Update glslang to e3bca2a

Filament specific changes
Update spirv-tools to 46492aa
2022-05-04 12:29:21 -07:00 · 2022-05-04 12:06:26 -07:00 · 2022-05-04 12:00:04 -07:00 · 2022-05-04 11:07:15 -07:00 · 2022-05-03 15:16:11 -07:00 · 2022-05-03 14:33:43 -07:00
1516 changed files with 64995 additions and 85627 deletions
@@ -349,7 +349,7 @@ same version that our continuous builds use.

 ```
 cd <your chosen parent folder for the emscripten SDK>
-curl -L https://github.com/emscripten-core/emsdk/archive/refs/tags/3.1.5.zip > emsdk.zip
+curl -L https://github.com/emscripten-core/emsdk/archive/refs/tags/3.1.9.zip > emsdk.zip
 unzip emsdk.zip ; mv emsdk-* emsdk ; cd emsdk
 python ./emsdk.py install latest
 python ./emsdk.py activate latest
@@ -654,6 +654,7 @@ add_subdirectory(${LIBRARIES}/utils)
 add_subdirectory(${LIBRARIES}/viewer)
 add_subdirectory(${FILAMENT}/filament)
 add_subdirectory(${FILAMENT}/shaders)
+add_subdirectory(${EXTERNAL}/basisu/tnt)
 add_subdirectory(${EXTERNAL}/civetweb/tnt)
 add_subdirectory(${EXTERNAL}/hat-trie/tnt)
 add_subdirectory(${EXTERNAL}/imgui/tnt)
@@ -706,9 +707,6 @@ if (IS_HOST_PLATFORM)

    add_subdirectory(${FILAMENT}/samples)

-    add_subdirectory(${EXTERNAL}/basisu/tnt)
-    add_subdirectory(${EXTERNAL}/astcenc/tnt)
-    add_subdirectory(${EXTERNAL}/etc2comp)
    add_subdirectory(${EXTERNAL}/libassimp/tnt)
    add_subdirectory(${EXTERNAL}/libpng/tnt)
    add_subdirectory(${EXTERNAL}/libsdl2/tnt)
@@ -31,7 +31,7 @@ repositories {
 }

 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.21.2'
+    implementation 'com.google.android.filament:filament-android:1.21.3'
 }
 ```

@@ -51,7 +51,7 @@ Here are all the libraries available in the group `com.google.android.filament`:
 iOS projects can use CocoaPods to install the latest release:

 ```
-pod 'Filament', '~> 1.21.2'
+pod 'Filament', '~> 1.21.3'
 ```

 ### Snapshots
@@ -177,6 +177,7 @@ steps:
  - [x] KHR_materials_variants
  - [x] KHR_materials_volume
  - [x] KHR_mesh_quantization
+  - [x] KHR_texture_basisu
  - [x] KHR_texture_transform


@@ -3,7 +3,25 @@
 This file contains one line summaries of commits that are worthy of mentioning in release notes.
 A new header is inserted each time a *tag* is created.

-## v1.21.3 (currently main branch)
+## v1.22.1 (currently main branch)
+
+## v1.22.0
+
+- engine: Changed UBOs layout [⚠️ **Material breakage**].
+- engine: Improve effects relying on mipmapping
+
+## v1.21.3
+
+- Java: Renamed the `KTXLoader` Kotlin class to `KTX1Loader`.  [⚠️ **API Change**].
+- libs: Added `Ktx2Reader` and `BasisEncoder` to support Basis-encoded KTX2 files.
+- engine: Remove deprecated `Stream` APIs, see `Texture::import()` for an alternative [⚠️ **API Change**].
+- tools: Removed KTX1 compression support from mipgen.  [⚠️ **API Change**].
+- tools: Added support for KTX2 to mipgen.
+- gltfio: Added `resetBoneMatrices()` method.
+- gltfio: Introduced `TextureProvider` interface.  [⚠️ **API Change**].
+- gltfio: Fix progress indicator when error occurs.
+- samples: Desktop and Web Suzanne samples now use KTX2 instead of KTX1.
+- samples: Fix regression with sample-gltf-viewer and zip files.

 ## v1.21.2

@@ -453,7 +471,7 @@ devices.
 - engine: Fix RenderTarget NPE when depth is not present.
 - engine: Improvements to Camera APIs. Move focus distance from DofOptions to Camera.
 - engine: VSM shadows now support `shadowMultiplier`.
- java: Expose severla MaterialInstance APIs (setColorWrite, setDepthWrite, setDepthCulling) that
+- java: Expose several MaterialInstance APIs (setColorWrite, setDepthWrite, setDepthCulling) that
  should have been public.
 - java: fix bug with Texture::setImage buffer size calculation.

@@ -11,6 +11,9 @@
 // com.google.android.filament.exclude-vulkan
 //     When set, support for Vulkan will be excluded.
 //
+// com.google.android.filament.matdbg
+//     When set, enables matdbg, disables shader optimizations
+//
 // com.google.android.filament.skip-samples
 //     Exclude samples from the project. Useful to speed up compilation.
 //
@@ -59,6 +62,11 @@ buildscript {
            .forUseAtConfigurationTime()
            .isPresent()

+    def matdbg = providers
+            .gradleProperty("com.google.android.filament.matdbg")
+            .forUseAtConfigurationTime()
+            .isPresent()
+
    def abis = ["arm64-v8a", "armeabi-v7a", "x86_64", "x86"]
    def newAbis = providers
            .gradleProperty("com.google.android.filament.abis")
@@ -102,7 +110,9 @@ buildscript {
        "-DANDROID_PLATFORM=21",
        "-DANDROID_STL=c++_static",
        "-DFILAMENT_DIST_DIR=${filamentPath}".toString(),
-        "-DFILAMENT_SUPPORTS_VULKAN=${excludeVulkan ? 'OFF' : 'ON'}".toString()
+        "-DFILAMENT_SUPPORTS_VULKAN=${excludeVulkan ? 'OFF' : 'ON'}".toString(),
+        "-DFILAMENT_ENABLE_MATDBG=${matdbg ? 'ON' : 'OFF'}".toString(),
+        "-DFILAMENT_DISABLE_MATOPT=${matdbg ? 'ON' : 'OFF'}".toString()
    ]

    ext.cppFlags = [
@@ -98,16 +98,6 @@ Java_com_google_android_filament_Stream_nBuilderStreamSource(JNIEnv* env,
    builder->setStreamSource(env, streamSource);
 }

-extern "C" JNIEXPORT void JNICALL
-Java_com_google_android_filament_Stream_nBuilderStream(JNIEnv*, jclass,
-        jlong nativeStreamBuilder, jlong externalTextureId) {
-    StreamBuilder* builder = (StreamBuilder*) nativeStreamBuilder;
-#pragma clang diagnostic push
-#pragma clang diagnostic ignored "-Wdeprecated-declarations"
-    builder->builder()->stream(externalTextureId);
-#pragma clang diagnostic pop
-}
-
 extern "C" JNIEXPORT void JNICALL
 Java_com_google_android_filament_Stream_nBuilderWidth(JNIEnv*, jclass,
        jlong nativeStreamBuilder, jint width) {
@@ -143,41 +133,6 @@ Java_com_google_android_filament_Stream_nSetDimensions(JNIEnv*, jclass, jlong na
    stream->setDimensions((uint32_t) width, (uint32_t) height);
 }

-extern "C" JNIEXPORT jint JNICALL
-Java_com_google_android_filament_Stream_nReadPixels(JNIEnv *env, jclass,
-        jlong nativeStream, jlong nativeEngine,
-        jint xoffset, jint yoffset, jint width, jint height,
-        jobject storage, jint remaining,
-        jint left, jint top, jint type, jint alignment, jint stride, jint format,
-        jobject handler, jobject runnable) {
-    Stream *stream = (Stream *) nativeStream;
-    Engine *engine = (Engine *) nativeEngine;
-
-    stride = stride ? stride : width;
-    size_t sizeInBytes = PixelBufferDescriptor::computeDataSize(
-            (PixelDataFormat) format, (PixelDataType) type,
-            (size_t) stride, (size_t) (height + top), (size_t) alignment);
-
-    AutoBuffer nioBuffer(env, storage, 0);
-    if (sizeInBytes > (remaining << nioBuffer.getShift())) {
-        // BufferOverflowException
-        return -1;
-    }
-
-    void *buffer = nioBuffer.getData();
-    auto *callback = JniBufferCallback::make(engine, env, handler, runnable, std::move(nioBuffer));
-
-    PixelBufferDescriptor desc(buffer, sizeInBytes, (backend::PixelDataFormat) format,
-            (backend::PixelDataType) type, (uint8_t) alignment, (uint32_t) left, (uint32_t) top,
-            (uint32_t) stride,
-            callback->getHandler(), &JniBufferCallback::postToJavaAndDestroy, callback);
-
-    stream->readPixels(uint32_t(xoffset), uint32_t(yoffset), uint32_t(width), uint32_t(height),
-            std::move(desc));
-
-    return 0;
-}
-
 extern "C" JNIEXPORT jlong JNICALL
 Java_com_google_android_filament_Stream_nGetTimestamp(JNIEnv*, jclass, jlong nativeStream) {
    Stream *stream = (Stream *) nativeStream;
@@ -531,6 +531,10 @@ public class Renderer {
     * Typically, this will happen after multiple calls to {@link #beginFrame},
     * {@link #render}, {@link #endFrame}.</p>
     * <br>
+     * <p>OpenGL only: if issuing a <code>readPixels</code> on a {@link RenderTarget} backed by a
+     * {@link Texture} that had data uploaded to it via {@link Texture#setImage}, the data returned
+     * from <code>readPixels</code> will be y-flipped with respect to the {@link Texture#setImage}
+     * call.</p>
     * <p><code>readPixels</code> is intended for debugging and testing.
     * It will impact performance significantly.</p>
     *
@@ -103,9 +103,6 @@ public class Stream {
        /** Not synchronized but copy-free. Good for video. */
        NATIVE,

-        /** Synchronized, but GL-only and incurs copies. Good for AR on devices before API 26. */
-        TEXTURE_ID,
-
        /** Synchronized, copy-free, and take a release callback. Good for AR but requires API 26+. */
        ACQUIRED,
    };
@@ -121,7 +118,7 @@ public class Stream {
     * By default, Stream objects are {@link StreamType#ACQUIRED ACQUIRED} and must have external images pushed to them via
     * {@link #setAcquiredImage}.
     *
-     * To create a {@link StreamType#NATIVE NATIVE} or {@link StreamType#TEXTURE_ID TEXTURE_ID} stream, call one of the <pre>stream</pre> methods
+     * To create a {@link StreamType#NATIVE NATIVE} stream, call one of the <pre>stream</pre> methods
     * on the builder.
     */
    public static class Builder {
@@ -156,27 +153,6 @@ public class Stream {
            throw new IllegalArgumentException("Invalid stream source: " + streamSource);
        }

