vk: testing a robin_map key hash failure

Executable on device with:

./build.sh -p android -q arm64-v8a release test_compiler && adb push out/cmake-android-release-aarch64/filament/test/test_compiler /data/local/tmp/ && adb shell chmod 777 /data/local/tmp/test_compiler && adb shell /data/local/tmp/test_compiler

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.53.0'
+    implementation 'com.google.android.filament:filament-android:1.53.2'
 }
 ```
 
@@ -51,7 +51,7 @@ Here are all the libraries available in the group `com.google.android.filament`:
 iOS projects can use CocoaPods to install the latest release:
 
 ```shell
-pod 'Filament', '~> 1.53.0'
+pod 'Filament', '~> 1.53.2'
 ```
 
 ### Snapshots
@@ -176,6 +176,7 @@ steps:
   - [x] KHR_materials_unlit
   - [x] KHR_materials_variants
   - [x] KHR_materials_volume
+  - [x] KHR_materials_specular
   - [x] KHR_mesh_quantization
   - [x] KHR_texture_basisu
   - [x] KHR_texture_transform

RELEASE_NOTES.md

@@ -7,6 +7,16 @@ A new header is inserted each time a *tag* is created.
 Instead, if you are authoring a PR for the main branch, add your release note to
 [NEW_RELEASE_NOTES.md](./NEW_RELEASE_NOTES.md).
 
+## v1.53.3
+
+- Add drag and drop support for IBL files for desktop gltf_viewer.
+
+## v1.53.2
+
+
+## v1.53.1
+
+
 ## v1.53.0
 
 - engine: fix skinning normals with large transforms (b/342459864) [⚠️ **New Material Version**]

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

@@ -571,3 +571,9 @@ Java_com_google_android_filament_Engine_nBuilderBuild(JNIEnv*, jclass, jlong nat
     Engine::Builder* builder = (Engine::Builder*) nativeBuilder;
     return (jlong) builder->build();
 }
+
+extern "C"
+JNIEXPORT jlong JNICALL
+Java_com_google_android_filament_Engine_getSteadyClockTimeNano(JNIEnv *env, jclass clazz) {
+    return (jlong)Engine::getSteadyClockTimeNano();
+}

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

@@ -25,12 +25,17 @@ using namespace filament;
 
 extern "C" JNIEXPORT jlong JNICALL
 Java_com_google_android_filament_Material_nBuilderBuild(JNIEnv *env, jclass,
-        jlong nativeEngine, jobject buffer_, jint size) {
+        jlong nativeEngine, jobject buffer_, jint size, jint shBandCount) {
     Engine* engine = (Engine*) nativeEngine;
     AutoBuffer buffer(env, buffer_, size);
-    Material* material = Material::Builder()
+    auto builder = Material::Builder();
+    if (shBandCount) {
+        builder.sphericalHarmonicsBandCount(shBandCount);
+    }
+    Material* material = builder
             .package(buffer.getData(), buffer.getSize())
             .build(*engine);
+
     return (jlong) material;
 }
 

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

@@ -253,16 +253,10 @@ Java_com_google_android_filament_RenderableManager_nBuilderMorphingStandard(JNIE
 }
 
 extern "C" JNIEXPORT void JNICALL
-Java_com_google_android_filament_RenderableManager_nBuilderSetMorphTargetBufferAt(JNIEnv*, jclass,
-        jlong nativeBuilder, int level, int primitiveIndex, jlong nativeMorphTargetBuffer,
-        int offset, int count) {
+Java_com_google_android_filament_RenderableManager_nBuilderSetMorphTargetBufferOffsetAt(JNIEnv*, jclass,
+        jlong nativeBuilder, int level, int primitiveIndex, int offset) {
     RenderableManager::Builder *builder = (RenderableManager::Builder *) nativeBuilder;
-    if (nativeMorphTargetBuffer) {
-        MorphTargetBuffer *morphTargetBuffer = (MorphTargetBuffer *) nativeMorphTargetBuffer;
-        builder->morphing(level, primitiveIndex, morphTargetBuffer, offset, count);
-    } else {
-        builder->morphing(level, primitiveIndex, offset, count);
-    }
+    builder->morphing(level, primitiveIndex, offset);
 }
 
 extern "C" JNIEXPORT void JNICALL
@@ -334,18 +328,12 @@ Java_com_google_android_filament_RenderableManager_nSetMorphWeights(JNIEnv* env,
 }
 
 extern "C" JNIEXPORT void JNICALL
-Java_com_google_android_filament_RenderableManager_nSetMorphTargetBufferAt(JNIEnv*,
+Java_com_google_android_filament_RenderableManager_nSetMorphTargetBufferOffsetAt(JNIEnv*,
         jclass, jlong nativeRenderableManager, jint i, int level, jint primitiveIndex,
-        jlong nativeMorphTargetBuffer, jint offset, jint count) {
+        jlong, jint offset) {
     RenderableManager *rm = (RenderableManager *) nativeRenderableManager;
-    if (nativeMorphTargetBuffer) {
-        MorphTargetBuffer *morphTargetBuffer = (MorphTargetBuffer *) nativeMorphTargetBuffer;
-        rm->setMorphTargetBufferAt((RenderableManager::Instance) i, (uint8_t) level,
-                (size_t) primitiveIndex, morphTargetBuffer, (size_t) offset, (size_t) count);
-    } else {
-        rm->setMorphTargetBufferAt((RenderableManager::Instance) i, (uint8_t) level,
-                (size_t) primitiveIndex, (size_t) offset, (size_t) count);
-    }
+    rm->setMorphTargetBufferOffsetAt((RenderableManager::Instance) i, (uint8_t) level,
+            (size_t) primitiveIndex, (size_t) offset);
 }
 
 extern "C" JNIEXPORT jint JNICALL

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

@@ -28,6 +28,14 @@
 using namespace filament;
 using namespace backend;
 
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_Renderer_nSkipFrame(JNIEnv *, jclass, jlong nativeRenderer,
+        jlong vsyncSteadyClockTimeNano) {
+    Renderer *renderer = (Renderer *) nativeRenderer;
+    renderer->skipFrame(uint64_t(vsyncSteadyClockTimeNano));
+}
+
 extern "C" JNIEXPORT jboolean JNICALL
 Java_com_google_android_filament_Renderer_nBeginFrame(JNIEnv *, jclass, jlong nativeRenderer,
         jlong nativeSwapChain, jlong frameTimeNanos) {
@@ -187,3 +195,10 @@ Java_com_google_android_filament_Renderer_nSetPresentationTime(JNIEnv *, jclass
     Renderer *renderer = (Renderer *) nativeRenderer;
     renderer->setPresentationTime(monotonicClockNanos);
 }
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_Renderer_nSetVsyncTime(JNIEnv *, jclass,
+    jlong nativeRenderer, jlong steadyClockTimeNano) {
+    Renderer *renderer = (Renderer *) nativeRenderer;
+    renderer->setVsyncTime(steadyClockTimeNano);
+}

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

@@ -531,3 +531,12 @@ Java_com_google_android_filament_View_nGetFogEntity(JNIEnv *env, jclass clazz,
     View *view = (View *) nativeView;
     return (jint)view->getFogEntity().getId();
 }
+
+extern "C"
+JNIEXPORT void JNICALL
+Java_com_google_android_filament_View_nClearFrameHistory(JNIEnv *env, jclass clazz,
+        jlong nativeView, jlong nativeEngine) {
+    View *view = (View *) nativeView;
+    Engine *engine = (Engine *) nativeEngine;
+    view->clearFrameHistory(*engine);
+}

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

@@ -425,9 +425,12 @@ public class Engine {
         public long resourceAllocatorCacheSizeMB = 64;
 
         /*
-         * This value determines for how many frames are texture entries kept in the cache.
+         * This value determines how many frames texture entries are kept for in the cache. This
+         * is a soft limit, meaning some texture older than this are allowed to stay in the cache.
+         * Typically only one texture is evicted per frame.
+         * The default is 1.
          */
-        public long resourceAllocatorCacheMaxAge = 2;
+        public long resourceAllocatorCacheMaxAge = 1;
 
         /*
          * Disable backend handles use-after-free checks.
@@ -1292,7 +1295,6 @@ public class Engine {
     /**
      * Switch the command queue to unprotected mode. Protected mode can be activated via
      * Renderer::beginFrame() using a protected SwapChain.
-
      * @see Renderer
      * @see SwapChain
      */
@@ -1300,6 +1302,13 @@ public class Engine {
         nUnprotected(getNativeObject());
     }
 
+    /**
+     * Get the current time. This is a convenience function that simply returns the
+     * time in nanosecond since epoch of std::chrono::steady_clock.
+     * @return current time in nanosecond since epoch of std::chrono::steady_clock.
+     * @see Renderer#beginFrame
+     */
+    public static native long getSteadyClockTimeNano();
 
     @UsedByReflection("TextureHelper.java")
     public long getNativeObject() {

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

@@ -346,6 +346,7 @@ public class Material {
     public static class Builder {
         private Buffer mBuffer;
         private int mSize;
+        private int mShBandCount = 0;
 
         /**
          * Specifies the material data. The material data is a binary blob produced by
@@ -361,6 +362,22 @@ public class Material {
             return this;
         }
 
+        /**
+         * Sets the quality of the indirect lights computations. This is only taken into account
+         * if this material is lit and in the surface domain. This setting will affect the
+         * IndirectLight computation if one is specified on the Scene and Spherical Harmonics
+         * are used for the irradiance.
+         *
+         * @param shBandCount Number of spherical harmonic bands. Must be 1, 2 or 3 (default).
+         * @return Reference to this Builder for chaining calls.
+         * @see IndirectLight
+         */
+        @NonNull
+        public Builder sphericalHarmonicsBandCount(@IntRange(from = 0) int shBandCount) {
+            mShBandCount = shBandCount;
+            return this;
+        }
+
         /**
          * Creates and returns the Material object.
          *
@@ -372,7 +389,8 @@ public class Material {
          */
         @NonNull
         public Material build(@NonNull Engine engine) {
-            long nativeMaterial = nBuilderBuild(engine.getNativeObject(), mBuffer, mSize);
+            long nativeMaterial = nBuilderBuild(engine.getNativeObject(),
+                mBuffer, mSize, mShBandCount);
             if (nativeMaterial == 0) throw new IllegalStateException("Couldn't create Material");
             return new Material(nativeMaterial);
         }
@@ -1023,7 +1041,7 @@ public class Material {
         mNativeObject = 0;
     }
 
-    private static native long nBuilderBuild(long nativeEngine, @NonNull Buffer buffer, int size);
+    private static native long nBuilderBuild(long nativeEngine, @NonNull Buffer buffer, int size, int shBandCount);
     private static native long nCreateInstance(long nativeMaterial);
     private static native long nCreateInstanceWithName(long nativeMaterial, @NonNull String name);
     private static native long nGetDefaultInstance(long nativeMaterial);

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

@@ -74,7 +74,7 @@ public class MorphTargetBuffer {
          *
          * @exception IllegalStateException if the MorphTargetBuffer could not be created
          *
-         * @see #setMorphTargetBufferAt
+         * @see #setMorphTargetBufferOffsetAt
          */
         @NonNull
         public MorphTargetBuffer build(@NonNull Engine engine) {

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

@@ -569,40 +569,13 @@ public class RenderableManager {
          *
          * @param level the level of detail (lod), only 0 can be specified
          * @param primitiveIndex zero-based index of the primitive, must be less than the count passed to Builder constructor
-         * @param morphTargetBuffer specifies the morph target buffer
          * @param offset specifies where in the morph target buffer to start reading (expressed as a number of vertices)
-         * @param count number of vertices in the morph target buffer to read, must equal the geometry's count (for triangles, this should be a multiple of 3)
          */
         @NonNull
         public Builder morphing(@IntRange(from = 0) int level,
                                 @IntRange(from = 0) int primitiveIndex,
-                                @IntRange(from = 0) int offset,
-                                @IntRange(from = 0) int count) {
-            nBuilderSetMorphTargetBufferAt(mNativeBuilder, level, primitiveIndex, 0, offset, count);
-            return this;
-        }
-
-        /** @deprecated */
-        @Deprecated
-        @NonNull
-        public Builder morphing(@IntRange(from = 0) int level,
-                                @IntRange(from = 0) int primitiveIndex,
-                                @NonNull MorphTargetBuffer morphTargetBuffer,
-                                @IntRange(from = 0) int offset,
-                                @IntRange(from = 0) int count) {
-            nBuilderSetMorphTargetBufferAt(mNativeBuilder, level, primitiveIndex,
-                    morphTargetBuffer.getNativeObject(), offset, count);
-            return this;
-        }
-
-        /** @deprecated */
-        @Deprecated
-        @NonNull
-        public Builder morphing(@IntRange(from = 0) int level,
-                                @IntRange(from = 0) int primitiveIndex,
-                                @NonNull MorphTargetBuffer morphTargetBuffer) {
-            nBuilderSetMorphTargetBufferAt(mNativeBuilder, level, primitiveIndex,
-                    morphTargetBuffer.getNativeObject(), 0, morphTargetBuffer.getVertexCount());
+                                @IntRange(from = 0) int offset) {
+            nBuilderSetMorphTargetBufferOffsetAt(mNativeBuilder, level, primitiveIndex, offset);
             return this;
         }
 
@@ -705,34 +678,11 @@ public class RenderableManager {
      *
      * @see Builder#morphing
      */
-    public void setMorphTargetBufferAt(@EntityInstance int i,
+    public void setMorphTargetBufferOffsetAt(@EntityInstance int i,
                                        @IntRange(from = 0) int level,
                                        @IntRange(from = 0) int primitiveIndex,
-                                       @IntRange(from = 0) int offset,
-                                       @IntRange(from = 0) int count) {
-        nSetMorphTargetBufferAt(mNativeObject, i, level, primitiveIndex, 0, offset, count);
-    }
-
-    /** @deprecated */
-    @Deprecated
-    public void setMorphTargetBufferAt(@EntityInstance int i,
-                                       @IntRange(from = 0) int level,
-                                       @IntRange(from = 0) int primitiveIndex,
-                                       @NonNull MorphTargetBuffer morphTargetBuffer,
-                                       @IntRange(from = 0) int offset,
-                                       @IntRange(from = 0) int count) {
-        nSetMorphTargetBufferAt(mNativeObject, i, level, primitiveIndex,
-                morphTargetBuffer.getNativeObject(), offset, count);
-    }
-
-    /** @deprecated */
-    @Deprecated
-    public void setMorphTargetBufferAt(@EntityInstance int i,
-                                       @IntRange(from = 0) int level,
-                                       @IntRange(from = 0) int primitiveIndex,
-                                       @NonNull MorphTargetBuffer morphTargetBuffer) {
-        nSetMorphTargetBufferAt(mNativeObject, i, level, primitiveIndex,
-                morphTargetBuffer.getNativeObject(), 0, morphTargetBuffer.getVertexCount());
+                                       @IntRange(from = 0) int offset) {
+        nSetMorphTargetBufferOffsetAt(mNativeObject, i, level, primitiveIndex, 0, offset);
     }
 
     /**
@@ -1033,7 +983,7 @@ public class RenderableManager {
     private static native void nBuilderSkinningBuffer(long nativeBuilder, long nativeSkinningBuffer, int boneCount, int offset);
     private static native void nBuilderMorphing(long nativeBuilder, int targetCount);
     private static native void nBuilderMorphingStandard(long nativeBuilder, long nativeMorphTargetBuffer);
-    private static native void nBuilderSetMorphTargetBufferAt(long nativeBuilder, int level, int primitiveIndex, long nativeMorphTargetBuffer, int offset, int count);
+    private static native void nBuilderSetMorphTargetBufferOffsetAt(long nativeBuilder, int level, int primitiveIndex, int offset);
     private static native void nBuilderEnableSkinningBuffers(long nativeBuilder, boolean enabled);
     private static native void nBuilderFog(long nativeBuilder, boolean enabled);
     private static native void nBuilderLightChannel(long nativeRenderableManager, int channel, boolean enable);
@@ -1043,7 +993,7 @@ public class RenderableManager {
     private static native int nSetBonesAsMatrices(long nativeObject, int i, Buffer matrices, int remaining, int boneCount, int offset);
     private static native int nSetBonesAsQuaternions(long nativeObject, int i, Buffer quaternions, int remaining, int boneCount, int offset);
     private static native void nSetMorphWeights(long nativeObject, int instance, float[] weights, int offset);
-    private static native void nSetMorphTargetBufferAt(long nativeObject, int i, int level, int primitiveIndex, long nativeMorphTargetBuffer, int offset, int count);
+    private static native void nSetMorphTargetBufferOffsetAt(long nativeObject, int i, int level, int primitiveIndex, long nativeMorphTargetBuffer, int offset);
     private static native int nGetMorphTargetCount(long nativeObject, int i);
     private static native void nSetAxisAlignedBoundingBox(long nativeRenderableManager, int i, float cx, float cy, float cz, float ex, float ey, float ez);
     private static native void nSetLayerMask(long nativeRenderableManager, int i, int select, int value);

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

@@ -284,6 +284,33 @@ public class Renderer {
         nSetPresentationTime(getNativeObject(), monotonicClockNanos);
     }
 
+    /**
+     * The use of this method is optional. It sets the VSYNC time expressed as the duration in
+     * nanosecond since epoch of std::chrono::steady_clock.
+     * If called, passing 0 to frameTimeNanos in Renderer.BeginFrame will use this
+     * time instead.
+     * @param steadyClockTimeNano duration in nanosecond since epoch of std::chrono::steady_clock
+     * @see Engine#getSteadyClockTimeNano
+     * @see Renderer#beginFrame
+     */
+    public void setVsyncTime(long steadyClockTimeNano) {
+        nSetVsyncTime(getNativeObject(), steadyClockTimeNano);
+    }
+
+    /**
+     * Call skipFrame when momentarily skipping frames, for instance if the content of the
+     * scene doesn't change.
+     *
+     * @param vsyncSteadyClockTimeNano The time in nanoseconds when the frame started being rendered,
+     *                       in the {@link System#nanoTime()} timebase. Divide this value by 1000000 to
+     *                       convert it to the {@link android.os.SystemClock#uptimeMillis()}
+     *                       time base. This typically comes from
+     *                       {@link android.view.Choreographer.FrameCallback}.
+     */
+    public void skipFrame(long vsyncSteadyClockTimeNano) {
+        nSkipFrame(getNativeObject(), vsyncSteadyClockTimeNano);
+    }
+
     /**
      * Sets up a frame for this <code>Renderer</code>.
      * <p><code>beginFrame</code> manages frame pacing, and returns whether or not a frame should be
@@ -702,6 +729,8 @@ public class Renderer {
     }
 
     private static native void nSetPresentationTime(long nativeObject, long monotonicClockNanos);
+    private static native void nSetVsyncTime(long nativeObject, long steadyClockTimeNano);
+    private static native void nSkipFrame(long nativeObject, long vsyncSteadyClockTimeNano);
     private static native boolean nBeginFrame(long nativeRenderer, long nativeSwapChain, long frameTimeNanos);
     private static native void nEndFrame(long nativeRenderer);
     private static native void nRender(long nativeRenderer, long nativeView);

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

@@ -1233,6 +1233,18 @@ public class View {
         return nGetFogEntity(getNativeObject());
     }
 
+    /**
+     * When certain temporal features are used (e.g.: TAA or Screen-space reflections), the view
+     * keeps a history of previous frame renders associated with the Renderer the view was last
+     * used with. When switching Renderer, it may be necessary to clear that history by calling
+     * this method. Similarly, if the whole content of the screen change, like when a cut-scene
+     * starts, clearing the history might be needed to avoid artifacts due to the previous frame
+     * being very different.
+     */
+    public void clearFrameHistory(Engine engine) {
+        nClearFrameHistory(getNativeObject(), engine.getNativeObject());
+    }
+
     public long getNativeObject() {
         if (mNativeObject == 0) {
             throw new IllegalStateException("Calling method on destroyed View");
@@ -1294,7 +1306,7 @@ public class View {
     private static native void nSetMaterialGlobal(long nativeView, int index, float x, float y, float z, float w);
     private static native void nGetMaterialGlobal(long nativeView, int index, float[] out);
     private static native int nGetFogEntity(long nativeView);
-
+    private static native void nClearFrameHistory(long nativeView, long nativeEngine);
 
     /**
      * List of available ambient occlusion techniques.

android/filament-utils-android/src/main/cpp/Manipulator.cpp

@@ -125,6 +125,11 @@ extern "C" JNIEXPORT void Java_com_google_android_filament_utils_Manipulator_nBu
     builder->groundPlane(a, b, c, d);
 }
 
+extern "C" JNIEXPORT void Java_com_google_android_filament_utils_Manipulator_nBuilderPanning(JNIEnv*, jclass, jlong nativeBuilder, jboolean enabled) {
+    Builder* builder = (Builder*) nativeBuilder;
+    builder->panning(enabled);
+}
+
 extern "C" JNIEXPORT long Java_com_google_android_filament_utils_Manipulator_nBuilderBuild(JNIEnv*, jclass, jlong nativeBuilder, jint mode) {
     Builder* builder = (Builder*) nativeBuilder;
     return (jlong) builder->build((Mode) mode);

android/filament-utils-android/src/main/java/com/google/android/filament/utils/Manipulator.java

@@ -274,6 +274,17 @@ public class Manipulator {
             return this;
         }
 
+        /**
+         * Sets whether panning is enabled in the manipulator.
+         *
+         * @return this <code>Builder</code> object for chaining calls
+         */
+        @NonNull
+        public Builder panning(Boolean enabled) {
+            nBuilderPanning(mNativeBuilder, enabled);
+            return this;
+        }
+
         /**
          * Creates and returns the <code>Manipulator</code> object.
          *
@@ -483,6 +494,7 @@ public class Manipulator {
     private static native void nBuilderFlightPanSpeed(long nativeBuilder, float x, float y);
     private static native void nBuilderFlightMoveDamping(long nativeBuilder, float damping);
     private static native void nBuilderGroundPlane(long nativeBuilder, float a, float b, float c, float d);
+    private static native void nBuilderPanning(long nativeBuilder, Boolean enabled);
     private static native long nBuilderBuild(long nativeBuilder, int mode);
 
     private static native void nDestroyManipulator(long nativeManip);

android/gradle.properties

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.53.0
+VERSION_NAME=1.53.2
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 

filament/backend/include/backend/Platform.h

@@ -75,6 +75,8 @@ public:
          */
         size_t textureUseAfterFreePoolSize = 0;
 
+        size_t metalUploadBufferSizeBytes = 512 * 1024;
+
         /**
          * Set to `true` to forcibly disable parallel shader compilation in the backend.
          * Currently only honored by the GL and Metal backends.

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

@@ -90,6 +90,20 @@ public:
         VkExtent2D extent = {0, 0};
     };
 
+    struct ImageSyncData {
+        static constexpr uint32_t INVALID_IMAGE_INDEX = UINT32_MAX;
+
+        // The index of the next image as returned by vkAcquireNextImage or equivalent.
+        uint32_t imageIndex = INVALID_IMAGE_INDEX;
+
+        // Semaphore to be signaled once the image is available.
+        VkSemaphore imageReadySemaphore = VK_NULL_HANDLE;
+
+        // A function called right before vkQueueSubmit. After this call, the image must be 
+        // available. This pointer can be null if imageReadySemaphore is not VK_NULL_HANDLE.
+        std::function<void(SwapChainPtr handle)> explicitImageReadyWait = nullptr;
+    };
+
     VulkanPlatform();
 
     ~VulkanPlatform() override;
@@ -127,6 +141,12 @@ public:
          * before recreating the swapchain. Default is true.
          */
         bool flushAndWaitOnWindowResize = true;
+
+        /**
+         * Whether the swapchain image should be transitioned to a layout suitable for
+         * presentation. Default is true.
+         */
+        bool transitionSwapChainImageLayoutForPresent = true;
     };
 
     /**
@@ -155,13 +175,10 @@ public:
      * corresponding VkImage will be used as the output color attachment. The client should signal
      * the `clientSignal` semaphore when the image is ready to be used by the backend.
      * @param handle         The handle returned by createSwapChain()
-     * @param clientSignal   The semaphore that the client will signal to indicate that the backend
-     *                       may render into the image.
-     * @param index          Pointer to memory that will be filled with the index that corresponding
-     *                       to an image in the `SwapChainBundle.colors` array.
+     * @param outImageSyncData The synchronization data used for image readiness
      * @return               Result of acquire
      */
-    virtual VkResult acquire(SwapChainPtr handle, VkSemaphore clientSignal, uint32_t* index);
+    virtual VkResult acquire(SwapChainPtr handle, ImageSyncData* outImageSyncData);
 
     /**
      * Present the image corresponding to `index` to the display. The client should wait on

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

@@ -181,6 +181,18 @@ public:
         return static_cast<Dp>(p);
     }
 
+    template<typename B>
+    bool is_valid(Handle<B>& handle) {
+        if (handle && isPoolHandle(handle.getId())) {
+            auto [p, tag] = handleToPointer(handle.getId());
+            uint8_t const age = (tag & HANDLE_AGE_MASK) >> HANDLE_AGE_SHIFT;
+            auto const pNode = static_cast<typename Allocator::Node*>(p);
+            uint8_t const expectedAge = pNode[-1].age;
+            return expectedAge == age;
+        }
+        return true;
+    }
+
     template<typename Dp, typename B>
     inline typename std::enable_if_t<
             std::is_pointer_v<Dp> &&

filament/backend/src/CircularBuffer.cpp

@@ -32,10 +32,11 @@
 #    define HAS_MMAP 0
 #endif
 
-#include <stdint.h>
 #include <stddef.h>
-#include <stdlib.h>
+#include <stdint.h>
 #include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 using namespace utils;
 
@@ -81,6 +82,9 @@ void* CircularBuffer::alloc(size_t size) noexcept {
             // map the circular buffer once...
             vaddr = mmap(reserve_vaddr, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
             if (vaddr != MAP_FAILED) {
+                // populate the address space with pages (because this is a circular buffer,
+                // all the pages will be allocated eventually, might as well do it now)
+                memset(vaddr, 0, size);
                 // and map the circular buffer again, behind the previous copy...
                 vaddr_shadow = mmap((char*)vaddr + size, size,
                         PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
@@ -101,7 +105,7 @@ void* CircularBuffer::alloc(size_t size) noexcept {
     if (UTILS_UNLIKELY(mAshmemFd < 0)) {
         // ashmem failed
         if (vaddr_guard != MAP_FAILED) {
-            munmap(vaddr_guard, size);
+            munmap(vaddr_guard, BLOCK_SIZE);
         }
 
         if (vaddr_shadow != MAP_FAILED) {
@@ -119,12 +123,11 @@ void* CircularBuffer::alloc(size_t size) noexcept {
         data = mmap(nullptr, size * 2 + BLOCK_SIZE,
                 PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
 
-        FILAMENT_CHECK_POSTCONDITION(data) <<
+        FILAMENT_CHECK_POSTCONDITION(data != MAP_FAILED) <<
                 "couldn't allocate " << (size * 2 / 1024) <<
                 " KiB of virtual address space for the command buffer";
 
-        slog.d << "WARNING: Using soft CircularBuffer (" << (size * 2 / 1024) << " KiB)"
-               << io::endl;
+        slog.w << "Using 'soft' CircularBuffer (" << (size * 2 / 1024) << " KiB)" << io::endl;
 
         // guard page at the end
         void* guard = (void*)(uintptr_t(data) + size * 2);

filament/backend/src/metal/MetalBuffer.h

@@ -204,6 +204,15 @@ public:
 
 private:
 
+    enum class UploadStrategy {
+        POOL,
+        BUMP_ALLOCATOR,
+    };
+
+    void uploadWithPoolBuffer(void* src, size_t size, size_t byteOffset) const;
+    void uploadWithBumpAllocator(void* src, size_t size, size_t byteOffset) const;
+
+    UploadStrategy mUploadStrategy;
     TrackedMetalBuffer mBuffer;
     size_t mBufferSize = 0;
     void* mCpuBuffer = nullptr;

filament/backend/src/metal/MetalBuffer.mm

@@ -27,7 +27,16 @@ MetalPlatform* TrackedMetalBuffer::platform = nullptr;
 MetalPlatform* ScopedAllocationTimer::platform = nullptr;
 
 MetalBuffer::MetalBuffer(MetalContext& context, BufferObjectBinding bindingType, BufferUsage usage,
-        size_t size, bool forceGpuBuffer) : mBufferSize(size), mContext(context) {
+        size_t size, bool forceGpuBuffer)
+    : mBufferSize(size), mContext(context) {
+    const MetalBumpAllocator& allocator = *mContext.bumpAllocator;
+    // VERTEX is also used for index buffers
+    if (allocator.getCapacity() > 0 && bindingType == BufferObjectBinding::VERTEX) {
+        mUploadStrategy = UploadStrategy::BUMP_ALLOCATOR;
+    } else {
+        mUploadStrategy = UploadStrategy::POOL;
+    }
+
     // If the buffer is less than 4K in size and is updated frequently, we don't use an explicit
     // buffer. Instead, we use immediate command encoder methods like setVertexBytes:length:atIndex:.
