getConstantInt(@NonNull String name) {
+        return nGetConstantInt(getNativeObject(), name);
+    }
+
     /**
      * Set-up a custom scissor rectangle; by default it is disabled.
      *
@@ -937,6 +1023,17 @@ public class MaterialInstance {
             @NonNull String name, int element, @NonNull @Size(min = 1) float[] v,
             @IntRange(from = 0) int offset, @IntRange(from = 1) int count);
 
+    private static native boolean nGetConstantBool(long nativeMaterialInstance, @NonNull String name);
+    private static native float nGetConstantFloat(long nativeMaterialInstance, @NonNull String name);
+    private static native int nGetConstantInt(long nativeMaterialInstance, @NonNull String name);
+
+    private static native void nSetConstantBool(long nativeMaterialInstance,
+            @NonNull String name, boolean x);
+    private static native void nSetConstantFloat(long nativeMaterialInstance,
+            @NonNull String name, float x);
+    private static native void nSetConstantInt(long nativeMaterialInstance,
+            @NonNull String name, int x);
+
     private static native void nSetParameterTexture(long nativeMaterialInstance,
             @NonNull String name, long nativeTexture, long sampler);
 
@@ -1000,4 +1097,5 @@ public class MaterialInstance {
     private static native int nGetDepthFunc(long nativeMaterialInstance);
     private static native void nSetTransparencyMode(long nativeMaterialInstance, int mode);
     private static native int nGetTransparencyMode(long nativeMaterialInstance);
+    private static native void nCompile(long nativeMaterialInstance, int priority, int variants, Object handler, Runnable callback);
 }

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

@@ -2121,7 +2121,7 @@ public class View {
          */
         public boolean highPrecision = false;
         /**
-         * VSM minimum variance scale, must be positive.
+         * @deprecated has no effect.
          */
         public float minVarianceScale = 0.5f;
         /**

android/gradle.properties

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.69.4
+VERSION_NAME=1.69.5
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 

android/gradle/gradle-mvn-push.gradle

@@ -94,6 +94,13 @@ afterEvaluate { project ->
     }
 
     signing {
+        def signingKey = findProperty("signingKey")
+        def signingPassword = findProperty("signingPassword")
+        if (signingKey && signingPassword) {
+            println("Signing with in-memory keys")
+            useInMemoryPgpKeys(signingKey, signingPassword)
+        }
+
         publishing.publications.all { publication ->
             sign publication
         }

android/samples/sample-render-validation/build.gradle

@@ -26,7 +26,8 @@ tasks.register('copyDamagedHelmetGltf', Copy) {
 preBuild.dependsOn copyDamagedHelmetGltf
 
 clean.doFirst {
-    delete "src/main/assets"
+    delete "src/main/assets/envs"
+    delete "src/main/assets/models"
 }
 
 android {
@@ -48,6 +49,7 @@ android {
 dependencies {
     implementation deps.kotlin
     implementation 'androidx.constraintlayout:constraintlayout:2.1.4'
+    implementation 'com.google.android.material:material:1.10.0'
     implementation deps.coroutines.core
     implementation project(':filament-android')
     implementation project(':gltfio-android')

android/samples/sample-render-validation/src/main/AndroidManifest.xml

@@ -12,10 +12,11 @@
         android:supportsRtl="true"
         android:largeHeap="true"
         android:requestLegacyExternalStorage="true"
-        android:theme="@android:style/Theme.NoTitleBar">
+        android:theme="@style/Theme.Material3.DayNight.NoActionBar">
         <activity
             android:name=".MainActivity"
             android:exported="true"
+            android:launchMode="singleTop"
             android:configChanges="orientation|screenSize|screenLayout|keyboardHidden"
             android:screenOrientation="fullSensor">
             <intent-filter>

android/samples/sample-render-validation/src/main/assets/default_test.json

@@ -0,0 +1,899 @@
+{
+  "name": "Default Test",
+  "backends": [
+    "opengl"
+  ],
+  "models": {
+    "DamagedHelmet": "helmet.glb"
+  },
+  "presets": [
+    {
+      "name": "base",
+      "models": [
+        "DamagedHelmet"
+      ],
+      "rendering": {
+        "camera": {
+          "enabled": true,
+          "horizontalFov": 45.0,
+          "center": [
+            0,
+            0,
+            0
+          ],
+          "lookAt": [
+            0,
+            0,
+            -1
+          ]
+        },
+        "view": {
+          "postProcessingEnabled": true,
+          "dithering": "NONE"
+        },
+        "tolerance": {
+          "maxAbsDiff": 0.1,
+          "maxFailingPixelsFraction": 0.0
+        }
+      }
+    },
+    {
+      "name": "tilted",
+      "rendering": {
+        "camera": {
+          "enabled": true,
+          "horizontalFov": 45.0,
+          "center": [
+            -4,
+            -2,
+            -3
+          ],
+          "lookAt": [
+            0,
+            0,
+            -4
+          ]
+        }
+      }
+    }
+  ],
+  "tests": [
+    {
+      "name": "basic",
+      "apply_presets": [
+        "base"
+      ]
+    },
+    {
+      "name": "rotated",
+      "apply_presets": [
+        "base",
+        "tilted"
+      ]
+    },
+    {
+      "name": "ssao",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true
+      }
+    },
+    {
+      "name": "msaa",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.msaa.enabled": true
+      }
+    },
+    {
+      "name": "bloom",
+      "apply_presets": [
+        "base",
+        "tilted"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true
+      }
+    },
+    {
+      "name": "aa_none",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.antiAliasing": "NONE"
+      }
+    },
+    {
+      "name": "aa_fxaa",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.antiAliasing": "FXAA"
+      }
+    },
+    {
+      "name": "dithering_none",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dithering": "NONE"
+      }
+    },
+    {
+      "name": "msaa_8",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.msaa.enabled": true,
+        "view.msaa.sampleCount": 8
+      }
+    },
+    {
+      "name": "taa_custom",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.taa.enabled": true,
+        "view.taa.feedback": 0.2,
+        "view.taa.jitterPattern": "HALTON_23_X16"
+      }
+    },
+    {
+      "name": "ssao_gtao",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.aoType": "GTAO"
+      }
+    },
+    {
+      "name": "ssao_sao",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.aoType": "SAO"
+      }
+    },
+    {
+      "name": "ssao_radius_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.radius": 1.0
+      }
+    },
+    {
+      "name": "ssao_radius_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.radius": 0.1
+      }
+    },
+    {
+      "name": "ssao_power_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.power": 2.0
+      }
+    },
+    {
+      "name": "ssao_power_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.power": 0.5
+      }
+    },
+    {
+      "name": "ssao_bias_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.bias": 0.05
+      }
+    },
+    {
+      "name": "ssao_resolution_half",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.resolution": 0.5
+      }
+    },
+    {
+      "name": "ssao_resolution_full",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.resolution": 1.0
+      }
+    },
+    {
+      "name": "ssao_intensity_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.intensity": 2.0
+      }
+    },
+    {
+      "name": "ssao_bent_normals",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.bentNormals": true
+      }
+    },
+    {
+      "name": "ssao_quality_ultra",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.quality": "ULTRA"
+      }
+    },
+    {
+      "name": "ssao_quality_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.quality": "LOW"
+      }
+    },
+    {
+      "name": "ssao_lowPassFilter_ultra",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.lowPassFilter": "ULTRA"
+      }
+    },
+    {
+      "name": "ssao_upsampling_ultra",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.ssao.enabled": true,
+        "view.ssao.upsampling": "ULTRA"
+      }
+    },
+    {
+      "name": "ssr_basic",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true
+      }
+    },
+    {
+      "name": "ssr_thickness_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true,
+        "view.screenSpaceReflections.thickness": 0.5
+      }
+    },
+    {
+      "name": "ssr_bias_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true,
+        "view.screenSpaceReflections.bias": 0.1
+      }
+    },
+    {
+      "name": "ssr_maxDistance_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true,
+        "view.screenSpaceReflections.maxDistance": 5.0
+      }
+    },
+    {
+      "name": "ssr_stride_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true,
+        "view.screenSpaceReflections.stride": 4.0
+      }
+    },
+    {
+      "name": "ssr_stride_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true,
+        "view.screenSpaceReflections.stride": 1.0
+      }
+    },
+    {
+      "name": "ssr_thickness_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true,
+        "view.screenSpaceReflections.thickness": 0.01
+      }
+    },
+    {
+      "name": "ssr_maxDistance_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true,
+        "view.screenSpaceReflections.maxDistance": 1.0
+      }
+    },
+    {
+      "name": "ssr_bias_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true,
+        "view.screenSpaceReflections.bias": 0.001
+      }
+    },
+    {
+      "name": "ssr_quality_combo",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.screenSpaceReflections.enabled": true,
+        "view.screenSpaceReflections.thickness": 0.2,
+        "view.screenSpaceReflections.stride": 2.0
+      }
+    },
+    {
+      "name": "bloom_levels_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.levels": 8
+      }
+    },
+    {
+      "name": "bloom_levels_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.levels": 4
+      }
+    },
+    {
+      "name": "bloom_resolution_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.resolution": 1024
+      }
+    },
+    {
+      "name": "bloom_strength_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.strength": 0.5
+      }
+    },
+    {
+      "name": "bloom_strength_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.strength": 0.01
+      }
+    },
+    {
+      "name": "bloom_blendMode_interpolate",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.blendMode": "INTERPOLATE"
+      }
+    },
+    {
+      "name": "bloom_no_threshold",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.threshold": false
+      }
+    },
+    {
+      "name": "bloom_quality_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.quality": "HIGH"
+      }
+    },
+    {
+      "name": "bloom_lensflare",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.lensFlare": true
+      }
+    },
+    {
+      "name": "bloom_lensflare_no_starburst",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.lensFlare": true,
+        "view.bloom.starburst": false
+      }
+    },
+    {
+      "name": "bloom_lensflare_chromatic",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.lensFlare": true,
+        "view.bloom.chromaticAberration": 0.05
+      }
+    },
+    {
+      "name": "bloom_lensflare_ghosts",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.lensFlare": true,
+        "view.bloom.ghostCount": 8
+      }
+    },
+    {
+      "name": "bloom_lensflare_ghostSpacing",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.lensFlare": true,
+        "view.bloom.ghostSpacing": 0.8
+      }
+    },
+    {
+      "name": "bloom_halo_thick",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.lensFlare": true,
+        "view.bloom.haloThickness": 0.2
+      }
+    },
+    {
+      "name": "bloom_halo_radius",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.bloom.enabled": true,
+        "view.bloom.lensFlare": true,
+        "view.bloom.haloRadius": 0.5
+      }
+    },
+    {
+      "name": "dof_basic",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true
+      }
+    },
+    {
+      "name": "dof_cocScale_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true,
+        "view.dof.cocScale": 2.0
+      }
+    },
+    {
+      "name": "dof_cocScale_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true,
+        "view.dof.cocScale": 0.5
+      }
+    },
+    {
+      "name": "dof_cocAspectRatio",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true,
+        "view.dof.cocAspectRatio": 2.0
+      }
+    },
+    {
+      "name": "dof_maxApertureDiameter",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true,
+        "view.dof.maxApertureDiameter": 0.05
+      }
+    },
+    {
+      "name": "dof_filter_none",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true,
+        "view.dof.filter": "NONE"
+      }
+    },
+    {
+      "name": "dof_nativeResolution",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true,
+        "view.dof.nativeResolution": true
+      }
+    },
+    {
+      "name": "dof_rings_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true,
+        "view.dof.foregroundRingCount": 5,
+        "view.dof.backgroundRingCount": 5
+      }
+    },
+    {
+      "name": "dof_rings_low",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true,
+        "view.dof.foregroundRingCount": 3,
+        "view.dof.backgroundRingCount": 3
+      }
+    },
+    {
+      "name": "dof_max_coc",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.dof.enabled": true,
+        "view.dof.maxForegroundCOC": 16,
+        "view.dof.maxBackgroundCOC": 16
+      }
+    },
+    {
+      "name": "fog_basic",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true
+      }
+    },
+    {
+      "name": "fog_distance",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true,
+        "view.fog.distance": 10.0
+      }
+    },
+    {
+      "name": "fog_cutOffDistance",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true,
+        "view.fog.cutOffDistance": 100.0
+      }
+    },
+    {
+      "name": "fog_maximumOpacity",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true,
+        "view.fog.maximumOpacity": 0.5
+      }
+    },
+    {
+      "name": "fog_height",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true,
+        "view.fog.height": 5.0
+      }
+    },
+    {
+      "name": "fog_heightFalloff",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true,
+        "view.fog.heightFalloff": 0.5
+      }
+    },
+    {
+      "name": "fog_density_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true,
+        "view.fog.density": 0.5
+      }
+    },
+    {
+      "name": "fog_inScatteringStart",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true,
+        "view.fog.inScatteringStart": 5.0
+      }
+    },
+    {
+      "name": "fog_inScatteringSize",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true,
+        "view.fog.inScatteringSize": 10.0
+      }
+    },
+    {
+      "name": "fog_fogColorFromIbl",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.fog.enabled": true,
+        "view.fog.fogColorFromIbl": true
+      }
+    },
+    {
+      "name": "vignette_basic",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.vignette.enabled": true
+      }
+    },
+    {
+      "name": "vignette_midPoint",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.vignette.enabled": true,
+        "view.vignette.midPoint": 0.8
+      }
+    },
+    {
+      "name": "vignette_roundness_circle",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.vignette.enabled": true,
+        "view.vignette.roundness": 1.0
+      }
+    },
+    {
+      "name": "vignette_roundness_rect",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.vignette.enabled": true,
+        "view.vignette.roundness": 0.0
+      }
+    },
+    {
+      "name": "vignette_feather_sharp",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.vignette.enabled": true,
+        "view.vignette.feather": 0.1
+      }
+    },
+    {
+      "name": "cg_filmic",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.colorGrading.toneMapping": "FILMIC"
+      }
+    },
+    {
+      "name": "cg_aces",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.colorGrading.toneMapping": "ACES"
+      }
+    },
+    {
+      "name": "cg_agx",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.colorGrading.toneMapping": "AGX"
+      }
+    },
+    {
+      "name": "cg_pbr_neutral",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.colorGrading.toneMapping": "PBR_NEUTRAL"
+      }
+    },
+    {
+      "name": "cg_contrast_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.colorGrading.contrast": 1.5
+      }
+    },
+    {
+      "name": "cg_saturation_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.colorGrading.saturation": 1.5
+      }
+    },
+    {
+      "name": "cg_exposure_high",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.colorGrading.exposure": 1.0
+      }
+    },
+    {
+      "name": "shadow_vsm",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.shadowType": "VSM"
+      }
+    },
+    {
+      "name": "shadow_vsm_anisotropy",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.shadowType": "VSM",
+        "view.vsmShadowOptions.anisotropy": 2
+      }
+    },
+    {
+      "name": "shadow_vsm_highPrecision",
+      "apply_presets": [
+        "base"
+      ],
+      "rendering": {
+        "view.shadowType": "VSM",
+        "view.vsmShadowOptions.highPrecision": true
+      }
+    }
+  ]
+}

android/samples/sample-render-validation/src/main/java/com/google/android/filament/validation/MainActivity.kt

@@ -17,36 +17,35 @@
 package com.google.android.filament.validation
 
 import android.app.Activity
+import android.app.AlertDialog
+import android.content.Intent
 import android.graphics.Bitmap
+import android.graphics.Color
 import android.os.Bundle
+import android.text.Html
 import android.util.Log
 import android.view.Choreographer
 import android.view.SurfaceView
 import android.view.View
 import android.view.WindowManager
+import android.widget.ArrayAdapter
+import android.widget.Button
 import android.widget.ImageView
 import android.widget.LinearLayout
-import android.widget.ScrollView
+import android.widget.Spinner
 import android.widget.TextView
+import com.google.android.filament.utils.KTX1Loader
 import com.google.android.filament.utils.ModelViewer
 import com.google.android.filament.utils.Utils
 import kotlinx.coroutines.CoroutineScope
 import kotlinx.coroutines.Dispatchers
 import kotlinx.coroutines.launch
 import kotlinx.coroutines.withContext
-import com.google.android.filament.utils.KTX1Loader
-import com.google.android.filament.IndirectLight
-import com.google.android.filament.Skybox
-import android.graphics.Color
 import java.io.File
-import java.io.FileOutputStream
-import java.net.HttpURLConnection
-import java.net.URL
 import java.nio.ByteBuffer
-import android.widget.ArrayAdapter
-import android.widget.Button
-import android.widget.Spinner
-import android.widget.AdapterView
+import java.text.SimpleDateFormat
+import java.util.Date
+import java.util.Locale
 
 class MainActivity : Activity(), ValidationRunner.Callback {
 
@@ -62,13 +61,17 @@ class MainActivity : Activity(), ValidationRunner.Callback {
     private lateinit var choreographer: Choreographer
     private lateinit var modelViewer: ModelViewer
     private lateinit var statusTextView: TextView
+    private lateinit var testResultsHeader: TextView
     private lateinit var resultsContainer: LinearLayout
     private lateinit var inputManager: ValidationInputManager
     private var currentInput: ValidationInputManager.ValidationInput? = null
-    private lateinit var modeSpinner: Spinner
-    private lateinit var runButton: Button
-    private var resultManager: ValidationResultManager? = null
 
+    // UI Elements
+    private lateinit var runButton: Button
+    private lateinit var loadButton: Button
+    private lateinit var optionsButton: Button
+
+    private var resultManager: ValidationResultManager? = null
     private var validationRunner: ValidationRunner? = null
 
     // Frame callback
@@ -89,25 +92,53 @@ class MainActivity : Activity(), ValidationRunner.Callback {
         surfaceView.holder.setFixedSize(512, 512)
 
         statusTextView = findViewById(R.id.status_text)
-        modeSpinner = findViewById(R.id.mode_spinner)
-        runButton = findViewById(R.id.run_button)
+        testResultsHeader = findViewById(R.id.test_results_header)
         resultsContainer = findViewById(R.id.results_container)
 
-        // Setup Spinner
-        val modes = arrayOf("Run Validation", "Generate Goldens")
-        val adapter = ArrayAdapter(this, android.R.layout.simple_spinner_item, modes)
-        adapter.setDropDownViewResource(android.R.layout.simple_spinner_dropdown_item)
-        modeSpinner.adapter = adapter
+        runButton = findViewById(R.id.run_button)
+        loadButton = findViewById(R.id.load_button)
+        optionsButton = findViewById(R.id.options_button)
 
         // Setup Run Button
         runButton.setOnClickListener {
             currentInput?.let { input ->
-                val generateGoldens = modeSpinner.selectedItemPosition == 1
-                val newInput = input.copy(generateGoldens = generateGoldens)
-                startValidation(newInput)
+                // Always use the generateGoldens flag from the intent/input
+                startValidation(input)
             }
         }
 
+        // Setup Load Button
+        loadButton.setOnClickListener {
+            showLoadDialog()
+        }
+
+        // Setup Options Menu Button
+        optionsButton.setOnClickListener { view ->
+            val popup = android.widget.PopupMenu(this, view)
+            popup.menu.add(0, 1, 0, "Generate Golden")
+            popup.menu.add(0, 2, 0, "Export Test")
+            popup.menu.add(0, 3, 0, "Export Result")
+            popup.menu.add(0, 4, 0, "Test ADB Info")
+            popup.menu.add(0, 5, 0, "Result ADB Info")
+
+            popup.setOnMenuItemClickListener { item ->
+                when (item.itemId) {
+                    1 -> {
+                        currentInput?.let { input ->
+                            val goldenInput = input.copy(generateGoldens = true)
+                            startValidation(goldenInput)
+                        }
+                    }
+                    2 -> exportTestBundleAction()
+                    3 -> exportTestResultsAction()
+                    4 -> showTestAdbInfo()
+                    5 -> showResultAdbInfo()
+                }
+                true
+            }
+            popup.show()
+        }
+
         window.addFlags(WindowManager.LayoutParams.FLAG_KEEP_SCREEN_ON)
 
         choreographer = Choreographer.getInstance()
@@ -120,6 +151,134 @@ class MainActivity : Activity(), ValidationRunner.Callback {
         handleIntent()
     }
 
+    private fun showLoadDialog() {
+        val exportDir = getExternalFilesDir(null) ?: filesDir
+        // Filter out result zips (starting with "results_") to only show test bundles
+        val zips = exportDir.listFiles { _, name ->
+            name.endsWith(".zip") && !name.startsWith("results_")
+        }?.sortedByDescending { it.lastModified() } ?: emptyList()
+
+        if (zips.isEmpty()) {
+            AlertDialog.Builder(this)
+                .setTitle("Load Test")
+                .setMessage("No test bundles found.")
+                .setPositiveButton("OK", null)
+                .show()
+            return
+        }
+
+        val builder = AlertDialog.Builder(this)
+        builder.setTitle("Select Test Bundle")
+
+        val items = zips.map { it.name }.toTypedArray()
+
+        builder.setItems(items) { dialog, which ->
+            val selectedFile = zips[which]
+            loadZipBundle(selectedFile)
+            dialog.dismiss()
+        }
+
+        builder.setNegativeButton("Cancel", null)
+        builder.show()
+    }
+
+    private fun showTestAdbInfo() {
+        val exportDir = getExternalFilesDir(null) ?: filesDir
+        val path = exportDir.absolutePath
+        val isInternal = path.startsWith(filesDir.absolutePath)
+        val message = StringBuilder()
+
+        message.append("Storage Path: $path<br><br>")
+
+        message.append("<b>--- PULL FROM DEVICE ---</b><br>")
+        if (isInternal) {
+            message.append("<tt>adb shell \"run-as $packageName cat files/&lt;filename&gt;\" &gt; &lt;filename&gt;</tt><br><br>")
+        } else {
+            message.append("<tt>adb pull $path/&lt;filename&gt; .</tt><br><br>")
+        }
+
+        message.append("<b>--- PUSH TO DEVICE ---</b><br>")
+        if (isInternal) {
+            message.append("1. <tt>adb push &lt;filename&gt; /sdcard/Download/</tt><br>")
+            message.append("2. <tt>adb shell \"run-as $packageName cp /sdcard/Download/&lt;filename&gt; files/\"</tt><br>")
+        } else {
+            message.append("<tt>adb push &lt;filename&gt; $path/</tt><br>")
+        }
+        message.append("<br>Note: Use underscores instead of spaces in &lt;filename&gt;.")
+
+        AlertDialog.Builder(this)
+            .setTitle("Test Bundle ADB Info")
+            .setMessage(Html.fromHtml(message.toString(), Html.FROM_HTML_MODE_LEGACY))
+            .setPositiveButton("OK", null)
+            .show()
+    }
+
+    private fun showResultAdbInfo() {
+        val exportDir = getExternalFilesDir(null) ?: filesDir
+        val path = exportDir.absolutePath
+        val isInternal = path.startsWith(filesDir.absolutePath)
+        val message = StringBuilder()
+
+        message.append("<b>--- PULL RESULTS ---</b><br>")
+        if (isInternal) {
+            message.append("<tt>adb shell \"run-as $packageName cat files/&lt;filename&gt;\" &gt; &lt;filename&gt;</tt><br><br>")
+        } else {
+            message.append("<tt>adb pull $path/&lt;filename&gt; .</tt><br><br>")
+        }
+
+        message.append("<b>--- AVAILABLE RESULTS ---</b><br>")
+        val zips = exportDir.listFiles { _, name ->
+            name.endsWith(".zip") && name.startsWith("results_")
+        }?.sortedByDescending { it.lastModified() } ?: emptyList()
+
+        if (zips.isEmpty()) {
+            message.append("No result zips found.<br>")
+        } else {
+            zips.forEach { file ->
+                message.append("${file.name}<br>")
+            }
+        }
+
+        AlertDialog.Builder(this)
+            .setTitle("Result ADB Info")
+            .setMessage(Html.fromHtml(message.toString(), Html.FROM_HTML_MODE_LEGACY))
+            .setPositiveButton("OK", null)
+            .show()
+    }
+
+    private fun loadZipBundle(file: File) {
+         statusTextView.text = "Loading ${file.name}..."
+         CoroutineScope(Dispatchers.Main).launch {
+             try {
+                 val config = inputManager.loadFromZip(file)
+                 val baseDir = getExternalFilesDir(null) ?: filesDir
+                 val outputDir = File(baseDir, "validation_results").apply { mkdirs() }
+
+                 // Clear existing results UI and state
+                 resultsContainer.removeAllViews()
+                 resultManager = null
+
+                 val newInput = ValidationInputManager.ValidationInput(
+                     config = config,
+                     outputDir = outputDir,
+                     generateGoldens = false,
+                     autoRun = false,
+                     autoExport = false,
+                     autoExportResults = false,
+                     sourceZip = file
+                 )
+
+                 currentInput = newInput
+                 statusTextView.text = "Loaded ${config.name}"
+                 Log.i(TAG, "Setting header to: Test Results: ${config.name}")
+                 testResultsHeader.text = "${config.name}"
+             } catch (e: Exception) {
+                 Log.e(TAG, "Failed to load zip", e)
+                 statusTextView.text = "Error: ${e.message}"
+             }
+         }
+    }
+
     private fun createIndirectLight() {
         try {
             val engine = modelViewer.engine
@@ -156,12 +315,18 @@ class MainActivity : Activity(), ValidationRunner.Callback {
         CoroutineScope(Dispatchers.Main).launch {
             try {
                 val input = inputManager.resolveConfig(intent)
+
+                // Update header
+                Log.i(TAG, "handleIntent: Setting header to: Test Results: ${input.config.name}")
+                testResultsHeader.text = "${input.config.name}"
                 currentInput = input
 
-                // Sync spinner with intent
-                modeSpinner.setSelection(if (input.generateGoldens) 1 else 0)
-
-                startValidation(input)
+                if (input.autoRun) {
+                    startValidation(input)
+                } else {
+                    // Just show status
+                    statusTextView.text = "Ready: ${input.config.name}"
+                }
             } catch (e: Exception) {
                 Log.e(TAG, "Failed to resolve config", e)
                 statusTextView.text = "Error: ${e.message}"
@@ -175,6 +340,8 @@ class MainActivity : Activity(), ValidationRunner.Callback {
             Log.i(TAG, "Starting validation with config: ${input.config.name}")
             Log.i(TAG, "Output dir: ${input.outputDir.absolutePath}")
 
+            testResultsHeader.text = "${input.config.name}"
+
             resultManager = ValidationResultManager(input.outputDir)
 
             validationRunner = ValidationRunner(this, modelViewer, input.config, resultManager!!)
@@ -182,9 +349,6 @@ class MainActivity : Activity(), ValidationRunner.Callback {
             validationRunner?.generateGoldens = input.generateGoldens
             validationRunner?.start()
 
-            // Sync spinner in case it was called programmatically or changed implicitly
-            modeSpinner.setSelection(if (input.generateGoldens) 1 else 0)
-
         } catch (e: Exception) {
             Log.e(TAG, "Failed to start validation", e)
             statusTextView.text = "Error: ${e.message}"
@@ -196,6 +360,12 @@ class MainActivity : Activity(), ValidationRunner.Callback {
         choreographer.postFrameCallback(frameScheduler)
     }
 
+    override fun onNewIntent(intent: Intent) {
+        super.onNewIntent(intent)
+        setIntent(intent)
+        handleIntent()
+    }
+
     override fun onPause() {
         super.onPause()
         choreographer.removeFrameCallback(frameScheduler)
@@ -221,13 +391,30 @@ class MainActivity : Activity(), ValidationRunner.Callback {
             resultContainer.orientation = LinearLayout.VERTICAL
             resultContainer.setPadding(0, 10, 0, 20)
 
-            // Header
-            val resultView = TextView(this)
-            resultView.text = "${result.testName}: ${if(result.passed) "PASS" else "FAIL"} (Diff: ${result.diffMetric})"
-            resultView.setTextColor(if(result.passed) Color.GREEN else Color.RED)
-            resultView.textSize = 16f
-            resultView.setTypeface(null, android.graphics.Typeface.BOLD)
-            resultContainer.addView(resultView)
+            // Header Layout
+            val headerRow = LinearLayout(this)
+            headerRow.orientation = LinearLayout.HORIZONTAL
+            headerRow.gravity = android.view.Gravity.CENTER_VERTICAL
+
+            // Status Icon + Name
+            val statusView = TextView(this)
+            val icon = if (result.passed) "✔" else "✖"
+
+            statusView.text = "$icon ${result.testName}"
+            statusView.setTextColor(
+                if (result.passed) Color.parseColor("#4CAF50") else Color.parseColor("#F44336")
+            )
+            statusView.textSize = 12f
+            statusView.setTypeface(null, android.graphics.Typeface.BOLD)
+            headerRow.addView(statusView)
+
+            // Diff Metric (only show if it's relevant/there's a diff or we just always show it like before)
+            val diffView = TextView(this)
+            diffView.text = "  (Diff: ${result.diffMetric})"
+            diffView.textSize = 12f
+            diffView.setTextColor(Color.GRAY)
+            headerRow.addView(diffView)
+            resultContainer.addView(headerRow)
 
             // Images Row
             val imagesRow = LinearLayout(this)
@@ -241,7 +428,7 @@ class MainActivity : Activity(), ValidationRunner.Callback {
 
                     val labelView = TextView(this)
                     labelView.text = label
-                    labelView.textSize = 12f
+                    labelView.textSize = 9f
                     container.addView(labelView)
 
                     val iv = ImageView(this)
@@ -274,7 +461,46 @@ class MainActivity : Activity(), ValidationRunner.Callback {
     override fun onAllTestsFinished() {
         runOnUiThread {
             statusTextView.text = "All tests finished!"
-            Log.i(TAG, "All tests finished")
+            Log.i(TAG, "All tests finished " + if (currentInput?.autoExport == true) "Exporting bundle" else "x")
+
+            if (currentInput?.autoExport == true) {
+                exportTestBundleAction()
+            }
+            if (currentInput?.autoExportResults == true) {
+                exportTestResultsAction()
+            }
+        }
+    }
+
+    private fun exportTestBundleAction() {
+        currentInput?.let { input ->
+            val timestamp = SimpleDateFormat("yyyyMMdd_HHmmss", Locale.US).format(Date())
+            val rm = resultManager ?: ValidationResultManager(input.outputDir)
+            val zip = rm.exportTestBundle(input.config, timestamp)
+            if (zip != null) {
+                val msg = "Exported Bundle: ${zip.name}"
+                statusTextView.text = msg
+                Log.i(TAG, "Exported test bundle to ${zip.absolutePath}")
+            } else {
+                statusTextView.text = "Export Bundle failed"
+                Log.e(TAG, "Export Bundle failed")
+            }
+        }
+    }
+
+    private fun exportTestResultsAction() {
+        currentInput?.let { input ->
+            val timestamp = SimpleDateFormat("yyyyMMdd_HHmmss", Locale.US).format(Date())
+            val rm = resultManager ?: ValidationResultManager(input.outputDir)
+            val zip = rm.exportTestResults(input.sourceZip, timestamp)
+            if (zip != null) {
+                val msg = "Exported Results: ${zip.name}"
+                statusTextView.text = msg
+                Log.i(TAG, "Exported results to ${zip.absolutePath}")
+            } else {
+                statusTextView.text = "Export Results failed"
+                Log.e(TAG, "Export Results failed")
+            }
         }
     }
 
@@ -301,50 +527,3 @@ class MainActivity : Activity(), ValidationRunner.Callback {
         }
     }
 }
-
-/*
- * Scripts for reference:
- *
- * generate_goldens.sh:
- * --------------------
- * #!/bin/bash
- * set -e
- *
- * # Config path (on device)
- * CONFIG_PATH=$1
- * if [ -z "$CONFIG_PATH" ]; then
- *     echo "Usage: $0 <device_config_path>"
- *     echo "Example: $0 /sdcard/Android/data/com.google.android.filament.validation/files/default_test.json"
- *     exit 1
- * fi
- *
- * echo "Starting Golden Generation for $CONFIG_PATH..."
- * adb shell am force-stop com.google.android.filament.validation
- * adb shell am start -n com.google.android.filament.validation/.MainActivity \
- *     -e test_config "$CONFIG_PATH" \
- *     --ez generate_goldens true
- *
- * echo "Check device or logcat for progress."
- * echo "adb logcat -s FilamentValidation:I ValidationRunner:I"
- * echo "To pull results: ./samples/sample-render-validation/pull_goldens.sh"
- *
- * pull_goldens.sh:
- * ----------------
- * #!/bin/bash
- * set -e
- *
- * # Default destination is local golden directory relative to script
- * SCRIPT_DIR=$(cd $(dirname $0); pwd)
- * DEST_DIR=${1:-"$SCRIPT_DIR/golden"}
- *
- * echo "Pulling goldens to $DEST_DIR..."
- * mkdir -p "$DEST_DIR"
- *
- * # Path on device
- * DEVICE_GOLDEN_DIR="/storage/emulated/0/Android/data/com.google.android.filament.validation/files/golden/."
- *
- * adb pull "$DEVICE_GOLDEN_DIR" "$DEST_DIR"
- *
- * echo "Done."
- * ls -l "$DEST_DIR"
- */

android/samples/sample-render-validation/src/main/java/com/google/android/filament/validation/ValidationInputManager.kt

@@ -26,6 +26,7 @@ import java.io.File
 import java.io.FileOutputStream
 import java.net.HttpURLConnection
 import java.net.URL
+import java.util.zip.ZipFile
 
 /**
  * Handles the retrieval and preparation of test configuration and assets.
@@ -42,7 +43,11 @@ class ValidationInputManager(private val context: Context) {
     data class ValidationInput(
         val config: RenderTestConfig,
         val outputDir: File,
-        val generateGoldens: Boolean
+        val generateGoldens: Boolean,
+        val autoRun: Boolean = false,
+        val autoExport: Boolean = false,
+        val autoExportResults: Boolean = false,
+        val sourceZip: File? = null
     )
 
     /**
@@ -53,22 +58,111 @@ class ValidationInputManager(private val context: Context) {
         val testConfigPath = intent.getStringExtra("test_config")
         val urlConfig = intent.getStringExtra("url_config")
         val urlModelsBase = intent.getStringExtra("url_models_base")
+        val zipPath = intent.getStringExtra("zip_path")
         val generateGoldens = intent.getBooleanExtra("generate_goldens", false)
+        val autoRun = intent.getBooleanExtra("auto_run", false)
+        val autoExport = intent.getBooleanExtra("auto_export", false)
+        val autoExportResults = intent.getBooleanExtra("auto_export_results", false)
         val outputPath = intent.getStringExtra("output_path")
 
-        val outputDir = if (outputPath != null) {
-            File(outputPath).apply { mkdirs() }
+        Log.i(TAG, "Resolving config with outputPath: $outputPath")
+
+        val baseDir = context.getExternalFilesDir(null) ?: context.filesDir
+        Log.i(TAG, "Base directory for resolution: ${baseDir.absolutePath}")
+        
+        val outputDir = if (!outputPath.isNullOrBlank()) {
+            val file = File(outputPath)
+            val resolved = if (file.isAbsolute) {
+                file
+            } else {
+                File(baseDir, outputPath)
+            }
+            
+            // Critical check: if resolved is root or very short, it's likely wrong
+            if (resolved.absolutePath == "/" || resolved.parent == null) {
+                Log.w(TAG, "Resolved outputDir is root ($resolved), defaulting to app-specific dir")
+                File(baseDir, "validation_results")
+            } else {
+                resolved
+            }
         } else {
-            File(context.getExternalFilesDir(null), "validation_results").apply { mkdirs() }
+            File(baseDir, "validation_results")
+        }
+        
+        if (!outputDir.exists() && !outputDir.mkdirs()) {
+            Log.e(TAG, "Failed to create outputDir: ${outputDir.absolutePath}")
+        }
+        Log.i(TAG, "Final outputDir: ${outputDir.absolutePath}")
+
+        val sourceZipFile = if (zipPath != null) {
+            val file = File(zipPath)
+            if (file.isAbsolute) {
+                file
+            } else {
+                File(baseDir, zipPath)
+            }
+        } else {
+            null
         }
 
         val config = when {
+            sourceZipFile != null && sourceZipFile.exists() -> loadFromZip(sourceZipFile)
             urlConfig != null -> downloadConfig(urlConfig, urlModelsBase)
             testConfigPath != null -> ConfigParser.parseFromPath(testConfigPath)
             else -> extractDefaultAssets()
         }
 
-        return@withContext ValidationInput(config, outputDir, generateGoldens)
+        return@withContext ValidationInput(config, outputDir, generateGoldens, autoRun, autoExport, autoExportResults, sourceZipFile)
+    }
+
+    suspend fun loadFromZip(zipFile: File): RenderTestConfig {
+        Log.i(TAG, "Unzipping validation bundle: ${zipFile.absolutePath}")
+        // Use a unique cache dir based on timestamp or just overwrite a common one
+        // Overwriting is safer to avoid filling up disk, but we must ensure we don't conflict with current run
+        val baseCacheDir = context.externalCacheDir ?: context.cacheDir
+        val cacheDir = File(baseCacheDir, "unzipped_validation")
+        if (cacheDir.exists()) cacheDir.deleteRecursively()
+        cacheDir.mkdirs()
+
+        Log.i(TAG, "Using cacheDir: ${cacheDir.absolutePath}")
+
+        ZipFile(zipFile).use { zip ->
+            val entries = zip.entries()
+            while (entries.hasMoreElements()) {
+                val entry = entries.nextElement()
+                val entryFile = File(cacheDir, entry.name)
+                // specific check to avoid zip slip vulnerability (though low risk here)
+                if (!entryFile.canonicalPath.startsWith(cacheDir.canonicalPath)) {
+                    throw SecurityException("Zip entry is outside of the target dir: ${entry.name}")
+                }
+                
+                if (entry.isDirectory) {
+                    entryFile.mkdirs()
+                } else {
+                    entryFile.parentFile?.mkdirs()
+                    zip.getInputStream(entry).use { input ->
+                         FileOutputStream(entryFile).use { output ->
+                             input.copyTo(output)
+                         }
+                    }
+                }
+            }
+        }
+        
+        // Find config.json
+        // We look for a file ending in .json within the unzipped structure
+        // Exclude results.json if it happened to be there
+        val jsonFiles = cacheDir.walkTopDown()
+            .filter { it.isFile && it.extension == "json" && it.name != "results.json" }
+            .toList()
+            
+        if (jsonFiles.isEmpty()) throw IllegalStateException("No config.json found in zip")
+        
+        // Prefer one named config.json or take the first one
+        val configFile = jsonFiles.find { it.name == "config.json" } ?: jsonFiles.first()
+
+        Log.i(TAG, "Parsed config from ${configFile.absolutePath}")
+        return ConfigParser.parseFromPath(configFile.absolutePath)
     }
 
     private suspend fun extractDefaultAssets(): RenderTestConfig {
@@ -90,9 +184,9 @@ class ValidationInputManager(private val context: Context) {
         // Copy DamagedHelmet.glb
         val modelsDir = File(filesDir, "models")
         modelsDir.mkdirs()
-        val modelOut = File(modelsDir, "DamagedHelmet.glb")
+        val modelOut = File(modelsDir, "helmet.glb")
 
-        assetManager.open("DamagedHelmet.glb").use { input ->
+        assetManager.open("models/helmet.glb").use { input ->
             FileOutputStream(modelOut).use { output ->
                 input.copyTo(output)
             }

android/samples/sample-render-validation/src/main/java/com/google/android/filament/validation/ValidationResultManager.kt

@@ -67,22 +67,201 @@ class ValidationResultManager(private val outputDir: File) {
     fun finalizeResults(): File? {
         // Write results JSON
         writeResultsJson()
+        return null
+    }
+
+    /**
+     * Exports a zip containing:
+     * - results.json
+     * - input test bundle (as nested zip), if provided
+     * - diff images (if any failure)
+     */
+    fun exportTestResults(sourceZip: File?, timestamp: String): File? {
+        // Safe parent dir resolution
+        val parentDir = outputDir.canonicalFile.parentFile ?: outputDir.parentFile
+        if (parentDir == null) return null
+
+        val resultZipName = "results_$timestamp"
+        val zipFile = File(parentDir, "$resultZipName.zip")
+
+        Log.i(TAG, "Exporting results to ${zipFile.absolutePath}")
 
-        // Zip results
-        val zipFile = File(outputDir, "results.zip")
         try {
             ZipOutputStream(FileOutputStream(zipFile)).use { zos ->
-                outputDir.walkTopDown().filter { it.isFile && it.name != "results.zip" }.forEach { file ->
-                    val entryName = file.relativeTo(outputDir).path
-                    zos.putNextEntry(ZipEntry(entryName))
-                    file.inputStream().use { it.copyTo(zos) }
+                // 1. Add results.json
+                val resultsJson = File(outputDir, "results.json")
+                if (resultsJson.exists()) {
+                    zos.putNextEntry(ZipEntry("results.json"))
+                    resultsJson.inputStream().use { it.copyTo(zos) }
+                    zos.closeEntry()
+                }
+
+                // 2. Add source zip if exists
+                if (sourceZip != null && sourceZip.exists()) {
+                    zos.putNextEntry(ZipEntry(sourceZip.name))
+                    sourceZip.inputStream().use { it.copyTo(zos) }
+                    zos.closeEntry()
+                }
+
+                // 3. Add diff images (any file ending in _diff.png in outputDir)
+                outputDir.listFiles { _, name -> name.endsWith("_diff.png") }?.forEach { diffFile ->
+                    zos.putNextEntry(ZipEntry(diffFile.name))
+                    diffFile.inputStream().use { it.copyTo(zos) }
                     zos.closeEntry()
                 }
             }
-            Log.i(TAG, "Zipped results to ${zipFile.absolutePath}")
+            Log.i(TAG, "Exported results to ${zipFile.absolutePath}")
             return zipFile
         } catch (e: Exception) {
-            Log.e(TAG, "Failed to zip results", e)
+            Log.e(TAG, "Failed to export results", e)
+            return null
+        }
+    }
+
+    /**
+     * Exports a zip bundle containing:
+     * - Modified config.json (with updated name and relative paths)
+     * - Models (in models/ subdirectory)
+     * - Golden images (in goldens/ subdirectory)
+     *
+     * Structure:
+     *   test_name_TIMESTAMP/
+     *     config.json
+     *     models/
+     *       model.glb
+     *     goldens/
+     *       test_result.png
+     */
+    fun exportTestBundle(config: RenderTestConfig, timestamp: String): File? {
+        Log.i(TAG, "Starting exportTestBundle for ${config.name} at $timestamp")
+        Log.i(TAG, "OutputDir: ${outputDir.absolutePath}")
+
+        val parentDir = outputDir.canonicalFile.parentFile ?: outputDir.parentFile
+        if (parentDir == null) {
+            Log.e(TAG, "OutputDir parent is null: ${outputDir.absolutePath}")
+            return null
+        }
+        Log.i(TAG, "Using parentDir for export: ${parentDir.absolutePath}")
+
+        val testNameWithTimestamp = "${config.name}_$timestamp"
+        val exportNameNoSpaces = testNameWithTimestamp.replace(" ", "_")
+
+        val exportDir = File(parentDir, "export_temp_$timestamp")
+
+        Log.i(TAG, "Creating export temp dir: ${exportDir.absolutePath}")
+        if (exportDir.exists()) exportDir.deleteRecursively()
+        if (!exportDir.mkdirs()) {
+             Log.e(TAG, "Failed to create export dir: ${exportDir.absolutePath}")
+             return null
+        }
+
+        val rootDir = File(exportDir, exportNameNoSpaces)
+        rootDir.mkdirs()
+
+        val modelsDir = File(rootDir, "models")
+        modelsDir.mkdirs()
+
+        val goldensDir = File(rootDir, "goldens")
+        goldensDir.mkdirs()
+
+        try {
+            // 1. Copy Models and update config map
+            val newModelsMap = mutableMapOf<String, String>()
+            Log.i(TAG, "Copying models...")
+            for ((modelName, modelPath) in config.models) {
+                val sourceFile = File(modelPath)
+                if (sourceFile.exists()) {
+                    val destFile = File(modelsDir, sourceFile.name)
+                    Log.d(TAG, "Copying model $modelName: $modelPath -> ${destFile.name}")
+                    sourceFile.copyTo(destFile, overwrite = true)
+                    // Use relative path for the new config
+                    newModelsMap[modelName] = "models/${sourceFile.name}"
+                } else {
+                    Log.w(TAG, "Model file not found: $modelPath")
+                }
+            }
+
+            // 2. Copy Golden Images
+            // We assume goldens are in outputDir with .png extension
+            Log.i(TAG, "Copying goldens from ${outputDir.absolutePath}...")
+            outputDir.listFiles { _, name -> name.endsWith(".png") }?.forEach { file ->
+                Log.d(TAG, "Copying golden: ${file.name}")
+                file.copyTo(File(goldensDir, file.name), overwrite = true)
+            }
+
+            // 3. Create modified config JSON
+            Log.i(TAG, "Creating config.json...")
+            val newConfigJson = JSONObject()
+            newConfigJson.put("name", testNameWithTimestamp) // Keep spaces in JSON name
+
+            // Reconstruct backends
+            val backendsArray = JSONArray()
+            config.backends.forEach { backendsArray.put(it) }
+            newConfigJson.put("backends", backendsArray)
+
+            // Reconstruct models
+            val modelsJson = JSONObject()
+            for ((k, v) in newModelsMap) {
+                modelsJson.put(k, v)
+            }
+            newConfigJson.put("models", modelsJson)
+
+            // Reconstruct tests
+            val testsArray = JSONArray()
+            for (test in config.tests) {
+                val testJson = JSONObject()
+                testJson.put("name", test.name)
+                if (test.description != null) testJson.put("description", test.description)
+
+                // Models for this test (set of strings)
+                val testModelsArray = JSONArray()
+                test.models.forEach { testModelsArray.put(it) }
+                testJson.put("models", testModelsArray)
+
+                // Rendering settings
+                testJson.put("rendering", test.rendering)
+
+                // Tolerance
+                if (test.tolerance != null) {
+                    testJson.put("tolerance", test.tolerance)
+                }
+
+                // Backends (optional override)
+                 val testBackends = JSONArray()
+                 test.backends.forEach { testBackends.put(it) }
+                 testJson.put("backends", testBackends)
+
+                testsArray.put(testJson)
+            }
+            newConfigJson.put("tests", testsArray)
+
+            // Write config.json
+            File(rootDir, "config.json").writeText(newConfigJson.toString(4))
+
+            // 4. Zip it
+            val zipFile = File(parentDir, "$exportNameNoSpaces.zip")
+            Log.i(TAG, "Zipping to ${zipFile.absolutePath}...")
+
+            ZipOutputStream(FileOutputStream(zipFile)).use { zos ->
+                rootDir.walkTopDown().forEach { file ->
+                    if (file.isFile) {
+                        val entryName = file.relativeTo(exportDir).path
+                        zos.putNextEntry(ZipEntry(entryName))
+                        file.inputStream().use { it.copyTo(zos) }
+                        zos.closeEntry()
+                    }
+                }
+            }
+
+            // Cleanup temp dir
+            exportDir.deleteRecursively()
+
+            Log.i(TAG, "Exported test bundle to ${zipFile.absolutePath}")
+            return zipFile
+
+        } catch (e: Exception) {
+            Log.e(TAG, "Failed to export test bundle", e)
+            exportDir.deleteRecursively()
             return null
         }
     }

android/samples/sample-render-validation/src/main/java/com/google/android/filament/validation/ValidationRunner.kt

@@ -41,18 +41,13 @@ class ValidationRunner(
     private var currentTestConfig: TestConfig? = null
     private var currentModelName: String? = null
 
-    private var loadStartFence: com.google.android.filament.Fence? = null
-    private var loadStartTime = 0L
     private var frameCounter = 0
 
     enum class State {
         IDLE,
-        LOADING_MODEL,
-        WAITING_FOR_FENCE,
         WAITING_FOR_RESOURCES,
         WARMUP,
-        RUNNING_TEST,
-        COMPARING
+        RUNNING_TEST
     }
 
     interface Callback {
@@ -93,8 +88,6 @@ class ValidationRunner(
             nextModel()
             return
         }
-
-        currentState = State.LOADING_MODEL
         callback?.onStatusChanged("Loading $modelName for ${currentTestConfig?.name}")
 
         // Load model on main thread (required by ModelViewer)
@@ -110,51 +103,24 @@ class ValidationRunner(
             Log.i("ValidationRunner", "Loading GLB buffer... (${bytes.size} bytes)")
             val buffer = ByteBuffer.wrap(bytes)
             modelViewer.loadModelGlb(buffer)
-            Log.i("ValidationRunner", "Model loaded. initializing fence.")
+            Log.i("ValidationRunner", "Model loaded.")
             modelViewer.transformToUnitCube()
-            loadStartFence = modelViewer.engine.createFence()
-            loadStartTime = System.nanoTime()
-            currentState = State.WAITING_FOR_FENCE
-            frameCounter = 0 // Reset for fence timeout tracking
-            Log.i("ValidationRunner", "State set to WAITING_FOR_FENCE")
+            currentState = State.WAITING_FOR_RESOURCES
+            frameCounter = 0
+            Log.i("ValidationRunner", "State set to WAITING_FOR_RESOURCES")
         } catch (e: Exception) {
-             Log.e("ValidationRunner", "Failed to load $path", e)
-             nextModel()
+            Log.e("ValidationRunner", "Failed to load $path", e)
+            nextModel()
         }
     }
 
     fun onFrame(frameTimeNanos: Long) {
         if (frameCounter % 60 == 0) {
-             Log.i("ValidationRunner", "onFrame: $currentState (frame: $frameCounter)")
+            Log.i("ValidationRunner", "onFrame: $currentState (frame: $frameCounter)")
         }
 
         when (currentState) {
             State.IDLE -> {}
-            State.WAITING_FOR_FENCE -> {
-                frameCounter++
-                if (frameCounter > 600) {
-                     Log.w("ValidationRunner", "Fence timed out after 600 frames! Forcing proceed.")
-                     modelViewer.engine.destroyFence(loadStartFence!!)
-                     loadStartFence = null
-                     currentState = State.WAITING_FOR_RESOURCES
-                     return
-                }
-
-                loadStartFence?.let { fence ->
-                     if (fence.wait(com.google.android.filament.Fence.Mode.FLUSH, 0) ==
-                            com.google.android.filament.Fence.FenceStatus.CONDITION_SATISFIED) {
-                         modelViewer.engine.destroyFence(fence)
-                         loadStartFence = null
-
-                         // Compile materials (simplified)
-                         modelViewer.scene.forEach { entity ->
-                             // ... existing material compilation logic ...
-                         }
-
-                         currentState = State.WAITING_FOR_RESOURCES
-                     }
-                }
-            }
             State.WAITING_FOR_RESOURCES -> {
                  val progress = modelViewer.progress
                  if (progress >= 1.0f) {
@@ -166,12 +132,12 @@ class ValidationRunner(
             State.WARMUP -> {
                 frameCounter++
                 if (frameCounter > 5) { // 5 frames warmup
-                     startAutomation()
+                    startAutomation()
                 }
             }
             State.RUNNING_TEST -> {
-                 // Log.i("ValidationRunner", "Running test...")
-                 currentEngine?.let { engine ->
+                // Log.i("ValidationRunner", "Running test...")
+                currentEngine?.let { engine ->
                     val content = AutomationEngine.ViewerContent()
                     content.view = modelViewer.view
                     content.renderer = modelViewer.renderer
@@ -186,13 +152,11 @@ class ValidationRunner(
                     if (engine.shouldClose()) {
                         Log.i("ValidationRunner", "Finishing test (frames: $frameCounter)")
                         // Test finished (for this spec)
-                        currentState = State.COMPARING
+                        currentState = State.IDLE
                         captureAndCompare()
                     }
                 }
             }
-            State.COMPARING -> {} // Busy
-            State.LOADING_MODEL -> {}
         }
     }
 
@@ -233,7 +197,36 @@ class ValidationRunner(
 
          // Golden path
          val modelFile = File(config.models.get(modelName)!!)
-         val goldenFile = modelFile.parentFile!!.parentFile!!.resolve("golden/${testFullName}.png")
+         val modelParent = modelFile.parentFile!!
+
+         // Search for golden in:
+         // 1. ../golden/ (standard structure)
+         // 2. ../goldens/ (exported structure, sibling of models)
+         // 3. ./goldens/ (backup)
+
+         val searchPaths = mutableListOf<File>()
+         modelParent.parentFile?.let {
+             searchPaths.add(it.resolve("golden"))
+             searchPaths.add(it.resolve("goldens"))
+         }
+         searchPaths.add(modelParent.resolve("goldens"))
+
+         var goldenFile: File? = null
+         for (path in searchPaths) {
+             val f = path.resolve("${testFullName}.png")
+             if (f.exists()) {
+                 goldenFile = f
+                 break
+             }
+         }
+
+         if (goldenFile != null) {
+             Log.i("ValidationRunner", "Found golden at ${goldenFile.absolutePath}")
+         } else {
+             Log.w("ValidationRunner", "Golden not found for $testFullName. Searched in: ${searchPaths.joinToString { it.absolutePath }}")
+             // Fallback to old behavior for reference if everything else fails
+             goldenFile = modelParent.parentFile?.resolve("golden/${testFullName}.png") ?: File("nonexistent")
+         }
 