-        /**
-         * Creates a {@link StreamType#TEXTURE_ID TEXTURE_ID} stream. A copy stream will sample data from the supplied
-         * external texture and copy it into an internal private texture.
-         *
-         * <p>Currently only OpenGL external texture ids are supported.</p>
-         *
-         * @param externalTextureId An opaque texture id (typically a GLuint created with
-         *                          <code>glGenTextures()</code>) in a context shared with
-         *                          filament -- in that case this texture's target must be
-         *                          <code>GL_TEXTURE_EXTERNAL_OES.</code>
-         * @return This Builder, for chaining calls.
-         * @see Texture#setExternalStream
-         * @deprecated this method existed only for ARCore which doesn't need this anymore, use {@link Texture.Builder#importTexture(long)} instead.
-         */
-        @Deprecated
-        @NonNull
-        public Builder stream(long externalTextureId) {
-            nBuilderStream(mNativeBuilder, externalTextureId);
-            return this;
-        }
-
        /**
         * @param width initial width of the incoming stream. Whether this value is used is
         *              stream dependent. On Android, it must be set when using
@@ -276,92 +252,6 @@ public class Stream {
        nSetDimensions(getNativeObject(), width, height);
    }

-    /**
-     * Reads back the content of the last frame of a <code>Stream</code> since the last call to
-     * {@link Renderer#beginFrame}.
-     *
-     * <p>The Stream must be a copy stream, which can be checked with {@link #getStreamType()}.
-     * This function is a no-op otherwise.</p>
-     *
-     * <pre>
-     *
-     *  Stream buffer                  User buffer (PixelBufferDescriptor)
-     *  +--------------------+
-     *  |                    |                .stride         .alignment
-     *  |                    |         ----------------------->-->
-     *  |                    |         O----------------------+--+   low addresses
-     *  |                    |         |          |           |  |
-     *  |             w      |         |          | .top      |  |
-     *  |       <--------->  |         |          V           |  |
-     *  |       +---------+  |         |     +---------+      |  |
-     *  |       |     ^   |  | ======> |     |         |      |  |
-     *  |   x   |    h|   |  |         |.left|         |      |  |
-     *  +------>|     v   |  |         +---->|         |      |  |
-     *  |       +.........+  |         |     +.........+      |  |
-     *  |            ^       |         |                      |  |
-     *  |          y |       |         +----------------------+--+  high addresses
-     *  O------------+-------+
-     *
-     * </pre>
-     *
-     * <p>Typically readPixels() will be called after {@link Renderer#beginFrame}.</p>
-     *
-     * <p>After calling this method, the callback associated with <code>buffer</code>
-     * will be invoked on the main thread, indicating that the read-back has completed.
-     * Typically, this will happen after multiple calls to {@link Renderer#beginFrame},
-     * {@link Renderer#render}, {@link Renderer#endFrame}.</p>
-     *
-     * <p><code>readPixels</code> is intended for debugging and testing.
-     * It will impact performance significantly.</p>
-     *
-     * @param xoffset   left offset of the sub-region to read back
-     * @param yoffset   bottom offset of the sub-region to read back
-     * @param width     width of the sub-region to read back
-     * @param height    height of the sub-region to read back
-     * @param buffer    client-side buffer where the read-back will be written
-     *
-     *                  <p>
-     *                  The following format are always supported:
-     *                      <li>{@link Texture.Format#RGBA}</li>
-     *                      <li>{@link Texture.Format#RGBA_INTEGER}</li>
-     *                  </p>
-     *
-     *                  <p>
-     *                  The following types are always supported:
-     *                      <li>{@link Texture.Type#UBYTE}</li>
-     *                      <li>{@link Texture.Type#UINT}</li>
-     *                      <li>{@link Texture.Type#INT}</li>
-     *                      <li>{@link Texture.Type#FLOAT}</li>
-     *                  </p>
-     *
-     *                  <p>Other combination of format/type may be supported. If a combination is
-     *                  not supported, this operation may fail silently. Use a DEBUG build
-     *                  to get some logs about the failure.</p>
-     *
-     * @exception BufferOverflowException if the specified parameters would result in reading
-     * outside of <code>buffer</code>.
-     */
-    public void readPixels(
-            @IntRange(from = 0) int xoffset, @IntRange(from = 0) int yoffset,
-            @IntRange(from = 0) int width, @IntRange(from = 0) int height,
-            @NonNull Texture.PixelBufferDescriptor buffer) {
-
-        if (buffer.storage.isReadOnly()) {
-            throw new ReadOnlyBufferException();
-        }
-
-        int result = nReadPixels(getNativeObject(), mNativeEngine,
-                xoffset, yoffset, width, height,
-                buffer.storage, buffer.storage.remaining(),
-                buffer.left, buffer.top, buffer.type.ordinal(), buffer.alignment,
-                buffer.stride, buffer.format.ordinal(),
-                buffer.handler, buffer.callback);
-
-        if (result < 0) {
-            throw new BufferOverflowException();
-        }
-    }
-
    /**
     * Returns the presentation time of the currently displayed frame in nanosecond.
     *
@@ -387,7 +277,6 @@ public class Stream {
    private static native long nCreateBuilder();
    private static native void nDestroyBuilder(long nativeStreamBuilder);
    private static native void nBuilderStreamSource(long nativeStreamBuilder, Object streamSource);
-    private static native void nBuilderStream(long nativeStreamBuilder, long externalTextureId);
    private static native void nBuilderWidth(long nativeStreamBuilder, int width);
    private static native void nBuilderHeight(long nativeStreamBuilder, int height);
    private static native long nBuilderBuild(long nativeStreamBuilder, long nativeEngine);
@@ -99,8 +99,8 @@ JNIEXPORT jint JNI_OnLoad(JavaVM* vm, void*) {

    int rc;

-    // KTXLoader
-    jclass ktxloaderClass = env->FindClass("com/google/android/filament/utils/KTXLoader");
+    // KTX1Loader
+    jclass ktxloaderClass = env->FindClass("com/google/android/filament/utils/KTX1Loader");
    if (ktxloaderClass == nullptr) return JNI_ERR;
    static const JNINativeMethod ktxMethods[] = {
        {(char*)"nCreateKTXTexture", (char*)"(JLjava/nio/Buffer;IZ)J", reinterpret_cast<void*>(nCreateKTXTexture)},
@@ -40,7 +40,7 @@ import com.google.android.filament.Renderer;
 *
 * When executing a test, clients should call tick() after each frame is rendered, which provides an
 * opportunity to push settings to Filament, increment the current test index (if enough time has
- * elapsed), and request an asychronous screenshot.
+ * elapsed), and request an asynchronous screenshot.
 *
 * The time to sleep between tests is configurable and can be set to zero. Automation also waits a
 * specified minimum number of frames between tests.
@@ -24,12 +24,12 @@ import com.google.android.filament.Texture
 import java.nio.Buffer

 /**
- * Utilities for consuming KTX files and producing Filament textures, IBLs, and sky boxes.
+ * Utilities for consuming KTX1 files and producing Filament textures, IBLs, and sky boxes.
 *
 * KTX is a simple container format that makes it easy to bundle miplevels and cubemap faces
 * into a single file.
 */
-object KTXLoader {
+object KTX1Loader {
    class Options {
        var srgb = false
    }
@@ -11,6 +11,18 @@ add_library(dracodec STATIC IMPORTED)
 set_target_properties(dracodec PROPERTIES IMPORTED_LOCATION
        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libdracodec.a)

+add_library(ktxreader STATIC IMPORTED)
+set_target_properties(ktxreader PROPERTIES IMPORTED_LOCATION
+        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libktxreader.a)
+
+add_library(stb STATIC IMPORTED)
+set_target_properties(stb PROPERTIES IMPORTED_LOCATION
+        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libstb.a)
+
+add_library(basis_transcoder STATIC IMPORTED)
+set_target_properties(basis_transcoder PROPERTIES IMPORTED_LOCATION
+        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libbasis_transcoder.a)
+
 add_library(utils STATIC IMPORTED)
 set_target_properties(utils PROPERTIES IMPORTED_LOCATION
        ${FILAMENT_DIR}/lib/${ANDROID_ABI}/libutils.a)
@@ -32,6 +44,7 @@ set(GLTFIO_SRCS
        ${GLTFIO_DIR}/include/gltfio/ResourceLoader.h
        ${GLTFIO_DIR}/include/gltfio/FilamentAsset.h
        ${GLTFIO_DIR}/include/gltfio/FilamentInstance.h
+        ${GLTFIO_DIR}/include/gltfio/TextureProvider.h
        ${GLTFIO_DIR}/include/gltfio/math.h

        ${GLTFIO_DIR}/src/Animator.cpp
@@ -45,15 +58,16 @@ set(GLTFIO_SRCS
        ${GLTFIO_DIR}/src/FFilamentInstance.h
        ${GLTFIO_DIR}/src/FilamentInstance.cpp
        ${GLTFIO_DIR}/src/GltfEnums.h
+        ${GLTFIO_DIR}/src/Ktx2Provider.cpp
        ${GLTFIO_DIR}/src/MaterialProvider.cpp
        ${GLTFIO_DIR}/src/ResourceLoader.cpp
+        ${GLTFIO_DIR}/src/StbProvider.cpp
        ${GLTFIO_DIR}/src/TangentsJob.h
        ${GLTFIO_DIR}/src/TangentsJob.cpp
        ${GLTFIO_DIR}/src/UbershaderLoader.cpp
        ${GLTFIO_DIR}/src/Wireframe.cpp
        ${GLTFIO_DIR}/src/Wireframe.h
        ${GLTFIO_DIR}/src/upcast.h
-        ${GLTFIO_DIR}/src/Image.cpp

        src/main/cpp/Animator.cpp
        src/main/cpp/AssetLoader.cpp
@@ -81,6 +95,7 @@ set(GLTFIO_INCLUDE_DIRS
        ../../third_party/hat-trie
        ../../third_party/stb
        ../../libs/utils/include
+        ../../libs/ktxreader/include
 )

 add_library(gltfio-jni SHARED ${GLTFIO_SRCS})
@@ -90,9 +105,9 @@ set_target_properties(gltfio-jni PROPERTIES LINK_DEPENDS ${CMAKE_CURRENT_SOURCE_

 if(GLTFIO_LITE)
        target_compile_definitions(gltfio-jni PUBLIC GLTFIO_LITE=1)
-        target_link_libraries(gltfio-jni filament-jni utils log gltfio_resources_lite)
+        target_link_libraries(gltfio-jni filament-jni utils log stb ktxreader basis_transcoder gltfio_resources_lite)
 else()
-        target_link_libraries(gltfio-jni filament-jni utils log gltfio_resources)
+        target_link_libraries(gltfio-jni filament-jni utils log stb ktxreader basis_transcoder gltfio_resources)

        # Enable Draco in the non-lite variant of gltfio.
        target_link_libraries(gltfio-jni dracodec)
@@ -36,6 +36,12 @@ Java_com_google_android_filament_gltfio_Animator_nUpdateBoneMatrices(JNIEnv*, jc
    animator->updateBoneMatrices();
 }