     // This won't work for SSBOs, since they are read/write.
@@ -61,34 +70,23 @@ void MetalBuffer::copyIntoBuffer(void* src, size_t size, size_t byteOffset) {
     FILAMENT_CHECK_PRECONDITION(size + byteOffset <= mBufferSize)
             << "Attempting to copy " << size << " bytes into a buffer of size " << mBufferSize
             << " at offset " << byteOffset;
+    // The copy blit requires that byteOffset be a multiple of 4.
+    FILAMENT_CHECK_PRECONDITION(!(byteOffset & 0x3)) << "byteOffset must be a multiple of 4";
 
-    // Either copy into the Metal buffer or into our cpu buffer.
+    // If we have a cpu buffer, we can directly copy into it.
     if (mCpuBuffer) {
         memcpy(static_cast<uint8_t*>(mCpuBuffer) + byteOffset, src, size);
         return;
     }
 
-    // Acquire a staging buffer to hold the contents of this update.
-    MetalBufferPool* bufferPool = mContext.bufferPool;
-    const MetalBufferPoolEntry* const staging = bufferPool->acquireBuffer(size);
-    memcpy(staging->buffer.get().contents, src, size);
-
-    // The blit below requires that byteOffset be a multiple of 4.
-    FILAMENT_CHECK_PRECONDITION(!(byteOffset & 0x3u)) << "byteOffset must be a multiple of 4";
-
-    // Encode a blit from the staging buffer into the private GPU buffer.
-    id<MTLCommandBuffer> cmdBuffer = getPendingCommandBuffer(&mContext);
-    id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
-    blitEncoder.label = @"Buffer upload blit";
-    [blitEncoder copyFromBuffer:staging->buffer.get()
-                   sourceOffset:0
-                       toBuffer:mBuffer.get()
-              destinationOffset:byteOffset
-                           size:size];
-    [blitEncoder endEncoding];
-    [cmdBuffer addCompletedHandler:^(id<MTLCommandBuffer> cb) {
-        bufferPool->releaseBuffer(staging);
-    }];
+    switch (mUploadStrategy) {
+        case UploadStrategy::BUMP_ALLOCATOR:
+            uploadWithBumpAllocator(src, size, byteOffset);
+            break;
+        case UploadStrategy::POOL:
+            uploadWithPoolBuffer(src, size, byteOffset);
+            break;
+    }
 }
 
 void MetalBuffer::copyIntoBufferUnsynchronized(void* src, size_t size, size_t byteOffset) {
@@ -199,5 +197,42 @@ void MetalBuffer::bindBuffers(id<MTLCommandBuffer> cmdBuffer, id<MTLCommandEncod
     }
 }
 
+void MetalBuffer::uploadWithPoolBuffer(void* src, size_t size, size_t byteOffset) const {
+    MetalBufferPool* bufferPool = mContext.bufferPool;
+    const MetalBufferPoolEntry* const staging = bufferPool->acquireBuffer(size);
+    memcpy(staging->buffer.get().contents, src, size);
+
+    // Encode a blit from the staging buffer into the private GPU buffer.
+    id<MTLCommandBuffer> cmdBuffer = getPendingCommandBuffer(&mContext);
+    id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
+    blitEncoder.label = @"Buffer upload blit - pool buffer";
+    [blitEncoder copyFromBuffer:staging->buffer.get()
+                   sourceOffset:0
+                       toBuffer:mBuffer.get()
+              destinationOffset:byteOffset
+                           size:size];
+    [blitEncoder endEncoding];
+    [cmdBuffer addCompletedHandler:^(id<MTLCommandBuffer> cb) {
+        bufferPool->releaseBuffer(staging);
+    }];
+}
+
+void MetalBuffer::uploadWithBumpAllocator(void* src, size_t size, size_t byteOffset) const {
+    MetalBumpAllocator& allocator = *mContext.bumpAllocator;
+    auto [buffer, offset] = allocator.allocateStagingArea(size);
+    memcpy(static_cast<char*>(buffer.contents) + offset, src, size);
+
+    // Encode a blit from the staging buffer into the private GPU buffer.
+    id<MTLCommandBuffer> cmdBuffer = getPendingCommandBuffer(&mContext);
+    id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
+    blitEncoder.label = @"Buffer upload blit - bump allocator";
+    [blitEncoder copyFromBuffer:buffer
+                   sourceOffset:offset
+                       toBuffer:mBuffer.get()
+              destinationOffset:byteOffset
+                           size:size];
+    [blitEncoder endEncoding];
+}
+
 } // namespace backend
 } // namespace filament

filament/backend/src/metal/MetalBufferPool.h

@@ -38,6 +38,28 @@ struct MetalBufferPoolEntry {
     mutable uint32_t referenceCount;
 };
 
+class MetalBumpAllocator {
+public:
+    MetalBumpAllocator(id<MTLDevice> device, size_t capacity);
+
+    /**
+     * Allocates a staging area of the given size. Returns a pair of the buffer and the offset
+     * within the buffer. The buffer is guaranteed to be at least the given size, but may be larger.
+     * Clients must not write to the buffer beyond the returned offset + size.
+     * Clients are responsible for holding a reference to the returned buffer.
+     * Allocations are guaranteed to be aligned to 4 bytes.
+     */
+    std::pair<id<MTLBuffer>, size_t> allocateStagingArea(size_t size);
+
+    size_t getCapacity() const noexcept { return mCapacity; }
+
+private:
+    id<MTLDevice> mDevice;
+    TrackedMetalBuffer mCurrentUploadBuffer = nil;
+    size_t mHead = 0;
+    size_t mCapacity;
+};
+
 // Manages a pool of Metal buffers, periodically releasing ones that have been unused for awhile.
 class MetalBufferPool {
 public:

filament/backend/src/metal/MetalBufferPool.mm

@@ -116,5 +116,39 @@ void MetalBufferPool::reset() noexcept {
     mFreeStages.clear();
 }
 
+MetalBumpAllocator::MetalBumpAllocator(id<MTLDevice> device, size_t capacity)
+    : mDevice(device), mCapacity(capacity) {
+    if (mCapacity > 0) {
+        mCurrentUploadBuffer = { [device newBufferWithLength:capacity options:MTLStorageModeShared],
+            TrackedMetalBuffer::Type::STAGING };
+    }
+}
+
+std::pair<id<MTLBuffer>, size_t> MetalBumpAllocator::allocateStagingArea(size_t size) {
+    if (size == 0) {
+        return { nil, 0 };
+    }
+    if (size > mCapacity) {
+        return { [mDevice newBufferWithLength:size options:MTLStorageModeShared], 0 };
+    }
+    assert_invariant(mCurrentUploadBuffer);
+
+    // Align the head to a 4-byte boundary.
+    mHead = (mHead + 3) & ~3;
+
+    if (UTILS_LIKELY(mHead + size <= mCapacity)) {
+        const size_t oldHead = mHead;
+        mHead += size;
+        return { mCurrentUploadBuffer.get(), oldHead };
+    }
+
+    // We're finished with the current allocation.
+    mCurrentUploadBuffer = { [mDevice newBufferWithLength:mCapacity options:MTLStorageModeShared],
+        TrackedMetalBuffer::Type::STAGING };
+    mHead = size;
+
+    return { mCurrentUploadBuffer.get(), 0 };
+}
+
 } // namespace backend
 } // namespace filament

filament/backend/src/metal/MetalContext.h

@@ -44,6 +44,7 @@ namespace backend {
 class MetalDriver;
 class MetalBlitter;
 class MetalBufferPool;
+class MetalBumpAllocator;
 class MetalRenderTarget;
 class MetalSamplerGroup;
 class MetalSwapChain;
@@ -141,6 +142,7 @@ struct MetalContext {
     utils::FixedCircularBuffer<Handle<HwTexture>> texturesToDestroy;
 
     MetalBufferPool* bufferPool;
+    MetalBumpAllocator* bumpAllocator;
 
     MetalSwapChain* currentDrawSwapChain = nil;
     MetalSwapChain* currentReadSwapChain = nil;

filament/backend/src/metal/MetalDriver.mm

@@ -171,6 +171,8 @@ MetalDriver::MetalDriver(MetalPlatform* platform, const Platform::DriverConfig&
     mContext->samplerStateCache.setDevice(mContext->device);
     mContext->argumentEncoderCache.setDevice(mContext->device);
     mContext->bufferPool = new MetalBufferPool(*mContext);
+    mContext->bumpAllocator =
+            new MetalBumpAllocator(mContext->device, driverConfig.metalUploadBufferSizeBytes);
     mContext->blitter = new MetalBlitter(*mContext);
 
     if (@available(iOS 12, *)) {
@@ -209,6 +211,7 @@ MetalDriver::~MetalDriver() noexcept {
     mContext->emptyTexture = nil;
     CFRelease(mContext->textureCache);
     delete mContext->bufferPool;
+    delete mContext->bumpAllocator;
     delete mContext->blitter;
     delete mContext->timerQueryImpl;
     delete mContext->shaderCompiler;

filament/backend/src/metal/MetalHandles.h

@@ -109,7 +109,7 @@ private:
     NSUInteger headlessWidth = 0;
     NSUInteger headlessHeight = 0;
     CAMetalLayer* layer = nullptr;
-    std::mutex layerDrawableMutex;
+    std::shared_ptr<std::mutex> layerDrawableMutex;
     MetalExternalImage externalImage;
     SwapChainType type;
 

filament/backend/src/metal/MetalHandles.mm

@@ -73,6 +73,7 @@ MetalSwapChain::MetalSwapChain(MetalContext& context, CAMetalLayer* nativeWindow
     : context(context),
       depthStencilFormat(decideDepthStencilFormat(flags)),
       layer(nativeWindow),
+      layerDrawableMutex(std::make_shared<std::mutex>()),
       externalImage(context),
       type(SwapChainType::CAMETALLAYER) {
 
@@ -179,7 +180,7 @@ id<MTLTexture> MetalSwapChain::acquireDrawable() {
     // calling -nextDrawable, or when releasing the last known reference
     // to any CAMetalDrawable returned from a previous -nextDrawable.
     {
-        std::lock_guard<std::mutex> lock(layerDrawableMutex);
+        std::lock_guard<std::mutex> lock(*layerDrawableMutex);
         drawable = [layer nextDrawable];
     }
 
@@ -188,8 +189,10 @@ id<MTLTexture> MetalSwapChain::acquireDrawable() {
 }
 
 void MetalSwapChain::releaseDrawable() {
-    std::lock_guard<std::mutex> lock(layerDrawableMutex);
-    drawable = nil;
+    if (drawable) {
+        std::lock_guard<std::mutex> lock(*layerDrawableMutex);
+        drawable = nil;
+    }
 }
 
 id<MTLTexture> MetalSwapChain::acquireDepthTexture() {
@@ -265,7 +268,7 @@ public:
     PresentDrawableData& operator=(const PresentDrawableData&) = delete;
 
     static PresentDrawableData* create(id<CAMetalDrawable> drawable,
-            std::mutex* drawableMutex, MetalDriver* driver) {
+            std::shared_ptr<std::mutex> drawableMutex, MetalDriver* driver) {
         assert_invariant(drawableMutex);
         assert_invariant(driver);
         return new PresentDrawableData(drawable, drawableMutex, driver);
@@ -287,22 +290,22 @@ public:
     }
 
 private:
-    PresentDrawableData(id<CAMetalDrawable> drawable, std::mutex* drawableMutex,
+    PresentDrawableData(id<CAMetalDrawable> drawable, std::shared_ptr<std::mutex> drawableMutex,
             MetalDriver* driver)
         : mDrawable(drawable), mDrawableMutex(drawableMutex), mDriver(driver) {}
 
     static void cleanupAndDestroy(PresentDrawableData *that) {
-        {
+        if (that->mDrawable) {
             std::lock_guard<std::mutex> lock(*(that->mDrawableMutex));
             that->mDrawable = nil;
         }
-        that->mDrawableMutex = nullptr;
+        that->mDrawableMutex.reset();
         that->mDriver = nullptr;
         delete that;
     }
 
     id<CAMetalDrawable> mDrawable;
-    std::mutex* mDrawableMutex = nullptr;
+    std::shared_ptr<std::mutex> mDrawableMutex;
     MetalDriver* mDriver = nullptr;
 };
 
@@ -320,7 +323,7 @@ void MetalSwapChain::scheduleFrameScheduledCallback() {
 
     struct Callback {
         Callback(std::shared_ptr<FrameScheduledCallback> callback, id<CAMetalDrawable> drawable,
-                std::mutex* drawableMutex, MetalDriver* driver)
+                 std::shared_ptr<std::mutex> drawableMutex, MetalDriver* driver)
             : f(callback), data(PresentDrawableData::create(drawable, drawableMutex, driver)) {}
         std::shared_ptr<FrameScheduledCallback> f;
         // PresentDrawableData* is destroyed by maybePresentAndDestroyAsync() later.
@@ -337,7 +340,7 @@ void MetalSwapChain::scheduleFrameScheduledCallback() {
     // This callback pointer will be captured by the block. Even if the scheduled handler is never
     // called, the unique_ptr will still ensure we don't leak memory.
     __block auto callback = std::make_unique<Callback>(
-        frameScheduled.callback, drawable, &layerDrawableMutex, context.driver);
+        frameScheduled.callback, drawable, layerDrawableMutex, context.driver);
 
     backend::CallbackHandler* handler = frameScheduled.handler;
     MetalDriver* driver = context.driver;
@@ -553,8 +556,6 @@ MetalTexture::MetalTexture(MetalContext& context, SamplerType target, uint8_t le
             descriptor.usage = getMetalTextureUsage(usage);
             descriptor.storageMode = MTLStorageModePrivate;
             texture = [context.device newTextureWithDescriptor:descriptor];
-            FILAMENT_CHECK_POSTCONDITION(texture != nil)
-                    << "Could not create Metal texture. Out of memory?";
             break;
         case SamplerType::SAMPLER_CUBEMAP:
         case SamplerType::SAMPLER_CUBEMAP_ARRAY:
@@ -569,8 +570,6 @@ MetalTexture::MetalTexture(MetalContext& context, SamplerType target, uint8_t le
             descriptor.usage = getMetalTextureUsage(usage);
             descriptor.storageMode = MTLStorageModePrivate;
             texture = [context.device newTextureWithDescriptor:descriptor];
-            FILAMENT_CHECK_POSTCONDITION(texture != nil)
-                    << "Could not create Metal texture. Out of memory?";
             break;
         case SamplerType::SAMPLER_3D:
             descriptor = [MTLTextureDescriptor new];
@@ -583,8 +582,6 @@ MetalTexture::MetalTexture(MetalContext& context, SamplerType target, uint8_t le
             descriptor.usage = getMetalTextureUsage(usage);
             descriptor.storageMode = MTLStorageModePrivate;
             texture = [context.device newTextureWithDescriptor:descriptor];
-            FILAMENT_CHECK_POSTCONDITION(texture != nil)
-                    << "Could not create Metal texture. Out of memory?";
             break;
         case SamplerType::SAMPLER_EXTERNAL:
             // If we're using external textures (CVPixelBufferRefs), we don't need to make any
@@ -593,6 +590,12 @@ MetalTexture::MetalTexture(MetalContext& context, SamplerType target, uint8_t le
             break;
     }
 
+    FILAMENT_CHECK_POSTCONDITION(target == SamplerType::SAMPLER_EXTERNAL || texture != nil)
+            << "Could not create Metal texture (SamplerType = " << int(target)
+            << ", levels = " << int(levels) << ", MTLPixelFormat = " << int(devicePixelFormat)
+            << ", width = " << width << ", height = " << height << ", depth = " << depth
+            << "). Out of memory?";
+
     // If swizzling is set, set up a swizzled texture view that we'll use when sampling this texture.
     const bool isDefaultSwizzle =
             r == TextureSwizzle::CHANNEL_0 &&

filament/backend/src/opengl/OpenGLDriver.cpp

@@ -304,6 +304,13 @@ void OpenGLDriver::bindSampler(GLuint unit, GLuint sampler) noexcept {
 void OpenGLDriver::setPushConstant(backend::ShaderStage stage, uint8_t index,
         backend::PushConstantVariant value) {
     assert_invariant(stage == ShaderStage::VERTEX || stage == ShaderStage::FRAGMENT);
+
+#if FILAMENT_ENABLE_MATDBG
+    if (UTILS_UNLIKELY(!mValidProgram)) {
+        return;
+    }
+#endif
+
     utils::Slice<std::pair<GLint, ConstantType>> constants;
     if (stage == ShaderStage::VERTEX) {
         constants = mCurrentPushConstants->vertexConstants;
@@ -340,15 +347,11 @@ void OpenGLDriver::bindTexture(GLuint unit, GLTexture const* t) noexcept {
 }
 
 bool OpenGLDriver::useProgram(OpenGLProgram* p) noexcept {
-    if (UTILS_UNLIKELY(!p->isValid())) {
-        // If the program is not valid, we can't call use().
-        return false;
-    }
-
     // set-up textures and samplers in the proper TMUs (as specified in setSamplers)
-    p->use(this, mContext);
+    bool const success = p->use(this, mContext);
+    assert_invariant(success == p->isValid());
 
-    if (UTILS_UNLIKELY(mContext.isES2())) {
+    if (UTILS_UNLIKELY(mContext.isES2() && success)) {
         for (uint32_t i = 0; i < Program::UNIFORM_BINDING_COUNT; i++) {
             auto [id, buffer, age] = mContext.getEs2UniformBinding(i);
             if (buffer) {
@@ -359,7 +362,8 @@ bool OpenGLDriver::useProgram(OpenGLProgram* p) noexcept {
         // when mPlatform.isSRGBSwapChainSupported() is false (no need to check though).
         p->setRec709ColorSpace(mRec709OutputColorspace);
     }
-    return true;
+
+    return success;
 }
 
 
@@ -2332,9 +2336,9 @@ void OpenGLDriver::updateSamplerGroup(Handle<HwSamplerGroup> sbh,
     auto const* const pSamplers = (SamplerDescriptor const*)data.buffer;
     for (size_t i = 0, c = sb->textureUnitEntries.size(); i < c; i++) {
         GLuint samplerId = 0u;
-        const GLTexture* t = nullptr;
-        if (UTILS_LIKELY(pSamplers[i].t)) {
-            t = handle_cast<const GLTexture*>(pSamplers[i].t);
+        Handle<HwTexture> th = pSamplers[i].t;
+        if (UTILS_LIKELY(th)) {
+            GLTexture const* const t = handle_cast<const GLTexture*>(th);
             assert_invariant(t);
 
             SamplerParams params = pSamplers[i].s;
@@ -2390,7 +2394,7 @@ void OpenGLDriver::updateSamplerGroup(Handle<HwSamplerGroup> sbh,
             // which is not an error.
         }
 
-        sb->textureUnitEntries[i] = { t, samplerId };
+        sb->textureUnitEntries[i] = { th, samplerId };
     }
     scheduleDestroy(std::move(data));
 }

filament/backend/src/opengl/OpenGLDriver.h

@@ -126,7 +126,7 @@ public:
     struct GLSamplerGroup : public HwSamplerGroup {
         using HwSamplerGroup::HwSamplerGroup;
         struct Entry {
-            GLTexture const* texture = nullptr;
+            Handle<HwTexture> th;
             GLuint sampler = 0u;
         };
         utils::FixedCapacityVector<Entry> textureUnitEntries;
@@ -256,6 +256,11 @@ private:
         return mHandleAllocator.handle_cast<Dp, B>(handle);
     }
 
+    template<typename B>
+    bool is_valid(Handle<B>& handle) {
+        return mHandleAllocator.is_valid(handle);
+    }
+
     template<typename Dp, typename B>
     inline typename std::enable_if_t<
             std::is_pointer_v<Dp> &&

filament/backend/src/opengl/OpenGLProgram.cpp

@@ -20,21 +20,24 @@
 #include "OpenGLDriver.h"
 #include "ShaderCompilerService.h"
 
+#include <backend/DriverEnums.h>
 #include <backend/Program.h>
+#include <backend/Handle.h>
 
-#include <private/backend/BackendUtils.h>
-
-#include <utils/debug.h>
 #include <utils/compiler.h>
+#include <utils/debug.h>
+#include <utils/FixedCapacityVector.h>
 #include <utils/Log.h>
 #include <utils/Systrace.h>
 
+#include <algorithm>
 #include <array>
 #include <string_view>
 #include <utility>
 #include <new>
 
 #include <stddef.h>
+#include <stdint.h>
 
 namespace filament::backend {
 
@@ -241,8 +244,9 @@ void OpenGLProgram::updateSamplers(OpenGLDriver* const gld) const noexcept {
         assert_invariant(sb);
         if (!sb) continue; // should never happen, this would be a user error.
         for (uint8_t j = 0, m = sb->textureUnitEntries.size(); j < m; ++j, ++tmu) { // "<=" on purpose here
-            const GLTexture* const t = sb->textureUnitEntries[j].texture;
-            if (t) { // program may not use all samplers of sampler group
+            Handle<HwTexture> th = sb->textureUnitEntries[j].th;
+            if (th) { // program may not use all samplers of sampler group
+                GLTexture const* const t = gld->handle_cast<GLTexture const*>(th);
                 gld->bindTexture(tmu, t);
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
                 if (UTILS_LIKELY(!es2)) {

filament/backend/src/opengl/OpenGLProgram.h

@@ -53,11 +53,21 @@ public:
 
     bool isValid() const noexcept { return mToken || gl.program != 0; }
 
-    void use(OpenGLDriver* const gld, OpenGLContext& context) noexcept {
-        if (UTILS_UNLIKELY(!gl.program)) {
+    bool use(OpenGLDriver* const gld, OpenGLContext& context) noexcept {
+        // both non-null is impossible by construction
+        assert_invariant(!mToken || !gl.program);
+
+        if (UTILS_UNLIKELY(mToken && !gl.program)) {
+            // first time a program is used
             initialize(*gld);
         }
 
+        if (UTILS_UNLIKELY(!gl.program)) {
+            // compilation failed (token should be null)
+            assert_invariant(!mToken);
+            return false;
+        }
+
         context.useProgram(gl.program);
         if (UTILS_UNLIKELY(mUsedBindingsCount)) {
             // We rely on GL state tracking to avoid unnecessary glBindTexture / glBindSampler
@@ -74,6 +84,7 @@ public:
 
             updateSamplers(gld);
         }
+        return true;
     }
 
     // For ES2 only

filament/backend/src/opengl/ShaderCompilerService.cpp

@@ -359,7 +359,7 @@ ShaderCompilerService::program_token_t ShaderCompilerService::createProgram(
 GLuint ShaderCompilerService::getProgram(ShaderCompilerService::program_token_t& token) {
     GLuint const program = initialize(token);
     assert_invariant(token == nullptr);
-#ifndef FILAMENT_ENABLE_MATDBG
+#if !FILAMENT_ENABLE_MATDBG
     assert_invariant(program);
 #endif
     return program;
@@ -572,16 +572,23 @@ void ShaderCompilerService::compileShaders(OpenGLContext& context,
 
             // split shader source, so we can insert the specialization constants and the packing
             // functions
-            auto const [prolog, body] = splitShaderSource({ shader_src, shader_len });
+            auto [version, prolog, body] = splitShaderSource({ shader_src, shader_len });
 
-            const std::array<const char*, 4> sources = {
+            // enable ESSL 3.10 if available
+            if (context.isAtLeastGLES<3, 1>()) {
+                version = "#version 310 es\n";
+            }
+
+            const std::array<const char*, 5> sources = {
+                    version.data(),
                     prolog.data(),
                     specializationConstantString.c_str(),
                     packingFunctions.data(),
                     body.data()
             };
 
-            const std::array<GLint, 4> lengths = {
+            const std::array<GLint, 5> lengths = {
+                    (GLint)version.length(),
                     (GLint)prolog.length(),
                     (GLint)specializationConstantString.length(),
                     (GLint)packingFunctions.length(),
@@ -661,6 +668,7 @@ void ShaderCompilerService::process_OVR_multiview2(OpenGLContext& context,
 
 // Tragically, OpenGL 4.1 doesn't support unpackHalf2x16 (appeared in 4.2) and
 // macOS doesn't support GL_ARB_shading_language_packing
+// Also GLES3.0 didn't have the full set of packing/unpacking functions
 std::string_view ShaderCompilerService::process_ARB_shading_language_packing(OpenGLContext& context) noexcept {
     using namespace std::literals;
 #ifdef BACKEND_OPENGL_VERSION_GL
@@ -700,31 +708,102 @@ highp uint packHalf2x16(vec2 v) {
     highp uint y = fp32tou16(v.y);
     return (y << 16u) | x;
 }
+highp uint packUnorm4x8(mediump vec4 v) {
+    v = round(clamp(v, 0.0, 1.0) * 255.0);
+    highp uint a = uint(v.x);
+    highp uint b = uint(v.y) <<  8;
+    highp uint c = uint(v.z) << 16;
+    highp uint d = uint(v.w) << 24;
+    return (a|b|c|d);
+}
+highp uint packSnorm4x8(mediump vec4 v) {
+    v = round(clamp(v, -1.0, 1.0) * 127.0);
+    highp uint a = uint((int(v.x) & 0xff));
+    highp uint b = uint((int(v.y) & 0xff)) <<  8;
+    highp uint c = uint((int(v.z) & 0xff)) << 16;
+    highp uint d = uint((int(v.w) & 0xff)) << 24;
+    return (a|b|c|d);
+}
+mediump vec4 unpackUnorm4x8(highp uint v) {
+    return vec4(float((v & 0x000000ffu)      ),
+                float((v & 0x0000ff00u) >>  8),
+                float((v & 0x00ff0000u) >> 16),
+                float((v & 0xff000000u) >> 24)) / 255.0;
+}
+mediump vec4 unpackSnorm4x8(highp uint v) {
+    int a = int(((v       ) & 0xffu) << 24u) >> 24 ;
+    int b = int(((v >>  8u) & 0xffu) << 24u) >> 24 ;
+    int c = int(((v >> 16u) & 0xffu) << 24u) >> 24 ;
+    int d = int(((v >> 24u) & 0xffu) << 24u) >> 24 ;
+    return clamp(vec4(float(a), float(b), float(c), float(d)) / 127.0, -1.0, 1.0);
+}
 )"sv;
     }
 #endif // BACKEND_OPENGL_VERSION_GL
+
+#ifdef BACKEND_OPENGL_VERSION_GLES
+    if (!context.isES2() && !context.isAtLeastGLES<3, 1>()) {
+        return R"(
+
+highp uint packUnorm4x8(mediump vec4 v) {
+    v = round(clamp(v, 0.0, 1.0) * 255.0);
+    highp uint a = uint(v.x);
+    highp uint b = uint(v.y) <<  8;
+    highp uint c = uint(v.z) << 16;
+    highp uint d = uint(v.w) << 24;
+    return (a|b|c|d);
+}
+highp uint packSnorm4x8(mediump vec4 v) {
+    v = round(clamp(v, -1.0, 1.0) * 127.0);
+    highp uint a = uint((int(v.x) & 0xff));
+    highp uint b = uint((int(v.y) & 0xff)) <<  8;
+    highp uint c = uint((int(v.z) & 0xff)) << 16;
+    highp uint d = uint((int(v.w) & 0xff)) << 24;
+    return (a|b|c|d);
+}
+mediump vec4 unpackUnorm4x8(highp uint v) {
+    return vec4(float((v & 0x000000ffu)      ),
+                float((v & 0x0000ff00u) >>  8),
+                float((v & 0x00ff0000u) >> 16),
+                float((v & 0xff000000u) >> 24)) / 255.0;
+}
+mediump vec4 unpackSnorm4x8(highp uint v) {
+    int a = int(((v       ) & 0xffu) << 24u) >> 24 ;
+    int b = int(((v >>  8u) & 0xffu) << 24u) >> 24 ;
+    int c = int(((v >> 16u) & 0xffu) << 24u) >> 24 ;
+    int d = int(((v >> 24u) & 0xffu) << 24u) >> 24 ;
+    return clamp(vec4(float(a), float(b), float(c), float(d)) / 127.0, -1.0, 1.0);
+}
+)"sv;
+    }
+#endif // BACKEND_OPENGL_VERSION_GLES
     return ""sv;
 }
 
-// split shader source code in two, the first section goes from the start to the line after the
-// last #extension, and the 2nd part goes from there to the end.