          Thread {
              try {
@@ -245,14 +238,15 @@ class ValidationRunner(
                 var diffMetric = 0f
 
                 if (generateGoldens) {
-                    goldenFile.parentFile?.mkdirs()
-                    FileOutputStream(goldenFile).use { out ->
+                    val targetGolden = goldenFile ?: modelParent.parentFile?.resolve("golden/${testFullName}.png") ?: File(modelParent, "golden/${testFullName}.png")
+                    targetGolden.parentFile?.mkdirs()
+                    FileOutputStream(targetGolden).use { out ->
                         flipped.compress(Bitmap.CompressFormat.PNG, 100, out)
                     }
                     passed = true // Generating goldens always passes if successful
                     callback?.onStatusChanged("Golden generated")
                 } else {
-                    if (goldenFile.exists()) {
+                    if (goldenFile != null && goldenFile.exists()) {
                         val golden = android.graphics.BitmapFactory.decodeFile(goldenFile.absolutePath)
                         if (golden != null) {
                             callback?.onImageResult("Golden", golden)
@@ -273,7 +267,7 @@ class ValidationRunner(
                             callback?.onStatusChanged("Failed to load golden")
                         }
                     } else {
-                        Log.w("ValidationRunner", "Golden not found: ${goldenFile.absolutePath}")
+                        Log.w("ValidationRunner", "Golden not found: ${goldenFile?.absolutePath}")
                         callback?.onStatusChanged("Golden not found")
                     }
                 }

android/samples/sample-render-validation/src/main/res/layout/activity_main.xml

@@ -1,5 +1,5 @@
 <?xml version="1.0" encoding="utf-8"?>
-<androidx.constraintlayout.widget.ConstraintLayout 
+<androidx.constraintlayout.widget.ConstraintLayout
     xmlns:android="http://schemas.android.com/apk/res/android"
     xmlns:app="http://schemas.android.com/apk/res-auto"
     android:layout_width="match_parent"
@@ -10,6 +10,8 @@
         android:layout_width="0dp"
         android:layout_height="0dp"
         app:layout_constraintDimensionRatio="1:1"
+        app:layout_constraintWidth_percent="0.6"
+        android:layout_marginTop="32dp"
         app:layout_constraintTop_toTopOf="parent"
         app:layout_constraintStart_toStartOf="parent"
         app:layout_constraintEnd_toEndOf="parent">
@@ -20,10 +22,83 @@
             android:layout_height="match_parent" />
     </FrameLayout>
 
+    <LinearLayout
+        android:id="@+id/controls_container"
+        android:layout_width="0dp"
+        android:layout_height="wrap_content"
+        android:orientation="vertical"
+        android:paddingStart="20dp"
+        android:paddingEnd="20dp"
+        android:paddingTop="20dp"
+        android:paddingBottom="0dp"
+        app:layout_constraintTop_toBottomOf="@id/surface_container"
+        app:layout_constraintStart_toStartOf="parent"
+        app:layout_constraintEnd_toEndOf="parent">
+
+        <TextView
+            android:id="@+id/status_text"
+            android:layout_width="match_parent"
+            android:layout_height="wrap_content"
+            android:text="Initializing..."
+            android:textSize="12sp" />
+
+        <LinearLayout
+            android:layout_width="match_parent"
+            android:layout_height="wrap_content"
+            android:orientation="vertical"
+            android:paddingTop="10dp"
+            android:paddingBottom="10dp">
+
+            <LinearLayout
+                android:layout_width="match_parent"
+                android:layout_height="wrap_content"
+                android:orientation="horizontal"
+                android:gravity="center_vertical"
+                android:paddingBottom="8dp">
+
+                <com.google.android.material.button.MaterialButton
+                    android:id="@+id/run_button"
+                    android:layout_width="0dp"
+                    android:layout_height="wrap_content"
+                    android:layout_weight="1"
+                    style="@style/Widget.Material3.Button"
+                    android:text="Run Test" />
+
+                <com.google.android.material.button.MaterialButton
+                    android:id="@+id/load_button"
+                    android:layout_width="wrap_content"
+                    android:layout_height="wrap_content"
+                    style="@style/Widget.Material3.Button.TonalButton"
+                    android:text="Load Test"
+                    android:layout_marginStart="8dp"/>
+
+                <com.google.android.material.button.MaterialButton
+                    android:id="@+id/options_button"
+                    android:layout_width="wrap_content"
+                    android:layout_height="wrap_content"
+                    style="@style/Widget.Material3.Button.IconButton.Outlined"
+                    app:icon="@android:drawable/ic_menu_more"
+                    android:contentDescription="More Options"
+                    android:layout_marginStart="8dp"/>
+            </LinearLayout>
+        </LinearLayout>
+
+        <TextView
+            android:id="@+id/test_results_header"
+            android:layout_width="match_parent"
+            android:layout_height="wrap_content"
+            android:text="Test Results"
+            android:gravity="center"
+            android:textSize="14sp"
+            android:paddingTop="0dp"
+            android:paddingBottom="8dp" />
+
+    </LinearLayout>
+
     <ScrollView
         android:layout_width="0dp"
         android:layout_height="0dp"
-        app:layout_constraintTop_toBottomOf="@id/surface_container"
+        app:layout_constraintTop_toBottomOf="@id/controls_container"
         app:layout_constraintBottom_toBottomOf="parent"
         app:layout_constraintStart_toStartOf="parent"
         app:layout_constraintEnd_toEndOf="parent">
@@ -32,41 +107,9 @@
             android:layout_width="match_parent"
             android:layout_height="wrap_content"
             android:orientation="vertical"
-            android:padding="20dp">
-
-            <TextView
-                android:id="@+id/status_text"
-                android:layout_width="match_parent"
-                android:layout_height="wrap_content"
-                android:text="Initializing..."
-                android:textSize="14sp" />
-
-            <LinearLayout
-                android:layout_width="match_parent"
-                android:layout_height="wrap_content"
-                android:orientation="horizontal"
-                android:paddingTop="10dp"
-                android:paddingBottom="10dp">
-
-                <Spinner
-                    android:id="@+id/mode_spinner"
-                    android:layout_width="wrap_content"
-                    android:layout_height="wrap_content" />
-
-                <Button
-                    android:id="@+id/run_button"
-                    android:layout_width="wrap_content"
-                    android:layout_height="wrap_content"
-                    android:text="Run" />
-            </LinearLayout>
-
-            <TextView
-                android:layout_width="match_parent"
-                android:layout_height="wrap_content"
-                android:text="Test Results"
-                android:textSize="18sp"
-                android:paddingTop="20dp"
-                android:paddingBottom="10dp" />
+            android:paddingStart="20dp"
+            android:paddingEnd="20dp"
+            android:paddingBottom="20dp">
 
             <LinearLayout
                 android:id="@+id/results_container"
@@ -74,7 +117,6 @@
                 android:layout_height="wrap_content"
                 android:orientation="vertical" />
 
-
         </LinearLayout>
     </ScrollView>
 

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

@@ -235,9 +235,9 @@
       "license": "MIT"
     },
     "node_modules/minimatch": {
-      "version": "3.1.2",
-      "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-3.1.2.tgz",
-      "integrity": "sha512-J7p63hRiAjw1NDEww1W7i37+ByIrOWO5XQQAzZ3VOcL0PNybwpfmV/N05zFAzwQ9USyEcX6t3UO+K5aqBQOIHw==",
+      "version": "3.1.3",
+      "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-3.1.3.tgz",
+      "integrity": "sha512-M2GCs7Vk83NxkUyQV1bkABc4yxgz9kILhHImZiBPAZ9ybuvCb0/H7lEl5XvIg3g+9d4eNotkZA5IWwYl0tibaA==",
       "dependencies": {
         "brace-expansion": "^1.1.7"
       },
@@ -429,9 +429,9 @@
       "integrity": "sha512-nGqtvLrj5w0naR6tDPfB4cUmYCqouzyQiz6C5y/LtcDllJdrcc6WaWW6iXyIIOErTa/XRybj28aasdn4LkVk6Q=="
     },
     "minimatch": {
-      "version": "3.1.2",
-      "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-3.1.2.tgz",
-      "integrity": "sha512-J7p63hRiAjw1NDEww1W7i37+ByIrOWO5XQQAzZ3VOcL0PNybwpfmV/N05zFAzwQ9USyEcX6t3UO+K5aqBQOIHw==",
+      "version": "3.1.3",
+      "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-3.1.3.tgz",
+      "integrity": "sha512-M2GCs7Vk83NxkUyQV1bkABc4yxgz9kILhHImZiBPAZ9ybuvCb0/H7lEl5XvIg3g+9d4eNotkZA5IWwYl0tibaA==",
       "requires": {
         "brace-expansion": "^1.1.7"
       }

docs/dup/intro.html

@@ -181,7 +181,7 @@ important for <code>matc</code> (material compiler).</p>
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.69.2'
+    implementation 'com.google.android.filament:filament-android:1.69.5'
 }
 </code></pre>
 <p>Here are all the libraries available in the group <code>com.google.android.filament</code>:</p>
@@ -195,7 +195,7 @@ dependencies {
 </div>
 <h3 id="ios"><a class="header" href="#ios">iOS</a></h3>
 <p>iOS projects can use CocoaPods to install the latest release:</p>
-<pre><code class="language-shell">pod 'Filament', '~&gt; 1.69.2'
+<pre><code class="language-shell">pod 'Filament', '~&gt; 1.69.5'
 </code></pre>
 <h2 id="documentation"><a class="header" href="#documentation">Documentation</a></h2>
 <ul>

docs/dup/test_ci_renderdiff.html

@@ -225,9 +225,10 @@ into <strong>branch</strong> of <code>filament-assets</code>. This branch is pai
 <code>filament</code> repo.</p>
 <p>As an example, imagine I am working on a PR, and I've uploaded my change, which is in a
 branch called <code>my-pr-branch</code>, to <code>filament</code>. This PR requires updating the golden. We would do
-it in the following fashion</p>
+it in the following fashion on a macOS machine:</p>
 <h3 id="using-a-script-to-update-the-golden-repo"><a class="header" href="#using-a-script-to-update-the-golden-repo">Using a script to update the golden repo</a></h3>
 <ul>
+<li>Make sure you've completed the steps in 'Setting up python'</li>
 <li>Run interactive mode in the <code>update_golden.py</code> script.
 <pre><code>python3 test/renderdiff/src/update_golden.py
 </code></pre>
@@ -263,6 +264,27 @@ branch of the golden repo (i.e. <code>my-pr-branch-golden</code>).</li>
 <li>If the PR is merged, then there is another workflow that will merge <code>my-pr-branch-golden</code>
 to the <code>main</code> branch of the golden repo.</li>
 </ul>
+<h3 id="automated-update-via-commit-message"><a class="header" href="#automated-update-via-commit-message">Automated update via commit message</a></h3>
+<p>Alternatively, if you are confident in your changes and want the CI to handle the update
+for you, you can use the following tag in your commit message:</p>
+<ul>
+<li>In the commit message of your working branch on <code>filament</code>, add the following line:
+<pre><code>RDIFF_ACCEPT_NEW_GOLDENS
+</code></pre>
+</li>
+</ul>
+<p>This has the following effects:</p>
+<ul>
+<li>The presubmit test <code>test-renderdiff</code> will be bypassed (it will not perform rendering or
+comparison).</li>
+<li>When the PR is merged, the postsubmit CI will automatically:
+<ol>
+<li>Build Filament and generate the new images.</li>
+<li>Upload them to a temporary branch in <code>filament-assets</code>.</li>
+<li>Merge that branch into <code>main</code>.</li>
+</ol>
+</li>
+</ul>
 <h2 id="viewing-test-results"><a class="header" href="#viewing-test-results">Viewing test results</a></h2>
 <p>We provide a viewer for looking at the result of a test run. The viewer is a webapp that can
 be used by pointing your browser to a localhost port. If you input the viewer with a PR or a