+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_gltfio_Animator_nResetBoneMatrices(JNIEnv*, jclass, jlong nativeAnimator) {
+    Animator* animator = (Animator*) nativeAnimator;
+    animator->resetBoneMatrices();
+}
+
 extern "C" JNIEXPORT jint JNICALL
 Java_com_google_android_filament_gltfio_Animator_nGetAnimationCount(JNIEnv*, jclass, jlong nativeAnimator) {
    Animator* animator = (Animator*) nativeAnimator;
@@ -19,6 +19,7 @@
 #include <filament/Engine.h>

 #include <gltfio/ResourceLoader.h>
+#include <gltfio/TextureProvider.h>

 #include <utils/Log.h>

@@ -114,3 +115,34 @@ Java_com_google_android_filament_gltfio_ResourceLoader_nAsyncCancelLoad(JNIEnv*,
    ResourceLoader* loader = (ResourceLoader*) nativeLoader;
    loader->asyncCancelLoad();
 }
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_com_google_android_filament_gltfio_ResourceLoader_nCreateStbProvider(JNIEnv*, jclass,
+        jlong nativeEngine) {
+    Engine* engine = (Engine*) nativeEngine;
+    return (jlong) createStbProvider(engine);
+}
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_com_google_android_filament_gltfio_ResourceLoader_nCreateKtx2Provider(JNIEnv*, jclass,
+        jlong nativeEngine) {
+    Engine* engine = (Engine*) nativeEngine;
+    return (jlong) createKtx2Provider(engine);
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_gltfio_ResourceLoader_nDestroyTextureProvider(JNIEnv*, jclass,
+        jlong nativeProvider) {
+    TextureProvider* provider = (TextureProvider*) nativeProvider;
+    delete provider;
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_google_android_filament_gltfio_ResourceLoader_nAddTextureProvider(JNIEnv* env, jclass,
+        jlong nativeLoader, jstring url, jlong nativeProvider) {
+    ResourceLoader* loader = (ResourceLoader*) nativeLoader;
+    TextureProvider* provider = (TextureProvider*) nativeProvider;
+    const char* cstring = env->GetStringUTFChars(url, nullptr);
+    loader->addTextureProvider(cstring, provider);
+    env->ReleaseStringUTFChars(url, cstring);
+}
@@ -66,6 +66,15 @@ public class Animator {
        nUpdateBoneMatrices(getNativeObject());
    }

+    /**
+     * Pass the identity matrix into all bone nodes, useful for returning to the T pose.
+     *
+     * <p>NOTE: this operation is independent of <code>animation</code>.</p>
+     */
+    public void resetBoneMatrices() {
+        nResetBoneMatrices(getNativeObject());
+    }
+
    /**
     * Returns the number of <code>animation</code> definitions in the glTF asset.
     */
@@ -109,6 +118,7 @@ public class Animator {

    private static native void nApplyAnimation(long nativeAnimator, int index, float time);
    private static native void nUpdateBoneMatrices(long nativeAnimator);
+    private static native void nResetBoneMatrices(long nativeAnimator);
    private static native int nGetAnimationCount(long nativeAnimator);
    private static native float nGetAnimationDuration(long nativeAnimator, int index);
    private static native String nGetAnimationName(long nativeAnimator, int index);
@@ -35,6 +35,8 @@ import java.nio.Buffer;
 */
 public class ResourceLoader {
    private final long mNativeObject;
+    private final long mNativeStbProvider;
+    private final long mNativeKtx2Provider;

    /**
     * Constructs a resource loader tied to the given Filament engine.
@@ -46,6 +48,11 @@ public class ResourceLoader {
    public ResourceLoader(@NonNull Engine engine) {
        long nativeEngine = engine.getNativeObject();
        mNativeObject = nCreateResourceLoader(nativeEngine, false, false, false);
+        mNativeStbProvider = nCreateStbProvider(nativeEngine);
+        mNativeKtx2Provider = nCreateKtx2Provider(nativeEngine);
+        nAddTextureProvider(mNativeObject, "image/jpeg", mNativeStbProvider);
+        nAddTextureProvider(mNativeObject, "image/png", mNativeStbProvider);
+        nAddTextureProvider(mNativeObject, "image/ktx2", mNativeKtx2Provider);
    }

    /**
@@ -63,6 +70,11 @@ public class ResourceLoader {
        long nativeEngine = engine.getNativeObject();
        mNativeObject = nCreateResourceLoader(nativeEngine, normalizeSkinningWeights,
                recomputeBoundingBoxes, ignoreBindTransform);
+        mNativeStbProvider = nCreateStbProvider(nativeEngine);
+        mNativeKtx2Provider = nCreateKtx2Provider(nativeEngine);
+        nAddTextureProvider(mNativeObject, "image/jpeg", mNativeStbProvider);
+        nAddTextureProvider(mNativeObject, "image/png", mNativeStbProvider);
+        nAddTextureProvider(mNativeObject, "image/ktx2", mNativeKtx2Provider);
    }

    /**
@@ -70,6 +82,8 @@ public class ResourceLoader {
     */
    public void destroy() {
        nDestroyResourceLoader(mNativeObject);
+        nDestroyTextureProvider(mNativeStbProvider);
+        nDestroyTextureProvider(mNativeKtx2Provider);
    }

    /**
@@ -178,4 +192,9 @@ public class ResourceLoader {
    private static native float nAsyncGetLoadProgress(long nativeLoader);
    private static native void nAsyncUpdateLoad(long nativeLoader);
    private static native void nAsyncCancelLoad(long nativeLoader);
+
+    private static native long nCreateStbProvider(long nativeEngine);
+    private static native long nCreateKtx2Provider(long nativeEngine);
+    private static native void nAddTextureProvider(long nativeLoader, String url, long nativeProvider);
+    private static native void nDestroyTextureProvider(long nativeProvider);
 }
@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.21.2
+VERSION_NAME=1.21.3

 POM_DESCRIPTION=Real-time physically based rendering engine for Android.

@@ -150,12 +150,12 @@ class MainActivity : Activity() {
        val scene = modelViewer.scene
        val ibl = "default_env"
        readCompressedAsset("envs/$ibl/${ibl}_ibl.ktx").let {
-            scene.indirectLight = KTXLoader.createIndirectLight(engine, it)
+            scene.indirectLight = KTX1Loader.createIndirectLight(engine, it)
            scene.indirectLight!!.intensity = 30_000.0f
            viewerContent.indirectLight = modelViewer.scene.indirectLight
        }
        readCompressedAsset("envs/$ibl/${ibl}_skybox.ktx").let {
-            scene.skybox = KTXLoader.createSkybox(engine, it)
+            scene.skybox = KTX1Loader.createSkybox(engine, it)
        }
    }

@@ -297,9 +297,10 @@ class MainActivity : Activity() {

        val gltfBuffer = pathToBufferMapping[gltfPath]!!

-        // The gltf is often not at the root level (e.g. if a folder is zipped) so
-        // we need to extract its path in order to resolve the embedded uri strings.
-        var prefix = URI(gltfPath!!).resolve("..")
+        // In a zip file, the gltf file might be in the same folder as resources, or in a different
+        // folder. It is crucial to test against both of these cases. In any case, the resource
+        // paths are all specified relative to the location of the gltf file.
+        var prefix = URI(gltfPath!!).resolve(".")

        withContext(Dispatchers.Main) {
            if (gltfPath!!.endsWith(".glb")) {
@@ -308,7 +309,7 @@ class MainActivity : Activity() {
                modelViewer.loadModelGltf(gltfBuffer) { uri ->
                    val path = prefix.resolve(uri).toString()
                    if (!pathToBufferMapping.contains(path)) {
-                        Log.e(TAG, "Could not find '$uri' in zip using prefix '$prefix'")
+                        Log.e(TAG, "Could not find '$uri' in zip using prefix '$prefix' and base path '${gltfPath!!}'")
                        setStatusText("Zip is missing $path")
                    }
                    pathToBufferMapping[path]
@@ -79,8 +79,6 @@ class MainActivity : Activity(), ActivityCompat.OnRequestPermissionsResultCallba
    // Performs the rendering and schedules new frames
    private val frameScheduler = FrameCallback()

-    private var externalTextureID: Int = 0
-
    @RequiresApi(30)
    class Api30Impl {
        companion object {
@@ -103,8 +101,6 @@ class MainActivity : Activity(), ActivityCompat.OnRequestPermissionsResultCallba
        setupView()
        setupScene()

-        externalTextureID = createExternalTexture()
-
        @Suppress("deprecation")
        val display = if (Build.VERSION.SDK_INT >= 30) {
            Api30Impl.getDisplay(this)
@@ -112,7 +108,7 @@ class MainActivity : Activity(), ActivityCompat.OnRequestPermissionsResultCallba
            windowManager.defaultDisplay!!
        }

-        streamHelper = StreamHelper(engine, materialInstance, display, externalTextureID)
+        streamHelper = StreamHelper(engine, materialInstance, display)
        this.title = streamHelper.getTestName()
    }

@@ -445,31 +441,4 @@ class MainActivity : Activity(), ActivityCompat.OnRequestPermissionsResultCallba
        check(EGL14.eglMakeCurrent(display, surface, surface, context)) { "Error making GL context." }
        return context
    }
-
-    private  fun createExternalTexture(): Int {
-        val textures = IntArray(1)
-        GLES30.glGenTextures(1, textures, 0)
-        val result = textures[0]
-
-        val textureTarget = GLES11Ext.GL_TEXTURE_EXTERNAL_OES
-        GLES30.glBindTexture(textureTarget, result)
-        GLES30.glTexParameteri(textureTarget, GLES30.GL_TEXTURE_WRAP_S, GLES30.GL_CLAMP_TO_EDGE)
-        GLES30.glTexParameteri(textureTarget, GLES30.GL_TEXTURE_WRAP_T, GLES30.GL_CLAMP_TO_EDGE)
-        GLES30.glTexParameteri(textureTarget, GLES30.GL_TEXTURE_MIN_FILTER, GLES30.GL_NEAREST)
-        GLES30.glTexParameteri(textureTarget, GLES30.GL_TEXTURE_MAG_FILTER, GLES30.GL_NEAREST)
-
-        if (!GLES30.glIsTexture(result)) {
-            throw RuntimeException("OpenGL error: $result is an invalid texture.")
-        }
-
-        val error = GLES30.glGetError()
-        if (error != GLES30.GL_NO_ERROR) {
-            val errorString = GLU.gluErrorString(error)
-            throw RuntimeException("OpenGL error: $errorString!")
-        }
-
-        return result
-    }
-
-
 }
@@ -37,7 +37,6 @@ class StreamHelper(
    private val filamentEngine: Engine,
    private val filamentMaterial: MaterialInstance,
    private val display: Display,
-    private val externalTextureId: Int
 ) {
    /**
     * The StreamSource configures the source data for the texture.
@@ -51,7 +50,6 @@ class StreamHelper(
     */
    enum class StreamSource {
        CANVAS_STREAM_NATIVE,     // copy-free but does not guarantee synchronization
-        CANVAS_STREAM_TEXID,      // synchronized but incurs a copy
        CANVAS_STREAM_ACQUIRED,   // synchronized and copy-free
    }

@@ -127,10 +125,6 @@ class StreamHelper(

            surface.unlockCanvasAndPost(canvas)