-std::array<std::string_view, 2> ShaderCompilerService::splitShaderSource(std::string_view source) noexcept {
-    auto start = source.find("#version");
-    assert_invariant(start != std::string_view::npos);
+// split shader source code in three:
+// - the version line
+// - extensions
+// - everything else
+std::array<std::string_view, 3> ShaderCompilerService::splitShaderSource(std::string_view source) noexcept {
+    auto version_start = source.find("#version");
+    assert_invariant(version_start != std::string_view::npos);
 
-    auto pos = source.rfind("\n#extension");
-    if (pos == std::string_view::npos) {
-        pos = start;
+    auto version_eol = source.find('\n', version_start) + 1;
+    assert_invariant(version_eol != std::string_view::npos);
+
+    auto prolog_start = version_eol;
+    auto prolog_eol = source.rfind("\n#extension"); // last #extension line
+    if (prolog_eol == std::string_view::npos) {
+        prolog_eol = prolog_start;
     } else {
-        ++pos;
+        prolog_eol = source.find('\n', prolog_eol + 1) + 1;
     }
+    auto body_start = prolog_eol;
 
-    auto eol = source.find('\n', pos) + 1;
-    assert_invariant(eol != std::string_view::npos);
-
-    std::string_view const version = source.substr(start, eol - start);
-    std::string_view const body = source.substr(version.length(), source.length() - version.length());
-    return { version, body };
+    std::string_view const version = source.substr(version_start, version_eol - version_start);
+    std::string_view const prolog = source.substr(prolog_start, prolog_eol - prolog_start);
+    std::string_view const body = source.substr(body_start, source.length() - body_start);
+    return { version, prolog, body };
 }
 
 /*

filament/backend/src/opengl/ShaderCompilerService.h

@@ -146,7 +146,7 @@ private:
 
     static std::string_view process_ARB_shading_language_packing(OpenGLContext& context) noexcept;
 
-    static std::array<std::string_view, 2> splitShaderSource(std::string_view source) noexcept;
+    static std::array<std::string_view, 3> splitShaderSource(std::string_view source) noexcept;
 
     static GLuint linkProgram(OpenGLContext& context,
             std::array<GLuint, Program::SHADER_TYPE_COUNT> shaders,

filament/backend/src/vulkan/VulkanBlitter.cpp

@@ -37,7 +37,7 @@ inline void blitFast(const VkCommandBuffer cmdbuffer, VkImageAspectFlags aspect,
         VulkanAttachment src, VulkanAttachment dst,
         const VkOffset3D srcRect[2], const VkOffset3D dstRect[2]) {
     if constexpr (FVK_ENABLED(FVK_DEBUG_BLITTER)) {
-        utils::slog.d << "Fast blit from=" << src.texture->getVkImage() << ",level=" << (int) src.level
+        FVK_LOGD << "Fast blit from=" << src.texture->getVkImage() << ",level=" << (int) src.level
                       << " layout=" << src.getLayout()
                       << " to=" << dst.texture->getVkImage() << ",level=" << (int) dst.level
                       << " layout=" << dst.getLayout() << utils::io::endl;
@@ -76,7 +76,7 @@ inline void blitFast(const VkCommandBuffer cmdbuffer, VkImageAspectFlags aspect,
 inline void resolveFast(const VkCommandBuffer cmdbuffer, VkImageAspectFlags aspect,
         VulkanAttachment src, VulkanAttachment dst) {
     if constexpr (FVK_ENABLED(FVK_DEBUG_BLITTER)) {
-        utils::slog.d << "Fast blit from=" << src.texture->getVkImage() << ",level=" << (int) src.level
+        FVK_LOGD << "Fast blit from=" << src.texture->getVkImage() << ",level=" << (int) src.level
                       << " layout=" << src.getLayout()
                       << " to=" << dst.texture->getVkImage() << ",level=" << (int) dst.level
                       << " layout=" << dst.getLayout() << utils::io::endl;

filament/backend/src/vulkan/VulkanBuffer.cpp

@@ -28,6 +28,7 @@ VulkanBuffer::VulkanBuffer(VmaAllocator allocator, VulkanStagePool& stagePool,
     : mAllocator(allocator),
       mStagePool(stagePool),
       mUsage(usage),
+	  mUpdatedOffset(0),
       mUpdatedBytes(0) {
     // for now make sure that only 1 bit is set in usage
     // (because loadFromCpu() assumes that somewhat)
@@ -80,6 +81,7 @@ void VulkanBuffer::loadFromCpu(VkCommandBuffer cmdbuf, const void* cpuData, uint
             .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
             .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
             .buffer = mGpuBuffer,
+			.offset = mUpdatedOffset,
             .size = mUpdatedBytes,
         };
         vkCmdPipelineBarrier(cmdbuf, srcStage, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 1,
@@ -93,6 +95,7 @@ void VulkanBuffer::loadFromCpu(VkCommandBuffer cmdbuf, const void* cpuData, uint
     };
     vkCmdCopyBuffer(cmdbuf, stage->buffer, mGpuBuffer, 1, &region);
 
+	mUpdatedOffset = byteOffset;
     mUpdatedBytes = numBytes;
 
     // Firstly, ensure that the copy finishes before the next draw call.

filament/backend/src/vulkan/VulkanBuffer.h

@@ -42,6 +42,7 @@ private:
     VmaAllocation mGpuMemory = VK_NULL_HANDLE;
     VkBuffer mGpuBuffer = VK_NULL_HANDLE;
     VkBufferUsageFlags mUsage = {};
+	uint32_t mUpdatedOffset = 0;
     uint32_t mUpdatedBytes = 0;
 };
 

filament/backend/src/vulkan/VulkanCommands.cpp

@@ -178,7 +178,7 @@ VulkanCommandBuffer& VulkanCommands::get() {
     // presenting the swap chain or waiting on a fence.
     while (mAvailableBufferCount == 0) {
 #if FVK_ENABLED(FVK_DEBUG_COMMAND_BUFFER)
-        slog.i << "VulkanCommands has stalled. "
+        FVK_LOGI << "VulkanCommands has stalled. "
                << "If this occurs frequently, consider increasing VK_MAX_COMMAND_BUFFERS."
                << io::endl;
 #endif
@@ -289,7 +289,7 @@ bool VulkanCommands::flush() {
     };
 
 #if FVK_ENABLED(FVK_DEBUG_COMMAND_BUFFER)
-    slog.i << "Submitting cmdbuffer=" << cmdbuffer
+    FVK_LOGI << "Submitting cmdbuffer=" << cmdbuffer
            << " wait=(" << signals[0] << ", " << signals[1] << ") "
            << " signal=" << renderingFinished
            << " fence=" << currentbuf->fence->fence
@@ -305,7 +305,7 @@ bool VulkanCommands::flush() {
 
 #if FVK_ENABLED(FVK_DEBUG_COMMAND_BUFFER)
     if (result != VK_SUCCESS) {
-        utils::slog.d << "Failed command buffer submission result: " << result << utils::io::endl;
+        FVK_LOGD << "Failed command buffer submission result: " << result << utils::io::endl;
     }
 #endif
     assert_invariant(result == VK_SUCCESS);
@@ -320,7 +320,7 @@ VkSemaphore VulkanCommands::acquireFinishedSignal() {
     VkSemaphore semaphore = mSubmissionSignal;
     mSubmissionSignal = VK_NULL_HANDLE;
 #if FVK_ENABLED(FVK_DEBUG_COMMAND_BUFFER)
-    slog.i << "Acquiring " << semaphore << " (e.g. for vkQueuePresentKHR)" << io::endl;
+    FVK_LOGI << "Acquiring " << semaphore << " (e.g. for vkQueuePresentKHR)" << io::endl;
 #endif
     return semaphore;
 }
@@ -329,7 +329,7 @@ void VulkanCommands::injectDependency(VkSemaphore next) {
     assert_invariant(mInjectedSignal == VK_NULL_HANDLE);
     mInjectedSignal = next;
 #if FVK_ENABLED(FVK_DEBUG_COMMAND_BUFFER)
-    slog.i << "Injecting " << next << " (e.g. due to vkAcquireNextImageKHR)" << io::endl;
+    FVK_LOGI << "Injecting " << next << " (e.g. due to vkAcquireNextImageKHR)" << io::endl;
 #endif
 }
 
@@ -398,7 +398,7 @@ void VulkanCommands::pushGroupMarker(char const* str, VulkanGroupMarkers::Timest
     // If the timestamp is not 0, then we are carrying over a marker across buffer submits.
     // If it is 0, then this is a normal marker push and we should just print debug line as usual.
     if (timestamp.time_since_epoch().count() == 0.0) {
-        utils::slog.d << "----> " << str << utils::io::endl;
+        FVK_LOGD << "----> " << str << utils::io::endl;
     }
 #endif
 
@@ -436,7 +436,7 @@ void VulkanCommands::popGroupMarker() {
         auto const [marker, startTime] = mGroupMarkers->pop();
         auto const endTime = std::chrono::high_resolution_clock::now();
         std::chrono::duration<double> diff = endTime - startTime;
-        utils::slog.d << "<---- " << marker << " elapsed: " << (diff.count() * 1000) << " ms"
+        FVK_LOGD << "<---- " << marker << " elapsed: " << (diff.count() * 1000) << " ms"
                       << utils::io::endl;
 #else
         mGroupMarkers->pop();

filament/backend/src/vulkan/VulkanConstants.h

@@ -17,6 +17,8 @@
 #ifndef TNT_FILAMENT_BACKEND_VULKANCONSTANTS_H
 #define TNT_FILAMENT_BACKEND_VULKANCONSTANTS_H
 
+#include <utils/Log.h>
+
 #include <stdint.h>
 
 // In debug builds, we enable validation layers and set up a debug callback.
@@ -70,6 +72,11 @@
 // the currently active resources.
 #define FVK_DEBUG_RESOURCE_LEAK           0x00010000
 
+// Set this to enable logging "only" to one output stream. This is useful in the case where we want
+// to debug with print statements and want ordered logging (e.g slog.i and slog.e will not appear in
+// order of calls).
+#define FVK_DEBUG_FORCE_LOG_TO_I          0x00020000
+
 // Useful default combinations
 #define FVK_DEBUG_EVERYTHING              0xFFFFFFFF
 #define FVK_DEBUG_PERFORMANCE     \
@@ -133,6 +140,18 @@ static_assert(FVK_ENABLED(FVK_DEBUG_VALIDATION));
 #define FVK_HANDLE_ARENA_SIZE_IN_MB 8
 #endif
 
+#if FVK_ENABLED(FVK_DEBUG_FORCE_LOG_TO_I)
+    #define FVK_LOGI (utils::slog.i)
+    #define FVK_LOGD FVK_LOGI
+    #define FVK_LOGE FVK_LOGI
+    #define FVK_LOGW FVK_LOGI
+#else
+    #define FVK_LOGE (utils::slog.e)
+    #define FVK_LOGW (utils::slog.w)
+    #define FVK_LOGD (utils::slog.d)
+    #define FVK_LOGI (utils::slog.i)
+#endif
+
 // All vkCreate* functions take an optional allocator. For now we select the default allocator by
 // passing in a null pointer, and we highlight the argument by using the VKALLOC constant.
 constexpr struct VkAllocationCallbacks* VKALLOC = nullptr;

filament/backend/src/vulkan/VulkanContext.cpp

@@ -100,7 +100,7 @@ std::tuple<uint32_t, uint32_t> VulkanTimestamps::getNextQuery() {
 	    return std::make_tuple(timerIndex * 2, timerIndex * 2 + 1);
         }
     }
-    utils::slog.e << "More than " << maxTimers << " timers are not supported." << utils::io::endl;
+    FVK_LOGE << "More than " << maxTimers << " timers are not supported." << utils::io::endl;
     return std::make_tuple(0, 1);
 }
 

filament/backend/src/vulkan/VulkanContext.h

@@ -131,6 +131,10 @@ public:
         return mDebugUtilsSupported;
     }
 
+    inline bool isMultiviewEnabled() const noexcept {
+        return mMultiviewEnabled;
+    }
+
     inline bool isClipDistanceSupported() const noexcept {
         return mPhysicalDeviceFeatures.shaderClipDistance == VK_TRUE;
     }
@@ -141,6 +145,7 @@ private:
     VkPhysicalDeviceFeatures mPhysicalDeviceFeatures = {};
     bool mDebugMarkersSupported = false;
     bool mDebugUtilsSupported = false;
+    bool mMultiviewEnabled = false;
 
     VkFormatList mDepthStencilFormats;
     VkFormatList mBlittableDepthStencilFormats;

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -115,15 +115,15 @@ VKAPI_ATTR VkBool32 VKAPI_CALL debugReportCallback(VkDebugReportFlagsEXT flags,
         VkDebugReportObjectTypeEXT objectType, uint64_t object, size_t location,
         int32_t messageCode, const char* pLayerPrefix, const char* pMessage, void* pUserData) {
     if (flags & VK_DEBUG_REPORT_ERROR_BIT_EXT) {
-        utils::slog.e << "VULKAN ERROR: (" << pLayerPrefix << ") " << pMessage << utils::io::endl;
+        FVK_LOGE << "VULKAN ERROR: (" << pLayerPrefix << ") " << pMessage << utils::io::endl;
     } else {
         // TODO: emit best practices warnings about aggressive pipeline barriers.
         if (strstr(pMessage, "ALL_GRAPHICS_BIT") || strstr(pMessage, "ALL_COMMANDS_BIT")) {
             return VK_FALSE;
         }
-        utils::slog.w << "VULKAN WARNING: (" << pLayerPrefix << ") " << pMessage << utils::io::endl;
+        FVK_LOGW << "VULKAN WARNING: (" << pLayerPrefix << ") " << pMessage << utils::io::endl;
     }
-    utils::slog.e << utils::io::endl;
+    FVK_LOGE << utils::io::endl;
     return VK_FALSE;
 }
 #endif // FVK_EANBLED(FVK_DEBUG_VALIDATION)
@@ -133,18 +133,18 @@ VKAPI_ATTR VkBool32 VKAPI_CALL debugUtilsCallback(VkDebugUtilsMessageSeverityFla
         VkDebugUtilsMessageTypeFlagsEXT types, const VkDebugUtilsMessengerCallbackDataEXT* cbdata,
         void* pUserData) {
     if (severity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT) {
-        utils::slog.e << "VULKAN ERROR: (" << cbdata->pMessageIdName << ") " << cbdata->pMessage
-                      << utils::io::endl;
+        FVK_LOGE << "VULKAN ERROR: (" << cbdata->pMessageIdName << ") " << cbdata->pMessage
+                 << utils::io::endl;
     } else {
         // TODO: emit best practices warnings about aggressive pipeline barriers.
         if (strstr(cbdata->pMessage, "ALL_GRAPHICS_BIT")
                 || strstr(cbdata->pMessage, "ALL_COMMANDS_BIT")) {
             return VK_FALSE;
         }
-        utils::slog.w << "VULKAN WARNING: (" << cbdata->pMessageIdName << ") " << cbdata->pMessage
-                      << utils::io::endl;
+        FVK_LOGW << "VULKAN WARNING: (" << cbdata->pMessageIdName << ") " << cbdata->pMessage
+                 << utils::io::endl;
     }
-    utils::slog.e << utils::io::endl;
+    FVK_LOGE << utils::io::endl;
     return VK_FALSE;
 }
 #endif // FVK_EANBLED(FVK_DEBUG_DEBUG_UTILS)
@@ -292,7 +292,7 @@ Driver* VulkanDriver::create(VulkanPlatform* platform, VulkanContext const& cont
     //    VulkanRenderTarget            : 312       few
     // -- less than or equal to 312 bytes
 
-    utils::slog.d
+    FVK_LOGD
            << "\nVulkanSwapChain: " << sizeof(VulkanSwapChain)
            << "\nVulkanBufferObject: " << sizeof(VulkanBufferObject)
            << "\nVulkanVertexBuffer: " << sizeof(VulkanVertexBuffer)
@@ -660,8 +660,8 @@ void VulkanDriver::createFenceR(Handle<HwFence> fh, int) {
 
 void VulkanDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow, uint64_t flags) {
     if ((flags & backend::SWAP_CHAIN_CONFIG_SRGB_COLORSPACE) != 0 && !isSRGBSwapChainSupported()) {
-        utils::slog.w << "sRGB swapchain requested, but Platform does not support it"
-                      << utils::io::endl;
+        FVK_LOGW << "sRGB swapchain requested, but Platform does not support it"
+                 << utils::io::endl;
         flags = flags | ~(backend::SWAP_CHAIN_CONFIG_SRGB_COLORSPACE);
     }
     auto swapChain = mResourceAllocator.construct<VulkanSwapChain>(sch, mPlatform, mContext,
@@ -672,7 +672,7 @@ void VulkanDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow,
 void VulkanDriver::createSwapChainHeadlessR(Handle<HwSwapChain> sch, uint32_t width,
         uint32_t height, uint64_t flags) {
     if ((flags & backend::SWAP_CHAIN_CONFIG_SRGB_COLORSPACE) != 0 && !isSRGBSwapChainSupported()) {
-        utils::slog.w << "sRGB swapchain requested, but Platform does not support it"
+        FVK_LOGW << "sRGB swapchain requested, but Platform does not support it"
                       << utils::io::endl;
         flags = flags | ~(backend::SWAP_CHAIN_CONFIG_SRGB_COLORSPACE);
     }
@@ -914,7 +914,7 @@ bool VulkanDriver::isStereoSupported() {
         case backend::StereoscopicType::INSTANCED:
             return mContext.isClipDistanceSupported();
         case backend::StereoscopicType::MULTIVIEW:
-            // TODO: implement multiview feature in Vulkan.
+            return mContext.isMultiviewEnabled();
         case backend::StereoscopicType::NONE:
             return false;
     }
@@ -986,15 +986,15 @@ FeatureLevel VulkanDriver::getFeatureLevel() {
 
     // If the max sampler counts do not meet FL2 standards, then this is an FL1 device.
     const auto& fl2 = FEATURE_LEVEL_CAPS[+FeatureLevel::FEATURE_LEVEL_2];
-    if (fl2.MAX_VERTEX_SAMPLER_COUNT < limits.maxPerStageDescriptorSamplers ||
-        fl2.MAX_FRAGMENT_SAMPLER_COUNT < limits.maxPerStageDescriptorSamplers) {
+    if (limits.maxPerStageDescriptorSamplers < fl2.MAX_VERTEX_SAMPLER_COUNT  ||
+        limits.maxPerStageDescriptorSamplers < fl2.MAX_FRAGMENT_SAMPLER_COUNT) {
         return FeatureLevel::FEATURE_LEVEL_1;
     }
 
     // If the max sampler counts do not meet FL3 standards, then this is an FL2 device.
     const auto& fl3 = FEATURE_LEVEL_CAPS[+FeatureLevel::FEATURE_LEVEL_3];
-    if (fl3.MAX_VERTEX_SAMPLER_COUNT < limits.maxPerStageDescriptorSamplers ||
-        fl3.MAX_FRAGMENT_SAMPLER_COUNT < limits.maxPerStageDescriptorSamplers) {
+    if (limits.maxPerStageDescriptorSamplers < fl3.MAX_VERTEX_SAMPLER_COUNT||
+        limits.maxPerStageDescriptorSamplers < fl3.MAX_FRAGMENT_SAMPLER_COUNT) {
         return FeatureLevel::FEATURE_LEVEL_2;
     }
 
@@ -1862,9 +1862,9 @@ void VulkanDriver::bindPipeline(PipelineState const& pipelineState) {
         // matching characteristics. (e.g. if the missing texture is a 3D texture)
         if (UTILS_UNLIKELY(texture->getPrimaryImageLayout() == VulkanLayout::UNDEFINED)) {
 #if FVK_ENABLED(FVK_DEBUG_TEXTURE) && FVK_ENABLED_DEBUG_SAMPLER_NAME
-            utils::slog.w << "Uninitialized texture bound to '" << bindingToName[binding] << "'";
-            utils::slog.w << " in material '" << program->name.c_str() << "'";
-            utils::slog.w << " at binding point " << +binding << utils::io::endl;
+            FVK_LOGW << "Uninitialized texture bound to '" << bindingToName[binding] << "'";
+            FVK_LOGW << " in material '" << program->name.c_str() << "'";
+            FVK_LOGW << " at binding point " << +binding << utils::io::endl;
 #endif
             texture = mEmptyTexture;
         }
@@ -2014,7 +2014,7 @@ void VulkanDriver::debugCommandBegin(CommandStream* cmds, bool synchronous, cons
         assert_invariant(inRenderPass);
         inRenderPass = false;
     } else if (inRenderPass && OUTSIDE_COMMANDS.find(command) != OUTSIDE_COMMANDS.end()) {
-        utils::slog.e << command.data() << " issued inside a render pass." << utils::io::endl;
+        FVK_LOGE << command.data() << " issued inside a render pass." << utils::io::endl;
     }
 #endif
 }

filament/backend/src/vulkan/VulkanFboCache.cpp

@@ -95,7 +95,7 @@ VkFramebuffer VulkanFboCache::getFramebuffer(FboKey config) noexcept {
     }
 
     #if FVK_ENABLED(FVK_DEBUG_FBO_CACHE)
-    utils::slog.d << "Creating framebuffer " << config.width << "x" << config.height << " "
+    FVK_LOGD << "Creating framebuffer " << config.width << "x" << config.height << " "
         << "for render pass " << config.renderPass << ", "
         << "samples = " << int(config.samples) << ", "
         << "depth = " << (config.depth ? 1 : 0) << ", "
@@ -310,7 +310,7 @@ VkRenderPass VulkanFboCache::getRenderPass(RenderPassKey config) noexcept {
     mRenderPassCache[config] = {renderPass, mCurrentTime};
 
     #if FVK_ENABLED(FVK_DEBUG_FBO_CACHE)
-    utils::slog.d << "Created render pass " << renderPass << " with "
+    FVK_LOGD << "Created render pass " << renderPass << " with "
         << "samples = " << int(config.samples) << ", "
         << "depth = " << (hasDepth ? 1 : 0) << ", "
         << "colorAttachmentCount[0] = " << subpasses[0].colorAttachmentCount

filament/backend/src/vulkan/VulkanHandles.cpp

@@ -292,8 +292,8 @@ VulkanProgram::VulkanProgram(VkDevice device, Program const& builder) noexcept
     }
 
 #if FVK_ENABLED(FVK_DEBUG_SHADER_MODULE)
-    utils::slog.d << "Created VulkanProgram " << builder << ", shaders = (" << modules[0]
-                  << ", " << modules[1] << ")" << utils::io::endl;
+    FVK_LOGD << "Created VulkanProgram " << builder << ", shaders = (" << modules[0]
+             << ", " << modules[1] << ")" << utils::io::endl;
 #endif
 }
 

filament/backend/src/vulkan/VulkanPipelineCache.cpp

@@ -230,14 +230,15 @@ VulkanPipelineCache::PipelineCacheEntry* VulkanPipelineCache::createPipeline() n
     PipelineCacheEntry cacheEntry = {};
 
     #if FVK_ENABLED(FVK_DEBUG_SHADER_MODULE)
-        utils::slog.d << "vkCreateGraphicsPipelines with shaders = ("
-                << shaderStages[0].module << ", " << shaderStages[1].module << ")" << utils::io::endl;
+        FVK_LOGD << "vkCreateGraphicsPipelines with shaders = ("
+                 << shaderStages[0].module << ", " << shaderStages[1].module << ")"
+                 << utils::io::endl;
     #endif
     VkResult error = vkCreateGraphicsPipelines(mDevice, VK_NULL_HANDLE, 1, &pipelineCreateInfo,
             VKALLOC, &cacheEntry.handle);
     assert_invariant(error == VK_SUCCESS);
     if (error != VK_SUCCESS) {
-        utils::slog.e << "vkCreateGraphicsPipelines error " << error << utils::io::endl;
+        FVK_LOGE << "vkCreateGraphicsPipelines error " << error << utils::io::endl;
         return nullptr;
     }
 
@@ -252,7 +253,7 @@ void VulkanPipelineCache::bindProgram(VulkanProgram* program) noexcept {
 #if FVK_ENABLED(FVK_DEBUG_SHADER_MODULE)
     if (mPipelineRequirements.shaders[0] == VK_NULL_HANDLE ||
             mPipelineRequirements.shaders[1] == VK_NULL_HANDLE) {
-        utils::slog.e << "Binding missing shader: " << program->name.c_str() << utils::io::endl;
+        FVK_LOGE << "Binding missing shader: " << program->name.c_str() << utils::io::endl;
     }
 #endif
 }

filament/backend/src/vulkan/VulkanReadPixels.cpp

@@ -167,9 +167,9 @@ void VulkanReadPixels::run(VulkanRenderTarget* srcTarget, uint32_t const x, uint
     vkCreateImage(device, &imageInfo, VKALLOC, &stagingImage);
 
 #if FVK_ENABLED(FVK_DEBUG_READ_PIXELS)
-    utils::slog.d << "readPixels created image=" << stagingImage
-                  << " to copy from image=" << srcTexture->getVkImage()
-                  << " src-layout=" << srcTexture->getLayout(0, 0) << utils::io::endl;
+    FVK_LOGD << "readPixels created image=" << stagingImage
+             << " to copy from image=" << srcTexture->getVkImage()
+             << " src-layout=" << srcTexture->getLayout(0, 0) << utils::io::endl;
 #endif
 
     VkMemoryRequirements memReqs;
@@ -185,7 +185,7 @@ void VulkanReadPixels::run(VulkanRenderTarget* srcTarget, uint32_t const x, uint
     if (memoryTypeIndex >= VK_MAX_MEMORY_TYPES) {
         memoryTypeIndex = selectMemoryFunc(memReqs.memoryTypeBits,
                 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
-        utils::slog.w
+        FVK_LOGW
                 << "readPixels is slow because VK_MEMORY_PROPERTY_HOST_CACHED_BIT is not available"
                 << utils::io::endl;
     }
@@ -303,7 +303,7 @@ void VulkanReadPixels::run(VulkanRenderTarget* srcTarget, uint32_t const x, uint
         VkResult status = vkWaitForFences(device, 1, &fence, VK_TRUE, UINT64_MAX);
         // Fence hasn't been reached. Try waiting again.