docs/main/filament.html

@@ -504,9 +504,9 @@ D_{GGX}(h,\alpha) = \frac{\aa}{\pi ( (\NoH)^2 (\aa - 1) + 1)^2}
 \end{equation}$$
 </p><p>
 The GLSL implementation of the NDF, shown in <a href="#listing_speculard">listing&nbsp;1</a>, is simple and efficient.
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-built_in">float</span> <span class="hljs-title">D_GGX</span>(<span class="hljs-params"><span class="hljs-built_in">float</span> NoH, <span class="hljs-built_in">float</span> roughness</span>)</span> {</span>
-<span class="line">    <span class="hljs-built_in">float</span> a = NoH * roughness;</span>
-<span class="line">    <span class="hljs-built_in">float</span> k = roughness / (<span class="hljs-number">1.0</span> - NoH * NoH + a * a);</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-type">float</span> <span class="hljs-title">D_GGX</span><span class="hljs-params">(<span class="hljs-type">float</span> NoH, <span class="hljs-type">float</span> roughness)</span> </span>{</span>
+<span class="line">    <span class="hljs-type">float</span> a = NoH * roughness;</span>
+<span class="line">    <span class="hljs-type">float</span> k = roughness / (<span class="hljs-number">1.0</span> - NoH * NoH + a * a);</span>
 <span class="line">    <span class="hljs-keyword">return</span> k * k * (<span class="hljs-number">1.0</span> / PI);</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_speculard">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;1:</b> Implementation of the specular D term in GLSL</div></center>
 <p>
@@ -590,10 +590,10 @@ V(v,l,\alpha) = \frac{0.5}{\NoL \sqrt{(\NoV)^2 (1 - \aa) + \aa} + \NoV \sqrt{(\N
 \end{equation}$$
 </p><p>
 The GLSL implementation of the visibility term, shown in <a href="#listing_specularv">listing&nbsp;3</a>, is a bit more expensive than we would like since it requires two <code>sqrt</code> operations.
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-keyword">float</span> <span class="hljs-title">V_SmithGGXCorrelated</span><span class="hljs-params">(<span class="hljs-keyword">float</span> NoV, <span class="hljs-keyword">float</span> NoL, <span class="hljs-keyword">float</span> roughness)</span> </span>{</span>
-<span class="line">    <span class="hljs-keyword">float</span> a2 = roughness * roughness;</span>
-<span class="line">    <span class="hljs-keyword">float</span> GGXV = NoL * <span class="hljs-built_in">sqrt</span>(NoV * NoV * (<span class="hljs-number">1.0</span> - a2) + a2);</span>
-<span class="line">    <span class="hljs-keyword">float</span> GGXL = NoV * <span class="hljs-built_in">sqrt</span>(NoL * NoL * (<span class="hljs-number">1.0</span> - a2) + a2);</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-type">float</span> <span class="hljs-title">V_SmithGGXCorrelated</span><span class="hljs-params">(<span class="hljs-type">float</span> NoV, <span class="hljs-type">float</span> NoL, <span class="hljs-type">float</span> roughness)</span> </span>{</span>
+<span class="line">    <span class="hljs-type">float</span> a2 = roughness * roughness;</span>
+<span class="line">    <span class="hljs-type">float</span> GGXV = NoL * <span class="hljs-built_in">sqrt</span>(NoV * NoV * (<span class="hljs-number">1.0</span> - a2) + a2);</span>
+<span class="line">    <span class="hljs-type">float</span> GGXL = NoV * <span class="hljs-built_in">sqrt</span>(NoL * NoL * (<span class="hljs-number">1.0</span> - a2) + a2);</span>
 <span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-number">0.5</span> / (GGXV + GGXL);</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_specularv">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;3:</b> Implementation of the specular V term in GLSL</div></center>
 <p>
@@ -604,10 +604,10 @@ V(v,l,\alpha) = \frac{0.5}{\NoL (\NoV (1 - \alpha) + \alpha) + \NoV (\NoL (1 - \
 \end{equation}$$
 </p><p>
 This approximation is mathematically wrong but saves two square root operations and is good enough for real-time mobile applications, as shown in <a href="#listing_approximatedspecularv">listing&nbsp;4</a>.
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-built_in">float</span> <span class="hljs-title">V_SmithGGXCorrelatedFast</span>(<span class="hljs-params"><span class="hljs-built_in">float</span> NoV, <span class="hljs-built_in">float</span> NoL, <span class="hljs-built_in">float</span> roughness</span>)</span> {</span>
-<span class="line">    <span class="hljs-built_in">float</span> a = roughness;</span>
-<span class="line">    <span class="hljs-built_in">float</span> GGXV = NoL * (NoV * (<span class="hljs-number">1.0</span> - a) + a);</span>
-<span class="line">    <span class="hljs-built_in">float</span> GGXL = NoV * (NoL * (<span class="hljs-number">1.0</span> - a) + a);</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-type">float</span> <span class="hljs-title">V_SmithGGXCorrelatedFast</span><span class="hljs-params">(<span class="hljs-type">float</span> NoV, <span class="hljs-type">float</span> NoL, <span class="hljs-type">float</span> roughness)</span> </span>{</span>
+<span class="line">    <span class="hljs-type">float</span> a = roughness;</span>
+<span class="line">    <span class="hljs-type">float</span> GGXV = NoL * (NoV * (<span class="hljs-number">1.0</span> - a) + a);</span>
+<span class="line">    <span class="hljs-type">float</span> GGXL = NoV * (NoL * (<span class="hljs-number">1.0</span> - a) + a);</span>
 <span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-number">0.5</span> / (GGXV + GGXL);</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_approximatedspecularv">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;4:</b> Implementation of the approximated specular V term in GLSL</div></center>
 <p>
@@ -659,7 +659,7 @@ $$\begin{equation}
 \end{equation}$$
 </p><p>
 In practice, the diffuse reflectance \(\sigma\) is multiplied later, as shown in <a href="#listing_diffusebrdf">listing&nbsp;8</a>.
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-built_in">float</span> <span class="hljs-title">Fd_Lambert</span>(<span class="hljs-params"></span>)</span> {</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-built_in">float</span> <span class="hljs-title">Fd_Lambert</span>()</span> {</span>
 <span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-number">1.0</span> / PI;</span>
 <span class="line">}</span>
 <span class="line"></span>
@@ -680,14 +680,14 @@ Where:
 $$\begin{equation}
 \fGrazing=0.5 + 2 \cdot \alpha cos^2(\theta_d)
 \end{equation}$$
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-keyword">float</span> <span class="hljs-title">F_Schlick</span><span class="hljs-params">(<span class="hljs-keyword">float</span> u, <span class="hljs-keyword">float</span> f0, <span class="hljs-keyword">float</span> f90)</span> </span>{</span>
-<span class="line">    <span class="hljs-keyword">return</span> f0 + (f90 - f0) * <span class="hljs-built_in">pow</span>(<span class="hljs-number">1.0</span> - u, <span class="hljs-number">5.0</span>);</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-type">float</span> <span class="hljs-title function_">F_Schlick</span><span class="hljs-params">(<span class="hljs-type">float</span> u, <span class="hljs-type">float</span> f0, <span class="hljs-type">float</span> f90)</span> {</span>
+<span class="line">    <span class="hljs-keyword">return</span> f0 + (f90 - f0) * pow(<span class="hljs-number">1.0</span> - u, <span class="hljs-number">5.0</span>);</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">float</span> <span class="hljs-title">Fd_Burley</span><span class="hljs-params">(<span class="hljs-keyword">float</span> NoV, <span class="hljs-keyword">float</span> NoL, <span class="hljs-keyword">float</span> LoH, <span class="hljs-keyword">float</span> roughness)</span> </span>{</span>
-<span class="line">    <span class="hljs-keyword">float</span> f90 = <span class="hljs-number">0.5</span> + <span class="hljs-number">2.0</span> * roughness * LoH * LoH;</span>
-<span class="line">    <span class="hljs-keyword">float</span> lightScatter = F_Schlick(NoL, <span class="hljs-number">1.0</span>, f90);</span>
-<span class="line">    <span class="hljs-keyword">float</span> viewScatter = F_Schlick(NoV, <span class="hljs-number">1.0</span>, f90);</span>
+<span class="line"><span class="hljs-type">float</span> <span class="hljs-title function_">Fd_Burley</span><span class="hljs-params">(<span class="hljs-type">float</span> NoV, <span class="hljs-type">float</span> NoL, <span class="hljs-type">float</span> LoH, <span class="hljs-type">float</span> roughness)</span> {</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">f90</span> <span class="hljs-operator">=</span> <span class="hljs-number">0.5</span> + <span class="hljs-number">2.0</span> * roughness * LoH * LoH;</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">lightScatter</span> <span class="hljs-operator">=</span> F_Schlick(NoL, <span class="hljs-number">1.0</span>, f90);</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">viewScatter</span> <span class="hljs-operator">=</span> F_Schlick(NoV, <span class="hljs-number">1.0</span>, f90);</span>
 <span class="line">    <span class="hljs-keyword">return</span> lightScatter * viewScatter * (<span class="hljs-number">1.0</span> / PI);</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_diffusebrdf">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;8:</b> Implementation of the diffuse Disney BRDF in GLSL</div></center>
 <p>
@@ -704,47 +704,47 @@ We could allow artists/developers to choose the Disney diffuse BRDF depending on
 <strong class="asterisk">Diffuse term</strong>: a Lambertian diffuse model.
 </p><p>
 The full GLSL implementation of the standard model is shown in <a href="#listing_glslbrdf">listing&nbsp;9</a>.
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-type">float</span> D_GGX(<span class="hljs-type">float</span> NoH, <span class="hljs-type">float</span> a) {</span>
-<span class="line">    <span class="hljs-type">float</span> a2 = a * a;</span>
-<span class="line">    <span class="hljs-type">float</span> f = (NoH * a2 - NoH) * NoH + <span class="hljs-number">1.0</span>;</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-type">float</span> <span class="hljs-title function_">D_GGX</span><span class="hljs-params">(<span class="hljs-type">float</span> NoH, <span class="hljs-type">float</span> a)</span> {</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">a2</span> <span class="hljs-operator">=</span> a * a;</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">f</span> <span class="hljs-operator">=</span> (NoH * a2 - NoH) * NoH + <span class="hljs-number">1.0</span>;</span>
 <span class="line">    <span class="hljs-keyword">return</span> a2 / (PI * f * f);</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-type">vec3</span> F_Schlick(<span class="hljs-type">float</span> u, <span class="hljs-type">vec3</span> f0) {</span>
-<span class="line">    <span class="hljs-keyword">return</span> f0 + (<span class="hljs-type">vec3</span>(<span class="hljs-number">1.0</span>) - f0) * <span class="hljs-built_in">pow</span>(<span class="hljs-number">1.0</span> - u, <span class="hljs-number">5.0</span>);</span>
+<span class="line">vec3 <span class="hljs-title function_">F_Schlick</span><span class="hljs-params">(<span class="hljs-type">float</span> u, vec3 f0)</span> {</span>
+<span class="line">    <span class="hljs-keyword">return</span> f0 + (vec3(<span class="hljs-number">1.0</span>) - f0) * pow(<span class="hljs-number">1.0</span> - u, <span class="hljs-number">5.0</span>);</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-type">float</span> V_SmithGGXCorrelated(<span class="hljs-type">float</span> NoV, <span class="hljs-type">float</span> NoL, <span class="hljs-type">float</span> a) {</span>
-<span class="line">    <span class="hljs-type">float</span> a2 = a * a;</span>
-<span class="line">    <span class="hljs-type">float</span> GGXL = NoV * <span class="hljs-built_in">sqrt</span>((-NoL * a2 + NoL) * NoL + a2);</span>
-<span class="line">    <span class="hljs-type">float</span> GGXV = NoL * <span class="hljs-built_in">sqrt</span>((-NoV * a2 + NoV) * NoV + a2);</span>
+<span class="line"><span class="hljs-type">float</span> <span class="hljs-title function_">V_SmithGGXCorrelated</span><span class="hljs-params">(<span class="hljs-type">float</span> NoV, <span class="hljs-type">float</span> NoL, <span class="hljs-type">float</span> a)</span> {</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">a2</span> <span class="hljs-operator">=</span> a * a;</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">GGXL</span> <span class="hljs-operator">=</span> NoV * sqrt((-NoL * a2 + NoL) * NoL + a2);</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">GGXV</span> <span class="hljs-operator">=</span> NoL * sqrt((-NoV * a2 + NoV) * NoV + a2);</span>
 <span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-number">0.5</span> / (GGXV + GGXL);</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-type">float</span> Fd_Lambert() {</span>
+<span class="line"><span class="hljs-type">float</span> <span class="hljs-title function_">Fd_Lambert</span><span class="hljs-params">()</span> {</span>
 <span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-number">1.0</span> / PI;</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-type">void</span> BRDF(...) {</span>
-<span class="line">    <span class="hljs-type">vec3</span> h = <span class="hljs-built_in">normalize</span>(v + l);</span>
+<span class="line"><span class="hljs-keyword">void</span> <span class="hljs-title function_">BRDF</span><span class="hljs-params">(...)</span> {</span>
+<span class="line">    <span class="hljs-type">vec3</span> <span class="hljs-variable">h</span> <span class="hljs-operator">=</span> normalize(v + l);</span>
 <span class="line"></span>
-<span class="line">    <span class="hljs-type">float</span> NoV = <span class="hljs-built_in">abs</span>(<span class="hljs-built_in">dot</span>(n, v)) + <span class="hljs-number">1e-5</span>;</span>
-<span class="line">    <span class="hljs-type">float</span> NoL = <span class="hljs-built_in">clamp</span>(<span class="hljs-built_in">dot</span>(n, l), <span class="hljs-number">0.0</span>, <span class="hljs-number">1.0</span>);</span>
-<span class="line">    <span class="hljs-type">float</span> NoH = <span class="hljs-built_in">clamp</span>(<span class="hljs-built_in">dot</span>(n, h), <span class="hljs-number">0.0</span>, <span class="hljs-number">1.0</span>);</span>
-<span class="line">    <span class="hljs-type">float</span> LoH = <span class="hljs-built_in">clamp</span>(<span class="hljs-built_in">dot</span>(l, h), <span class="hljs-number">0.0</span>, <span class="hljs-number">1.0</span>);</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">NoV</span> <span class="hljs-operator">=</span> abs(dot(n, v)) + <span class="hljs-number">1e-5</span>;</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">NoL</span> <span class="hljs-operator">=</span> clamp(dot(n, l), <span class="hljs-number">0.0</span>, <span class="hljs-number">1.0</span>);</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">NoH</span> <span class="hljs-operator">=</span> clamp(dot(n, h), <span class="hljs-number">0.0</span>, <span class="hljs-number">1.0</span>);</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">LoH</span> <span class="hljs-operator">=</span> clamp(dot(l, h), <span class="hljs-number">0.0</span>, <span class="hljs-number">1.0</span>);</span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-comment">// perceptually linear roughness to roughness (see parameterization)</span></span>
-<span class="line">    <span class="hljs-type">float</span> roughness = perceptualRoughness * perceptualRoughness;</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">roughness</span> <span class="hljs-operator">=</span> perceptualRoughness * perceptualRoughness;</span>
 <span class="line"></span>
-<span class="line">    <span class="hljs-type">float</span> D = D_GGX(NoH, roughness);</span>
-<span class="line">    <span class="hljs-type">vec3</span>  F = F_Schlick(LoH, f0);</span>
-<span class="line">    <span class="hljs-type">float</span> V = V_SmithGGXCorrelated(NoV, NoL, roughness);</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">D</span> <span class="hljs-operator">=</span> D_GGX(NoH, roughness);</span>
+<span class="line">    <span class="hljs-type">vec3</span>  <span class="hljs-variable">F</span> <span class="hljs-operator">=</span> F_Schlick(LoH, f0);</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">V</span> <span class="hljs-operator">=</span> V_SmithGGXCorrelated(NoV, NoL, roughness);</span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-comment">// specular BRDF</span></span>
-<span class="line">    <span class="hljs-type">vec3</span> Fr = (D * V) * F;</span>
+<span class="line">    <span class="hljs-type">vec3</span> <span class="hljs-variable">Fr</span> <span class="hljs-operator">=</span> (D * V) * F;</span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-comment">// diffuse BRDF</span></span>
-<span class="line">    <span class="hljs-type">vec3</span> Fd = diffuseColor * Fd_Lambert();</span>
+<span class="line">    <span class="hljs-type">vec3</span> <span class="hljs-variable">Fd</span> <span class="hljs-operator">=</span> diffuseColor * Fd_Lambert();</span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-comment">// apply lighting...</span></span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_glslbrdf">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;9:</b> Evaluation of the BRDF in GLSL</div></center>
@@ -965,7 +965,7 @@ $$\begin{equation}
 \end{equation}$$
 </p><p>
 <a href="#listing_fnormal">Listing&nbsp;12</a> shows how \(\fNormal\) is computed for both dielectric and metallic materials. It shows that the color of the specular reflectance is derived from the base color in the metallic case.
-</p><pre class="listing tilde"><code><span class="line">vec3 f0 = 0.16 <span class="hljs-emphasis">* reflectance *</span> reflectance <span class="hljs-emphasis">* (1.0 - metallic) + baseColor *</span> metallic;</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_fnormal">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;12:</b> Computing \(\fNormal\) for dielectric and metallic materials in GLSL</div></center>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-symbol">vec3</span> <span class="hljs-built_in">f0</span> = <span class="hljs-number">0</span>.<span class="hljs-number">16</span> * reflectance * reflectance * (<span class="hljs-number">1</span>.<span class="hljs-number">0</span> - metallic) + baseColor * metallic<span class="hljs-comment">;</span></span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_fnormal">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;12:</b> Computing \(\fNormal\) for dielectric and metallic materials in GLSL</div></center>
 <a class="target" name="roughnessremappingandclamping">&nbsp;</a><a class="target" name="materialsystem/parameterization/remapping/roughnessremappingandclamping">&nbsp;</a><a class="target" name="toc4.8.3.3">&nbsp;</a><h4 id="roughness-remapping-and-clamping"><a class="header" href="#roughness-remapping-and-clamping">Roughness remapping and clamping</a></h4>
 <p>
 <p>The roughness set by the user, called <code>perceptualRoughness</code> here, is remapped to a perceptually linear range using the following formulation:</p>
@@ -1054,7 +1054,7 @@ V(l,h) = \frac{1}{4(\LoH)^2}
 This masking-shadowing function is not physically based, as shown in [<a href="#citation-heitz14">Heitz14</a>], but its simplicity makes it desirable for real-time rendering.
 </p><p>
 In summary, our clear coat BRDF is a Cook-Torrance specular microfacet model, with a GGX normal distribution function, a Kelemen visibility function, and a Schlick Fresnel function. <a href="#listing_kelemen">Listing&nbsp;13</a> shows how trivial the GLSL implementation is.
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-built_in">float</span> <span class="hljs-title">V_Kelemen</span>(<span class="hljs-params"><span class="hljs-built_in">float</span> LoH</span>)</span> {</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-type">float</span> <span class="hljs-title">V_Kelemen</span><span class="hljs-params">(<span class="hljs-type">float</span> LoH)</span> </span>{</span>
 <span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-number">0.25</span> / (LoH * LoH);</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_kelemen">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;13:</b> Implementation of the Kelemen visibility term in GLSL</div></center>
 <p>
@@ -1097,18 +1097,18 @@ The clear coat roughness parameter is remapped and clamped in a similar way to t
 <center><div class="image" style=""><a href="../images/material_clear_coat2.png" target="_blank"><img class="markdeep" src="../images/material_clear_coat2.png" /></a><center><span class="imagecaption"><a class="target" name="figure_clearcoatroughness">&nbsp;</a><b style="font-style:normal;">Figure&nbsp;26:</b> Clear coat roughness varying from 0.0 (left) to 1.0 (right) with metallic set to 1.0, roughness to 0.8 and clear coat to 1.0</span></center></div></center>
 </p><p>
 <a href="#listing_clearcoatbrdf">Listing&nbsp;14</a> shows the GLSL implementation of the clear coat material model after remapping, parameterization and integration in the standard surface response.
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-keyword">void</span> <span class="hljs-title">BRDF</span>(<span class="hljs-params">...</span>)</span> {</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-type">void</span> <span class="hljs-title">BRDF</span><span class="hljs-params">(...)</span> </span>{</span>
 <span class="line">    <span class="hljs-comment">// compute Fd and Fr from standard model</span></span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-comment">// remapping and linearization of clear coat roughness</span></span>
-<span class="line">    clearCoatPerceptualRoughness = clamp(clearCoatPerceptualRoughness, <span class="hljs-number">0.089</span>, <span class="hljs-number">1.0</span>);</span>
+<span class="line">    clearCoatPerceptualRoughness = <span class="hljs-built_in">clamp</span>(clearCoatPerceptualRoughness, <span class="hljs-number">0.089</span>, <span class="hljs-number">1.0</span>);</span>
 <span class="line">    clearCoatRoughness = clearCoatPerceptualRoughness * clearCoatPerceptualRoughness;</span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-comment">// clear coat BRDF</span></span>
-<span class="line">    <span class="hljs-built_in">float</span>  Dc = D_GGX(clearCoatRoughness, NoH);</span>
-<span class="line">    <span class="hljs-built_in">float</span>  Vc = V_Kelemen(clearCoatRoughness, LoH);</span>
-<span class="line">    <span class="hljs-built_in">float</span>  Fc = F_Schlick(<span class="hljs-number">0.04</span>, LoH) * clearCoat; <span class="hljs-comment">// clear coat strength</span></span>
-<span class="line">    <span class="hljs-built_in">float</span> Frc = (Dc * Vc) * Fc;</span>
+<span class="line">    <span class="hljs-type">float</span>  Dc = <span class="hljs-built_in">D_GGX</span>(clearCoatRoughness, NoH);</span>
+<span class="line">    <span class="hljs-type">float</span>  Vc = <span class="hljs-built_in">V_Kelemen</span>(clearCoatRoughness, LoH);</span>
+<span class="line">    <span class="hljs-type">float</span>  Fc = <span class="hljs-built_in">F_Schlick</span>(<span class="hljs-number">0.04</span>, LoH) * clearCoat; <span class="hljs-comment">// clear coat strength</span></span>
+<span class="line">    <span class="hljs-type">float</span> Frc = (Dc * Vc) * Fc;</span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-comment">// account for energy loss in the base layer</span></span>
 <span class="line">    <span class="hljs-keyword">return</span> color * ((Fd + Fr * (<span class="hljs-number">1.0</span> - Fc)) * (<span class="hljs-number">1.0</span> - Fc) + Frc);</span>
@@ -1294,14 +1294,14 @@ f_{r}(v,h,\alpha) = \frac{D_{velvet}(v,h,\alpha)}{4(\NoL + \NoV - (\NoL)(\NoV))}
 \end{equation}$$
 </p><p>
 The implementation of the velvet NDF is presented in <a href="#listing_clothbrdf">listing&nbsp;17</a>, optimized to properly fit in half float formats and to avoid computing a costly cotangent, relying instead on trigonometric identities. Note that we removed the Fresnel component from this BRDF.
-</p><pre class="listing tilde"><code><span class="line">float D_Ashikhmin(float roughness, float NoH) {</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-type">float</span> <span class="hljs-title">D_Ashikhmin</span><span class="hljs-params">(<span class="hljs-type">float</span> roughness, <span class="hljs-type">float</span> NoH)</span> </span>{</span>
 <span class="line">    <span class="hljs-comment">// Ashikhmin 2007, &quot;Distribution-based BRDFs&quot;</span></span>
-<span class="line"> float a<span class="hljs-number">2</span> = roughness * roughness;</span>
-<span class="line"> float <span class="hljs-keyword">cos</span><span class="hljs-number">2</span>h = NoH * NoH;</span>
-<span class="line"> float <span class="hljs-keyword">sin</span><span class="hljs-number">2</span>h = <span class="hljs-keyword">max</span>(<span class="hljs-number">1.0</span> - <span class="hljs-keyword">cos</span><span class="hljs-number">2</span>h, <span class="hljs-number">0.0078125</span>); <span class="hljs-comment">// 2^(-14/2), so sin2h^2 &gt; 0 in fp16</span></span>
-<span class="line"> float <span class="hljs-keyword">sin</span><span class="hljs-number">4</span>h = <span class="hljs-keyword">sin</span><span class="hljs-number">2</span>h * <span class="hljs-keyword">sin</span><span class="hljs-number">2</span>h;</span>
-<span class="line"> float cot<span class="hljs-number">2</span> = -<span class="hljs-keyword">cos</span><span class="hljs-number">2</span>h / (a<span class="hljs-number">2</span> * <span class="hljs-keyword">sin</span><span class="hljs-number">2</span>h);</span>
-<span class="line"> <span class="hljs-keyword">return</span> <span class="hljs-number">1.0</span> / (PI * (<span class="hljs-number">4.0</span> * a<span class="hljs-number">2</span> + <span class="hljs-number">1.0</span>) * <span class="hljs-keyword">sin</span><span class="hljs-number">4</span>h) * (<span class="hljs-number">4.0</span> * exp(cot<span class="hljs-number">2</span>) + <span class="hljs-keyword">sin</span><span class="hljs-number">4</span>h);</span>
+<span class="line"> <span class="hljs-type">float</span> a2 = roughness * roughness;</span>
+<span class="line"> <span class="hljs-type">float</span> cos2h = NoH * NoH;</span>
+<span class="line"> <span class="hljs-type">float</span> sin2h = <span class="hljs-built_in">max</span>(<span class="hljs-number">1.0</span> - cos2h, <span class="hljs-number">0.0078125</span>); <span class="hljs-comment">// 2^(-14/2), so sin2h^2 &gt; 0 in fp16</span></span>
+<span class="line"> <span class="hljs-type">float</span> sin4h = sin2h * sin2h;</span>
+<span class="line"> <span class="hljs-type">float</span> cot2 = -cos2h / (a2 * sin2h);</span>
+<span class="line"> <span class="hljs-keyword">return</span> <span class="hljs-number">1.0</span> / (PI * (<span class="hljs-number">4.0</span> * a2 + <span class="hljs-number">1.0</span>) * sin4h) * (<span class="hljs-number">4.0</span> * <span class="hljs-built_in">exp</span>(cot2) + sin4h);</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_clothbrdf">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;17:</b> Implementation of Ashikhmin's velvet NDF in GLSL</div></center>
 <p>
 <p>In [<a href="#citation-estevez17">Estevez17</a>] Estevez and Kulla propose a different NDF (called the “Charlie” sheen) that is based on an exponentiated sinusoidal instead of an inverted Gaussian. This NDF is appealing for several reasons: its parameterization feels more natural and intuitive, it provides a softer appearance and, as shown in equation (\ref{charlieNDF}), its implementation is simpler:</p>
@@ -1312,11 +1312,11 @@ D(m) = \frac{(2 + \frac{1}{\alpha}) sin(\theta)^{\frac{1}{\alpha}}}{2 \pi}
 </p><p>
 [<a href="#citation-estevez17">Estevez17</a>] also presents a new shadowing term that we omit here because of its cost. We instead rely on the visibility term from [<a href="#citation-neubelt13">Neubelt13</a>] (shown in equation \(\ref{clothSpecularBRDF}\) above).
 The implementation of this NDF is presented in <a href="#listing_clothcharliebrdf">listing&nbsp;18</a>, optimized to properly fit in half float formats.
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-keyword">float</span> <span class="hljs-title">D_Charlie</span><span class="hljs-params">(<span class="hljs-keyword">float</span> roughness, <span class="hljs-keyword">float</span> NoH)</span> </span>{</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-type">float</span> <span class="hljs-title">D_Charlie</span><span class="hljs-params">(<span class="hljs-type">float</span> roughness, <span class="hljs-type">float</span> NoH)</span> </span>{</span>
 <span class="line">    <span class="hljs-comment">// Estevez and Kulla 2017, &quot;Production Friendly Microfacet Sheen BRDF&quot;</span></span>
-<span class="line">    <span class="hljs-keyword">float</span> invAlpha  = <span class="hljs-number">1.0</span> / roughness;</span>
-<span class="line">    <span class="hljs-keyword">float</span> cos2h = NoH * NoH;</span>
-<span class="line">    <span class="hljs-keyword">float</span> sin2h = max(<span class="hljs-number">1.0</span> - cos2h, <span class="hljs-number">0.0078125</span>); <span class="hljs-comment">// 2^(-14/2), so sin2h^2 &gt; 0 in fp16</span></span>
+<span class="line">    <span class="hljs-type">float</span> invAlpha  = <span class="hljs-number">1.0</span> / roughness;</span>
+<span class="line">    <span class="hljs-type">float</span> cos2h = NoH * NoH;</span>
+<span class="line">    <span class="hljs-type">float</span> sin2h = <span class="hljs-built_in">max</span>(<span class="hljs-number">1.0</span> - cos2h, <span class="hljs-number">0.0078125</span>); <span class="hljs-comment">// 2^(-14/2), so sin2h^2 &gt; 0 in fp16</span></span>
 <span class="line">    <span class="hljs-keyword">return</span> (<span class="hljs-number">2.0</span> + invAlpha) * <span class="hljs-built_in">pow</span>(sin2h, invAlpha * <span class="hljs-number">0.5</span>) / (<span class="hljs-number">2.0</span> * PI);</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_clothcharliebrdf">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;18:</b> Implementation of the &ldquo;Charlie&rdquo; NDF in GLSL</div></center>
 <a class="target" name="sheencolor">&nbsp;</a><a class="target" name="materialsystem/clothmodel/clothspecularbrdf/sheencolor">&nbsp;</a><a class="target" name="toc4.12.1.1">&nbsp;</a><h4 id="sheen-color"><a class="header" href="#sheen-color">Sheen color</a></h4>
@@ -1745,21 +1745,21 @@ The photometric attenuation function can be easily implemented in GLSL by adding
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_glslphotometricpunctuallight">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;22:</b> Implementation of attenuation from photometric profiles in GLSL</div></center>
 <p>
 <p>The light intensity is computed CPU-side (<a href="#listing_photometriclightintensity">listing 23</a>) and depends on whether the photometric profile is used as a mask.</p>
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-keyword">float</span> multiplier;</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-type">float</span> multiplier;</span>
 <span class="line"><span class="hljs-comment">// Photometric profile used as a mask</span></span>
-<span class="line"><span class="hljs-keyword">if</span> (photometricLight.isMasked()) {</span>
+<span class="line"><span class="hljs-keyword">if</span> (photometricLight.<span class="hljs-built_in">isMasked</span>()) {</span>
 <span class="line">    <span class="hljs-comment">// The desired intensity is set by the artist</span></span>
 <span class="line">    <span class="hljs-comment">// The integrated intensity comes from a Monte-Carlo</span></span>
 <span class="line">    <span class="hljs-comment">// integration over the unit sphere around the luminaire</span></span>
-<span class="line">    multiplier = photometricLight.getDesiredIntensity() /</span>
-<span class="line">            photometricLight.getIntegratedIntensity();</span>
+<span class="line">    multiplier = photometricLight.<span class="hljs-built_in">getDesiredIntensity</span>() /</span>
+<span class="line">            photometricLight.<span class="hljs-built_in">getIntegratedIntensity</span>();</span>
 <span class="line">} <span class="hljs-keyword">else</span> {</span>
 <span class="line">    <span class="hljs-comment">// Multiplier provided for convenience, set to 1.0 by default</span></span>
-<span class="line">    multiplier = photometricLight.getMultiplier();</span>
+<span class="line">    multiplier = photometricLight.<span class="hljs-built_in">getMultiplier</span>();</span>
 <span class="line">}</span>
 <span class="line"></span>
 <span class="line"><span class="hljs-comment">// The max intensity in cd comes from the IES profile</span></span>
-<span class="line"><span class="hljs-keyword">float</span> lightIntensity = photometricLight.getMaxIntensity() * multiplier;</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_photometriclightintensity">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;23:</b> Computing the intensity of a photometric light on the CPU</div></center>
+<span class="line"><span class="hljs-type">float</span> lightIntensity = photometricLight.<span class="hljs-built_in">getMaxIntensity</span>() * multiplier;</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_photometriclightintensity">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;23:</b> Computing the intensity of a photometric light on the CPU</div></center>
 <p>
 <div class="endnote"><a class="target" name="endnote-xarrowintensity">&nbsp;</a><sup>4</sup> The XArrow profile declares a luminous intensity of 1,750 lm but a Monte-Carlo integration shows an intensity of only 350 lm.
 </div>
@@ -2058,19 +2058,19 @@ In practice only 4 or 9 coefficients (i.e.: 2 or 3 bands) are enough for \(\cosT
 <center><div class="image" style=""><a href="../images/ibl/ibl_irradiance_sh2.png" target="_blank"><img class="markdeep" src="../images/ibl/ibl_irradiance_sh2.png" style="max-width:100%;" /></a><center><span class="imagecaption"><a class="target" name="figure_iblsh2">&nbsp;</a><b style="font-style:normal;">Figure&nbsp;52:</b> 2 bands (4 coefficients)</span></center></div></center>
 </p><p>
 In practice we pre-convolve \(\Lt\) with \(\cosTheta\) and pre-scale these coefficients by the basis scaling factors \(K_l^m\) so that the reconstruction code is as simple as possible in the shader:
-</p><pre class="listing tilde"><code><span class="line">vec3 <span class="hljs-function"><span class="hljs-title">irradianceSH</span>(<span class="hljs-params">vec3 n</span>)</span> {</span>
-<span class="line">    <span class="hljs-comment">// uniform vec3 sphericalHarmonics[9]</span></span>
-<span class="line">    <span class="hljs-comment">// We can use only the first 2 bands for better performance</span></span>
-<span class="line">    <span class="hljs-keyword">return</span></span>
-<span class="line">          sphericalHarmonics[<span class="hljs-number">0</span>]</span>
-<span class="line">        + sphericalHarmonics[<span class="hljs-number">1</span>] * (n.y)</span>
-<span class="line">        + sphericalHarmonics[<span class="hljs-number">2</span>] * (n.z)</span>
-<span class="line">        + sphericalHarmonics[<span class="hljs-number">3</span>] * (n.x)</span>
-<span class="line">        + sphericalHarmonics[<span class="hljs-number">4</span>] * (n.y * n.x)</span>
-<span class="line">        + sphericalHarmonics[<span class="hljs-number">5</span>] * (n.y * n.z)</span>
-<span class="line">        + sphericalHarmonics[<span class="hljs-number">6</span>] * (<span class="hljs-number">3.0</span> * n.z * n.z - <span class="hljs-number">1.0</span>)</span>
-<span class="line">        + sphericalHarmonics[<span class="hljs-number">7</span>] * (n.z * n.x)</span>
-<span class="line">        + sphericalHarmonics[<span class="hljs-number">8</span>] * (n.x * n.x - n.y * n.y);</span>
+</p><pre class="listing tilde"><code><span class="line">vec3 irradianceSH(vec3 n) {</span>
+<span class="line">    // uniform vec3 sphericalHarmonics<span class="hljs-selector-attr">[9]</span></span>
+<span class="line">    // We can <span class="hljs-selector-tag">use</span> only the first <span class="hljs-number">2</span> bands for better performance</span>
+<span class="line">    return</span>
+<span class="line">          sphericalHarmonics<span class="hljs-selector-attr">[0]</span></span>
+<span class="line">        + sphericalHarmonics<span class="hljs-selector-attr">[1]</span> * (n<span class="hljs-selector-class">.y</span>)</span>
+<span class="line">        + sphericalHarmonics<span class="hljs-selector-attr">[2]</span> * (n<span class="hljs-selector-class">.z</span>)</span>
+<span class="line">        + sphericalHarmonics<span class="hljs-selector-attr">[3]</span> * (n<span class="hljs-selector-class">.x</span>)</span>
+<span class="line">        + sphericalHarmonics<span class="hljs-selector-attr">[4]</span> * (n<span class="hljs-selector-class">.y</span> * n<span class="hljs-selector-class">.x</span>)</span>
+<span class="line">        + sphericalHarmonics<span class="hljs-selector-attr">[5]</span> * (n<span class="hljs-selector-class">.y</span> * n<span class="hljs-selector-class">.z</span>)</span>
+<span class="line">        + sphericalHarmonics<span class="hljs-selector-attr">[6]</span> * (<span class="hljs-number">3.0</span> * n<span class="hljs-selector-class">.z</span> * n<span class="hljs-selector-class">.z</span> - <span class="hljs-number">1.0</span>)</span>
+<span class="line">        + sphericalHarmonics<span class="hljs-selector-attr">[7]</span> * (n<span class="hljs-selector-class">.z</span> * n<span class="hljs-selector-class">.x</span>)</span>
+<span class="line">        + sphericalHarmonics<span class="hljs-selector-attr">[8]</span> * (n<span class="hljs-selector-class">.x</span> * n<span class="hljs-selector-class">.x</span> - n<span class="hljs-selector-class">.y</span> * n<span class="hljs-selector-class">.y</span>);</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_irradiancesh">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;26:</b> GLSL code to reconstruct the irradiance from the pre-scaled SH</div></center>
 <p>
 <p>Note that with 2 bands, the computation above becomes a single (4 \times 4) matrix-by-vector multiply.</p>
@@ -2443,7 +2443,7 @@ LD(n, \alpha)                                   &=      \frac{\sum_i^N V(l_i, n,
 $$
 </p><p>
 These two new \(DFG\) terms simply need to replace the ones used in the implementation shown in section  <a href="#toc9.5">9.5</a>:
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-keyword">float</span> Fc = <span class="hljs-built_in">pow</span>(<span class="hljs-number">1</span> - VoH, <span class="hljs-number">5.0f</span>);</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-type">float</span> Fc = <span class="hljs-built_in">pow</span>(<span class="hljs-number">1</span> - VoH, <span class="hljs-number">5.0f</span>);</span>
 <span class="line">r.x += Gv * Fc;</span>
 <span class="line">r.y += Gv;</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_multiscatteriblpreintegration">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;29:</b> C++ implementation of the \(L_{DFG}\) term for multiscattering</div></center>
 <p>
@@ -2507,11 +2507,11 @@ using an environment made of colored vertical stripes (skybox hidden).</span></c
 <p>
 <p>When sampling the IBL, the clear coat layer is calculated as a second specular lobe. This specular lobe is oriented along the view direction since we cannot reasonably integrate over the hemisphere. <a href="#listing_clearcoatibl">Listing 31</a> demonstrates this approximation in practice. It also shows the energy conservation step. It is important to note that this second specular lobe is computed exactly the same way as the main specular lobe, using the same DFG approximation.</p>
 </p><pre class="listing tilde"><code><span class="line"><span class="hljs-comment">// clearCoat_NoV == shading_NoV if the clear coat layer doesn&#x27;t have its own normal map</span></span>
-<span class="line"><span class="hljs-built_in">float</span> Fc = F_Schlick(<span class="hljs-number">0.04</span>, <span class="hljs-number">1.0</span>, clearCoat_NoV) * clearCoat;</span>
+<span class="line"><span class="hljs-type">float</span> Fc = <span class="hljs-built_in">F_Schlick</span>(<span class="hljs-number">0.04</span>, <span class="hljs-number">1.0</span>, clearCoat_NoV) * clearCoat;</span>
 <span class="line"><span class="hljs-comment">// base layer attenuation for energy compensation</span></span>
 <span class="line">iblDiffuse  *= <span class="hljs-number">1.0</span> - Fc;</span>
-<span class="line">iblSpecular *= sq(<span class="hljs-number">1.0</span> - Fc);</span>
-<span class="line">iblSpecular += specularIBL(r, clearCoatPerceptualRoughness) * Fc;</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_clearcoatibl">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;31:</b> GLSL implementation of the clear coat specular lobe for image-based lighting</div></center>
+<span class="line">iblSpecular *= <span class="hljs-built_in">sq</span>(<span class="hljs-number">1.0</span> - Fc);</span>
+<span class="line">iblSpecular += <span class="hljs-built_in">specularIBL</span>(r, clearCoatPerceptualRoughness) * Fc;</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_clearcoatibl">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;31:</b> GLSL implementation of the clear coat specular lobe for image-based lighting</div></center>
 <a class="target" name="anisotropy">&nbsp;</a><a class="target" name="lighting/imagebasedlights/anisotropy">&nbsp;</a><a class="target" name="toc5.3.6">&nbsp;</a><h3 id="anisotropy"><a class="header" href="#anisotropy">Anisotropy </a></h3>
 <p>
 <p>[<a href="#citation-mcauley15">McAuley15</a>] describes a technique called “bent reflection vector”, based [<a href="#citation-revie12">Revie12</a>]. The bent reflection vector is a rough approximation of anisotropic lighting but the alternative is to use importance sampling. This approximation is sufficiently cheap to compute and provides good results, as shown in <a href="#figure_anisotropicibl1">figure 59</a> and <a href="#figure_anisotropicibl2">figure 60</a>.</p>
@@ -2550,17 +2550,17 @@ The DG term is generated using uniform sampling as recommended in [<a href="#cit
 <center><div class="image" style=""><a href="../images/ibl/dfg_cloth.png" target="_blank"><img class="markdeep" src="../images/ibl/dfg_cloth.png" /></a><center><span class="imagecaption"><a class="target" name="figure_dfgclothlut">&nbsp;</a><b style="font-style:normal;">Figure&nbsp;62:</b> DFG LUT with a 3rd channel encoding the DG term of the cloth BRDF</span></center></div></center>
 </p><p>
 The remainder of the image-based lighting implementation follows the same steps as the implementation of regular lights, including the optional subsurface scattering term and its wrap diffuse component. Just as with the clear coat IBL implementation, we cannot integrate over the hemisphere and use the view direction as the dominant light direction to compute the wrap diffuse component.
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-built_in">float</span> diffuse = Fd_Lambert() * ambientOcclusion;</span>
-<span class="line"><span class="hljs-meta">#<span class="hljs-meta-keyword">if</span> defined(SHADING_MODEL_CLOTH)</span></span>
-<span class="line"><span class="hljs-meta">#<span class="hljs-meta-keyword">if</span> defined(MATERIAL_HAS_SUBSURFACE_COLOR)</span></span>
-<span class="line">diffuse *= saturate((NoV + <span class="hljs-number">0.5</span>) / <span class="hljs-number">2.25</span>);</span>
-<span class="line"><span class="hljs-meta">#<span class="hljs-meta-keyword">endif</span></span></span>
-<span class="line"><span class="hljs-meta">#<span class="hljs-meta-keyword">endif</span></span></span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-type">float</span> diffuse = <span class="hljs-built_in">Fd_Lambert</span>() * ambientOcclusion;</span>
+<span class="line"><span class="hljs-meta">#<span class="hljs-keyword">if</span> defined(SHADING_MODEL_CLOTH)</span></span>
+<span class="line"><span class="hljs-meta">#<span class="hljs-keyword">if</span> defined(MATERIAL_HAS_SUBSURFACE_COLOR)</span></span>
+<span class="line">diffuse *= <span class="hljs-built_in">saturate</span>((NoV + <span class="hljs-number">0.5</span>) / <span class="hljs-number">2.25</span>);</span>
+<span class="line"><span class="hljs-meta">#<span class="hljs-keyword">endif</span></span></span>
+<span class="line"><span class="hljs-meta">#<span class="hljs-keyword">endif</span></span></span>
 <span class="line"></span>
-<span class="line">vec3 indirectDiffuse = irradianceIBL(n) * diffuse;</span>
-<span class="line"><span class="hljs-meta">#<span class="hljs-meta-keyword">if</span> defined(SHADING_MODEL_CLOTH) &amp;&amp; defined(MATERIAL_HAS_SUBSURFACE_COLOR)</span></span>
-<span class="line">indirectDiffuse *= saturate(subsurfaceColor + NoV);</span>
-<span class="line"><span class="hljs-meta">#<span class="hljs-meta-keyword">endif</span></span></span>
+<span class="line">vec3 indirectDiffuse = <span class="hljs-built_in">irradianceIBL</span>(n) * diffuse;</span>
+<span class="line"><span class="hljs-meta">#<span class="hljs-keyword">if</span> defined(SHADING_MODEL_CLOTH) &amp;&amp; defined(MATERIAL_HAS_SUBSURFACE_COLOR)</span></span>
+<span class="line">indirectDiffuse *= <span class="hljs-built_in">saturate</span>(subsurfaceColor + NoV);</span>
+<span class="line"><span class="hljs-meta">#<span class="hljs-keyword">endif</span></span></span>
 <span class="line"></span>
 <span class="line">vec3 ibl = diffuseColor * indirectDiffuse + indirectSpecular * specularColor;</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_clothapprox">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;34:</b> GLSL implementation of the DFG approximation for the cloth NDF</div></center>
 <p>
@@ -2982,20 +2982,20 @@ L_{max} &= 2^{EV_{100}} \times 1.2
 <span class="line"><span class="hljs-comment">// aperture in f-stops</span></span>
 <span class="line"><span class="hljs-comment">// shutterSpeed in seconds</span></span>
 <span class="line"><span class="hljs-comment">// sensitivity in ISO</span></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">float</span> <span class="hljs-title">exposureSettings</span><span class="hljs-params">(<span class="hljs-keyword">float</span> aperture, <span class="hljs-keyword">float</span> shutterSpeed, <span class="hljs-keyword">float</span> sensitivity)</span> </span>{</span>
+<span class="line"><span class="hljs-type">float</span> <span class="hljs-title function_">exposureSettings</span><span class="hljs-params">(<span class="hljs-type">float</span> aperture, <span class="hljs-type">float</span> shutterSpeed, <span class="hljs-type">float</span> sensitivity)</span> {</span>
 <span class="line">    <span class="hljs-keyword">return</span> log2((aperture * aperture) / shutterSpeed * <span class="hljs-number">100.0</span> / sensitivity);</span>
 <span class="line">}</span>
 <span class="line"></span>
 <span class="line"><span class="hljs-comment">// Computes the exposure normalization factor from</span></span>
 <span class="line"><span class="hljs-comment">// the camera&#x27;s EV100</span></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">float</span> <span class="hljs-title">exposure</span><span class="hljs-params">(<span class="hljs-keyword">float</span> ev100)</span> </span>{</span>
-<span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-number">1.0</span> / (<span class="hljs-built_in">pow</span>(<span class="hljs-number">2.0</span>, ev100) * <span class="hljs-number">1.2</span>);</span>
+<span class="line"><span class="hljs-type">float</span> <span class="hljs-title function_">exposure</span><span class="hljs-params">(<span class="hljs-type">float</span> ev100)</span> {</span>
+<span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-number">1.0</span> / (pow(<span class="hljs-number">2.0</span>, ev100) * <span class="hljs-number">1.2</span>);</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-keyword">float</span> ev100 = exposureSettings(aperture, shutterSpeed, sensitivity);</span>
-<span class="line"><span class="hljs-keyword">float</span> exposure = exposure(ev100);</span>
+<span class="line"><span class="hljs-type">float</span> <span class="hljs-variable">ev100</span> <span class="hljs-operator">=</span> exposureSettings(aperture, shutterSpeed, sensitivity);</span>
+<span class="line"><span class="hljs-type">float</span> <span class="hljs-variable">exposure</span> <span class="hljs-operator">=</span> exposure(ev100);</span>
 <span class="line"></span>
-<span class="line">vec4 color = evaluateLighting();</span>
+<span class="line"><span class="hljs-type">vec4</span> <span class="hljs-variable">color</span> <span class="hljs-operator">=</span> evaluateLighting();</span>
 <span class="line">color.rgb *= exposure;</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_fragmentexposure">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;42:</b> Implementation of exposure in GLSL</div></center>
 <p>
 <p>In practice the exposure factor can be pre-computed on the CPU to save shader instructions.</p>
@@ -3466,9 +3466,9 @@ Our implementation is presented in <a href="#listing_specularcolorimpl">listing&
 <span class="line"><span class="hljs-comment">// Data source:</span></span>
 <span class="line"><span class="hljs-comment">//     http://cvrl.ioo.ucl.ac.uk/cmfs.htm</span></span>
 <span class="line"><span class="hljs-comment">//     http://cvrl.ioo.ucl.ac.uk/database/text/cmfs/ciexyz31.htm</span></span>
-<span class="line"><span class="hljs-keyword">const</span> <span class="hljs-keyword">size_t</span> CIE_XYZ_START = <span class="hljs-number">360</span>;</span>
-<span class="line"><span class="hljs-keyword">const</span> <span class="hljs-keyword">size_t</span> CIE_XYZ_COUNT = <span class="hljs-number">471</span>;</span>
-<span class="line"><span class="hljs-keyword">const</span> float3 CIE_XYZ[CIE_XYZ_COUNT] = { ... };</span>
+<span class="line"><span class="hljs-type">const</span> <span class="hljs-type">size_t</span> CIE_XYZ_START = <span class="hljs-number">360</span>;</span>
+<span class="line"><span class="hljs-type">const</span> <span class="hljs-type">size_t</span> CIE_XYZ_COUNT = <span class="hljs-number">471</span>;</span>
+<span class="line"><span class="hljs-type">const</span> float3 CIE_XYZ[CIE_XYZ_COUNT] = { ... };</span>
 <span class="line"></span>
 <span class="line"><span class="hljs-comment">// CIE Standard Illuminant D65 relative spectral power distribution,</span></span>
 <span class="line"><span class="hljs-comment">// from 300nm to 830, at 5nm intervals</span></span>
@@ -3476,51 +3476,51 @@ Our implementation is presented in <a href="#listing_specularcolorimpl">listing&
 <span class="line"><span class="hljs-comment">// Data source:</span></span>
 <span class="line"><span class="hljs-comment">//     https://en.wikipedia.org/wiki/Illuminant_D65</span></span>
 <span class="line"><span class="hljs-comment">//     https://cielab.xyz/pdf/CIE_sel_colorimetric_tables.xls</span></span>
-<span class="line"><span class="hljs-keyword">const</span> <span class="hljs-keyword">size_t</span> CIE_D65_INTERVAL = <span class="hljs-number">5</span>;</span>
-<span class="line"><span class="hljs-keyword">const</span> <span class="hljs-keyword">size_t</span> CIE_D65_START = <span class="hljs-number">300</span>;</span>
-<span class="line"><span class="hljs-keyword">const</span> <span class="hljs-keyword">size_t</span> CIE_D65_END = <span class="hljs-number">830</span>;</span>
-<span class="line"><span class="hljs-keyword">const</span> <span class="hljs-keyword">size_t</span> CIE_D65_COUNT = <span class="hljs-number">107</span>;</span>
-<span class="line"><span class="hljs-keyword">const</span> <span class="hljs-keyword">float</span> CIE_D65[CIE_D65_COUNT] = { ... };</span>
+<span class="line"><span class="hljs-type">const</span> <span class="hljs-type">size_t</span> CIE_D65_INTERVAL = <span class="hljs-number">5</span>;</span>
+<span class="line"><span class="hljs-type">const</span> <span class="hljs-type">size_t</span> CIE_D65_START = <span class="hljs-number">300</span>;</span>
+<span class="line"><span class="hljs-type">const</span> <span class="hljs-type">size_t</span> CIE_D65_END = <span class="hljs-number">830</span>;</span>
+<span class="line"><span class="hljs-type">const</span> <span class="hljs-type">size_t</span> CIE_D65_COUNT = <span class="hljs-number">107</span>;</span>
+<span class="line"><span class="hljs-type">const</span> <span class="hljs-type">float</span> CIE_D65[CIE_D65_COUNT] = { ... };</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-class"><span class="hljs-keyword">struct</span> <span class="hljs-title">Sample</span> {</span></span>
-<span class="line">    <span class="hljs-keyword">float</span> w = <span class="hljs-number">0.0f</span>; <span class="hljs-comment">// wavelength</span></span>
-<span class="line">    <span class="hljs-built_in">std</span>::<span class="hljs-built_in">complex</span>&lt;<span class="hljs-keyword">float</span>&gt; ior; <span class="hljs-comment">// complex IOR, n + ik</span></span>
+<span class="line"><span class="hljs-keyword">struct</span> <span class="hljs-title class_">Sample</span> {</span>
+<span class="line">    <span class="hljs-type">float</span> w = <span class="hljs-number">0.0f</span>; <span class="hljs-comment">// wavelength</span></span>
+<span class="line">    std::complex&lt;<span class="hljs-type">float</span>&gt; ior; <span class="hljs-comment">// complex IOR, n + ik</span></span>
 <span class="line">};</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">static</span> <span class="hljs-keyword">float</span> <span class="hljs-title">illuminantD65</span><span class="hljs-params">(<span class="hljs-keyword">float</span> w)</span> </span>{</span>
-<span class="line">    <span class="hljs-keyword">auto</span> i0 = <span class="hljs-keyword">size_t</span>((w - CIE_D65_START) / CIE_D65_INTERVAL);</span>
-<span class="line">    uint2 indexBounds{i0, <span class="hljs-built_in">std</span>::min(i0 + <span class="hljs-number">1</span>, CIE_D65_END)};</span>
+<span class="line"><span class="hljs-function"><span class="hljs-type">static</span> <span class="hljs-type">float</span> <span class="hljs-title">illuminantD65</span><span class="hljs-params">(<span class="hljs-type">float</span> w)</span> </span>{</span>
+<span class="line">    <span class="hljs-keyword">auto</span> i0 = <span class="hljs-built_in">size_t</span>((w - CIE_D65_START) / CIE_D65_INTERVAL);</span>
+<span class="line">    uint2 indexBounds{i0, std::<span class="hljs-built_in">min</span>(i0 + <span class="hljs-number">1</span>, CIE_D65_END)};</span>
 <span class="line"></span>
 <span class="line">    float2 wavelengthBounds = CIE_D65_START + float2{indexBounds} * CIE_D65_INTERVAL;</span>
-<span class="line">    <span class="hljs-keyword">float</span> t = (w - wavelengthBounds.x) / (wavelengthBounds.y - wavelengthBounds.x);</span>
-<span class="line">    <span class="hljs-keyword">return</span> lerp(CIE_D65[indexBounds.x], CIE_D65[indexBounds.y], t);</span>
+<span class="line">    <span class="hljs-type">float</span> t = (w - wavelengthBounds.x) / (wavelengthBounds.y - wavelengthBounds.x);</span>
+<span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-built_in">lerp</span>(CIE_D65[indexBounds.x], CIE_D65[indexBounds.y], t);</span>
 <span class="line">}</span>
 <span class="line"></span>
 <span class="line"><span class="hljs-comment">// For std::lower_bound</span></span>
-<span class="line"><span class="hljs-keyword">bool</span> <span class="hljs-keyword">operator</span>&lt;(<span class="hljs-keyword">const</span> Sample&amp; lhs, <span class="hljs-keyword">const</span> Sample&amp; rhs) {</span>
+<span class="line"><span class="hljs-type">bool</span> <span class="hljs-keyword">operator</span>&lt;(<span class="hljs-type">const</span> Sample&amp; lhs, <span class="hljs-type">const</span> Sample&amp; rhs) {</span>
 <span class="line">    <span class="hljs-keyword">return</span> lhs.w &lt; rhs.w;</span>
 <span class="line">}</span>
 <span class="line"></span>
 <span class="line"><span class="hljs-comment">// The wavelength w must be between 360nm and 830nm</span></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">static</span> <span class="hljs-built_in">std</span>::<span class="hljs-built_in">complex</span>&lt;<span class="hljs-keyword">float</span>&gt; <span class="hljs-title">findSample</span><span class="hljs-params">(<span class="hljs-keyword">const</span> <span class="hljs-built_in">std</span>::<span class="hljs-built_in">vector</span>&lt;sample&gt;&amp; samples, <span class="hljs-keyword">float</span> w)</span> </span>{</span>
-<span class="line">    <span class="hljs-keyword">auto</span> i1 = <span class="hljs-built_in">std</span>::lower_bound(</span>
-<span class="line">         samples.begin(), samples.end(), Sample{w, <span class="hljs-number">0.0f</span> + <span class="hljs-number">0.0</span><span class="hljs-keyword">if</span>});</span>
+<span class="line"><span class="hljs-function"><span class="hljs-type">static</span> std::complex&lt;<span class="hljs-type">float</span>&gt; <span class="hljs-title">findSample</span><span class="hljs-params">(<span class="hljs-type">const</span> std::vector&lt;sample&gt;&amp; samples, <span class="hljs-type">float</span> w)</span> </span>{</span>
+<span class="line">    <span class="hljs-keyword">auto</span> i1 = std::<span class="hljs-built_in">lower_bound</span>(</span>
+<span class="line">         samples.<span class="hljs-built_in">begin</span>(), samples.<span class="hljs-built_in">end</span>(), Sample{w, <span class="hljs-number">0.0f</span> + <span class="hljs-number">0.0</span><span class="hljs-keyword">if</span>});</span>
 <span class="line">    <span class="hljs-keyword">auto</span> i0 = i1 - <span class="hljs-number">1</span>;</span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-comment">// Interpolate the complex IORs</span></span>
-<span class="line">    <span class="hljs-keyword">float</span> t = (w - i0-&gt;w) / (i1-&gt;w - i0-&gt;w);</span>
-<span class="line">    <span class="hljs-keyword">float</span> n = lerp(i0-&gt;ior.real(), i1-&gt;ior.real(), t);</span>
-<span class="line">    <span class="hljs-keyword">float</span> k = lerp(i0-&gt;ior.imag(), i1-&gt;ior.imag(), t);</span>
+<span class="line">    <span class="hljs-type">float</span> t = (w - i0-&gt;w) / (i1-&gt;w - i0-&gt;w);</span>
+<span class="line">    <span class="hljs-type">float</span> n = <span class="hljs-built_in">lerp</span>(i0-&gt;ior.<span class="hljs-built_in">real</span>(), i1-&gt;ior.<span class="hljs-built_in">real</span>(), t);</span>
+<span class="line">    <span class="hljs-type">float</span> k = <span class="hljs-built_in">lerp</span>(i0-&gt;ior.<span class="hljs-built_in">imag</span>(), i1-&gt;ior.<span class="hljs-built_in">imag</span>(), t);</span>
 <span class="line">    <span class="hljs-keyword">return</span> { n, k };</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">static</span> <span class="hljs-keyword">float</span> <span class="hljs-title">fresnel</span><span class="hljs-params">(<span class="hljs-keyword">const</span> <span class="hljs-built_in">std</span>::<span class="hljs-built_in">complex</span>&lt;<span class="hljs-keyword">float</span>&gt;&amp; sample)</span> </span>{</span>
-<span class="line">    <span class="hljs-keyword">return</span> (((sample - (<span class="hljs-number">1.0f</span> + <span class="hljs-number">0</span><span class="hljs-keyword">if</span>)) * (<span class="hljs-built_in">std</span>::conj(sample) - (<span class="hljs-number">1.0f</span> + <span class="hljs-number">0</span><span class="hljs-keyword">if</span>))) /</span>
-<span class="line">            ((sample + (<span class="hljs-number">1.0f</span> + <span class="hljs-number">0</span><span class="hljs-keyword">if</span>)) * (<span class="hljs-built_in">std</span>::conj(sample) + (<span class="hljs-number">1.0f</span> + <span class="hljs-number">0</span><span class="hljs-keyword">if</span>)))).real();</span>
+<span class="line"><span class="hljs-function"><span class="hljs-type">static</span> <span class="hljs-type">float</span> <span class="hljs-title">fresnel</span><span class="hljs-params">(<span class="hljs-type">const</span> std::complex&lt;<span class="hljs-type">float</span>&gt;&amp; sample)</span> </span>{</span>
+<span class="line">    <span class="hljs-keyword">return</span> (((sample - (<span class="hljs-number">1.0f</span> + <span class="hljs-number">0</span><span class="hljs-keyword">if</span>)) * (std::<span class="hljs-built_in">conj</span>(sample) - (<span class="hljs-number">1.0f</span> + <span class="hljs-number">0</span><span class="hljs-keyword">if</span>))) /</span>
+<span class="line">            ((sample + (<span class="hljs-number">1.0f</span> + <span class="hljs-number">0</span><span class="hljs-keyword">if</span>)) * (std::<span class="hljs-built_in">conj</span>(sample) + (<span class="hljs-number">1.0f</span> + <span class="hljs-number">0</span><span class="hljs-keyword">if</span>)))).<span class="hljs-built_in">real</span>();</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">static</span> float3 <span class="hljs-title">XYZ_to_sRGB</span><span class="hljs-params">(<span class="hljs-keyword">const</span> float3&amp; v)</span> </span>{</span>
-<span class="line">    <span class="hljs-keyword">const</span> mat3f XYZ_sRGB{</span>
+<span class="line"><span class="hljs-function"><span class="hljs-type">static</span> float3 <span class="hljs-title">XYZ_to_sRGB</span><span class="hljs-params">(<span class="hljs-type">const</span> float3&amp; v)</span> </span>{</span>
+<span class="line">    <span class="hljs-type">const</span> mat3f XYZ_sRGB{</span>
 <span class="line">             <span class="hljs-number">3.2404542f</span>, <span class="hljs-number">-0.9692660f</span>,  <span class="hljs-number">0.0556434f</span>,</span>
 <span class="line">            <span class="hljs-number">-1.5371385f</span>,  <span class="hljs-number">1.8760108f</span>, <span class="hljs-number">-0.2040259f</span>,</span>
 <span class="line">            <span class="hljs-number">-0.4985314f</span>,  <span class="hljs-number">0.0415560f</span>,  <span class="hljs-number">1.0572252f</span></span>
@@ -3529,21 +3529,21 @@ Our implementation is presented in <a href="#listing_specularcolorimpl">listing&
 <span class="line">}</span>
 <span class="line"></span>
 <span class="line"><span class="hljs-comment">// Outputs a linear sRGB color</span></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">static</span> float3 <span class="hljs-title">computeColor</span><span class="hljs-params">(<span class="hljs-keyword">const</span> <span class="hljs-built_in">std</span>::<span class="hljs-built_in">vector</span>&lt;sample&gt;&amp; samples)</span> </span>{</span>
+<span class="line"><span class="hljs-function"><span class="hljs-type">static</span> float3 <span class="hljs-title">computeColor</span><span class="hljs-params">(<span class="hljs-type">const</span> std::vector&lt;sample&gt;&amp; samples)</span> </span>{</span>
 <span class="line">    float3 xyz{<span class="hljs-number">0.0f</span>};</span>
-<span class="line">    <span class="hljs-keyword">float</span> y = <span class="hljs-number">0.0f</span>;</span>
+<span class="line">    <span class="hljs-type">float</span> y = <span class="hljs-number">0.0f</span>;</span>
 <span class="line"></span>
-<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-keyword">size_t</span> i = <span class="hljs-number">0</span>; i &lt; CIE_XYZ_COUNT; i++) {</span>
+<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-type">size_t</span> i = <span class="hljs-number">0</span>; i &lt; CIE_XYZ_COUNT; i++) {</span>
 <span class="line">        <span class="hljs-comment">// Current wavelength</span></span>
-<span class="line">        <span class="hljs-keyword">float</span> w = CIE_XYZ_START + i;</span>
+<span class="line">        <span class="hljs-type">float</span> w = CIE_XYZ_START + i;</span>
 <span class="line"></span>
 <span class="line">        <span class="hljs-comment">// Find most appropriate CIE XYZ sample for the wavelength</span></span>
-<span class="line">        <span class="hljs-keyword">auto</span> sample = findSample(samples, w);</span>
+<span class="line">        <span class="hljs-keyword">auto</span> sample = <span class="hljs-built_in">findSample</span>(samples, w);</span>
 <span class="line">        <span class="hljs-comment">// Compute Fresnel reflectance at normal incidence</span></span>
-<span class="line">        <span class="hljs-keyword">float</span> f0 = fresnel(sample);</span>
+<span class="line">        <span class="hljs-type">float</span> f0 = <span class="hljs-built_in">fresnel</span>(sample);</span>
 <span class="line"></span>
 <span class="line">        <span class="hljs-comment">// We need to multiply by the spectral power distribution of the illuminant</span></span>
-<span class="line">        <span class="hljs-keyword">float</span> d65 = illuminantD65(w);</span>
+<span class="line">        <span class="hljs-type">float</span> d65 = <span class="hljs-built_in">illuminantD65</span>(w);</span>
 <span class="line"></span>
 <span class="line">        xyz += f0 * CIE_XYZ[i] * d65;</span>
 <span class="line">        y += CIE_XYZ[i].y * d65;</span>
@@ -3552,10 +3552,10 @@ Our implementation is presented in <a href="#listing_specularcolorimpl">listing&
 <span class="line">    <span class="hljs-comment">// Normalize so that 100% reflectance at every wavelength yields Y=1</span></span>
 <span class="line">    xyz /= y;</span>
 <span class="line"></span>
-<span class="line">    float3 linear = XYZ_to_sRGB(xyz);</span>
+<span class="line">    float3 linear = <span class="hljs-built_in">XYZ_to_sRGB</span>(xyz);</span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-comment">// Normalize out-of-gamut values</span></span>
-<span class="line">    <span class="hljs-keyword">if</span> (any(greaterThan(linear, float3{<span class="hljs-number">1.0f</span>}))) linear *= <span class="hljs-number">1.0f</span> / max(linear);</span>
+<span class="line">    <span class="hljs-keyword">if</span> (<span class="hljs-built_in">any</span>(<span class="hljs-built_in">greaterThan</span>(linear, float3{<span class="hljs-number">1.0f</span>}))) linear *= <span class="hljs-number">1.0f</span> / <span class="hljs-built_in">max</span>(linear);</span>
 <span class="line"></span>
 <span class="line">    <span class="hljs-keyword">return</span> linear;</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde"><a class="target" name="listing_specularcolorimpl">&nbsp;</a><b style="font-style:normal;">Listing&nbsp;46:</b> C++ implementation to compute the base color of a metallic surface from spectral data</div></center>
@@ -3735,47 +3735,47 @@ l &= \{ cos\phi sin\theta,sin\phi sin\theta,cos\theta \}
 <pre class="listing tilde"><code><span class="line">vec2f hammersley(uint i, <span class="hljs-built_in">float</span> numSamples) {</span>
 <span class="line">    uint bits = i;</span>
 <span class="line">    bits = (bits &lt;&lt; <span class="hljs-string">16) | (bits &gt;&gt; 16</span>);</span>
-<span class="line">    bits = ((bits &amp; 0x55555555) &lt;&lt; <span class="hljs-string">1) | ((bits &amp; 0xAAAAAAAA) &gt;&gt; 1</span>);</span>
-<span class="line">    bits = ((bits &amp; 0x33333333) &lt;&lt; <span class="hljs-string">2) | ((bits &amp; 0xCCCCCCCC) &gt;&gt; 2</span>);</span>
-<span class="line">    bits = ((bits &amp; 0x0F0F0F0F) &lt;&lt; <span class="hljs-string">4) | ((bits &amp; 0xF0F0F0F0) &gt;&gt; 4</span>);</span>
-<span class="line">    bits = ((bits &amp; 0x00FF00FF) &lt;&lt; <span class="hljs-string">8) | ((bits &amp; 0xFF00FF00) &gt;&gt; 8</span>);</span>
-<span class="line">    <span class="hljs-built_in">return</span> vec2f(i / numSamples, bits / exp2(32));</span>
+<span class="line">    bits = ((bits &amp; <span class="hljs-number">0</span>x55555555) &lt;&lt; <span class="hljs-number">1</span>) | ((bits &amp; <span class="hljs-number">0</span>xAAAAAAAA) &gt;&gt; <span class="hljs-number">1</span>);</span>
+<span class="line">    bits = ((bits &amp; <span class="hljs-number">0</span>x33333333) &lt;&lt; <span class="hljs-number">2</span>) | ((bits &amp; <span class="hljs-number">0</span>xCCCCCCCC) &gt;&gt; <span class="hljs-number">2</span>);</span>
+<span class="line">    bits = ((bits &amp; <span class="hljs-number">0</span>x0F0F0F0F) &lt;&lt; <span class="hljs-number">4</span>) | ((bits &amp; <span class="hljs-number">0</span>xF0F0F0F0) &gt;&gt; <span class="hljs-number">4</span>);</span>
+<span class="line">    bits = ((bits &amp; <span class="hljs-number">0</span>x00FF00FF) &lt;&lt; <span class="hljs-number">8</span>) | ((bits &amp; <span class="hljs-number">0</span>xFF00FF00) &gt;&gt; <span class="hljs-number">8</span>);</span>
+<span class="line">    return vec2f(i / numSamples, bits / exp2(<span class="hljs-number">32</span>));</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde">C++ implementation of a Hammersley sequence generator</div></center>
 <a class="target" name="precomputinglforimage-basedlighting">&nbsp;</a><a class="target" name="annex/precomputinglforimage-basedlighting">&nbsp;</a><a class="target" name="toc9.5">&nbsp;</a><h2 id="precomputing-l-for-image-based-lighting"><a class="header" href="#precomputing-l-for-image-based-lighting">Precomputing L for image-based lighting</a></h2>
 <p>
 <p>The term ( L_{DFG} ) is only dependent on ( \NoV ). Below, the normal is arbitrarily set to ( n=\left[0, 0, 1\right] ) and (v) is chosen to satisfy ( \NoV ). The vector ( h_i ) is the ( D_{GGX}(\alpha) ) important direction sample (i).</p>
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-type">float</span> GDFG(<span class="hljs-type">float</span> NoV, <span class="hljs-type">float</span> NoL, <span class="hljs-type">float</span> a) {</span>
-<span class="line">    <span class="hljs-type">float</span> a2 = a * a;</span>
-<span class="line">    <span class="hljs-type">float</span> GGXL = NoV * <span class="hljs-built_in">sqrt</span>((-NoL * a2 + NoL) * NoL + a2);</span>
-<span class="line">    <span class="hljs-type">float</span> GGXV = NoL * <span class="hljs-built_in">sqrt</span>((-NoV * a2 + NoV) * NoV + a2);</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-type">float</span> <span class="hljs-title function_">GDFG</span><span class="hljs-params">(<span class="hljs-type">float</span> NoV, <span class="hljs-type">float</span> NoL, <span class="hljs-type">float</span> a)</span> {</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">a2</span> <span class="hljs-operator">=</span> a * a;</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">GGXL</span> <span class="hljs-operator">=</span> NoV * sqrt((-NoL * a2 + NoL) * NoL + a2);</span>
+<span class="line">    <span class="hljs-type">float</span> <span class="hljs-variable">GGXV</span> <span class="hljs-operator">=</span> NoL * sqrt((-NoV * a2 + NoV) * NoV + a2);</span>
 <span class="line">    <span class="hljs-keyword">return</span> (<span class="hljs-number">2</span> * NoL) / (GGXV + GGXL);</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line">float2 DFG(<span class="hljs-type">float</span> NoV, <span class="hljs-type">float</span> a) {</span>
+<span class="line">float2 <span class="hljs-title function_">DFG</span><span class="hljs-params">(<span class="hljs-type">float</span> NoV, <span class="hljs-type">float</span> a)</span> {</span>
 <span class="line">    float3 V;</span>
-<span class="line">    V.x = <span class="hljs-built_in">sqrt</span>(<span class="hljs-number">1.0</span>f - NoV*NoV);</span>
-<span class="line">    V.y = <span class="hljs-number">0.0</span>f;</span>
+<span class="line">    V.x = sqrt(<span class="hljs-number">1.0f</span> - NoV*NoV);</span>
+<span class="line">    V.y = <span class="hljs-number">0.0f</span>;</span>
 <span class="line">    V.z = NoV;</span>
 <span class="line"></span>
-<span class="line">    float2 r = <span class="hljs-number">0.0</span>f;</span>
-<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-type">uint</span> i = <span class="hljs-number">0</span>; i &lt; sampleCount; i++) {</span>
-<span class="line">        float2 Xi = hammersley(i, sampleCount);</span>
-<span class="line">        float3 H = importanceSampleGGX(Xi, a, N);</span>
-<span class="line">        float3 L = <span class="hljs-number">2.0</span>f * <span class="hljs-built_in">dot</span>(V, H) * H - V;</span>
+<span class="line">    <span class="hljs-type">float2</span> <span class="hljs-variable">r</span> <span class="hljs-operator">=</span> <span class="hljs-number">0.0f</span>;</span>
+<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-type">uint</span> <span class="hljs-variable">i</span> <span class="hljs-operator">=</span> <span class="hljs-number">0</span>; i &lt; sampleCount; i++) {</span>
+<span class="line">        <span class="hljs-type">float2</span> <span class="hljs-variable">Xi</span> <span class="hljs-operator">=</span> hammersley(i, sampleCount);</span>
+<span class="line">        <span class="hljs-type">float3</span> <span class="hljs-variable">H</span> <span class="hljs-operator">=</span> importanceSampleGGX(Xi, a, N);</span>
+<span class="line">        <span class="hljs-type">float3</span> <span class="hljs-variable">L</span> <span class="hljs-operator">=</span> <span class="hljs-number">2.0f</span> * dot(V, H) * H - V;</span>
 <span class="line"></span>
-<span class="line">        <span class="hljs-type">float</span> VoH = saturate(<span class="hljs-built_in">dot</span>(V, H));</span>
-<span class="line">        <span class="hljs-type">float</span> NoL = saturate(L.z);</span>
-<span class="line">        <span class="hljs-type">float</span> NoH = saturate(H.z);</span>
+<span class="line">        <span class="hljs-type">float</span> <span class="hljs-variable">VoH</span> <span class="hljs-operator">=</span> saturate(dot(V, H));</span>
+<span class="line">        <span class="hljs-type">float</span> <span class="hljs-variable">NoL</span> <span class="hljs-operator">=</span> saturate(L.z);</span>
+<span class="line">        <span class="hljs-type">float</span> <span class="hljs-variable">NoH</span> <span class="hljs-operator">=</span> saturate(H.z);</span>
 <span class="line"></span>
-<span class="line">        <span class="hljs-keyword">if</span> (NoL &gt; <span class="hljs-number">0.0</span>f) {</span>
-<span class="line">            <span class="hljs-type">float</span> G = GDFG(NoV, NoL, a);</span>
-<span class="line">            <span class="hljs-type">float</span> Gv = G * VoH / NoH;</span>
-<span class="line">            <span class="hljs-type">float</span> Fc = <span class="hljs-built_in">pow</span>(<span class="hljs-number">1</span> - VoH, <span class="hljs-number">5.0</span>f);</span>
+<span class="line">        <span class="hljs-keyword">if</span> (NoL &gt; <span class="hljs-number">0.0f</span>) {</span>
+<span class="line">            <span class="hljs-type">float</span> <span class="hljs-variable">G</span> <span class="hljs-operator">=</span> GDFG(NoV, NoL, a);</span>
+<span class="line">            <span class="hljs-type">float</span> <span class="hljs-variable">Gv</span> <span class="hljs-operator">=</span> G * VoH / NoH;</span>
+<span class="line">            <span class="hljs-type">float</span> <span class="hljs-variable">Fc</span> <span class="hljs-operator">=</span> pow(<span class="hljs-number">1</span> - VoH, <span class="hljs-number">5.0f</span>);</span>
 <span class="line">            r.x += Gv * (<span class="hljs-number">1</span> - Fc);</span>
 <span class="line">            r.y += Gv * Fc;</span>
 <span class="line">        }</span>
 <span class="line">    }</span>
-<span class="line">    <span class="hljs-keyword">return</span> r * (<span class="hljs-number">1.0</span>f / sampleCount);</span>
+<span class="line">    <span class="hljs-keyword">return</span> r * (<span class="hljs-number">1.0f</span> / sampleCount);</span>
 <span class="line">}</span></code></pre><center><div class="listingcaption tilde">C++ implementation of the \( L_{DFG} \) term</div></center>
 <a class="target" name="sphericalharmonics">&nbsp;</a><a class="target" name="annex/sphericalharmonics">&nbsp;</a><a class="target" name="toc9.6">&nbsp;</a><h2 id="spherical-harmonics"><a class="header" href="#spherical-harmonics">Spherical Harmonics</a></h2>
 <p>
@@ -3851,56 +3851,56 @@ sin(m \phi + \phi) &= sin(m \phi) cos(\phi) + cos(m \phi) sin(\phi) \Leftrightar
 \end{align*}$$
 </p><p>
 <a href="#listing_nonnormalizedshbasis">Listing&nbsp;47</a> shows the C++ code to compute the non-normalized SH basis \(\frac{y^m_l(s)}{\sqrt{2} K^m_l}\):
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-keyword">static</span> <span class="hljs-keyword">inline</span> <span class="hljs-keyword">size_t</span> <span class="hljs-title">SHindex</span><span class="hljs-params">(<span class="hljs-keyword">ssize_t</span> m, <span class="hljs-keyword">size_t</span> l)</span> </span>{</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-type">static</span> <span class="hljs-keyword">inline</span> <span class="hljs-type">size_t</span> <span class="hljs-title">SHindex</span><span class="hljs-params">(<span class="hljs-type">ssize_t</span> m, <span class="hljs-type">size_t</span> l)</span> </span>{</span>
 <span class="line">    <span class="hljs-keyword">return</span> l * (l + <span class="hljs-number">1</span>) + m;</span>
 <span class="line">}</span>
 <span class="line"></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">void</span> <span class="hljs-title">computeShBasis</span><span class="hljs-params">(</span>
-<span class="line">        <span class="hljs-keyword">double</span>* <span class="hljs-keyword">const</span> SHb,</span>
-<span class="line">        <span class="hljs-keyword">size_t</span> numBands,</span>
-<span class="line">        <span class="hljs-keyword">const</span> vec3&amp; s)</span></span>
+<span class="line"><span class="hljs-function"><span class="hljs-type">void</span> <span class="hljs-title">computeShBasis</span><span class="hljs-params">(</span>
+<span class="line">        <span class="hljs-type">double</span>* <span class="hljs-type">const</span> SHb,</span>
+<span class="line">        <span class="hljs-type">size_t</span> numBands,</span>
+<span class="line">        <span class="hljs-type">const</span> vec3&amp; s)</span></span>
 <span class="line"></span>{</span>
 <span class="line">    <span class="hljs-comment">// handle m=0 separately, since it produces only one coefficient</span></span>
-<span class="line">    <span class="hljs-keyword">double</span> Pml_2 = <span class="hljs-number">0</span>;</span>
-<span class="line">    <span class="hljs-keyword">double</span> Pml_1 = <span class="hljs-number">1</span>;</span>
+<span class="line">    <span class="hljs-type">double</span> Pml_2 = <span class="hljs-number">0</span>;</span>
+<span class="line">    <span class="hljs-type">double</span> Pml_1 = <span class="hljs-number">1</span>;</span>
 <span class="line">    SHb[<span class="hljs-number">0</span>] =  Pml_1;</span>
-<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-keyword">ssize_t</span> l = <span class="hljs-number">1</span>; l &lt; numBands; l++) {</span>
-<span class="line">        <span class="hljs-keyword">double</span> Pml = ((<span class="hljs-number">2</span> * l - <span class="hljs-number">1</span>) * Pml_1 * s.z - (l - <span class="hljs-number">1</span>) * Pml_2) / l;</span>
+<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-type">ssize_t</span> l = <span class="hljs-number">1</span>; l &lt; numBands; l++) {</span>
+<span class="line">        <span class="hljs-type">double</span> Pml = ((<span class="hljs-number">2</span> * l - <span class="hljs-number">1</span>) * Pml_1 * s.z - (l - <span class="hljs-number">1</span>) * Pml_2) / l;</span>
 <span class="line">        Pml_2 = Pml_1;</span>
 <span class="line">        Pml_1 = Pml;</span>
-<span class="line">        SHb[SHindex(<span class="hljs-number">0</span>, l)] = Pml;</span>
+<span class="line">        SHb[<span class="hljs-built_in">SHindex</span>(<span class="hljs-number">0</span>, l)] = Pml;</span>
 <span class="line">    }</span>
-<span class="line">    <span class="hljs-keyword">double</span> Pmm = <span class="hljs-number">1</span>;</span>
-<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-keyword">ssize_t</span> m = <span class="hljs-number">1</span>; m &lt; numBands ; m++) {</span>
+<span class="line">    <span class="hljs-type">double</span> Pmm = <span class="hljs-number">1</span>;</span>
+<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-type">ssize_t</span> m = <span class="hljs-number">1</span>; m &lt; numBands ; m++) {</span>
 <span class="line">        Pmm = (<span class="hljs-number">1</span> - <span class="hljs-number">2</span> * m) * Pmm;</span>
-<span class="line">        <span class="hljs-keyword">double</span> Pml_2 = Pmm;</span>
-<span class="line">        <span class="hljs-keyword">double</span> Pml_1 = (<span class="hljs-number">2</span> * m + <span class="hljs-number">1</span>)*Pmm*s.z;</span>
+<span class="line">        <span class="hljs-type">double</span> Pml_2 = Pmm;</span>
+<span class="line">        <span class="hljs-type">double</span> Pml_1 = (<span class="hljs-number">2</span> * m + <span class="hljs-number">1</span>)*Pmm*s.z;</span>
 <span class="line">        <span class="hljs-comment">// l == m</span></span>
-<span class="line">        SHb[SHindex(-m, m)] = Pml_2;</span>
-<span class="line">        SHb[SHindex( m, m)] = Pml_2;</span>
+<span class="line">        SHb[<span class="hljs-built_in">SHindex</span>(-m, m)] = Pml_2;</span>
+<span class="line">        SHb[<span class="hljs-built_in">SHindex</span>( m, m)] = Pml_2;</span>
 <span class="line">        <span class="hljs-keyword">if</span> (m + <span class="hljs-number">1</span> &lt; numBands) {</span>
 <span class="line">            <span class="hljs-comment">// l == m+1</span></span>
-<span class="line">            SHb[SHindex(-m, m + <span class="hljs-number">1</span>)] = Pml_1;</span>
-<span class="line">            SHb[SHindex( m, m + <span class="hljs-number">1</span>)] = Pml_1;</span>
-<span class="line">            <span class="hljs-keyword">for</span> (<span class="hljs-keyword">ssize_t</span> l = m + <span class="hljs-number">2</span>; l &lt; numBands; l++) {</span>
-<span class="line">                <span class="hljs-keyword">double</span> Pml = ((<span class="hljs-number">2</span> * l - <span class="hljs-number">1</span>) * Pml_1 * s.z - (l + m - <span class="hljs-number">1</span>) * Pml_2)</span>
+<span class="line">            SHb[<span class="hljs-built_in">SHindex</span>(-m, m + <span class="hljs-number">1</span>)] = Pml_1;</span>
+<span class="line">            SHb[<span class="hljs-built_in">SHindex</span>( m, m + <span class="hljs-number">1</span>)] = Pml_1;</span>
+<span class="line">            <span class="hljs-keyword">for</span> (<span class="hljs-type">ssize_t</span> l = m + <span class="hljs-number">2</span>; l &lt; numBands; l++) {</span>
+<span class="line">                <span class="hljs-type">double</span> Pml = ((<span class="hljs-number">2</span> * l - <span class="hljs-number">1</span>) * Pml_1 * s.z - (l + m - <span class="hljs-number">1</span>) * Pml_2)</span>
 <span class="line">                        / (l - m);</span>
 <span class="line">                Pml_2 = Pml_1;</span>
 <span class="line">                Pml_1 = Pml;</span>
-<span class="line">                SHb[SHindex(-m, l)] = Pml;</span>
-<span class="line">                SHb[SHindex( m, l)] = Pml;</span>
+<span class="line">                SHb[<span class="hljs-built_in">SHindex</span>(-m, l)] = Pml;</span>
+<span class="line">                SHb[<span class="hljs-built_in">SHindex</span>( m, l)] = Pml;</span>
 <span class="line">            }</span>
 <span class="line">        }</span>
 <span class="line">    }</span>
-<span class="line">    <span class="hljs-keyword">double</span> Cm = s.x;</span>
-<span class="line">    <span class="hljs-keyword">double</span> Sm = s.y;</span>
-<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-keyword">ssize_t</span> m = <span class="hljs-number">1</span>; m &lt;= numBands ; m++) {</span>
-<span class="line">        <span class="hljs-keyword">for</span> (<span class="hljs-keyword">ssize_t</span> l = m; l &lt; numBands ; l++) {</span>
-<span class="line">            SHb[SHindex(-m, l)] *= Sm;</span>
-<span class="line">            SHb[SHindex( m, l)] *= Cm;</span>
+<span class="line">    <span class="hljs-type">double</span> Cm = s.x;</span>
+<span class="line">    <span class="hljs-type">double</span> Sm = s.y;</span>
+<span class="line">    <span class="hljs-keyword">for</span> (<span class="hljs-type">ssize_t</span> m = <span class="hljs-number">1</span>; m &lt;= numBands ; m++) {</span>
+<span class="line">        <span class="hljs-keyword">for</span> (<span class="hljs-type">ssize_t</span> l = m; l &lt; numBands ; l++) {</span>
+<span class="line">            SHb[<span class="hljs-built_in">SHindex</span>(-m, l)] *= Sm;</span>
+<span class="line">            SHb[<span class="hljs-built_in">SHindex</span>( m, l)] *= Cm;</span>
 <span class="line">        }</span>
-<span class="line">        <span class="hljs-keyword">double</span> Cm1 = Cm * s.x - Sm * s.y;</span>
-<span class="line">        <span class="hljs-keyword">double</span> Sm1 = Sm * s.x + Cm * s.y;</span>
+<span class="line">        <span class="hljs-type">double</span> Cm1 = Cm * s.x - Sm * s.y;</span>
+<span class="line">        <span class="hljs-type">double</span> Sm1 = Sm * s.x + Cm * s.y;</span>
 <span class="line">        Cm = Cm1;</span>
 <span class="line">        Sm = Sm1;</span>
 <span class="line">    }</span>
@@ -3979,10 +3979,10 @@ $$\begin{equation}
 \end{equation}$$
 </p><p>
 Here is the C++ code to compute \(\hat{C}_l\):
-</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-keyword">static</span> <span class="hljs-keyword">double</span> <span class="hljs-title">factorial</span><span class="hljs-params">(<span class="hljs-keyword">size_t</span> n, <span class="hljs-keyword">size_t</span> d = <span class="hljs-number">1</span>)</span></span>;</span>
+</p><pre class="listing tilde"><code><span class="line"><span class="hljs-function"><span class="hljs-type">static</span> <span class="hljs-type">double</span> <span class="hljs-title">factorial</span><span class="hljs-params">(<span class="hljs-type">size_t</span> n, <span class="hljs-type">size_t</span> d = <span class="hljs-number">1</span>)</span></span>;</span>
 <span class="line"></span>
 <span class="line"><span class="hljs-comment">// &lt; cos(theta) &gt; SH coefficients pre-multiplied by 1 / K(0,l)</span></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">double</span> <span class="hljs-title">computeTruncatedCosSh</span><span class="hljs-params">(<span class="hljs-keyword">size_t</span> l)</span> </span>{</span>
+<span class="line"><span class="hljs-function"><span class="hljs-type">double</span> <span class="hljs-title">computeTruncatedCosSh</span><span class="hljs-params">(<span class="hljs-type">size_t</span> l)</span> </span>{</span>
 <span class="line">    <span class="hljs-keyword">if</span> (l == <span class="hljs-number">0</span>) {</span>
 <span class="line">        <span class="hljs-keyword">return</span> M_PI;</span>
 <span class="line">    } <span class="hljs-keyword">else</span> <span class="hljs-keyword">if</span> (l == <span class="hljs-number">1</span>) {</span>
@@ -3990,17 +3990,17 @@ Here is the C++ code to compute \(\hat{C}_l\):
 <span class="line">    } <span class="hljs-keyword">else</span> <span class="hljs-keyword">if</span> (l &amp; <span class="hljs-number">1</span>) {</span>
 <span class="line">        <span class="hljs-keyword">return</span> <span class="hljs-number">0</span>;</span>
 <span class="line">    }</span>
-<span class="line">    <span class="hljs-keyword">const</span> <span class="hljs-keyword">size_t</span> l_2 = l / <span class="hljs-number">2</span>;</span>
-<span class="line">    <span class="hljs-keyword">double</span> A0 = ((l_2 &amp; <span class="hljs-number">1</span>) ? <span class="hljs-number">1.0</span> : <span class="hljs-number">-1.0</span>) / ((l + <span class="hljs-number">2</span>) * (l - <span class="hljs-number">1</span>));</span>
-<span class="line">    <span class="hljs-keyword">double</span> A1 = factorial(l, l_2) / (factorial(l_2) * (<span class="hljs-number">1</span> &lt;&lt; l));</span>
+<span class="line">    <span class="hljs-type">const</span> <span class="hljs-type">size_t</span> l_2 = l / <span class="hljs-number">2</span>;</span>
+<span class="line">    <span class="hljs-type">double</span> A0 = ((l_2 &amp; <span class="hljs-number">1</span>) ? <span class="hljs-number">1.0</span> : <span class="hljs-number">-1.0</span>) / ((l + <span class="hljs-number">2</span>) * (l - <span class="hljs-number">1</span>));</span>
+<span class="line">    <span class="hljs-type">double</span> A1 = <span class="hljs-built_in">factorial</span>(l, l_2) / (<span class="hljs-built_in">factorial</span>(l_2) * (<span class="hljs-number">1</span> &lt;&lt; l));</span>
 <span class="line">    <span class="hljs-keyword">return</span> <span class="hljs-number">2</span> * M_PI * A0 * A1;</span>
 <span class="line">}</span>
 <span class="line"></span>
 <span class="line"><span class="hljs-comment">// returns n! / d!</span></span>
-<span class="line"><span class="hljs-function"><span class="hljs-keyword">double</span> <span class="hljs-title">factorial</span><span class="hljs-params">(<span class="hljs-keyword">size_t</span> n, <span class="hljs-keyword">size_t</span> d )</span> </span>{</span>
-<span class="line">   d = <span class="hljs-built_in">std</span>::max(<span class="hljs-keyword">size_t</span>(<span class="hljs-number">1</span>), d);</span>
-<span class="line">   n = <span class="hljs-built_in">std</span>::max(<span class="hljs-keyword">size_t</span>(<span class="hljs-number">1</span>), n);</span>
-<span class="line">   <span class="hljs-keyword">double</span> r = <span class="hljs-number">1.0</span>;</span>
+<span class="line"><span class="hljs-function"><span class="hljs-type">double</span> <span class="hljs-title">factorial</span><span class="hljs-params">(<span class="hljs-type">size_t</span> n, <span class="hljs-type">size_t</span> d )</span> </span>{</span>
+<span class="line">   d = std::<span class="hljs-built_in">max</span>(<span class="hljs-built_in">size_t</span>(<span class="hljs-number">1</span>), d);</span>
+<span class="line">   n = std::<span class="hljs-built_in">max</span>(<span class="hljs-built_in">size_t</span>(<span class="hljs-number">1</span>), n);</span>
+<span class="line">   <span class="hljs-type">double</span> r = <span class="hljs-number">1.0</span>;</span>
 <span class="line">   <span class="hljs-keyword">if</span> (n == d) {</span>
 <span class="line">       <span class="hljs-comment">// intentionally left blank</span></span>
 <span class="line">   } <span class="hljs-keyword">else</span> <span class="hljs-keyword">if</span> (n &gt; d) {</span>

docs/wip/sky/SimulatedSkybox.js

@@ -16,12 +16,13 @@ class SimulatedSkybox {
     this.ozone = 0.0;
     this.msFactors = [0.1, 0.5, 0.0];
     this.contrast = 1.0;
-    this.nightColor = [0.0, 0.0003, 0.00075];
+    this.nightColor = [0.0, 3.0e-9, 7.5e-9];
     this.shimmerControl = [0.0, 20.0, 0.1];
     this.cloudControl = [0.0, 0.1, 8000.0, 0.0];
     this.cloudControl2 = [0.0, 0.0, 0.0, 0.0];
     this.waterControl = [50.0, 1.0, 1.0, 4.0]; // x=Strength, y=Speed, z=DerivativeTrick, w=Octaves
     this.starControl = [0.001, 1.0, 350.0, 0.01]; // x=Density, y=Enabled, z=Frequency, w=PixelScale
+    this.starIntensity = 1.0;
     this.focalLength = 24.0;
     this.height = 1000.0;
     this.planetRadius = 6360.0;
@@ -43,7 +44,7 @@ class SimulatedSkybox {
 
     // Milky Way Parameters
     // x=Intensity, y=Saturation, z=Unused
-    this.milkyWayControl = [1.0, 1.0, 0.07];
+    this.milkyWayControl = [1.0, 1.2, 0.05];
     this.milkyWayEnabled = true;
     this.milkyWayRotation = [1, 0, 0, 0, 1, 0, 0, 0, 1]; // Identity by default
     this.initEntity();
@@ -453,6 +454,11 @@ class SimulatedSkybox {
     this.updateCoefficients();
   }
 
+  setStarIntensity(intensity) {
+    this.starIntensity = Math.max(0.0, intensity);
+    this.updateCoefficients();
+  }
+
   setMoonPosition(direction) {
     // normalize
     const len = Math.hypot(direction[0], direction[1], direction[2]);
@@ -500,6 +506,13 @@ class SimulatedSkybox {
     }
   }
 
+
+
+  setExposure(exposure) {
+    this.exposure = exposure;
+    this.updateCoefficients();
+  }
+
   updateCoefficients() {
     if (!this.materialInstance) {
       console.warn("updateCoefficients called before material loaded");
@@ -585,6 +598,7 @@ class SimulatedSkybox {
     this.materialInstance.setFloat4Parameter('cloudControl2', new Float32Array(this.cloudControl2));
     this.materialInstance.setFloat4Parameter('waterControl', new Float32Array(this.waterControl));
     this.materialInstance.setFloat4Parameter('starControl', new Float32Array(this.starControl));
+    this.materialInstance.setFloatParameter('starIntensity', this.starIntensity);
 
     this.materialInstance.setFloatParameter('sunIntensity', physicalSunIntensity);
 
@@ -615,12 +629,13 @@ class SimulatedSkybox {
 
     this.materialInstance.setFloatParameter('sunIntensity2', finalMoonIntensity);
 
-    // Scale Milky Way by Sun Intensity (Pre-exposed) to match dynamic range
-    // Calibration: 1.0 User Intensity ~ 0.025 Lux Relative (approx 2.5% of Sun Pixel Value at Day)
-    // At Sunny 16 (Sun ~ 2.6), this gives ~0.065 pixel brightness, which is visible but dim.
+    // Scale Milky Way by Sun Intensity
+    // Calibration: 1.0 User Intensity = 1.5e-3 cd/m^2 (Nits) approx.
+    // Target: 1.5e-3 Nits. SunIntensity = 100,000.
+    // Scale = 1.5e-3 / 1.0e5 = 1.5e-8.
     const mwIntensity = this.milkyWayEnabled ? this.milkyWayControl[0] : 0.0;
     const mwUniform = [
-      mwIntensity * this.sunIntensity * 0.025,
+      mwIntensity * this.sunIntensity * 1.5e-8,
       this.milkyWayControl[1],
       this.milkyWayControl[2]
     ];
@@ -634,6 +649,7 @@ class SimulatedSkybox {
     const moonHaloUpload = [...this.moonHalo];
     moonHaloUpload[2] *= moonRadConv;
     this.materialInstance.setFloat4Parameter('sunHalo2', new Float32Array(moonHaloUpload));
+    this.materialInstance.setFloatParameter('exposure', this.exposure !== undefined ? this.exposure : 1.0);
 
     // Solar Eclipse (CPU Calculation)
     const sunRadius = Math.acos(this.sunHalo[0]);

docs/wip/sky/main.js

@@ -218,6 +218,7 @@ class App {
     const exposure = this.getExposure();
     const preExposedIntensity = this.params.sunIntensity * exposure;
     this.skybox.setSunIntensity(preExposedIntensity);
+    this.skybox.setExposure(exposure); // Update Skybox Exposure Uniform
 