-            if (streamSource == StreamSource.CANVAS_STREAM_TEXID) {
-                surfaceTexture!!.updateTexImage()
-            }
-
            if (streamSource == StreamSource.CANVAS_STREAM_ACQUIRED) {
                val image = imageReader!!.acquireLatestImage()
                filamentStream!!.setAcquiredImage(
@@ -146,7 +140,7 @@ class StreamHelper(

    fun nextTest() {
        stopTest()
-        streamSource = StreamSource.values()[(streamSource.ordinal + 1) % 3]
+        streamSource = StreamSource.values()[(streamSource.ordinal + 1) % StreamSource.values().size]
        startTest()
    }

@@ -213,23 +207,6 @@ class StreamHelper(
            filamentTexture.setExternalStream(filamentEngine, filamentStream!!)
        }

-        if (streamSource == StreamSource.CANVAS_STREAM_TEXID) {
-
-            // Create the Android surface that will hold the canvas image.
-            surfaceTexture = SurfaceTexture(externalTextureId)
-            surfaceTexture!!.setDefaultBufferSize(resolution.width, resolution.height)
-            canvasSurface = Surface(surfaceTexture)
-
-            // Create the Filament Stream object that gets bound to the Texture.
-            filamentStream = Stream.Builder()
-                    .stream(externalTextureId.toLong())
-                    .width(resolution.width)
-                    .height(resolution.height)
-                    .build(filamentEngine)
-
-            filamentTexture.setExternalStream(filamentEngine, filamentStream!!)
-        }
-
        if (streamSource == StreamSource.CANVAS_STREAM_ACQUIRED) {
            filamentStream = Stream.Builder()
                    .width(resolution.width)
@@ -98,9 +98,9 @@ function print_matdbg_help {
    echo ""
    echo "FOR ANDROID BUILDS:"
    echo ""
-    echo "1) The most reliable way to enable matdbg is to bypass Gradle and"
-    echo "   directly modify the appropriate two lines in the following file:"
-    echo "       ./android/filament-android/CMakeLists.txt"
+    echo "1) For Android Studio builds, make sure to set:"
+    echo "       -Pcom.google.android.filament.matdbg"
+    echo "   option in Preferences > Build > Compiler > Command line options."
    echo ""
    echo "2) The port number is hardcoded to 8081 so you will need to do:"
    echo "       adb forward tcp:8081 tcp:8081"
@@ -156,6 +156,7 @@ VULKAN_ANDROID_GRADLE_OPTION=""
 SWIFTSHADER_OPTION="-DFILAMENT_USE_SWIFTSHADER=OFF"

 MATDBG_OPTION="-DFILAMENT_ENABLE_MATDBG=OFF"
+MATDBG_GRADLE_OPTION=""

 IOS_BUILD_SIMULATOR=false
 BUILD_UNIVERSAL_LIBRARIES=false
@@ -303,7 +304,7 @@ function build_webgl_with_target {
 }

 function build_webgl {
-    # For the host tools, supress install and always use Release.
+    # For the host tools, suppress install and always use Release.
    local old_install_command=${INSTALL_COMMAND}; INSTALL_COMMAND=
    local old_issue_debug_build=${ISSUE_DEBUG_BUILD}; ISSUE_DEBUG_BUILD=false
    local old_issue_release_build=${ISSUE_RELEASE_BUILD}; ISSUE_RELEASE_BUILD=true
@@ -402,7 +403,7 @@ function ensure_android_build {
 function build_android {
    ensure_android_build

-    # Supress intermediate desktop tools install
+    # Suppress intermediate desktop tools install
    local old_install_command=${INSTALL_COMMAND}
    INSTALL_COMMAND=

@@ -456,6 +457,7 @@ function build_android {
            -Pcom.google.android.filament.dist-dir=../out/android-debug/filament \
            -Pcom.google.android.filament.abis=${ABI_GRADLE_OPTION} \
            ${VULKAN_ANDROID_GRADLE_OPTION} \
+            ${MATDBG_GRADLE_OPTION} \
            :filament-android:assembleDebug \
            :gltfio-android:assembleDebug \
            :filament-utils-android:assembleDebug
@@ -504,6 +506,7 @@ function build_android {
            -Pcom.google.android.filament.dist-dir=../out/android-release/filament \
            -Pcom.google.android.filament.abis=${ABI_GRADLE_OPTION} \
            ${VULKAN_ANDROID_GRADLE_OPTION} \
+            ${MATDBG_GRADLE_OPTION} \
            :filament-android:assembleRelease \
            :gltfio-android:assembleRelease \
            :filament-utils-android:assembleRelease
@@ -598,7 +601,7 @@ function archive_ios {
 }

 function build_ios {
-    # Supress intermediate desktop tools install
+    # Suppress intermediate desktop tools install
    local old_install_command=${INSTALL_COMMAND}
    INSTALL_COMMAND=

@@ -754,6 +757,7 @@ while getopts ":hacCfijmp:q:uvslwtdk:" opt; do
        d)
            PRINT_MATDBG_HELP=true
            MATDBG_OPTION="-DFILAMENT_ENABLE_MATDBG=ON, -DFILAMENT_DISABLE_MATOPT=ON, -DFILAMENT_BUILD_FILAMAT=ON"
+            MATDBG_GRADLE_OPTION="-Pcom.google.android.filament.matdbg"
            ;;
        f)
            ISSUE_CMAKE_ALWAYS=true
@@ -17,7 +17,7 @@ export PATH="$PWD:$PATH"
 # npm install -g typescript

 # Install emscripten.
-curl -L https://github.com/emscripten-core/emsdk/archive/refs/tags/3.1.5.zip > emsdk.zip
+curl -L https://github.com/emscripten-core/emsdk/archive/refs/tags/3.1.9.zip > emsdk.zip
 unzip emsdk.zip ; mv emsdk-* emsdk ; cd emsdk
 ./emsdk install latest
 ./emsdk activate latest
@@ -2169,7 +2169,7 @@ type aliases:

                 Name               |   Type   |            Description
 :-----------------------------------|:--------:|:------------------------------------
-**getResolution()**                 | float4   |  Resolution of the view in pixels: `width`, `height`, `1 / width`, `1 / height`
+**getResolution()**                 | float4   |  Dimensions of the view's effective viewport in pixels: `width`, `height`, `1 / width`, `1 / height`. This might be different from `View::getViewport()` for instance because of added rendering guard-bands. This can be used in conjunction with `getNormalizedViewportCoord()` to generate pixel coordinates.
 **getWorldCameraPosition()**        | float3   |  Position of the camera/eye in world space
 **getWorldOffset()**                | float3   |  The shift required to obtain API-level world space
 **getTime()**                       | float    |  Current time as a remainder of 1 second. Yields a value between 0 and 1
@@ -77,6 +77,7 @@ set(SRCS
        src/RenderTarget.cpp
        src/RenderableManager.cpp
        src/Renderer.cpp
+        src/RendererUtils.cpp
        src/ResourceAllocator.cpp
        src/ResourceList.cpp
        src/Scene.cpp
@@ -143,6 +144,7 @@ set(PRIVATE_HDRS
        src/PerViewUniforms.h
        src/PIDController.h
        src/PostProcessManager.h
+        src/RendererUtils.h
        src/RenderPass.h
        src/RenderPrimitive.h
        src/ResourceAllocator.h
@@ -65,8 +65,6 @@ if (FILAMENT_SUPPORTS_OPENGL AND NOT FILAMENT_USE_EXTERNAL_GLES3 AND NOT FILAMEN
            src/opengl/gl_headers.h
            src/opengl/GLUtils.cpp
            src/opengl/GLUtils.h
-            src/opengl/OpenGLBlitter.cpp
-            src/opengl/OpenGLBlitter.h
            src/opengl/OpenGLContext.cpp
            src/opengl/OpenGLContext.h
            src/opengl/OpenGLDriver.cpp
@@ -89,7 +87,6 @@ if (FILAMENT_SUPPORTS_OPENGL AND NOT FILAMENT_USE_EXTERNAL_GLES3 AND NOT FILAMEN
    endif ()
    if (ANDROID)
        list(APPEND SRCS src/opengl/platforms/ExternalStreamManagerAndroid.cpp)
-        list(APPEND SRCS src/android/ExternalTextureManagerAndroid.cpp)
        list(APPEND SRCS src/opengl/platforms/PlatformEGLAndroid.cpp)
    elseif (IOS)
        list(APPEND SRCS src/opengl/platforms/PlatformCocoaTouchGL.mm)
@@ -42,22 +42,22 @@
 */
 namespace filament::backend {

-static constexpr uint64_t SWAP_CHAIN_CONFIG_TRANSPARENT = 0x1;
-static constexpr uint64_t SWAP_CHAIN_CONFIG_READABLE = 0x2;
-static constexpr uint64_t SWAP_CHAIN_CONFIG_ENABLE_XCB = 0x4;
+static constexpr uint64_t SWAP_CHAIN_CONFIG_TRANSPARENT         = 0x1;
+static constexpr uint64_t SWAP_CHAIN_CONFIG_READABLE            = 0x2;
+static constexpr uint64_t SWAP_CHAIN_CONFIG_ENABLE_XCB          = 0x4;
 static constexpr uint64_t SWAP_CHAIN_CONFIG_APPLE_CVPIXELBUFFER = 0x8;

-static constexpr size_t MAX_VERTEX_ATTRIBUTE_COUNT = 16; // This is guaranteed by OpenGL ES.
-static constexpr size_t MAX_VERTEX_SAMPLER_COUNT = 16;   // This is guaranteed by OpenGL ES.
-static constexpr size_t MAX_FRAGMENT_SAMPLER_COUNT = 16; // This is guaranteed by OpenGL ES.
-static constexpr size_t MAX_SAMPLER_COUNT = 32;          // This is guaranteed by OpenGL ES.
-static constexpr size_t MAX_VERTEX_BUFFER_COUNT = 16;    // Max number of bound buffer objects.
+static constexpr size_t MAX_VERTEX_ATTRIBUTE_COUNT  = 16;   // This is guaranteed by OpenGL ES.
+static constexpr size_t MAX_VERTEX_SAMPLER_COUNT    = 16;   // This is guaranteed by OpenGL ES.
+static constexpr size_t MAX_FRAGMENT_SAMPLER_COUNT  = 16;   // This is guaranteed by OpenGL ES.
+static constexpr size_t MAX_SAMPLER_COUNT           = 32;   // This is guaranteed by OpenGL ES.
+static constexpr size_t MAX_VERTEX_BUFFER_COUNT     = 16;   // Max number of bound buffer objects.

 static_assert(MAX_VERTEX_BUFFER_COUNT <= MAX_VERTEX_ATTRIBUTE_COUNT,
        "The number of buffer objects that can be attached to a VertexBuffer must be "
        "less than or equal to the maximum number of vertex attributes.");

-static constexpr size_t CONFIG_BINDING_COUNT = 8;
+static constexpr size_t CONFIG_BINDING_COUNT = 12;  // This is guaranteed by OpenGL ES.