         if (status != VK_SUCCESS) {
-            utils::slog.e << "Failed to wait for readPixels fence" << utils::io::endl;
+            FVK_LOGE << "Failed to wait for readPixels fence" << utils::io::endl;
             return;
         }
 
@@ -321,7 +321,7 @@ void VulkanReadPixels::run(VulkanRenderTarget* srcTarget, uint32_t const x, uint
                     getComponentCount(srcFormat), srcPixels,
                     static_cast<int>(subResourceLayout.rowPitch), static_cast<int>(width),
                     static_cast<int>(height), swizzle)) {
-            utils::slog.e << "Unsupported PixelDataFormat or PixelDataType" << utils::io::endl;
+            FVK_LOGE << "Unsupported PixelDataFormat or PixelDataType" << utils::io::endl;
         }
 
         vkUnmapMemory(device, stagingMemory);

filament/backend/src/vulkan/VulkanResourceAllocator.h

@@ -111,11 +111,11 @@ private:
 #if DEBUG_RESOURCE_LEAKS
 public:
     void print() {
-        utils::slog.d << "Resource Allocator state (debug only)" << utils::io::endl;
+        FVK_LOGD << "Resource Allocator state (debug only)" << utils::io::endl;
         for (size_t i = 0; i < RESOURCE_TYPE_COUNT; i++) {
-            utils::slog.d << "[" << i << "]=" << mDebugOnlyResourceCount[i] << utils::io::endl;
+            FVK_LOGD << "[" << i << "]=" << mDebugOnlyResourceCount[i] << utils::io::endl;
         }
-        utils::slog.d << "+++++++++++++++++++++++++++++++++++++" << utils::io::endl;
+        FVK_LOGD << "+++++++++++++++++++++++++++++++++++++" << utils::io::endl;
     }
 private:
     utils::FixedCapacityVector<size_t> mDebugOnlyResourceCount;

filament/backend/src/vulkan/VulkanStagePool.cpp

@@ -61,7 +61,7 @@ VulkanStage const* VulkanStagePool::acquireStage(uint32_t numBytes) {
 
 #if FVK_ENABLED(FVK_DEBUG_ALLOCATION)
     if (result != VK_SUCCESS) {
-        utils::slog.e << "Allocation error: " << result << utils::io::endl;
+        FVK_LOGE << "Allocation error: " << result << utils::io::endl;
     }
 #endif
 

filament/backend/src/vulkan/VulkanSwapChain.cpp

@@ -35,29 +35,13 @@ VulkanSwapChain::VulkanSwapChain(VulkanPlatform* platform, VulkanContext const&
       mStagePool(stagePool),
       mHeadless(extent.width != 0 && extent.height != 0 && !nativeWindow),
       mFlushAndWaitOnResize(platform->getCustomization().flushAndWaitOnWindowResize),
-      mCurrentImageReadyIndex(0),
+      mTransitionSwapChainImageLayoutForPresent(
+            platform->getCustomization().transitionSwapChainImageLayoutForPresent),
       mAcquired(false),
       mIsFirstRenderPass(true) {
     swapChain = mPlatform->createSwapChain(nativeWindow, flags, extent);
     FILAMENT_CHECK_POSTCONDITION(swapChain) << "Unable to create swapchain";
 
-    VkSemaphoreCreateInfo const createInfo = {
-            .sType = VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO,
-    };
-
-    // No need to wait on this semaphore before drawing when in Headless mode.
-    if (mHeadless) {
-		// Set all sempahores to VK_NULL_HANDLE
-		memset(mImageReady, 0, sizeof(mImageReady[0]) * IMAGE_READY_SEMAPHORE_COUNT);
-	} else {
-		for (uint32_t i = 0; i < IMAGE_READY_SEMAPHORE_COUNT; ++i) {
-			VkResult result =
-					vkCreateSemaphore(mPlatform->getDevice(), &createInfo, nullptr, mImageReady + i);
-                        FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-                                << "Failed to create semaphore";
-                }
-    }
-
     update();
 }
 
@@ -68,11 +52,6 @@ VulkanSwapChain::~VulkanSwapChain() {
     mCommands->wait();
 
     mPlatform->destroy(swapChain);
-	for (uint32_t i = 0; i < IMAGE_READY_SEMAPHORE_COUNT; ++i) {
-		if (mImageReady[i] != VK_NULL_HANDLE) {
-			vkDestroySemaphore(mPlatform->getDevice(), mImageReady[i], VKALLOC);
-		}
-	}
 }
 
 void VulkanSwapChain::update() {
@@ -95,7 +74,7 @@ void VulkanSwapChain::update() {
 }
 
 void VulkanSwapChain::present() {
-    if (!mHeadless) {
+    if (!mHeadless && mTransitionSwapChainImageLayoutForPresent) {
         VkCommandBuffer const cmdbuf = mCommands->get().buffer();
         VkImageSubresourceRange const subresources{
                 .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
@@ -106,10 +85,15 @@ void VulkanSwapChain::present() {
         };
         mColors[mCurrentSwapIndex]->transitionLayout(cmdbuf, subresources, VulkanLayout::PRESENT);
     }
-    mCommands->flush();
 
-    // We only present if it is not headless.  No-op for headless (but note that we still need the
-    // flush() in the above line).
+    mCommands->flush();
+    
+    // call the image ready wait function
+    if (mExplicitImageReadyWait != nullptr) {
+        mExplicitImageReadyWait(swapChain);
+    }
+
+    // We only present if it is not headless. No-op for headless.
     if (!mHeadless) {
         VkSemaphore const finishedDrawing = mCommands->acquireFinishedSignal();
         VkResult const result = mPlatform->present(swapChain, mCurrentSwapIndex, finishedDrawing);
@@ -139,13 +123,14 @@ void VulkanSwapChain::acquire(bool& resized) {
         update();
     }
 
-	mCurrentImageReadyIndex = (mCurrentImageReadyIndex + 1) % IMAGE_READY_SEMAPHORE_COUNT;
-	const VkSemaphore imageReady = mImageReady[mCurrentImageReadyIndex];
-    VkResult const result = mPlatform->acquire(swapChain, imageReady, &mCurrentSwapIndex);
-    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS || result == VK_SUBOPTIMAL_KHR)
-            << "Cannot acquire in swapchain.";
-    if (imageReady != VK_NULL_HANDLE) {
-        mCommands->injectDependency(imageReady);
+    VulkanPlatform::ImageSyncData imageSyncData;
+    VkResult const result = mPlatform->acquire(swapChain, &imageSyncData);
+    mCurrentSwapIndex = imageSyncData.imageIndex;
+    mExplicitImageReadyWait = imageSyncData.explicitImageReadyWait;
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS || result == VK_SUBOPTIMAL_KHR) 
+        << "Cannot acquire in swapchain.";
+    if (imageSyncData.imageReadySemaphore != VK_NULL_HANDLE) {
+        mCommands->injectDependency(imageSyncData.imageReadySemaphore);
     }
     mAcquired = true;
 }

filament/backend/src/vulkan/VulkanSwapChain.h

@@ -50,7 +50,10 @@ struct VulkanSwapChain : public HwSwapChain, VulkanResource {
     void acquire(bool& reized);
 
     inline VulkanTexture* getCurrentColor() const noexcept {
-        return mColors[mCurrentSwapIndex].get();
+        uint32_t const imageIndex = mCurrentSwapIndex;
+        FILAMENT_CHECK_PRECONDITION(
+            imageIndex != VulkanPlatform::ImageSyncData::INVALID_IMAGE_INDEX);
+        return mColors[imageIndex].get();
     }
 
     inline VulkanTexture* getDepth() const noexcept {
@@ -80,15 +83,15 @@ private:
     VulkanStagePool& mStagePool;
     bool const mHeadless;
     bool const mFlushAndWaitOnResize;
+    bool const mTransitionSwapChainImageLayoutForPresent;
 
     // We create VulkanTextures based on VkImages. VulkanTexture has facilities for doing layout
     // transitions, which are useful here.
     utils::FixedCapacityVector<std::unique_ptr<VulkanTexture>> mColors;
     std::unique_ptr<VulkanTexture> mDepth;
     VkExtent2D mExtent;
-    VkSemaphore mImageReady[IMAGE_READY_SEMAPHORE_COUNT];
-	uint32_t mCurrentImageReadyIndex;
     uint32_t mCurrentSwapIndex;
+    std::function<void(Platform::SwapChain* handle)> mExplicitImageReadyWait = nullptr;
     bool mAcquired;
     bool mIsFirstRenderPass;
 };

filament/backend/src/vulkan/VulkanTexture.cpp

@@ -113,7 +113,7 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
         VkFormatProperties props;
         vkGetPhysicalDeviceFormatProperties(physicalDevice, mVkFormat, &props);
         if (!(props.optimalTilingFeatures & VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT)) {
-            utils::slog.w << "Texture usage is SAMPLEABLE but format " << mVkFormat << " is not "
+            FVK_LOGW << "Texture usage is SAMPLEABLE but format " << mVkFormat << " is not "
                     "sampleable with optimal tiling." << utils::io::endl;
         }
 #endif
@@ -163,7 +163,7 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
 
     VkResult error = vkCreateImage(mDevice, &imageInfo, VKALLOC, &mTextureImage);
     if (error || FVK_ENABLED(FVK_DEBUG_TEXTURE)) {
-        utils::slog.d << "vkCreateImage: "
+        FVK_LOGD << "vkCreateImage: "
             << "image = " << mTextureImage << ", "
             << "result = " << error << ", "
             << "handle = " << utils::io::hex << mTextureImage << utils::io::dec << ", "
@@ -450,7 +450,7 @@ void VulkanTexture::transitionLayout(VkCommandBuffer cmdbuf, const VkImageSubres
     }
 
 #if FVK_ENABLED(FVK_DEBUG_LAYOUT_TRANSITION)
-    utils::slog.d << "transition texture=" << mTextureImage
+    FVK_LOGD << "transition texture=" << mTextureImage
                   << " (" << range.baseArrayLayer
                   << "," << range.baseMipLevel << ")"
                   << " count=(" << range.layerCount
@@ -539,7 +539,7 @@ void VulkanTexture::print() const {
                 layer < (mPrimaryViewRange.baseArrayLayer + mPrimaryViewRange.layerCount) &&
                 level >= mPrimaryViewRange.baseMipLevel &&
                 level < (mPrimaryViewRange.baseMipLevel + mPrimaryViewRange.levelCount);
-            utils::slog.d << "[" << mTextureImage << "]: (" << layer << "," << level
+            FVK_LOGD << "[" << mTextureImage << "]: (" << layer << "," << level
                           << ")=" << getLayout(layer, level)
                           << " primary=" << primary
                           << utils::io::endl;
@@ -548,7 +548,7 @@ void VulkanTexture::print() const {
 
     for (auto view: mCachedImageViews) {
         auto& range = view.first.range;
-        utils::slog.d << "[" << mTextureImage << ", imageView=" << view.second << "]=>"
+        FVK_LOGD << "[" << mTextureImage << ", imageView=" << view.second << "]=>"
                       << " (" << range.baseArrayLayer << "," << range.baseMipLevel << ")"
                       << " count=(" << range.layerCount << "," << range.levelCount << ")"
                       << " aspect=" << range.aspectMask << " viewType=" << view.first.type

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

@@ -33,11 +33,11 @@ VkPipelineLayout VulkanPipelineLayoutCache::getLayout(
 
     // build the push constant layout key
     uint32_t pushConstantRangeCount = program->getPushConstantRangeCount();
-    auto const& pushCostantRanges = program->getPushConstantRanges();    
+    auto const& pushConstantRanges = program->getPushConstantRanges();    
     if (pushConstantRangeCount > 0) {
         assert_invariant(pushConstantRangeCount <= Program::SHADER_TYPE_COUNT);
         for (uint8_t i = 0; i < pushConstantRangeCount; ++i) {
-            auto const& range = pushCostantRanges[i];
+            auto const& range = pushConstantRanges[i];
             auto& pushConstant = key.pushConstant[i];
             if (range.stageFlags &  VK_SHADER_STAGE_VERTEX_BIT) {
                 pushConstant.stage = static_cast<uint8_t>(ShaderStage::VERTEX);
@@ -52,9 +52,8 @@ VkPipelineLayout VulkanPipelineLayoutCache::getLayout(
         }
     }
 
-    auto iter = mPipelineLayouts.find(key);
-    if (iter != mPipelineLayouts.end()) {
-        PipelineLayoutCacheEntry& entry = mPipelineLayouts[key];
+    if (PipelineLayoutMap::iterator iter = mPipelineLayouts.find(key); iter != mPipelineLayouts.end()) {
+        PipelineLayoutCacheEntry& entry = iter.value();
         entry.lastUsed = mTimestamp++;
         return entry.handle;
     }
@@ -65,7 +64,7 @@ VkPipelineLayout VulkanPipelineLayoutCache::getLayout(
         .setLayoutCount = (uint32_t) descSetLayoutCount,
         .pSetLayouts = key.descSetLayouts.data(),
         .pushConstantRangeCount = pushConstantRangeCount,
-        .pPushConstantRanges = pushCostantRanges,
+        .pPushConstantRanges = pushConstantRanges,
     };
 
     VkPipelineLayout layout;

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

@@ -95,7 +95,7 @@ void printDeviceInfo(VkInstance instance, VkPhysicalDevice device) {
                 .pNext = &driverProperties,
         };
         vkGetPhysicalDeviceProperties2(device, &physicalDeviceProperties2);
-        utils::slog.i << "Vulkan device driver: " << driverProperties.driverName << " "
+        FVK_LOGI << "Vulkan device driver: " << driverProperties.driverName << " "
                       << driverProperties.driverInfo << utils::io::endl;
     }
 
@@ -128,7 +128,7 @@ void printDeviceInfo(VkInstance instance, VkPhysicalDevice device) {
     const FixedCapacityVector<VkPhysicalDevice> physicalDevices
             = filament::backend::enumerate(vkEnumeratePhysicalDevices, instance);
 
-    utils::slog.i << "Selected physical device '" << deviceProperties.deviceName << "' from "
+    FVK_LOGI << "Selected physical device '" << deviceProperties.deviceName << "' from "
                   << physicalDevices.size() << " physical devices. "
                   << "(vendor " << utils::io::hex << vendorID << ", "
                   << "device " << deviceID << ", "
@@ -142,15 +142,15 @@ void printDepthFormats(VkPhysicalDevice device) {
     // Note that Vulkan is more constrained than OpenGL ES 3.1 in this area.
     const VkFormatFeatureFlags required = VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT
                                               | VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT;
-    utils::slog.i << "Sampleable depth formats: ";
+    FVK_LOGI << "Sampleable depth formats: ";
     for (VkFormat format : ALL_VK_FORMATS) {
         VkFormatProperties props;
         vkGetPhysicalDeviceFormatProperties(device, format, &props);
         if ((props.optimalTilingFeatures & required) == required) {
-            utils::slog.i << format << " ";
+            FVK_LOGI << format << " ";
         }
     }
-    utils::slog.i << utils::io::endl;
+    FVK_LOGI << utils::io::endl;
 }
 #endif
 
@@ -198,6 +198,7 @@ ExtensionSet getDeviceExtensions(VkPhysicalDevice device) {
             VK_KHR_MAINTENANCE1_EXTENSION_NAME,
             VK_KHR_MAINTENANCE2_EXTENSION_NAME,
             VK_KHR_MAINTENANCE3_EXTENSION_NAME,
+            VK_KHR_MULTIVIEW_EXTENSION_NAME,
     };
     ExtensionSet exts;
     // Identify supported physical device extensions
@@ -241,11 +242,11 @@ VkInstance createInstance(ExtensionSet const& requiredExts) {
         instanceCreateInfo.ppEnabledLayerNames = enabledLayers.data();
     } else {
 #if defined(__ANDROID__)
-        utils::slog.d << "Validation layers are not available; did you set jniLibs in your "
-                      << "gradle file?" << utils::io::endl;
+        FVK_LOGD << "Validation layers are not available; did you set jniLibs in your "
+                 << "gradle file?" << utils::io::endl;
 #else
-        utils::slog.d << "Validation layer not available; did you install the Vulkan SDK?\n"
-                      << "Please ensure that VK_LAYER_PATH is set correctly." << utils::io::endl;
+        FVK_LOGD << "Validation layer not available; did you install the Vulkan SDK?\n"
+                 << "Please ensure that VK_LAYER_PATH is set correctly." << utils::io::endl;
 #endif // __ANDROID__
 
     }
@@ -334,6 +335,7 @@ VkDevice createLogicalDevice(VkPhysicalDevice physicalDevice,
     deviceCreateInfo.enabledExtensionCount = (uint32_t) requestExtensions.size();
     deviceCreateInfo.ppEnabledExtensionNames = requestExtensions.data();
 
+    void* pNext = nullptr;
     VkPhysicalDevicePortabilitySubsetFeaturesKHR portability = {
             .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PORTABILITY_SUBSET_FEATURES_KHR,
             .pNext = nullptr,
@@ -341,9 +343,24 @@ VkDevice createLogicalDevice(VkPhysicalDevice physicalDevice,
             .mutableComparisonSamplers = VK_TRUE,
     };
     if (setContains(deviceExtensions, VK_KHR_PORTABILITY_SUBSET_EXTENSION_NAME)) {
-        deviceCreateInfo.pNext = &portability;
+        portability.pNext = pNext;
+        pNext = &portability;
     }
 
+    VkPhysicalDeviceMultiviewFeaturesKHR multiview = {
+            .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES_KHR,
+            .pNext = nullptr,
+            .multiview = VK_TRUE,
+            .multiviewGeometryShader = VK_FALSE,
+            .multiviewTessellationShader = VK_FALSE
+    };
+    if (setContains(deviceExtensions, VK_KHR_MULTIVIEW_EXTENSION_NAME)) {
+        multiview.pNext = pNext;
+        pNext = &multiview;
+    }
+
+    deviceCreateInfo.pNext = pNext;
+
     VkResult result = vkCreateDevice(physicalDevice, &deviceCreateInfo, VKALLOC, &device);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "vkCreateDevice error=" << result << ".";
 
@@ -418,8 +435,8 @@ inline int deviceTypeOrder(VkPhysicalDeviceType deviceType) {
         case VK_PHYSICAL_DEVICE_TYPE_OTHER:
             return 1;
         default:
-            utils::slog.w << "deviceTypeOrder: Unexpected deviceType: " << deviceType
-                          << utils::io::endl;
+            FVK_LOGW << "deviceTypeOrder: Unexpected deviceType: " << deviceType
+                     << utils::io::endl;
             return -1;
     }
 }
@@ -701,6 +718,10 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
                 = pruneExtensions(mImpl->mPhysicalDevice, instExts, deviceExts);
         instExts = prunedInstExts;
         deviceExts = prunedDeviceExts;
+
+        if (driverConfig.stereoscopicType != StereoscopicType::MULTIVIEW) {
+            deviceExts.erase(VK_KHR_MULTIVIEW_EXTENSION_NAME);
+        }
     }
 
     mImpl->mDevice
@@ -716,6 +737,7 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
     // Store the extension support in the context
     context.mDebugUtilsSupported = setContains(instExts, VK_EXT_DEBUG_UTILS_EXTENSION_NAME);
     context.mDebugMarkersSupported = setContains(deviceExts, VK_EXT_DEBUG_MARKER_EXTENSION_NAME);
+    context.mMultiviewEnabled = setContains(deviceExts, VK_KHR_MULTIVIEW_EXTENSION_NAME);
 
 #ifdef NDEBUG
     // If we are in release build, we should not have turned on debug extensions
@@ -750,8 +772,8 @@ VulkanPlatform::SwapChainBundle VulkanPlatform::getSwapChainBundle(SwapChainPtr
     SWAPCHAIN_RET_FUNC(getSwapChainBundle, handle, )
 }
 
-VkResult VulkanPlatform::acquire(SwapChainPtr handle, VkSemaphore clientSignal, uint32_t* index) {
-    SWAPCHAIN_RET_FUNC(acquire, handle, clientSignal, index)
+VkResult VulkanPlatform::acquire(SwapChainPtr handle, ImageSyncData* outImageSyncData) {
+    SWAPCHAIN_RET_FUNC(acquire, handle, outImageSyncData)
 }
 
 VkResult VulkanPlatform::present(SwapChainPtr handle, uint32_t index,

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

@@ -133,6 +133,7 @@ VulkanPlatformSurfaceSwapChain::VulkanPlatformSurfaceSwapChain(VulkanContext con
 VulkanPlatformSurfaceSwapChain::~VulkanPlatformSurfaceSwapChain() {
     vkDestroySwapchainKHR(mDevice, mSwapchain, VKALLOC);
     vkDestroySurfaceKHR(mInstance, mSurface, VKALLOC);
+    destroy();
 }
 
 VkResult VulkanPlatformSurfaceSwapChain::create() {
@@ -153,7 +154,7 @@ VkResult VulkanPlatformSurfaceSwapChain::create() {
     // the number of images, though there may be limits related to the total amount of memory used
     // by presentable images."
     if (maxImageCount != 0 && desiredImageCount > maxImageCount) {
-        utils::slog.e << "Swap chain does not support " << desiredImageCount << " images."
+        FVK_LOGE << "Swap chain does not support " << desiredImageCount << " images."
                       << utils::io::endl;
         desiredImageCount = caps.minImageCount;
     }
@@ -246,7 +247,7 @@ VkResult VulkanPlatformSurfaceSwapChain::create() {
             selectDepthFormat(mContext.getAttachmentDepthStencilFormats(), mHasStencil);
     mSwapChainBundle.depth = createImage(mSwapChainBundle.extent, mSwapChainBundle.depthFormat);
 
-    slog.i << "vkCreateSwapchain"
+    FVK_LOGI << "vkCreateSwapchain"
            << ": " << mSwapChainBundle.extent.width << "x" << mSwapChainBundle.extent.height << ", "
            << surfaceFormat.format << ", " << surfaceFormat.colorSpace << ", "
            << "swapchain-size=" << mSwapChainBundle.colors.size() << ", "
@@ -254,19 +255,29 @@ VkResult VulkanPlatformSurfaceSwapChain::create() {
            << "depth=" << mSwapChainBundle.depthFormat
            << io::endl;
 
+    VkSemaphoreCreateInfo const semaphoreCreateInfo = {
+            .sType = VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO,
+    };
+
+    for (uint32_t i = 0; i < IMAGE_READY_SEMAPHORE_COUNT; ++i) {
+        VkResult result = vkCreateSemaphore(mDevice, &semaphoreCreateInfo, nullptr,
+                mImageReady + i);
+        FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "Failed to create semaphore";
+    }
+
     return result;
 }
 
-VkResult VulkanPlatformSurfaceSwapChain::acquire(VkSemaphore clientSignal, uint32_t* index) {
-    // This immediately retrieves the index of the next available presentable image, and
-    // asynchronously requests the GPU to trigger the "imageAvailable" semaphore.