     // Moon Exposure
     if (this.mParams) {
@@ -389,7 +390,7 @@ class App {
     };
 
     mwFolder.add(this.mwParams, 'enabled').name('Enabled').onChange(updateMW);
-    mwFolder.add(this.mwParams, 'intensity', 0.0, 5.0).onChange(updateMW);
+    mwFolder.add(this.mwParams, 'intensity', 0.0, 100.0).onChange(updateMW);
     mwFolder.add(this.mwParams, 'saturation', 0.0, 2.0).onChange(updateMW);
     mwFolder.add(this.mwParams, 'blackPoint', 0.0, 0.5).name('Black Point').onChange(updateMW);
     mwFolder.add(this.mwParams, 'siderealTime', 0.0, 24.0).name('Sidereal Time').listen().onChange(updateMW);
@@ -464,18 +465,23 @@ class App {
     const starFolder = gui.addFolder('Stars');
     this.sParams = {
       enabled: true,
-      density: 0.001
+      density: 0.001,
+      intensity: 0.0 // 2^0 = 1.0
     };
     // Initialize defaults (Density 0.001, Enabled True)
     sky.setStarControl(0.001, true);
+    sky.setStarIntensity(Math.pow(2.0, 0.0));
 
     const updateStars = () => {
       sky.setStarControl(this.sParams.density, this.sParams.enabled);
+      sky.setStarIntensity(Math.pow(2.0, this.sParams.intensity));
     };
 
     starFolder.add(this.sParams, 'enabled').name('Enabled').onChange(updateStars);
     starFolder.add(this.sParams, 'density', 0.0, 0.01, 0.0001).name('Density').onChange(updateStars);
+    starFolder.add(this.sParams, 'intensity', 0.0, 24.0).name('Intensity (Exp)').onChange(updateStars);
     starFolder.close();
+    this.updateStars = updateStars;
 
     const artFolder = gui.addFolder('Artistic');
     // Set Horizon Glow default to 0.0
@@ -489,7 +495,7 @@ class App {
     artFolder.add(sky.msFactors, 2, 0.0, 1.0).name('Horizon Glow').onChange(v => sky.setHorizonGlow(v));
     artFolder.add(sky, 'contrast', 0.1, 2.0).onChange(v => sky.setContrast(v));
 
-    artFolder.addColor(sky, 'nightColor').onChange(v => sky.setNightColor(v));
+
 
     const shmFolder = artFolder.addFolder('Shimmer');
     // Set Shimmer Strength default to 0.0
@@ -502,7 +508,7 @@ class App {
     const camFolder = gui.addFolder('Camera');
     camFolder.add(this.params, 'focalLength', 8.0, 300.0).name('Focal Length').onChange(() => this.updateCameraProjection());
     camFolder.add(this.params, 'aperture', 1.4, 32.0).onChange(() => this.updateCameraExposure());
-    camFolder.add(this.params, 'shutterSpeed', 1.0, 1000.0).onChange(() => this.updateCameraExposure());
+    camFolder.add(this.params, 'shutterSpeed', 0.05, 1000.0).onChange(() => this.updateCameraExposure());
     camFolder.add(this.params, 'iso', 50.0, 3200.0).onChange(() => this.updateCameraExposure());
 
     const bloomFolder = camFolder.addFolder('Bloom');
@@ -692,13 +698,15 @@ class App {
     const s = this.sParams;
     const b = this.bParams;
     const m = this.mParams;
+    const mw = this.mwParams;
     const sk = this.skybox;
 
     return {
       p: { a: p.aperture, ss: p.shutterSpeed, i: p.iso, st: p.sunTheta, sp: p.sunPhi, fl: p.focalLength, si: p.sunIntensity },
       c: { v: c.volumetrics, co: c.coverage, d: c.density, h: c.height, s: c.speed, e: c.evolution },
       w: { dt: w.derivativeTrick, st: w.strength, s: w.speed, o: w.octaves },
-      s: { e: s.enabled, d: s.density },
+      s: { e: s.enabled, d: s.density, i: s.intensity },
+      mw: { e: mw.enabled, i: mw.intensity, s: mw.saturation, bp: mw.blackPoint, st: mw.siderealTime, l: mw.latitude },
       b: { e: b.enabled, lf: b.lensFlare },
       m: { e: m.enabled, az: m.azimuth, h: m.height, r: m.radius, i: m.intensity },
       cm: { t: this.camState.theta, p: this.camState.phi },
@@ -723,6 +731,7 @@ class App {
     const c = state.c;
     const w = state.w;
     const s = state.s;
+    const mw = state.mw;
     const b = state.b;
     const m = state.m;
     const k = state.k;
@@ -757,6 +766,18 @@ class App {
     if (s) {
       if (s.e !== undefined) this.sParams.enabled = s.e;
       if (s.d !== undefined) this.sParams.density = s.d;
+      if (s.i !== undefined) this.sParams.intensity = s.i;
+      if (this.updateStars) this.updateStars();
+    }
+
+    if (mw) {
+      if (mw.e !== undefined) this.mwParams.enabled = mw.e;
+      if (mw.i !== undefined) this.mwParams.intensity = mw.i;
+      if (mw.s !== undefined) this.mwParams.saturation = mw.s;
+      if (mw.bp !== undefined) this.mwParams.blackPoint = mw.bp;
+      if (mw.st !== undefined) this.mwParams.siderealTime = mw.st;
+      if (mw.l !== undefined) this.mwParams.latitude = mw.l;
+      if (this.updateMW) this.updateMW();
     }
 
     if (b) {

docs/wip/sky/simulated_skybox.mat

@@ -96,10 +96,20 @@ material {
             name : starControl, // x=Density, y=Enabled, z=Frequency, w=PixelScale
             precision : high
         },
+        {
+            type : float,
+            name : starIntensity,
+            precision : high
+        },
         {
             type : sampler2d,
             name : moonTexture
         },
+        {
+            type : float,
+            name : exposure,
+            precision : high
+        },
         {
             type : sampler2d,
             name : moonNormal
@@ -162,9 +172,9 @@ fragment {
     // --- CONFIGURATION ---
 
     // Stars
-    #define STAR_GLOBAL_INTENSITY 150.0      // Master brightness multiplier [0.0 - 500.0]
-    #define STAR_BRIGHTNESS_BASE 0.5         // Minimum random brightness [0.0 - 1.0]
-    #define STAR_BRIGHTNESS_VAR 4.0          // Random brightness variance range [0.0 - 10.0]
+    #define STAR_GLOBAL_INTENSITY 100.0      // Master brightness multiplier [0.0 - 500.0]
+    #define STAR_BRIGHTNESS_BASE 0.01        // Minimum random brightness [0.0 - 1.0]
+    #define STAR_BRIGHTNESS_VAR 15.0         // Random brightness variance range [0.0 - 10.0]
     #define STAR_FADE_SUN_ELV_HIGH 0.10      // Sun elevation (sin) where stars are fully hidden [0.0 - 0.5]
     #define STAR_FADE_SUN_ELV_LOW -0.20      // Sun elevation (sin) where stars are fully visible [-0.5 - 0.0]
     #define STAR_CLOUD_OCCLUSION 0.1         // Visibility when covered by clouds [0.0 - 1.0]
@@ -961,7 +971,11 @@ fragment {
                              highp vec4 sunHalo, highp vec4 cloudControl, highp vec4 cloudControl2, 
                              highp vec4 shimmerControl, highp vec4 waterControl,
                              highp vec3 L2, highp float sunIntensity2, highp vec4 sunHalo2,
-                             sampler2D moonTex, sampler2D moonNormal) {
+                             sampler2D moonTex, sampler2D moonNormal,
+                             highp vec3 nightColor,
+                             highp mat3 milkyWayRotation,
+                             highp vec3 milkyWayControl,
+                             sampler2D milkyWayTexture) {
 
         // Project to plane y=0
         highp float t = WATER_PLANE_HEIGHT / min(V.y, -0.0002); // Reduced clamp to minimize "wall" artifact
@@ -1057,7 +1071,7 @@ fragment {
          highp float rHorizonMask = 1.0 - smoothstep(0.0, REFLECTION_HORIZON_FADE, R.y);
         
         if (rHorizonMask > 0.0) {
-             reflection += getStarLayer(R, L, reflCloudDensity, outTransmittance, materialParams.starControl) * rHorizonMask;
+             reflection += getStarLayer(R, L, reflCloudDensity, outTransmittance, materialParams.starControl) * rHorizonMask * materialParams.exposure;
         }
 
         // Sun Disk Reflection
@@ -1088,6 +1102,29 @@ fragment {
         // Apply clouds to reflection
         reflection = mix(reflection, reflCloudLayer, reflCloudDensity);
                 
+        // Add Milky Way to Reflection
+        // Calculate Fade based on Sun Elevation
+        highp float sunElvSin = L.y;
+        highp float mwFade = smoothstep(STAR_FADE_SUN_ELV_HIGH, STAR_FADE_SUN_ELV_LOW, sunElvSin);
+        
+        if (mwFade > 0.0) {
+                 highp vec3 mwColor = getMilkyWay(R, milkyWayRotation, milkyWayTexture, milkyWayControl);
+                 
+                 // Apply Atmosphere Transmittance (approximate)
+                 mwColor *= outTransmittance;
+                 
+                 // Apply Fades
+                 mwColor *= mwFade;
+                 mwColor *= (1.0 - reflMoonOcclusion); // Occlude by Moon
+                 mwColor *= (1.0 - reflCloudDensity);  // Occlude by Clouds
+                 
+                 reflection += mwColor;
+        }
+                
+
+        // Add Night Color to Reflection
+        reflection += nightColor;
+                
         // Fresnel
         highp float F0 = WATER_FRESNEL_F0; // Water
         highp float cosTheta = clamp(dot(-V, N_water), 0.0, 1.0);
@@ -1187,7 +1224,7 @@ fragment {
 
         // 7. Stars
         // Add stars before clouds (clouds cover stars)
-        highp vec3 starColor = getStarLayer(V, L, cloudDensityVal, transmittance, materialParams.starControl);
+        highp vec3 starColor = getStarLayer(V, L, cloudDensityVal, transmittance, materialParams.starControl) * materialParams.exposure * materialParams.starIntensity;
         
         // 7b. Milky Way
         // Add Milky Way behind stars (conceptually) but handled similarly
@@ -1232,7 +1269,11 @@ fragment {
                                        materialParams.shimmerControl,
                                        materialParams.waterControl,
                                        L2, materialParams.sunIntensity2, materialParams.sunHalo2,
-                                       materialParams_moonTexture, materialParams_moonNormal);
+                                       materialParams_moonTexture, materialParams_moonNormal,
+                                       materialParams.nightColor,
+                                       materialParams.milkyWayRotation,
+                                       materialParams.milkyWayControl,
+                                       materialParams_milkyWayTexture);
             color = applyDynamicToneMapping(color, L, materialParams.contrast);
         }
 

docs_src/src_raw/wip/sky/SimulatedSkybox.js

@@ -16,12 +16,13 @@ class SimulatedSkybox {
     this.ozone = 0.0;
     this.msFactors = [0.1, 0.5, 0.0];
     this.contrast = 1.0;
-    this.nightColor = [0.0, 0.0003, 0.00075];
+    this.nightColor = [0.0, 3.0e-9, 7.5e-9];
     this.shimmerControl = [0.0, 20.0, 0.1];
     this.cloudControl = [0.0, 0.1, 8000.0, 0.0];
     this.cloudControl2 = [0.0, 0.0, 0.0, 0.0];
     this.waterControl = [50.0, 1.0, 1.0, 4.0]; // x=Strength, y=Speed, z=DerivativeTrick, w=Octaves
     this.starControl = [0.001, 1.0, 350.0, 0.01]; // x=Density, y=Enabled, z=Frequency, w=PixelScale
+    this.starIntensity = 1.0;
     this.focalLength = 24.0;
     this.height = 1000.0;
     this.planetRadius = 6360.0;
@@ -43,7 +44,7 @@ class SimulatedSkybox {
 
     // Milky Way Parameters
     // x=Intensity, y=Saturation, z=Unused
-    this.milkyWayControl = [1.0, 1.0, 0.07];
+    this.milkyWayControl = [1.0, 1.2, 0.05];
     this.milkyWayEnabled = true;
     this.milkyWayRotation = [1, 0, 0, 0, 1, 0, 0, 0, 1]; // Identity by default
     this.initEntity();
@@ -453,6 +454,11 @@ class SimulatedSkybox {
     this.updateCoefficients();
   }
 
+  setStarIntensity(intensity) {
+    this.starIntensity = Math.max(0.0, intensity);
+    this.updateCoefficients();
+  }
+
   setMoonPosition(direction) {
     // normalize
     const len = Math.hypot(direction[0], direction[1], direction[2]);
@@ -500,6 +506,13 @@ class SimulatedSkybox {
     }
   }
 
+
+
+  setExposure(exposure) {
+    this.exposure = exposure;
+    this.updateCoefficients();
+  }
+
   updateCoefficients() {
     if (!this.materialInstance) {
       console.warn("updateCoefficients called before material loaded");
@@ -585,6 +598,7 @@ class SimulatedSkybox {
     this.materialInstance.setFloat4Parameter('cloudControl2', new Float32Array(this.cloudControl2));
     this.materialInstance.setFloat4Parameter('waterControl', new Float32Array(this.waterControl));
     this.materialInstance.setFloat4Parameter('starControl', new Float32Array(this.starControl));
+    this.materialInstance.setFloatParameter('starIntensity', this.starIntensity);
 
     this.materialInstance.setFloatParameter('sunIntensity', physicalSunIntensity);
 
@@ -615,12 +629,13 @@ class SimulatedSkybox {
 
     this.materialInstance.setFloatParameter('sunIntensity2', finalMoonIntensity);
 
-    // Scale Milky Way by Sun Intensity (Pre-exposed) to match dynamic range
-    // Calibration: 1.0 User Intensity ~ 0.025 Lux Relative (approx 2.5% of Sun Pixel Value at Day)
-    // At Sunny 16 (Sun ~ 2.6), this gives ~0.065 pixel brightness, which is visible but dim.
+    // Scale Milky Way by Sun Intensity
+    // Calibration: 1.0 User Intensity = 1.5e-3 cd/m^2 (Nits) approx.
+    // Target: 1.5e-3 Nits. SunIntensity = 100,000.
+    // Scale = 1.5e-3 / 1.0e5 = 1.5e-8.
     const mwIntensity = this.milkyWayEnabled ? this.milkyWayControl[0] : 0.0;
     const mwUniform = [
-      mwIntensity * this.sunIntensity * 0.025,
+      mwIntensity * this.sunIntensity * 1.5e-8,
       this.milkyWayControl[1],
       this.milkyWayControl[2]
     ];
@@ -634,6 +649,7 @@ class SimulatedSkybox {
     const moonHaloUpload = [...this.moonHalo];
     moonHaloUpload[2] *= moonRadConv;
     this.materialInstance.setFloat4Parameter('sunHalo2', new Float32Array(moonHaloUpload));
+    this.materialInstance.setFloatParameter('exposure', this.exposure !== undefined ? this.exposure : 1.0);
 
     // Solar Eclipse (CPU Calculation)
     const sunRadius = Math.acos(this.sunHalo[0]);

docs_src/src_raw/wip/sky/main.js

@@ -218,6 +218,7 @@ class App {
     const exposure = this.getExposure();
     const preExposedIntensity = this.params.sunIntensity * exposure;
     this.skybox.setSunIntensity(preExposedIntensity);
+    this.skybox.setExposure(exposure); // Update Skybox Exposure Uniform
 
     // Moon Exposure
     if (this.mParams) {
@@ -389,7 +390,7 @@ class App {
     };
 
     mwFolder.add(this.mwParams, 'enabled').name('Enabled').onChange(updateMW);
-    mwFolder.add(this.mwParams, 'intensity', 0.0, 5.0).onChange(updateMW);
+    mwFolder.add(this.mwParams, 'intensity', 0.0, 100.0).onChange(updateMW);
     mwFolder.add(this.mwParams, 'saturation', 0.0, 2.0).onChange(updateMW);
     mwFolder.add(this.mwParams, 'blackPoint', 0.0, 0.5).name('Black Point').onChange(updateMW);
     mwFolder.add(this.mwParams, 'siderealTime', 0.0, 24.0).name('Sidereal Time').listen().onChange(updateMW);
@@ -464,18 +465,23 @@ class App {
     const starFolder = gui.addFolder('Stars');
     this.sParams = {
       enabled: true,
-      density: 0.001
+      density: 0.001,
+      intensity: 0.0 // 2^0 = 1.0
     };
     // Initialize defaults (Density 0.001, Enabled True)
     sky.setStarControl(0.001, true);
+    sky.setStarIntensity(Math.pow(2.0, 0.0));
 
     const updateStars = () => {
       sky.setStarControl(this.sParams.density, this.sParams.enabled);
+      sky.setStarIntensity(Math.pow(2.0, this.sParams.intensity));
     };
 
     starFolder.add(this.sParams, 'enabled').name('Enabled').onChange(updateStars);
     starFolder.add(this.sParams, 'density', 0.0, 0.01, 0.0001).name('Density').onChange(updateStars);
+    starFolder.add(this.sParams, 'intensity', 0.0, 24.0).name('Intensity (Exp)').onChange(updateStars);
     starFolder.close();
+    this.updateStars = updateStars;
 
     const artFolder = gui.addFolder('Artistic');
     // Set Horizon Glow default to 0.0
@@ -489,7 +495,7 @@ class App {
     artFolder.add(sky.msFactors, 2, 0.0, 1.0).name('Horizon Glow').onChange(v => sky.setHorizonGlow(v));
     artFolder.add(sky, 'contrast', 0.1, 2.0).onChange(v => sky.setContrast(v));
 
-    artFolder.addColor(sky, 'nightColor').onChange(v => sky.setNightColor(v));
+
 
     const shmFolder = artFolder.addFolder('Shimmer');
     // Set Shimmer Strength default to 0.0
@@ -502,7 +508,7 @@ class App {
     const camFolder = gui.addFolder('Camera');
     camFolder.add(this.params, 'focalLength', 8.0, 300.0).name('Focal Length').onChange(() => this.updateCameraProjection());
     camFolder.add(this.params, 'aperture', 1.4, 32.0).onChange(() => this.updateCameraExposure());
-    camFolder.add(this.params, 'shutterSpeed', 1.0, 1000.0).onChange(() => this.updateCameraExposure());
+    camFolder.add(this.params, 'shutterSpeed', 0.05, 1000.0).onChange(() => this.updateCameraExposure());
     camFolder.add(this.params, 'iso', 50.0, 3200.0).onChange(() => this.updateCameraExposure());
 
     const bloomFolder = camFolder.addFolder('Bloom');
@@ -692,13 +698,15 @@ class App {
     const s = this.sParams;
     const b = this.bParams;
     const m = this.mParams;
+    const mw = this.mwParams;
     const sk = this.skybox;
 
     return {
       p: { a: p.aperture, ss: p.shutterSpeed, i: p.iso, st: p.sunTheta, sp: p.sunPhi, fl: p.focalLength, si: p.sunIntensity },
       c: { v: c.volumetrics, co: c.coverage, d: c.density, h: c.height, s: c.speed, e: c.evolution },
       w: { dt: w.derivativeTrick, st: w.strength, s: w.speed, o: w.octaves },
-      s: { e: s.enabled, d: s.density },
+      s: { e: s.enabled, d: s.density, i: s.intensity },
+      mw: { e: mw.enabled, i: mw.intensity, s: mw.saturation, bp: mw.blackPoint, st: mw.siderealTime, l: mw.latitude },
       b: { e: b.enabled, lf: b.lensFlare },
       m: { e: m.enabled, az: m.azimuth, h: m.height, r: m.radius, i: m.intensity },
       cm: { t: this.camState.theta, p: this.camState.phi },
@@ -723,6 +731,7 @@ class App {
     const c = state.c;
     const w = state.w;
     const s = state.s;
+    const mw = state.mw;
     const b = state.b;
     const m = state.m;
     const k = state.k;
@@ -757,6 +766,18 @@ class App {
     if (s) {
       if (s.e !== undefined) this.sParams.enabled = s.e;
       if (s.d !== undefined) this.sParams.density = s.d;
+      if (s.i !== undefined) this.sParams.intensity = s.i;
+      if (this.updateStars) this.updateStars();
+    }
+
+    if (mw) {
+      if (mw.e !== undefined) this.mwParams.enabled = mw.e;
+      if (mw.i !== undefined) this.mwParams.intensity = mw.i;
+      if (mw.s !== undefined) this.mwParams.saturation = mw.s;
+      if (mw.bp !== undefined) this.mwParams.blackPoint = mw.bp;
+      if (mw.st !== undefined) this.mwParams.siderealTime = mw.st;
+      if (mw.l !== undefined) this.mwParams.latitude = mw.l;
+      if (this.updateMW) this.updateMW();
     }
 
     if (b) {

docs_src/src_raw/wip/sky/simulated_skybox.mat

@@ -96,10 +96,20 @@ material {
             name : starControl, // x=Density, y=Enabled, z=Frequency, w=PixelScale
             precision : high
         },
+        {
+            type : float,
+            name : starIntensity,
+            precision : high
+        },
         {
             type : sampler2d,
             name : moonTexture
         },
+        {
+            type : float,
+            name : exposure,
+            precision : high
+        },
         {
             type : sampler2d,
             name : moonNormal
@@ -162,9 +172,9 @@ fragment {
     // --- CONFIGURATION ---
 
     // Stars
-    #define STAR_GLOBAL_INTENSITY 150.0      // Master brightness multiplier [0.0 - 500.0]
-    #define STAR_BRIGHTNESS_BASE 0.5         // Minimum random brightness [0.0 - 1.0]
-    #define STAR_BRIGHTNESS_VAR 4.0          // Random brightness variance range [0.0 - 10.0]
+    #define STAR_GLOBAL_INTENSITY 100.0      // Master brightness multiplier [0.0 - 500.0]
+    #define STAR_BRIGHTNESS_BASE 0.01        // Minimum random brightness [0.0 - 1.0]
+    #define STAR_BRIGHTNESS_VAR 15.0         // Random brightness variance range [0.0 - 10.0]
     #define STAR_FADE_SUN_ELV_HIGH 0.10      // Sun elevation (sin) where stars are fully hidden [0.0 - 0.5]
     #define STAR_FADE_SUN_ELV_LOW -0.20      // Sun elevation (sin) where stars are fully visible [-0.5 - 0.0]
     #define STAR_CLOUD_OCCLUSION 0.1         // Visibility when covered by clouds [0.0 - 1.0]
@@ -961,7 +971,11 @@ fragment {
                              highp vec4 sunHalo, highp vec4 cloudControl, highp vec4 cloudControl2, 
                              highp vec4 shimmerControl, highp vec4 waterControl,
                              highp vec3 L2, highp float sunIntensity2, highp vec4 sunHalo2,
-                             sampler2D moonTex, sampler2D moonNormal) {
+                             sampler2D moonTex, sampler2D moonNormal,
+                             highp vec3 nightColor,
+                             highp mat3 milkyWayRotation,
+                             highp vec3 milkyWayControl,
+                             sampler2D milkyWayTexture) {
 
         // Project to plane y=0
         highp float t = WATER_PLANE_HEIGHT / min(V.y, -0.0002); // Reduced clamp to minimize "wall" artifact
@@ -1057,7 +1071,7 @@ fragment {
          highp float rHorizonMask = 1.0 - smoothstep(0.0, REFLECTION_HORIZON_FADE, R.y);
         
         if (rHorizonMask > 0.0) {
-             reflection += getStarLayer(R, L, reflCloudDensity, outTransmittance, materialParams.starControl) * rHorizonMask;
+             reflection += getStarLayer(R, L, reflCloudDensity, outTransmittance, materialParams.starControl) * rHorizonMask * materialParams.exposure;
         }
 
         // Sun Disk Reflection
@@ -1088,6 +1102,29 @@ fragment {
         // Apply clouds to reflection
         reflection = mix(reflection, reflCloudLayer, reflCloudDensity);
                 
+        // Add Milky Way to Reflection
+        // Calculate Fade based on Sun Elevation
+        highp float sunElvSin = L.y;
+        highp float mwFade = smoothstep(STAR_FADE_SUN_ELV_HIGH, STAR_FADE_SUN_ELV_LOW, sunElvSin);
+        
+        if (mwFade > 0.0) {
+                 highp vec3 mwColor = getMilkyWay(R, milkyWayRotation, milkyWayTexture, milkyWayControl);
+                 
+                 // Apply Atmosphere Transmittance (approximate)
+                 mwColor *= outTransmittance;
+                 
+                 // Apply Fades
+                 mwColor *= mwFade;
+                 mwColor *= (1.0 - reflMoonOcclusion); // Occlude by Moon
+                 mwColor *= (1.0 - reflCloudDensity);  // Occlude by Clouds
+                 
+                 reflection += mwColor;
+        }
+                
+
+        // Add Night Color to Reflection
+        reflection += nightColor;
+                
         // Fresnel
         highp float F0 = WATER_FRESNEL_F0; // Water
         highp float cosTheta = clamp(dot(-V, N_water), 0.0, 1.0);
@@ -1187,7 +1224,7 @@ fragment {
 
         // 7. Stars
         // Add stars before clouds (clouds cover stars)
-        highp vec3 starColor = getStarLayer(V, L, cloudDensityVal, transmittance, materialParams.starControl);
+        highp vec3 starColor = getStarLayer(V, L, cloudDensityVal, transmittance, materialParams.starControl) * materialParams.exposure * materialParams.starIntensity;
         
         // 7b. Milky Way
         // Add Milky Way behind stars (conceptually) but handled similarly
@@ -1232,7 +1269,11 @@ fragment {
                                        materialParams.shimmerControl,
                                        materialParams.waterControl,
                                        L2, materialParams.sunIntensity2, materialParams.sunHalo2,
-                                       materialParams_moonTexture, materialParams_moonNormal);
+                                       materialParams_moonTexture, materialParams_moonNormal,
+                                       materialParams.nightColor,
+                                       materialParams.milkyWayRotation,
+                                       materialParams.milkyWayControl,
+                                       materialParams_milkyWayTexture);
             color = applyDynamicToneMapping(color, L, materialParams.contrast);
         }
 

filament/CMakeLists.txt

@@ -93,6 +93,7 @@ set(SRCS
         src/MorphTargetBuffer.cpp
         src/PostProcessManager.cpp
         src/ProgramSpecialization.cpp
+        src/LocalProgramCache.cpp
         src/RenderPass.cpp
         src/RenderPrimitive.cpp
         src/RenderTarget.cpp
@@ -186,6 +187,7 @@ set(PRIVATE_HDRS
         src/MaterialCache.h
         src/MaterialDefinition.h
         src/MaterialParser.h
+        src/LocalProgramCache.h
         src/MaterialInstanceManager.h
         src/PIDController.h
         src/PostProcessManager.h

filament/backend/CMakeLists.txt

@@ -564,169 +564,170 @@ endif()
 # ==================================================================================================
 # Test
 # ==================================================================================================
-option(INSTALL_BACKEND_TEST "Install the backend test library so it can be consumed on iOS" OFF)
+if (FILAMENT_BUILD_TESTING)
+    option(INSTALL_BACKEND_TEST "Install the backend test library so it can be consumed on iOS" OFF)
 
-if (APPLE OR LINUX)
-    set(BACKEND_TEST_SRC
-        test/BackendTest.cpp
-        test/ShaderGenerator.cpp
-        test/TrianglePrimitive.cpp
-        test/Arguments.cpp
-        test/ImageExpectations.cpp
-        test/Lifetimes.cpp
-        test/PlatformRunner.cpp
-        test/Shader.cpp
-        test/SharedShaders.cpp
-        test/Skip.cpp
-        test/test_FeedbackLoops.cpp
-        test/test_Blit.cpp
-        test/test_MissingRequiredAttributes.cpp
-        test/test_ReadPixels.cpp
-        test/test_ReadTexture.cpp
-        test/test_BufferUpdates.cpp
-        test/test_Callbacks.cpp
-        test/test_JobQueue.cpp
-        test/test_MemoryMappedBuffer.cpp
-        test/test_MsaaSwapChain.cpp
-        test/test_MRT.cpp
-        test/test_PushConstants.cpp
-        test/test_LoadImage.cpp
-        test/test_StencilBuffer.cpp
-        test/test_Scissor.cpp
-        test/test_MipLevels.cpp
-        test/test_Handles.cpp
-        test/test_CircularBuffer.cpp
-        test/test_CommandBufferQueue.cpp
-        test/test_Template.cpp
-    )
-    if (FILAMENT_SUPPORTS_WEBGPU)
-        list(APPEND BACKEND_TEST_SRC
-                test/test_WebGPUAsyncTaskCounter.cpp
-                test/test_WebGPUComposeSwizzle.cpp)
-    endif()
-    if (APPLE)
-        # Metal-specific tests
-        list(APPEND BACKEND_TEST_SRC
-            test/test_MetalBlitter.mm
+    if (APPLE OR LINUX)
+        set(BACKEND_TEST_SRC
+            test/BackendTest.cpp
+            test/ShaderGenerator.cpp
+            test/TrianglePrimitive.cpp
+            test/Arguments.cpp
+            test/ImageExpectations.cpp
+            test/Lifetimes.cpp
+            test/PlatformRunner.cpp
+            test/Shader.cpp
+            test/SharedShaders.cpp
+            test/Skip.cpp
+            test/test_Autoresolve.cpp
+            test/test_FeedbackLoops.cpp
+            test/test_Blit.cpp
+            test/test_MissingRequiredAttributes.cpp
+            test/test_ReadPixels.cpp
+            test/test_ReadTexture.cpp
+            test/test_BufferUpdates.cpp
+            test/test_Callbacks.cpp
+            test/test_JobQueue.cpp
+            test/test_MemoryMappedBuffer.cpp
+            test/test_MsaaSwapChain.cpp
+            test/test_MRT.cpp
+            test/test_PushConstants.cpp
+            test/test_LoadImage.cpp
+            test/test_StencilBuffer.cpp
+            test/test_Scissor.cpp
+            test/test_MipLevels.cpp
+            test/test_Handles.cpp
+            test/test_CircularBuffer.cpp
+            test/test_CommandBufferQueue.cpp
+            test/test_Template.cpp
         )
-    endif()
-    set(BACKEND_TEST_LIBS
-        absl::str_format
-        backend
-        getopt
-        gtest
-        imageio
-        filamat
-        SPIRV
-        spirv-cross-glsl)
-    # Create input/output directories for test result images.
-    file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/images/actual_images)
-    file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/images/diff_images)
-    file(COPY test/expected_images DESTINATION ${CMAKE_CURRENT_BINARY_DIR}/images)
-endif()
-
-# TODO: Disabling IOS test due to breakage wrt glslang update
-if (APPLE AND NOT IOS)
-    # TODO: we should expand this test to Linux and other platforms.
-    list(APPEND BACKEND_TEST_SRC
-         test/test_RenderExternalImage.cpp)
-    add_library(backend_test STATIC ${BACKEND_TEST_SRC})
-    target_link_libraries(backend_test PUBLIC ${BACKEND_TEST_LIBS})
-    target_compile_options(backend_test PRIVATE "-fobjc-arc")
-
-    set(BACKEND_TEST_DEPS
-            OSDependent
-            SPIRV
-            SPIRV-Tools
-            SPIRV-Tools-opt
-            backend_test
+        if (FILAMENT_SUPPORTS_WEBGPU)
+            list(APPEND BACKEND_TEST_SRC
+                    test/test_WebGPUAsyncTaskCounter.cpp
+                    test/test_WebGPUComposeSwizzle.cpp)
+        endif()
+        if (APPLE)
+            # Metal-specific tests
+            list(APPEND BACKEND_TEST_SRC
+                test/test_MetalBlitter.mm
+            )
+        endif()
+        set(BACKEND_TEST_LIBS
+            absl::str_format
+            backend
             getopt
             gtest
-            glslang
-            spirv-cross-core
-            spirv-cross-glsl
-            spirv-cross-msl)
-
-    if (NOT IOS)
-        target_link_libraries(backend_test PRIVATE image imageio)
-        list(APPEND BACKEND_TEST_DEPS image)
-    endif()
-    if (FILAMENT_SUPPORTS_WEBGPU)
-        target_link_libraries(backend_test PRIVATE webgpu_dawn dawncpp_headers)
-        list(APPEND BACKEND_TEST_DEPS webgpu_dawn dawncpp_headers)
+            imageio
+            filamat
+            SPIRV
+            spirv-cross-glsl)
+        # Create input/output directories for test result images.
+        file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/images/actual_images)
+        file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/images/diff_images)
+        file(COPY test/expected_images DESTINATION ${CMAKE_CURRENT_BINARY_DIR}/images)
     endif()
 
-    set(BACKEND_TEST_COMBINED_OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/libbackendtest_combined.a")
-    combine_static_libs(backend_test "${BACKEND_TEST_COMBINED_OUTPUT}" "${BACKEND_TEST_DEPS}")
+    # TODO: Disabling IOS test due to breakage wrt glslang update
+    if (APPLE AND NOT IOS)
+        # TODO: we should expand this test to Linux and other platforms.
+        list(APPEND BACKEND_TEST_SRC
+            test/test_RenderExternalImage.cpp)
+        add_library(backend_test STATIC ${BACKEND_TEST_SRC})
+        target_link_libraries(backend_test PUBLIC ${BACKEND_TEST_LIBS})
+        target_compile_options(backend_test PRIVATE "-fobjc-arc")
 
-    set(BACKEND_TEST_LIB_NAME ${CMAKE_STATIC_LIBRARY_PREFIX}backend_test${CMAKE_STATIC_LIBRARY_SUFFIX})
+        set(BACKEND_TEST_DEPS
+                OSDependent
+                SPIRV
+                SPIRV-Tools
+                SPIRV-Tools-opt
+                backend_test
+                getopt
+                gtest
+                glslang
+                spirv-cross-core
+                spirv-cross-glsl
+                spirv-cross-msl)
 
-    if (INSTALL_BACKEND_TEST)
-        install(FILES "${BACKEND_TEST_COMBINED_OUTPUT}" DESTINATION lib/${DIST_DIR} RENAME ${BACKEND_TEST_LIB_NAME})
-        install(FILES test/PlatformRunner.h DESTINATION include/backend_test)
+        if (NOT IOS)
+            target_link_libraries(backend_test PRIVATE image imageio)
+            list(APPEND BACKEND_TEST_DEPS image)
+        endif()
+        if (FILAMENT_SUPPORTS_WEBGPU)
+            target_link_libraries(backend_test PRIVATE webgpu_dawn dawncpp_headers)
+            list(APPEND BACKEND_TEST_DEPS webgpu_dawn dawncpp_headers)
+        endif()
+
+        set(BACKEND_TEST_COMBINED_OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/libbackendtest_combined.a")
+        combine_static_libs(backend_test "${BACKEND_TEST_COMBINED_OUTPUT}" "${BACKEND_TEST_DEPS}")
+
+        set(BACKEND_TEST_LIB_NAME ${CMAKE_STATIC_LIBRARY_PREFIX}backend_test${CMAKE_STATIC_LIBRARY_SUFFIX})
+
+        if (INSTALL_BACKEND_TEST)
+            install(FILES "${BACKEND_TEST_COMBINED_OUTPUT}" DESTINATION lib/${DIST_DIR} RENAME ${BACKEND_TEST_LIB_NAME})
+            install(FILES test/PlatformRunner.h DESTINATION include/backend_test)
+        endif()
+
+        set_target_properties(backend_test PROPERTIES FOLDER Tests)
+
+        if (APPLE AND NOT IOS)
+            add_executable(backend_test_mac test/mac_runner.mm)
+            target_link_libraries(backend_test_mac PRIVATE "-framework Metal -framework AppKit -framework QuartzCore")
+            # Because each test case is a separate file, the -force_load flag is necessary to prevent the
+            # linker from removing "unused" symbols.
+            target_link_libraries(backend_test_mac PRIVATE -force_load backend_test)
+            set_target_properties(backend_test_mac PROPERTIES FOLDER Tests)
+
+            # This is needed after XCode 15.3
+            set_target_properties(backend_test_mac PROPERTIES BUILD_WITH_INSTALL_RPATH TRUE)
+            set_target_properties(backend_test_mac PROPERTIES INSTALL_RPATH /usr/local/lib)
+        endif()
     endif()
 
-    set_target_properties(backend_test PROPERTIES FOLDER Tests)
+    if (LINUX)
+        add_executable(backend_test_linux test/linux_runner.cpp ${BACKEND_TEST_SRC})
+        target_link_libraries(backend_test_linux PRIVATE ${BACKEND_TEST_LIBS})
+        set_target_properties(backend_test_linux PROPERTIES FOLDER Tests)
+        if (FILAMENT_SUPPORTS_WEBGPU)
+            target_link_libraries(backend_test_linux PRIVATE webgpu_dawn dawncpp_headers)
+        endif()
+    endif()
+
+    # ==================================================================================================
+    # Compute tests
+    #
+    #if (NOT IOS AND NOT WEBGL)
+    #
+    #add_executable(compute_test
+    #        test/ComputeTest.cpp
+    #        test/Arguments.cpp
+    #        test/test_ComputeBasic.cpp
+    #        )
+    #
+    #target_link_libraries(compute_test PRIVATE
+    #        backend
+    #        getopt
+    #        gtest
+    #        )
+    #
+    #set_target_properties(compute_test PROPERTIES FOLDER Tests)
+    #
+    #endif()
+
+    # ==================================================================================================
+    # Metal utils tests
 
     if (APPLE AND NOT IOS)
-       add_executable(backend_test_mac test/mac_runner.mm)
-       target_link_libraries(backend_test_mac PRIVATE "-framework Metal -framework AppKit -framework QuartzCore")
-       # Because each test case is a separate file, the -force_load flag is necessary to prevent the
-       # linker from removing "unused" symbols.
-       target_link_libraries(backend_test_mac PRIVATE -force_load backend_test)
-       set_target_properties(backend_test_mac PROPERTIES FOLDER Tests)
+        add_executable(metal_utils_test test/MetalTest.mm)
 
-       # This is needed after XCode 15.3
-       set_target_properties(backend_test_mac PROPERTIES BUILD_WITH_INSTALL_RPATH TRUE)
-       set_target_properties(backend_test_mac PROPERTIES INSTALL_RPATH /usr/local/lib)
+        target_compile_options(metal_utils_test PRIVATE "-fobjc-arc")
+
+        target_link_libraries(metal_utils_test PRIVATE
+                backend
+                getopt
+                gtest
+                )
+
+        set_target_properties(metal_utils_test PROPERTIES FOLDER Tests)
     endif()
-endif()
-
-if (LINUX)
-    add_executable(backend_test_linux test/linux_runner.cpp ${BACKEND_TEST_SRC})
-    target_link_libraries(backend_test_linux PRIVATE ${BACKEND_TEST_LIBS})
-    set_target_properties(backend_test_linux PROPERTIES FOLDER Tests)
-    if (FILAMENT_SUPPORTS_WEBGPU)
-        target_link_libraries(backend_test_linux PRIVATE webgpu_dawn dawncpp_headers)
-    endif()
-endif()
-
-# ==================================================================================================
-# Compute tests
-#
-#if (NOT IOS AND NOT WEBGL)
-#
-#add_executable(compute_test
-#        test/ComputeTest.cpp
-#        test/Arguments.cpp
-#        test/test_ComputeBasic.cpp
-#        )
-#
-#target_link_libraries(compute_test PRIVATE
-#        backend
-#        getopt
-#        gtest
-#        )
-#
-#set_target_properties(compute_test PROPERTIES FOLDER Tests)
-#
-#endif()
-
-# ==================================================================================================
-# Metal utils tests
-
-if (APPLE AND NOT IOS)
-
-add_executable(metal_utils_test test/MetalTest.mm)
-
-target_compile_options(metal_utils_test PRIVATE "-fobjc-arc")
-
-target_link_libraries(metal_utils_test PRIVATE
-        backend
-        getopt
-        gtest
-        )
-
-set_target_properties(metal_utils_test PROPERTIES FOLDER Tests)
-
-endif()
+endif()

filament/backend/src/VirtualMachineEnv.cpp

@@ -25,6 +25,11 @@
 
 #include <mutex>
 
+#ifdef __ANDROID__
+#include <pthread.h>
+#endif
+
+
 namespace filament {
 
 using namespace utils;
@@ -105,8 +110,21 @@ JNIEnv* VirtualMachineEnv::getEnvironmentSlow() {
     FILAMENT_CHECK_PRECONDITION(mVirtualMachine)
             << "JNI_OnLoad() has not been called";
 
-#if defined(__ANDROID__)
-    jint const result = mVirtualMachine->AttachCurrentThread(&mJniEnv, nullptr);
+#ifdef __ANDROID__
+    JavaVMAttachArgs args;
+    args.version = JNI_VERSION_1_6;
+    args.group = nullptr;
+    char threadName[16]; // pthread_getname_np returns at most 16 bytes
+    if (__builtin_available(android 26, *)) {
+        if (pthread_getname_np(pthread_self(), threadName, sizeof(threadName)) == 0) {
+            args.name = threadName;
+        } else {
+            args.name = nullptr;
+        }
+    } else {
+        args.name = nullptr;
+    }
+    jint const result = mVirtualMachine->AttachCurrentThread(&mJniEnv, &args);
 #else
     jint const result = mVirtualMachine->AttachCurrentThread(reinterpret_cast<void**>(&mJniEnv), nullptr);
 #endif

filament/backend/src/metal/MetalHandles.mm

@@ -1092,7 +1092,7 @@ MetalRenderTarget::MetalRenderTarget(MetalContext* context, uint32_t width, uint
         // a multisampled sidecar texture and do a resolve automatically.
         if (samples > 1 && texture->samples == 1) {
             auto& sidecar = texture->msaaSidecar;
-            if (!sidecar) {
+            if (!sidecar || sidecar.sampleCount != samples) {
                 sidecar = createMultisampledTexture(mtlTexture.pixelFormat, texture->width,
                         texture->height, samples);
             }
@@ -1123,7 +1123,7 @@ MetalRenderTarget::MetalRenderTarget(MetalContext* context, uint32_t width, uint
         // a multisampled sidecar texture and do a resolve automatically.
         if (samples > 1 && texture->samples == 1) {
             auto& sidecar = texture->msaaSidecar;
-            if (!sidecar) {
+            if (!sidecar || sidecar.sampleCount != samples) {
                 sidecar = createMultisampledTexture(mtlTexture.pixelFormat, texture->width,
                         texture->height, samples);
             }
@@ -1155,7 +1155,7 @@ MetalRenderTarget::MetalRenderTarget(MetalContext* context, uint32_t width, uint
         // a multisampled sidecar texture and do a resolve automatically.
         if (samples > 1 && texture->samples == 1) {
             auto& sidecar = texture->msaaSidecar;
-            if (!sidecar) {
+            if (!sidecar || sidecar.sampleCount != samples) {
                 sidecar = createMultisampledTexture(mtlTexture.pixelFormat, texture->width,
                         texture->height, samples);
             }

filament/backend/src/opengl/OpenGLContext.h

@@ -91,7 +91,7 @@ public:
         GLenum getIndicesType() const noexcept {
             return indicesType;
         }
-    } gl;
+    };
 
     static bool queryOpenGLVersion(GLint* major, GLint* minor) noexcept;
 

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

@@ -120,33 +120,27 @@ bool PlatformEGL::isOpenGL() const noexcept {
 PlatformEGL::ExternalImageEGL::~ExternalImageEGL() = default;
 
 Driver* PlatformEGL::createDriver(void* sharedContext, const DriverConfig& driverConfig) {
-    static constexpr int kMaxNumEGLDevices = 32;
+    mEGLDisplay = eglGetDisplay(EGL_DEFAULT_DISPLAY);
+    assert_invariant(mEGLDisplay != EGL_NO_DISPLAY);
 
     EGLint major, minor;
-    EGLBoolean initialized = false;
+    EGLBoolean initialized = eglInitialize(mEGLDisplay, &major, &minor);
 
-    PFNEGLQUERYDEVICESEXTPROC const eglQueryDevicesEXT =
-        PFNEGLQUERYDEVICESEXTPROC(eglGetProcAddress("eglQueryDevicesEXT"));
-    PFNEGLGETPLATFORMDISPLAYEXTPROC const getPlatformDisplay =
-        PFNEGLGETPLATFORMDISPLAYEXTPROC(eglGetProcAddress("eglGetPlatformDisplay"));
-
-    if (eglQueryDevicesEXT != nullptr && getPlatformDisplay != nullptr) {
-        EGLint numDevices = 0;
-        EGLDeviceEXT eglDevices[kMaxNumEGLDevices];
-        if (eglQueryDevicesEXT(kMaxNumEGLDevices, eglDevices, &numDevices)) {
-            for (int i = 0; i < numDevices && !initialized; ++i) {
-                mEGLDisplay = getPlatformDisplay(EGL_PLATFORM_DEVICE_EXT, eglDevices[i], nullptr);
+    if (!initialized) {
+        EGLDeviceEXT eglDevice;
+        EGLint numDevices;
+        PFNEGLQUERYDEVICESEXTPROC const eglQueryDevicesEXT =
+                PFNEGLQUERYDEVICESEXTPROC(eglGetProcAddress("eglQueryDevicesEXT"));
+        if (eglQueryDevicesEXT != nullptr) {
+            eglQueryDevicesEXT(1, &eglDevice, &numDevices);
+            if(auto* getPlatformDisplay = reinterpret_cast<PFNEGLGETPLATFORMDISPLAYEXTPROC>(
+                    eglGetProcAddress("eglGetPlatformDisplay"))) {
+                mEGLDisplay = getPlatformDisplay(EGL_PLATFORM_DEVICE_EXT, eglDevice, nullptr);
                 initialized = eglInitialize(mEGLDisplay, &major, &minor);
             }
         }
     }
 
-    if (!initialized) {
-        mEGLDisplay = eglGetDisplay(EGL_DEFAULT_DISPLAY);
-        assert_invariant(mEGLDisplay != EGL_NO_DISPLAY);
-        initialized = eglInitialize(mEGLDisplay, &major, &minor);
-    }
-
     if (UTILS_UNLIKELY(!initialized)) {
         LOG(ERROR) << "eglInitialize failed";
         return nullptr;

filament/backend/src/vulkan/VulkanAsyncHandles.cpp

@@ -29,6 +29,12 @@ using namespace bluevk;
 
 namespace filament::backend {
 
+std::shared_ptr<VulkanCmdFence> VulkanCmdFence::completed() noexcept {
+    auto cmdFence = std::make_shared<VulkanCmdFence>(VK_NULL_HANDLE);
+    cmdFence->mStatus = VK_SUCCESS;
+    return cmdFence;
+}
+
 FenceStatus VulkanCmdFence::wait(VkDevice device, uint64_t const timeout,
     std::chrono::steady_clock::time_point const until) {
 

filament/backend/src/vulkan/VulkanAsyncHandles.h

@@ -40,6 +40,13 @@ struct VulkanCmdFence {
     explicit VulkanCmdFence(VkFence fence) : mFence(fence) { }
     ~VulkanCmdFence() = default;
 
+    // Creates a VulkanCmdFence with its status set to VK_SUCCESS. It holds
+    // a null handle; it is assumed that any user of this object will avoid
+    // using the fence handle directly if getStatus() returns VK_SUCCESS, as
+    // in that case, it's likely the fence is being reused for other passes,
+    // and is not in the expected state anyway.
+    static std::shared_ptr<VulkanCmdFence> completed() noexcept;
+
     void setStatus(VkResult const value) {
         std::lock_guard const l(mLock);
         mStatus = value;

filament/backend/src/vulkan/VulkanCommands.h

@@ -277,7 +277,11 @@ private:
     fvkmemory::resource_ptr<VulkanSemaphore> mLastSubmit;
 
     VkFence mLastFence = VK_NULL_HANDLE;
-    std::shared_ptr<VulkanCmdFence> mLastFenceStatus;
+    // Start out with a completed fence, because if no commands have
+    // been queued or submited, then by definition, all pending work
+    // is complete.
+    std::shared_ptr<VulkanCmdFence> mLastFenceStatus =
+            VulkanCmdFence::completed();
 
     VkPipelineStageFlags mInjectedDependencyWaitStage = 0;
 };

filament/backend/src/vulkan/VulkanContext.h

@@ -36,6 +36,7 @@ VK_DEFINE_HANDLE(VmaPool)
 namespace filament::backend {
 
 struct VulkanCommandBuffer;
+struct VulkanRenderPass;
 struct VulkanRenderTarget;
 struct VulkanSwapChain;
 struct VulkanTexture;
@@ -58,11 +59,11 @@ struct VulkanAttachment {
     VkImageSubresourceRange getSubresourceRange() const;
 };
 
-struct VulkanRenderPass {
+struct VulkanRenderPassContext {
     // Between the begin and end command render pass we cache the command buffer
     VulkanCommandBuffer* commandBuffer;
     fvkmemory::resource_ptr<VulkanRenderTarget> renderTarget;
-    VkRenderPass renderPass;
+    fvkmemory::resource_ptr<VulkanRenderPass> renderPass;
     RenderPassParams params;
     int currentSubpass;
 };

filament/backend/src/vulkan/VulkanDescriptorSetCache.cpp

@@ -355,35 +355,8 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
 }
 
 void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture, VkSampler sampler,
-        VkDescriptorSetLayout externalSamplerLayout) noexcept {
-
-    // We have to update a bound set for two use cases
-    //   - streaming API (a changing feed of AHardwareBuffer)
-    //   - external samplers - potential changing of dataspace per-frame
-
-    // TODO: Fix the stream flow case base on the above comment!!
-    if (set->isAnExternalSamplerBound) {
-        auto layout = set->getLayout();
-        // Build a new descriptor set from the new layout
-        VkDescriptorSetLayout const genLayout = set->boundLayout;
-        VkDescriptorSet const newSet = getVkSet(layout->count, genLayout);
-        Bitmask const ubo = layout->bitmask.ubo | layout->bitmask.dynamicUbo;
-        Bitmask samplers = layout->bitmask.sampler;
-        samplers.unset(binding);
-
-        // Each bitmask denotes a binding index, and separated into two stages - vertex and buffer
-        // We fold the two stages into just the lower half of the bits to denote a combined set of
-        // bindings.
-        Bitmask const copyBindings = foldBitsInHalf(ubo | samplers);
-        VkDescriptorSet const srcSet = set->getVkSet();
-        copySet(srcSet, newSet, copyBindings);
-        set->addNewSet(newSet,
-                [this, layoutCount = layout->count, genLayout, newSet](VulkanDescriptorSet*) {
-                    this->manualRecycle(layoutCount, genLayout, newSet);
-                });
-    }
-
+        uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
+        VkSampler sampler, VkDescriptorSetLayout externalSamplerLayout) noexcept {
     VkDescriptorSet const vkset = set->getVkSet();
     VkImageSubresourceRange range = texture->getPrimaryViewRange();
     VkImageViewType const expectedType = texture->getViewType();
@@ -432,6 +405,28 @@ fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetCache::createSet
     return set;
 }
 
+void VulkanDescriptorSetCache::cloneSet(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+        fvkutils::SamplerBitmask samplerMask) noexcept {
+    auto const& layout = set->getLayout();
+    // Build a new descriptor set from layout
+    VkDescriptorSetLayout const genLayout = set->boundLayout;
+    VkDescriptorSet const newSet = getVkSet(layout->count, genLayout);
+
+    // Each bitmask denotes a binding index, and separated into two stages - vertex and buffer
+    // We fold the two stages into just the lower half of the bits to denote a combined set of
+    // bindings.
+    Bitmask const ubo = layout->bitmask.ubo | layout->bitmask.dynamicUbo;
+    // Don't copy the samplers in the mask
+    Bitmask samplers = layout->bitmask.sampler ^ samplerMask;
+    Bitmask const copyBindings = foldBitsInHalf(ubo | samplers);
+
+    VkDescriptorSet const srcSet = set->getVkSet();
+    copySet(srcSet, newSet, copyBindings);
+    set->addNewSet(newSet,
+            [this, layoutCount = layout->count, genLayout, newSet](
+                    VulkanDescriptorSet*) { this->manualRecycle(layoutCount, genLayout, newSet); });
+}
+
 VkDescriptorSet VulkanDescriptorSetCache::getVkSet(DescriptorCount const& count,
         VkDescriptorSetLayout vklayout) {
     return mDescriptorPool->obtainSet(count, vklayout);

filament/backend/src/vulkan/VulkanDescriptorSetCache.h

@@ -76,6 +76,11 @@ public:
     fvkmemory::resource_ptr<VulkanDescriptorSet> createSet(Handle<HwDescriptorSet> handle,
             fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
 
+    // Create and set as current a new VkDescriptorSet using the `set` currently bound layout and
+    // copy all the bindings and ignoring the samplers bindings in the `samplerMask`.
+    void cloneSet(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
+            fvkutils::SamplerBitmask samplerMask) noexcept;
+
     // This method is meant to be used with external samplers
     VkDescriptorSet getVkSet(DescriptorCount const& count, VkDescriptorSetLayout vklayout);
 

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -1913,9 +1913,6 @@ void VulkanDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassP
 
     auto rt = resource_ptr<VulkanRenderTarget>::cast(&mResourceManager, rth);
 
-    VulkanCommandBuffer* commandBuffer = rt->isProtected() ?
-           &mCommands.getProtected() : &mCommands.get();
-
     // Filament has the expectation that the contents of the swap chain are not preserved on the
     // first render pass. Note however that its contents are often preserved on subsequent render
     // passes, due to multiple views.
@@ -1930,6 +1927,11 @@ void VulkanDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassP
         }
     }
 
+    // Note that this needs to come after the acquireNextswapchainImage() above because that path
+    // might flush the current command buffer.
+    VulkanCommandBuffer* commandBuffer =
+            rt->isProtected() ? &mCommands.getProtected() : &mCommands.get();
+
     // Note that retrieving the extent must come after the acquireNextSwapchainImage() above;
     // otherwise it might be 0.
     VkExtent2D const extent = rt->getExtent();
@@ -1975,17 +1977,19 @@ void VulkanDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassP
     rpkey.initialDepthLayout = currentDepthLayout;
     rpkey.subpassMask = uint8_t(params.subpassMask);
 