 /**
 * Selects which driver a particular Engine should use.
@@ -788,7 +788,6 @@ enum class BlendFunction : uint8_t {
 //! Stream for external textures
 enum class StreamType {
    NATIVE,     //!< Not synchronized but copy-free. Good for video.
-    TEXTURE_ID, //!< Synchronized, but GL-only and incurs copies. Good for AR on devices before API 26.
    ACQUIRED,   //!< Synchronized, copy-free, and take a release callback. Good for AR but requires API 26+.
 };

@@ -913,7 +912,7 @@ struct ShaderStageFlags {
               (fragment && type == ShaderType::FRAGMENT);
    }
 };
-static constexpr ShaderStageFlags ALL_SHADER_STAGE_FLAGS = { .vertex = true, .fragment = true };
+static constexpr ShaderStageFlags ALL_SHADER_STAGE_FLAGS = { true, true };

 /**
 * Selects which buffers to clear at the beginning of the render pass, as well as which buffers
@@ -246,11 +246,6 @@ DECL_DRIVER_API_R_N(backend::SwapChainHandle, createSwapChainHeadless,
        uint32_t, height,
        uint64_t, flags)

-DECL_DRIVER_API_R_N(backend::StreamHandle, createStreamFromTextureId,
-        intptr_t, externalTextureId,
-        uint32_t, width,
-        uint32_t, height)
-
 DECL_DRIVER_API_R_0(backend::TimerQueryHandle, createTimerQuery)


@@ -461,14 +456,6 @@ DECL_DRIVER_API_N(readPixels,
        uint32_t, height,
        backend::PixelBufferDescriptor&&, data)

-DECL_DRIVER_API_N(readStreamPixels,
-        backend::StreamHandle, sh,
-        uint32_t, x,
-        uint32_t, y,
-        uint32_t, width,
-        uint32_t, height,
-        backend::PixelBufferDescriptor&&, data)
-
 /*
 * Rendering operations
 * --------------------
@@ -82,15 +82,6 @@ public:
    virtual void detach(Stream* stream) noexcept = 0;
    virtual void updateTexImage(Stream* stream, int64_t* timestamp) noexcept = 0;

-    // external texture storage
-    virtual ExternalTexture* createExternalTextureStorage() noexcept = 0;
-
-    // this is called synchronously in the application thread (NOT the Driver thread)
-    virtual void reallocateExternalStorage(ExternalTexture* ets,
-            uint32_t w, uint32_t h, TextureFormat format) noexcept = 0;
-
-    virtual void destroyExternalTextureStorage(ExternalTexture* ets) noexcept = 0;
-
    // The method allows platforms to convert a user-supplied external image object into a new type
    // (e.g. HardwareBuffer => EGLImage). It makes sense for the default implementation to do nothing.
    virtual AcquiredImage transformAcquiredImage(AcquiredImage source) noexcept { return source; }
@@ -1,222 +0,0 @@
-/*
- * Copyright (C) 2018 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-
-#include "ExternalTextureManagerAndroid.h"
-
-#include <utils/api_level.h>
-#include <utils/compiler.h>
-#include <utils/Log.h>
-
-#include <dlfcn.h>
-
-using namespace utils;
-
-namespace filament::backend {
-
-template <typename T>
-static void loadSymbol(T*& pfn, const char *symbol) noexcept {
-    pfn = (T*)dlsym(RTLD_DEFAULT, symbol);
-}
-
-// ------------------------------------------------------------------------------------------------
-
-namespace ndk {
-/*
- * This mimics the GraphicBuffer class for pre-API 26
- */
-template<typename NATIVE_TYPE, typename REF>
-struct ANativeObjectBase : public NATIVE_TYPE, public REF {
-};
-
-struct android_native_base_t {
-    unsigned int magic;
-    unsigned int version;
-    void* reserved[4];
-    void (* incRef)(struct android_native_base_t* base);
-    void (* decRef)(struct android_native_base_t* base);
-};
-
-struct ANativeWindowBuffer {
-    struct android_native_base_t common;
-    unsigned int width;
-    unsigned int height;
-    unsigned int stride;
-    unsigned int format;
-    unsigned int usage_deprecated;
-    uintptr_t layerCount;
-    void* reserved[1];
-    const void* handle;
-    uint64_t usage;
-    void* reserved_proc[8 - (sizeof(uint64_t) / sizeof(void*))];
-};
-
-struct RefBase {
-    virtual ~RefBase() = default;
-    void* ref;
-};
-
-struct GraphicBuffer : public ANativeObjectBase<ANativeWindowBuffer, RefBase> {
-};
-struct GraphicBufferWrapper {
-    GraphicBuffer* graphicBuffer;
-};
-} // namespace ndk
-
-// ------------------------------------------------------------------------------------------------
-using namespace ndk;
-
-struct EGLExternalTexture : public ExternalTextureManagerAndroid::ExternalTexture {
-    GraphicBufferWrapper* graphicBufferWrapper = nullptr;
-};
-
-ExternalTextureManagerAndroid& ExternalTextureManagerAndroid::create() noexcept {
-    return *(new ExternalTextureManagerAndroid{});
-}
-
-void ExternalTextureManagerAndroid::destroy(ExternalTextureManagerAndroid* pExternalTextureManager) noexcept {
-    delete pExternalTextureManager;
-}
-
-// called on gl thread
-ExternalTextureManagerAndroid::ExternalTextureManagerAndroid() noexcept
-        : mVm(VirtualMachineEnv::get()) {
-}
-
-// not quite sure on which thread this is going to be called
-ExternalTextureManagerAndroid::~ExternalTextureManagerAndroid() noexcept {
-    if (__builtin_available(android 26, *)) {
-    } else {
-        if (mGraphicBufferClass) {
-            JNIEnv* env = VirtualMachineEnv::getThreadEnvironment();
-            if (env) {
-                env->DeleteGlobalRef(mGraphicBufferClass);
-            }
-        }
-    }
-}
-
-// called on gl thread
-Platform::ExternalTexture* ExternalTextureManagerAndroid::createExternalTexture() noexcept {
-    if (__builtin_available(android 26, *)) {
-    } else {
-        // initialize java stuff on-demand
-        if (!mGraphicBufferClass) {
-            JNIEnv* env = mVm.getEnvironment();
-            mGraphicBufferClass = env->FindClass("android/view/GraphicBuffer");
-            mGraphicBuffer_nCreateGraphicBuffer = env->GetStaticMethodID(
-                    mGraphicBufferClass, "nCreateGraphicBuffer", "(IIII)J");
-            mGraphicBuffer_nDestroyGraphicBuffer = env->GetStaticMethodID(
-                    mGraphicBufferClass, "nDestroyGraphicBuffer", "(J)V");
-
-            mGraphicBufferClass = static_cast<jclass>(env->NewGlobalRef(mGraphicBufferClass));
-        }
-    }
-    EGLExternalTexture* ets = new EGLExternalTexture;
-    return ets;
-}
-
-// called on app thread
-void ExternalTextureManagerAndroid::reallocate(
-        Platform::ExternalTexture* ets, uint32_t w, uint32_t h,
-        TextureFormat format, uint64_t usage) noexcept {
-    destroyStorage(ets);
-    alloc(ets, w, h, format, usage);
-}
-
-// called on gl thread
-void ExternalTextureManagerAndroid::destroy(Platform::ExternalTexture* ets) noexcept {
-    destroyStorage(ets);
-    delete static_cast<EGLExternalTexture*>(ets);
-}
-
-// called on app thread
-void ExternalTextureManagerAndroid::alloc(
-        Platform::ExternalTexture* t,
-        uint32_t w, uint32_t h, TextureFormat format, uint64_t usage) noexcept {
-
-    EGLExternalTexture* ets = static_cast<EGLExternalTexture*>(t);
-
-    AHardwareBuffer_Desc desc = { w, h, 1, 0, usage, 0, 0, 0 };
-    switch (format) {
-        case TextureFormat::RGB8:
-            // don't use R8G8B8 here because some drivers produce garbled images
-            desc.format = AHARDWAREBUFFER_FORMAT_R8G8B8X8_UNORM;
-            break;
-        case TextureFormat::RGBA8:
-            desc.format = AHARDWAREBUFFER_FORMAT_R8G8B8A8_UNORM;
-            break;
-        case TextureFormat::RGB565:
-            desc.format = AHARDWAREBUFFER_FORMAT_R5G6B5_UNORM;
-            break;
-        case TextureFormat::RGB10_A2:
-            desc.format = AHARDWAREBUFFER_FORMAT_R10G10B10A2_UNORM;
-            break;
-        default:
-            slog.e << "Unsupported format " << (int)format << ", use RGBA8 or RGB8 only."
-                   << io::endl;
-            return;
-    }
-
-    if (__builtin_available(android 26, *)) {
-        // allocate new storage...
-        AHardwareBuffer* buffer = nullptr;
-        if (AHardwareBuffer_allocate(&desc, &buffer) < 0) {
-            slog.e << "AHardwareBuffer_allocate() failed" << io::endl;
-            return;
-        }
-        ets->hardwareBuffer = buffer;
-    } else {
-        // note: This is called on the application thread (not the GL thread)
-        JNIEnv* env = VirtualMachineEnv::getThreadEnvironment();
-        if (!env) {
-            return; // this should not happen
-        }
-
-        jlong graphicBufferWrapperJni = env->CallStaticLongMethod(mGraphicBufferClass,
-                mGraphicBuffer_nCreateGraphicBuffer, w, h, desc.format, usage);
-        VirtualMachineEnv::handleException(env);
-
-        if (graphicBufferWrapperJni) {
-            GraphicBuffer* const gb = ((GraphicBufferWrapper*)graphicBufferWrapperJni)->graphicBuffer;
-            ets->graphicBufferWrapper = (GraphicBufferWrapper*)graphicBufferWrapperJni;
-            ets->clientBuffer = static_cast<ANativeWindowBuffer*>(gb);
-        }
-    }
-}
-
-// called on gl thread
-void ExternalTextureManagerAndroid::destroyStorage(Platform::ExternalTexture* t) noexcept {
-        EGLExternalTexture* ets = static_cast<EGLExternalTexture*>(t);
-    if (__builtin_available(android 26, *)) {
-        // destroy the current storage if any
-        if (ets->hardwareBuffer) {
-            AHardwareBuffer_release(ets->hardwareBuffer);
-            ets->hardwareBuffer = nullptr;
-        }
-    } else {
-        if (ets->graphicBufferWrapper) {
-            JNIEnv* env = VirtualMachineEnv::get().getEnvironment();
-            env->CallStaticVoidMethod(mGraphicBufferClass,
-                    mGraphicBuffer_nDestroyGraphicBuffer, (jlong)ets->graphicBufferWrapper);
-            VirtualMachineEnv::handleException(env);
-            ets->graphicBufferWrapper = nullptr;
-        }
-    }
-}
-
-} // namespace filament::backend
-
@@ -1,84 +0,0 @@
-/*
- * Copyright (C) 2018 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef TNT_FILAMENT_DRIVER_ANDROID_EXTERNAL_TEXTURE_MANAGER_ANDROID_H
-#define TNT_FILAMENT_DRIVER_ANDROID_EXTERNAL_TEXTURE_MANAGER_ANDROID_H
-
-#include "private/backend/VirtualMachineEnv.h"
-
-#include <backend/DriverEnums.h>
-#include <backend/Platform.h>
-
-#include <android/api-level.h>
-#include <android/hardware_buffer.h>
-
-namespace filament::backend {
-
-/*
- * ExternalTextureManagerAndroid::ExternalTexture is basically a wrapper for AHardwareBuffer.
- *
- * This class doesn't rely on GL or EGL, and could be used for other Android platform if needed
- * (e.g. Vulkan).
- *
- * ExternalTextureManagerAndroid handle allocation/destruction using either Java or the NDK,
- * whichever is available.
- */
-class ExternalTextureManagerAndroid {
-public:
-
-    struct ExternalTexture : public Platform::ExternalTexture {
-        void* clientBuffer = nullptr;
-        AHardwareBuffer* hardwareBuffer = nullptr;
-    };
-
-    // must be called on backend thread
-    static ExternalTextureManagerAndroid& create() noexcept;
-
-    // must be called on backend thread
-    static void destroy(ExternalTextureManagerAndroid* pExternalTextureManager) noexcept;
-
-    // must be called on backend thread (only because we don't synchronize
-    Platform::ExternalTexture* createExternalTexture() noexcept;
-
-    // called on app thread
-    void reallocate(
-        Platform::ExternalTexture* ets, uint32_t w, uint32_t h,
-        TextureFormat format, uint64_t usage) noexcept;
-
-    // must be called on backend thread
-    void destroy(Platform::ExternalTexture* ets) noexcept;
-
-private:
-    ExternalTextureManagerAndroid() noexcept;
-    ~ExternalTextureManagerAndroid() noexcept;
-
-    // called on app thread
-    void alloc(Platform::ExternalTexture* ets,
-               uint32_t w, uint32_t h, TextureFormat format, uint64_t usage) noexcept;
-
-    // called on gl thread
-    void destroyStorage(Platform::ExternalTexture* ets) noexcept;
-
-    VirtualMachineEnv& mVm;
-
-    jclass mGraphicBufferClass = nullptr;
-    jmethodID mGraphicBuffer_nCreateGraphicBuffer = nullptr;
-    jmethodID mGraphicBuffer_nDestroyGraphicBuffer = nullptr;
-};
-
-} // namespace filament::backend
-
-#endif // TNT_FILAMENT_DRIVER_ANDROID_EXTERNAL_TEXTURE_MANAGER_ANDROID_H
@@ -359,10 +359,6 @@ void MetalDriver::createSwapChainHeadlessR(Handle<HwSwapChain> sch,
    construct_handle<MetalSwapChain>(sch, *mContext, width, height, flags);
 }