-    VkResult result = vkAcquireNextImageKHR(mDevice, mSwapchain, UINT64_MAX, clientSignal,
-            VK_NULL_HANDLE, index);
+VkResult VulkanPlatformSurfaceSwapChain::acquire(VulkanPlatform::ImageSyncData* outImageSyncData) {
+    mCurrentImageReadyIndex = (mCurrentImageReadyIndex + 1) % IMAGE_READY_SEMAPHORE_COUNT;
+    outImageSyncData->imageReadySemaphore = mImageReady[mCurrentImageReadyIndex];
+    VkResult result = vkAcquireNextImageKHR(mDevice, mSwapchain, UINT64_MAX,
+            outImageSyncData->imageReadySemaphore, VK_NULL_HANDLE, &outImageSyncData->imageIndex);
 
     // Users should be notified of a suboptimal surface, but it should not cause a cascade of
     // log messages or a loop of re-creations.
     if (result == VK_SUBOPTIMAL_KHR && !mSuboptimal) {
-        slog.w << "Vulkan Driver: Suboptimal swap chain." << io::endl;
+        FVK_LOGW << "Vulkan Driver: Suboptimal swap chain." << io::endl;
         mSuboptimal = true;
     }
     return result;
@@ -288,7 +299,7 @@ VkResult VulkanPlatformSurfaceSwapChain::present(uint32_t index, VkSemaphore fin
     // On Android Q and above, a suboptimal surface is always reported after screen rotation:
     // https://android-developers.googleblog.com/2020/02/handling-device-orientation-efficiently.html
     if (result == VK_SUBOPTIMAL_KHR && !mSuboptimal) {
-        utils::slog.w << "Vulkan Driver: Suboptimal swap chain." << utils::io::endl;
+        FVK_LOGW << "Vulkan Driver: Suboptimal swap chain." << utils::io::endl;
         mSuboptimal = true;
     }
     return result;
@@ -312,6 +323,17 @@ VkResult VulkanPlatformSurfaceSwapChain::recreate() {
     return create();
 }
 
+void VulkanPlatformSurfaceSwapChain::destroy() {
+    VulkanPlatformSwapChainImpl::destroy();
+
+    for (uint32_t i = 0; i < IMAGE_READY_SEMAPHORE_COUNT; ++i) {
+        if (mImageReady[i] != VK_NULL_HANDLE) {
+            vkDestroySemaphore(mDevice, mImageReady[i], VKALLOC);
+            mImageReady[i] = VK_NULL_HANDLE;
+        }
+    }
+}
+
 VulkanPlatformHeadlessSwapChain::VulkanPlatformHeadlessSwapChain(VulkanContext const& context,
         VkDevice device, VkQueue queue, VkExtent2D extent, uint64_t flags)
     : VulkanPlatformSwapChainImpl(context, device, queue),
@@ -343,8 +365,8 @@ VkResult VulkanPlatformHeadlessSwapChain::present(uint32_t index, VkSemaphore fi
     return VK_SUCCESS;
 }
 
-VkResult VulkanPlatformHeadlessSwapChain::acquire(VkSemaphore clientSignal, uint32_t* index) {
-    *index = mCurrentIndex;
+VkResult VulkanPlatformHeadlessSwapChain::acquire(VulkanPlatform::ImageSyncData* outImageSyncData) {
+    outImageSyncData->imageIndex = mCurrentIndex;
     mCurrentIndex = (mCurrentIndex + 1) % HEADLESS_SWAPCHAIN_SIZE;
     return VK_SUCCESS;
 }

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

@@ -42,7 +42,7 @@ struct VulkanPlatformSwapChainImpl : public Platform::SwapChain {
     ~VulkanPlatformSwapChainImpl();
 
     // Non-virtual override-able method
-    VkResult acquire(VkSemaphore clientSignal, uint32_t* index) {
+    VkResult acquire(VulkanPlatform::ImageSyncData* outImageSyncData) {
         PANIC_PRECONDITION("Should not be called");
         return VK_ERROR_UNKNOWN;
     }
@@ -86,7 +86,7 @@ struct VulkanPlatformSurfaceSwapChain : public VulkanPlatformSwapChainImpl {
     ~VulkanPlatformSurfaceSwapChain();
 
     // Non-virtual override
-    VkResult acquire(VkSemaphore clientSignal, uint32_t* index);
+    VkResult acquire(VulkanPlatform::ImageSyncData* outImageSyncData);
 
     // Non-virtual override
     VkResult present(uint32_t index, VkSemaphore finished);
@@ -97,7 +97,13 @@ struct VulkanPlatformSurfaceSwapChain : public VulkanPlatformSwapChainImpl {
     // Non-virtual override-able method
     bool hasResized();
 
+protected:
+    // Non-virtual override-able method
+    void destroy();
+
 private:
+    static constexpr int IMAGE_READY_SEMAPHORE_COUNT = FVK_MAX_COMMAND_BUFFERS;
+    
     VkResult create();
 
     VkInstance mInstance;
@@ -106,6 +112,8 @@ private:
     VkSurfaceKHR mSurface;
     VkSwapchainKHR mSwapchain = VK_NULL_HANDLE;
     VkExtent2D const mFallbackExtent;
+    VkSemaphore mImageReady[IMAGE_READY_SEMAPHORE_COUNT];
+    uint32_t mCurrentImageReadyIndex;
 
     bool mUsesRGB = false;
     bool mHasStencil = false;
@@ -121,7 +129,7 @@ struct VulkanPlatformHeadlessSwapChain : public VulkanPlatformSwapChainImpl {
     ~VulkanPlatformHeadlessSwapChain();
 
     // Non-virtual override
-    VkResult acquire(VkSemaphore clientSignal, uint32_t* index);
+    VkResult acquire(VulkanPlatform::ImageSyncData* outImageSyncData);
 
     // Non-virtual override
     VkResult present(uint32_t index, VkSemaphore finished);

filament/docs/SpirvDebugging.md

@@ -64,14 +64,15 @@ First determine the Filament material and variant that causes the problem.
 What we want is the "raw" GLSL version of the shader, before any optimizations / cross-compilation
 happens.
 
-The easiest way I've found to do this is by adding a `std::cout` statement inside `MaterialBuilder`,
-right after `sg.createFragmentProgram` or `sg.createVertexProgram` is called. For example,
+We can use the `--save-raw-variants` debug flag in matc to export each GLSL
+shader to a file. For example:
 
 ```
-if (v.variant.key == 8u) { std::cout << shader << std::endl; }
+matc --save-raw-variants --optimize-size --variant-filter fog,ssr,vsm,stereo \
+        -a all -p all -o mymaterial.filamat mymaterial.mat
 ```
 
-Then, I invoke `matc` and redirect stdout to a .frag or .vert file.
+Files will be named like `mymaterial_0x05.frag` or `mymaterial_0x05.vert`.
 
 Note that gltfio material "templates" first go through a build step. After building gltfio, the
 gltfio Filament materials are output to:

filament/include/filament/Engine.h

@@ -298,6 +298,24 @@ public:
          */
         size_t textureUseAfterFreePoolSize = 0;
 
+        /**
+         * When uploading vertex or index data, the Filament Metal backend copies data
+         * into a shared staging area before transferring it to the GPU. This setting controls
+         * the total size of the buffer used to perform these allocations.
+         *
+         * Higher values can improve performance when performing many uploads across a small
+         * number of frames.
+         *
+         * This buffer remains alive throughout the lifetime of the Engine, so this size adds to the
+         * memory footprint of the app and should be set as conservative as possible.
+         *
+         * A value of 0 disables the shared staging buffer entirely; uploads will acquire an
+         * individual buffer from a pool of shared buffers.
+         *
+         * Only respected by the Metal backend.
+         */
+        size_t metalUploadBufferSizeBytes = 512 * 1024;
+
         /**
          * Set to `true` to forcibly disable parallel shader compilation in the backend.
          * Currently only honored by the GL and Metal backends.
@@ -332,9 +350,12 @@ public:
         uint32_t resourceAllocatorCacheSizeMB = 64;
 
         /*
-         * This value determines for how many frames are texture entries kept in the cache.
+         * This value determines how many frames texture entries are kept for in the cache. This
+         * is a soft limit, meaning some texture older than this are allowed to stay in the cache.
+         * Typically only one texture is evicted per frame.
+         * The default is 1.
          */
-        uint32_t resourceAllocatorCacheMaxAge = 2;
+        uint32_t resourceAllocatorCacheMaxAge = 1;
 
         /*
          * Disable backend handles use-after-free checks.
@@ -1021,6 +1042,21 @@ public:
     void resetBackendState() noexcept;
 #endif
 
+    /**
+     * Get the current time. This is a convenience function that simply returns the
+     * time in nanosecond since epoch of std::chrono::steady_clock.
+     * A possible implementation is:
+     *
+     * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+     *     return std::chrono::steady_clock::now().time_since_epoch().count();
+     * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+     *
+     * @return current time in nanosecond since epoch of std::chrono::steady_clock.
+     * @see Renderer::beginFrame()
+     */
+    static uint64_t getSteadyClockTimeNano() noexcept;
+
+
     DebugRegistry& getDebugRegistry() noexcept;
 
 protected:

filament/include/filament/IndirectLight.h

@@ -158,6 +158,8 @@ public:
          *
          * @return This Builder, for chaining calls.
          *
+         * @see Material::Builder::sphericalHarmonicsBandCount()
+         *
          * @note
          * Because the coefficients are pre-scaled, `sh[0]` is the environment's
          * average irradiance.

filament/include/filament/Material.h

@@ -140,6 +140,18 @@ public:
             return constant(name, strlen(name), value);
         }
 
+        /**
+         * Sets the quality of the indirect lights computations. This is only taken into account
+         * if this material is lit and in the surface domain. This setting will affect the
+         * IndirectLight computation if one is specified on the Scene and Spherical Harmonics
+         * are used for the irradiance.
+         *
+         * @param shBandCount Number of spherical harmonic bands. Must be 1, 2 or 3 (default).
+         * @return Reference to this Builder for chaining calls.
+         * @see IndirectLight
+         */
+        Builder& sphericalHarmonicsBandCount(size_t shBandCount) noexcept;
+
         /**
          * Creates the Material object and returns a pointer to it.
          *

filament/include/filament/MaterialInstance.h

@@ -140,6 +140,7 @@ public:
      * @param values        Array of values to set to the named parameter array.
      * @param count         Size of the array to set.
      * @throws utils::PreConditionPanic if name doesn't exist or no-op if exceptions are disabled.
+     * @see Material::hasParameter
      */
     template<typename T, typename = is_supported_parameter_t<T>>
     void setParameter(const char* UTILS_NONNULL name, size_t nameLength,
@@ -234,6 +235,32 @@ public:
         setParameter(name, strlen(name), type, color);
     }
 
+    /**
+     * Gets the value of a parameter by name.
+     * 
+     * Note: Only supports non-texture parameters such as numeric and math types.
+     * 
+     * @param name          Name of the parameter as defined by Material. Cannot be nullptr.
+     * @param nameLength    Length in `char` of the name parameter.
+     * @throws utils::PreConditionPanic if name doesn't exist or no-op if exceptions are disabled.
+     * 
+     * @see Material::hasParameter
+     */
+    template<typename T>
+    T getParameter(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_parameter_t<T>>
+    inline T getParameter(StringLiteral name) const {
+        return getParameter<T>(name.data, name.size);
+    }
+
+    /** inline helper to provide the name as a null-terminated C string */
+    template<typename T, typename = is_supported_parameter_t<T>>
+    inline T getParameter(const char* UTILS_NONNULL name) const {
+        return getParameter<T>(name, strlen(name));
+    }
+
     /**
      * Set-up a custom scissor rectangle; by default it is disabled.
      *

filament/include/filament/RenderableManager.h

@@ -489,43 +489,20 @@ public:
          * For standard morphing, A MorphTargetBuffer must be provided.
          * Standard morphing supports up to \c CONFIG_MAX_MORPH_TARGET_COUNT morph targets.
          *
-         * For legacy morphing, the attached VertexBuffer must provide data in the
-         * appropriate VertexAttribute slots (\c MORPH_POSITION_0 etc). Legacy morphing only
-         * supports up to 4 morph targets and will be deprecated in the future. Legacy morphing must
-         * be enabled on the material definition: either via the legacyMorphing material attribute
-         * or by calling filamat::MaterialBuilder::useLegacyMorphing().
-         *
          * See also RenderableManager::setMorphWeights(), which can be called on a per-frame basis
          * to advance the animation.
          */
         Builder& morphing(MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer) noexcept;
 
-        /**
-         * @deprecated Use morphing(uint8_t level, size_t primitiveIndex, size_t offset, size_t count) instead
-         */
-        Builder& morphing(uint8_t level, size_t primitiveIndex,
-                MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer,
-                size_t offset, size_t count) noexcept;
-
-        /**
-         * @deprecated Use morphing(uint8_t level, size_t primitiveIndex, size_t offset, size_t count) instead
-         */
-        inline Builder& morphing(uint8_t level, size_t primitiveIndex,
-                MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer) noexcept {
-            return morphing(level, primitiveIndex, morphTargetBuffer, 0,
-                    morphTargetBuffer->getVertexCount());
-        }
-
         /**
          * Specifies the the range of the MorphTargetBuffer to use with this primitive.
          *
          * @param level the level of detail (lod), only 0 can be specified
          * @param primitiveIndex zero-based index of the primitive, must be less than the count passed to Builder constructor
          * @param offset specifies where in the morph target buffer to start reading (expressed as a number of vertices)
-         * @param count number of vertices in the morph target buffer to read, must equal the geometry's count (for triangles, this should be a multiple of 3)
          */
-        Builder& morphing(uint8_t level, size_t primitiveIndex,
-                size_t offset, size_t count) noexcept;
+        RenderableManager::Builder& morphing(uint8_t level,
+                size_t primitiveIndex, size_t offset) noexcept;
 
 
         /**
@@ -786,25 +763,13 @@ public:
     /**
      * Associates a MorphTargetBuffer to the given primitive.
      */
-    void setMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex,
-            size_t offset, size_t count);
-
-    /** @deprecated */
-    void setMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex,
-            MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer, size_t offset, size_t count);
-
-    /** @deprecated */
-    inline void setMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex,
-            MorphTargetBuffer* UTILS_NONNULL morphTargetBuffer) {
-        setMorphTargetBufferAt(instance, level, primitiveIndex, morphTargetBuffer, 0,
-                morphTargetBuffer->getVertexCount());
-    }
+    void setMorphTargetBufferOffsetAt(Instance instance, uint8_t level, size_t primitiveIndex,
+            size_t offset);
 
     /**
-     * Get a MorphTargetBuffer to the given primitive or null if it doesn't exist.
+     * Get a MorphTargetBuffer to the given renderable or null if it doesn't exist.
      */
-    MorphTargetBuffer* UTILS_NULLABLE getMorphTargetBufferAt(Instance instance,
-            uint8_t level, size_t primitiveIndex) const noexcept;
+    MorphTargetBuffer* UTILS_NULLABLE getMorphTargetBuffer(Instance instance) const noexcept;
 
     /**
      * Gets the number of morphing in the given entity.

filament/include/filament/Renderer.h

@@ -22,9 +22,11 @@
 #include <filament/FilamentAPI.h>
 
 #include <utils/compiler.h>
+#include <utils/FixedCapacityVector.h>
 
 #include <math/vec4.h>
 
+#include <stddef.h>
 #include <stdint.h>
 
 namespace filament {
@@ -81,6 +83,37 @@ public:
         UTILS_DEPRECATED uint64_t vsyncOffsetNanos = 0;
     };
 
+    /**
+     * Timing information about a frame
+     * @see getFrameInfoHistory()
+     */
+    struct FrameInfo {
+        using time_point_ns = int64_t;
+        using duration_ns = int64_t;
+        uint32_t frameId;                   //!< monotonically increasing frame identifier
+        duration_ns frameTime;              //!< frame duration on the GPU in nanosecond [ns]
+        duration_ns denoisedFrameTime;      //!< denoised frame duration on the GPU in [ns]
+        time_point_ns beginFrame;           //!< Renderer::beginFrame() time since epoch [ns]
+        time_point_ns endFrame;             //!< Renderer::endFrame() time since epoch [ns]
+        time_point_ns backendBeginFrame;    //!< Backend thread time of frame start since epoch [ns]
+        time_point_ns backendEndFrame;      //!< Backend thread time of frame end since epoch [ns]
+    };
+
+    /**
+     * Retrieve an historic of frame timing information. The maximum frame history size is
+     * given by getMaxFrameHistorySize().
+     * @param historySize requested history size. The returned vector could be smaller.
+     * @return A vector of FrameInfo.
+     */
+    utils::FixedCapacityVector<Renderer::FrameInfo> getFrameInfoHistory(
+            size_t historySize = 1) const noexcept;
+
+    /**
+     * @return the maximum supported frame history size.
+     * @see getFrameInfoHistory()
+     */
+    size_t getMaxFrameHistorySize() const noexcept;
+
     /**
      * Use FrameRateOptions to set the desired frame rate and control how quickly the system
      * reacts to GPU load changes.
@@ -227,6 +260,25 @@ public:
     static constexpr CopyFrameFlag CLEAR = 0x4;
 
 
+    /**
+     * The use of this method is optional. It sets the VSYNC time expressed as the duration in
+     * nanosecond since epoch of std::chrono::steady_clock.
+     * If called, passing 0 to vsyncSteadyClockTimeNano in Renderer::BeginFrame will use this
+     * time instead.
+     * @param steadyClockTimeNano duration in nanosecond since epoch of std::chrono::steady_clock
+     * @see Engine::getSteadyClockTimeNano()
+     * @see Renderer::BeginFrame()
+     */
+    void setVsyncTime(uint64_t steadyClockTimeNano) noexcept;
+
+    /**
+     * Call skipFrame when momentarily skipping frames, for instance if the content of the
+     * scene doesn't change.
+     *
+     * @param vsyncSteadyClockTimeNano
+     */
+    void skipFrame(uint64_t vsyncSteadyClockTimeNano = 0u);
+
     /**
      * Set-up a frame for this Renderer.
      *

filament/include/filament/View.h

@@ -40,6 +40,7 @@ class CallbackHandler;
 
 class Camera;
 class ColorGrading;
+class Engine;
 class MaterialInstance;
 class RenderTarget;
 class Scene;
@@ -878,6 +879,17 @@ public:
      */
     utils::Entity getFogEntity() const noexcept;
 
+
+    /**
+     * When certain temporal features are used (e.g.: TAA or Screen-space reflections), the view
+     * keeps a history of previous frame renders associated with the Renderer the view was last
+     * used with. When switching Renderer, it may be necessary to clear that history by calling
+     * this method. Similarly, if the whole content of the screen change, like when a cut-scene
+     * starts, clearing the history might be needed to avoid artifacts due to the previous frame
+     * being very different.
+     */
+    void clearFrameHistory(Engine& engine) noexcept;
+
     /**
      * List of available ambient occlusion techniques
      * @deprecated use AmbientOcclusionOptions::enabled instead

filament/src/Engine.cpp

@@ -39,6 +39,8 @@
 #include <utils/compiler.h>
 #include <utils/Panic.h>
 
+#include <chrono>
+
 #include <stddef.h>
 #include <stdint.h>
 
@@ -361,7 +363,7 @@ const Engine::Config& Engine::getConfig() const noexcept {
     return downcast(this)->getConfig();
 }
 
-bool Engine::isStereoSupported(StereoscopicType stereoscopicType) const noexcept {
+bool Engine::isStereoSupported(StereoscopicType) const noexcept {
     return downcast(this)->isStereoSupported();
 }
 
@@ -369,6 +371,10 @@ size_t Engine::getMaxStereoscopicEyes() noexcept {
     return FEngine::getMaxStereoscopicEyes();
 }
 
+uint64_t Engine::getSteadyClockTimeNano() noexcept {
+    return std::chrono::steady_clock::now().time_since_epoch().count();
+}
+
 #if defined(__EMSCRIPTEN__)
 void Engine::resetBackendState() noexcept {
     downcast(this)->resetBackendState();

filament/src/FrameHistory.h

@@ -21,6 +21,9 @@
 #include <fg/FrameGraphTexture.h>
 
 #include <math/mat4.h>
+#include <math/vec2.h>
+
+#include <stdint.h>
 
 namespace filament {
 

filament/src/FrameInfo.cpp

@@ -16,30 +16,33 @@
 
 #include "FrameInfo.h"
 
+#include <filament/Renderer.h>
+
+#include <backend/DriverEnums.h>
+
+#include <utils/compiler.h>
+#include <utils/debug.h>
+#include <utils/FixedCapacityVector.h>
 #include <utils/Log.h>
-#include <utils/Systrace.h>
+#include <utils/ostream.h>
 
-#include <math/scalar.h>
+#include <algorithm>
+#include <atomic>
+#include <chrono>
+#include <memory>
+#include <ratio>
 
-#include <cmath>
+#include <stdint.h>
+#include <stddef.h>
 
 namespace filament {
 
 using namespace utils;
 using namespace backend;
 
-// this is to avoid a call to memmove
-template<class InputIterator, class OutputIterator>
-static inline
-void move_backward(InputIterator first, InputIterator last, OutputIterator result) {
-    while (first != last) {
-        *--result = *--last;
-    }
-}
-
 FrameInfoManager::FrameInfoManager(DriverApi& driver) noexcept {
     for (auto& query : mQueries) {
-        query = driver.createTimerQuery();
+        query.handle = driver.createTimerQuery();
     }
 }
 
@@ -47,14 +50,36 @@ FrameInfoManager::~FrameInfoManager() noexcept = default;
 
 void FrameInfoManager::terminate(DriverApi& driver) noexcept {
     for (auto& query : mQueries) {
-        driver.destroyTimerQuery(query);
+        driver.destroyTimerQuery(query.handle);
     }
 }
 
-void FrameInfoManager::beginFrame(DriverApi& driver,Config const& config, uint32_t) noexcept {
-    driver.beginTimerQuery(mQueries[mIndex]);
+void FrameInfoManager::beginFrame(DriverApi& driver, Config const& config, uint32_t frameId) noexcept {
+    auto& history = mFrameTimeHistory;
+    // don't exceed the capacity, drop the oldest entry
+    if (UTILS_LIKELY(history.size() == history.capacity())) {
+        history.pop_back();
+    }
+
+    // create a new entry
+    auto& front = history.emplace_front(frameId);
+
+    // store the current time
+    front.beginFrame = std::chrono::steady_clock::now();
+
+    // references are not invalidated by CircularQueue<>, so we can associate a reference to
+    // the slot we created to the timer query used to find the frame time.
+    mQueries[mIndex].pInfo = std::addressof(front);
+    // issue the timer query
+    driver.beginTimerQuery(mQueries[mIndex].handle);
+    // issue the custom backend command to get the backend time
+    driver.queueCommand([&front](){
+        front.backendBeginFrame = std::chrono::steady_clock::now();
+    });
+
+    // now is a good time to check the oldest active query
     uint64_t elapsed = 0;
-    TimerQueryResult const result = driver.getTimerQueryValue(mQueries[mLast], &elapsed);
+    TimerQueryResult const result = driver.getTimerQueryValue(mQueries[mLast].handle, &elapsed);
     switch (result) {
         case TimerQueryResult::NOT_READY:
             // nothing to do
@@ -63,49 +88,111 @@ void FrameInfoManager::beginFrame(DriverApi& driver,Config const& config, uint32
             mLast = (mLast + 1) % POOL_COUNT;
             break;
         case TimerQueryResult::AVAILABLE:
-            mLast = (mLast + 1) % POOL_COUNT;
             // conversion to our duration happens here
-            mFrameTime = std::chrono::duration<uint64_t, std::nano>(elapsed);
+            pFront = mQueries[mLast].pInfo;
+            pFront->frameTime = std::chrono::duration<uint64_t, std::nano>(elapsed);
+            mLast = (mLast + 1) % POOL_COUNT;
+            denoiseFrameTime(config);
             break;
     }
-    update(config, mFrameTime);
+
+
+    // keep this just for debugging
+    if constexpr (false) {
+        using namespace utils;
+        auto h = getFrameInfoHistory(1);
+        if (!h.empty()) {
+            slog.d << frameId << ": "
+                   << h[0].frameId << " (" << frameId - h[0].frameId << ")"
+                   << ", Dm=" << h[0].endFrame - h[0].beginFrame
+                   << ", L =" << h[0].backendBeginFrame - h[0].beginFrame
+                   << ", Db=" << h[0].backendEndFrame - h[0].backendBeginFrame
+                   << ", T =" << h[0].frameTime
+                   << io::endl;
+        }
+    }
 }
 
 void FrameInfoManager::endFrame(DriverApi& driver) noexcept {
-    driver.endTimerQuery(mQueries[mIndex]);
+    auto& front = mFrameTimeHistory.front();
+    // close the timer query
+    driver.endTimerQuery(mQueries[mIndex].handle);
+    // queue custom backend command to query the current time
+    driver.queueCommand([&front](){
+        // backend frame end-time
+        front.backendEndFrame = std::chrono::steady_clock::now();
+        // signal that the data is available
+        front.ready.store(true, std::memory_order_release);
+    });
+    // and finally acquire the time on the main thread
+    front.endFrame = std::chrono::steady_clock::now();
     mIndex = (mIndex + 1) % POOL_COUNT;
 }
 
-void FrameInfoManager::update(Config const& config,
-        FrameInfoManager::duration lastFrameTime) noexcept {
-    // keep an history of frame times
+void FrameInfoManager::denoiseFrameTime(Config const& config) noexcept {
     auto& history = mFrameTimeHistory;
+    assert_invariant(!history.empty());
 
-    // this is like doing { pop_back(); push_front(); }
-    filament::move_backward(history.begin(), history.end() - 1, history.end());
-    history[0].frameTime = lastFrameTime;
-
-    mFrameTimeHistorySize = std::min(++mFrameTimeHistorySize, uint32_t(MAX_FRAMETIME_HISTORY));
-    if (UTILS_UNLIKELY(mFrameTimeHistorySize < 3)) {
-        // not enough history to do anything useful
-        history[0].valid = false;
-        return;
+    // find the first slot that has a valid frame duration
+    size_t first = history.size();
+    for (size_t i = 0, c = history.size(); i < c; ++i) {
+        if (history[i].frameTime != duration(0)) {
+            assert_invariant(std::addressof(history[i]) == pFront);
+            first = i;
+            break;
+        }
     }
+    assert_invariant(first != history.size());
 
-    // apply a median filter to get a good representation of the frame time of the last
-    // N frames.
-    std::array<duration, MAX_FRAMETIME_HISTORY> median; // NOLINT -- it's initialized below
-    size_t const size = std::min(mFrameTimeHistorySize,
-            std::min(config.historySize, (uint32_t)median.size()));
-    for (size_t i = 0; i < size; ++i) {
-        median[i] = history[i].frameTime;
-    }
-    std::sort(median.begin(), median.begin() + size);
-    duration const denoisedFrameTime = median[size / 2];
+    // we need at least 3 valid frame time to calculate the median
+    if (history.size() >= first + 3) {
+        // apply a median filter to get a good representation of the frame time of the last
+        // N frames.