-    VkRenderPass renderPass = mFramebufferCache.getRenderPass(rpkey);
+    fvkmemory::resource_ptr<VulkanRenderPass> renderPass =
+            mFramebufferCache.getRenderPass(rpkey, &mResourceManager);
     mPipelineCache.bindRenderPass(renderPass, 0);
 
     // Create the VkFramebuffer or fetch it from cache.
     VulkanFboCache::FboKey fbkey = rt->getFboKey();
-    fbkey.renderPass = renderPass;
+    fbkey.renderPass = renderPass->getVkRenderPass();
     fbkey.layers = 1;
 
     rt->emitBarriersBeginRenderPass(*commandBuffer);
 
-    VkFramebuffer vkfb = mFramebufferCache.getFramebuffer(fbkey);
+    fvkmemory::resource_ptr<VulkanFramebuffer> vkfb =
+            mFramebufferCache.getFramebuffer(fbkey, &mResourceManager, rt);
 
 // Assign a label to the framebuffer for debugging purposes.
 #if FVK_ENABLED(FVK_DEBUG_GROUP_MARKERS | FVK_DEBUG_DEBUG_UTILS)
@@ -1998,12 +2002,14 @@ void VulkanDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassP
 
     // The current command buffer now has references to the render target and its attachments.
     commandBuffer->acquire(rt);
+    commandBuffer->acquire(renderPass);
+    commandBuffer->acquire(vkfb);
 
     // Populate the structures required for vkCmdBeginRenderPass.
     VkRenderPassBeginInfo renderPassInfo {
         .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
-        .renderPass = renderPass,
-        .framebuffer = vkfb,
+        .renderPass = renderPass->getVkRenderPass(),
+        .framebuffer = vkfb->getVkFramebuffer(),
 
         // The renderArea field constrains the LoadOp, but scissoring does not.
         // Therefore, we do not set the scissor rect here, we only need it in draw().
@@ -2058,7 +2064,7 @@ void VulkanDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassP
     mCurrentRenderPass = {
         .commandBuffer = commandBuffer,
         .renderTarget = rt,
-        .renderPass = renderPassInfo.renderPass,
+        .renderPass = renderPass,
         .params = params,
         .currentSubpass = 0,
     };
@@ -2078,8 +2084,7 @@ void VulkanDriver::endRenderPass(int) {
     rt->emitBarriersEndRenderPass(*mCurrentRenderPass.commandBuffer);
 
     mCurrentRenderPass.renderTarget = {};
-    mCurrentRenderPass.renderPass = VK_NULL_HANDLE;
-
+    mCurrentRenderPass.renderPass = {};
     mCurrentRenderPass.commandBuffer = nullptr;
 }
 
@@ -2244,7 +2249,7 @@ void VulkanDriver::resolve(
         Handle<HwTexture> src, uint8_t dstLevel, uint8_t dstLayer) {
     FVK_SYSTRACE_SCOPE();
 
-    FILAMENT_CHECK_PRECONDITION(mCurrentRenderPass.renderPass == VK_NULL_HANDLE)
+    FILAMENT_CHECK_PRECONDITION(!mCurrentRenderPass.renderPass)
             << "resolve() cannot be invoked inside a render pass.";
 
     auto srcTexture = resource_ptr<VulkanTexture>::cast(&mResourceManager, src);
@@ -2287,7 +2292,7 @@ void VulkanDriver::blit(
         math::uint2 size) {
     FVK_SYSTRACE_SCOPE();
 
-    FILAMENT_CHECK_PRECONDITION(mCurrentRenderPass.renderPass == VK_NULL_HANDLE)
+    FILAMENT_CHECK_PRECONDITION(!mCurrentRenderPass.renderPass)
             << "blit() cannot be invoked inside a render pass.";
 
     auto srcTexture = resource_ptr<VulkanTexture>::cast(&mResourceManager, src);
@@ -2329,7 +2334,7 @@ void VulkanDriver::blitDEPRECATED(TargetBufferFlags buffers,
 
     // Note: blitDEPRECATED is only used for Renderer::copyFrame()
 
-    FILAMENT_CHECK_PRECONDITION(mCurrentRenderPass.renderPass == VK_NULL_HANDLE)
+    FILAMENT_CHECK_PRECONDITION(!mCurrentRenderPass.renderPass)
             << "blitDEPRECATED() cannot be invoked inside a render pass.";
 
     FILAMENT_CHECK_PRECONDITION(buffers == TargetBufferFlags::COLOR0)

filament/backend/src/vulkan/VulkanDriver.h

@@ -139,7 +139,7 @@ private:
 
     resource_ptr<VulkanSwapChain> mCurrentSwapChain;
     resource_ptr<VulkanRenderTarget> mDefaultRenderTarget;
-    VulkanRenderPass mCurrentRenderPass = {};
+    VulkanRenderPassContext mCurrentRenderPass = {};
     VmaAllocator mAllocator = VK_NULL_HANDLE;
     VkDebugReportCallbackEXT mDebugCallback = VK_NULL_HANDLE;
 

filament/backend/src/vulkan/VulkanExternalImageManager.cpp

@@ -94,6 +94,9 @@ void VulkanExternalImageManager::updateSetAndLayout(
     fvkmemory::resource_ptr<VulkanDescriptorSetLayout> const& layout = set->getLayout();
     set->boundLayout = mDescriptorSetLayoutCache->getVkLayout(layout->bitmask,
             actualExternalSamplers, outSamplers);
+
+    mDescriptorSetCache->cloneSet(set, actualExternalSamplers);
+
     // Update the external samplers in the set
     for (auto& [binding, sampler, image]: samplerAndBindings) {
         // We cannot call updateSamplerForExternalSamplerSet because some samplers are non NULL

filament/backend/src/vulkan/VulkanFboCache.cpp

@@ -17,6 +17,7 @@
 #include "VulkanFboCache.h"
 
 #include "VulkanConstants.h"
+#include "VulkanHandles.h"
 #include "vulkan/utils/Image.h"
 
 #include <utils/Panic.h>
@@ -69,9 +70,11 @@ VulkanFboCache::~VulkanFboCache() {
             << "Please explicitly call terminate() while the VkDevice is still alive.";
 }
 
-VkFramebuffer VulkanFboCache::getFramebuffer(FboKey const& config) noexcept {
+fvkmemory::resource_ptr<VulkanFramebuffer> VulkanFboCache::getFramebuffer(FboKey const& config,
+        fvkmemory::ResourceManager* resManager,
+        fvkmemory::resource_ptr<VulkanRenderTarget> renderTarget) noexcept {
     FboMap::iterator iter = mFramebufferCache.find(config);
-    if (UTILS_LIKELY(iter != mFramebufferCache.end() && iter->second.handle != VK_NULL_HANDLE)) {
+    if (UTILS_LIKELY(iter != mFramebufferCache.end())) {
         iter.value().timestamp = mCurrentTime;
         return iter->second.handle;
     }
@@ -117,13 +120,17 @@ VkFramebuffer VulkanFboCache::getFramebuffer(FboKey const& config) noexcept {
     VkResult error = vkCreateFramebuffer(mDevice, &info, VKALLOC, &framebuffer);
     FILAMENT_CHECK_POSTCONDITION(error == VK_SUCCESS) << "Unable to create framebuffer."
                                                      << " error=" << static_cast<int32_t>(error);
-    mFramebufferCache[config] = {framebuffer, mCurrentTime};
-    return framebuffer;
+    fvkmemory::resource_ptr<VulkanFramebuffer> fbh =
+            fvkmemory::resource_ptr<VulkanFramebuffer>::construct(resManager, mDevice, framebuffer,
+                    renderTarget);
+    mFramebufferCache[config] = { fbh, mCurrentTime };
+    return fbh;
 }
 
-VkRenderPass VulkanFboCache::getRenderPass(RenderPassKey const& config) noexcept {
+fvkmemory::resource_ptr<VulkanRenderPass> VulkanFboCache::getRenderPass(
+        RenderPassKey const& config, fvkmemory::ResourceManager* resManager) noexcept {
     auto iter = mRenderPassCache.find(config);
-    if (UTILS_LIKELY(iter != mRenderPassCache.end() && iter->second.handle != VK_NULL_HANDLE)) {
+    if (UTILS_LIKELY(iter != mRenderPassCache.end())) {
         iter.value().timestamp = mCurrentTime;
         return iter->second.handle;
     }
@@ -326,7 +333,9 @@ VkRenderPass VulkanFboCache::getRenderPass(RenderPassKey const& config) noexcept
     VkResult error = vkCreateRenderPass(mDevice, &renderPassInfo, VKALLOC, &renderPass);
     FILAMENT_CHECK_POSTCONDITION(error == VK_SUCCESS) << "Unable to create render pass."
                                                       << " error=" << error;
-    mRenderPassCache[config] = {renderPass, mCurrentTime};
+    fvkmemory::resource_ptr<VulkanRenderPass> rph =
+        fvkmemory::resource_ptr<VulkanRenderPass>::construct(resManager, mDevice, renderPass);
+    mRenderPassCache[config] = {rph, mCurrentTime};
 
 #if FVK_ENABLED(FVK_DEBUG_FBO_CACHE)
     FVK_LOGD << "Created render pass " << renderPass << " with ";
@@ -343,13 +352,12 @@ VkRenderPass VulkanFboCache::getRenderPass(RenderPassKey const& config) noexcept
              << "colorAttachmentCount[0] = " << subpasses[0].colorAttachmentCount;
 #endif
 
-    return renderPass;
+    return rph;
 }
 
 void VulkanFboCache::resetFramebuffers() noexcept {
     for (const auto& pair: mFramebufferCache) {
         mRenderPassRefCount[pair.first.renderPass]--;
-        vkDestroyFramebuffer(mDevice, pair.second.handle, VKALLOC);
     }
     mFramebufferCache.clear();
 }
@@ -357,9 +365,6 @@ void VulkanFboCache::resetFramebuffers() noexcept {
 void VulkanFboCache::terminate() noexcept {
     resetFramebuffers();
 
-    for (const auto& pair: mRenderPassCache) {
-        vkDestroyRenderPass(mDevice, pair.second.handle, VKALLOC);
-    }
     mRenderPassRefCount.clear();
     mRenderPassCache.clear();
 }
@@ -380,8 +385,6 @@ void VulkanFboCache::gc() noexcept {
         const FboVal fbo = iter->second;
         if (fbo.timestamp < evictTime && fbo.handle) {
             mRenderPassRefCount[iter->first.renderPass]--;
-            vkDestroyFramebuffer(mDevice, fbo.handle, VKALLOC);
-            iter.value().handle = VK_NULL_HANDLE;
 
             // erase(iterator) returns the iterator to the next element.
             iter = mFramebufferCache.erase(iter);
@@ -391,11 +394,8 @@ void VulkanFboCache::gc() noexcept {
     }
 
     for (RenderPassMap::iterator iter = mRenderPassCache.begin(); iter != mRenderPassCache.end(); ) {
-        const VkRenderPass handle = iter->second.handle;
+        const VkRenderPass handle = iter->second.handle->getVkRenderPass();
         if (iter->second.timestamp < evictTime && handle && mRenderPassRefCount[handle] == 0) {
-            vkDestroyRenderPass(mDevice, handle, VKALLOC);
-            iter.value().handle = VK_NULL_HANDLE;
-
             // erase(iterator) returns the iterator to the next element.
             iter = mRenderPassCache.erase(iter);
             mRenderPassRefCount.erase(handle);

filament/backend/src/vulkan/VulkanFboCache.h

@@ -18,6 +18,9 @@
 #define TNT_FILAMENT_BACKEND_VULKANFBOCACHE_H
 
 #include "VulkanContext.h"
+#include "vulkan/memory/Resource.h"
+#include "vulkan/memory/ResourceManager.h"
+#include "vulkan/memory/ResourcePointer.h"
 
 #include <utils/Hash.h>
 
@@ -27,6 +30,9 @@
 
 namespace filament::backend {
 
+struct VulkanFramebuffer;
+struct VulkanRenderPass;
+
 // Simple manager for VkFramebuffer and VkRenderPass objects.
 //
 // Note that a VkFramebuffer is just a binding between a render pass and a set of image views. So,
@@ -57,7 +63,7 @@ public:
         uint8_t padding[2];
     };
     struct RenderPassVal {
-        VkRenderPass handle;
+        fvkmemory::resource_ptr<VulkanRenderPass> handle;
         uint32_t timestamp;
     };
     static_assert(0 == MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT % 8);
@@ -83,7 +89,7 @@ public:
         VkImageView depth; // 8 bytes
     };
     struct FboVal {
-        VkFramebuffer handle;
+        fvkmemory::resource_ptr<VulkanFramebuffer> handle;
         uint32_t timestamp;
     };
     static_assert(sizeof(VkRenderPass) == 8, "VkRenderPass has unexpected size.");
@@ -98,10 +104,13 @@ public:
     ~VulkanFboCache();
 
     // Retrieves or creates a VkFramebuffer handle.
-    VkFramebuffer getFramebuffer(FboKey const& config) noexcept;
+    fvkmemory::resource_ptr<VulkanFramebuffer> getFramebuffer(FboKey const& config,
+            fvkmemory::ResourceManager* resManager,
+            fvkmemory::resource_ptr<VulkanRenderTarget> renderTarget) noexcept;
 
     // Retrieves or creates a VkRenderPass handle.
-    VkRenderPass getRenderPass(RenderPassKey const& config) noexcept;
+    fvkmemory::resource_ptr<VulkanRenderPass> getRenderPass(
+        RenderPassKey const& config, fvkmemory::ResourceManager* resManager) noexcept;
 
     // Evicts old unused Vulkan objects. Call this once per frame.
     void gc() noexcept;

filament/backend/src/vulkan/VulkanHandles.cpp

@@ -565,7 +565,7 @@ void VulkanRenderTarget::transformViewportToPlatform(VkViewport* bounds) const {
     flipVertically(bounds, getExtent().height);
 }
 
-uint8_t VulkanRenderTarget::getColorTargetCount(const VulkanRenderPass& pass) const {
+uint8_t VulkanRenderTarget::getColorTargetCount(const VulkanRenderPassContext& pass) const {
     if (!mOffscreen) {
         return 1;
     }
@@ -715,4 +715,21 @@ VulkanRenderPrimitive::VulkanRenderPrimitive(PrimitiveType pt,
       vertexBuffer(vb),
       indexBuffer(ib) {}
 
+VulkanFramebuffer::VulkanFramebuffer(VkDevice device, VkFramebuffer framebuffer,
+        fvkmemory::resource_ptr<VulkanRenderTarget> renderTarget)
+        : mDevice(device),
+          mFramebuffer(framebuffer),
+          mRenderTarget(renderTarget) {}
+
+VulkanFramebuffer::~VulkanFramebuffer() {
+    vkDestroyFramebuffer(mDevice, mFramebuffer, VKALLOC);
+}
+
+VulkanRenderPass::VulkanRenderPass(VkDevice device, VkRenderPass renderPass)
+    : mDevice(device), mRenderPass(renderPass) {}
+
+VulkanRenderPass::~VulkanRenderPass() {
+    vkDestroyRenderPass(mDevice, mRenderPass, VKALLOC);
+}
+
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanHandles.h

@@ -195,7 +195,7 @@ public:
     }
 
     // The current layout used by the descriptor set. This one will match the bindings, including
-    // external samplers data. 
+    // external samplers data.
     // This will not necessarilly be the same as `mLayout`.
     VkDescriptorSetLayout boundLayout = VK_NULL_HANDLE;
 
@@ -203,7 +203,7 @@ public:
 
     fvkutils::UniformBufferBitmask const& dynamicUboMask;
     uint8_t const uniqueDynamicUboCount;
-    
+
     // Flag to indicate if the current layout needs to be recreated or not.
     // This should only set to `true` when a external sampler image is bound to the descriptor set.
     bool isLayoutDirty = false;
@@ -419,7 +419,7 @@ struct VulkanRenderTarget : private HwRenderTarget, fvkmemory::Resource {
         return mInfo->fbkey.samples;
     }
 
-    uint8_t getColorTargetCount(VulkanRenderPass const& pass) const;
+    uint8_t getColorTargetCount(VulkanRenderPassContext const& pass) const;
 
     inline bool hasDepth() const { return mInfo->depthIndex != Auxiliary::UNDEFINED_INDEX; }
 
@@ -585,6 +585,39 @@ struct VulkanRenderPrimitive : public HwRenderPrimitive, fvkmemory::Resource {
     fvkmemory::resource_ptr<VulkanIndexBuffer> indexBuffer;
 };
 
+struct VulkanFramebuffer : public fvkmemory::Resource {
+    VulkanFramebuffer(VkDevice device,
+            VkFramebuffer framebuffer,
+            fvkmemory::resource_ptr<VulkanRenderTarget> renderTarget);
+    ~VulkanFramebuffer();
+
+    inline VkFramebuffer getVkFramebuffer() const noexcept {
+        return mFramebuffer;
+    }
+
+private:
+    VkDevice mDevice;
+    VkFramebuffer mFramebuffer;
+
+    // We need to keep a reference to the renderTarget because the key of the framebuffer in the
+    // cache has references to the image views that are derived from the textures of the render
+    // target.
+    fvkmemory::resource_ptr<VulkanRenderTarget> mRenderTarget;
+};
+
+struct VulkanRenderPass : public fvkmemory::Resource {
+    VulkanRenderPass(VkDevice device, VkRenderPass renderPass);
+    ~VulkanRenderPass();
+
+    inline VkRenderPass getVkRenderPass() const noexcept {
+        return mRenderPass;
+    }
+
+private:
+    VkDevice mDevice;
+    VkRenderPass mRenderPass;
+};
+
 } // namespace filament::backend
 
 #endif // TNT_FILAMENT_BACKEND_VULKANHANDLES_H

filament/backend/src/vulkan/VulkanPipelineCache.cpp

@@ -459,8 +459,10 @@ void VulkanPipelineCache::bindStencilState(StencilState const& stencilState) noe
     mPipelineRequirements.stencilState = stencilState;
 }
 
-void VulkanPipelineCache::bindRenderPass(VkRenderPass renderPass, int subpassIndex) noexcept {
-    mPipelineRequirements.renderPass = renderPass;
+void VulkanPipelineCache::bindRenderPass(
+        fvkmemory::resource_ptr<VulkanRenderPass> renderPass,
+        int subpassIndex) noexcept {
+    mPipelineRequirements.renderPass = renderPass->getVkRenderPass();
     mPipelineRequirements.subpassIndex = subpassIndex;
 }
 

filament/backend/src/vulkan/VulkanPipelineCache.h

@@ -123,7 +123,8 @@ public:
     void bindProgram(fvkmemory::resource_ptr<VulkanProgram> program) noexcept;
     void bindRasterState(RasterState const& rasterState) noexcept;
     void bindStencilState(StencilState const& stencilState) noexcept;
-    void bindRenderPass(VkRenderPass renderPass, int subpassIndex) noexcept;
+    void bindRenderPass(fvkmemory::resource_ptr<VulkanRenderPass> renderPass,
+            int subpassIndex) noexcept;
     void bindPrimitiveTopology(VkPrimitiveTopology topology) noexcept;
     void bindVertexArray(VkVertexInputAttributeDescription const* attribDesc,
             VkVertexInputBindingDescription const* bufferDesc, uint8_t count);

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

@@ -42,6 +42,8 @@ template ResourceType getTypeEnum<VulkanSync>() noexcept;
 template ResourceType getTypeEnum<VulkanMemoryMappedBuffer>() noexcept;
 template ResourceType getTypeEnum<VulkanSemaphore>() noexcept;
 template ResourceType getTypeEnum<VulkanStream>() noexcept;
+template ResourceType getTypeEnum<VulkanFramebuffer>() noexcept;
+template ResourceType getTypeEnum<VulkanRenderPass>() noexcept;
 
 template<typename D>
 ResourceType getTypeEnum() noexcept {
@@ -108,6 +110,12 @@ ResourceType getTypeEnum() noexcept {
     if constexpr (std::is_same_v<D, VulkanStream>) {
         return ResourceType::STREAM;
     }
+    if constexpr (std::is_same_v<D, VulkanFramebuffer>) {
+        return ResourceType::FRAMEBUFFER;
+    }
+    if constexpr (std::is_same_v<D, VulkanRenderPass>) {
+        return ResourceType::RENDER_PASS;
+    }
     return ResourceType::UNDEFINED_TYPE;
 }
 
@@ -155,6 +163,10 @@ std::string_view getTypeStr(ResourceType type) {
             return "Semaphore";
         case ResourceType::STREAM:
             return "VulkanStream";
+        case ResourceType::FRAMEBUFFER:
+            return "Framebuffer";
+        case ResourceType::RENDER_PASS:
+            return "RenderPass";
         case ResourceType::UNDEFINED_TYPE:
             return "";
     }

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

@@ -56,7 +56,9 @@ enum class ResourceType : uint8_t {
     MEMORY_MAPPED_BUFFER = 18,
     SEMAPHORE = 19,
     STREAM = 20,
-    UNDEFINED_TYPE = 21,    // Must be the last enum because we use it for iterating over the enums.
+    FRAMEBUFFER = 21,
+    RENDER_PASS = 22,
+    UNDEFINED_TYPE = 23,    // Must be the last enum because we use it for iterating over the enums.
 };
 
 template<typename D>

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

@@ -126,6 +126,12 @@ void ResourceManager::destroyWithType(ResourceType type, HandleId id) {
         case ResourceType::STREAM:
             destruct<VulkanStream>(Handle<VulkanStream>(id));
             break;
+        case ResourceType::FRAMEBUFFER:
+            destruct<VulkanFramebuffer>(Handle<VulkanFramebuffer>(id));
+            break;
+        case ResourceType::RENDER_PASS:
+            destruct<VulkanRenderPass>(Handle<VulkanRenderPass>(id));
+            break;
         case ResourceType::UNDEFINED_TYPE:
             break;
     }

filament/backend/test/test_Autoresolve.cpp

@@ -0,0 +1,120 @@
+/*
+ * 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 "BackendTest.h"
+
+#include "Shader.h"
+#include "SharedShaders.h"
+#include "Skip.h"
+#include "TrianglePrimitive.h"
+
+#include <backend/PixelBufferDescriptor.h>
+
+#include <utils/FixedCapacityVector.h>
+
+namespace test {
+
+using namespace filament;
+using namespace filament::backend;
+using namespace filament::math;
+
+TEST_F(BackendTest, AutoresolveDifferingSampleCounts) {
+    SKIP_IF(SkipEnvironment(OperatingSystem::CI, Backend::OPENGL), "see b/486954356");
+    SKIP_IF(SkipEnvironment(OperatingSystem::CI, Backend::VULKAN), "see b/486954356");
+
+    auto& api = getDriverApi();
+    constexpr int kRenderTargetSize = 512;
+
+    auto swapChain = addCleanup(createSwapChain());
+    api.makeCurrent(swapChain, swapChain);
+
+    Shader shader = SharedShaders::makeShader(api, *mCleanup,
+            {
+                .mVertexType = VertexShaderType::Simple,
+                .mFragmentType = FragmentShaderType::SolidColored,
+                .mUniformType = ShaderUniformType::Simple,
+            });
+
+    TrianglePrimitive triangle(api);
+    PipelineState ps = getColorWritePipelineState();
+    shader.addProgramToPipelineState(ps);
+
+    auto ubuffer = addCleanup(api.createBufferObject(sizeof(SimpleMaterialParams),
+            BufferObjectBinding::UNIFORM, BufferUsage::STATIC));
+    shader.bindUniform<SimpleMaterialParams>(api, ubuffer);
+
+    // Create a texture with sample count = 1.
+    Handle<HwTexture> texture = addCleanup(api.createTexture(SamplerType::SAMPLER_2D, 1,
+            TextureFormat::RGBA8, 1, kRenderTargetSize, kRenderTargetSize, 1,
+            TextureUsage::COLOR_ATTACHMENT | TextureUsage::SAMPLEABLE));
+
+    // First render pass: render a red triangle to a RenderTarget with 8 samples.
+    {
+        Handle<HwRenderTarget> renderTarget =
+                addCleanup(api.createRenderTarget(TargetBufferFlags::COLOR, kRenderTargetSize,
+                        kRenderTargetSize, 8, 1, { { texture } }, {}, {}));
+
+        shader.uploadUniform(api, ubuffer,
+                SimpleMaterialParams{
+                    .color = float4(1, 0, 0, 1),
+                });
+
+        RenderPassParams params = getClearColorRenderPass(float4(0));
+        params.viewport = { 0, 0, kRenderTargetSize, kRenderTargetSize };
+
+        RenderFrame frame(api);
+        api.beginRenderPass(renderTarget, params);
+        ps.primitiveType = PrimitiveType::TRIANGLES;
+        ps.vertexBufferInfo = triangle.getVertexBufferInfo();
+        api.bindPipeline(ps);
+        api.bindRenderPrimitive(triangle.getRenderPrimitive());
+        api.draw2(0, 3, 1);
+        api.endRenderPass();
+        api.commit(swapChain);
+    }
+
+    // Second render pass: render a green triangle to a RenderTarget with 4 samples, attached to
+    // the same texture.
+    {
+        Handle<HwRenderTarget> renderTarget =
+                addCleanup(api.createRenderTarget(TargetBufferFlags::COLOR, kRenderTargetSize,
+                        kRenderTargetSize, 4, 1, { { texture } }, {}, {}));
+
+        shader.uploadUniform(api, ubuffer,
+                SimpleMaterialParams{
+                    .color = float4(0, 1, 0, 1),
+                });
+
+        RenderPassParams params = getClearColorRenderPass(float4(0));
+        params.viewport = { 0, 0, kRenderTargetSize, kRenderTargetSize };
+
+        RenderFrame frame(api);
+        api.beginRenderPass(renderTarget, params);
+        ps.primitiveType = PrimitiveType::TRIANGLES;
+        ps.vertexBufferInfo = triangle.getVertexBufferInfo();
+        api.bindPipeline(ps);
+        api.bindRenderPrimitive(triangle.getRenderPrimitive());
+        api.draw2(0, 3, 1);
+        api.endRenderPass();
+
+        EXPECT_IMAGE(renderTarget, ScreenshotParams(kRenderTargetSize, kRenderTargetSize,
+                                           "AutoresolveDifferingSampleCounts", 1048576));
+
+        api.commit(swapChain);
+    }
+}
+
+} // namespace test

filament/include/filament/LightManager.h

@@ -372,7 +372,7 @@ public:
         } vsm;
 
         /**
-         * Light bulb radius used for soft shadows. Currently this is only used when DPCF or PCSS is
+         * Light bulb radius used for soft shadows. Currently, this is only used when DPCF or PCSS is
          * enabled. (2cm by default).
          */
         float shadowBulbRadius = 0.02f;

filament/include/filament/MaterialInstance.h

@@ -94,6 +94,12 @@ public:
             std::is_same_v<math::mat3f, T>
     >;
 
+    template<typename T>
+    using is_supported_constant_parameter_t = std::enable_if_t<
+            std::is_same_v<int32_t, T> ||
+            std::is_same_v<float, T> ||
+            std::is_same_v<bool, T>>;
+
     /**
      * Creates a new MaterialInstance using another MaterialInstance as a template for initialization.
      * The new MaterialInstance is an instance of the same Material of the template instance and
@@ -276,6 +282,91 @@ public:
         return getParameter<T>(name, strlen(name));
     }
 
+    /**
+     * Overrides a specialization constant of this material instance.
+     *
+     * @tparam T    The type of the constant. Must be int32_t, float, or bool.
+     * @param name  The name of the constant as defined in the material. Cannot be nullptr.
+     * @param nameLength Length in `char` of the name parameter.
+     * @param value The value of the constant.
+     *
+     * @see Material::Builder::constant
+     */
+    template<typename T, typename = is_supported_constant_parameter_t<T>>
+    void setConstant(const char* UTILS_NONNULL name, size_t nameLength, T value);
+
+    /** inline helper to provide the name as a null-terminated string literal */
+    template<typename T, typename = is_supported_constant_parameter_t<T>>
+    void setConstant(StringLiteral const name, T value) {
+        setConstant<T>(name.data, name.size, value);
+    }
+
+    /** inline helper to provide the name as a null-terminated C string */
+    template<typename T, typename = is_supported_constant_parameter_t<T>>
+    void setConstant(const char* UTILS_NONNULL name, T value) {
+        setConstant<T>(name, strlen(name), value);
+    }
+
+    /**
+     * Gets the value of a specialization constant by name.
+     *
+     * @tparam T    The type of the constant. Must be int32_t, float, or bool.
+     * @param name  The name of the constant as defined in the material. Cannot be nullptr.
+     * @param nameLength Length in `char` of the name parameter.
+     * @return The value of the constant.
+     */
+    template<typename T, typename = is_supported_constant_parameter_t<T>>
+    T getConstant(const char* UTILS_NONNULL name, size_t nameLength) const;
+
+    /** inline helper to provide the name as a null-terminated C string */
+    template<typename T, typename = is_supported_constant_parameter_t<T>>
+    T getConstant(StringLiteral const name) const {
+        return getConstant<T>(name.data, name.size);
+    }
+
+    /** inline helper to provide the name as a null-terminated C string */
+    template<typename T, typename = is_supported_constant_parameter_t<T>>
+    T getConstant(const char* UTILS_NONNULL name) const {
+        return getConstant<T>(name, strlen(name));
+    }
+
+    using CompilerPriorityQueue = backend::CompilerPriorityQueue;
+
+    /**
+     * Asynchronously ensures that a subset of this MaterialInstance's variants are compiled.
+     *
+     * This function behaves identically to Material::compile(), but takes into account the
+     * specific constants overridden by setConstant().
+     *
+     * @param priority      Which priority queue to use, LOW or HIGH.
+     * @param variants      Variants to include to the compile command.
+     * @param handler       Handler to dispatch the callback or nullptr for the default handler
+     * @param callback      callback called on the main thread when the compilation is done on
+     *                      by backend.
+     *
+     * @see Material::compile
+     * @see setConstant
+     */
+    void compile(CompilerPriorityQueue priority,
+            UserVariantFilterMask variants,
+            backend::CallbackHandler* UTILS_NULLABLE handler = nullptr,
+            utils::Invocable<void(MaterialInstance* UTILS_NONNULL)>&& callback = {}) noexcept;
+
+    inline void compile(CompilerPriorityQueue priority,
+            UserVariantFilterBit variants,
+            backend::CallbackHandler* UTILS_NULLABLE handler = nullptr,
+            utils::Invocable<void(MaterialInstance* UTILS_NONNULL)>&& callback = {}) noexcept {
+        compile(priority, UserVariantFilterMask(variants), handler,
+                std::forward<utils::Invocable<void(MaterialInstance* UTILS_NONNULL)>>(callback));
+    }
+
+    inline void compile(CompilerPriorityQueue priority,
+            backend::CallbackHandler* UTILS_NULLABLE handler = nullptr,
+            utils::Invocable<void(MaterialInstance* UTILS_NONNULL)>&& callback = {}) noexcept {
+        compile(priority, UserVariantFilterBit::ALL, handler,
+                std::forward<utils::Invocable<void(MaterialInstance* UTILS_NONNULL)>>(callback));
+    }
+
     /**
      * Set-up a custom scissor rectangle; by default it is disabled.
      *

filament/include/filament/Options.h

@@ -735,7 +735,7 @@ struct VsmShadowOptions {
     bool highPrecision = false;
 
     /**
-     * VSM minimum variance scale, must be positive.
+     * @deprecated has no effect.
      */
     float minVarianceScale = 0.5f;
 

filament/src/LocalProgramCache.cpp

@@ -0,0 +1,268 @@
+/*
+ * Copyright (C) 2026 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 "LocalProgramCache.h"
+#include "MaterialParser.h"
+
+#include "details/Engine.h"
+#include "details/Material.h"
+
+#include <backend/DriverApiForward.h>
+
+namespace filament {
+
+using namespace backend;
+using namespace utils;
+
+LocalProgramCache::LocalProgramCache(LocalProgramCache const& other)
+        : mMaterial(other.mMaterial),
+          mCachedPrograms(other.mCachedPrograms.size()),
+          mSpecializationConstants((other.mMaterial != nullptr)
+                                           ? other.mMaterial->getEngine()
+                                                     .getMaterialCache()
+                                                     .getSpecializationConstantsInternPool()
+                                                     .acquire(other.mSpecializationConstants)
+                                           : SpecializationConstants()) {}
+
+LocalProgramCache& LocalProgramCache::operator=(LocalProgramCache const& other) {
+    assert_invariant(mMaterial == nullptr);
+    assert_invariant(mCachedPrograms.empty());
+    assert_invariant(mSpecializationConstants.empty());
+
+    mMaterial = other.mMaterial;
+    if (mMaterial != nullptr) {
+        mCachedPrograms = FixedCapacityVector<Handle<HwProgram>>(other.mCachedPrograms.size());
+        mSpecializationConstants = other.mMaterial->getEngine()
+                .getMaterialCache()
+                .getSpecializationConstantsInternPool()
+                .acquire(other.mSpecializationConstants);
+    }
+
+    return *this;
+}
+
+void LocalProgramCache::initializeForMaterial(FEngine& engine, FMaterial const& material,
+        utils::FixedCapacityVector<backend::Program::SpecializationConstant>
+                specializationConstants) {
+    assert_invariant(mMaterial == nullptr);
+    assert_invariant(mCachedPrograms.empty());
+    assert_invariant(mSpecializationConstants.empty());
+
+    mMaterial = &material;
+
+    mSpecializationConstants =
+            engine.getMaterialCache().getSpecializationConstantsInternPool().acquire(
+                    std::move(specializationConstants));
+
+    size_t cachedProgramsSize;
+    switch (material.getMaterialDomain()) {
+        case filament::MaterialDomain::SURFACE:
+            cachedProgramsSize = 1 << VARIANT_BITS;
+            break;
+        case filament::MaterialDomain::POST_PROCESS:
+            cachedProgramsSize = 1 << POST_PROCESS_VARIANT_BITS;
+            break;
+        case filament::MaterialDomain::COMPUTE:
+            cachedProgramsSize = 1;
+            break;
+    }
+    mCachedPrograms = FixedCapacityVector<Handle<HwProgram>>(cachedProgramsSize);
+
+    material.getDefinition().acquirePrograms(engine, mCachedPrograms.as_slice(),
+            material.getMaterialParser(), mSpecializationConstants, material.isDefaultMaterial());
+}
+
+void LocalProgramCache::initializeForMaterialInstance(FEngine& engine, FMaterial const& material) {
+    assert_invariant(mMaterial == nullptr);
+    assert_invariant(mCachedPrograms.empty());
+    assert_invariant(mSpecializationConstants.empty());
+
+    mMaterial = &material;
+    LocalProgramCache const& programs = material.getPrograms();
+
+    mSpecializationConstants =
+            engine.getMaterialCache().getSpecializationConstantsInternPool().acquire(
+                    programs.getSpecializationConstants());
+
+    mCachedPrograms =
+            FixedCapacityVector<Handle<HwProgram>>(material.getPrograms().mCachedPrograms.size());
+}
+
+Handle<HwProgram> LocalProgramCache::prepareProgramSlow(DriverApi& driver, Variant const variant,
+        CompilerPriorityQueue const priorityQueue) const noexcept {
+    assert_invariant(mMaterial != nullptr);
+
+    FEngine& engine = mMaterial->getEngine();
+    if (mMaterial->isSharedVariant(variant)) {
+        FMaterial const* defaultMaterial = engine.getDefaultMaterial();
+        assert_invariant(defaultMaterial);
+        LocalProgramCache const& defaultPrograms = defaultMaterial->getPrograms();
+        Handle<HwProgram> program = defaultPrograms.mCachedPrograms[variant.key];
+        if (program) {
+            return mCachedPrograms[variant.key] = program;
+        }
+        return mCachedPrograms[variant.key] =
+                defaultPrograms.prepareProgram(driver, variant, priorityQueue);
+    }
+    return mCachedPrograms[variant.key] = mMaterial->getDefinition().prepareProgram(engine, driver,
+                   mMaterial->getMaterialParser(), getProgramSpecialization(variant), priorityQueue);
+}
+
+ProgramSpecialization LocalProgramCache::getProgramSpecialization(Variant variant) const noexcept {
+    assert_invariant(mMaterial != nullptr);
+
+    return ProgramSpecialization {
+        .materialCrc32 = mMaterial->getMaterialParser().getCrc32(),
+        .variant = variant,
+        .specializationConstants = mSpecializationConstants,
+    };
+}
+
+void LocalProgramCache::terminate(FEngine& engine) {
+    assert_invariant(mMaterial != nullptr);
+
+    mMaterial->getDefinition().releasePrograms(engine, mCachedPrograms.as_slice(),
+            mMaterial->getMaterialParser(), mSpecializationConstants,
+            mMaterial->isDefaultMaterial());
+    engine.getMaterialCache().releaseMaterial(engine, mMaterial->getDefinition());
+    engine.getMaterialCache().getSpecializationConstantsInternPool().release(
+            mSpecializationConstants);
+}
+
+void LocalProgramCache::clear(FEngine& engine) {
+    assert_invariant(mMaterial != nullptr);
+
+    mMaterial->getDefinition().releasePrograms(engine, mCachedPrograms.as_slice(),
+            mMaterial->getMaterialParser(), mSpecializationConstants,
+            mMaterial->isDefaultMaterial());
+}
+
+Variant LocalProgramCache::filterVariantForGetProgram(Variant variant) const noexcept {
+    if (UTILS_UNLIKELY(mMaterial->getEngine().features.material.enable_fog_as_postprocess)) {
+        // if the fog as post-process feature is enabled, we need to proceed "as-if" the material
+        // didn't have the FOG variant bit.
+        if (mMaterial->getMaterialDomain() == MaterialDomain::SURFACE) {
+            BlendingMode const blendingMode = mMaterial->getBlendingMode();
+            bool const hasScreenSpaceRefraction =
+                    mMaterial->getRefractionMode() == RefractionMode::SCREEN_SPACE;
+            bool const isBlendingCommand = !hasScreenSpaceRefraction &&
+                    (blendingMode != BlendingMode::OPAQUE && blendingMode != BlendingMode::MASKED);
+            if (!isBlendingCommand) {
+                variant.setFog(false);
+            }
+        }
+    }
+    return variant;
+}
+
+Program::SpecializationConstant LocalProgramCache::getConstantImpl(uint32_t id) const noexcept {
+    return mSpecializationConstants[id];
+}
+
+Program::SpecializationConstant LocalProgramCache::getConstantImpl(
+        std::string_view name) const noexcept {
+    assert_invariant(mMaterial != nullptr);
+
+    auto const& constants = mMaterial->getDefinition().specializationConstantsNameToIndex;
+    auto it = constants.find(name);
+    FILAMENT_CHECK_PRECONDITION(it != constants.end()) << "Constant " << name << " does not exist";
+
+    return getConstantImpl(it->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS);
+}
+
+void LocalProgramCache::setConstants(
+        std::initializer_list<std::pair<uint32_t, Program::SpecializationConstant>>
+                constants) noexcept {
+    assert_invariant(mMaterial != nullptr);
+
+    auto newSpecializationConstants =
+            FixedCapacityVector<Program::SpecializationConstant>(mSpecializationConstants);
+
+    bool hasChanged = false;
+    for (const auto& [id, value] : constants) {
+        if (newSpecializationConstants[id] != value) {
+            newSpecializationConstants[id] = value;
+            hasChanged = true;
+        }
+    }
+
+    if (hasChanged) {
+        setConstantsImpl(std::move(newSpecializationConstants));
+    }
+}
+
+void LocalProgramCache::setConstants(
+        std::initializer_list<std::pair<std::string_view, Program::SpecializationConstant>>
+                constants) noexcept {
+    assert_invariant(mMaterial != nullptr);
+
+    auto newSpecializationConstants =
+            FixedCapacityVector<Program::SpecializationConstant>(mSpecializationConstants);
+
+    bool hasChanged = false;
+    for (const auto& [name, value] : constants) {
+        MaterialDefinition const& definition = mMaterial->getDefinition();
+        auto it = definition.specializationConstantsNameToIndex.find(name);
+        if (it != definition.specializationConstantsNameToIndex.cend()) {
+            uint32_t id = it->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS;
+            if (newSpecializationConstants[id] != value) {
+                newSpecializationConstants[id] = value;
+                hasChanged = true;
+            }
+        }
+    }
+
+    if (hasChanged) {
+        setConstantsImpl(std::move(newSpecializationConstants));
+    }
+}
+
+void LocalProgramCache::setConstants(
+        FixedCapacityVector<Program::SpecializationConstant> constants) noexcept {
+    assert_invariant(mMaterial != nullptr);
+
+    setConstantsImpl(std::move(constants));
+}
+
+void LocalProgramCache::setConstantsImpl(
+        FixedCapacityVector<Program::SpecializationConstant> constants) noexcept {
+    FEngine& engine = mMaterial->getEngine();
+
+    auto& internPool = engine.getMaterialCache().getSpecializationConstantsInternPool();
+    MaterialParser const& materialParser = mMaterial->getMaterialParser();
+    MaterialDefinition const& definition = mMaterial->getDefinition();
+    const bool isDefaultMaterial = mMaterial->isDefaultMaterial();
+
+    // Release old resources...
+    definition.releasePrograms(engine, mCachedPrograms.as_slice(), materialParser,
+            mSpecializationConstants, isDefaultMaterial);
+    internPool.release(mSpecializationConstants);
+
+    // Then acquire new ones.
+    mSpecializationConstants = internPool.acquire(std::move(constants));
+    definition.acquirePrograms(engine, mCachedPrograms.as_slice(), materialParser,
+            mSpecializationConstants, isDefaultMaterial);
+}
+
+template int32_t LocalProgramCache::getConstant<int32_t>(uint32_t id) const noexcept;
+template float LocalProgramCache::getConstant<float>(uint32_t id) const noexcept;
+template bool LocalProgramCache::getConstant<bool>(uint32_t id) const noexcept;
+
+template int32_t LocalProgramCache::getConstant<int32_t>(std::string_view name) const noexcept;
+template float LocalProgramCache::getConstant<float>(std::string_view name) const noexcept;
+template bool LocalProgramCache::getConstant<bool>(std::string_view name) const noexcept;
+
+} // namespace filament

filament/src/LocalProgramCache.h

@@ -0,0 +1,151 @@
+/*
+ * Copyright (C) 2026 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_LOCALPROGRAMCACHE_H
+#define TNT_FILAMENT_LOCALPROGRAMCACHE_H
+
+#include "MaterialDefinition.h"
+
+#include <backend/Handle.h>
+
+#include <private/filament/Variant.h>
+
+#include <backend/DriverEnums.h>
+#include <backend/DriverApiForward.h>
+#include <backend/Program.h>
+
+#include <utility>
+
+namespace filament {
+
+class FMaterial;
+
+// L0 cache for material programs. Manages recompiling them on-demand; owned by Material and
+// MaterialInstance.
+class LocalProgramCache {
+    template<typename T>
+    using is_supported_constant_parameter_t =
+            std::enable_if_t<std::is_same_v<int32_t, T> || std::is_same_v<float, T> ||
+                             std::is_same_v<bool, T>>;
+
+public:
+    using Programs = utils::Slice<const backend::Handle<backend::HwProgram>>;
+    using SpecializationConstants = utils::Slice<const backend::Program::SpecializationConstant>;
+
+    LocalProgramCache() = default;
+    LocalProgramCache(LocalProgramCache const& other);
+
+    LocalProgramCache& operator=(LocalProgramCache const& other);
+
+    // Initialize for use in a Material.
+    void initializeForMaterial(FEngine& engine, FMaterial const& material,
+            utils::FixedCapacityVector<backend::Program::SpecializationConstant>
+                    specializationConstants);
+
+    // Initialize for use in a MaterialInstance. Copies the set of spec constants currently in use
+    // from its Material.
+    void initializeForMaterialInstance(FEngine& engine, FMaterial const& material);
+
+    bool isInitialized() const noexcept { return mMaterial != nullptr; }
+
+    // prepareProgram creates the program for the material's given variant at the backend level.
+    // Must be called outside of backend render pass.
+    // Must be called before getProgram() below.
+    backend::Handle<backend::HwProgram> prepareProgram(backend::DriverApi& driver,
+            Variant const variant,
+            backend::CompilerPriorityQueue const priorityQueue) const noexcept {
+        backend::Handle<backend::HwProgram> program = mCachedPrograms[variant.key];
+        if (UTILS_LIKELY(program)) {
+            return program;
+        }
+        return prepareProgramSlow(driver, variant, priorityQueue);
+    }
+
+    // getProgram returns the backend program for the material's given variant.
+    // Must be called after prepareProgram().
+    [[nodiscard]]
+    backend::Handle<backend::HwProgram> getProgram(Variant variant) const noexcept {
+        variant = filterVariantForGetProgram(variant);
+        backend::Handle<backend::HwProgram> program = mCachedPrograms[variant.key];
+        assert_invariant(program);
+        return program;
+    }
+
+    SpecializationConstants getSpecializationConstants() const noexcept {
+        return mSpecializationConstants;
+    }
+
+    Programs getPrograms() const noexcept { return mCachedPrograms.as_slice(); }
+
+    // Free all engine resources associated with this instance.
+    void terminate(FEngine& engine);
+
+    // Clear all cached programs. Used primarily by matdbg to "sever" a Material's connection to the
+    // global material cache.
+    void clear(FEngine& engine);
+
+    // Get constant by ID.
+    template<typename T, typename = is_supported_constant_parameter_t<T>>
+    T getConstant(uint32_t id) const noexcept {
+        return std::get<T>(getConstantImpl(id));
+    }
+
+    // Get constant by name.
+    template<typename T, typename = is_supported_constant_parameter_t<T>>
+    T getConstant(std::string_view name) const noexcept {
+        return std::get<T>(getConstantImpl(name));
+    }
+
+    // Set constants by ID.
+    void setConstants(
+            std::initializer_list<std::pair<uint32_t, backend::Program::SpecializationConstant>>
+                    constants) noexcept;
+
+    // Set constants by name.
+    void setConstants(std::initializer_list<
+            std::pair<std::string_view, backend::Program::SpecializationConstant>>
+                    constants) noexcept;
+
+    // Set constants list directly.
+    void setConstants(utils::FixedCapacityVector<backend::Program::SpecializationConstant>
+                    constants) noexcept;
+
+private:
+    // Apply any pending specialization constants. Invalidates programs as necessary.
+    void flushConstants() const;
+
+    backend::Handle<backend::HwProgram> prepareProgramSlow(backend::DriverApi& driver,
+            Variant const variant,
+            backend::CompilerPriorityQueue const priorityQueue) const noexcept;
+
+    ProgramSpecialization getProgramSpecialization(Variant variant) const noexcept;
+
+    Variant filterVariantForGetProgram(Variant const variant) const noexcept;
+
+    void setConstantsImpl(utils::FixedCapacityVector<backend::Program::SpecializationConstant>
+                    constants) noexcept;
+
+    backend::Program::SpecializationConstant getConstantImpl(uint32_t id) const noexcept;
+    backend::Program::SpecializationConstant getConstantImpl(std::string_view name) const noexcept;
+
+    FMaterial const* mMaterial = nullptr;
+    mutable utils::FixedCapacityVector<backend::Handle<backend::HwProgram>> mCachedPrograms;
+    SpecializationConstants mSpecializationConstants;
+};
+
+} // namespace filament
+
+#endif  // TNT_FILAMENT_LOCALPROGRAMCACHE_H

filament/src/MaterialCache.cpp

@@ -39,12 +39,25 @@ bool MaterialCache::MaterialKey::operator==(MaterialKey const& rhs) const noexce
     return parser == rhs.parser;
 }
 
+MaterialCache::MaterialCache()
+        : mDefinitions("MaterialCache::mDefinitions", 0),
+          mPrograms("MaterialCache::mPrograms", 0) {}
+
 MaterialCache::~MaterialCache() {
     assert_invariant(mDefinitions.empty());
     assert_invariant(mPrograms.empty());
     assert_invariant(mSpecializationConstantsInternPool.empty());
 }
 
+void MaterialCache::terminate(FEngine& engine) {
+    mPrograms.clearLruCache([&engine](backend::Handle<backend::HwProgram>& program) {
+        engine.getDriverApi().destroyProgram(program);
+    });
+    mDefinitions.clearLruCache([&engine](MaterialDefinition& definition) {
+        definition.terminate(engine);
+    });
+}
+
 MaterialDefinition* UTILS_NULLABLE MaterialCache::acquireMaterial(FEngine& engine,
         const void* UTILS_NONNULL data, size_t size) noexcept {
     std::unique_ptr<MaterialParser> parser = MaterialDefinition::createParser(engine.getBackend(),

filament/src/MaterialCache.h

@@ -56,8 +56,13 @@ public:
     using ProgramCache =
             utils::RefCountedMap<ProgramSpecialization, backend::Handle<backend::HwProgram>>;
 
+    MaterialCache();
     ~MaterialCache();
 
+    // All reference-counted resources should be freed by the time MaterialCache is destructed, but
+    // the LRU cache needs to be explicitly freed in addition.
+    void terminate(FEngine& engine);
+
     SpecializationConstantInternPool& getSpecializationConstantsInternPool() {
         return mSpecializationConstantsInternPool;
     }

filament/src/MaterialDefinition.cpp

@@ -271,8 +271,8 @@ std::unique_ptr<MaterialDefinition> MaterialDefinition::create(FEngine& engine,
 
 void MaterialDefinition::terminate(FEngine& engine) {
     DriverApi& driver = engine.getDriverApi();
-    perViewDescriptorSetLayout.terminate(engine.getDescriptorSetLayoutFactory(), driver);
-    perViewDescriptorSetLayoutVsm.terminate(engine.getDescriptorSetLayoutFactory(), driver);
+    perViewDescriptorSetLayoutPcf.terminate(engine.getDescriptorSetLayoutFactory(), driver);
+    perViewDescriptorSetLayoutS2d.terminate(engine.getDescriptorSetLayoutFactory(), driver);
     descriptorSetLayout.terminate(engine.getDescriptorSetLayoutFactory(), driver);
 }
 
@@ -607,23 +607,23 @@ void MaterialDefinition::processDescriptorSets(FEngine& engine) {
             refractionMode == RefractionMode::SCREEN_SPACE;
     bool const hasFog = !(variantFilterMask & UserVariantFilterMask(UserVariantFilterBit::FOG));
 
-    this->perViewDescriptorSetLayoutDescription = descriptor_sets::getPerViewDescriptorSetLayout(
+    this->perViewDescriptorSetLayoutPcfDescription = descriptor_sets::getPerViewDescriptorSetLayout(
             materialDomain, isLit, isSSR, hasFog, false);
 
-    this->perViewDescriptorSetLayoutVsmDescription = descriptor_sets::getPerViewDescriptorSetLayout(
+    this->perViewDescriptorSetLayoutS2dDescription = descriptor_sets::getPerViewDescriptorSetLayout(
             materialDomain, isLit, isSSR, hasFog, true);
 
     // set the labels
     this->descriptorSetLayoutDescription.label = CString{ name }.append("_perMat");
-    this->perViewDescriptorSetLayoutDescription.label = CString{ name }.append("_perView");
-    this->perViewDescriptorSetLayoutVsmDescription.label = CString{ name }.append("_perViewVsm");
+    this->perViewDescriptorSetLayoutPcfDescription.label = CString{ name }.append("_perView");
+    this->perViewDescriptorSetLayoutS2dDescription.label = CString{ name }.append("_perViewVsm");
 
     // get the PER_RENDERABLE and PER_VIEW descriptor binding info
     for (auto&& [bindingPoint, dsl] : {
             std::pair{ DescriptorSetBindingPoints::PER_RENDERABLE,
                     descriptor_sets::getPerRenderableLayout() },
             std::pair{ DescriptorSetBindingPoints::PER_VIEW,
-                    this->perViewDescriptorSetLayoutDescription }}) {
+                    this->perViewDescriptorSetLayoutPcfDescription }}) {
         Program::DescriptorBindingsInfo& descriptors = programDescriptorBindings[+bindingPoint];
         descriptors.reserve(dsl.descriptors.size());
         for (auto const& entry: dsl.descriptors) {
@@ -636,17 +636,17 @@ void MaterialDefinition::processDescriptorSets(FEngine& engine) {
             descriptorSetLayoutFactory, driver,
             this->descriptorSetLayoutDescription };
 
-    this->perViewDescriptorSetLayout = {
+    this->perViewDescriptorSetLayoutPcf = {
             descriptorSetLayoutFactory, driver,
-            this->perViewDescriptorSetLayoutDescription };
+            this->perViewDescriptorSetLayoutPcfDescription };
 
-    this->perViewDescriptorSetLayoutVsm = {
+    this->perViewDescriptorSetLayoutS2d = {
             descriptorSetLayoutFactory, driver,
-            this->perViewDescriptorSetLayoutVsmDescription };
+            this->perViewDescriptorSetLayoutS2dDescription };
 }
 
 backend::DescriptorSetLayout const& MaterialDefinition::getPerViewDescriptorSetLayoutDescription(
-        Variant const variant, bool const useVsmDescriptorSetLayout) const noexcept {
+        Variant const variant, bool const useS2dDescriptorSetLayout) const noexcept {
     if (materialDomain == MaterialDomain::SURFACE) {
         if (Variant::isValidDepthVariant(variant)) {
             // Use the layout description used to create the per view depth variant layout.
@@ -657,10 +657,10 @@ backend::DescriptorSetLayout const& MaterialDefinition::getPerViewDescriptorSetL
             return descriptor_sets::getSsrVariantLayout();
         }
     }
-    if (useVsmDescriptorSetLayout) {
-        return perViewDescriptorSetLayoutVsmDescription;
+    if (useS2dDescriptorSetLayout) {
+        return perViewDescriptorSetLayoutS2dDescription;
     }
-    return perViewDescriptorSetLayoutDescription;
+    return perViewDescriptorSetLayoutPcfDescription;
 }
 