-void MetalDriver::createStreamFromTextureIdR(Handle<HwStream>, intptr_t externalTextureId,
-        uint32_t width, uint32_t height) {
-}
-
 void MetalDriver::createTimerQueryR(Handle<HwTimerQuery> tqh, int) {
    // nothing to do, timer query was constructed in createTimerQueryS
 }
@@ -431,10 +427,6 @@ Handle<HwSwapChain> MetalDriver::createSwapChainHeadlessS() noexcept {
    return alloc_handle<MetalSwapChain>();
 }

-Handle<HwStream> MetalDriver::createStreamFromTextureIdS() noexcept {
-    return {};
-}
-
 Handle<HwTimerQuery> MetalDriver::createTimerQueryS() noexcept {
    // The handle must be constructed here, as a synchronous call to getTimerQueryValue might happen
    // before createTimerQueryR is executed.
@@ -1053,11 +1045,6 @@ void MetalDriver::readPixels(Handle<HwRenderTarget> src, uint32_t x, uint32_t y,
    }];
 }

-void MetalDriver::readStreamPixels(Handle<HwStream> sh, uint32_t x, uint32_t y, uint32_t width,
-        uint32_t height, PixelBufferDescriptor&& data) {
-
-}
-
 void MetalDriver::blit(TargetBufferFlags buffers,
        Handle<HwRenderTarget> dst, Viewport dstRect,
        Handle<HwRenderTarget> src, Viewport srcRect,
@@ -367,7 +367,7 @@ MetalProgram::MetalProgram(id<MTLDevice> device, const Program& program) noexcep
        *shaderFunctions[i] = [library newFunctionWithName:@"main0"];
    }

-    // All stages of the program have compiled successfuly, this is a valid program.
+    // All stages of the program have compiled successfully, this is a valid program.
    isValid = true;

    // This calculates Metal resource binding indices. Filament's sampler bindings range from 0-31
@@ -170,7 +170,7 @@ private:
 };

 // StateTracker keeps track of state changes made to a Metal command encoder.
-// Different kinds of state, like pipeline state, uniform buffer state, etc, are passed to the
+// Different kinds of state, like pipeline state, uniform buffer state, etc., are passed to the
 // current Metal command encoder and persist throughout the lifetime of the encoder (a frame).
 // StateTracker is used to prevent calling redundant state change methods.
 template<typename StateType>
@@ -308,11 +308,6 @@ void NoopDriver::readPixels(Handle<HwRenderTarget> src,
    scheduleDestroy(std::move(p));
 }

-void NoopDriver::readStreamPixels(Handle<HwStream> sh, uint32_t x, uint32_t y, uint32_t width,
-        uint32_t height, PixelBufferDescriptor&& p) {
-    scheduleDestroy(std::move(p));
-}
-
 void NoopDriver::blit(TargetBufferFlags buffers,
        Handle<HwRenderTarget> dst, Viewport dstRect,
        Handle<HwRenderTarget> src, Viewport srcRect,
@@ -42,9 +42,11 @@ void assertFramebufferStatus(utils::io::ostream& out, GLenum target, const char*

 #ifdef NDEBUG
 #   define CHECK_GL_ERROR(out)
+#   define CHECK_GL_ERROR_NON_FATAL(out)
 #   define CHECK_GL_FRAMEBUFFER_STATUS(out, target)
 #else
 #   define CHECK_GL_ERROR(out) { GLUtils::assertGLError(out, __func__, __LINE__); }
+#   define CHECK_GL_ERROR_NON_FATAL(out) { GLUtils::checkGLError(out, __func__, __LINE__); }
 #   define CHECK_GL_FRAMEBUFFER_STATUS(out, target) { GLUtils::checkFramebufferStatus(out, target, __func__, __LINE__); }
 #endif