+        std::array<duration, MAX_FRAMETIME_HISTORY> median; // NOLINT -- it's initialized below
+        size_t const size = std::min({
+            history.size() - first,
+            median.size(),
+            size_t(config.historySize) });
 
-    history[0].denoisedFrameTime = denoisedFrameTime;
-    history[0].valid = true;
+        for (size_t i = 0; i < size; ++i) {
+            median[i] = history[first + i].frameTime;
+        }
+        std::sort(median.begin(), median.begin() + size);
+        duration const denoisedFrameTime = median[size / 2];
+
+        history[first].denoisedFrameTime = denoisedFrameTime;
+        history[first].valid = true;
+     }
 }
 
+utils::FixedCapacityVector<Renderer::FrameInfo> FrameInfoManager::getFrameInfoHistory(
+        size_t historySize) const noexcept {
+    auto result = utils::FixedCapacityVector<Renderer::FrameInfo>::with_capacity(MAX_FRAMETIME_HISTORY);
+    auto const& history = mFrameTimeHistory;
+    size_t i = 0;
+    size_t const c = history.size();
+    for (; i < c; ++i) {
+        auto const& entry = history[i];
+        if (entry.ready.load(std::memory_order_acquire) && entry.valid) {
+            // once we found an entry ready,
+            // we know by construction that all following ones are too
+            break;
+        }
+    }
+    for (; i < c && historySize; ++i, --historySize) {
+        auto const& entry = history[i];
+        using namespace std::chrono;
+        result.push_back({
+                entry.frameId,
+                duration_cast<nanoseconds>(entry.frameTime).count(),
+                duration_cast<nanoseconds>(entry.denoisedFrameTime).count(),
+                duration_cast<nanoseconds>(entry.beginFrame.time_since_epoch()).count(),
+                duration_cast<nanoseconds>(entry.endFrame.time_since_epoch()).count(),
+                duration_cast<nanoseconds>(entry.backendBeginFrame.time_since_epoch()).count(),
+                duration_cast<nanoseconds>(entry.backendEndFrame.time_since_epoch()).count()
+        });
+    }
+    return result;
+}
 
 } // namespace filament

filament/src/FrameInfo.h

@@ -17,30 +17,136 @@
 #ifndef TNT_FILAMENT_FRAMEINFO_H
 #define TNT_FILAMENT_FRAMEINFO_H
 
-#include "backend/Handle.h"
+#include <filament/Renderer.h>
+
+#include <backend/Handle.h>
+
 #include <private/backend/DriverApi.h>
 
+#include <utils/compiler.h>
+#include <utils/debug.h>
+#include <utils/FixedCapacityVector.h>
+
 #include <array>
+#include <atomic>
 #include <chrono>
+#include <ratio>
+#include <type_traits>
 
 #include <stdint.h>
+#include <stddef.h>
 
 namespace filament {
 class FEngine;
 
+namespace details {
 struct FrameInfo {
     using duration = std::chrono::duration<float, std::milli>;
     duration frameTime{};            // frame period
     duration denoisedFrameTime{};    // frame period (median filter)
-    bool valid = false;
+    bool valid = false;              // true if the data of the structure is valid
+};
+} // namespace details
+
+struct FrameInfoImpl : public details::FrameInfo {
+    using clock = std::chrono::steady_clock;
+    using time_point = clock::time_point;
+    uint32_t const frameId;
+    time_point beginFrame;           // main thread beginFrame time
+    time_point endFrame;             // main thread endFrame time
+    time_point backendBeginFrame;    // backend thread beginFrame time (makeCurrent time)
+    time_point backendEndFrame;      // backend thread endFrame time (present time)
+    std::atomic_bool ready{};        // true once backend thread has populated its data
+    explicit FrameInfoImpl(uint32_t frameId) noexcept
+        : frameId(frameId) {
+    }
+};
+
+template<typename T, size_t CAPACITY>
+class CircularQueue {
+public:
+    using value_type = T;
+    using reference = value_type&;
+    using const_reference = value_type const&;
+
+    size_t capacity() const {
+        return CAPACITY;
+    }
+
+    size_t size() const {
+        return mSize;
+    }
+
+    bool empty() const noexcept {
+        return !size();
+    }
+
+    void pop_back() noexcept {
+        assert_invariant(!empty());
+        --mSize;
+        std::destroy_at(&mStorage[(mFront - mSize) % CAPACITY]);
+    }
+
+    void push_front(T const& v) noexcept {
+        assert_invariant(size() < CAPACITY);
+        mFront = advance(mFront);
+        new(&mStorage[mFront]) T(v);
+        ++mSize;
+    }
+
+    void push_front(T&& v) noexcept {
+        assert_invariant(size() < CAPACITY);
+        mFront = advance(mFront);
+        new(&mStorage[mFront]) T(std::move(v));
+        ++mSize;
+    }
+
+    template<typename ...Args>
+    T& emplace_front(Args&&... args) noexcept {
+        assert_invariant(size() < CAPACITY);
+        mFront = advance(mFront);
+        new(&mStorage[mFront]) T(std::forward<Args>(args)...);
+        ++mSize;
+        return front();
+    }
+
+    T& operator[](size_t pos) noexcept {
+        assert_invariant(pos < size());
+        size_t const index = (mFront + CAPACITY - pos) % CAPACITY;
+        return *std::launder(reinterpret_cast<T*>(&mStorage[index]));
+    }
+
+    T const& operator[](size_t pos) const noexcept {
+        return const_cast<CircularQueue&>(*this)[pos];
+    }
+
+    T const& front() const noexcept {
+        assert_invariant(!empty());
+        return operator[](0);
+    }
+
+    T& front() noexcept {
+        assert_invariant(!empty());
+        return operator[](0);
+    }
+
+private:
+    using Storage = std::aligned_storage_t<sizeof(T), alignof(T)>;
+    Storage mStorage[CAPACITY];
+    uint32_t mFront = 0;    // always index 0
+    uint32_t mSize = 0;
+    [[nodiscard]] inline uint32_t advance(uint32_t v) noexcept {
+        return (v + 1) % CAPACITY;
+    }
 };
 
 class FrameInfoManager {
     static constexpr size_t POOL_COUNT = 4;
-    static constexpr size_t MAX_FRAMETIME_HISTORY = 31u;
+    static constexpr size_t MAX_FRAMETIME_HISTORY = 16u;
 
 public:
-    using duration = FrameInfo::duration;
+    using duration = FrameInfoImpl::duration;
+    using clock = FrameInfoImpl::clock;
 
     struct Config {
         uint32_t historySize;
@@ -57,23 +163,24 @@ public:
     // call this immediately before "swap buffers"
     void endFrame(backend::DriverApi& driver) noexcept;
 
-    FrameInfo const& getLastFrameInfo() const noexcept {
-        return mFrameTimeHistory[0];
+    details::FrameInfo const& getLastFrameInfo() const noexcept {
+        // if pFront is not set yet, return front() but in this case front().valid will be false
+        return pFront ? *pFront : mFrameTimeHistory.front();
     }
 
-    duration getLastFrameTime() const noexcept {
-        return getLastFrameInfo().frameTime;
-    }
+    utils::FixedCapacityVector<Renderer::FrameInfo> getFrameInfoHistory(size_t historySize) const noexcept;
 
 private:
-    void update(Config const& config, duration lastFrameTime) noexcept;
-    backend::Handle<backend::HwTimerQuery> mQueries[POOL_COUNT];
-    duration mFrameTime{};
-    uint32_t mIndex = 0;
-    uint32_t mLast = 0;
-
-    std::array<FrameInfo, MAX_FRAMETIME_HISTORY> mFrameTimeHistory;
-    uint32_t mFrameTimeHistorySize = 0;
+    void denoiseFrameTime(Config const& config) noexcept;
+    struct Query {
+        backend::Handle<backend::HwTimerQuery> handle{};
+        FrameInfoImpl* pInfo = nullptr;
+    };
+    std::array<Query, POOL_COUNT> mQueries;
+    uint32_t mIndex = 0;                // index of current query
+    uint32_t mLast = 0;                 // index of oldest query still active
+    FrameInfoImpl* pFront = nullptr;    // the most recent slot with a valid frame time
+    CircularQueue<FrameInfoImpl, MAX_FRAMETIME_HISTORY> mFrameTimeHistory;
 };
 
 

filament/src/Froxelizer.cpp

@@ -26,6 +26,7 @@
 #include <filament/Viewport.h>
 
 #include <utils/BinaryTreeArray.h>
+#include <utils/Log.h>
 #include <utils/Systrace.h>
 #include <utils/debug.h>
 

filament/src/MaterialInstance.cpp

@@ -183,6 +183,35 @@ template UTILS_PUBLIC void MaterialInstance::setParameter<mat4f>   (const char*
 
 // ------------------------------------------------------------------------------------------------
 
+template<typename T>
+T FMaterialInstance::getParameterImpl(std::string_view name) const {
+    ssize_t offset = mMaterial->getUniformInterfaceBlock().getFieldOffset(name, 0);
+    assert_invariant(offset>=0);
+    return downcast(this)->getUniformBuffer().getUniform<T>(offset);
+}
+
+template<typename T>
+T MaterialInstance::getParameter(const char* name, size_t nameLength) const {
+    return downcast(this)->getParameterImpl<T>({ name, nameLength });
+}
+
+// explicit template instantiation of our supported types
+template UTILS_PUBLIC float MaterialInstance::getParameter<float>        (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC int32_t MaterialInstance::getParameter<int32_t>  (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC uint32_t MaterialInstance::getParameter<uint32_t>(const char* name, size_t nameLength) const;
+template UTILS_PUBLIC int2 MaterialInstance::getParameter<int2>        (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC int3 MaterialInstance::getParameter<int3>        (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC int4 MaterialInstance::getParameter<int4>        (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC uint2 MaterialInstance::getParameter<uint2>      (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC uint3 MaterialInstance::getParameter<uint3>      (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC uint4 MaterialInstance::getParameter<uint4>      (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC float2 MaterialInstance::getParameter<float2>      (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC float3 MaterialInstance::getParameter<float3>      (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC float4 MaterialInstance::getParameter<float4>      (const char* name, size_t nameLength) const;
+template UTILS_PUBLIC mat3f MaterialInstance::getParameter<mat3f>      (const char* name, size_t nameLength) const;
+
+// ------------------------------------------------------------------------------------------------
+
 Material const* MaterialInstance::getMaterial() const noexcept {
     return downcast(this)->getMaterial();
 }

filament/src/PostProcessManager.cpp

@@ -1965,8 +1965,8 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloom(FrameGraph& fg,
         // - visible bloom size changes with dynamic resolution in non-homogenous mode
         // This allows us to use the 9 sample downsampling filter (instead of 13)
         // for at least 4 levels.
-        uint32_t width  = std::max(1u, uint32_t(std::floor(bloomWidth)));
-        uint32_t height = std::max(1u, uint32_t(std::floor(bloomHeight)));
+        uint32_t width  = std::max(16u, uint32_t(std::floor(bloomWidth)));
+        uint32_t height = std::max(16u, uint32_t(std::floor(bloomHeight)));
         width  &= ~((1 << 4) - 1);  // at least 4 levels
         height &= ~((1 << 4) - 1);
         bloomWidth  = float(width);
@@ -1978,6 +1978,8 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloom(FrameGraph& fg,
     // we don't need to do the fireflies reduction if we have TAA (it already does it)
     bool fireflies = threshold && !taaOptions.enabled;
 
+    assert_invariant(bloomWidth && bloomHeight);
+
     while (2 * bloomWidth < float(desc.width) || 2 * bloomHeight < float(desc.height)) {
         if (inoutBloomOptions.quality == QualityLevel::LOW ||
             inoutBloomOptions.quality == QualityLevel::MEDIUM) {

filament/src/RenderPass.cpp

@@ -38,6 +38,7 @@
 #include <backend/PipelineState.h>
 
 #include "private/backend/CircularBuffer.h"
+#include "private/backend/CommandStream.h"
 
 #include <utils/compiler.h>
 #include <utils/debug.h>
@@ -153,7 +154,7 @@ RenderPass::RenderPass(FEngine& engine, RenderPassBuilder const& builder) noexce
     sortCommands(builder.mArena);
 
     if (engine.isAutomaticInstancingEnabled()) {
-        uint32_t stereoscopicEyeCount = 1;
+        int32_t stereoscopicEyeCount = 1;
         if (builder.mFlags & IS_INSTANCED_STEREOSCOPIC) {
             stereoscopicEyeCount *= engine.getConfig().stereoscopicEyeCount;
         }
@@ -162,8 +163,7 @@ RenderPass::RenderPass(FEngine& engine, RenderPassBuilder const& builder) noexce
 }
 
 // this destructor is actually heavy because it inlines ~vector<>
-RenderPass::~RenderPass() noexcept {
-}
+RenderPass::~RenderPass() noexcept = default;
 
 void RenderPass::resize(Arena& arena, size_t count) noexcept {
     if (mCommandBegin) {
@@ -221,7 +221,7 @@ void RenderPass::appendCommands(FEngine& engine,
         work(vr.first, vr.size());
     } else {
         auto* jobCommandsParallel = jobs::parallel_for(js, nullptr, vr.first, (uint32_t)vr.size(),
-                std::cref(work), jobs::CountSplitter<JOBS_PARALLEL_FOR_COMMANDS_COUNT, 5>());
+                std::cref(work), jobs::CountSplitter<JOBS_PARALLEL_FOR_COMMANDS_COUNT>());
         js.runAndWait(jobCommandsParallel);
     }
 
@@ -671,7 +671,6 @@ RenderPass::Command* RenderPass::generateCommandsImpl(RenderPass::CommandTypeFla
          */
         for (size_t pi = 0, c = primitives.size(); pi < c; ++pi) {
             auto const& primitive = primitives[pi];
-            auto const& morphTargets = morphing.targets[pi];
             FMaterialInstance const* const mi = primitive.getMaterialInstance();
             FMaterial const* const ma = mi->getMaterial();
 
@@ -684,8 +683,9 @@ RenderPass::Command* RenderPass::generateCommandsImpl(RenderPass::CommandTypeFla
             cmd.info.indexOffset = primitive.getIndexOffset();
             cmd.info.indexCount = primitive.getIndexCount();
             cmd.info.type = primitive.getPrimitiveType();
-            cmd.info.morphTargetBuffer = morphTargets.buffer->getHwHandle();
-            cmd.info.morphingOffset = morphTargets.offset;
+            cmd.info.morphTargetBuffer = morphing.morphTargetBuffer ?
+                    morphing.morphTargetBuffer->getHwHandle() : SamplerGroupHandle{};
+            cmd.info.morphingOffset = primitive.getMorphingBufferOffset();
 
             if constexpr (isColorPass) {
                 RenderPass::setupColorCommand(cmd, renderableVariant, mi, inverseFrontFaces);
@@ -885,14 +885,31 @@ void RenderPass::Executor::execute(FEngine& engine,
 
         // Maximum space occupied in the CircularBuffer by a single `Command`. This must be
         // reevaluated when the inner loop below adds DriverApi commands or when we change the
-        // CommandStream protocol. Currently, the maximum is 240 bytes, and we use 256 to be on
-        // the safer side.
-        size_t const maxCommandSizeInBytes = 256;
+        // CommandStream protocol. Currently, the maximum is 320 bytes.
+        // The batch size is calculated by adding the size of all commands that can possibly be
+        // emitted per draw call:
+        constexpr size_t const maxCommandSizeInBytes =
+                sizeof(CustomCommand) +
+                sizeof(COMMAND_TYPE(scissor)) +
+                sizeof(COMMAND_TYPE(bindUniformBuffer)) +
+                sizeof(COMMAND_TYPE(bindSamplers)) +
+                sizeof(COMMAND_TYPE(bindBufferRange)) +
+                sizeof(COMMAND_TYPE(bindBufferRange)) +
+                sizeof(COMMAND_TYPE(bindSamplers)) +
+                sizeof(COMMAND_TYPE(bindSamplers)) +
+                sizeof(COMMAND_TYPE(bindUniformBuffer)) +
+                sizeof(COMMAND_TYPE(bindSamplers)) +
+                sizeof(COMMAND_TYPE(bindSamplers)) +
+                sizeof(COMMAND_TYPE(bindPipeline)) +
+                sizeof(COMMAND_TYPE(setPushConstant)) +
+                sizeof(COMMAND_TYPE(bindRenderPrimitive)) +
+                sizeof(COMMAND_TYPE(draw2));
+
 
         // Number of Commands that can be issued and guaranteed to fit in the current
         // CircularBuffer allocation. In practice, we'll have tons of headroom especially if
         // skinning and morphing aren't used. With a 2 MiB buffer (the default) a batch is
-        // 8192 commands (i.e. draw calls).
+        // 6553 commands (i.e. draw calls).
         size_t const batchCommandCount = capacity / maxCommandSizeInBytes;
         while(first != last) {
             Command const* const batchLast = std::min(first + batchCommandCount, last);

filament/src/RenderPass.h

@@ -417,8 +417,7 @@ private:
     void instanceify(FEngine& engine, Arena& arena, int32_t eyeCount) noexcept;
 
     // We choose the command count per job to minimize JobSystem overhead.
-    // On a Pixel 4, 2048 commands is about half a millisecond of processing.
-    static constexpr size_t JOBS_PARALLEL_FOR_COMMANDS_COUNT = 2048;
+    static constexpr size_t JOBS_PARALLEL_FOR_COMMANDS_COUNT = 128;
     static constexpr size_t JOBS_PARALLEL_FOR_COMMANDS_SIZE  =
             sizeof(Command) * JOBS_PARALLEL_FOR_COMMANDS_COUNT;
 

filament/src/RenderPrimitive.h

@@ -56,6 +56,7 @@ public:
     backend::VertexBufferInfoHandle getVertexBufferInfoHandle() const { return mVertexBufferInfoHandle; }
     uint32_t getIndexOffset() const noexcept { return mIndexOffset; }
     uint32_t getIndexCount() const noexcept { return mIndexCount; }
+    uint32_t getMorphingBufferOffset() const noexcept { return mMorphingBufferOffset; }
 
     backend::PrimitiveType getPrimitiveType() const noexcept { return mPrimitiveType; }
     AttributeBitset getEnabledAttributes() const noexcept { return mEnabledAttributes; }
@@ -72,6 +73,10 @@ public:
         mGlobalBlendOrderEnabled = enabled;
     }
 
+    void setMorphingBufferOffset(uint32_t offset) noexcept {
+        mMorphingBufferOffset = offset;
+    }
+
 private:
     // These first fields are dereferences from PrimitiveInfo, keep them together
     struct {
@@ -80,6 +85,7 @@ private:
         backend::Handle<backend::HwVertexBufferInfo> mVertexBufferInfoHandle = {};
         uint32_t mIndexOffset = 0;
         uint32_t mIndexCount = 0;
+        uint32_t mMorphingBufferOffset = 0;
     };
 
     AttributeBitset mEnabledAttributes = {};

filament/src/RenderableManager.cpp

@@ -158,20 +158,14 @@ void RenderableManager::setMorphWeights(Instance instance, float const* weights,
     downcast(this)->setMorphWeights(instance, weights, count, offset);
 }
 
-void RenderableManager::setMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex,
-        MorphTargetBuffer* morphTargetBuffer, size_t offset, size_t count) {
-    downcast(this)->setMorphTargetBufferAt(instance, level, primitiveIndex,
-            downcast(morphTargetBuffer), offset, count);
+void RenderableManager::setMorphTargetBufferOffsetAt(Instance instance, uint8_t level,
+        size_t primitiveIndex,
+        size_t offset) {
+    downcast(this)->setMorphTargetBufferOffsetAt(instance, level, primitiveIndex, offset);
 }
 
-void RenderableManager::setMorphTargetBufferAt(
-        Instance instance, uint8_t level, size_t primitiveIndex, size_t offset, size_t count) {
-    downcast(this)->setMorphTargetBufferAt(instance, level, primitiveIndex, offset, count);
-}
-
-MorphTargetBuffer* RenderableManager::getMorphTargetBufferAt(Instance instance, uint8_t level,
-        size_t primitiveIndex) const noexcept {
-    return downcast(this)->getMorphTargetBufferAt(instance, level, primitiveIndex);
+MorphTargetBuffer* RenderableManager::getMorphTargetBuffer(Instance instance) const noexcept {
+    return downcast(this)->getMorphTargetBuffer(instance);
 }
 
 size_t RenderableManager::getMorphTargetCount(Instance instance) const noexcept {

filament/src/Renderer.cpp

@@ -14,11 +14,20 @@
  * limitations under the License.
  */
 
+#include <filament/Renderer.h>
+
 #include "details/Renderer.h"
 
 #include "details/Engine.h"
 #include "details/View.h"
 
+#include <utils/FixedCapacityVector.h>
+
+#include <utility>
+
+#include <stddef.h>
+#include <stdint.h>
+
 namespace filament {
 
 using namespace math;
@@ -36,6 +45,10 @@ void Renderer::setPresentationTime(int64_t monotonic_clock_ns) {
     downcast(this)->setPresentationTime(monotonic_clock_ns);
 }
 
+void Renderer::skipFrame(uint64_t vsyncSteadyClockTimeNano) {
+    downcast(this)->skipFrame(vsyncSteadyClockTimeNano);
+}
+
 bool Renderer::beginFrame(SwapChain* swapChain, uint64_t vsyncSteadyClockTimeNano) {
     return downcast(this)->beginFrame(downcast(swapChain), vsyncSteadyClockTimeNano);
 }
@@ -89,4 +102,16 @@ void Renderer::renderStandaloneView(View const* view) {
     downcast(this)->renderStandaloneView(downcast(view));
 }
 
+void Renderer::setVsyncTime(uint64_t steadyClockTimeNano) noexcept {
+    downcast(this)->setVsyncTime(steadyClockTimeNano);
+}
+
+utils::FixedCapacityVector<Renderer::FrameInfo> Renderer::getFrameInfoHistory(size_t historySize) const noexcept {
+    return downcast(this)->getFrameInfoHistory(historySize);
+}
+
+size_t Renderer::getMaxFrameHistorySize() const noexcept {
+    return downcast(this)->getMaxFrameHistorySize();
+}
+
 } // namespace filament

filament/src/RendererUtils.cpp

@@ -172,7 +172,7 @@ FrameGraphId<FrameGraphTexture> RendererUtils::colorPass(
 
                 data.color = builder.write(data.color, FrameGraphTexture::Usage::COLOR_ATTACHMENT);
                 data.depth = builder.write(data.depth, FrameGraphTexture::Usage::DEPTH_ATTACHMENT);
-                if (engine.getConfig().stereoscopicType == StereoscopicType::MULTIVIEW) {
+                if (view.hasStereo() && engine.getConfig().stereoscopicType == StereoscopicType::MULTIVIEW) {
                     layerCount = engine.getConfig().stereoscopicEyeCount;
                 }
 

filament/src/ResourceAllocator.cpp

@@ -27,15 +27,18 @@
 
 #include "private/backend/DriverApi.h"
 
+#include <utils/algorithm.h>
+#include <utils/bitset.h>
 #include <utils/compiler.h>
 #include <utils/debug.h>
-#include <utils/FixedCapacityVector.h>
 #include <utils/Log.h>
 #include <utils/ostream.h>
 
 #include <array>
 #include <algorithm>
 #include <iterator>
+#include <memory>
+#include <optional>
 #include <utility>
 
 #include <stddef.h>
@@ -89,8 +92,15 @@ void ResourceAllocator::AssociativeContainer<K, V, H>::emplace(ARGS&& ... args)
 }
 
 // ------------------------------------------------------------------------------------------------
+
 ResourceAllocatorInterface::~ResourceAllocatorInterface() = default;
 
+// ------------------------------------------------------------------------------------------------
+
+ResourceAllocatorDisposerInterface::~ResourceAllocatorDisposerInterface() = default;
+
+// ------------------------------------------------------------------------------------------------
+
 size_t ResourceAllocator::TextureKey::getSize() const noexcept {
     size_t const pixelCount = width * height * depth;
     size_t size = pixelCount * FTexture::getFormatSize(format);
@@ -110,16 +120,22 @@ size_t ResourceAllocator::TextureKey::getSize() const noexcept {
 
 ResourceAllocator::ResourceAllocator(Engine::Config const& config, DriverApi& driverApi) noexcept
         : mCacheMaxAge(config.resourceAllocatorCacheMaxAge),
-          mBackend(driverApi) {
+          mBackend(driverApi),
+          mDisposer(std::make_shared<ResourceAllocatorDisposer>(driverApi)) {
+}
+
+ResourceAllocator::ResourceAllocator(std::shared_ptr<ResourceAllocatorDisposer> disposer,
+        Engine::Config const& config, DriverApi& driverApi) noexcept
+        : mCacheMaxAge(config.resourceAllocatorCacheMaxAge),
+          mBackend(driverApi),
+          mDisposer(std::move(disposer)) {
 }
 
 ResourceAllocator::~ResourceAllocator() noexcept {
-    assert_invariant(!mTextureCache.size());
-    assert_invariant(!mInUseTextures.size());
+    assert_invariant(mTextureCache.empty());
 }
 
 void ResourceAllocator::terminate() noexcept {
-    assert_invariant(!mInUseTextures.size());
     auto& textureCache = mTextureCache;
     for (auto it = textureCache.begin(); it != textureCache.end();) {
         mBackend.destroyTexture(it->second.handle);
@@ -174,7 +190,7 @@ backend::TextureHandle ResourceAllocator::createTexture(const char* name,
                         swizzle[0], swizzle[1], swizzle[2], swizzle[3]);
             }
         }
-        mInUseTextures.emplace(handle, key);
+        mDisposer->checkout(handle, key);
     } else {
         if (swizzle == defaultSwizzle) {
             handle = mBackend.createTexture(
@@ -190,58 +206,99 @@ backend::TextureHandle ResourceAllocator::createTexture(const char* name,
 
 void ResourceAllocator::destroyTexture(TextureHandle h) noexcept {
     if constexpr (mEnabled) {
-        // find the texture in the in-use list (it must be there!)
-        auto it = mInUseTextures.find(h);
-        assert_invariant(it != mInUseTextures.end());
-
-        // move it to the cache
-        const TextureKey key = it->second;
-        uint32_t const size = key.getSize();
-
-        mTextureCache.emplace(key, TextureCachePayload{ h, mAge, size });
-        mCacheSize += size;
-
-        // remove it from the in-use list
-        mInUseTextures.erase(it);
+        auto const key = mDisposer->checkin(h);
+        if (UTILS_LIKELY(key.has_value())) {
+            uint32_t const size = key.value().getSize();
+            mTextureCache.emplace(key.value(), TextureCachePayload{ h, mAge, size });
+            mCacheSize += size;
+            mCacheSizeHiWaterMark = std::max(mCacheSizeHiWaterMark, mCacheSize);
+        }
     } else {
         mBackend.destroyTexture(h);
     }
 }
 
-void ResourceAllocator::gc() noexcept {
-    // this is called regularly -- usually once per frame of each Renderer
+ResourceAllocatorDisposerInterface& ResourceAllocator::getDisposer() noexcept {
+    return *mDisposer;
+}
 
-    // increase our age
-    const size_t age = mAge++;
+void ResourceAllocator::gc(bool skippedFrame) noexcept {
+    // this is called regularly -- usually once per frame
+
+    // increase our age at each (non-skipped) frame
+    const size_t age = mAge;
+    if (!skippedFrame) {
+        mAge++;
+    }
 
     // Purging strategy:
-    //  - remove entries that are older than a certain age
-    //      - remove only one entry per gc(),
+    //  - remove all entries older than MAX_AGE_SKIPPED_FRAME when skipping a frame
+    //  - remove entries older than mCacheMaxAgeSoft
+    //      - remove only MAX_EVICTION_COUNT entry per gc(),
+    //  - look for the number of unique resource ages present in the cache (this basically gives
+    //    us how many buckets of resources we have corresponding to previous frames.