 Handle<HwProgram> MaterialDefinition::compileProgram(
@@ -689,7 +689,7 @@ Handle<HwProgram> MaterialDefinition::compileProgram(
     pb.descriptorLayout(+DescriptorSetBindingPoints::PER_VIEW,
             getPerViewDescriptorSetLayoutDescription(
                     specialization.variant,
-                    Variant::isVSMVariant(specialization.variant)));
+                    Variant::isShadowSampler2DVariant(specialization.variant)));
     pb.descriptorLayout(+DescriptorSetBindingPoints::PER_RENDERABLE,
             descriptor_sets::getPerRenderableLayout());
     pb.descriptorLayout(

filament/src/MaterialDefinition.h

@@ -94,17 +94,17 @@ struct MaterialDefinition {
             backend::ShaderModel const sm, bool isStereoSupported) const noexcept;
 
     backend::DescriptorSetLayout const& getPerViewDescriptorSetLayoutDescription(
-            Variant const variant, bool useVsmDescriptorSetLayout) const noexcept;
+            Variant const variant, bool useS2dDescriptorSetLayout) const noexcept;
 
     // Keep track of the definitions of the descriptor set layouts, as these
     // may be used by some backends in parallel compilation of programs.
-    backend::DescriptorSetLayout perViewDescriptorSetLayoutDescription;
-    backend::DescriptorSetLayout perViewDescriptorSetLayoutVsmDescription;
+    backend::DescriptorSetLayout perViewDescriptorSetLayoutPcfDescription;
+    backend::DescriptorSetLayout perViewDescriptorSetLayoutS2dDescription;
     backend::DescriptorSetLayout descriptorSetLayoutDescription;
 
     // try to order by frequency of use
-    filament::DescriptorSetLayout perViewDescriptorSetLayout;
-    filament::DescriptorSetLayout perViewDescriptorSetLayoutVsm;
+    filament::DescriptorSetLayout perViewDescriptorSetLayoutPcf;
+    filament::DescriptorSetLayout perViewDescriptorSetLayoutS2d;
     filament::DescriptorSetLayout descriptorSetLayout;
     backend::Program::DescriptorSetInfo programDescriptorBindings;
 

filament/src/MaterialInstance.cpp

@@ -233,6 +233,26 @@ template UTILS_PUBLIC mat3f MaterialInstance::getParameter<mat3f>      (const ch
 
 // ------------------------------------------------------------------------------------------------
 
+template<typename T, typename>
+void MaterialInstance::setConstant(const char* name, size_t nameLength, T value) {
+    downcast(this)->setConstantImpl(std::string_view{name, nameLength}, value);
+}
+
+template UTILS_PUBLIC void MaterialInstance::setConstant<int32_t>(const char* name, size_t nameLength, int32_t value);
+template UTILS_PUBLIC void MaterialInstance::setConstant<float>(const char* name, size_t nameLength, float value);
+template UTILS_PUBLIC void MaterialInstance::setConstant<bool>(const char* name, size_t nameLength, bool value);
+
+template<typename T, typename>
+T MaterialInstance::getConstant(const char* name, size_t nameLength) const {
+    return downcast(this)->getConstantImpl<T>(std::string_view{name, nameLength});
+}
+
+template UTILS_PUBLIC int32_t MaterialInstance::getConstant<int32_t>(const char* name, size_t nameLength) const;
+template UTILS_PUBLIC float MaterialInstance::getConstant<float>(const char* name, size_t nameLength) const;
+template UTILS_PUBLIC bool MaterialInstance::getConstant<bool>(const char* name, size_t nameLength) const;
+
+// ------------------------------------------------------------------------------------------------
+
 Material const* MaterialInstance::getMaterial() const noexcept {
     return downcast(this)->getMaterial();
 }
@@ -403,4 +423,10 @@ void MaterialInstance::commit(Engine& engine) const {
     downcast(this)->commit(downcast(engine));
 }
 
+void MaterialInstance::compile(CompilerPriorityQueue const priority,
+        UserVariantFilterMask const variants, CallbackHandler* handler,
+        utils::Invocable<void(MaterialInstance*)>&& callback) noexcept {
+    downcast(this)->compile(priority, variants, handler, std::move(callback));
+}
+
 } // namespace filament

filament/src/MaterialInstanceManager.cpp

@@ -19,90 +19,61 @@
 
 #include "details/Engine.h"
 #include "details/Material.h"
-#include "details/MaterialInstance.h"
 
-#include <utils/debug.h>
-
-#include <iterator>
-#include <utility>
-
-#include <stdint.h>
+#include <cstdint>
 
 namespace filament {
 
-using Record = MaterialInstanceManager::Record;
+using namespace utils;
 
-std::pair<FMaterialInstance*, int32_t> Record::getInstance() {
-    if (mAvailable < mInstances.size()) {
-        auto index = mAvailable++;
-        return { mInstances[index], index };
-    }
-    assert_invariant(mAvailable == mInstances.size());
-    auto& name = mMaterial->getName();
-    FMaterialInstance* inst = mMaterial->createInstance(name.c_str_safe());
-    mInstances.push_back(inst);
-    return { inst, mAvailable++ };
-}
+MaterialInstanceManager::MaterialInstanceManager() noexcept = default;
 
-FMaterialInstance* Record::getInstance(int32_t const fixedInstanceindex) const {
-    assert_invariant(fixedInstanceindex >= 0 &&  fixedInstanceindex < int32_t(mInstances.size()));
-    return mInstances[fixedInstanceindex];
-}
-
-// Defined in cpp to avoid inlining
-Record::Record(Record const& rhs) noexcept = default;
-Record& Record::operator=(Record const& rhs) noexcept = default;
-Record::Record(Record&& rhs) noexcept = default;
-Record& Record::operator=(Record&& rhs) noexcept = default;
-
-void Record::terminate(FEngine& engine) {
-    std::for_each(mInstances.begin(), mInstances.end(),
-            [&engine](auto instance) { engine.destroy(instance); });
-}
-
-MaterialInstanceManager::MaterialInstanceManager() noexcept {}
-
-MaterialInstanceManager::MaterialInstanceManager(
-        MaterialInstanceManager const& rhs) noexcept = default;
 MaterialInstanceManager::MaterialInstanceManager(MaterialInstanceManager&& rhs) noexcept = default;
-MaterialInstanceManager& MaterialInstanceManager::operator=(
-        MaterialInstanceManager const& rhs) noexcept = default;
-MaterialInstanceManager& MaterialInstanceManager::operator=(
-        MaterialInstanceManager&& rhs) noexcept = default;
-
+MaterialInstanceManager& MaterialInstanceManager::operator=(MaterialInstanceManager&& rhs) noexcept = default;
 MaterialInstanceManager::~MaterialInstanceManager() = default;
 
 void MaterialInstanceManager::terminate(FEngine& engine) {
-    std::for_each(mMaterials.begin(), mMaterials.end(), [&engine](auto& record) {
-        record.terminate(engine);
-    });
+    for (auto const& [key, instance] : mMaterialInstances) {
+        engine.destroy(instance);
+    }
+    mMaterialInstances.clear();
+
+    for (auto const& [material, pool] : mAnonymousMaterialInstances) {
+        for (auto instance : pool.instances) {
+            engine.destroy(instance);
+        }
+    }
+    mAnonymousMaterialInstances.clear();
 }
 
-Record& MaterialInstanceManager::getRecord(FMaterial const* const ma) const {
-    auto itr = std::find_if(mMaterials.begin(), mMaterials.end(), [ma](auto& record) {
-        return ma == record.mMaterial;
-    });
-    if (itr == mMaterials.end()) {
-        mMaterials.emplace_back(ma);
-        itr = std::prev(mMaterials.end());
+void MaterialInstanceManager::reset() {
+    for (auto& [material, pool] : mAnonymousMaterialInstances) {
+        pool.highWaterMark = 0;
     }
-    return *itr;
+}
+
+FMaterialInstance* MaterialInstanceManager::getMaterialInstance(FMaterial const* ma, uint32_t tag) const {
+    Key const key{ma, tag};
+    auto it = mMaterialInstances.find(key);
+    if (it != mMaterialInstances.end()) {
+        return it->second;
+    }
+
+    FMaterialInstance* const instance = ma->createInstance(ma->getName().c_str_safe());
+    mMaterialInstances.emplace(key, instance);
+    return instance;
 }
 
 FMaterialInstance* MaterialInstanceManager::getMaterialInstance(FMaterial const* ma) const {
-    auto [inst, index] = getRecord(ma).getInstance();
-    return inst;
-}
+    AnonymousPool& pool = mAnonymousMaterialInstances[ma];
+    if (pool.highWaterMark < pool.instances.size()) {
+        return pool.instances[pool.highWaterMark++];
+    }
 
-FMaterialInstance* MaterialInstanceManager::getMaterialInstance(FMaterial const* ma,
-        int32_t const fixedIndex) const {
-    return getRecord(ma).getInstance(fixedIndex);
+    FMaterialInstance* const instance = ma->createInstance(ma->getName().c_str_safe());
+    pool.instances.push_back(instance);
+    pool.highWaterMark++;
+    return instance;
 }
 
-std::pair<FMaterialInstance*, int32_t> MaterialInstanceManager::getFixedMaterialInstance(
-        FMaterial const* ma) {
-    return getRecord(ma).getInstance();
-}
-
-
 } // namespace filament

filament/src/MaterialInstanceManager.h

@@ -16,10 +16,13 @@
 
 #pragma once
 
-#include <utils/bitset.h>
-
+#include <cstddef>
+#include <cstdint>
+#include <unordered_map>
 #include <vector>
 
+#include <utils/Hash.h>
+
 namespace filament {
 
 class FMaterial;
@@ -30,37 +33,10 @@ class FEngine;
 // re-use instances across frames.
 class MaterialInstanceManager {
 public:
-    class Record {
-    public:
-        Record(FMaterial const* material)
-            : mMaterial(material),
-              mAvailable(0) {}
-        ~Record() = default;
-
-        Record(Record const& rhs) noexcept;
-        Record& operator=(Record const& rhs) noexcept;
-        Record(Record&& rhs) noexcept;
-        Record& operator=(Record&& rhs) noexcept;
-
-        void terminate(FEngine& engine);
-        void reset() { mAvailable = 0; }
-        std::pair<FMaterialInstance*, int32_t> getInstance();
-        FMaterialInstance* getInstance(int32_t fixedInstanceindex) const;
-
-    private:
-        FMaterial const* mMaterial = nullptr;
-        std::vector<FMaterialInstance*> mInstances;
-        uint32_t mAvailable;
-
-        friend class MaterialInstanceManager;
-    };
-
-    constexpr static int32_t INVALID_FIXED_INDEX = -1;
-
     MaterialInstanceManager() noexcept;
-    MaterialInstanceManager(MaterialInstanceManager const& rhs) noexcept;
+    MaterialInstanceManager(MaterialInstanceManager const& rhs) = delete;
     MaterialInstanceManager(MaterialInstanceManager&& rhs) noexcept;
-    MaterialInstanceManager& operator=(MaterialInstanceManager const& rhs) noexcept;
+    MaterialInstanceManager& operator=(MaterialInstanceManager const& rhs) = delete;
     MaterialInstanceManager& operator=(MaterialInstanceManager&& rhs) noexcept;
 
     ~MaterialInstanceManager();
@@ -71,40 +47,60 @@ public:
      */
     void terminate(FEngine& engine);
 
-    /*
-     * This returns a material instance given a material. The implementation will try to find an
-     * available instance in the cache. If one is not found, then a new instance will be created and
-     * added to the cache.
+    /**
+     * Resets the anonymous material instances cache.
+     */
+    void reset();
+
+    /**
+     * This returns a material instance given a material and a tag.
+     *
+     * If the material instance doesn't exist in the cache, it is created and cached.
+     *
+     * @param ma    FMaterial to get a MaterialInstance for
+     * @param tag   A unique tag identifying the MaterialInstance
+     * @return      A FMaterialInstance pointer
+     */
+    FMaterialInstance* getMaterialInstance(FMaterial const* ma, uint32_t tag) const;
+
+    /**
+     * This returns a material instance given a material from a cache.
+     *
+     * If the material instance doesn't exist in the cache, it is created and cached.
+     *
+     * It is permissible to call the method several times, in which case a different MaterialInstance will be returned.
+     * It is guaranteed to be different from MaterialInstances returned with a tag.
+     *
+     * @param ma    FMaterial to get a MaterialInstance for
+     * @return      A FMaterialInstance pointer
      */
     FMaterialInstance* getMaterialInstance(FMaterial const* ma) const;
 
-    /*
-     * This returns a material instance given a material and an index. The `fixedIndex` should be
-     * a value returned by getiFixedMaterialInstance.
-     */
-    FMaterialInstance* getMaterialInstance(FMaterial const* ma, int32_t const fixedIndex) const;
-
-    /*
-     * This returns a material instance and an index given a material. This is needed for the
-     * case when two framegraph passes need to refer to the same material instance.
-     * The returned index can be used with `getFixedMaterialInstance` to get a specific instance
-     * of a material (and not a random entry in the record cache).
-     */
-    std::pair<FMaterialInstance*, int32_t> getFixedMaterialInstance(FMaterial const* ma);
-
-
-    /*
-     * Marks all of the material instances as unused. Typically, you'd call this at the beginning of
-     * a frame.
-     */
-    void reset() {
-        std::for_each(mMaterials.begin(), mMaterials.end(), [](auto& record) { record.reset(); });
-    }
-
 private:
-    Record& getRecord(FMaterial const* material) const;
+    struct Key {
+        FMaterial const* material;
+        uint32_t tag;
+        bool operator==(Key const& rhs) const noexcept {
+            return material == rhs.material && tag == rhs.tag;
+        }
+    };
 
-    mutable std::vector<Record> mMaterials;
+    struct Hasher {
+        std::size_t operator()(Key const& key) const noexcept {
+            std::size_t seed = 0;
+            utils::hash::combine(seed, key.material);
+            utils::hash::combine(seed, key.tag);
+            return seed;
+        }
+    };
+
+    struct AnonymousPool {
+        std::vector<FMaterialInstance*> instances;
+        uint32_t highWaterMark = 0;
+    };
+
+    mutable std::unordered_map<Key, FMaterialInstance*, Hasher> mMaterialInstances;
+    mutable std::unordered_map<FMaterial const*, AnonymousPool> mAnonymousMaterialInstances;
 };
 
 } // namespace filament

filament/src/PostProcessManager.cpp

@@ -184,12 +184,10 @@ void PostProcessManager::PostProcessMaterial::loadMaterial(FEngine& engine) cons
 }
 
 UTILS_NOINLINE
-FMaterial* PostProcessManager::PostProcessMaterial::getMaterial(FEngine& engine,
-        DriverApi& driver, Variant::type_t const variant) const noexcept {
+FMaterial* PostProcessManager::PostProcessMaterial::getMaterial(FEngine& engine) const noexcept {
     if (UTILS_UNLIKELY(mSize)) {
         loadMaterial(engine);
     }
-    mMaterial->prepareProgram(driver, Variant{ variant }, CompilerPriorityQueue::CRITICAL);
     return mMaterial;
 }
 
@@ -225,7 +223,6 @@ const PostProcessManager::JitterSequence<32>
 
 PostProcessManager::PostProcessManager(FEngine& engine) noexcept
         : mEngine(engine),
-          mFixedMaterialInstanceIndex {},
           mWorkaroundSplitEasu(false),
           mWorkaroundAllowReadOnlyAncillaryFeedbackLoop(false) {
     // don't use Engine here, it's not fully initialized yet
@@ -248,6 +245,20 @@ void PostProcessManager::bindPerRenderableDescriptorSet(DriverApi& driver) const
             { { 0, 0 }, driver });
 }
 
+FMaterialInstance* PostProcessManager::getMaterialInstance(backend::DriverApi& driver,
+        FMaterial const* ma, Variant::type_t variant) const {
+    FMaterialInstance* mi = mMaterialInstanceManager.getMaterialInstance(ma);
+    mi->prepareProgram(driver, Variant{ variant }, backend::CompilerPriorityQueue::CRITICAL);
+    return mi;
+}
+
+FMaterialInstance* PostProcessManager::getMaterialInstanceWithTag(backend::DriverApi& driver,
+        FMaterial const* ma, uint32_t tag, Variant::type_t variant) const {
+    FMaterialInstance* mi = mMaterialInstanceManager.getMaterialInstance(ma, tag);
+    mi->prepareProgram(driver, Variant { variant }, backend::CompilerPriorityQueue::CRITICAL);
+    return mi;
+}
+
 UboManager* PostProcessManager::getUboManager() const noexcept {
     return mEngine.getUboManager();
 }
@@ -455,7 +466,6 @@ Handle<HwTexture> PostProcessManager::getZeroTextureArray() const {
 
 void PostProcessManager::resetForRender() {
     mMaterialInstanceManager.reset();
-    mFixedMaterialInstanceIndex = {};
 }
 
 void PostProcessManager::unbindAllDescriptorSets(DriverApi& driver) noexcept {
@@ -466,9 +476,10 @@ void PostProcessManager::unbindAllDescriptorSets(DriverApi& driver) noexcept {
 
 UTILS_NOINLINE
 PipelineState PostProcessManager::getPipelineState(
-        FMaterial const* const ma, Variant::type_t const variant) const noexcept {
+        FMaterialInstance const* const mi, Variant::type_t const variant) const noexcept {
+    FMaterial const* const ma = mi->getMaterial();
     return {
-            .program = ma->getProgram(Variant{ variant }),
+            .program = mi->getProgram(Variant{ variant }),
             .vertexBufferInfo = mFullScreenQuadVbih,
             .pipelineLayout = {
                     .setLayout = {
@@ -476,7 +487,7 @@ PipelineState PostProcessManager::getPipelineState(
                             mPerRenderableDslh,
                             ma->getDescriptorSetLayout().getHandle()
                     }},
-            .rasterState = ma->getRasterState()
+            .rasterState = mi->getRasterState()
     };
 }
 
@@ -516,8 +527,7 @@ void PostProcessManager::commitAndRenderFullScreenQuad(DriverApi& driver,
         PostProcessVariant const variant) const noexcept {
     mi->commit(driver, getUboManager());
     mi->use(driver);
-    FMaterial const* const ma = mi->getMaterial();
-    PipelineState const pipeline = getPipelineState(ma, variant);
+    PipelineState const pipeline = getPipelineState(mi, variant);
 
     assert_invariant(
             ((out.params.readOnlyDepthStencil & RenderPassParams::READONLY_DEPTH)
@@ -634,12 +644,12 @@ PostProcessManager::StructurePassOutput PostProcessManager::structure(FrameGraph
 
                 auto in = resources.getTexture(data.depth);
                 auto& material = getPostProcessMaterial("mipmapDepth");
-                FMaterial const* const ma = material.getMaterial(mEngine, driver);
-                FMaterialInstance* const mi = getMaterialInstance(ma);
+                FMaterial const* const ma = material.getMaterial(mEngine);
+                FMaterialInstance* const mi = getMaterialInstance(driver, ma);
                 // Only the depth texture is changing in the material instance (no UBO updates),
                 // we do not move getMaterialInstance() inside the loop.
 
-                auto pipeline = getPipelineState(ma);
+                auto pipeline = getPipelineState(mi);
 
                 // The first mip already exists, so we process n-1 lods
                 for (size_t level = 0; level < levelCount - 1; level++) {
@@ -802,7 +812,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::ssr(FrameGraph& fg,
 
                 // use our special SSR variant, it can only be applied to object that have
                 // the SCREEN_SPACE ReflectionMode.
-                passBuilder.variant(Variant{ Variant::SPECIAL_SSR });
+                passBuilder.variant(Variant{ Variant::SPECIAL_SSR_VARIANT });
 
                 // generate all our drawing commands, except blended objects.
                 passBuilder.commandTypeFlags(RenderPass::CommandTypeFlags::SCREEN_SPACE_REFLECTIONS);
@@ -999,17 +1009,13 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::screenSpaceAmbientOcclusion(
 #endif
                 auto& material = getPostProcessMaterial(materialName);
 
-                FMaterial* ma = material.getMaterial(mEngine, driver);
-                {
-                    FMaterial::SpecializationConstantsBuilder maConstants =
-                            ma->getSpecializationConstantsBuilder();
-                    maConstants.set("useVisibilityBitmasks", options.gtao.useVisibilityBitmasks);
-                    maConstants.set("linearThickness", options.gtao.linearThickness);
-                    ma->setSpecializationConstants(std::move(maConstants));
-                }
+                FMaterial* ma = material.getMaterial(mEngine);
+                ma->getPrograms().setConstants({
+                    { "useVisibilityBitmasks", options.gtao.useVisibilityBitmasks },
+                    { "linearThickness", options.gtao.linearThickness },
+                });
 
-                ma = material.getMaterial(mEngine, driver);
-                FMaterialInstance* const mi = getMaterialInstance(ma);
+                FMaterialInstance* const mi = getMaterialInstance(driver, ma);
 
                 // Set AO type specific material parameters
                 switch (aoType) {
@@ -1085,7 +1091,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::screenSpaceAmbientOcclusion(
                 mi->commit(driver, getUboManager());
                 mi->use(driver);
 
-                auto pipeline = getPipelineState(ma);
+                auto pipeline = getPipelineState(mi);
                 pipeline.rasterState.depthFunc = RasterState::DepthFunc::L;
                 assert_invariant(ssao.params.readOnlyDepthStencil & RenderPassParams::READONLY_DEPTH);
                 renderFullScreenQuad(ssao, pipeline, driver);
@@ -1195,8 +1201,8 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::bilateralBlurPass(FrameGraph
                 auto& material = config.bentNormals ?
                         getPostProcessMaterial("bilateralBlurBentNormals") :
                         getPostProcessMaterial("bilateralBlur");
-                FMaterial const* const ma = material.getMaterial(mEngine, driver);
-                FMaterialInstance* const mi = getMaterialInstance(ma);
+                FMaterial const* const ma = material.getMaterial(mEngine);
+                FMaterialInstance* const mi = getMaterialInstance(driver, ma);
                 mi->setParameter("ssao", ssao, { /* only reads level 0 */ });
                 mi->setParameter("axis", axis / float2{desc.width, desc.height});
                 mi->setParameter("kernel", kGaussianSamples, kGaussianCount);
@@ -1206,7 +1212,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::bilateralBlurPass(FrameGraph
                 mi->commit(driver, getUboManager());
                 mi->use(driver);
 
-                auto pipeline = getPipelineState(ma);
+                auto pipeline = getPipelineState(mi);
                 pipeline.rasterState.depthFunc = RasterState::DepthFunc::L;
                 renderFullScreenQuad(blurred, pipeline, driver);
                 unbindAllDescriptorSets(driver);
@@ -1364,7 +1370,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::gaussianBlurPass(FrameGraph&
                             "separableGaussianBlur4L"sv : "separableGaussianBlur4"sv;   break;
                 }
                 auto const& separableGaussianBlur = getPostProcessMaterial(materialName);
-                auto ma = separableGaussianBlur.getMaterial(mEngine, driver);
+                auto ma = separableGaussianBlur.getMaterial(mEngine);
 
                 const size_t kernelStorageSize = ma->reflect("kernel")->size;
                 float2 kernel[64];
@@ -1880,9 +1886,10 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::dof(FrameGraph& fg,
                 auto inOutCoc   = resources.getTexture(data.inOutCoc);
 
                 auto const& material = getPostProcessMaterial("dofMipmap");
-                FMaterial const* const ma = material.getMaterial(mEngine, driver);
+                FMaterial const* const ma = material.getMaterial(mEngine);
+                FMaterialInstance* const mi = getMaterialInstance(driver, ma);
 
-                auto const pipeline = getPipelineState(ma, variant);
+                auto const pipeline = getPipelineState(mi, variant);
 
                 for (size_t level = 0 ; level < mipmapCount - 1u ; level++) {
                     const float w = FTexture::valueForLevel(level, desc.width);
@@ -1890,7 +1897,8 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::dof(FrameGraph& fg,
                     auto const& out = resources.getRenderPassInfo(data.rp[level]);
                     auto inColor = driver.createTextureView(inOutColor, level, 1);
                     auto inCoc = driver.createTextureView(inOutCoc, level, 1);
-                    FMaterialInstance* const mi = getMaterialInstance(ma);
+                    // FIXME: is this necessary?
+                    FMaterialInstance* const mi = getMaterialInstance(driver, ma);
 
                     mi->setParameter("color", inColor, SamplerParams{
                             .filterMin = SamplerMinFilter::NEAREST_MIPMAP_NEAREST });
@@ -2407,9 +2415,9 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloom(FrameGraph& fg,
                 auto const& outDesc = resources.getDescriptor(data.out);
 
                 auto const& material = getPostProcessMaterial("bloomUpsample");
-                FMaterial const* const ma = material.getMaterial(mEngine, driver);
+                FMaterial const* const ma = material.getMaterial(mEngine);
 
-                auto pipeline = getPipelineState(ma);
+                auto pipeline = getPipelineState(getMaterialInstance(driver, ma));
                 pipeline.rasterState.blendFunctionSrcRGB = BlendFunction::ONE;
                 pipeline.rasterState.blendFunctionDstRGB = BlendFunction::ONE;
 
@@ -2417,7 +2425,7 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloom(FrameGraph& fg,
                     // Note that we wouldn't want to use the same instance for each pass since that
                     // would imply using the same UBOs, which implies synchronization across the
                     // passes.
-                    FMaterialInstance* mi = getMaterialInstance(ma);
+                    FMaterialInstance* mi = getMaterialInstance(driver, ma);
                     auto hwDstRT = resources.getRenderPassInfo(data.outRT[i - 1]);
                     hwDstRT.params.flags.discardStart = TargetBufferFlags::NONE; // b/c we'll blend
                     hwDstRT.params.flags.discardEnd = TargetBufferFlags::NONE;
@@ -2550,15 +2558,12 @@ void PostProcessManager::colorGradingSubpass(DriverApi& driver,
             PostProcessVariant::TRANSLUCENT : PostProcessVariant::OPAQUE;
 
     auto const& material = getPostProcessMaterial("colorGradingAsSubpass");
-    FMaterial const* const ma = material.getMaterial(mEngine, driver, variant);
+    FMaterial const* const ma = material.getMaterial(mEngine);
     // the UBO has been set and committed in colorGradingPrepareSubpass()
-    int32_t const fixedIndex = colorGradingConfig.translucent
-                                       ? mFixedMaterialInstanceIndex.colorGradingTranslucent
-                                       : mFixedMaterialInstanceIndex.colorGradingOpaque;
-
-    FMaterialInstance const* mi = mMaterialInstanceManager.getMaterialInstance(ma, fixedIndex);
+    FMaterialInstance const* mi =
+            getMaterialInstanceWithTag(driver, ma, colorGradingConfig.translucent, variant);
     mi->use(driver);
-    auto const pipeline = getPipelineState(ma, variant);
+    auto const pipeline = getPipelineState(mi, variant);
     driver.nextSubpass();
     driver.scissor(mi->getScissor());
     driver.draw(pipeline, mFullScreenQuadRph, 0, 3, 1);
@@ -2566,9 +2571,8 @@ void PostProcessManager::colorGradingSubpass(DriverApi& driver,
 
 void PostProcessManager::customResolvePrepareSubpass(DriverApi& driver, CustomResolveOp const op) noexcept {
     auto const& material = getPostProcessMaterial("customResolveAsSubpass");
-    auto const ma = material.getMaterial(mEngine, driver, PostProcessVariant::OPAQUE);
-    auto [mi, fixedIndex] = mMaterialInstanceManager.getFixedMaterialInstance(ma);
-    mFixedMaterialInstanceIndex.customResolve = fixedIndex;
+    auto const ma = material.getMaterial(mEngine);
+    auto* const mi = getMaterialInstance(driver, ma, 0);
     mi->setParameter("direction", op == CustomResolveOp::COMPRESS ? 1.0f : -1.0f),
     mi->commit(driver, getUboManager());
 }
@@ -2578,13 +2582,12 @@ void PostProcessManager::customResolveSubpass(DriverApi& driver) noexcept {
     bindPerRenderableDescriptorSet(driver);
 
     auto const& material = getPostProcessMaterial("customResolveAsSubpass");
-    FMaterial const* const ma = material.getMaterial(mEngine, driver);
+    FMaterial const* const ma = material.getMaterial(mEngine);
     // the UBO has been set and committed in customResolvePrepareSubpass()
-    FMaterialInstance const* mi = mMaterialInstanceManager.getMaterialInstance(ma,
-            mFixedMaterialInstanceIndex.customResolve);
+    FMaterialInstance const* mi = getMaterialInstance(driver, ma, 0);
     mi->use(driver);
 
-    auto const pipeline = getPipelineState(ma);
+    auto const pipeline = getPipelineState(mi);
     driver.nextSubpass();
     driver.scissor(mi->getScissor());
     driver.draw(pipeline, mFullScreenQuadRph, 0, 3, 1);
@@ -2620,9 +2623,8 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::customResolveUncompressPass(
 void PostProcessManager::clearAncillaryBuffersPrepare(DriverApi& driver,
         Variant::type_t variant) noexcept {
     auto const& material = getPostProcessMaterial("clearDepth");
-    auto ma = material.getMaterial(mEngine, driver, variant);
-    auto [mi, fixedIndex] = mMaterialInstanceManager.getFixedMaterialInstance(ma);
-    mFixedMaterialInstanceIndex.clearDepth = fixedIndex;
+    auto const ma = material.getMaterial(mEngine);
+    auto const mi = getMaterialInstanceWithTag(driver, ma, 0, variant);
     mi->commit(driver, getUboManager());
 }
 
@@ -2637,14 +2639,13 @@ void PostProcessManager::clearAncillaryBuffers(DriverApi& driver,
     bindPerRenderableDescriptorSet(driver);
 
     auto const& material = getPostProcessMaterial("clearDepth");
-    FMaterial const* const ma = material.getMaterial(mEngine, driver, variant);
+    FMaterial const* const ma = material.getMaterial(mEngine);
 
     // the UBO has been set and committed in clearAncillaryBuffersPrepare()
-    FMaterialInstance const* const mi = mMaterialInstanceManager.getMaterialInstance(ma,
-            mFixedMaterialInstanceIndex.clearDepth);
+    FMaterialInstance const* const mi = getMaterialInstanceWithTag(driver, ma, 0, variant);
     mi->use(driver);
 
-    auto pipeline = getPipelineState(ma, variant);
+    auto pipeline = getPipelineState(mi, variant);
     pipeline.rasterState.depthFunc = RasterState::DepthFunc::A;
 
     driver.scissor(mi->getScissor());
@@ -2654,7 +2655,7 @@ void PostProcessManager::clearAncillaryBuffers(DriverApi& driver,
 void PostProcessManager::fogPrepare(DriverApi& driver) noexcept {
     // ensures the material is loaded and material instance created
     auto const& material = getPostProcessMaterial("fog");
-    FMaterial const* const ma = material.getMaterial(mEngine, driver, PostProcessVariant::OPAQUE);
+    FMaterial const* const ma = material.getMaterial(mEngine);
     FMaterialInstance const* mi = ma->getDefaultInstance();
     mi->commit(driver, getUboManager());
 }
@@ -2665,11 +2666,11 @@ void PostProcessManager::fog(DriverApi& driver) noexcept {
     bindPerRenderableDescriptorSet(driver);
 
     auto const& material = getPostProcessMaterial("fog");
-    FMaterial const* const ma = material.getMaterial(mEngine, driver);
+    FMaterial const* const ma = material.getMaterial(mEngine);
     FMaterialInstance const* mi = ma->getDefaultInstance();
     mi->use(driver);
 
-    auto pipeline = getPipelineState(ma, Variant::NO_VARIANT);
+    auto pipeline = getPipelineState(mi, Variant::NO_VARIANT);
     driver.scissor(mi->getScissor());
     driver.draw(pipeline, mFullScreenQuadRph, 0, 3, 1);
 }
@@ -2926,45 +2927,37 @@ void PostProcessManager::TaaJitterCamera(
 void PostProcessManager::configureTemporalAntiAliasingMaterial(backend::DriverApi& driver,
         TemporalAntiAliasingOptions const& taaOptions) noexcept {
 
-    FMaterial* const ma = getPostProcessMaterial("taa").getMaterial(mEngine, driver);
-    FMaterial::SpecializationConstantsBuilder maConstants = ma->getSpecializationConstantsBuilder();
-
-    maConstants.set("upscaling", taaOptions.upscaling > 1.0f);
-    maConstants.set("historyReprojection", taaOptions.historyReprojection);
-    maConstants.set("filterHistory", taaOptions.filterHistory);
-    maConstants.set("filterInput", taaOptions.filterInput);
-    maConstants.set("useYCoCg", taaOptions.useYCoCg);
-    maConstants.set("hdr", taaOptions.hdr);
-    maConstants.set("preventFlickering", taaOptions.preventFlickering);
-    maConstants.set("boxType", int32_t(taaOptions.boxType));
-    maConstants.set("boxClipping", int32_t(taaOptions.boxClipping));
-    maConstants.set("varianceGamma", taaOptions.varianceGamma);
-
-    ma->setSpecializationConstants(std::move(maConstants));
+    FMaterial* const ma = getPostProcessMaterial("taa").getMaterial(mEngine);
+    ma->getPrograms().setConstants({
+        { "upscaling", taaOptions.upscaling > 1.0f },
+        { "historyReprojection", taaOptions.historyReprojection },
+        { "filterHistory", taaOptions.filterHistory },
+        { "filterInput", taaOptions.filterInput },
+        { "useYCoCg", taaOptions.useYCoCg },
+        { "hdr", taaOptions.hdr },
+        { "preventFlickering", taaOptions.preventFlickering },
+        { "boxType", int32_t(taaOptions.boxType) },
+        { "boxClipping", int32_t(taaOptions.boxClipping) },
+        { "varianceGamma", taaOptions.varianceGamma },
+    });
 }
 
 FMaterialInstance* PostProcessManager::configureColorGradingMaterial(backend::DriverApi& driver,
         PostProcessMaterial const& material, FColorGrading const* colorGrading,
         ColorGradingConfig const& colorGradingConfig, VignetteOptions const& vignetteOptions,
         uint32_t const width, uint32_t const height) noexcept {
-    FMaterial* ma = material.getMaterial(mEngine, driver);
-    {
-        FMaterial::SpecializationConstantsBuilder maConstants =
-                ma->getSpecializationConstantsBuilder();
-        maConstants.set("isOneDimensional", colorGrading->isOneDimensional());
-        maConstants.set("isLDR", colorGrading->isLDR());
-        ma->setSpecializationConstants(std::move(maConstants));
-    }
+    FMaterial* ma = material.getMaterial(mEngine);
+    ma->getPrograms().setConstants({
+        { "isOneDimensional", colorGrading->isOneDimensional() },
+        { "isLDR", colorGrading->isLDR() },
+    });
 
     PostProcessVariant const variant = colorGradingConfig.translucent
                                                ? PostProcessVariant::TRANSLUCENT
                                                : PostProcessVariant::OPAQUE;
-    ma = material.getMaterial(mEngine, driver, variant);
-    FMaterialInstance* mi = nullptr;
-    int32_t& fixedIndex = colorGradingConfig.translucent
-                                  ? mFixedMaterialInstanceIndex.colorGradingTranslucent
-                                  : mFixedMaterialInstanceIndex.colorGradingOpaque;
-    std::tie(mi, fixedIndex) = mMaterialInstanceManager.getFixedMaterialInstance(ma);
+    ma = material.getMaterial(mEngine);
+    FMaterialInstance* mi =
+            getMaterialInstanceWithTag(driver, ma, colorGradingConfig.translucent, variant);
 
     const SamplerParams params = SamplerParams{
             .filterMag = SamplerMagFilter::LINEAR,
@@ -3076,9 +3069,9 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
                 PostProcessVariant const variant = colorGradingConfig.translucent ?
                         PostProcessVariant::TRANSLUCENT : PostProcessVariant::OPAQUE;
 
-                FMaterial const* const ma = material.getMaterial(mEngine, driver, variant);
+                FMaterial const* const ma = material.getMaterial(mEngine);
 
-                FMaterialInstance* mi = getMaterialInstance(ma);
+                FMaterialInstance* mi = getMaterialInstance(driver, ma);
                 mi->setParameter("color",  color, SamplerParams{});  // nearest
                 mi->setParameter("depth",  depth, SamplerParams{});  // nearest
                 mi->setParameter("history", history, SamplerParams{
@@ -3118,7 +3111,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
                 if (colorGradingConfig.asSubpass) {
                     out.params.subpassMask = 1;
                 }
-                auto const pipeline = getPipelineState(ma, variant);
+                auto const pipeline = getPipelineState(mi, variant);
 
                 driver.beginRenderPass(out.target, out.params);
                 driver.draw(pipeline, mFullScreenQuadRph, 0, 3, 1);
@@ -3200,7 +3193,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::rcas(
                 mi->commit(driver, getUboManager());
                 mi->use(driver);
 
-                auto pipeline = getPipelineState(material.getMaterial(mEngine, driver), variant);
+                auto pipeline = getPipelineState(mi, variant);
                 if (mode == RcasMode::BLENDED) {
                     pipeline.rasterState.blendFunctionSrcRGB = BlendFunction::ONE;
                     pipeline.rasterState.blendFunctionSrcAlpha = BlendFunction::ONE;
@@ -3293,7 +3286,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::upscaleBilinear(FrameGraph&
 
                 auto out = resources.getRenderPassInfo();
 
-                auto pipeline = getPipelineState(material.getMaterial(mEngine, driver));
+                auto pipeline = getPipelineState(mi);
                 if (blended) {
                     pipeline.rasterState.blendFunctionSrcRGB = BlendFunction::ONE;
                     pipeline.rasterState.blendFunctionSrcAlpha = BlendFunction::ONE;
@@ -3507,15 +3500,15 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::upscaleFSR1(FrameGraph& fg,
                 auto out = resources.getRenderPassInfo();
 
                 if (UTILS_UNLIKELY(twoPassesEASU)) {
-                    auto pipeline0 = getPipelineState(splitEasuMaterial->getMaterial(mEngine, driver));
-                    auto pipeline1 = getPipelineState(easuMaterial->getMaterial(mEngine, driver));
+                    auto pipeline0 = getPipelineState(getMaterialInstance(mEngine, driver, *splitEasuMaterial));
+                    auto pipeline1 = getPipelineState(getMaterialInstance(mEngine, driver, *easuMaterial));
                     pipeline1.rasterState.depthFunc = SamplerCompareFunc::NE;
                     driver.beginRenderPass(out.target, out.params);
                     driver.draw(pipeline0, mFullScreenQuadRph, 0, 3, 1);
                     driver.draw(pipeline1, mFullScreenQuadRph, 0, 3, 1);
                     driver.endRenderPass();
                 } else {
-                    auto pipeline = getPipelineState(easuMaterial->getMaterial(mEngine, driver));
+                    auto pipeline = getPipelineState(getMaterialInstance(mEngine, driver, *easuMaterial));
                     renderFullScreenQuad(out, pipeline, driver);
                 }
                 unbindAllDescriptorSets(driver);
@@ -3567,8 +3560,8 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::blit(FrameGraph& fg, bool co
 
                 PostProcessMaterial const& material =
                         getPostProcessMaterial(layer ? "blitArray" : "blitLow");
-                FMaterial const* const ma = material.getMaterial(mEngine, driver);
-                auto* mi = getMaterialInstance(ma);
+                FMaterial const* const ma = material.getMaterial(mEngine);
+                auto* mi = getMaterialInstance(driver, ma);
                 mi->setParameter("color", color, SamplerParams{
                         .filterMag = filterMag,
                         .filterMin = filterMin
@@ -3586,7 +3579,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::blit(FrameGraph& fg, bool co
                 mi->commit(driver, getUboManager());
                 mi->use(driver);
 
-                auto pipeline = getPipelineState(ma);
+                auto pipeline = getPipelineState(mi);
                 if (translucent) {
                     pipeline.rasterState.blendFunctionSrcRGB = BlendFunction::ONE;
                     pipeline.rasterState.blendFunctionSrcAlpha = BlendFunction::ONE;
@@ -3823,17 +3816,15 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::vsmMipmapPass(FrameGraph& fg
                 auto const& inDesc = resources.getDescriptor(data.in);
                 auto width = inDesc.width;
                 assert_invariant(width == inDesc.height);
-                int const dim = width >> (level + 1);
+                uint32_t const dim = std::max(1u, width >> (level + 1));
 
                 auto& material = getPostProcessMaterial("vsmMipmap");
-                FMaterial const* const ma = material.getMaterial(mEngine, driver);
+                FMaterial const* const ma = material.getMaterial(mEngine);
+                FMaterialInstance* const mi = getMaterialInstance(driver, ma);
 
-                // When generating shadow map mip levels, we want to preserve the 1 texel border.
-                // (note clearing never respects the scissor in Filament)
-                auto const pipeline = getPipelineState(ma);
-                backend::Viewport const scissor = { 1u, 1u, dim - 2u, dim - 2u };
+                auto const pipeline = getPipelineState(mi);
+                backend::Viewport const scissor = { 0, 0, dim, dim };
 
-                FMaterialInstance* const mi = getMaterialInstance(ma);
                 mi->setParameter("color", in, SamplerParams{
                         .filterMag = SamplerMagFilter::LINEAR,
                         .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
@@ -3928,10 +3919,10 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::debugCombineArrayTexture(Fra
                 // set uniforms
 
                 PostProcessMaterial const& material = getPostProcessMaterial("blitArray");
-                FMaterial const* const ma = material.getMaterial(mEngine, driver);
+                FMaterial const* const ma = material.getMaterial(mEngine);
                 // It should be ok to not move this getMaterialInstance to inside the loop, since
                 // this is a pass meant for debug.
-                auto* mi = getMaterialInstance(ma);
+                auto* mi = getMaterialInstance(driver, ma);
                 mi->setParameter("color", color, SamplerParams{
                         .filterMag = filterMag,
                         .filterMin = filterMin
@@ -3945,7 +3936,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::debugCombineArrayTexture(Fra
                 mi->commit(driver, getUboManager());
                 mi->use(driver);
 
-                auto pipeline = getPipelineState(ma);
+                auto pipeline = getPipelineState(mi);
                 if (translucent) {
                     pipeline.rasterState.blendFunctionSrcRGB = BlendFunction::ONE;
                     pipeline.rasterState.blendFunctionSrcAlpha = BlendFunction::ONE;

filament/src/PostProcessManager.h

@@ -357,13 +357,7 @@ public:
 
         void terminate(FEngine& engine) noexcept;
 
-        FMaterial* getMaterial(FEngine& engine, backend::DriverApi& driver,
-                Variant::type_t variant) const noexcept;
-
-        FMaterial* getMaterial(FEngine& engine, backend::DriverApi& driver,
-                PostProcessVariant variant = PostProcessVariant::OPAQUE) const noexcept {
-                return getMaterial(engine, driver, Variant::type_t(variant));
-        }
+        FMaterial* getMaterial(FEngine& engine) const noexcept;
 
     private:
         void loadMaterial(FEngine& engine) const noexcept;
@@ -388,11 +382,12 @@ public:
 
     void bindPostProcessDescriptorSet(backend::DriverApi& driver) const noexcept;
 
-    backend::PipelineState getPipelineState(FMaterial const* ma, Variant::type_t variant) const noexcept;
+    backend::PipelineState getPipelineState(FMaterialInstance const* mi,
+            Variant::type_t variant) const noexcept;
 
-    backend::PipelineState getPipelineState(FMaterial const* ma,
-                    PostProcessVariant variant = PostProcessVariant::OPAQUE) const noexcept {
-            return getPipelineState(ma, Variant::type_t(variant));
+    backend::PipelineState getPipelineState(FMaterialInstance const* mi,
+            PostProcessVariant variant = PostProcessVariant::OPAQUE) const noexcept {
+        return getPipelineState(mi, Variant::type_t(variant));
     }
 
     void renderFullScreenQuad(FrameGraphResources::RenderPassInfo const& out,
@@ -427,17 +422,37 @@ private:
 
     void bindPerRenderableDescriptorSet(backend::DriverApi& driver) const noexcept;
 
-    // Helper to get a MaterialInstance from a FMaterial
-    // This currently just call FMaterial::getDefaultInstance().
-    FMaterialInstance* getMaterialInstance(FMaterial const* ma) {
-        return mMaterialInstanceManager.getMaterialInstance(ma);
+    // Helpers to get MaterialInstances.
+    //
+    // These funcions additionally ensure that the necessary shader programs are compiled via
+    // prepareProgram().
+    FMaterialInstance* getMaterialInstance(backend::DriverApi& driver, FMaterial const* ma,
+            Variant::type_t variant) const;
+
+    FMaterialInstance* getMaterialInstance(backend::DriverApi& driver, FMaterial const* ma,
+            PostProcessVariant variant = PostProcessVariant::OPAQUE) const {
+        return getMaterialInstance(driver, ma, Variant::type_t(variant));
     }
 
-    // Helper to get a MaterialInstance from a PostProcessMaterial.
-    FMaterialInstance* getMaterialInstance(FEngine& engine, backend::DriverApi& driver, PostProcessMaterial const& material,
-            PostProcessVariant variant = PostProcessVariant::OPAQUE) {
-        FMaterial const* ma = material.getMaterial(engine, driver, variant);
-        return getMaterialInstance(ma);
+    FMaterialInstance* getMaterialInstance(FEngine& engine, backend::DriverApi& driver,
+            PostProcessMaterial const& material,
+            PostProcessVariant variant = PostProcessVariant::OPAQUE) const {
+        return getMaterialInstance(driver, material.getMaterial(engine), Variant::type_t(variant));
+    }
+
+    FMaterialInstance* getMaterialInstanceWithTag(backend::DriverApi& driver, FMaterial const* ma,
+            uint32_t tag, Variant::type_t variant) const;
+
+    FMaterialInstance* getMaterialInstanceWithTag(backend::DriverApi& driver, FMaterial const* ma,
+            uint32_t tag, PostProcessVariant variant = PostProcessVariant::OPAQUE) const {
+        return getMaterialInstanceWithTag(driver, ma, tag, Variant::type_t(variant));
+    }
+
+    FMaterialInstance* getMaterialInstanceWithTag(FEngine& engine, backend::DriverApi& driver,
+            PostProcessMaterial const& material, uint32_t tag,
+            PostProcessVariant variant = PostProcessVariant::OPAQUE) const {
+        return getMaterialInstanceWithTag(driver, material.getMaterial(engine), tag,
+                Variant::type_t(variant));
     }
 
     UboManager* getUboManager() const noexcept;
@@ -486,13 +501,6 @@ private:
 
     MaterialInstanceManager mMaterialInstanceManager;
 
-    struct {
-        int32_t colorGradingTranslucent = MaterialInstanceManager::INVALID_FIXED_INDEX;
-        int32_t colorGradingOpaque = MaterialInstanceManager::INVALID_FIXED_INDEX;
-        int32_t customResolve = MaterialInstanceManager::INVALID_FIXED_INDEX;
-        int32_t clearDepth = MaterialInstanceManager::INVALID_FIXED_INDEX;
-    } mFixedMaterialInstanceIndex;
-
     backend::Handle<backend::HwTexture> mStarburstTexture;
 
     std::uniform_real_distribution<float> mUniformDistribution{0.0f, 1.0f};

filament/src/RenderPass.cpp

@@ -243,8 +243,8 @@ void RenderPass::appendCommands(FEngine const& engine, backend::DriverApi& drive
     // This must be done from the main thread.
     for (Command const* first = curr, *last = curr + commandCount ; first != last ; ++first) {
         if (UTILS_LIKELY((first->key & CUSTOM_MASK) == uint64_t(CustomCommand::PASS))) {
-            auto ma = first->info.mi->getMaterial();
-            ma->prepareProgram(driver, first->info.materialVariant, CompilerPriorityQueue::CRITICAL);
+            first->info.mi->prepareProgram(driver, first->info.materialVariant,
+                    CompilerPriorityQueue::CRITICAL);
         }
     }
 }
@@ -426,16 +426,18 @@ void RenderPass::setupColorCommand(Command& cmdDraw, Variant variant,
     bool const isBlendingCommand = !hasScreenSpaceRefraction &&
             (blendingMode != BlendingMode::OPAQUE && blendingMode != BlendingMode::MASKED);
 
+    RasterState rasterState = mi->getRasterState();
+
     uint64_t keyDraw = cmdDraw.key;
     keyDraw &= ~(PASS_MASK | BLENDING_MASK | MATERIAL_MASK);
     keyDraw |= uint64_t(hasScreenSpaceRefraction ? Pass::REFRACT : Pass::COLOR);
     keyDraw |= uint64_t(CustomCommand::PASS);
     keyDraw |= mi->getSortingKey(); // already all set-up for direct or'ing
     keyDraw |= makeField(variant.key, MATERIAL_VARIANT_KEY_MASK, MATERIAL_VARIANT_KEY_SHIFT);
-    keyDraw |= makeField(ma->getRasterState().alphaToCoverage, BLENDING_MASK, BLENDING_SHIFT);
+    keyDraw |= makeField(rasterState.alphaToCoverage, BLENDING_MASK, BLENDING_SHIFT);
 
     cmdDraw.key = isBlendingCommand ? keyBlending : keyDraw;
-    cmdDraw.info.rasterState = ma->getRasterState();
+    cmdDraw.info.rasterState = rasterState;
 
     // for SSR pass, the blending mode of opaques (including MASKED) must be off
     // see Material.cpp.
@@ -446,10 +448,6 @@ void RenderPass::setupColorCommand(Command& cmdDraw, Variant variant,
             BlendFunction::ZERO : cmdDraw.info.rasterState.blendFunctionDstAlpha;
 
     cmdDraw.info.rasterState.inverseFrontFaces = inverseFrontFaces;
-    cmdDraw.info.rasterState.culling = mi->getCullingMode();
-    cmdDraw.info.rasterState.colorWrite = mi->isColorWriteEnabled();
-    cmdDraw.info.rasterState.depthWrite = mi->isDepthWriteEnabled();
-    cmdDraw.info.rasterState.depthFunc = mi->getDepthFunc();
     cmdDraw.info.rasterState.depthClamp = hasDepthClamp;
     cmdDraw.info.materialVariant = variant;
     // we keep "RasterState::colorWrite" to the value set by material (could be disabled)
@@ -568,7 +566,7 @@ RenderPass::Command* RenderPass::generateCommandsImpl(CommandTypeFlags extraFlag
     if constexpr (isDepthPass) {
         cmd.info.materialVariant = variant;
         cmd.info.rasterState = {};
-        cmd.info.rasterState.colorWrite = Variant::isPickingVariant(variant) || Variant::isVSMVariant(variant);
+        cmd.info.rasterState.colorWrite = Variant::isPickingVariant(variant) || Variant::isDepthMomentsVariant(variant);
         cmd.info.rasterState.depthWrite = true;
         cmd.info.rasterState.depthFunc = RasterState::DepthFunc::GE;
         cmd.info.rasterState.alphaToCoverage = false;
@@ -616,7 +614,7 @@ RenderPass::Command* RenderPass::generateCommandsImpl(CommandTypeFlags extraFlag
         bool const hasSkinningOrMorphing = hasSkinning || hasMorphing;
 
         // if we are already an SSR variant, the SRE bit is already set
-        static_assert(Variant::SPECIAL_SSR & Variant::SRE);
+        static_assert(Variant::SPECIAL_SSR_VARIANT & Variant::SRE);
         Variant renderableVariant{ variant };
 
         // we can't have SSR and shadowing together by construction
@@ -798,16 +796,19 @@ RenderPass::Command* RenderPass::generateCommandsImpl(CommandTypeFlags extraFlag
                 cmd.info.rasterState.culling = cullingMode;
 
                 // FIXME: should writeDepthForShadowCasters take precedence over mi->getDepthWrite()?
-                cmd.info.rasterState.depthWrite = (1 // only keep bit 0
-                        & (mi->isDepthWriteEnabled() | (mode == TransparencyMode::TWO_PASSES_ONE_SIDE)
-                                                     | isPickingVariant)
-                                                   & !(filterTranslucentObjects & translucent)
-                                                   & !(depthFilterAlphaMaskedObjects & rs.alphaToCoverage))
-                                                  | writeDepthForShadowCasters;
+                cmd.info.rasterState.depthWrite =
+                        (1 // only keep bit 0
+                                & (mi->isDepthWriteEnabled() |
+                                          (mode == TransparencyMode::TWO_PASSES_ONE_SIDE) |
+                                          isPickingVariant) &
+                                !(depthFilterAlphaMaskedObjects & rs.alphaToCoverage)) |
+                        writeDepthForShadowCasters;
 
                 *curr = cmd;
                 // cancel command if both front and back faces are culled
                 curr->key |= select(cullingMode == CullingMode::FRONT_AND_BACK);
+                // cancel command if asked to filter translucent objects
+                curr->key |= select(filterTranslucentObjects & translucent);
             }
 
             ++curr;
@@ -1087,8 +1088,7 @@ void RenderPass::Executor::execute(FEngine const& engine, DriverApi& driver,
                     mi->use(driver, info.materialVariant);
                 }
 
-                assert_invariant(ma);
-                pipeline.program = ma->getProgram(info.materialVariant);
+                pipeline.program = mi->getProgram(info.materialVariant);
 
                 if (UTILS_UNLIKELY(memcmp(&pipeline, &currentPipeline, sizeof(PipelineState)) != 0)) {
                     currentPipeline = pipeline;

filament/src/RenderableManager.cpp

@@ -54,7 +54,7 @@ RenderableManager::getInstance(Entity const e) const noexcept {
 }
 
 void RenderableManager::destroy(Entity const e) noexcept {
-    return downcast(this)->destroy(e);
+    return downcast(this)->clientDestroy(e);
 }
 
 void RenderableManager::setAxisAlignedBoundingBox(Instance const instance, const Box& aabb) {

filament/src/ShadowMap.cpp

@@ -62,6 +62,8 @@ ShadowMap::ShadowMap(FEngine& engine) noexcept
         : mPerShadowMapUniforms(engine),
           mShadowType(ShadowType::DIRECTIONAL),
           mHasVisibleShadows(false),
+          mVsm(false),
+          mReservedBit(false),
           mFace(0) {
     Entity entities[2];
     engine.getEntityManager().create(2, entities);
@@ -85,7 +87,7 @@ void ShadowMap::terminate(FEngine& engine) {
 