@@ -1,215 +0,0 @@
-/*
- * Copyright (C) 2017 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 "OpenGLBlitter.h"
-
-#include "GLUtils.h"
-#include "OpenGLContext.h"
-
-#include <utils/compiler.h>
-#include <utils/Log.h>
-
-#include <math/vec2.h>
-
-using namespace filament::math;
-using namespace utils;
-
-namespace filament::backend {
-
-static const char s_vertexES[] = R"SHADER(#version 300 es
-in vec4 pos;
-void main() {
-    gl_Position = pos;
-}
-)SHADER";
-
-static const char s_vertexGL[] = R"SHADER(#version 410 core
-in vec4 pos;
-void main() {
-    gl_Position = pos;
-}
-)SHADER";
-
-static const char s_fragmentES[] = R"SHADER(#version 300 es
-#extension GL_OES_EGL_image_external_essl3 : enable
-precision mediump float;
-uniform samplerExternalOES sampler;
-out vec4 fragColor;
-void main() {
-    fragColor = texelFetch(sampler, ivec2(gl_FragCoord.xy), 0);
-}
-)SHADER";
-
-static const char s_fragmentGL[] = R"SHADER(#version 410 core
-precision mediump float;
-uniform sampler2D sampler;
-out vec4 fragColor;
-void main() {
-    fragColor = texelFetch(sampler, ivec2(gl_FragCoord.xy), 0);
-}
-)SHADER";
-
-void OpenGLBlitter::init() noexcept {
-    glGenSamplers(1, &mSampler);
-    glSamplerParameteri(mSampler, GL_TEXTURE_MIN_FILTER,   GL_NEAREST);
-    glSamplerParameteri(mSampler, GL_TEXTURE_MAG_FILTER,   GL_NEAREST);
-    glSamplerParameteri(mSampler, GL_TEXTURE_WRAP_S,       GL_CLAMP_TO_EDGE);
-    glSamplerParameteri(mSampler, GL_TEXTURE_WRAP_T,       GL_CLAMP_TO_EDGE);
-    glSamplerParameteri(mSampler, GL_TEXTURE_WRAP_R,       GL_CLAMP_TO_EDGE);
-
-    GLint status;
-    char const* vsource[2] = { s_vertexES, s_vertexGL };
-    char const* fsource[2] = { s_fragmentES, s_fragmentGL };
-    const size_t index = BACKEND_OPENGL_VERSION == BACKEND_OPENGL_VERSION_GLES ? 0 : 1;
-
-    mVertexShader = glCreateShader(GL_VERTEX_SHADER);
-    glShaderSource(mVertexShader, 1, vsource + index, nullptr);
-    glCompileShader(mVertexShader);
-    glGetShaderiv(mVertexShader, GL_COMPILE_STATUS, &status);
-    assert_invariant(status == GL_TRUE);
-
-    mFragmentShader = glCreateShader(GL_FRAGMENT_SHADER);
-    glShaderSource(mFragmentShader, 1, fsource + index, nullptr);
-    glCompileShader(mFragmentShader);
-    glGetShaderiv(mFragmentShader, GL_COMPILE_STATUS, &status);
-    assert_invariant(status == GL_TRUE);
-
-    mProgram = glCreateProgram();
-    glAttachShader(mProgram, mVertexShader);
-    glAttachShader(mProgram, mFragmentShader);
-    glLinkProgram(mProgram);
-    glGetProgramiv(mProgram, GL_LINK_STATUS, &status);
-    assert_invariant(status == GL_TRUE);
-
-    glUseProgram(mProgram);
-    GLint loc = glGetUniformLocation(mProgram, "sampler");
-    GLuint tmu = 0;
-    glUniform1i(loc, tmu);
-
-    glGenFramebuffers(1, &mFBO);
-
-    CHECK_GL_ERROR(utils::slog.e)
-}
-
-void OpenGLBlitter::terminate() noexcept {
-    glDeleteSamplers(1, &mSampler);
-    glDetachShader(mProgram, mVertexShader);
-    glDetachShader(mProgram, mFragmentShader);
-    glDeleteShader(mVertexShader);
-    glDeleteShader(mFragmentShader);
-    glDeleteProgram(mProgram);
-    glDeleteFramebuffers(1, &mFBO);
-}
-
-void OpenGLBlitter::blit(GLuint srcTextureExternal, GLuint dstTexture2d, GLuint w, GLuint h) noexcept {
-    const float2 vtx[3] = {{ -1.0f,  3.0f },
-                           { -1.0f, -1.0f },
-                           {  3.0f, -1.0f }};
-
-    // we're using tmu 0 as the source texture
-    GLuint tmu = 0;
-
-    // source texture
-    glBindSampler(tmu, mSampler);
-    glBindTexture(GL_TEXTURE_EXTERNAL_OES, srcTextureExternal);
-    CHECK_GL_ERROR(utils::slog.e)
-
-    // destination texture
-    glBindFramebuffer(GL_FRAMEBUFFER, mFBO);
-    glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, dstTexture2d, 0);
-    CHECK_GL_ERROR(utils::slog.e)
-    CHECK_GL_FRAMEBUFFER_STATUS(utils::slog.e, GL_FRAMEBUFFER)
-
-    // geometry
-    glBindBuffer(GL_ARRAY_BUFFER, 0);
-    glEnableVertexAttribArray(0);
-    glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 0, vtx);
-
-    // blit...
-    glViewport(0, 0, w, h);
-    glUseProgram(mProgram);
-    glDrawArrays(GL_TRIANGLES, 0, 3);
-
-    CHECK_GL_ERROR(utils::slog.e)
-}
-
-void OpenGLBlitter::State::save() noexcept {
-    mHasState = true;
-    // TODO: technically we should also save glVertexAttribPointer
-    GLuint tmu = 0;
-    glGetIntegerv(GL_DRAW_FRAMEBUFFER_BINDING, &framebuffer);
-    glGetIntegerv(GL_ELEMENT_ARRAY_BUFFER_BINDING, &array);
-    glGetIntegerv(GL_CURRENT_PROGRAM, &program);
-    glGetIntegerv(GL_VIEWPORT, viewport);
-    glGetIntegerv(GL_COLOR_WRITEMASK, writeMask);
-    scissorTest = glIsEnabled(GL_SCISSOR_TEST);
-    stencilTest = glIsEnabled(GL_STENCIL_TEST);
-    cullFace = glIsEnabled(GL_CULL_FACE);
-    glGetVertexAttribiv(0, GL_VERTEX_ATTRIB_ARRAY_ENABLED, &vertexAttrib);
-
-    // save the current active texture first
-    glGetIntegerv(GL_ACTIVE_TEXTURE, &activeTexture);
-
-    // we're using tmu 0 as the source texture
-    glActiveTexture(GL_TEXTURE0 + tmu);
-
-    // save what depends on glActiveTexture
-    glGetIntegerv(GL_SAMPLER_BINDING, &sampler);
-    glGetIntegerv(GL_TEXTURE_BINDING_2D, &texture);
-
-    /*
-     * Set the state
-     */
-
-    // we're using tmu 0 as the source texture
-    glDisable(GL_SCISSOR_TEST);
-    glDisable(GL_STENCIL_TEST);
-    glDisable(GL_CULL_FACE);
-    glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
-    CHECK_GL_ERROR(utils::slog.e)
-}
-
-void OpenGLBlitter::State::restore() noexcept {
-    GLuint tmu = 0;
-    glColorMask(
-            (GLboolean)writeMask[0],
-            (GLboolean)writeMask[1],
-            (GLboolean)writeMask[2],
-            (GLboolean)writeMask[3]);
-    if (cullFace) {
-        glEnable(GL_CULL_FACE);
-    }
-    if (stencilTest) {
-        glEnable(GL_STENCIL_TEST);
-    }
-    if (scissorTest) {
-        glEnable(GL_SCISSOR_TEST);
-    }
-    glViewport(viewport[0], viewport[1], viewport[2], viewport[3]);
-    glUseProgram(static_cast<GLuint>(program));
-    glBindBuffer(GL_ARRAY_BUFFER, static_cast<GLuint>(array));
-    glBindFramebuffer(GL_FRAMEBUFFER, static_cast<GLuint>(framebuffer));
-    if (!vertexAttrib) {
-        glDisableVertexAttribArray(0);
-    }
-
-    glBindTexture(GL_TEXTURE_2D, static_cast<GLuint>(texture));
-    glBindSampler(tmu, static_cast<GLuint>(sampler));
-    glActiveTexture(static_cast<GLenum>(activeTexture));
-    CHECK_GL_ERROR(utils::slog.e)
-}
-
-} // namespace filament::backend
@@ -1,67 +0,0 @@
-/*
- * Copyright (C) 2017 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef TNT_FILAMENT_BACKEND_OPENGL_OPENGLBLITTER_H
-#define TNT_FILAMENT_BACKEND_OPENGL_OPENGLBLITTER_H
-
-#include <utils/compiler.h>
-#include "gl_headers.h"
-
-namespace filament::backend {
-
-class OpenGLContext;
-
-class OpenGLBlitter {
-public:
-    explicit OpenGLBlitter(OpenGLContext& context) noexcept : mContext(context) {}
-
-    void init() noexcept;
-    void terminate() noexcept;
-
-    void blit(GLuint srcTextureExternal, GLuint dstTexture2d, GLuint w, GLuint h) noexcept;
-
-    class State {
-        bool mHasState = false;
-        GLint activeTexture, sampler, texture, framebuffer, array, vertexAttrib, program;
-        GLboolean stencilTest, scissorTest, cullFace;
-        GLint viewport[4], writeMask[4];
-        void save() noexcept;
-        void restore() noexcept;
-    public:
-        void setup() noexcept {
-            if (UTILS_UNLIKELY(!mHasState)) {
-                save();
-            }
-        }
-        ~State() noexcept {
-            if (UTILS_UNLIKELY(mHasState)) {
-                restore();
-            }
-        }
-    };
-
-private:
-    UTILS_UNUSED OpenGLContext& mContext;
-    GLuint mSampler{};
-    GLuint mVertexShader{};
-    GLuint mFragmentShader{};
-    GLuint mProgram{};
-    GLuint mFBO{};
-};
-
-} // namespace filament::backend
-
-#endif // TNT_FILAMENT_BACKEND_OPENGL_OPENGLBLITTER_H
@@ -81,75 +81,87 @@ OpenGLContext::OpenGLContext() noexcept {
     * Figure out which driver bugs we need to workaround
     */

-    if (strstr(state.renderer, "Adreno")) {
-        // Qualcomm GPU
-        bugs.invalidate_end_only_if_invalidate_start = true;
+    const bool isAngle = strstr(state.renderer, "ANGLE");
+    if (!isAngle) {
+        if (strstr(state.renderer, "Adreno")) {
+            // Qualcomm GPU
+            bugs.invalidate_end_only_if_invalidate_start = true;

-        // On Adreno (As of 3/20) timer query seem to return the CPU time, not the GPU time.
-        bugs.dont_use_timer_query = true;
-
-        // Blits to texture arrays are failing
-        //   This bug continues to reproduce, though at times we've seen it appear to "go away". The
-        //   standalone sample app that was written to show this problem still reproduces.
-        //   The working hypthesis is that some other state affects this behavior.
-        bugs.disable_sidecar_blit_into_texture_array = true;
-
-        // early exit condition is flattened in EASU code
-        bugs.split_easu = true;
-
-        int maj, min, driverMajor, driverMinor;
-        int c = sscanf(state.version, "OpenGL ES %d.%d V@%d.%d", // NOLINT(cert-err34-c)
-                &maj, &min, &driverMajor, &driverMinor);
-        if (c == 4) {
-            // workarounds based on version here.
-            // notes:
-            //  bugs.invalidate_end_only_if_invalidate_start
-            //  - appeared at least in
-            //      "OpenGL ES 3.2 V@0490.0 (GIT@85da404, I46ff5fc46f, 1606794520) (Date:11/30/20)"
-            //  - wasn't present in
-            //      "OpenGL ES 3.2 V@0490.0 (GIT@0905e9f, Ia11ce2d146, 1599072951) (Date:09/02/20)"
-            //  - has been confirmed fixed in V@570.1 by Qualcomm
-            if (driverMajor < 490 || driverMajor > 570 ||
-                (driverMajor == 570 && driverMinor >= 1)) {
-                bugs.invalidate_end_only_if_invalidate_start = false;
-            }
-        }
-
-        // qualcomm seems to have no problem with this (which is good for us)
-        bugs.allow_read_only_ancillary_feedback_loop = true;
-    } else if (strstr(state.renderer, "Mali")) {
-        // ARM GPU
-        bugs.vao_doesnt_store_element_array_buffer_binding = true;
-        if (strstr(state.renderer, "Mali-T")) {
-            bugs.disable_glFlush = true;
-            bugs.disable_shared_context_draws = true;
-            bugs.texture_external_needs_rebind = true;
-            // We have not verified that timer queries work on Mali-T, so we disable to be safe.
+            // On Adreno (As of 3/20) timer query seem to return the CPU time, not the GPU time.
            bugs.dont_use_timer_query = true;
+
+            // Blits to texture arrays are failing
+            //   This bug continues to reproduce, though at times we've seen it appear to "go away".
+            //   The standalone sample app that was written to show this problem still reproduces.
+            //   The working hypthesis is that some other state affects this behavior.
+            bugs.disable_sidecar_blit_into_texture_array = true;
+
+            // early exit condition is flattened in EASU code
+            bugs.split_easu = true;
+
+            int maj, min, driverMajor, driverMinor;
+            int c = sscanf(state.version, "OpenGL ES %d.%d V@%d.%d", // NOLINT(cert-err34-c)
+                    &maj, &min, &driverMajor, &driverMinor);
+            if (c == 4) {
+                // workarounds based on version here.
+                // notes:
+                //  bugs.invalidate_end_only_if_invalidate_start
+                //  - appeared at least in
+                //      "OpenGL ES 3.2 V@0490.0 (GIT@85da404, I46ff5fc46f, 1606794520) (Date:11/30/20)"
+                //  - wasn't present in
+                //      "OpenGL ES 3.2 V@0490.0 (GIT@0905e9f, Ia11ce2d146, 1599072951) (Date:09/02/20)"
+                //  - has been confirmed fixed in V@570.1 by Qualcomm
+                if (driverMajor < 490 || driverMajor > 570 ||
+                    (driverMajor == 570 && driverMinor >= 1)) {
+                    bugs.invalidate_end_only_if_invalidate_start = false;
+                }
+            }
+
+            // qualcomm seems to have no problem with this (which is good for us)
+            bugs.allow_read_only_ancillary_feedback_loop = true;
+        } else if (strstr(state.renderer, "Mali")) {
+            // ARM GPU
+            bugs.vao_doesnt_store_element_array_buffer_binding = true;
+            if (strstr(state.renderer, "Mali-T")) {
+                bugs.disable_glFlush = true;
+                bugs.disable_shared_context_draws = true;
+                bugs.texture_external_needs_rebind = true;
+                // We have not verified that timer queries work on Mali-T, so we disable to be safe.
+                bugs.dont_use_timer_query = true;
+            }
+            if (strstr(state.renderer, "Mali-G")) {
+                // note: We have verified that timer queries work well at least on some Mali-G.
+            }
+            // Mali seems to have no problem with this (which is good for us)
+            bugs.allow_read_only_ancillary_feedback_loop = true;
+        } else if (strstr(state.renderer, "Intel")) {
+            // Intel GPU
+            bugs.vao_doesnt_store_element_array_buffer_binding = true;
+        } else if (strstr(state.renderer, "PowerVR")) {
+            // PowerVR GPU
+        } else if (strstr(state.renderer, "Apple")) {
+            // Apple GPU
+        } else if (strstr(state.renderer, "Tegra") ||
+                   strstr(state.renderer, "GeForce") ||
+                   strstr(state.renderer, "NV")) {
+            // NVIDIA GPU
+        } else if (strstr(state.renderer, "Vivante")) {
+            // Vivante GPU
+        } else if (strstr(state.renderer, "AMD") ||
+                   strstr(state.renderer, "ATI")) {
+            // AMD/ATI GPU
+        } else if (strstr(state.renderer, "Mozilla")) {
+            bugs.disable_invalidate_framebuffer = true;
        }
-        if (strstr(state.renderer, "Mali-G")) {
-            // note: We have verified that timer queries work well at least on some Mali-G.
+    } else {
+        // When running under ANGLE, it's a different set of workaround that we need.
+        if (strstr(state.renderer, "Adreno")) {
+            // Qualcomm GPU
+            // early exit condition is flattened in EASU code
+            // (that should be regardless of ANGLE, but we should double check)
+            bugs.split_easu = true;
        }
-        // Mali seems to have no problem with this (which is good for us)
-        bugs.allow_read_only_ancillary_feedback_loop = true;
-    } else if (strstr(state.renderer, "Intel")) {
-        // Intel GPU
-        bugs.vao_doesnt_store_element_array_buffer_binding = true;
-    } else if (strstr(state.renderer, "PowerVR")) {
-        // PowerVR GPU
-    } else if (strstr(state.renderer, "Apple")) {
-        // Apple GPU
-    } else if (strstr(state.renderer, "Tegra") ||
-               strstr(state.renderer, "GeForce") ||
-               strstr(state.renderer, "NV")) {
-        // NVIDIA GPU
-    } else if (strstr(state.renderer, "Vivante")) {
-        // Vivante GPU
-    } else if (strstr(state.renderer, "AMD") ||
-               strstr(state.renderer, "ATI")) {
-        // AMD/ATI GPU
-    } else if (strstr(state.renderer, "Mozilla")) {
-        bugs.disable_invalidate_framebuffer = true;
+        // TODO: see if we could use `bugs.allow_read_only_ancillary_feedback_loop = true`
    }