+    //      - remove all resources that have an age older than the MAX_UNIQUE_AGE_COUNT'th bucket
 
     auto& textureCache = mTextureCache;
+
+    // when skipping a frame, the maximum age to keep in the cache
+    constexpr size_t MAX_AGE_SKIPPED_FRAME = 1;
+
+    // maximum entry count to evict per GC, under the mCacheMaxAgeSoft limit
+    constexpr size_t MAX_EVICTION_COUNT = 1;
+
+    // maximum number of unique ages in the cache
+    constexpr size_t MAX_UNIQUE_AGE_COUNT = 3;
+
+    utils::bitset32 ages;
+    uint32_t evictedCount = 0;
     for (auto it = textureCache.begin(); it != textureCache.end();) {
-        const size_t ageDiff = age - it->second.age;
-        if (ageDiff >= mCacheMaxAge) {
+        size_t const ageDiff = age - it->second.age;
+        if ((ageDiff >= MAX_AGE_SKIPPED_FRAME && skippedFrame) ||
+            (ageDiff >= mCacheMaxAge && evictedCount < MAX_EVICTION_COUNT)) {
+            evictedCount++;
             purge(it);
-            // only purge a single entry per gc
-            break;
         } else {
+            // build the set of ages present in the cache after eviction
+            ages.set(std::min(size_t(31), ageDiff));
             ++it;
         }
     }
+
+    // if we have MAX_UNIQUE_AGE_COUNT ages or more, we evict all the resources that
+    // are older than the MAX_UNIQUE_AGE_COUNT'th age.
+    if (!skippedFrame && ages.count() >= MAX_UNIQUE_AGE_COUNT) {
+        uint32_t bits = ages.getValue();
+        // remove from the set the ages we keep
+        for (size_t i = 0; i < MAX_UNIQUE_AGE_COUNT - 1; i++) {
+            bits &= ~(1 << utils::ctz(bits));
+        }
+        size_t const maxAge = utils::ctz(bits);
+        for (auto it = textureCache.begin(); it != textureCache.end();) {
+            const size_t ageDiff = age - it->second.age;
+            if (ageDiff >= maxAge) {
+                purge(it);
+            } else {
+                ++it;
+            }
+        }
+    }
 }
 
 UTILS_NOINLINE
 void ResourceAllocator::dump(bool brief) const noexcept {
-    slog.d << "# entries=" << mTextureCache.size() << ", sz=" << mCacheSize / float(1u << 20u)
-           << " MiB" << io::endl;
+    constexpr float MiB = 1.0f / float(1u << 20u);
+    slog.d  << "# entries=" << mTextureCache.size()
+            << ", sz=" << (float)mCacheSize * MiB << " MiB"
+            << ", max=" << (float)mCacheSizeHiWaterMark * MiB << " MiB"
+            << io::endl;
     if (!brief) {
         for (auto const& it : mTextureCache) {
             auto w = it.first.width;
             auto h = it.first.height;
             auto f = FTexture::getFormatSize(it.first.format);
             slog.d << it.first.name << ": w=" << w << ", h=" << h << ", f=" << f << ", sz="
-                   << it.second.size / float(1u << 20u) << io::endl;
+                   << (float)it.second.size * MiB << io::endl;
         }
     }
 }
@@ -254,4 +311,46 @@ void ResourceAllocator::purge(
     mTextureCache.erase(pos);
 }
 
+// ------------------------------------------------------------------------------------------------
+
+ResourceAllocatorDisposer::ResourceAllocatorDisposer(DriverApi& driverApi) noexcept
+        : mBackend(driverApi) {
+}
+
+ResourceAllocatorDisposer::~ResourceAllocatorDisposer() noexcept {
+     assert_invariant(mInUseTextures.empty());
+}
+
+void ResourceAllocatorDisposer::terminate() noexcept {
+    assert_invariant(mInUseTextures.empty());
+}
+
+void ResourceAllocatorDisposer::destroy(backend::TextureHandle handle) noexcept {
+    if (handle) {
+        auto r = checkin(handle);
+        if (r.has_value()) {
+            mBackend.destroyTexture(handle);
+        }
+    }
+}
+
+void ResourceAllocatorDisposer::checkout(backend::TextureHandle handle,
+        ResourceAllocator::TextureKey key) {
+    mInUseTextures.emplace(handle, key);
+}
+
+std::optional<ResourceAllocator::TextureKey> ResourceAllocatorDisposer::checkin(
+        backend::TextureHandle handle) {
+    // find the texture in the in-use list (it must be there!)
+    auto it = mInUseTextures.find(handle);
+    assert_invariant(it != mInUseTextures.end());
+    if (it == mInUseTextures.end()) {
+        return std::nullopt;
+    }
+    TextureKey const key = it->second;
+    // remove it from the in-use list
+    mInUseTextures.erase(it);
+    return key;
+}
+
 } // namespace filament

filament/src/ResourceAllocator.h

@@ -28,16 +28,29 @@
 #include <utils/Hash.h>
 
 #include <array>
-#include <vector>
+#include <memory>
+#include <optional>
 #include <utility>
+#include <vector>
 
+#include <cstddef>
 #include <stddef.h>
 #include <stdint.h>
 
 namespace filament {
 
+class ResourceAllocatorDisposer;
+
 // The only reason we use an interface here is for unit-tests, so we can mock this allocator.
 // This is not too time-critical, so that's okay.
+
+class ResourceAllocatorDisposerInterface {
+public:
+    virtual void destroy(backend::TextureHandle handle) noexcept = 0;
+protected:
+    virtual ~ResourceAllocatorDisposerInterface();
+};
+
 class ResourceAllocatorInterface {
 public:
     virtual backend::RenderTargetHandle createRenderTarget(const char* name,
@@ -60,15 +73,20 @@ public:
 
     virtual void destroyTexture(backend::TextureHandle h) noexcept = 0;
 
+    virtual ResourceAllocatorDisposerInterface& getDisposer() noexcept = 0;
+
 protected:
     virtual ~ResourceAllocatorInterface();
 };
 
-
 class ResourceAllocator final : public ResourceAllocatorInterface {
 public:
+    explicit ResourceAllocator(std::shared_ptr<ResourceAllocatorDisposer> disposer,
+            Engine::Config const& config, backend::DriverApi& driverApi) noexcept;
+
     explicit ResourceAllocator(
             Engine::Config const& config, backend::DriverApi& driverApi) noexcept;
+
     ~ResourceAllocator() noexcept override;
 
     void terminate() noexcept;
@@ -93,7 +111,9 @@ public:
 
     void destroyTexture(backend::TextureHandle h) noexcept override;
 
-    void gc() noexcept;
+    ResourceAllocatorDisposerInterface& getDisposer() noexcept override;
+
+    void gc(bool skippedFrame = false) noexcept;
 
 private:
     size_t const mCacheMaxAge;
@@ -181,6 +201,7 @@ private:
         using value_type = typename Container::value_type::second_type;
 
         size_t size() const { return mContainer.size(); }
+        bool empty() const { return size() == 0; }
         iterator begin() { return mContainer.begin(); }
         const_iterator begin() const { return mContainer.begin(); }
         iterator end() { return mContainer.end(); }
@@ -193,16 +214,36 @@ private:
     };
 
     using CacheContainer = AssociativeContainer<TextureKey, TextureCachePayload>;
-    using InUseContainer = AssociativeContainer<backend::TextureHandle, TextureKey>;
 
     void purge(ResourceAllocator::CacheContainer::iterator const& pos);
 
     backend::DriverApi& mBackend;
+    std::shared_ptr<ResourceAllocatorDisposer> mDisposer;
     CacheContainer mTextureCache;
-    InUseContainer mInUseTextures;
     size_t mAge = 0;
     uint32_t mCacheSize = 0;
+    uint32_t mCacheSizeHiWaterMark = 0;
     static constexpr bool mEnabled = true;
+
+    friend class ResourceAllocatorDisposer;
+};
+
+class ResourceAllocatorDisposer final : public ResourceAllocatorDisposerInterface {
+    using TextureKey = ResourceAllocator::TextureKey;
+public:
+    explicit ResourceAllocatorDisposer(backend::DriverApi& driverApi) noexcept;
+    ~ResourceAllocatorDisposer() noexcept override;
+    void terminate() noexcept;
+    void destroy(backend::TextureHandle handle) noexcept override;
+
+private:
+    friend class ResourceAllocator;
+    void checkout(backend::TextureHandle handle, TextureKey key);
+    std::optional<TextureKey> checkin(backend::TextureHandle handle);
+
+    using InUseContainer = ResourceAllocator::AssociativeContainer<backend::TextureHandle, TextureKey>;
+    backend::DriverApi& mBackend;
+    InUseContainer mInUseTextures;
 };
 
 } // namespace filament

filament/src/View.cpp

@@ -15,6 +15,8 @@
  */
 
 #include "details/View.h"
+#include "filament/View.h"
+
 
 namespace filament {
 
@@ -312,4 +314,8 @@ utils::Entity View::getFogEntity() const noexcept {
     return downcast(this)->getFogEntity();
 }
 
+void View::clearFrameHistory(Engine& engine) noexcept {
+    downcast(this)->clearFrameHistory(downcast(engine));
+}
+
 } // namespace filament

filament/src/components/LightManager.cpp

@@ -20,12 +20,13 @@
 
 #include "details/Engine.h"
 
-#include <math/fast.h>
-#include <math/scalar.h>
-
-#include <utils/debug.h>
 #include <filament/LightManager.h>
 
+#include <utils/debug.h>
+#include <utils/Log.h>
+
+#include <math/fast.h>
+#include <math/scalar.h>
 
 using namespace filament::math;
 using namespace utils;

filament/src/components/RenderableManager.cpp

@@ -281,19 +281,13 @@ RenderableManager::Builder& RenderableManager::Builder::morphing(
     return *this;
 }
 
-RenderableManager::Builder& RenderableManager::Builder::morphing(
-        uint8_t level, size_t primitiveIndex, size_t offset, size_t count) noexcept {
-    return morphing(level, primitiveIndex, mImpl->mMorphTargetBuffer, offset, count);
-}
-
-RenderableManager::Builder& RenderableManager::Builder::morphing(uint8_t, size_t primitiveIndex,
-        MorphTargetBuffer* morphTargetBuffer, size_t offset, size_t count) noexcept {
-    std::vector<Entry>& entries = mImpl->mEntries;
+RenderableManager::Builder& RenderableManager::Builder::morphing(uint8_t level,
+        size_t primitiveIndex, size_t offset) noexcept {
+    // the last parameter "count" is unused, because it must be equal to the primitive's vertex count
+    std::vector<BuilderDetails::Entry>& entries = mImpl->mEntries;
     if (primitiveIndex < entries.size()) {
         auto& morphing = entries[primitiveIndex].morphing;
-        morphing.buffer = morphTargetBuffer;
-        morphing.offset = offset;
-        morphing.count = count;
+        morphing.offset = uint32_t(offset);
     }
     return *this;
 }
@@ -449,10 +443,10 @@ RenderableManager::Builder::Result RenderableManager::Builder::build(Engine& eng
             << ") instances when supplying transforms via an InstanceBuffer.";
 
     if (mImpl->mGeometryType == GeometryType::STATIC) {
-        FILAMENT_CHECK_PRECONDITION(mImpl->mSkinningBoneCount > 0)
+        FILAMENT_CHECK_PRECONDITION(mImpl->mSkinningBoneCount == 0)
                 << "Skinning can't be used with STATIC geometry";
 
-        FILAMENT_CHECK_PRECONDITION(mImpl->mMorphTargetCount > 0)
+        FILAMENT_CHECK_PRECONDITION(mImpl->mMorphTargetCount == 0)
                 << "Morphing can't be used with STATIC geometry";
     }
 
@@ -668,13 +662,6 @@ void FRenderableManager::create(
         if (morphTargetBuffer == nullptr) {
             morphTargetBuffer = mEngine.getDummyMorphTargetBuffer();
         }
-        MorphTargets* const morphTargets = new MorphTargets[entryCount];
-        std::generate_n(morphTargets, entryCount,
-                [morphTargetBuffer]() -> MorphTargets {
-                    return { morphTargetBuffer, 0, 0 };
-                });
-
-        mManager[ci].morphTargets = { morphTargets, size_type(entryCount) };
 
         // Always create skinning and morphing resources if one of them is enabled because
         // the shader always handles both. See Variant::SKINNING_OR_MORPHING.
@@ -700,13 +687,13 @@ void FRenderableManager::create(
                         backend::BufferUsage::DYNAMIC),
                 .count = targetCount };
 
-            for (size_t i = 0; i < entryCount; ++i) {
-                const auto& morphing = builder->mEntries[i].morphing;
-                if (!morphing.buffer) {
-                    continue;
+            Slice<FRenderPrimitive>& primitives = mManager[ci].primitives;
+            mManager[ci].morphTargetBuffer = morphTargetBuffer;
+            if (builder->mMorphTargetBuffer) {
+                for (size_t i = 0; i < entryCount; ++i) {
+                    const auto& morphing = builder->mEntries[i].morphing;
+                    primitives[i].setMorphingBufferOffset(morphing.offset);
                 }
-                morphTargets[i] = { downcast(morphing.buffer), (uint32_t)morphing.offset,
-                                    (uint32_t)morphing.count };
             }
             
             // When targetCount equal 0, boneCount>0 in this case, do an initialization for the
@@ -762,7 +749,6 @@ void FRenderableManager::destroyComponent(Instance ci) noexcept {
 
     // See create(RenderableManager::Builder&, Entity)
     destroyComponentPrimitives(mHwRenderPrimitiveFactory, driver, manager[ci].primitives);
-    destroyComponentMorphTargets(engine, manager[ci].morphTargets);
 
     // destroy the bones structures if any
     Bones const& bones = manager[ci].bones;
@@ -795,11 +781,6 @@ void FRenderableManager::destroyComponentPrimitives(
     delete[] primitives.data();
 }
 
-void FRenderableManager::destroyComponentMorphTargets(FEngine&,
-        utils::Slice<MorphTargets>& morphTargets) noexcept {
-    delete[] morphTargets.data();
-}
-
 void FRenderableManager::setMaterialInstanceAt(Instance instance, uint8_t level,
         size_t primitiveIndex, FMaterialInstance const* mi) {
     if (instance) {
@@ -968,42 +949,21 @@ void FRenderableManager::setMorphWeights(Instance instance, float const* weights
     }
 }
 
-void FRenderableManager::setMorphTargetBufferAt(Instance instance, uint8_t level,
-        size_t primitiveIndex, FMorphTargetBuffer* morphTargetBuffer, size_t offset, size_t count) {
+void FRenderableManager::setMorphTargetBufferOffsetAt(Instance instance, uint8_t level,
+        size_t primitiveIndex,
+        size_t offset) {
     if (instance) {
-        assert_invariant(morphTargetBuffer);
-
-        MorphWeights const& morphWeights = mManager[instance].morphWeights;
-        FILAMENT_CHECK_PRECONDITION(morphWeights.count == count)
-                << "Only " << morphWeights.count << " morph targets can be set (count=" << count
-                << ")";
-
-        Slice<MorphTargets>& morphTargets = getMorphTargets(instance, level);
-        if (primitiveIndex < morphTargets.size()) {
-            morphTargets[primitiveIndex] = { morphTargetBuffer, (uint32_t)offset,
-                                             (uint32_t)count };
+        assert_invariant(mManager[instance].morphTargetBuffer);
+        Slice<FRenderPrimitive>& primitives = mManager[instance].primitives;
+        if (primitiveIndex < primitives.size()) {
+            primitives[primitiveIndex].setMorphingBufferOffset(offset);
         }
     }
 }
 
-void FRenderableManager::setMorphTargetBufferAt(Instance instance, uint8_t level,
-        size_t primitiveIndex, size_t offset, size_t count) {
+MorphTargetBuffer* FRenderableManager::getMorphTargetBuffer(Instance instance) const noexcept {
     if (instance) {
-        Slice<MorphTargets>& morphTargets = getMorphTargets(instance, level);
-        if (primitiveIndex < morphTargets.size()) {
-            setMorphTargetBufferAt(instance, level,
-                    primitiveIndex, morphTargets[primitiveIndex].buffer, offset, count);
-        }
-    }
-}
-
-MorphTargetBuffer* FRenderableManager::getMorphTargetBufferAt(Instance instance, uint8_t level,
-        size_t primitiveIndex) const noexcept {
-    if (instance) {
-        const Slice<MorphTargets>& morphTargets = getMorphTargets(instance, level);
-        if (primitiveIndex < morphTargets.size()) {
-            return morphTargets[primitiveIndex].buffer;
-        }
+        return mManager[instance].morphTargetBuffer;
     }
     return nullptr;
 }

filament/src/components/RenderableManager.h

@@ -80,12 +80,6 @@ public:
 
     static_assert(sizeof(Visibility) == sizeof(uint16_t), "Visibility should be 16 bits");
 
-    struct MorphTargets {
-        FMorphTargetBuffer* buffer = nullptr;
-        uint32_t offset = 0;
-        uint32_t count = 0;
-    };
-
     explicit FRenderableManager(FEngine& engine) noexcept;
     ~FRenderableManager();
 
@@ -155,11 +149,9 @@ public:
 
     inline void setMorphing(Instance instance, bool enable);
     void setMorphWeights(Instance instance, float const* weights, size_t count, size_t offset);
-    void setMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex,
-            FMorphTargetBuffer* morphTargetBuffer, size_t offset, size_t count);
-    void setMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex,
-            size_t offset, size_t count);
-    MorphTargetBuffer* getMorphTargetBufferAt(Instance instance, uint8_t level, size_t primitiveIndex) const noexcept;
+    void setMorphTargetBufferOffsetAt(Instance instance, uint8_t level, size_t primitiveIndex,
+            size_t offset);
+    MorphTargetBuffer* getMorphTargetBuffer(Instance instance) const noexcept;
     size_t getMorphTargetCount(Instance instance) const noexcept;
 
     void setLightChannel(Instance instance, unsigned int channel, bool enable) noexcept;
@@ -189,7 +181,7 @@ public:
     struct MorphingBindingInfo {
         backend::Handle<backend::HwBufferObject> handle;
         uint32_t count;
-        MorphTargets const* targets; // Pointer to Slice<MorphTargets> at a renderable.
+        FMorphTargetBuffer const* morphTargetBuffer;
     };
     inline MorphingBindingInfo getMorphingBufferInfo(Instance instance) const noexcept;
 
@@ -218,16 +210,12 @@ public:
     AttributeBitset getEnabledAttributesAt(Instance instance, uint8_t level, size_t primitiveIndex) const noexcept;
     inline utils::Slice<FRenderPrimitive> const& getRenderPrimitives(Instance instance, uint8_t level) const noexcept;
     inline utils::Slice<FRenderPrimitive>& getRenderPrimitives(Instance instance, uint8_t level) noexcept;
-    inline utils::Slice<MorphTargets> const& getMorphTargets(Instance instance, uint8_t level) const noexcept;
-    inline utils::Slice<MorphTargets>& getMorphTargets(Instance instance, uint8_t level) noexcept;
 
 private:
     void destroyComponent(Instance ci) noexcept;
     static void destroyComponentPrimitives(
             HwRenderPrimitiveFactory& factory, backend::DriverApi& driver,
             utils::Slice<FRenderPrimitive>& primitives) noexcept;
-    static void destroyComponentMorphTargets(FEngine& engine,
-            utils::Slice<MorphTargets>& morphTargets) noexcept;
 
     struct Bones {
         backend::Handle<backend::HwBufferObject> handle;
@@ -254,7 +242,7 @@ private:
         VISIBILITY,             // user data
         PRIMITIVES,             // user data
         BONES,                  // filament data, UBO storing a pointer to the bones information
-        MORPH_TARGETS
+        MORPHTARGET_BUFFER      // morphtarget buffer for the component
     };
 
     using Base = utils::SingleInstanceComponentManager<
@@ -266,7 +254,7 @@ private:
             Visibility,                      // VISIBILITY
             utils::Slice<FRenderPrimitive>,  // PRIMITIVES
             Bones,                           // BONES
-            utils::Slice<MorphTargets>       // MORPH_TARGETS
+            FMorphTargetBuffer*              // MORPHTARGET_BUFFER
     >;
 
     struct Sim : public Base {
@@ -289,7 +277,7 @@ private:
                 Field<VISIBILITY>           visibility;
                 Field<PRIMITIVES>           primitives;
                 Field<BONES>                bones;
-                Field<MORPH_TARGETS>        morphTargets;
+                Field<MORPHTARGET_BUFFER>   morphTargetBuffer;
             };
         };
 
@@ -461,8 +449,8 @@ inline uint32_t FRenderableManager::getBoneCount(Instance instance) const noexce
 FRenderableManager::MorphingBindingInfo
 FRenderableManager::getMorphingBufferInfo(Instance instance) const noexcept {
     MorphWeights const& morphWeights = mManager[instance].morphWeights;
-    utils::Slice<MorphTargets> const& morphTargets = getMorphTargets(instance, 0);
-    return { morphWeights.handle, morphWeights.count, morphTargets.data() };
+    FMorphTargetBuffer const* const buffer = mManager[instance].morphTargetBuffer;
+    return { morphWeights.handle, morphWeights.count, buffer  };
 }
 
 FRenderableManager::InstancesInfo
@@ -480,16 +468,6 @@ utils::Slice<FRenderPrimitive>& FRenderableManager::getRenderPrimitives(
     return mManager[instance].primitives;
 }
 
-utils::Slice<FRenderableManager::MorphTargets> const& FRenderableManager::getMorphTargets(
-        Instance instance, UTILS_UNUSED uint8_t level) const noexcept {
-    return mManager[instance].morphTargets;
-}
-
-utils::Slice<FRenderableManager::MorphTargets>& FRenderableManager::getMorphTargets(
-        Instance instance, UTILS_UNUSED uint8_t level) noexcept {
-    return mManager[instance].morphTargets;
-}
-
 } // namespace filament
 
 #endif // TNT_FILAMENT_COMPONENTS_RENDERABLEMANAGER_H

filament/src/details/Engine.cpp

@@ -102,6 +102,7 @@ Engine* FEngine::create(Engine::Builder const& builder) {
         DriverConfig const driverConfig{
                 .handleArenaSize = instance->getRequestedDriverHandleArenaSize(),
                 .textureUseAfterFreePoolSize = instance->getConfig().textureUseAfterFreePoolSize,
+                .metalUploadBufferSizeBytes = instance->getConfig().metalUploadBufferSizeBytes,
                 .disableParallelShaderCompile = instance->getConfig().disableParallelShaderCompile,
                 .disableHandleUseAfterFreeCheck = instance->getConfig().disableHandleUseAfterFreeCheck,
                 .forceGLES2Context = instance->getConfig().forceGLES2Context,
@@ -162,8 +163,7 @@ FEngine* FEngine::getEngine(void* token) {
     FILAMENT_CHECK_PRECONDITION(ThreadUtils::isThisThread(instance->mMainThreadId))
             << "Engine::createAsync() and Engine::getEngine() must be called on the same thread.";
 
-    // we use mResourceAllocator as a proxy for "am I already initialized"
-    if (!instance->mResourceAllocator) {
+    if (!instance->mInitialized) {
         if (UTILS_UNLIKELY(!instance->mDriver)) {
             // something went horribly wrong during driver initialization
             instance->mDriverThread.join();
@@ -261,7 +261,7 @@ void FEngine::init() {
     slog.i << "FEngine feature level: " << int(mActiveFeatureLevel) << io::endl;
 
 
-    mResourceAllocator = new ResourceAllocator(mConfig, driverApi);
+    mResourceAllocatorDisposer = std::make_shared<ResourceAllocatorDisposer>(driverApi);
 
     mFullScreenTriangleVb = downcast(VertexBuffer::Builder()
             .vertexCount(3)
@@ -424,11 +424,13 @@ void FEngine::init() {
                     }
                 });
             });
+
+    mInitialized = true;
 }
 
 FEngine::~FEngine() noexcept {
     SYSTRACE_CALL();
-    delete mResourceAllocator;
+    assert_invariant(!mResourceAllocatorDisposer);
     delete mDriver;
     if (mOwnPlatform) {
         PlatformFactory::destroy(&mPlatform);
@@ -439,7 +441,7 @@ void FEngine::shutdown() {
     SYSTRACE_CALL();
 
     // by construction this should never be nullptr
-    assert_invariant(mResourceAllocator);
+    assert_invariant(mResourceAllocatorDisposer);
 
     FILAMENT_CHECK_PRECONDITION(ThreadUtils::isThisThread(mMainThreadId))
             << "Engine::shutdown() called from the wrong thread!";
@@ -459,7 +461,8 @@ void FEngine::shutdown() {
      */
 
     mPostProcessManager.terminate(driver);  // free-up post-process manager resources
-    mResourceAllocator->terminate();
+    mResourceAllocatorDisposer->terminate();
+    mResourceAllocatorDisposer.reset();
     mDFG.terminate(*this);                  // free-up the DFG
     mRenderableManager.terminate();         // free-up all renderables
     mLightManager.terminate();              // free-up all lights
@@ -678,6 +681,7 @@ int FEngine::loop() {
     DriverConfig const driverConfig {
             .handleArenaSize = getRequestedDriverHandleArenaSize(),
             .textureUseAfterFreePoolSize = mConfig.textureUseAfterFreePoolSize,
+            .metalUploadBufferSizeBytes = mConfig.metalUploadBufferSizeBytes,
             .disableParallelShaderCompile = mConfig.disableParallelShaderCompile,
             .disableHandleUseAfterFreeCheck = mConfig.disableHandleUseAfterFreeCheck,
             .forceGLES2Context = mConfig.forceGLES2Context,

filament/src/details/Engine.h

@@ -86,6 +86,7 @@ namespace filament {
 
 class Renderer;
 class MaterialParser;
+class ResourceAllocatorDisposer;
 
 namespace backend {
 class Driver;
@@ -254,9 +255,13 @@ public:
         }
     }
 
-    ResourceAllocator& getResourceAllocator() noexcept {
-        assert_invariant(mResourceAllocator);
-        return *mResourceAllocator;
+    ResourceAllocatorDisposer& getResourceAllocatorDisposer() noexcept {
+        assert_invariant(mResourceAllocatorDisposer);
+        return *mResourceAllocatorDisposer;
+    }
+
+    std::shared_ptr<ResourceAllocatorDisposer> const& getSharedResourceAllocatorDisposer() noexcept {
+        return mResourceAllocatorDisposer;
     }
 
     void* streamAlloc(size_t size, size_t alignment) noexcept;
@@ -490,7 +495,7 @@ private:
     FTransformManager mTransformManager;
     FLightManager mLightManager;
     FCameraManager mCameraManager;
-    ResourceAllocator* mResourceAllocator = nullptr;
+    std::shared_ptr<ResourceAllocatorDisposer> mResourceAllocatorDisposer;
     HwVertexBufferInfoFactory mHwVertexBufferInfoFactory;
 
     ResourceList<FBufferObject> mBufferObjects{ "BufferObject" };
@@ -562,6 +567,8 @@ private:
 
     std::thread::id mMainThreadId{};
 
+    bool mInitialized = false;
+
     // Creation parameters
     Config mConfig;
 

filament/src/details/Material.cpp

@@ -119,6 +119,7 @@ struct Material::BuilderDetails {
     const void* mPayload = nullptr;
     size_t mSize = 0;
     bool mDefaultMaterial = false;
+    int32_t mShBandsCount = 3;
     std::unordered_map<
         utils::CString,
         std::variant<int32_t, float, bool>,
@@ -143,6 +144,11 @@ Material::Builder& Material::Builder::package(const void* payload, size_t size)
     return *this;
 }
 
+Material::Builder& Material::Builder::sphericalHarmonicsBandCount(size_t shBandCount) noexcept {
+    mImpl->mShBandsCount = math::clamp(shBandCount, size_t(1), size_t(3));
+    return *this;
+}
+
 template<typename T, typename>
 Material::Builder& Material::Builder::constant(const char* name, size_t nameLength, T value) {
     FILAMENT_CHECK_PRECONDITION(name != nullptr) << "name cannot be null";
@@ -897,6 +903,8 @@ void FMaterial::processSpecializationConstants(FEngine& engine, Material::Builde
     mSpecializationConstants.push_back({
             +ReservedSpecializationConstants::CONFIG_STEREO_EYE_COUNT,
             (int)engine.getConfig().stereoscopicEyeCount });
+    mSpecializationConstants.push_back({
+            +ReservedSpecializationConstants::CONFIG_SH_BANDS_COUNT, builder->mShBandsCount });
     if (UTILS_UNLIKELY(parser->getShaderLanguage() == ShaderLanguage::ESSL1)) {
         // The actual value of this spec-constant is set in the OpenGLDriver backend.