 ShadowMap::~ShadowMap() = default;
 
-void ShadowMap::initialize(size_t const lightIndex, ShadowType const shadowType,
+void ShadowMap::initialize(size_t const lightIndex, ShadowType const shadowType, bool const vsm,
         uint16_t const shadowIndex, uint8_t const face,
         LightManager::ShadowOptions const* options) {
     mLightIndex = lightIndex;
@@ -93,6 +95,7 @@ void ShadowMap::initialize(size_t const lightIndex, ShadowType const shadowType,
     mOptions = options;
     mShadowType = shadowType;
     mFace = face;
+    mVsm = vsm;
 }
 
 mat4f ShadowMap::getDirectionalLightViewMatrix(float3 direction, float3 up,
@@ -213,12 +216,14 @@ ShadowMap::ShaderParameters ShadowMap::updateDirectional(FEngine& engine,
      * Final shadow map transform
      */
 
-    // Final shadow transform (focused warped light-space)
+    // Final shadow transform (focused warped light-space), world space to clip space
     const mat4f S = F * (W * LMpMv);
 
     // Computes St the transform to use in the shader to access the shadow map texture
     // i.e. it transforms a world-space vertex to a texture coordinate in the shadowmap
     const auto [Mt, Mn] = getTextureCoordsMapping(shadowMapInfo, getViewport());
+
+    // world space to texture atlas space
     const mat4f St = highPrecisionMultiply(Mt, S);
 
     ShaderParameters shaderParameters;
@@ -226,14 +231,12 @@ ShadowMap::ShaderParameters ShadowMap::updateDirectional(FEngine& engine,
 
     // note: in texelSizeWorldSpace() below, we can use Mb * Mt * F * W because
     // L * Mp * Mv is a rigid transform for directional lights, and doesn't matter.
-    // if Wp[3][1] is 0, then LISPSM was cancelled.
+    // if Wp[3][1] is 0, then LiSPSM was canceled.
     if (useLispsm && Wp[3][1] != 0.0f) {
-        shaderParameters.texelSizeAtOneMeterWs =
-                texelSizeWorldSpace(Wp, mat4f(Mt * F), shadowMapInfo.shadowDimension);
+        shaderParameters.texelSizeAtOneMeterWs = texelSizeWorldSpace(S, shadowMapInfo.shadowDimension);
     } else {
         // We know we're using an ortho projection
-        shaderParameters.texelSizeAtOneMeterWs =
-                texelSizeWorldSpace(St.upperLeft(), shadowMapInfo.shadowDimension);
+        shaderParameters.texelSizeAtOneMeterWs = texelSizeWorldSpace(S.upperLeft(), shadowMapInfo.shadowDimension);
     }
     if (!shadowMapInfo.vsm) {
         shaderParameters.lightSpace = St;
@@ -1117,90 +1120,72 @@ bool ShadowMap::intersectSegmentWithPlanarQuad(float3& UTILS_RESTRICT p,
     return hit;
 }
 
-float ShadowMap::texelSizeWorldSpace(const mat3f& worldToShadowTexture,
-        uint16_t shadowDimension) noexcept {
+float2 ShadowMap::texelSizeWorldSpace(const mat3f& clipFromWorld, uint16_t shadowDimension) noexcept {
     // The Jacobian of the transformation from texture-to-world is the matrix itself for
-    // orthographic projections. We just need to inverse worldToShadowTexture,
+    // orthographic projections. We just need to inverse shadowMapFromWorld,
     // which is guaranteed to be orthographic.
     // The two first columns give us how a texel maps in world-space.
-    const float ures = 1.0f / float(shadowDimension);
-    const float vres = 1.0f / float(shadowDimension);
-    const mat3f shadowTextureToWorld(inverse(worldToShadowTexture));
-    const float3 Jx = shadowTextureToWorld[0];
-    const float3 Jy = shadowTextureToWorld[1];
-    const float s = std::max(length(Jx) * ures, length(Jy) * vres);
+    float const oneTexel = 2.0f / float(shadowDimension);
+    mat3f const worldFromClip(inverse(clipFromWorld));
+    float3 const Jx = worldFromClip[0];
+    float3 const Jy = worldFromClip[1];
+    float2 const s = float2{ length(Jx), length(Jy) } * oneTexel;
     return s;
 }
 
-float ShadowMap::texelSizeWorldSpace(const mat4f& Wp, const mat4f& MbMtF,
-        uint16_t shadowDimension) noexcept {
-    // Here we compute the Jacobian of inverse(MbMtF * Wp).
-    // The expression below has been computed with Mathematica. However, it's not very hard,
-    // albeit error-prone, to do it by hand because MbMtF is a linear transform.
-    // So we really only need to calculate the Jacobian of inverse(Wp) at inverse(MbMtF).
-    //
-    // Because we're only interested in the length of the columns of the Jacobian, we can use
-    // Mb * Mt * F * Wp instead of the full expression Mb * Mt * F * Wp * Wv * L * Mp * Mv,
-    // because Wv * L * Mp * Mv is a rigid transform, which doesn't affect the length of
-    // the Jacobian's column vectors.
+/**
+ * Calculates the Jacobian matrix J = ∂(u,v,d)/∂(x,y,z) for a 4x4 perspective projection.
+ *
+ * Given a view-space point P = [x, y, z, 1]^T and a projection matrix M,
+ * the projected homogeneous coordinates are [X, Y, Z, W]^T = M * P.
+ * The resulting NDC coordinates are u = X/W, v = Y/W, and depth d = Z/W.
+ *
+ * To find the Jacobian on the CPU, we apply the quotient rule to each component:
+ * ∂(X/W) / ∂xi = ( (∂X/∂xi) * W - X * (∂W/∂xi) ) / W^2
+ *
+ * In matrix form, this can be expressed as:
+ * J = (1/W) * [ M_sub - (1/W) * (T ⊗ w_grad) ]
+ *
+ * Where:
+ * - W: The homogeneous w-component after projection (usually -z for standard mats).
+ * - M_sub: The top-left 3x3 submatrix of M.
+ * - T: The column vector [X, Y, Z]^T (pre-perspective divide).
+ * - w_grad: The row vector [m30, m31, m32] (the first three elements of M's last row).
+ * - ⊗: The outer product, resulting in a 3x3 matrix.
+ *
+ * This Jacobian describes the local "stretch" of the projection. For LiSPSM or
+ * shadow mapping, the inverse Jacobian J^-1 provides the world-space footprint
+ * of a shadow texel, which is essential for calculating an accurate,
+ * non-constant depth bias to eliminate shadow acne.
+ *
+ * @param M The 4x4 projection matrix.
+ * @param p The 3D point in view-space where the Jacobian is evaluated.
+ * @return A 3x3 matrix representing the partial derivatives of (u,v,d) w.r.t (x,y,z).
+ */
+static mat3f jacobian(mat4f const& M, float3 const& p) noexcept {
+    float4 const T = M * p;
+    mat3f const M_sub = M.upperLeft();
+    float3 const w_grad = { M[0].w, M[1].w, M[2].w };
+    mat3f const t_cross_w{
+        w_grad.x * T.xyz,
+        w_grad.y * T.xyz,
+        w_grad.z * T.xyz
+    };
+    return (M_sub - t_cross_w / T.w) / T.w;
+}
 
+float2 ShadowMap::texelSizeWorldSpace(mat4f const& S, uint16_t const shadowDimension) noexcept {
     // The Jacobian is not constant, so we evaluate it in the center of the shadow-map texture.
     // It might be better to do this computation in the vertex shader.
-    float3 const p = { 0.5f, 0.5f, 0.0f };
-
-    const float ures = 1.0f / float(shadowDimension);
-    const float vres = 1.0f / float(shadowDimension);
-    const float dres = 1.0f / 65536.0f;
-
-    constexpr bool JACOBIAN_ESTIMATE = false;
-    if constexpr (JACOBIAN_ESTIMATE) {
-        // This estimates the Jacobian -- this is a lot heavier. This is mostly for reference
-        // and testing.
-        const mat4f Si(inverse(MbMtF * Wp));
-        const float3 p0 = mat4f::project(Si, p);
-        const float3 p1 = mat4f::project(Si, p + float3{ 1, 0, 0 } * ures);
-        const float3 p2 = mat4f::project(Si, p + float3{ 0, 1, 0 } * vres);
-        const float3 p3 = mat4f::project(Si, p + float3{ 0, 0, 1 } * dres);
-        const float3 Jx = p1 - p0;
-        const float3 Jy = p2 - p0;
-        const float3 UTILS_UNUSED Jz = p3 - p0;
-        const float s = std::max(length(Jx), length(Jy));
-        return s;
-    }
-
-    const float n = Wp[0][0];
-    const float A = Wp[1][1];
-    const float B = Wp[3][1];
-    const float sx = MbMtF[0][0];
-    const float sy = MbMtF[1][1];
-    const float sz = MbMtF[2][2];
-    const float ox = MbMtF[3][0];
-    const float oy = MbMtF[3][1];
-    const float oz = MbMtF[3][2];
-
-    const float X = p.x - ox;
-    const float Y = p.y - oy;
-    const float Z = p.z - oz;
-
-    const float dz = A * sy - Y;
-    const float nsxsz = n * sx * sz;
-    const float j = -(B * sy) / (nsxsz * dz * dz);
-    const mat3f J(mat3f::row_major_init{
-            j * dz * sz,    j * X * sz,     0.0f,
-            0.0f,           j * nsxsz,      0.0f,
-            0.0f,           j * Z * sx,     j * dz * sx
-    });
-
-    float3 const Jx = J[0] * ures;
-    float3 const Jy = J[1] * vres;
-    UTILS_UNUSED float3 const Jz = J[2] * dres;
-    const float s = std::max(length(Jx), length(Jy));
+    float3 const p = { 0.0f, 0.0f, 0.0f }; // clip-space
+    float const oneTexel = 2.0f / float(shadowDimension);
+    mat3f const J = jacobian(inverse(S), p);
+    float2 const s = float2{ length(J[0]), length(J[1]) } * oneTexel;
     return s;
 }
 
-template<typename Casters, typename Receivers>
-void ShadowMap::visitScene(const FScene& scene, uint32_t const visibleLayers,
-        Casters casters, Receivers receivers) noexcept {
+template<typename Visitor>
+void ShadowMap::visitScene(const FScene& scene, uint32_t const visibleLayers, Visitor visitor) noexcept {
     FILAMENT_TRACING_CALL(FILAMENT_TRACING_CATEGORY_FILAMENT);
 
     using State = FRenderableManager::Visibility;
@@ -1213,14 +1198,11 @@ void ShadowMap::visitScene(const FScene& scene, uint32_t const visibleLayers,
     size_t const c = soa.size();
     for (size_t i = 0; i < c; i++) {
         if (layers[i] & visibleLayers) {
-            const Aabb aabb{ worldAABBCenter[i] - worldAABBExtent[i],
-                             worldAABBCenter[i] + worldAABBExtent[i] };
-            if (visibility[i].castShadows) {
-                casters(aabb, visibleMasks[i]);
-            }
-            if (visibility[i].receiveShadows) {
-                receivers(aabb, visibleMasks[i]);
-            }
+            Aabb const aabb{
+                worldAABBCenter[i] - worldAABBExtent[i],
+                worldAABBCenter[i] + worldAABBExtent[i]
+            };
+            visitor(aabb, visibleMasks[i], visibility[i]);
         }
     }
 }
@@ -1239,13 +1221,15 @@ ShadowMap::SceneInfo::SceneInfo(
     wsShadowCastersVolume = {};
     wsShadowReceiversVolume = {};
     visitScene(scene, visibleLayers,
-            [&](Aabb caster, Culler::result_type) {
-                wsShadowCastersVolume.min = min(wsShadowCastersVolume.min, caster.min);
-                wsShadowCastersVolume.max = max(wsShadowCastersVolume.max, caster.max);
-            },
-            [&](Aabb receiver, Culler::result_type) {
-                wsShadowReceiversVolume.min = min(wsShadowReceiversVolume.min, receiver.min);
-                wsShadowReceiversVolume.max = max(wsShadowReceiversVolume.max, receiver.max);
+            [this](Aabb const& aabb, Culler::result_type, FRenderableManager::Visibility const visibility) {
+                if (visibility.castShadows) {
+                    wsShadowCastersVolume.min = min(wsShadowCastersVolume.min, aabb.min);
+                    wsShadowCastersVolume.max = max(wsShadowCastersVolume.max, aabb.max);
+                }
+                if (visibility.receiveShadows) {
+                    wsShadowReceiversVolume.min = min(wsShadowReceiversVolume.min, aabb.min);
+                    wsShadowReceiversVolume.max = max(wsShadowReceiversVolume.max, aabb.max);
+                }
             }
     );
 }
@@ -1259,18 +1243,20 @@ void ShadowMap::updateSceneInfoDirectional(mat4f const& Mv, FScene const& scene,
     sceneInfo.lsCastersNearFar = { std::numeric_limits<float>::lowest(), std::numeric_limits<float>::max() };
     sceneInfo.lsReceiversNearFar = { std::numeric_limits<float>::lowest(), std::numeric_limits<float>::max() };
     visitScene(scene, sceneInfo.visibleLayers,
-            [&](Aabb caster, Culler::result_type) {
-                auto r = Aabb::transform(Mv.upperLeft(), Mv[3].xyz, caster);
-                sceneInfo.lsCastersNearFar.x = max(sceneInfo.lsCastersNearFar.x, r.max.z);
-                sceneInfo.lsCastersNearFar.y = min(sceneInfo.lsCastersNearFar.y, r.min.z);
-            },
-            [&](Aabb receiver, Culler::result_type const vis) {
-                // account only for objects that are visible by the camera
-                auto mask = 1u << VISIBLE_RENDERABLE_BIT;
-                if ((vis & mask) == mask) {
-                    auto r = Aabb::transform(Mv.upperLeft(), Mv[3].xyz, receiver);
-                    sceneInfo.lsReceiversNearFar.x = max(sceneInfo.lsReceiversNearFar.x, r.max.z);
-                    sceneInfo.lsReceiversNearFar.y = min(sceneInfo.lsReceiversNearFar.y, r.min.z);
+            [&](Aabb const& aabb, Culler::result_type const vis, FRenderableManager::Visibility const visibility) {
+                if (visibility.castShadows) {
+                    auto const r = Aabb::transform(Mv.upperLeft(), Mv[3].xyz, aabb);
+                    sceneInfo.lsCastersNearFar.x = max(sceneInfo.lsCastersNearFar.x, r.max.z);
+                    sceneInfo.lsCastersNearFar.y = min(sceneInfo.lsCastersNearFar.y, r.min.z);
+                }
+                if (visibility.castShadows) {
+                    // account only for objects that are visible by the camera
+                    constexpr auto mask = 1u << VISIBLE_RENDERABLE_BIT;
+                    if ((vis & mask) == mask) {
+                        auto r = Aabb::transform(Mv.upperLeft(), Mv[3].xyz, aabb);
+                        sceneInfo.lsReceiversNearFar.x = max(sceneInfo.lsReceiversNearFar.x, r.max.z);
+                        sceneInfo.lsReceiversNearFar.y = min(sceneInfo.lsReceiversNearFar.y, r.min.z);
+                    }
                 }
             }
     );
@@ -1285,15 +1271,15 @@ void ShadowMap::updateSceneInfoSpot(mat4f const& Mv, FScene const& scene,
     sceneInfo.lsCastersNearFar = { std::numeric_limits<float>::lowest(), std::numeric_limits<float>::max() };
     // account only for objects that are visible by both the camera and the light
     visitScene(scene, sceneInfo.visibleLayers,
-            [&](Aabb caster, Culler::result_type const vis) {
-                auto mask = VISIBLE_DYN_SHADOW_RENDERABLE;
-                if ((vis & mask) == mask) {
-                    auto r = Aabb::transform(Mv.upperLeft(), Mv[3].xyz, caster);
-                    sceneInfo.lsCastersNearFar.x = std::max(sceneInfo.lsCastersNearFar.x, r.max.z);  // near
-                    sceneInfo.lsCastersNearFar.y = std::min(sceneInfo.lsCastersNearFar.y, r.min.z);  // far
+            [&](Aabb const& aabb, Culler::result_type const vis, FRenderableManager::Visibility const visibility) {
+                if (visibility.castShadows) {
+                    constexpr auto mask = VISIBLE_DYN_SHADOW_RENDERABLE;
+                    if ((vis & mask) == mask) {
+                        auto const r = Aabb::transform(Mv.upperLeft(), Mv[3].xyz, aabb);
+                        sceneInfo.lsCastersNearFar.x = std::max(sceneInfo.lsCastersNearFar.x, r.max.z);  // near
+                        sceneInfo.lsCastersNearFar.y = std::min(sceneInfo.lsCastersNearFar.y, r.min.z);  // far
+                    }
                 }
-            },
-            [&](Aabb receiver, Culler::result_type) {
             }
     );
 }
@@ -1303,47 +1289,56 @@ void ShadowMap::setAllocation(uint8_t const layer, backend::Viewport viewport) n
 }
 
 backend::Viewport ShadowMap::getViewport() const noexcept {
-    // We set a viewport with a 1-texel border for when we index outside the texture.
-    // This happens only for directional lights when "focus shadow casters" is used,
-    // or when shadowFar is smaller than the camera far.
-    // For spot- and point-lights we also use a 1-texel border, so that bilinear filtering
-    // can work properly if the shadowmap is in an atlas (and we can't rely on h/w clamp).
+    // This is used for calculating the light space; in particular, if the shadowmap is in a 2D atlas, this
+    // returns the valid area of the shadowmap. By definition all values must be integer.
+
+    // We set a viewport with a 1-texel border for when we index outside the texture, which should only happen for
+    // directional lights when "focus shadow casters" is used, or when shadowFar is smaller than the camera far.
+    // For directional lights, the border is filed with "fully lit".
+    //
+    // For point-lights, we also use a 1-texel border, but for a different reason; the border is filled with data
+    // so that bilinear (PCF) filtering works properly.
+    //
+    // Spot-light a treated like point lights (the border will be filled with "fully-lit" anyways)
+
     const uint32_t dim = mOptions->mapSize;
     const uint16_t border = 1u;
     return { mOffset.x + border, mOffset.y + border, dim - 2u * border, dim - 2u * border };
 }
 
 backend::Viewport ShadowMap::getScissor() const noexcept {
-    // We set a viewport with a 1-texel border for when we index outside the texture.
-    // This happens only for directional lights when "focus shadow casters" is used,
-    // or when shadowFar is smaller than the camera far.
-    // For spot- and point-lights we also use a 1-texel border, so that bilinear filtering
-    // can work properly if the shadowmap is in an atlas (and we can't rely on h/w clamp), so we
-    // don't scissor the border, so it gets filled with correct neighboring texels.
+    // This is used while rendering the shadowmap.
+
     const uint32_t dim = mOptions->mapSize;
     const uint16_t border = 1u;
-    switch (mShadowType) {
-        case ShadowType::DIRECTIONAL:
-            return { mOffset.x + border, mOffset.y + border, dim - 2u * border, dim - 2u * border };
+    switch (mShadowType) { // NOLINT(*-multiway-paths-covered)
+        case ShadowType::DIRECTIONAL: {
+            // Don't render anything into the border; it's already filled with "fully lit".
+            return {mOffset.x + border, mOffset.y + border, dim - 2u * border, dim - 2u * border};
+        }
         case ShadowType::SPOT:
-        case ShadowType::POINT:
-            return { mOffset.x, mOffset.y, dim, dim };
+        case ShadowType::POINT: {
+            // Render into the border (which will get the adjacent faces texels)
+            return {mOffset.x, mOffset.y, dim, dim};
+        }
     }
+    return {};
 }
 
 float4 ShadowMap::getClampToEdgeCoords(ShadowMapInfo const& shadowMapInfo) const noexcept {
-    float border; // shadowmap border in texels
-    switch (mShadowType) {
+    // This is used to clamp the texture coordinates when sampling the shadowmap
+
+    float border = 0; // shadowmap border in texels
+    switch (mShadowType) { // NOLINT(*-multiway-paths-covered)
         case ShadowType::DIRECTIONAL:
-            // For directional lights, we need to allow the sampling to reach the border, it
-            // happens when "focus shadow casters" is used for instance.
+            // For directional lights, we need to allow the sampling to reach the center of border texels,
+            // but no further, so we don't read into the adjacent texture in the 2D atlas (taking bilinear filtering
+            // into account).
             border = 0.5f;
             break;
         case ShadowType::SPOT:
         case ShadowType::POINT:
-            // For spot and point light, this is equal to the viewport. i.e. the valid
-            // texels are inside the viewport (w/ 1-texel border), the border will be used
-            // for bilinear filtering.
+            // For point-light, the border is only needed for bilinear filtering (of the other faces)
             border = 1.0f;
             break;
     }
@@ -1372,8 +1367,8 @@ void ShadowMap::prepareCamera(Transaction const& transaction,
 }
 
 void ShadowMap::prepareViewport(Transaction const& transaction,
-        backend::Viewport const& viewport) noexcept {
-    ShadowMapDescriptorSet::prepareViewport(transaction, viewport);
+        backend::Viewport const& physicalViewport, backend::Viewport const& logicalViewport) noexcept {
+    ShadowMapDescriptorSet::prepareViewport(transaction, physicalViewport, logicalViewport);
 }
 
 void ShadowMap::prepareTime(Transaction const& transaction,
@@ -1387,8 +1382,8 @@ void ShadowMap::prepareMaterialGlobals(Transaction const& transaction,
 }
 
 void ShadowMap::prepareShadowMapping(Transaction const& transaction,
-        bool const highPrecision) noexcept {
-    ShadowMapDescriptorSet::prepareShadowMapping(transaction, highPrecision);
+        float const vsmExponent, float const vsmMaxMoment) noexcept {
+    ShadowMapDescriptorSet::prepareShadowMapping(transaction, vsmExponent, vsmMaxMoment);
 }
 
 ShadowMapDescriptorSet::Transaction ShadowMap::open(DriverApi& driver) noexcept {

filament/src/ShadowMap.h

@@ -128,14 +128,14 @@ public:
     static math::mat4f getPointLightViewMatrix(backend::TextureCubemapFace face,
             math::float3 position) noexcept;
 
-    void initialize(size_t lightIndex, ShadowType shadowType, uint16_t shadowIndex, uint8_t face,
+    void initialize(size_t lightIndex, ShadowType shadowType, bool vsm, uint16_t shadowIndex, uint8_t face,
             LightManager::ShadowOptions const* options);
 
     struct ShaderParameters {
         math::mat4f lightSpace{};
         math::float4 lightFromWorldZ{};
         math::float4 scissorNormalized{};
-        float texelSizeAtOneMeterWs{};
+        math::float2 texelSizeAtOneMeterWs{};
     };
 
     // Call once per frame if the light, scene (or visible layers) or camera changes.
@@ -173,7 +173,10 @@ public:
     static void updateSceneInfoSpot(const math::mat4f& Mv, FScene const& scene,
             SceneInfo& sceneInfo);
 
-    LightManager::ShadowOptions const* getShadowOptions() const noexcept { return mOptions; }
+    LightManager::ShadowOptions const& getShadowOptions() const noexcept {
+            assert_invariant(mOptions);
+            return *mOptions;
+    }
     size_t getLightIndex() const { return mLightIndex; }
     uint16_t getShadowIndex() const { return mShadowIndex; }
     void setAllocation(uint8_t layer, backend::Viewport viewport) noexcept;
@@ -193,13 +196,13 @@ public:
     static void prepareCamera(Transaction const& transaction,
             FEngine const& engine, const CameraInfo& cameraInfo) noexcept;
     static void prepareViewport(Transaction const& transaction,
-            backend::Viewport const& viewport) noexcept;
+            backend::Viewport const& physicalViewport, backend::Viewport const& logicalViewport) noexcept;
     static void prepareTime(Transaction const& transaction,
             FEngine const& engine, math::float4 const& userTime) noexcept;
     static void prepareMaterialGlobals(Transaction const& transaction,
             std::array<math::float4, 4> const& materialGlobals) noexcept;
     static void prepareShadowMapping(Transaction const& transaction,
-            bool highPrecision) noexcept;
+            float vsmExponent, float vsmMaxMoment) noexcept;
     static ShadowMapDescriptorSet::Transaction open(backend::DriverApi& driver) noexcept;
     void commit(Transaction& transaction, FEngine& engine, backend::DriverApi& driver) const noexcept;
     void bind(backend::DriverApi& driver) const noexcept;
@@ -273,9 +276,8 @@ private:
     static inline math::float4 computeBoundingSphere(
             math::float3 const* vertices, size_t count) noexcept;
 
-    template<typename Casters, typename Receivers>
-    static void visitScene(FScene const& scene, uint32_t visibleLayers,
-            Casters casters, Receivers receivers) noexcept;
+    template<typename Visitor>
+    static void visitScene(FScene const& scene, uint32_t visibleLayers, Visitor visitor) noexcept;
 
     static inline Aabb compute2DBounds(const math::mat4f& lightView,
             math::float3 const* wsVertices, size_t count) noexcept;
@@ -319,11 +321,8 @@ private:
 
     math::float4 getClampToEdgeCoords(ShadowMapInfo const& shadowMapInfo) const noexcept;
 
-    static float texelSizeWorldSpace(const math::mat3f& worldToShadowTexture,
-            uint16_t shadowDimension) noexcept;
-
-    static float texelSizeWorldSpace(const math::mat4f& W, const math::mat4f& MbMtF,
-            uint16_t shadowDimension) noexcept;
+    static math::float2 texelSizeWorldSpace(const math::mat3f& clipFromWorld, uint16_t shadowDimension) noexcept;
+    static math::float2 texelSizeWorldSpace(const math::mat4f& clipFromWorld, uint16_t shadowDimension) noexcept;
 
     static constexpr Segment sBoxSegments[12] = {
             { 0, 1 }, { 1, 3 }, { 3, 2 }, { 2, 0 },
@@ -352,9 +351,11 @@ private:
     uint8_t mLayer = 0;         // our layer in the shadowMap texture       // 1
     ShadowType mShadowType  : 2;                                            // :2
     bool mHasVisibleShadows : 1;                                            // :1
+    bool mVsm               : 1;                                            // :1
+    UTILS_UNUSED bool mReservedBit : 1;                                     // :1
     uint8_t mFace           : 3;                                            // :3
     math::ushort2 mOffset{};                                                // 4
-    UTILS_UNUSED uint8_t reserved[4];                                       // 4
+    UTILS_UNUSED uint8_t reserved[4] = {};                                  // 4
 };
 
 } // namespace filament

filament/src/ShadowMapManager.cpp

@@ -151,12 +151,14 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::update(
 
     mDirectionalShadowMapCount = builder.mDirectionalShadowMapCount;
     mSpotShadowMapCount = builder.mSpotShadowMapCount;
+    const bool vsm = view.hasVSM();
 
     for (auto const& entry : builder.mShadowMaps) {
         auto& shadowMap = getShadowMap(entry.shadowIndex);
         shadowMap.initialize(
                 entry.lightIndex,
                 entry.shadowType,
+                vsm,
                 entry.shadowIndex,
                 entry.face,
                 entry.options);
@@ -230,15 +232,12 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FEngine& engine, FrameG
         FView& view, CameraInfo const& mainCameraInfo,
         float4 const& userTime) noexcept {
 
-    const float moment2 = std::numeric_limits<half>::max();
-    const float moment1 = std::sqrt(moment2);
-    const float4 vsmClearColor{ moment1, moment2, -moment1, moment2 };
-
     FScene* scene = view.getScene();
     assert_invariant(scene);
 
     // make a copy here, because it's a very small structure
-    const TextureAtlasRequirements textureRequirements = mTextureAtlasRequirements;
+    VsmShadowOptions const& vsmShadowOptions = view.getVsmShadowOptions();
+    TextureAtlasRequirements const textureRequirements = mTextureAtlasRequirements;
     assert_invariant(textureRequirements.layers <= CONFIG_MAX_SHADOW_LAYERS);
 
     // -------------------------------------------------------------------------------------------
@@ -258,8 +257,6 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FEngine& engine, FrameG
         utils::FixedCapacityVector<ShadowPass> passList;
     };
 
-    VsmShadowOptions const& vsmShadowOptions = view.getVsmShadowOptions();
-
     auto& prepareShadowPass = fg.addPass<PrepareShadowPassData>("Prepare Shadow Pass",
             [&](FrameGraph::Builder& builder, auto& data) {
                 data.passList.reserve(getMaxShadowMapCount());
@@ -392,16 +389,18 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FEngine& engine, FrameG
 
                     auto transaction = ShadowMap::open(driver);
                     ShadowMap::prepareCamera(transaction, engine, cameraInfo);
-                    ShadowMap::prepareViewport(transaction, shadowMap.getViewport());
+                    ShadowMap::prepareViewport(transaction,
+                            { 0, 0, textureRequirements.size, textureRequirements.size },
+                            shadowMap.getViewport());
                     ShadowMap::prepareTime(transaction, engine, userTime);
                     ShadowMap::prepareMaterialGlobals(transaction, view.getMaterialGlobals());
                     ShadowMap::prepareShadowMapping(transaction,
-                            vsmShadowOptions.highPrecision);
+                            getWrapExponentEVSM(vsmShadowOptions, shadowMap.getShadowOptions()),
+                            getMaxMomentEVSM(vsmShadowOptions));
                     shadowMap.commit(transaction, engine, driver);
 
                     // updatePrimitivesLod must be run before RenderPass::appendCommands.
-                    FView::updatePrimitivesLod(scene->getRenderableData(),
-                            engine, cameraInfo, entry.range);
+                    FView::updatePrimitivesLod(scene->getRenderableData(), engine, cameraInfo, entry.range);
 
                     // generate and sort the commands for rendering the shadow map
 
@@ -430,10 +429,10 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FEngine& engine, FrameG
                     entry.executor = pass.getExecutor();
 
                     if (!view.hasVSM()) {
-                        auto const* options = shadowMap.getShadowOptions();
+                        auto const& options = shadowMap.getShadowOptions();
                         PolygonOffset const polygonOffset = { // handle reversed Z
-                                .slope    = -options->polygonOffsetSlope,
-                                .constant = -options->polygonOffsetConstant
+                                .slope    = -options.polygonOffsetSlope,
+                                .constant = -options.polygonOffsetConstant
                         };
                         entry.executor.overridePolygonOffset(&polygonOffset);
                     }
@@ -465,10 +464,10 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FEngine& engine, FrameG
         auto const& entry = *first;
 
         const uint8_t layer = entry.shadowMap->getLayer();
-        const auto* options = entry.shadowMap->getShadowOptions();
+        const auto& options = entry.shadowMap->getShadowOptions();
         const auto msaaSamples = textureRequirements.msaaSamples;
         const bool blur = entry.shadowMap->hasVisibleShadows() &&
-                view.hasVSM() && options->vsm.blurWidth > 0.0f;
+                view.hasVSM() && options.vsm.blurWidth > 0.0f;
 
         auto last = first;
         // loop over each shadow pass to find its layer range
@@ -508,10 +507,9 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FEngine& engine, FrameG
 
                         renderTargetDesc.attachments.color[0] = data.output;
                         renderTargetDesc.attachments.depth = depth;
-                        renderTargetDesc.clearFlags =
-                                TargetBufferFlags::COLOR | TargetBufferFlags::DEPTH;
+                        renderTargetDesc.clearFlags = TargetBufferFlags::COLOR | TargetBufferFlags::DEPTH;
                         // we need to clear the shadow map with the max EVSM moments
-                        renderTargetDesc.clearColor = vsmClearColor;
+                        renderTargetDesc.clearColor = textureRequirements.clearColor;
                         renderTargetDesc.samples = msaaSamples;
 
                         if (UTILS_UNLIKELY(blur)) {
@@ -530,16 +528,14 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FEngine& engine, FrameG
                                     .attachments = {
                                             .color = { data.tempBlurSrc },
                                             .depth = depth },
-                                    .clearColor = vsmClearColor,
+                                    .clearColor = textureRequirements.clearColor,
                                     .samples = msaaSamples,
-                                    .clearFlags = TargetBufferFlags::COLOR
-                                            | TargetBufferFlags::DEPTH
+                                    .clearFlags = TargetBufferFlags::COLOR | TargetBufferFlags::DEPTH
                             });
                         }
                     } else {
                         // the shadowmap layer
-                        data.output = builder.write(data.output,
-                                FrameGraphTexture::Usage::DEPTH_ATTACHMENT);
+                        data.output = builder.write(data.output, FrameGraphTexture::Usage::DEPTH_ATTACHMENT);
                         renderTargetDesc.attachments.depth = data.output;
                         renderTargetDesc.clearFlags = TargetBufferFlags::DEPTH;
                     }
@@ -594,27 +590,27 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FEngine& engine, FrameG
             //        the whole layer. Blurring should happen per shadowmap, not for the whole
             //        layer.
 
-            const float blurWidth = options->vsm.blurWidth;
+            // FIXME: this Gaussian blur is not precise enough for EVSM
+
+            const float blurWidth = options.vsm.blurWidth;
             if (blurWidth > 0.0f) {
                 const float sigma = (blurWidth + 1.0f) / 6.0f;
-                size_t kernelWidth = std::ceil((blurWidth - 5.0f) / 4.0f);
+                size_t kernelWidth = size_t(std::ceil((blurWidth - 5.0f) / 4.0f));
                 kernelWidth = kernelWidth * 4 + 5;
                 ppm.gaussianBlurPass(fg,
                         shadowPass->tempBlurSrc,
                         shadowPass->output,
                         false, kernelWidth, sigma);
             }
+        }
 
-            // FIXME: mipmapping here is broken because it'll access texels from adjacent
-            //        shadow maps.
-
-            // If the shadow texture has more than one level, mipmapping was requested, either directly
-            // or indirectly via anisotropic filtering.
-            // So generate the mipmaps for each layer
-            if (textureRequirements.levels > 1) {
-                for (size_t level = 0; level < textureRequirements.levels - 1; level++) {
-                    ppm.vsmMipmapPass(fg, prepareShadowPass->shadows, layer, level, vsmClearColor);
-                }
+        // If the shadow texture has more than one level, mipmapping was requested, either directly
+        // or indirectly via anisotropic filtering.
+        // So generate the mipmaps for each layer
+        if (UTILS_UNLIKELY(textureRequirements.levels > 1)) {
+            auto& ppm = engine.getPostProcessManager();
+            for (size_t level = 0; level < textureRequirements.levels - 1; level++) {
+                ppm.vsmMipmapPass(fg, prepareShadowPass->shadows, layer, level, textureRequirements.clearColor);
             }
         }
     }
@@ -626,7 +622,8 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
         FView& view, CameraInfo cameraInfo, FScene::RenderableSoa& renderableData,
         FScene::LightSoa const& lightData, ShadowMap::SceneInfo sceneInfo) noexcept {
 
-    FScene* scene = view.getScene();
+    FScene const* const scene = view.getScene();
+    auto const& vsmShadowOptions = view.getVsmShadowOptions();
     auto& lcm = engine.getLightManager();
 
     FLightManager::Instance const directionalLight = lightData.elementAt<FScene::LIGHT_INSTANCE>(0);
@@ -650,7 +647,7 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
     };
 
     bool hasVisibleShadows = false;
-    utils::Slice<ShadowMap> cascadedShadowMaps = getCascadedShadowMap();
+    utils::Slice<ShadowMap> const cascadedShadowMaps = getCascadedShadowMap();
     if (!cascadedShadowMaps.empty()) {
         // Even if we have more than one cascade, we cull directional shadow casters against the
         // entire camera frustum, as if we only had a single cascade.
@@ -742,17 +739,17 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
 
                 // Texel size is constant for directional light (although that's not true when LISPSM
                 // is used, but in that case we're pretending it is).
-                const float wsTexelSize = shaderParameters.texelSizeAtOneMeterWs;
+                float2 const wsTexelSize = shaderParameters.texelSizeAtOneMeterWs;
 
                 auto& s = mShadowUb.edit();
                 s.shadows[shadowIndex].layer = shadowMap.getLayer();
                 s.shadows[shadowIndex].lightFromWorldMatrix = shaderParameters.lightSpace;
                 s.shadows[shadowIndex].scissorNormalized = shaderParameters.scissorNormalized;
-                s.shadows[shadowIndex].normalBias = normalBias * wsTexelSize;
-                s.shadows[shadowIndex].texelSizeAtOneMeter = wsTexelSize;
+                s.shadows[shadowIndex].normalBias = wsTexelSize * normalBias;
                 s.shadows[shadowIndex].elvsm = options.vsm.elvsm;
+                s.shadows[shadowIndex].vsmExponent = getWrapExponentEVSM(vsmShadowOptions, options);
                 s.shadows[shadowIndex].bulbRadiusLs =
-                        mSoftShadowOptions.penumbraScale * options.shadowBulbRadius / wsTexelSize;
+                        mSoftShadowOptions.penumbraScale * options.shadowBulbRadius / length(wsTexelSize);
 
                 shadowTechnique |= ShadowTechnique::SHADOW_MAP;
                 cascadeHasVisibleShadows |= 0x1u << i;
@@ -802,10 +799,8 @@ void ShadowMapManager::updateSpotVisibilityMasks(
         const bool visSpotShadowRenderable = v.castShadows && inVisibleLayer &&
                 (!v.culling || (mask & VISIBLE_DYN_SHADOW_RENDERABLE));
 
-        using Type = Culler::result_type;
-
-        visibleMask[i] &= ~Type(VISIBLE_DYN_SHADOW_RENDERABLE);
-        visibleMask[i] |= Type(visSpotShadowRenderable << VISIBLE_DYN_SHADOW_RENDERABLE_BIT);
+        visibleMask[i] &= ~Culler::result_type(VISIBLE_DYN_SHADOW_RENDERABLE);
+        visibleMask[i] |= Culler::result_type(visSpotShadowRenderable << VISIBLE_DYN_SHADOW_RENDERABLE_BIT);
     }
 }
 
@@ -814,13 +809,14 @@ void ShadowMapManager::prepareSpotShadowMap(ShadowMap& shadowMap, FEngine& engin
         FScene::LightSoa const& lightData, ShadowMap::SceneInfo const& sceneInfo) noexcept {
 
     const size_t lightIndex = shadowMap.getLightIndex();
-    FLightManager::ShadowOptions const* const options = shadowMap.getShadowOptions();
+    FLightManager::ShadowOptions const& options = shadowMap.getShadowOptions();
+    auto const& vsmShadowOptions = view.getVsmShadowOptions();
 
     // update the shadow map frustum/camera
     const ShadowMap::ShadowMapInfo shadowMapInfo{
             .atlasDimension      = mTextureAtlasRequirements.size,
-            .textureDimension    = uint16_t(options->mapSize),
-            .shadowDimension     = uint16_t(options->mapSize - 2u),
+            .textureDimension    = uint16_t(options.mapSize),
+            .shadowDimension     = uint16_t(options.mapSize - 2u),
             .textureSpaceFlipped = engine.getBackend() == Backend::METAL ||
                                    engine.getBackend() == Backend::VULKAN ||
                                    engine.getBackend() == Backend::WEBGPU,
@@ -833,9 +829,9 @@ void ShadowMapManager::prepareSpotShadowMap(ShadowMap& shadowMap, FEngine& engin
     // and if we need to generate it, update all the UBO data
     if (shadowMap.hasVisibleShadows()) {
         const size_t shadowIndex = shadowMap.getShadowIndex();
-        const float wsTexelSizeAtOneMeter = shaderParameters.texelSizeAtOneMeterWs;
+        const float2 wsTexelSizeAtOneMeter = shaderParameters.texelSizeAtOneMeterWs;
         // note: normalBias is set to zero for VSM
-        const float normalBias = shadowMapInfo.vsm ? 0.0f : options->normalBias;
+        const float normalBias = shadowMapInfo.vsm ? 0.0f : options.normalBias;
 
         auto& s = mShadowUb.edit();
         const double n = shadowMap.getCamera().getNear();
@@ -845,12 +841,12 @@ void ShadowMapManager::prepareSpotShadowMap(ShadowMap& shadowMap, FEngine& engin
         s.shadows[shadowIndex].scissorNormalized = shaderParameters.scissorNormalized;
         s.shadows[shadowIndex].normalBias = normalBias * wsTexelSizeAtOneMeter;
         s.shadows[shadowIndex].lightFromWorldZ = shaderParameters.lightFromWorldZ;
-        s.shadows[shadowIndex].texelSizeAtOneMeter = wsTexelSizeAtOneMeter;
         s.shadows[shadowIndex].nearOverFarMinusNear = float(n / (f - n));
-        s.shadows[shadowIndex].elvsm = options->vsm.elvsm;
+        s.shadows[shadowIndex].elvsm = options.vsm.elvsm;
+        s.shadows[shadowIndex].vsmExponent = getWrapExponentEVSM(vsmShadowOptions, options);
         s.shadows[shadowIndex].bulbRadiusLs =
-                mSoftShadowOptions.penumbraScale * options->shadowBulbRadius
-                        / wsTexelSizeAtOneMeter;
+                mSoftShadowOptions.penumbraScale * options.shadowBulbRadius
+                        / length(wsTexelSizeAtOneMeter);
 
     }
 }
@@ -907,13 +903,14 @@ void ShadowMapManager::preparePointShadowMap(ShadowMap& shadowMap,
 
     const uint8_t face = shadowMap.getFace();
     const size_t lightIndex = shadowMap.getLightIndex();
-    FLightManager::ShadowOptions const* const options = shadowMap.getShadowOptions();
+    FLightManager::ShadowOptions const& options = shadowMap.getShadowOptions();
+    auto const& vsmShadowOptions = view.getVsmShadowOptions();
 
     // update the shadow map frustum/camera
     const ShadowMap::ShadowMapInfo shadowMapInfo{
             .atlasDimension      = mTextureAtlasRequirements.size,
-            .textureDimension    = uint16_t(options->mapSize),
-            .shadowDimension     = uint16_t(options->mapSize), // point-lights don't have a border
+            .textureDimension    = uint16_t(options.mapSize),
+            .shadowDimension     = uint16_t(options.mapSize), // point-lights don't have a border
             .textureSpaceFlipped = engine.getBackend() == Backend::METAL ||
                                    engine.getBackend() == Backend::VULKAN ||
                                    engine.getBackend() == Backend::WEBGPU,
@@ -926,9 +923,9 @@ void ShadowMapManager::preparePointShadowMap(ShadowMap& shadowMap,
     // and if we need to generate it, update all the UBO data
     if (shadowMap.hasVisibleShadows()) {
         const size_t shadowIndex = shadowMap.getShadowIndex();
-        const float wsTexelSizeAtOneMeter = shaderParameters.texelSizeAtOneMeterWs;
+        const float2 wsTexelSizeAtOneMeter = shaderParameters.texelSizeAtOneMeterWs;
         // note: normalBias is set to zero for VSM
-        const float normalBias = shadowMapInfo.vsm ? 0.0f : options->normalBias;
+        const float normalBias = shadowMapInfo.vsm ? 0.0f : options.normalBias;
 
         auto& s = mShadowUb.edit();
         const double n = shadowMap.getCamera().getNear();
@@ -938,12 +935,12 @@ void ShadowMapManager::preparePointShadowMap(ShadowMap& shadowMap,
         s.shadows[shadowIndex].scissorNormalized = shaderParameters.scissorNormalized;
         s.shadows[shadowIndex].normalBias = normalBias * wsTexelSizeAtOneMeter;
         s.shadows[shadowIndex].lightFromWorldZ = shaderParameters.lightFromWorldZ;
-        s.shadows[shadowIndex].texelSizeAtOneMeter = wsTexelSizeAtOneMeter;
         s.shadows[shadowIndex].nearOverFarMinusNear = float(n / (f - n));
-        s.shadows[shadowIndex].elvsm = options->vsm.elvsm;
+        s.shadows[shadowIndex].elvsm = options.vsm.elvsm;
+        s.shadows[shadowIndex].vsmExponent = getWrapExponentEVSM(vsmShadowOptions, options);
         s.shadows[shadowIndex].bulbRadiusLs =
-                mSoftShadowOptions.penumbraScale * options->shadowBulbRadius
-                        / wsTexelSizeAtOneMeter;
+                mSoftShadowOptions.penumbraScale * options.shadowBulbRadius
+                        / length(wsTexelSizeAtOneMeter);
     }
 }
 
@@ -998,7 +995,7 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateSpotShadowMaps(FEngine
             lightData.data<FScene::SHADOW_INFO>());
 
     ShadowTechnique shadowTechnique{};
-    utils::Slice<const ShadowMap> spotShadowMaps = getSpotShadowMaps();
+    utils::Slice<const ShadowMap> const spotShadowMaps = getSpotShadowMaps();
     if (!spotShadowMaps.empty()) {
         shadowTechnique |= ShadowTechnique::SHADOW_MAP;
         for (ShadowMap const& shadowMap : spotShadowMaps) {
@@ -1038,14 +1035,14 @@ void ShadowMapManager::calculateTextureRequirements(FEngine& engine, FView& view
     for (ShadowMap const& shadowMap : getCascadedShadowMap()) {
         // Shadow map size should be the same for all cascades.
         auto const& options = shadowMap.getShadowOptions();
-        maxDimension = std::max(maxDimension, options->mapSize);
-        elvsm = elvsm || options->vsm.elvsm;
+        maxDimension = std::max(maxDimension, options.mapSize);
+        elvsm = elvsm || options.vsm.elvsm;
     }
 
     for (ShadowMap const& shadowMap : getSpotShadowMaps()) {
         auto const& options = shadowMap.getShadowOptions();
-        maxDimension = std::max(maxDimension, options->mapSize);
-        elvsm = elvsm || options->vsm.elvsm;
+        maxDimension = std::max(maxDimension, options.mapSize);
+        elvsm = elvsm || options.vsm.elvsm;
     }
 
     uint8_t layersNeeded = 0;
@@ -1055,7 +1052,7 @@ void ShadowMapManager::calculateTextureRequirements(FEngine& engine, FView& view
         ShadowMap* pShadowMap) mutable {
         // Allocate shadowmap from our Atlas Allocator
         auto const& options = pShadowMap->getShadowOptions();
-        auto allocation = allocator.allocate(options->mapSize);
+        auto allocation = allocator.allocate(options.mapSize);
         assert_invariant(allocation.isValid());
         assert_invariant(!allocation.viewport.empty());
         pShadowMap->setAllocation(allocation.layer, allocation.viewport);
@@ -1091,6 +1088,7 @@ void ShadowMapManager::calculateTextureRequirements(FEngine& engine, FView& view
         msaaSamples = 1;
     }
 
+    float4 clearColor{};
     TextureFormat format = TextureFormat::DEPTH16;
     if (view.hasVSM()) {
         if (vsmShadowOptions.highPrecision) {
@@ -1098,6 +1096,9 @@ void ShadowMapManager::calculateTextureRequirements(FEngine& engine, FView& view
         } else {
             format = elvsm ? TextureFormat::RGBA16F : TextureFormat::RG16F;
         }
+        const float maxMoment2 = getMaxMomentEVSM(vsmShadowOptions);
+        const float maxMoment1 = std::sqrt(maxMoment2);
+        clearColor = { maxMoment1, maxMoment2, 0, 0 };
     }
 
     mSoftShadowOptions = view.getSoftShadowOptions();
@@ -1119,7 +1120,8 @@ void ShadowMapManager::calculateTextureRequirements(FEngine& engine, FView& view
             layersNeeded,
             mipLevels,
             msaaSamples,
-            format
+            format,
+            clearColor
     };
 }
 

filament/src/ShadowMapManager.h

@@ -47,9 +47,12 @@
 #include <utils/Slice.h>
 
 #include <math/mat4.h>
+#include <math/half.h>
 #include <math/vec4.h>
 
+#include <algorithm>
 #include <array>
+#include <limits>
 #include <memory>
 #include <new>
 #include <type_traits>
@@ -145,6 +148,28 @@ public:
     // for debugging only
     utils::FixedCapacityVector<Camera const*> getDirectionalShadowCameras() const noexcept;
 
+    static float getMaxMomentEVSM(VsmShadowOptions const& vsmShadowOptions) noexcept {
+        return vsmShadowOptions.highPrecision ?
+                std::numeric_limits<float>::max() :
+                float(std::numeric_limits<math::half>::max());
+    }
+
+    static float getMaxWrapExponentEVSM(VsmShadowOptions const& vsmShadowOptions) noexcept {
+        constexpr float low  = 5.2f;  // ~ std::log(std::numeric_limits<math::half>::max()) * 0.5f;
+        constexpr float high = 40.0f; // ~ std::log(std::numeric_limits<float>::max()) * 0.5f;
+        return vsmShadowOptions.highPrecision ? high : low;
+    }
+
+    static float getWrapExponentEVSM(
+            VsmShadowOptions const& vsmShadowOptions,
+            LightManager::ShadowOptions const& options) noexcept {
+        constexpr float ABSOLUTE_FILTER_LIMIT = 42.0f;
+        float const targetExponent = getMaxWrapExponentEVSM(vsmShadowOptions);
+        float const effectiveFilterRadius = std::max(1.0f, options.vsm.blurWidth);
+        float const filterCeiling = ABSOLUTE_FILTER_LIMIT / effectiveFilterRadius;
+        return std::min(targetExponent, filterCeiling);
+    }
+
 private:
     explicit ShadowMapManager(FEngine& engine);
 
@@ -216,6 +241,7 @@ private:
         uint8_t levels = 0;
         uint8_t msaaSamples = 1;
         backend::TextureFormat format = backend::TextureFormat::DEPTH16;
+        math::float4 clearColor{};
     } mTextureAtlasRequirements;
 
     SoftShadowOptions mSoftShadowOptions;

filament/src/components/CameraManager.cpp

@@ -85,7 +85,6 @@ void FCameraManager::destroy(FEngine& engine, Entity const e) noexcept {
         { // scope for camera -- it's invalid after this scope.
             FCamera* const camera = manager.elementAt<CAMERA>(i);
             assert_invariant(camera);
-            camera->terminate(engine);
             engine.getHeapAllocator().destroy(camera);
 
             // Remove the camera component

filament/src/components/RenderableManager.cpp

@@ -577,7 +577,7 @@ void FRenderableManager::create(
     FEngine::DriverApi& driver = engine.getDriverApi();
 
     if (UTILS_UNLIKELY(manager.hasComponent(entity))) {
-        destroy(entity);
+        destroy(entity, driver);
     }
     Instance const ci = manager.addComponent(entity);
     assert_invariant(ci);
@@ -733,41 +733,42 @@ void FRenderableManager::create(
 }
 
 // this destroys a single component from an entity
-void FRenderableManager::destroy(Entity const e) noexcept {
+void FRenderableManager::destroy(Entity const e, DriverApi& driver) noexcept {
     Instance const ci = getInstance(e);
     if (ci) {
-        destroyComponent(ci);
+        destroyComponent(ci, driver);
         mManager.removeComponent(e);
     }
 }
 
+void FRenderableManager::clientDestroy(utils::Entity e) noexcept {
+    destroy(e, mEngine.getDriverApi());
+}
+
 // this destroys all components in this manager
-void FRenderableManager::terminate() noexcept {
+void FRenderableManager::terminate(DriverApi& driver) noexcept {
     auto& manager = mManager;
     if (!manager.empty()) {
         DLOG(INFO) << "cleaning up " << manager.getComponentCount()
                    << " leaked Renderable components";
         while (!manager.empty()) {
             Instance const ci = manager.end() - 1;
-            destroyComponent(ci);
+            destroyComponent(ci, driver);
             manager.removeComponent(manager.getEntity(ci));
         }
     }
-    mHwRenderPrimitiveFactory.terminate(mEngine.getDriverApi());
+    mHwRenderPrimitiveFactory.terminate(driver);
 }
 
-void FRenderableManager::gc(EntityManager& em) noexcept {
-    mManager.gc(em, [this](Entity const e) {
-        destroy(e);
+void FRenderableManager::gc(EntityManager& em, DriverApi& driver) noexcept {
+    mManager.gc(em, [this, &driver](Entity const e) {
+        destroy(e, driver);
     });
 }
 
 // This is basically a Renderable's destructor.
-void FRenderableManager::destroyComponent(Instance const ci) noexcept {
+void FRenderableManager::destroyComponent(Instance const ci, DriverApi& driver) noexcept {
     auto& manager = mManager;
-    FEngine& engine = mEngine;
-
-    FEngine::DriverApi& driver = engine.getDriverApi();
 
     // See create(RenderableManager::Builder&, Entity)
     destroyComponentPrimitives(mHwRenderPrimitiveFactory, driver, manager[ci].primitives);

filament/src/components/RenderableManager.h

@@ -91,9 +91,9 @@ public:
     ~FRenderableManager();
 
     // free-up all resources
-    void terminate() noexcept;
+    void terminate(backend::DriverApi& driver) noexcept;
 
-    void gc(utils::EntityManager& em) noexcept;
+    void gc(utils::EntityManager& em, backend::DriverApi& driver) noexcept;
 
     /*
      * Component Manager APIs
@@ -125,7 +125,11 @@ public:
 
     void create(const Builder& builder, utils::Entity entity);
 