         mSpecializationConstants.push_back({

filament/src/details/MaterialInstance.h

@@ -231,6 +231,9 @@ private:
     void setParameterImpl(std::string_view name,
             FTexture const* texture, TextureSampler const& sampler);
 
+    template<typename T>
+    T getParameterImpl(std::string_view name) const;
+
     // initialize the default instance
     FMaterialInstance(FEngine& engine, FMaterial const* material) noexcept;
 

filament/src/details/MorphTargetBuffer.h

@@ -29,7 +29,8 @@
 
 #include <utils/Allocator.h>
 
-#include <vector>
+#include <stddef.h>
+#include <stdint.h>
 
 namespace filament {
 
@@ -59,16 +60,21 @@ public:
     inline size_t getVertexCount() const noexcept { return mVertexCount; }
     inline size_t getCount() const noexcept { return mCount; }
 
-private:
-    friend class FView;
-    friend class RenderPass;
+    backend::TextureHandle getPositionsHandle() const noexcept {
+        return mPbHandle;
+    }
 
+    backend::TextureHandle getTangentsHandle() const noexcept {
+        return mTbHandle;
+    }
+
+    inline backend::Handle<backend::HwSamplerGroup> getHwHandle() const noexcept { return mSbHandle; }
+
+private:
     void updateDataAt(backend::DriverApi& driver, backend::Handle <backend::HwTexture> handle,
             backend::PixelDataFormat format, backend::PixelDataType type, const char* out,
             size_t elementSize, size_t targetIndex, size_t count, size_t offset);
 
-    inline backend::Handle<backend::HwSamplerGroup> getHwHandle() const noexcept { return mSbHandle; }
-
     backend::Handle<backend::HwSamplerGroup> mSbHandle;
     backend::Handle<backend::HwTexture> mPbHandle;
     backend::Handle<backend::HwTexture> mTbHandle;

filament/src/details/Renderer.cpp

@@ -60,6 +60,7 @@
 
 #include <chrono>
 #include <limits>
+#include <memory>
 #include <utility>
 
 #include <stddef.h>
@@ -88,7 +89,11 @@ FRenderer::FRenderer(FEngine& engine) :
         mHdrQualityMedium(TextureFormat::R11F_G11F_B10F),
         mHdrQualityHigh(TextureFormat::RGB16F),
         mIsRGB8Supported(false),
-        mUserEpoch(engine.getEngineEpoch())
+        mUserEpoch(engine.getEngineEpoch()),
+        mResourceAllocator(std::make_unique<ResourceAllocator>(
+                engine.getSharedResourceAllocatorDisposer(),
+                engine.getConfig(),
+                engine.getDriverApi()))
 {
     FDebugRegistry& debugRegistry = engine.getDebugRegistry();
     debugRegistry.registerProperty("d.renderer.doFrameCapture",
@@ -176,6 +181,7 @@ void FRenderer::terminate(FEngine& engine) {
     }
     mFrameInfoManager.terminate(driver);
     mFrameSkipper.terminate(driver);
+    mResourceAllocator->terminate();
 }
 
 void FRenderer::resetUserTime() {
@@ -232,6 +238,43 @@ void FRenderer::setPresentationTime(int64_t monotonic_clock_ns) {
     driver.setPresentationTime(monotonic_clock_ns);
 }
 
+void FRenderer::setVsyncTime(uint64_t steadyClockTimeNano) noexcept {
+    mVsyncSteadyClockTimeNano = steadyClockTimeNano;
+}
+
+void FRenderer::skipFrame(uint64_t vsyncSteadyClockTimeNano) {
+    SYSTRACE_CALL();
+
+    FILAMENT_CHECK_PRECONDITION(!mSwapChain) <<
+            "skipFrame() can't be called between beginFrame() and endFrame()";
+
+    if (!vsyncSteadyClockTimeNano) {
+        vsyncSteadyClockTimeNano = mVsyncSteadyClockTimeNano;
+        mVsyncSteadyClockTimeNano = 0;
+    }
+
+    FEngine& engine = mEngine;
+    FEngine::DriverApi& driver = engine.getDriverApi();
+
+    // Gives the backend a chance to execute periodic tasks. This must be called before
+    // the frame skipper.
+    driver.tick();
+
+    // do this before engine.flush()
+    mResourceAllocator->gc(true);
+
+    // Run the component managers' GC in parallel
+    // WARNING: while doing this we can't access any component manager
+    auto& js = engine.getJobSystem();
+
+    auto *job = js.runAndRetain(jobs::createJob(js, nullptr, &FEngine::gc, &engine)); // gc all managers
+
+    engine.flush();     // flush command stream
+
+    // make sure we're done with the gcs
+    js.waitAndRelease(job);
+}
+
 bool FRenderer::beginFrame(FSwapChain* swapChain, uint64_t vsyncSteadyClockTimeNano) {
     assert_invariant(swapChain);
 
@@ -252,6 +295,11 @@ bool FRenderer::beginFrame(FSwapChain* swapChain, uint64_t vsyncSteadyClockTimeN
     }
 #endif
 
+    if (!vsyncSteadyClockTimeNano) {
+        vsyncSteadyClockTimeNano = mVsyncSteadyClockTimeNano;
+        mVsyncSteadyClockTimeNano = 0;
+    }
+
     // get the timestamp as soon as possible
     using namespace std::chrono;
     const steady_clock::time_point now{ steady_clock::now() };
@@ -371,7 +419,7 @@ void FRenderer::endFrame() {
     }
 
     // do this before engine.flush()
-    engine.getResourceAllocator().gc();
+    mResourceAllocator->gc();
 
     // Run the component managers' GC in parallel
     // WARNING: while doing this we can't access any component manager
@@ -459,6 +507,9 @@ void FRenderer::renderStandaloneView(FView const* view) {
     FILAMENT_CHECK_PRECONDITION(view->getRenderTarget())
             << "View \"" << view->getName() << "\" must have a RenderTarget associated";
 
+    FILAMENT_CHECK_PRECONDITION(!mSwapChain)
+            << "renderStandaloneView() must be called outside of beginFrame() / endFrame()";
+
     if (UTILS_LIKELY(view->getScene())) {
         mPreviousRenderTargets.clear();
         mFrameId++;
@@ -738,7 +789,7 @@ void FRenderer::renderJob(RootArenaScope& rootArenaScope, FView& view) {
      * Frame graph
      */
 
-    FrameGraph fg(engine.getResourceAllocator(),
+    FrameGraph fg(*mResourceAllocator,
             isProtectedContent ? FrameGraph::Mode::PROTECTED : FrameGraph::Mode::UNPROTECTED);
     auto& blackboard = fg.getBlackboard();
 

filament/src/details/Renderer.h

@@ -29,9 +29,6 @@
 
 #include "backend/DriverApiForward.h"
 
-#include <fg/FrameGraphId.h>
-#include <fg/FrameGraphTexture.h>
-
 #include <filament/Renderer.h>
 #include <filament/Viewport.h>
 
@@ -40,11 +37,25 @@
 
 #include <utils/compiler.h>
 #include <utils/Allocator.h>
+#include <utils/FixedCapacityVector.h>
+
+#include <math/vec4.h>
 
 #include <tsl/robin_set.h>
 
+#include <algorithm>
+#include <chrono>
+#include <functional>
+#include <memory>
+#include <utility>
+
+#include <stddef.h>
+#include <stdint.h>
+
 namespace filament {
 
+class ResourceAllocator;
+
 namespace backend {
 class Driver;
 } // namespace backend
@@ -78,6 +89,11 @@ public:
 
     void setPresentationTime(int64_t monotonic_clock_ns);
 
+    void setVsyncTime(uint64_t steadyClockTimeNano) noexcept;
+
+    // skip a frame
+    void skipFrame(uint64_t vsyncSteadyClockTimeNano);
+
     // start a frame
     bool beginFrame(FSwapChain* swapChain, uint64_t vsyncSteadyClockTimeNano);
 
@@ -131,6 +147,14 @@ public:
         return mClearOptions;
     }
 
+    utils::FixedCapacityVector<Renderer::FrameInfo> getFrameInfoHistory(size_t historySize) const noexcept {
+        return mFrameInfoManager.getFrameInfoHistory(historySize);
+    }
+
+    size_t getMaxFrameHistorySize() const noexcept {
+        return MAX_FRAMETIME_HISTORY;
+    }
+
 private:
     friend class Renderer;
     using Command = RenderPass::Command;
@@ -187,6 +211,8 @@ private:
     backend::TargetBufferFlags mClearFlags{};
     tsl::robin_set<FRenderTarget*> mPreviousRenderTargets;
     std::function<void()> mBeginFrameInternal;
+    uint64_t mVsyncSteadyClockTimeNano = 0;
+    std::unique_ptr<ResourceAllocator> mResourceAllocator{};
 };
 
 FILAMENT_DOWNCAST(Renderer)

filament/src/details/Scene.cpp

@@ -248,7 +248,7 @@ void FScene::prepare(utils::JobSystem& js,
 
     auto* renderableJob = jobs::parallel_for(js, rootJob,
             renderableInstances.data(), renderableInstances.size(),
-            std::cref(renderableWork), jobs::CountSplitter<128, 5>());
+            std::cref(renderableWork), jobs::CountSplitter<64>());
 
     auto* lightJob = jobs::parallel_for(js, rootJob,
             lightInstances.data(), lightInstances.size(),

filament/src/details/View.cpp

@@ -17,6 +17,7 @@
 #include "details/View.h"
 
 #include "Culler.h"
+#include "FrameHistory.h"
 #include "Froxelizer.h"
 #include "RenderPrimitive.h"
 #include "ResourceAllocator.h"
@@ -118,7 +119,7 @@ void FView::terminate(FEngine& engine) {
     DriverApi& driver = engine.getDriverApi();
     driver.destroyBufferObject(mLightUbh);
     driver.destroyBufferObject(mRenderableUbh);
-    drainFrameHistory(engine);
+    clearFrameHistory(engine);
 
     ShadowMapManager::terminate(engine, mShadowMapManager);
     mPerViewUniforms.terminate(driver);
@@ -162,7 +163,7 @@ void FView::setDynamicLightingOptions(float zLightNear, float zLightFar) noexcep
 }
 
 float2 FView::updateScale(FEngine& engine,
-        FrameInfo const& info,
+        filament::details::FrameInfo const& info,
         Renderer::FrameRateOptions const& frameRateOptions,
         Renderer::DisplayInfo const& displayInfo) noexcept {
 
@@ -1007,20 +1008,29 @@ FrameGraphId<FrameGraphTexture> FView::renderShadowMaps(FEngine& engine, FrameGr
 
 void FView::commitFrameHistory(FEngine& engine) noexcept {
     // Here we need to destroy resources in mFrameHistory.back()
+    auto& disposer = engine.getResourceAllocatorDisposer();
     auto& frameHistory = mFrameHistory;
 
     FrameHistoryEntry& last = frameHistory.back();
-    last.taa.color.destroy(engine.getResourceAllocator());
-    last.ssr.color.destroy(engine.getResourceAllocator());
+    disposer.destroy(last.taa.color.handle);
+    disposer.destroy(last.ssr.color.handle);
+    last.taa.color.handle.clear();
+    last.ssr.color.handle.clear();
 
     // and then push the new history entry to the history stack
     frameHistory.commit();
 }
 
-void FView::drainFrameHistory(FEngine& engine) noexcept {
+void FView::clearFrameHistory(FEngine& engine) noexcept {
     // make sure we free all resources in the history
-    for (size_t i = 0; i < mFrameHistory.size(); ++i) {
-        commitFrameHistory(engine);
+    auto& disposer = engine.getResourceAllocatorDisposer();
+    auto& frameHistory = mFrameHistory;
+    for (size_t i = 0; i < frameHistory.size(); ++i) {
+        FrameHistoryEntry& last = frameHistory[i];
+        disposer.destroy(last.taa.color.handle);
+        disposer.destroy(last.ssr.color.handle);
+        last.taa.color.handle.clear();
+        last.ssr.color.handle.clear();
     }
 }
 

filament/src/details/View.h

@@ -283,7 +283,7 @@ public:
     }
 
     math::float2 updateScale(FEngine& engine,
-            FrameInfo const& info,
+            details::FrameInfo const& info,
             Renderer::FrameRateOptions const& frameRateOptions,
             Renderer::DisplayInfo const& displayInfo) noexcept;
 
@@ -422,6 +422,10 @@ public:
     // (e.g.: after the FrameGraph execution).
     void commitFrameHistory(FEngine& engine) noexcept;
 
+    // Clean-up the whole history, free all resources. This is typically called when the View is
+    // being terminated. Or we're changing Renderer.
+    void clearFrameHistory(FEngine& engine) noexcept;
+
     // create the picking query
     View::PickingQuery& pick(uint32_t x, uint32_t y, backend::CallbackHandler* handler,
             View::PickingQueryResultCallback callback) noexcept;
@@ -485,10 +489,6 @@ private:
             Culler::result_type* visibleMask,
             size_t count);
 
-    // Clean-up the whole history, free all resources. This is typically called when the View is
-    // being terminated.
-    void drainFrameHistory(FEngine& engine) noexcept;
-
     // we don't inline this one, because the function is quite large and there is not much to
     // gain from inlining.
     static FScene::RenderableSoa::iterator partition(

filament/src/fg/FrameGraph.cpp

@@ -24,6 +24,7 @@
 #include "FrameGraphPass.h"
 #include "FrameGraphRenderPass.h"
 #include "FrameGraphTexture.h"
+#include "ResourceAllocator.h"
 
 #include "details/Engine.h"
 

filament/test/CMakeLists.txt

@@ -62,6 +62,7 @@ if (ANDROID)
     target_link_libraries(test_compiler PRIVATE utils)
     target_link_libraries(test_compiler PRIVATE EGL)
     target_link_libraries(test_compiler PRIVATE GLESv3)
+    target_link_libraries(test_compiler PRIVATE bluevk)    
 endif()
 
 add_executable(test_material_parser

filament/test/compiler_test.cpp

@@ -1,181 +1,82 @@
-/*
- * Copyright (C) 2023 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 <iostream>
+#include <utils/Hash.h>
+#include <tsl/robin_map.h>
+#include <bluevk/BlueVK.h>
 
-#include <utils/Log.h>
+namespace {
 
-#include <GLES3/gl3.h>
-#include <EGL/egl.h>
-#include <EGL/eglext.h>
-
-#include <gtest/gtest.h>
-
-using namespace utils;
-using namespace std::literals;
-
-class CompilerTest : public testing::Test {
-protected:
-    void SetUp() override {
-        EGLBoolean success;
-        EGLint major, minor;
-        dpy = eglGetDisplay(EGL_DEFAULT_DISPLAY);
-        ASSERT_NE(dpy, EGL_NO_DISPLAY);
-
-        EGLBoolean const initialized = eglInitialize(dpy, &major, &minor);
-        ASSERT_TRUE(initialized);
-
-        EGLint const contextAttribs[] = {
-                EGL_CONTEXT_CLIENT_VERSION, 3,
-                EGL_NONE
-        };
-
-        EGLint configsCount;
-        EGLint configAttribs[] = {
-                EGL_RENDERABLE_TYPE, EGL_OPENGL_ES3_BIT_KHR,        //  0
-                EGL_RED_SIZE,    8,                                 //  2
-                EGL_GREEN_SIZE,  8,                                 //  4
-                EGL_BLUE_SIZE,   8,                                 //  6
-                EGL_NONE                                            // 14
-        };
-        success = eglChooseConfig(dpy, configAttribs, &config, 1, &configsCount);
-        ASSERT_TRUE(success);
-
-        context = eglCreateContext(dpy, config, EGL_NO_CONTEXT, contextAttribs);
-        ASSERT_NE(context, EGL_NO_CONTEXT);
-
-        success = eglMakeCurrent(dpy, EGL_NO_SURFACE, EGL_NO_SURFACE, context);
-        ASSERT_TRUE(success);
-
-        ASSERT_EQ(eglGetError(), EGL_SUCCESS);
-    }
-
-    void TearDown() override {
-        eglMakeCurrent(dpy, EGL_NO_SURFACE, EGL_NO_SURFACE, EGL_NO_CONTEXT);
-        eglDestroyContext(dpy, context);
-        eglTerminate(dpy);
-    }
-
-private:
-    EGLDisplay dpy;
-    EGLContext context;
-    EGLConfig config;
+struct T {
 };
 
-TEST_F(CompilerTest, Simple) {
-    auto shader = R"(
-#version 300 es
-void main()
-{
-})"sv;
+constexpr uint8_t const UNIQUE_DESCRIPTOR_SET_COUNT = 3;
+constexpr uint8_t const SHADER_TYPE_COUNT = 3;
+using Timestamp = uint64_t;
+using VulkanResourceAllocator = T;
 
-    const char* const src = shader.data();
-    GLint const len = (GLint)shader.size();
+struct PushConstantKey {
+    uint8_t stage;// We have one set of push constant per shader stage (fragment, vertex, etc).
+    uint8_t size;
+};
 
-    GLuint const id = glCreateShader(GL_VERTEX_SHADER);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
+struct PipelineLayoutKey {
+    using DescriptorSetLayoutArray = std::array<VkDescriptorSetLayout, UNIQUE_DESCRIPTOR_SET_COUNT>;
+    DescriptorSetLayoutArray descSetLayouts = {};        // 8 * 3
+    PushConstantKey pushConstant[SHADER_TYPE_COUNT] = {};// 2 * 3
+    uint16_t padding = 0;
+};
+static_assert(sizeof(PipelineLayoutKey) == 32);
 
-    glShaderSource(id, 1, &src, &len);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
+using PipelineLayoutKeyHashFn = utils::hash::MurmurHashFn<PipelineLayoutKey>;
+struct PipelineLayoutKeyEqual {
+    bool operator()(PipelineLayoutKey const& k1, PipelineLayoutKey const& k2) const {
+        return 0 == memcmp((const void*) &k1, (const void*) &k2, sizeof(PipelineLayoutKey));
+    }
+};
 
-    glCompileShader(id);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
+struct PipelineLayoutCacheEntry {
+    VkPipelineLayout handle;
+    Timestamp lastUsed;
+};
 
-    GLint result = 0;
-    glGetShaderiv(id, GL_COMPILE_STATUS, &result);
-    EXPECT_EQ(result, GL_TRUE);
+using PipelineLayoutMap = tsl::robin_map<PipelineLayoutKey, PipelineLayoutCacheEntry,
+        PipelineLayoutKeyHashFn, PipelineLayoutKeyEqual>;
 
-    glDeleteShader(id);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
+class TestClass {
+public:
+    VkDevice mDevice;
+    VulkanResourceAllocator* mAllocator;
+    Timestamp mTimestamp;
+    PipelineLayoutMap layout;
+};
+
+void insert(PipelineLayoutMap& layout, uint64_t val) {
+    PipelineLayoutKey key{};
+    key.descSetLayouts[0] = (VkDescriptorSetLayout) val;
+    layout[key] = {(VkPipelineLayout) val, val};
 }
 
-TEST_F(CompilerTest, CrashPVRUniFlexCompileToHw) {
-
-    // Some PowerVR driver crash with this shader
-
-    auto shader = R"(
-#version 300 es
-
-layout(location = 0) in vec4 mesh_position;
-
-layout(std140) uniform FrameUniforms {
-    vec2 i;
-} frameUniforms;
-
-void main() {
-    gl_Position = mesh_position;
-    gl_Position.z = dot(gl_Position.zw, frameUniforms.i);
-})"sv;
-
-    const char* const src = shader.data();
-    GLint const len = (GLint)shader.size();
-
-    GLuint const id = glCreateShader(GL_VERTEX_SHADER);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
-
-    glShaderSource(id, 1, &src, &len);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
-
-    glCompileShader(id);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
-
-    GLint result = 0;
-    glGetShaderiv(id, GL_COMPILE_STATUS, &result);
-    EXPECT_EQ(result, GL_TRUE);
-
-    glDeleteShader(id);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
-}
-
-TEST_F(CompilerTest, ConstParameters) {
-
-    // Some PowerVR driver fail to compile this shader
-
-    auto shader = R"(
-#version 300 es
-
-highp mat3 m;
-void buggy(const mediump vec3 n) {
-    m*n;
-}
-
-void main() {
-})"sv;
-
-    const char* const src = shader.data();
-    GLint const len = (GLint)shader.size();
-
-    GLuint const id = glCreateShader(GL_FRAGMENT_SHADER);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
-
-    glShaderSource(id, 1, &src, &len);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
-
-    glCompileShader(id);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
-
-    GLint result = 0;
-    glGetShaderiv(id, GL_COMPILE_STATUS, &result);
-    EXPECT_EQ(result, GL_TRUE);
-
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
-
-    glDeleteShader(id);
-    EXPECT_EQ(glGetError(), GL_NO_ERROR);
-}
+}// namespace
 
 int main(int argc, char** argv) {
-    ::testing::InitGoogleTest(&argc, argv);
-    return RUN_ALL_TESTS();
+    constexpr uint32_t START = 1000;
+    TestClass c;
+    auto& layout = c.layout;
+
+    uint64_t count = 0;
+    for (; count < 20; count++) {
+        insert(layout, count + START);
+    }
+
+    for (uint64_t t = 0; t < count; ++t) {
+        PipelineLayoutKey key = {};
+        key.descSetLayouts[0] = (VkDescriptorSetLayout)(START + t);
+
+        if (layout.find(key) != layout.end()) {
+            std::cout << "found " << t << std::endl;
+        } else {
+            std::cout << "not found" << t << std::endl;
+        }
+    }
+
+    return 0;
 }

filament/test/filament_framegraph_test.cpp

@@ -34,6 +34,10 @@ using namespace backend;
 
 class MockResourceAllocator : public ResourceAllocatorInterface {
     uint32_t handle = 0;
+    struct MockDisposer : public ResourceAllocatorDisposerInterface {
+        void destroy(backend::TextureHandle) noexcept override {}
+    } disposer;
+
 public:
     backend::RenderTargetHandle createRenderTarget(const char* name,
             backend::TargetBufferFlags targetBufferFlags,
@@ -60,6 +64,10 @@ public:
 
     void destroyTexture(backend::TextureHandle h) noexcept override {
     }
+
+    ResourceAllocatorDisposerInterface& getDisposer() noexcept override {
+        return disposer;
+    }
 };
 
 class FrameGraphTest : public testing::Test {

ios/CocoaPods/Filament.podspec

@@ -1,12 +1,12 @@
 Pod::Spec.new do |spec|
   spec.name = "Filament"
-  spec.version = "1.53.0"
+  spec.version = "1.53.2"
   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/v1.53.0/filament-v1.53.0-ios.tgz" }
+  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.53.2/filament-v1.53.2-ios.tgz" }
 
   # Fix linking error with Xcode 12; we do not yet support the simulator on Apple silicon.
   spec.pod_target_xcconfig = {

libs/camutils/include/camutils/Manipulator.h

@@ -109,6 +109,7 @@ public:
         vec4 groundPlane;
         RayCallback raycastCallback;
         void* raycastUserdata;
+        bool panning = true;
     };
 
     struct Builder {
@@ -143,6 +144,8 @@ public:
         Builder& groundPlane(FLOAT a, FLOAT b, FLOAT c, FLOAT d);  //! Plane equation used as a raycast fallback
         Builder& raycastCallback(RayCallback cb, void* userdata);  //! Raycast function for accurate grab-and-pan
 
+        Builder& panning(bool enabled);  //! Sets whether panning is enabled
+
         /**
          * Creates a new camera manipulator, either ORBIT, MAP, or FREE_FLIGHT.
          *

libs/camutils/src/Manipulator.cpp

@@ -144,6 +144,14 @@ Manipulator<FLOAT>::Builder& Manipulator<FLOAT>::Builder::raycastCallback(RayCal
     return *this;
 }
 
+
+
+template <typename FLOAT> typename
+Manipulator<FLOAT>::Builder& Manipulator<FLOAT>::Builder::panning(bool enabled) {
+    details.panning = enabled;
+    return *this;
+}
+
 template <typename FLOAT>
 Manipulator<FLOAT>* Manipulator<FLOAT>::Builder::build(Mode mode) {
     switch (mode) {

libs/camutils/src/OrbitManipulator.h

@@ -69,7 +69,7 @@ public:
     }
 
     void grabBegin(int x, int y, bool strafe) override {
-        mGrabState = strafe ? PANNING : ORBITING;
+        mGrabState = strafe && Base::mProps.panning ? PANNING : ORBITING;
         mGrabPivot = mPivot;
         mGrabEye = Base::mEye;
         mGrabTarget = Base::mTarget;