-    void destroy(utils::Entity e) noexcept;
+    void destroy(utils::Entity e, backend::DriverApi& driver) noexcept;
+
+    // The client API should have taken an Engine&, but it doesn't, so that's a workaround
+    // so we can keep the API as it is.
+    void clientDestroy(utils::Entity e) noexcept;
 
     inline void setAxisAlignedBoundingBox(Instance instance, const Box& aabb);
 
@@ -239,7 +243,7 @@ public:
     };
 
 private:
-    void destroyComponent(Instance ci) noexcept;
+    void destroyComponent(Instance ci, backend::DriverApi& driver) noexcept;
     static void destroyComponentPrimitives(
             HwRenderPrimitiveFactory& factory, backend::DriverApi& driver,
             utils::Slice<FRenderPrimitive> primitives) noexcept;

filament/src/details/Camera.h

@@ -46,9 +46,6 @@ public:
 
     FCamera(FEngine& engine, utils::Entity e);
 
-    void terminate(FEngine&) noexcept { }
-
-
     // Sets the projection matrices (viewing and culling). The viewing matrice has infinite far.
     void setProjection(Projection projection,
                        double left, double right, double bottom, double top,

filament/src/details/Engine.cpp

@@ -20,6 +20,11 @@
 #include "TextureCache.h"
 #include "RenderPrimitive.h"
 
+#include "components/CameraManager.h"
+#include "components/LightManager.h"
+#include "components/RenderableManager.h"
+#include "components/TransformManager.h"
+
 #include "details/BufferObject.h"
 #include "details/Camera.h"
 #include "details/Fence.h"
@@ -58,6 +63,7 @@
 #include <utils/CallStack.h>
 #include <utils/CString.h>
 #include <utils/Invocable.h>
+#include <utils/JobSystem.h>
 #include <utils/Logger.h>
 #include <utils/Panic.h>
 #include <utils/PrivateImplementation-impl.h>
@@ -538,12 +544,10 @@ void FEngine::init() {
             &debug.shadowmap.debug_directional_shadowmap, [this] {
                 mMaterials.forEach([this](FMaterial* material) {
                     if (material->getMaterialDomain() == MaterialDomain::SURFACE) {
-                        FMaterial::SpecializationConstantsBuilder constants =
-                                material->getSpecializationConstantsBuilder();
-                        constants.set(+ReservedSpecializationConstants::
-                                              CONFIG_DEBUG_DIRECTIONAL_SHADOWMAP,
-                                debug.shadowmap.debug_directional_shadowmap);
-                        material->setSpecializationConstants(std::move(constants));
+                        material->getPrograms().setConstants({
+                            { +ReservedSpecializationConstants::CONFIG_DEBUG_DIRECTIONAL_SHADOWMAP,
+                              debug.shadowmap.debug_directional_shadowmap },
+                        });
                     }
                 });
             });
@@ -552,12 +556,10 @@ void FEngine::init() {
             &debug.lighting.debug_froxel_visualization, [this] {
                 mMaterials.forEach([this](FMaterial* material) {
                     if (material->getMaterialDomain() == MaterialDomain::SURFACE) {
-                        FMaterial::SpecializationConstantsBuilder constants =
-                                material->getSpecializationConstantsBuilder();
-                        constants.set(+ReservedSpecializationConstants::
-                                              CONFIG_DEBUG_FROXEL_VISUALIZATION,
-                                debug.lighting.debug_froxel_visualization);
-                        material->setSpecializationConstants(std::move(constants));
+                        material->getPrograms().setConstants({
+                            { +ReservedSpecializationConstants::CONFIG_DEBUG_FROXEL_VISUALIZATION,
+                              debug.lighting.debug_froxel_visualization },
+                        });
                     }
                 });
             });
@@ -590,6 +592,9 @@ void FEngine::shutdown() {
     DLOG(INFO) << "CircularBuffer: High watermark " << wm / 1024 << " KiB (" << wmpct << "%)";
 #endif
 
+    /* Destroy any leftover items in the cache. */
+    mMaterialCache.terminate(*this);
+
     DriverApi& driver = getDriverApi();
 
     /*
@@ -600,7 +605,7 @@ void FEngine::shutdown() {
     mResourceAllocatorDisposer->terminate();
     mResourceAllocatorDisposer.reset();
     mDFG.terminate(*this);                  // free-up the DFG
-    mRenderableManager.terminate();         // free-up all renderables
+    mRenderableManager.terminate(driver);   // free-up all renderables
     mLightManager.terminate();              // free-up all lights
     mCameraManager.terminate(*this);        // free-up all cameras
 
@@ -764,13 +769,30 @@ void FEngine::prepare(DriverApi& driver) {
     });
 }
 
-void FEngine::gcManagers() {
-    // Note: this runs in a Job
-    auto& em = mEntityManager;
-    mRenderableManager.gc(em);
-    mLightManager.gc(em);
-    mTransformManager.gc(em);
-    mCameraManager.gc(*this, em);
+void FEngine::gc() {
+    JobSystem& js = getJobSystem();
+    auto *rootJob = js.createJob();
+
+    js.run(
+            jobs::createJob(js, rootJob, [this] {
+                // These are safe to GC *sequentially* from a worker thread.
+                // The JobSystem acts as a release/acquire operation.
+                mLightManager.gc(mEntityManager);
+                mTransformManager.gc(mEntityManager);
+                mCameraManager.gc(*this, mEntityManager);
+            }));
+
+    if (isAsynchronousModeEnabled()) {
+        // gcDeferredAsyncObjectDestruction() is thread-safe in this context because
+        // 1. JobSystem provides the release/acquire operation
+        // 2. it's only used from the main thread, which we are blocking during the gc
+        js.run(jobs::createJob(js, rootJob, &FEngine::gcDeferredAsyncObjectDestruction, this));
+    }
+
+    // RenderableManager cannot be gc'ed from a different thread, because it needs the DriverAPI
+    mRenderableManager.gc(mEntityManager, getDriverApi());
+
+    js.runAndWait(rootJob);
 }
 
 void FEngine::gcDeferredAsyncObjectDestruction() {
@@ -1399,7 +1421,7 @@ bool FEngine::destroy(const FMaterialInstance* p) {
 
 UTILS_NOINLINE
 void FEngine::destroy(Entity const e) {
-    mRenderableManager.destroy(e);
+    mRenderableManager.destroy(e, getDriverApi());
     mLightManager.destroy(e);
     mTransformManager.destroy(e);
     mCameraManager.destroy(*this, e);

filament/src/details/Engine.h

@@ -456,8 +456,7 @@ public:
     }
 
     void prepare(backend::DriverApi& driver);
-    void gcManagers();
-    void gcDeferredAsyncObjectDestruction();
+    void gc();
     void submitFrame();
 
     using ShaderContent = utils::FixedCapacityVector<uint8_t>;
@@ -583,6 +582,8 @@ private:
     int loop();
     void flushCommandBuffer(backend::CommandBufferQueue& commandBufferQueue) const;
 
+    void gcDeferredAsyncObjectDestruction();
+
     template<typename T>
     bool isValid(const T* ptr, ResourceList<T> const& list) const;
 

filament/src/details/Material.cpp

@@ -170,34 +170,11 @@ FMaterial::FMaterial(FEngine& engine, const Builder& builder, MaterialDefinition
 
     DriverApi& driver = engine.getDriverApi();
 
-    mIsStereoSupported = driver.isStereoSupported();
-    mIsParallelShaderCompileSupported = driver.isParallelShaderCompileSupported();
     mDepthPrecacheDisabled =
             driver.isWorkaroundNeeded(Workaround::DISABLE_DEPTH_PRECACHE_FOR_DEFAULT_MATERIAL);
     mDefaultMaterial = engine.getDefaultMaterial();
 
-    FixedCapacityVector<Program::SpecializationConstant> specializationConstants =
-            processSpecializationConstants(builder);
-    mSpecializationConstants =
-            engine.getMaterialCache().getSpecializationConstantsInternPool().acquire(
-                    std::move(specializationConstants));
-
-    size_t cachedProgramsSize;
-    switch (mDefinition.materialDomain) {
-        case filament::MaterialDomain::SURFACE:
-            cachedProgramsSize = 1 << VARIANT_BITS;
-            break;
-        case filament::MaterialDomain::POST_PROCESS:
-            cachedProgramsSize = 1 << POST_PROCESS_VARIANT_BITS;
-            break;
-        case filament::MaterialDomain::COMPUTE:
-            cachedProgramsSize = 1;
-            break;
-    }
-    mCachedPrograms = FixedCapacityVector<backend::Handle<backend::HwProgram>>(cachedProgramsSize);
-
-    mDefinition.acquirePrograms(engine, mCachedPrograms.as_slice(), getMaterialParser(),
-            mSpecializationConstants, mIsDefaultMaterial);
+    mPrograms.initializeForMaterial(engine, *this, processSpecializationConstants(builder));
 
 #if FILAMENT_ENABLE_MATDBG
     // Register the material with matdbg.
@@ -237,11 +214,7 @@ void FMaterial::terminate(FEngine& engine) {
     }
 #endif
 
-    mDefinition.releasePrograms(engine, mCachedPrograms.as_slice(),
-            getMaterialParser(), mSpecializationConstants, mIsDefaultMaterial);
-    engine.getMaterialCache().releaseMaterial(engine, mDefinition);
-    engine.getMaterialCache().getSpecializationConstantsInternPool().release(
-            mSpecializationConstants);
+    mPrograms.terminate(engine);
 }
 
 filament::DescriptorSetLayout const& FMaterial::getPerViewDescriptorSetLayout(
@@ -257,71 +230,24 @@ filament::DescriptorSetLayout const& FMaterial::getPerViewDescriptorSetLayout(
     }
     // mDefinition.perViewDescriptorSetLayout{Vsm} is already resolved for MaterialDomain
     if (useVsmDescriptorSetLayout) {
-        return mDefinition.perViewDescriptorSetLayoutVsm;
+        return mDefinition.perViewDescriptorSetLayoutS2d;
     }
-    return mDefinition.perViewDescriptorSetLayout;
+    return mDefinition.perViewDescriptorSetLayoutPcf;
 }
 
-void FMaterial::compile(CompilerPriorityQueue const priority,
-        UserVariantFilterMask variantSpec,
-        CallbackHandler* handler,
-        Invocable<void(Material*)>&& callback) noexcept {
-
-    DriverApi& driver = mEngine.getDriverApi();
-
-    // Turn off the STE variant if stereo is not supported.
-    if (!mIsStereoSupported) {
-        variantSpec &= ~UserVariantFilterMask(UserVariantFilterBit::STE);
-    }
-
-    UserVariantFilterMask const variantFilter =
-            ~variantSpec & UserVariantFilterMask(UserVariantFilterBit::ALL);
-    ShaderModel const shaderModel = mEngine.getShaderModel();
-    bool const isStereoSupported = mEngine.getDriverApi().isStereoSupported();
-
-    if (UTILS_LIKELY(mIsParallelShaderCompileSupported)) {
-        for (auto const variant: mDefinition.getVariants()) {
-            if (!variantFilter || variant == Variant::filterUserVariant(variant, variantFilter)) {
-                if (mDefinition.hasVariant(variant, shaderModel, isStereoSupported)) {
-                    prepareProgram(driver, variant, priority);
-                }
-            }
-        }
-    }
-
+void FMaterial::compile(CompilerPriorityQueue const priority, UserVariantFilterMask variantSpec,
+        CallbackHandler* handler, Invocable<void(Material*)>&& callback) noexcept {
+    FMaterialInstance* mi = getDefaultInstance();
     if (callback) {
-        struct Callback {
-            Invocable<void(Material*)> f;
-            Material* m;
-            static void func(void* user) {
-                auto* const c = static_cast<Callback*>(user);
-                c->f(c->m);
-                delete c;
-            }
-        };
-        auto* const user = new(std::nothrow) Callback{ std::move(callback), this };
-        driver.compilePrograms(priority, handler, &Callback::func, user);
+        mi->compile(priority, variantSpec, handler,
+                [this, callback = std::move(callback)](MaterialInstance*) {
+                    callback(this);
+                });
     } else {
-        driver.compilePrograms(priority, nullptr, nullptr, nullptr);
+        mi->compile(priority, variantSpec, handler, {});
     }
 }
 
-Handle<HwProgram> FMaterial::prepareProgramSlow(DriverApi& driver, Variant const variant,
-        CompilerPriorityQueue const priorityQueue) const noexcept {
-    if (isSharedVariant(variant)) {
-        FMaterial const* defaultMaterial = mEngine.getDefaultMaterial();
-        FILAMENT_CHECK_PRECONDITION(defaultMaterial);
-        Handle<HwProgram> program = defaultMaterial->mCachedPrograms[variant.key];
-        if (program) {
-            return mCachedPrograms[variant.key] = program;
-        }
-        return mCachedPrograms[variant.key] =
-                defaultMaterial->prepareProgram(driver, variant, priorityQueue);
-    }
-    return mCachedPrograms[variant.key] = mDefinition.prepareProgram(mEngine, driver,
-            getMaterialParser(), getProgramSpecialization(variant), priorityQueue);
-}
-
 FMaterialInstance* FMaterial::createInstance(const char* name) const noexcept {
     if (mDefaultMaterialInstance) {
         // if we have a default instance, use it to create a new one
@@ -356,14 +282,6 @@ BufferInterfaceBlock::FieldInfo const* FMaterial::reflect(
     return mDefinition.uniformInterfaceBlock.getFieldInfo(name);
 }
 
-ProgramSpecialization FMaterial::getProgramSpecialization(Variant const variant) const noexcept {
-    return ProgramSpecialization {
-        .materialCrc32 = mDefinition.getMaterialParser().getCrc32(),
-        .variant = variant,
-        .specializationConstants = mSpecializationConstants,
-    };
-}
-
 size_t FMaterial::getParameters(ParameterInfo* parameters, size_t count) const noexcept {
     count = std::min(count, getParameterCount());
 
@@ -430,8 +348,7 @@ void FMaterial::updateActiveProgramsForMatdbg(Variant const variant) const noexc
 void FMaterial::applyPendingEdits() noexcept {
     const char* name = mDefinition.name.c_str();
     DLOG(INFO) << "Applying edits to " << (name ? name : "(untitled)");
-    mDefinition.releasePrograms(mEngine, mCachedPrograms.as_slice(), getMaterialParser(),
-            mSpecializationConstants, mIsDefaultMaterial);
+    mPrograms.clear(mEngine);
     latchPendingEdits();
 }
 
@@ -480,26 +397,6 @@ void FMaterial::onQueryCallback(void* userdata, VariantList* pActiveVariants) {
 
 #endif // FILAMENT_ENABLE_MATDBG
 
-void FMaterial::setSpecializationConstants(SpecializationConstantsBuilder&& builder) noexcept {
-    if (builder.mConstants.empty()) {
-        // Nothing was changed.
-        return;
-    }
-
-    auto& internPool = mEngine.getMaterialCache().getSpecializationConstantsInternPool();
-    MaterialParser const& materialParser = getMaterialParser();
-
-    // Release old resources...
-    mDefinition.releasePrograms(mEngine, mCachedPrograms.as_slice(), materialParser,
-            mSpecializationConstants, mIsDefaultMaterial);
-    internPool.release(mSpecializationConstants);
-
-    // Then acquire new ones.
-    mSpecializationConstants = internPool.acquire(std::move(builder.mConstants));
-    mDefinition.acquirePrograms(mEngine, mCachedPrograms.as_slice(), materialParser,
-            mSpecializationConstants, mIsDefaultMaterial);
-}
-
 FixedCapacityVector<Program::SpecializationConstant> FMaterial::processSpecializationConstants(
         Builder const& builder) {
     FixedCapacityVector<Program::SpecializationConstant> specializationConstants =
@@ -560,12 +457,4 @@ const char* FMaterial::getParameterTransformName(std::string_view samplerName) c
     return info->transformName.c_str();
 }
 
-template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<int32_t>(uint32_t id, int32_t value) noexcept;
-template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<float>(uint32_t id, float value) noexcept;
-template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<bool>(uint32_t id, bool value) noexcept;
-
-template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<int32_t>(std::string_view name, int32_t value) noexcept;
-template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<float>(std::string_view name, float value) noexcept;
-template FMaterial::SpecializationConstantsBuilder& FMaterial::SpecializationConstantsBuilder::set<bool>(std::string_view name, bool value) noexcept;
-
 } // namespace filament

filament/src/details/Material.h

@@ -18,6 +18,7 @@
 #define TNT_FILAMENT_DETAILS_MATERIAL_H
 
 #include "downcast.h"
+#include "LocalProgramCache.h"
 
 #include "details/MaterialInstance.h"
 
@@ -45,7 +46,6 @@
 #include <utils/Invocable.h>
 #include <utils/Mutex.h>
 
-#include <optional>
 #include <string_view>
 
 #include <stddef.h>
@@ -72,62 +72,6 @@ public:
         DefaultMaterialBuilder();
     };
 
-    // Used to change specialization constants at runtime internally by Filament.
-    //
-    // Call Material::getSpecializationConstantsBuilder() to return an instance of
-    // SpecializationConstantsBuilder. The lifetime of this object must not exceed the lifetime of
-    // the corresponding Material. After spec constants are changed with calls to set(), apply the
-    // changes to the original Material via setSpecializationConstants().
-    //
-    // It would be nice to reuse Builder for this, but Builder allocates an entire map and asserts
-    // if specified constants don't exist. In constrast, SpecializationConstantsBuilder is
-    // copy-on-write and allows non-existent specialization constants to be named.
-    class SpecializationConstantsBuilder {
-        friend class FMaterial;
-
-    public:
-        SpecializationConstantsBuilder(SpecializationConstantsBuilder const& rhs) = delete;
-        SpecializationConstantsBuilder& operator=(
-                SpecializationConstantsBuilder const& rhs) = delete;
-
-        SpecializationConstantsBuilder(SpecializationConstantsBuilder&& rhs) = default;
-        SpecializationConstantsBuilder& operator=(SpecializationConstantsBuilder&& rhs) = default;
-
-        template<typename T, typename = Builder::is_supported_constant_parameter_t<T>>
-        SpecializationConstantsBuilder& set(uint32_t id, T value) noexcept {
-            // Don't allocate if we can help it.
-            if (mConstants.empty()) {
-                if (std::get<T>(mDefaultConstants[id]) == value) {
-                    return *this;
-                }
-                mConstants = utils::FixedCapacityVector<backend::Program::SpecializationConstant>(
-                        mDefaultConstants);
-            }
-            mConstants[id] = value;
-            return *this;
-        }
-
-        template<typename T, typename = Builder::is_supported_constant_parameter_t<T>>
-        SpecializationConstantsBuilder& set(std::string_view name, T value) noexcept {
-            auto it = mDefinition->specializationConstantsNameToIndex.find(name);
-            if (it != mDefinition->specializationConstantsNameToIndex.cend()) {
-                set(it->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS, value);
-            }
-            return *this;
-        }
-
-    private:
-        SpecializationConstantsBuilder(MaterialDefinition const& definition,
-                utils::Slice<const backend::Program::SpecializationConstant> defaultConstants)
-                : mDefinition(&definition),
-                  mDefaultConstants(defaultConstants) {}
-
-        MaterialDefinition const* mDefinition;
-        utils::Slice<const backend::Program::SpecializationConstant> mDefaultConstants;
-        // Copy-on-write vector.
-        utils::FixedCapacityVector<backend::Program::SpecializationConstant> mConstants;
-    };
-
     void terminate(FEngine& engine);
 
     // return the uniform interface block for this material
@@ -139,7 +83,7 @@ public:
         // This is mostly intended to be used for post-process materials; but it's also useful for
         // Surface material that behave like post-process material (i.e. that don't really have variants or for
         // which VSM is nonsensical, like for unlit materials)
-        return mDefinition.perViewDescriptorSetLayout;
+        return mDefinition.perViewDescriptorSetLayoutPcf;
     }
 
     DescriptorSetLayout const& getPerViewDescriptorSetLayout(
@@ -181,46 +125,6 @@ public:
 
     FEngine& getEngine() const noexcept  { return mEngine; }
 
-    // prepareProgram creates the program for the material's given variant at the backend level.
-    // Must be called outside of backend render pass.
-    // Must be called before getProgram() below.
-    backend::Handle<backend::HwProgram> prepareProgram(backend::DriverApi& driver,
-            Variant const variant,
-            backend::CompilerPriorityQueue const priorityQueue) const noexcept {
-        backend::Handle<backend::HwProgram> program = mCachedPrograms[variant.key];
-        if (UTILS_LIKELY(program)) {
-            return program;
-        }
-        return prepareProgramSlow(driver, variant, priorityQueue);
-    }
-
-    // getProgram returns the backend program for the material's given variant.
-    // Must be called after prepareProgram().
-    [[nodiscard]]
-    backend::Handle<backend::HwProgram> getProgram(Variant variant) const noexcept {
-
-        if (UTILS_UNLIKELY(mEngine.features.material.enable_fog_as_postprocess)) {
-            // if the fog as post-process feature is enabled, we need to proceed "as-if" the material
-            // didn't have the FOG variant bit.
-            if (getMaterialDomain() == MaterialDomain::SURFACE) {
-                BlendingMode const blendingMode = getBlendingMode();
-                bool const hasScreenSpaceRefraction = getRefractionMode() == RefractionMode::SCREEN_SPACE;
-                bool const isBlendingCommand = !hasScreenSpaceRefraction &&
-                        (blendingMode != BlendingMode::OPAQUE && blendingMode != BlendingMode::MASKED);
-                if (!isBlendingCommand) {
-                    variant.setFog(false);
-                }
-            }
-        }
-
-#if FILAMENT_ENABLE_MATDBG
-        updateActiveProgramsForMatdbg(variant);
-#endif
-        backend::Handle<backend::HwProgram> program = mCachedPrograms[variant.key];
-        assert_invariant(program);
-        return program;
-    }
-
     // MaterialInstance::use() binds descriptor sets before drawing. For shared variants,
     // however, the material instance will call useShared() to bind the default material's sets
     // instead.
@@ -275,6 +179,8 @@ public:
     float getSpecularAntiAliasingVariance() const noexcept { return mDefinition.specularAntiAliasingVariance; }
     float getSpecularAntiAliasingThreshold() const noexcept { return mDefinition.specularAntiAliasingThreshold; }
 
+    bool isDefaultMaterial() const noexcept { return mIsDefaultMaterial; }
+
     backend::descriptor_binding_t getSamplerBinding(
             std::string_view const& name) const;
 
@@ -297,12 +203,7 @@ public:
 
     uint32_t generateMaterialInstanceId() const noexcept { return mMaterialInstanceId++; }
 
-    // Update specialization constants.
-    SpecializationConstantsBuilder getSpecializationConstantsBuilder() const noexcept {
-        return SpecializationConstantsBuilder(mDefinition, mSpecializationConstants);
-    }
-
-    void setSpecializationConstants(SpecializationConstantsBuilder&& builder) noexcept;
+    LocalProgramCache& getPrograms() noexcept { return mPrograms; }
 
     uint8_t getPerViewLayoutIndex() const noexcept {
         return mDefinition.perViewLayoutIndex;
@@ -316,6 +217,30 @@ public:
         return mDefinition.source.c_str_safe();
     }
 
+    inline bool isSharedVariant(Variant const variant) const {
+        // HACK: The default material "should" have MNT | DEP, but then we'd have to compile it as a
+        // lit material, which would increase binary size. Perhaps we could specially compile it
+        // with this variant, but with the shader program cache in active development, the days of
+        // the default material are numbered anyway.
+        constexpr Variant::type_t vsmAndDep = Variant::MNT | Variant::DEP;
+        return mDefinition.materialDomain == MaterialDomain::SURFACE && !mIsDefaultMaterial &&
+               !mDefinition.hasCustomDepthShader && Variant::isValidDepthVariant(variant) &&
+               (variant.key & vsmAndDep) != vsmAndDep;
+    }
+
+    MaterialParser const& getMaterialParser() const noexcept {
+#if FILAMENT_ENABLE_MATDBG
+        if (mEditedMaterialParser) {
+            return *mEditedMaterialParser;
+        }
+#endif
+        return mDefinition.getMaterialParser();
+    }
+
+    MaterialDefinition const& getDefinition() const noexcept { return mDefinition; }
+
+    LocalProgramCache const& getPrograms() const noexcept { return mPrograms; }
+
 #if FILAMENT_ENABLE_MATDBG
     void applyPendingEdits() noexcept;
 
@@ -346,24 +271,12 @@ public:
     }
 
     /** @}*/
+
+    // Called by getProgram() to update active program list for matdbg UI.
+    void updateActiveProgramsForMatdbg(Variant const variant) const noexcept;
 #endif
 
 private:
-    MaterialParser const& getMaterialParser() const noexcept {
-#if FILAMENT_ENABLE_MATDBG
-        if (mEditedMaterialParser) {
-            return *mEditedMaterialParser;
-        }
-#endif
-        return mDefinition.getMaterialParser();
-    }
-
-    ProgramSpecialization getProgramSpecialization(Variant const variant) const noexcept;
-
-    backend::Handle<backend::HwProgram> prepareProgramSlow(backend::DriverApi& driver,
-            Variant const variant,
-            backend::CompilerPriorityQueue const priorityQueue) const noexcept;
-
     utils::FixedCapacityVector<backend::Program::SpecializationConstant>
             processSpecializationConstants(Builder const& builder);
 
@@ -371,31 +284,11 @@ private:
 
     void createAndCacheProgram(backend::DriverApi& driver, backend::Program&& p, Variant variant) const noexcept;
 
-    backend::DescriptorSetLayout const& getPerViewDescriptorSetLayoutDescription(
-            Variant variant, bool useVsmDescriptorSetLayout) const noexcept;
-
-    backend::DescriptorSetLayout const& getDescriptorSetLayoutDescription(
-            Variant variant = {}) const noexcept;
-
-    bool isSharedVariant(Variant const variant) const {
-        // HACK: The default material "should" have VSM | DEP, but then we'd have to compile it as a
-        // lit material, which would increase binary size. Perhaps we could specially compile it
-        // with this variant, but with the shader program cache in active development, the days of
-        // the default material are numbered anyway.
-        constexpr Variant::type_t vsmAndDep = Variant::VSM | Variant::DEP;
-        return mDefinition.materialDomain == MaterialDomain::SURFACE && !mIsDefaultMaterial &&
-               !mDefinition.hasCustomDepthShader && Variant::isValidDepthVariant(variant) &&
-               (variant.key & vsmAndDep) != vsmAndDep;
-    }
-
-    mutable utils::FixedCapacityVector<backend::Handle<backend::HwProgram>> mCachedPrograms;
     MaterialDefinition const& mDefinition;
 
     bool mIsDefaultMaterial = false;
 
     bool mUseUboBatching = false;
-    bool mIsStereoSupported = false;
-    bool mIsParallelShaderCompileSupported = false;
     bool mDepthPrecacheDisabled = false;
 
     FMaterial const* mDefaultMaterial = nullptr;
@@ -403,9 +296,6 @@ private:
     // reserve some space to construct the default material instance
     mutable FMaterialInstance* mDefaultMaterialInstance = nullptr;
 
-    // current specialization constants for the HwProgram
-    utils::Slice<const backend::Program::SpecializationConstant> mSpecializationConstants;
-
 #if FILAMENT_ENABLE_MATDBG
     matdbg::MaterialKey mDebuggerId;
     mutable utils::Mutex mActiveProgramsLock;
@@ -413,8 +303,6 @@ private:
     mutable utils::Mutex mPendingEditsLock;
     std::unique_ptr<MaterialParser> mPendingEdits;
     std::unique_ptr<MaterialParser> mEditedMaterialParser;
-    // Called by getProgram() to update active program list for matdbg UI.
-    void updateActiveProgramsForMatdbg(Variant const variant) const noexcept;
     void setPendingEdits(std::unique_ptr<MaterialParser> pendingEdits) noexcept;
     bool hasPendingEdits() const noexcept;
     void latchPendingEdits() noexcept;
@@ -423,6 +311,8 @@ private:
     FEngine& mEngine;
     const uint32_t mMaterialId;
     mutable uint32_t mMaterialInstanceId = 0;
+
+    LocalProgramCache mPrograms;
 };
 
 

filament/src/details/MaterialInstance.cpp

@@ -17,6 +17,7 @@
 #include <filament/MaterialInstance.h>
 
 #include "RenderPass.h"
+#include "MaterialParser.h"
 
 #include "ds/DescriptorSetLayout.h"
 
@@ -131,6 +132,7 @@ FMaterialInstance::FMaterialInstance(FEngine& engine,
           mTextureParameters(other->mTextureParameters),
           mDescriptorSet(other->mDescriptorSet.duplicate(
                 "MaterialInstance", mMaterial->getDescriptorSetLayout())),
+          mPrograms(other->mPrograms),
           mPolygonOffset(other->mPolygonOffset),
           mStencilState(other->mStencilState),
           mMaskThreshold(other->mMaskThreshold),
@@ -207,6 +209,10 @@ void FMaterialInstance::terminate(FEngine& engine) {
     if (ubHandle){
         driver.destroyBufferObject(*ubHandle);
     }
+
+    if (mPrograms.isInitialized()) {
+        mPrograms.terminate(engine);
+    }
 }
 
 void FMaterialInstance::commit(FEngine& engine) const {
@@ -259,6 +265,39 @@ void FMaterialInstance::commit(FEngine::DriverApi& driver, UboManager* uboManage
 
 // ------------------------------------------------------------------------------------------------
 
+template<typename T>
+void FMaterialInstance::setConstantImpl(std::string_view name, T value) {
+    auto const& constants = mMaterial->getDefinition().specializationConstantsNameToIndex;
+    auto it = constants.find(name);
+    FILAMENT_CHECK_PRECONDITION(it != constants.end()) << "Constant " << name << " does not exist";
+
+    if (UTILS_UNLIKELY(mPendingSpecializationConstants.empty())) {
+        mPendingSpecializationConstants =
+                FixedCapacityVector<backend::Program::SpecializationConstant>(
+                        getPrograms().getSpecializationConstants());
+    }
+
+    uint32_t id = it->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS;
+    mPendingSpecializationConstants[id] = value;
+}
+
+template<typename T>
+T FMaterialInstance::getConstantImpl(std::string_view name) const {
+    auto const& constants = mMaterial->getDefinition().specializationConstantsNameToIndex;
+    auto it = constants.find(name);
+    FILAMENT_CHECK_PRECONDITION(it != constants.end()) << "Constant " << name << " does not exist";
+
+    uint32_t id = it->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS;
+
+    if (UTILS_UNLIKELY(!mPendingSpecializationConstants.empty())) {
+        return std::get<T>(mPendingSpecializationConstants[id]);
+    }
+
+    return getPrograms().getConstant<T>(id);
+}
+
+// ------------------------------------------------------------------------------------------------
+
 void FMaterialInstance::setParameter(std::string_view const name,
         Handle<HwTexture> texture, SamplerParams const params) {
     auto const binding = mMaterial->getSamplerBinding(name);
@@ -378,6 +417,15 @@ void FMaterialInstance::setTransparencyMode(TransparencyMode const mode) noexcep
     mTransparencyMode = mode;
 }
 
+RasterState FMaterialInstance::getRasterState() const noexcept {
+    RasterState rs = mMaterial->getRasterState();
+    rs.culling = mCulling;
+    rs.depthWrite = mDepthWrite;
+    rs.depthFunc = mDepthFunc;
+    rs.colorWrite = mColorWrite;
+    return rs;
+}
+
 void FMaterialInstance::setDepthCulling(bool const enable) noexcept {
     mDepthFunc = enable ? RasterState::DepthFunc::GE : RasterState::DepthFunc::A;
 }
@@ -398,6 +446,51 @@ const char* FMaterialInstance::getName() const noexcept {
 
 // ------------------------------------------------------------------------------------------------
 
+void FMaterialInstance::compile(CompilerPriorityQueue const priority,
+        UserVariantFilterMask variantSpec, CallbackHandler* handler,
+        Invocable<void(MaterialInstance*)>&& callback) noexcept {
+    FEngine& engine = mMaterial->getEngine();
+    DriverApi& driver = engine.getDriverApi();
+    MaterialDefinition const& definition = mMaterial->getDefinition();
+
+    bool const isStereoSupported = driver.isStereoSupported();
+
+    // Turn off the STE variant if stereo is not supported.
+    if (UTILS_LIKELY(!isStereoSupported)) {
+        variantSpec &= ~UserVariantFilterMask(UserVariantFilterBit::STE);
+    }
+
+    UserVariantFilterMask const variantFilter =
+            ~variantSpec & UserVariantFilterMask(UserVariantFilterBit::ALL);
+    ShaderModel const shaderModel = engine.getShaderModel();
+
+    if (UTILS_LIKELY(driver.isParallelShaderCompileSupported())) {
+        for (auto const variant: definition.getVariants()) {
+            if (!variantFilter || variant == Variant::filterUserVariant(variant, variantFilter)) {
+                if (definition.hasVariant(variant, shaderModel, isStereoSupported)) {
+                    prepareProgram(driver, variant, priority);
+                }
+            }
+        }
+    }
+
+    if (callback) {
+        struct Callback {
+            Invocable<void(MaterialInstance*)> f;
+            MaterialInstance* m;
+            static void func(void* user) {
+                auto* const c = static_cast<Callback*>(user);
+                c->f(c->m);
+                delete c;
+            }
+        };
+        auto* const user = new (std::nothrow) Callback{ std::move(callback), this };
+        driver.compilePrograms(priority, handler, &Callback::func, user);
+    } else {
+        driver.compilePrograms(priority, nullptr, nullptr, nullptr);
+    }
+}
+
 void FMaterialInstance::use(FEngine::DriverApi& driver, Variant variant) const {
     assert_invariant(mDescriptorSet.getHandle());
     assert_invariant(!isUsingUboBatching() || BufferAllocator::isValid(getAllocationId()));
@@ -502,4 +595,34 @@ void FMaterialInstance::fixMissingSamplers() const {
     }
 }
 
+LocalProgramCache const& FMaterialInstance::getPrograms() const noexcept {
+    return mPrograms.isInitialized() ? mPrograms : mMaterial->getPrograms();
+}
+
+void FMaterialInstance::flushSpecializationConstants() const noexcept {
+    assert_invariant(!mPendingSpecializationConstants.empty());
+
+    if (!mPrograms.isInitialized()) {
+        mPrograms.initializeForMaterialInstance(mMaterial->getEngine(), *mMaterial);
+    }
+    mPrograms.setConstants(std::move(mPendingSpecializationConstants));
+    mPendingSpecializationConstants.clear();
+}
+
+#if FILAMENT_ENABLE_MATDBG
+
+void FMaterialInstance::updateActiveProgramsForMatdbg(Variant const variant) const noexcept {
+    mMaterial->updateActiveProgramsForMatdbg(variant);
+}
+
+#endif // FILAMENT_ENABLE_MATDBG
+
+template void FMaterialInstance::setConstantImpl<int32_t>(std::string_view name, int32_t value);
+template void FMaterialInstance::setConstantImpl<float>(std::string_view name, float value);
+template void FMaterialInstance::setConstantImpl<bool>(std::string_view name, bool value);
+
+template int32_t FMaterialInstance::getConstantImpl<int32_t>(std::string_view name) const;
+template float FMaterialInstance::getConstantImpl<float>(std::string_view name) const;
+template bool FMaterialInstance::getConstantImpl<bool>(std::string_view name) const;
+
 } // namespace filament

filament/src/details/MaterialInstance.h

@@ -18,7 +18,7 @@
 #define TNT_FILAMENT_DETAILS_MATERIALINSTANCE_H
 
 #include "downcast.h"
-
+#include "LocalProgramCache.h"
 #include "UniformBuffer.h"
 
 #include "ds/DescriptorSet.h"
@@ -26,8 +26,6 @@
 #include "details/BufferAllocator.h"
 #include "details/Engine.h"
 
-#include "private/backend/DriverApi.h"
-
 #include <filament/MaterialInstance.h>
 
 #include <private/filament/Variant.h>
@@ -67,7 +65,7 @@ public:
     ~FMaterialInstance() noexcept;
 
     void terminate(FEngine& engine);
-    
+
     void commit(FEngine& engine) const;
 
     void commit(FEngine::DriverApi& driver, UboManager* uboManager) const;
@@ -86,6 +84,34 @@ public:
 
     UniformBuffer const& getUniformBuffer() const noexcept { return mUniforms; }
 
+    void compile(backend::CompilerPriorityQueue priority, UserVariantFilterMask variantSpec,
+            backend::CallbackHandler* handler,
+            utils::Invocable<void(MaterialInstance*)>&& callback) noexcept;
+
+    // prepareProgram creates the program for the material's given variant at the backend level.
+    // Must be called outside of backend render pass.
+    // Must be called before getProgram() below.
+    backend::Handle<backend::HwProgram> prepareProgram(backend::DriverApi& driver,
+            Variant const variant,
+            backend::CompilerPriorityQueue const priorityQueue) const noexcept {
+        if (UTILS_UNLIKELY(!mPendingSpecializationConstants.empty())) {
+            flushSpecializationConstants();
+        }
+        return getPrograms().prepareProgram(driver, variant, priorityQueue);
+    }
+
+    // getProgram returns the backend program for the material's given variant.
+    // Must be called after prepareProgram().
+    //
+    // See also Material::getProgram().
+    [[nodiscard]]
+    backend::Handle<backend::HwProgram> getProgram(Variant const variant) const noexcept {
+#if FILAMENT_ENABLE_MATDBG
+        updateActiveProgramsForMatdbg(variant);
+#endif
+        return getPrograms().getProgram(variant);
+    }
+
     void setScissor(uint32_t const left, uint32_t const bottom, uint32_t const width, uint32_t const height) noexcept {
         constexpr uint32_t maxvalu = std::numeric_limits<int32_t>::max();
         mScissorRect = { int32_t(left), int32_t(bottom),
@@ -93,6 +119,10 @@ public:
         mHasScissor = true;
     }
 
+    void setScissor(backend::Viewport const& viewport) noexcept {
+        setScissor(viewport.left, viewport.bottom, viewport.width, viewport.height);
+    }
+
     void unsetScissor() noexcept {
         constexpr uint32_t maxvalu = std::numeric_limits<int32_t>::max();
         mScissorRect = { 0, 0, maxvalu, maxvalu };
@@ -103,6 +133,8 @@ public:
 
     bool hasScissor() const noexcept { return mHasScissor; }
 
+    backend::RasterState getRasterState() const noexcept;
+
     backend::CullingMode getCullingMode() const noexcept { return mCulling; }
 
     backend::CullingMode getShadowCullingMode() const noexcept { return mShadowCulling; }
@@ -250,11 +282,15 @@ public:
             backend::Handle<backend::HwTexture> texture, backend::SamplerParams params);
 
     using MaterialInstance::setParameter;
+    using MaterialInstance::setConstant;
 
 private:
     friend class FMaterial;
     friend class MaterialInstance;
 
+    // Cannot inline since it inspects the FMaterial class.
+    LocalProgramCache const& getPrograms() const noexcept;
+
     template<size_t Size>
     void setParameterUntypedImpl(std::string_view name, const void* value);
 
@@ -273,6 +309,19 @@ private:
     template<typename T>
     T getParameterImpl(std::string_view name) const;
 
+    template<typename T>
+    void setConstantImpl(std::string_view name, T value);
+
+    template<typename T>
+    T getConstantImpl(std::string_view name) const;
+
+    void flushSpecializationConstants() const noexcept;
+
+#if FILAMENT_ENABLE_MATDBG
+    // Called by getProgram() to update active program list for matdbg UI.
+    void updateActiveProgramsForMatdbg(Variant const variant) const noexcept;
+#endif
+
     // keep these grouped, they're accessed together in the render-loop
     FMaterial const* mMaterial = nullptr;
 
@@ -287,6 +336,11 @@ private:
     mutable DescriptorSet mDescriptorSet;
     UniformBuffer mUniforms;
 
+    // HACK: Mutable so that prepareProgram() can update specialization constants.
+    mutable LocalProgramCache mPrograms;
+    mutable utils::FixedCapacityVector<backend::Program::SpecializationConstant>
+            mPendingSpecializationConstants;
+
     backend::PolygonOffset mPolygonOffset{};
     backend::StencilState mStencilState{};
 

filament/src/details/Renderer.cpp

@@ -310,13 +310,7 @@ void FRenderer::skipFrame(uint64_t vsyncSteadyClockTimeNano) {
     engine.flush();     // flush command stream
 
     // Run GC
-    JobSystem& js = engine.getJobSystem();
-    auto *rootJob = js.createJob();
-    js.run(jobs::createJob(js, rootJob, &FEngine::gcManagers, &engine)); // gc all component managers
-    if (engine.isAsynchronousModeEnabled()) {
-        js.run(jobs::createJob(js, rootJob, &FEngine::gcDeferredAsyncObjectDestruction, &engine));
-    }
-    js.runAndWait(rootJob);
+    engine.gc();
 
     mFrameSkipper.frameSkipped();
 }
@@ -518,13 +512,7 @@ void FRenderer::endFrame() {
     engine.flush();     // flush command stream
 
     // Run GC
-    JobSystem& js = engine.getJobSystem();
-    auto *rootJob = js.createJob();
-    js.run(jobs::createJob(js, rootJob, &FEngine::gcManagers, &engine)); // gc all component managers
-    if (engine.isAsynchronousModeEnabled()) {
-        js.run(jobs::createJob(js, rootJob, &FEngine::gcDeferredAsyncObjectDestruction, &engine));
-    }
-    js.runAndWait(rootJob);
+    engine.gc();
 }
 
 void FRenderer::readPixels(
@@ -969,10 +957,7 @@ void FRenderer::renderJob(DriverApi& driver, RootArenaScope& rootArenaScope, FVi
     variant.setDirectionalLighting(view.hasDirectionalLighting());
     variant.setDynamicLighting(view.hasDynamicLighting());
     variant.setFog(view.hasFog());
-    // The VSM bit has a different meaning for STANDARD_VARIANT (as opposed to DEPTH_VARIANT),
-    // In the STANDARD_VARIANT case, we are *using* the shadow-map, and the VSM only decides which
-    // type of sampler is used (samplerShadow or sampler2D).
-    variant.setVsm(view.hasShadowing() && view.getShadowType() != ShadowType::PCF);
+    variant.setShadowSampler2D(view.hasShadowing() && view.getShadowType() != ShadowType::PCF);
     variant.setStereo(view.hasStereo());
 
     /*
@@ -981,10 +966,7 @@ void FRenderer::renderJob(DriverApi& driver, RootArenaScope& rootArenaScope, FVi
 
     if (view.needsShadowMap()) {
         Variant shadowVariant(Variant::DEPTH_VARIANT);
-        // The VSM bit has a different meaning for DEPTH_VARIANT (as opposed to STANDARD_VARIANT),
-        // In the DEPTH_VARIANT case, we are *generating* the shadow-map, and some computations
-        // are handled differently. In addition, the color buffer is used.
-        shadowVariant.setVsm(view.getShadowType() == ShadowType::VSM);
+        shadowVariant.setDepthMoments(view.getShadowType() == ShadowType::VSM);
 
         auto shadows = view.renderShadowMaps(engine, fg, cameraInfo, mShaderUserTime,
                 RenderPassBuilder{ commandArena }

filament/src/details/View.cpp

@@ -76,6 +76,7 @@
 #include <cmath>
 #include <chrono>
 #include <functional>
+#include <limits>
 #include <memory>
 #include <new>
 #include <ratio>
@@ -143,7 +144,7 @@ FView::FView(FEngine& engine)
 #ifndef NDEBUG
     // This can fail if another view has already registered this data source
     mDebugState->owner = debugRegistry.registerDataSource("d.view.frame_info",
-            [weak = std::weak_ptr(mDebugState)]() -> DebugRegistry::DataSource {
+            [weak = std::weak_ptr<DebugState>(mDebugState)]() -> DebugRegistry::DataSource {
                 // the View could have been destroyed by the time we do this
                 auto const state = weak.lock();
                 if (!state) {
@@ -1131,8 +1132,6 @@ void FView::prepareShadowMapping() const noexcept {
         uniforms = mShadowMapManager->getShadowMappingUniforms();
     }
 
-    constexpr float low  = 5.54f; // ~ std::log(std::numeric_limits<math::half>::max()) * 0.5f;
-    constexpr float high = 42.0f; // ~ std::log(std::numeric_limits<float>::max()) * 0.5f;
     constexpr uint32_t SHADOW_SAMPLING_RUNTIME_PCF = 0u;
     constexpr uint32_t SHADOW_SAMPLING_RUNTIME_EVSM = 1u;
     constexpr uint32_t SHADOW_SAMPLING_RUNTIME_DPCF = 2u;
@@ -1148,8 +1147,8 @@ void FView::prepareShadowMapping() const noexcept {
             break;
         case ShadowType::VSM:
             s.shadowSamplingType = SHADOW_SAMPLING_RUNTIME_EVSM;
-            s.vsmExponent = mVsmShadowOptions.highPrecision ? high : low;
-            s.vsmDepthScale = mVsmShadowOptions.minVarianceScale * 0.01f * s.vsmExponent;
+            s.vsmExponent = 0; // this is only used when rendering the shadowmap, not when using it
+            s.vsmMaxMoment = ShadowMapManager::getMaxMomentEVSM(mVsmShadowOptions);
             s.vsmLightBleedReduction = mVsmShadowOptions.lightBleedReduction;
             break;
         case ShadowType::DPCF:

filament/src/ds/ShadowMapDescriptorSet.cpp

@@ -28,9 +28,11 @@
 
 #include <utils/debug.h>
 
+#include <math/half.h>
 #include <math/vec4.h>
 
 #include <array>
+#include <limits>
 
 namespace filament {
 
@@ -74,8 +76,8 @@ void ShadowMapDescriptorSet::prepareLodBias(Transaction const& transaction, floa
 }
 
 void ShadowMapDescriptorSet::prepareViewport(Transaction const& transaction,
-        backend::Viewport const& viewport) noexcept {
-    PerViewDescriptorSetUtils::prepareViewport(edit(transaction), viewport, viewport);
+        backend::Viewport const& physicalViewport, backend::Viewport const& logicalViewport) noexcept {
+    PerViewDescriptorSetUtils::prepareViewport(edit(transaction), physicalViewport, logicalViewport);
     // TODO: offset calculation is now different
 }
 
@@ -90,11 +92,10 @@ void ShadowMapDescriptorSet::prepareMaterialGlobals(Transaction const& transacti
 }
 
 void ShadowMapDescriptorSet::prepareShadowMapping(Transaction const& transaction,
-        bool const highPrecision) noexcept {
+        float const vsmExponent, float const vsmMaxMoment) noexcept {
     auto& s = edit(transaction);
-    constexpr float low  = 5.54f; // ~ std::log(std::numeric_limits<math::half>::max()) * 0.5f;
-    constexpr float high = 42.0f; // ~ std::log(std::numeric_limits<float>::max()) * 0.5f;
-    s.vsmExponent = highPrecision ? high : low;
+    s.vsmExponent = vsmExponent;
+    s.vsmMaxMoment = vsmMaxMoment;
 }
 
 ShadowMapDescriptorSet::Transaction ShadowMapDescriptorSet::open(DriverApi& driver) noexcept {

filament/src/ds/ShadowMapDescriptorSet.h

@@ -66,7 +66,7 @@ public:
             float bias) noexcept;
 
     static void prepareViewport(Transaction const& transaction,
-            backend::Viewport const& viewport) noexcept;
+            backend::Viewport const& physicalViewport, backend::Viewport const& logicalViewport) noexcept;
 
     static void prepareTime(Transaction const& transaction,
             FEngine const& engine, math::float4 const& userTime) noexcept;
@@ -75,7 +75,7 @@ public:
             std::array<math::float4, 4> const& materialGlobals) noexcept;
 
     static void prepareShadowMapping(Transaction const& transaction,
-            bool highPrecision) noexcept;
+            float vsmExponent, float vsmMaxMoment) noexcept;
 
     static Transaction open(backend::DriverApi& driver) noexcept;
 

filament/src/materials/vsmMipmap.mat

@@ -19,7 +19,8 @@ material {
         {
             name : color,
             target : color,
-            type : float4
+            type : float4,
+            precision : high
         }
     ],
     domain : postprocess,
@@ -30,8 +31,29 @@ material {
 
 fragment {
     void postProcess(inout PostProcessInputs postProcess) {
-        highp vec2 uv = gl_FragCoord.xy * materialParams.uvscale;
-        postProcess.color = textureLod(materialParams_color,
-                vec3(uv, materialParams.layer), 0.0);
+        // For EVSM mipmapping, we use a custom box filter (instead of the GPU's bilinear filter), because GPUs often
+        // use a lower precision for texture interpolation and EVSM moments are very sensitive to that.
+
+        highp ivec2 destCoord = ivec2(gl_FragCoord.xy);
+        highp ivec3 srcCoord = ivec3(destCoord * 2, materialParams.layer);
+
+#if defined(TARGET_WEBGPU_ENVIRONMENT)
+        // it looks like WebGPU doesn't support texelFetchOffset
+        highp vec4 m0 = texelFetch(materialParams_color, ivec3(srcCoord.xy + ivec2(0, 0), srcCoord.z), 0);
+        highp vec4 m1 = texelFetch(materialParams_color, ivec3(srcCoord.xy + ivec2(1, 0), srcCoord.z), 0);
+        highp vec4 m2 = texelFetch(materialParams_color, ivec3(srcCoord.xy + ivec2(0, 1), srcCoord.z), 0);
+        highp vec4 m3 = texelFetch(materialParams_color, ivec3(srcCoord.xy + ivec2(1, 1), srcCoord.z), 0);
+#else
+        highp vec4 m0 = texelFetchOffset(materialParams_color, srcCoord, 0, ivec2(0, 0));
+        highp vec4 m1 = texelFetchOffset(materialParams_color, srcCoord, 0, ivec2(1, 0));
+        highp vec4 m2 = texelFetchOffset(materialParams_color, srcCoord, 0, ivec2(0, 1));
+        highp vec4 m3 = texelFetchOffset(materialParams_color, srcCoord, 0, ivec2(1, 1));
+#endif
+
+        // Make sure to not overflow the moments when averaging them
+        postProcess.color = (m0 * 0.25) +
+                            (m1 * 0.25) +
+                            (m2 * 0.25) +
+                            (m3 * 0.25);
     }
 }

filament/test/filament_test_material.cpp

@@ -19,6 +19,7 @@
 
 #include <filament/Engine.h>
 #include <filament/Material.h>
+#include <filament/MaterialInstance.h>
 
 #include "filament_test_resources.h"
 
@@ -167,3 +168,52 @@ TEST(Material, MaterialSettingInvalidApiLevelReturnsAnInvalidPackage) {
 
     Engine::destroy(engine);
 }
+
+TEST(MaterialInstanceTest, SetConstant) {
+    Engine* engine = Engine::create(Engine::Backend::NOOP);
+
+    std::string shaderCode(R"(
+        void material(inout MaterialInputs material) {
+            prepareMaterial(material);
+            material.baseColor = vec4(1.0);
+        }
+    )");
+
+    filamat::MaterialBuilder builder;
+    builder.init();
+    builder.name("MaterialInstanceTest");
+    builder.material(shaderCode.c_str());
+    builder.constant("myFloat", filamat::MaterialBuilder::ConstantType::FLOAT, 1.0f);
+    builder.constant("myInt", filamat::MaterialBuilder::ConstantType::INT, 2);
+    builder.constant("myBool", filamat::MaterialBuilder::ConstantType::BOOL, false);
+
+    filamat::Package result = builder.build(engine->getJobSystem());
+    ASSERT_TRUE(result.isValid());
+
+    Material* material = Material::Builder()
+            .package(result.getData(), result.getSize())
+            .build(*engine);
+    ASSERT_NE(material, nullptr);
+
+    MaterialInstance* instance = material->createInstance();
+    ASSERT_NE(instance, nullptr);
+
+    // Verify default values
+    EXPECT_EQ(instance->getConstant<float>("myFloat"), 1.0f);
+    EXPECT_EQ(instance->getConstant<int32_t>("myInt"), 2);
+    EXPECT_EQ(instance->getConstant<bool>("myBool"), false);
+
+    // Set new values
+    instance->setConstant("myFloat", 3.0f);
+    instance->setConstant("myInt", 4);
+    instance->setConstant("myBool", true);
+
+    // Verify new values
+    EXPECT_EQ(instance->getConstant<float>("myFloat"), 3.0f);
+    EXPECT_EQ(instance->getConstant<int32_t>("myInt"), 4);
+    EXPECT_EQ(instance->getConstant<bool>("myBool"), true);
+
+    engine->destroy(instance);
+    engine->destroy(material);
+    Engine::destroy(engine);
+}

ios/CocoaPods/Filament.podspec

@@ -1,12 +1,12 @@
 Pod::Spec.new do |spec|
   spec.name = "Filament"
-  spec.version = "1.69.4"
+  spec.version = "1.69.5"
   spec.license = { :type => "Apache 2.0", :file => "LICENSE" }
   spec.homepage = "https://google.github.io/filament"
   spec.authors = "Google LLC."
   spec.summary = "Filament is a real-time physically based rendering engine for Android, iOS, Windows, Linux, macOS, and WASM/WebGL."
   spec.platform = :ios, "11.0"
-  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.69.4/filament-v1.69.4-ios.tgz" }
+  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.69.5/filament-v1.69.5-ios.tgz" }
 
   spec.libraries = 'c++'
 

libs/filabridge/include/private/filament/DescriptorSets.h

@@ -36,7 +36,7 @@ backend::DescriptorSetLayout const& getPerRenderableLayout() noexcept;
 backend::DescriptorSetLayout getPerViewDescriptorSetLayout(
         MaterialDomain domain,
         bool isLit, bool isSSR, bool hasFog,
-        bool isVSM) noexcept;
+        bool isShadowSampler2D) noexcept;
 
 backend::DescriptorSetLayout getPerViewDescriptorSetLayoutWithVariant(
         Variant variant,