    slog.v << "Active workarounds: " << '\n';
@@ -294,7 +306,9 @@ void OpenGLContext::initExtensionsGLES() noexcept {
    ext.EXT_multisampled_render_to_texture = exts.has("GL_EXT_multisampled_render_to_texture"sv);
    ext.EXT_multisampled_render_to_texture2 = exts.has("GL_EXT_multisampled_render_to_texture2"sv);
    ext.EXT_shader_framebuffer_fetch = exts.has("GL_EXT_shader_framebuffer_fetch"sv);
+#if !defined(__EMSCRIPTEN__)
    ext.EXT_texture_compression_etc2 = true;
+#endif
    ext.EXT_texture_filter_anisotropic = exts.has("GL_EXT_texture_filter_anisotropic"sv);
    ext.GOOGLE_cpp_style_line_directive = exts.has("GL_GOOGLE_cpp_style_line_directive"sv);
    ext.KHR_debug = exts.has("GL_KHR_debug"sv);
@@ -20,7 +20,6 @@
 #include "private/backend/OpenGLPlatform.h"

 #include "CommandStreamDispatcher.h"
-#include "OpenGLBlitter.h"
 #include "OpenGLDriverFactory.h"
 #include "OpenGLProgram.h"
 #include "OpenGLTimerQuery.h"
@@ -168,13 +167,6 @@ OpenGLDriver::OpenGLDriver(OpenGLPlatform* platform) noexcept
    slog.i << "OS version: " << mPlatform.getOSVersion() << io::endl;
 #endif

-    // Initialize the blitter only if we have OES_EGL_image_external_essl3
-    if (mContext.ext.OES_EGL_image_external_essl3) {
-        mOpenGLBlitter = new OpenGLBlitter(mContext);
-        mOpenGLBlitter->init();
-        mContext.resetProgram();
-    }
-
    if (mContext.ext.EXT_disjoint_timer_query ||
            BACKEND_OPENGL_VERSION == BACKEND_OPENGL_VERSION_GL) {
        // timer queries are available
@@ -197,7 +189,6 @@ OpenGLDriver::OpenGLDriver(OpenGLPlatform* platform) noexcept
 }

 OpenGLDriver::~OpenGLDriver() noexcept {
-    delete mOpenGLBlitter;
 }

 Dispatcher OpenGLDriver::getDispatcher() const noexcept {
@@ -223,9 +214,6 @@ void OpenGLDriver::terminate() {
        glDeleteSamplers(1, &item.second);
    }
    mSamplerMap.clear();
-    if (mOpenGLBlitter) {
-        mOpenGLBlitter->terminate();
-    }

    delete mTimerQueryImpl;

@@ -381,10 +369,6 @@ Handle<HwSwapChain> OpenGLDriver::createSwapChainHeadlessS() noexcept {
    return initHandle<HwSwapChain>();
 }

-Handle<HwStream> OpenGLDriver::createStreamFromTextureIdS() noexcept {
-    return initHandle<GLStream>();
-}
-
 Handle<HwTimerQuery> OpenGLDriver::createTimerQueryS() noexcept {
    return initHandle<GLTimerQuery>();
 }
@@ -1137,24 +1121,6 @@ void OpenGLDriver::createSwapChainHeadlessR(Handle<HwSwapChain> sch,
    sc->swapChain = mPlatform.createSwapChain(width, height, flags);
 }

-void OpenGLDriver::createStreamFromTextureIdR(Handle<HwStream> sh,
-        intptr_t externalTextureId, uint32_t width, uint32_t height) {
-    DEBUG_MARKER()
-
-    GLStream* s = handle_cast<GLStream*>(sh);
-    // It would be better if we could query the externalTextureId size, unfortunately
-    // this is not supported in GL for GL_TEXTURE_EXTERNAL_OES targets
-    s->width = width;
-    s->height = height;
-    s->gl.externalTextureId = static_cast<GLuint>(externalTextureId);
-    s->streamType = StreamType::TEXTURE_ID;
-    glGenTextures(GLStream::ROUND_ROBIN_TEXTURE_COUNT, s->user_thread.read);
-    glGenTextures(GLStream::ROUND_ROBIN_TEXTURE_COUNT, s->user_thread.write);
-    for (auto& info : s->user_thread.infos) {
-        info.ets = mPlatform.createExternalTextureStorage();
-    }
-}
-
 void OpenGLDriver::createTimerQueryR(Handle<HwTimerQuery> tqh, int) {
    DEBUG_MARKER()

@@ -1301,15 +1267,6 @@ void OpenGLDriver::destroyStream(Handle<HwStream> sh) {
        }
        if (s->streamType == StreamType::NATIVE) {
            mPlatform.destroyStream(s->stream);
-        } else if (s->streamType == StreamType::TEXTURE_ID) {
-            glDeleteTextures(GLStream::ROUND_ROBIN_TEXTURE_COUNT, s->user_thread.read);
-            glDeleteTextures(GLStream::ROUND_ROBIN_TEXTURE_COUNT, s->user_thread.write);
-            if (s->gl.fbo) {
-                glDeleteFramebuffers(1, &s->gl.fbo);
-            }
-            for (auto const& info : s->user_thread.infos) {
-                mPlatform.destroyExternalTextureStorage(info.ets);
-            }
        }
        destruct(sh, s);
    }
@@ -1368,7 +1325,6 @@ void OpenGLDriver::setAcquiredImage(Handle<HwStream> sh, void* hwbuffer,

 void OpenGLDriver::updateStreams(DriverApi* driver) {
    if (UTILS_UNLIKELY(!mExternalStreams.empty())) {
-        OpenGLBlitter::State state;
        for (GLTexture* t : mExternalStreams) {
            assert_invariant(t);

@@ -1379,11 +1335,6 @@ void OpenGLDriver::updateStreams(DriverApi* driver) {
                continue;
            }

-            if (s->streamType == StreamType::TEXTURE_ID) {
-                state.setup();
-                updateStreamTexId(t, driver);
-            }
-
            if (s->streamType == StreamType::ACQUIRED) {
                updateStreamAcquired(t, driver);
            }
@@ -2080,20 +2031,12 @@ void OpenGLDriver::setExternalStream(Handle<HwTexture> th, Handle<HwStream> sh)

 UTILS_NOINLINE
 void OpenGLDriver::attachStream(GLTexture* t, GLStream* hwStream) noexcept {
-    auto& gl = mContext;
    mExternalStreams.push_back(t);

    switch (hwStream->streamType) {
        case StreamType::NATIVE:
            mPlatform.attach(hwStream->stream, t->gl.id);
            break;
-        case StreamType::TEXTURE_ID:
-            assert_invariant(t->target == SamplerType::SAMPLER_EXTERNAL);
-            // The texture doesn't need a texture name anymore, get rid of it
-            gl.unbindTexture(t->gl.target, t->gl.id);
-            glDeleteTextures(1, &t->gl.id);
-            t->gl.id = hwStream->user_thread.read[hwStream->user_thread.cur];
-            break;
        case StreamType::ACQUIRED:
            break;
    }
@@ -2115,8 +2058,6 @@ void OpenGLDriver::detachStream(GLTexture* t) noexcept {
            mPlatform.detach(t->hwStream->stream);
            // ^ this deletes the texture id
            break;
-        case StreamType::TEXTURE_ID:
-            break;
        case StreamType::ACQUIRED:
            gl.unbindTexture(t->gl.target, t->gl.id);
            glDeleteTextures(1, &t->gl.id);
@@ -2141,7 +2082,6 @@ void OpenGLDriver::replaceStream(GLTexture* texture, GLStream* newStream) noexce
            mPlatform.detach(texture->hwStream->stream);
            // ^ this deletes the texture id
            break;
-        case StreamType::TEXTURE_ID:
        case StreamType::ACQUIRED:
            break;
    }
@@ -2151,10 +2091,6 @@ void OpenGLDriver::replaceStream(GLTexture* texture, GLStream* newStream) noexce
            glGenTextures(1, &texture->gl.id);
            mPlatform.attach(newStream->stream, texture->gl.id);
            break;
-        case StreamType::TEXTURE_ID:
-            assert_invariant(texture->target == SamplerType::SAMPLER_EXTERNAL);
-            texture->gl.id = newStream->user_thread.read[newStream->user_thread.cur];
-            break;
        case StreamType::ACQUIRED:
            // Just re-use the old texture id.
            break;
@@ -2524,179 +2460,6 @@ void OpenGLDriver::updateStreamAcquired(GLTexture* gltexture, DriverApi* driver)
    });
 }

-#define DEBUG_NO_EXTERNAL_STREAM_COPY false
-
-void OpenGLDriver::updateStreamTexId(GLTexture* t, DriverApi* driver) noexcept {
-    SYSTRACE_CALL();
-    auto& gl = mContext;
-
-    GLStream* s = static_cast<GLStream*>(t->hwStream);
-    assert_invariant(s);
-    assert_invariant(s->streamType == StreamType::TEXTURE_ID);
-
-    // round-robin to the next texture name
-    if (UTILS_UNLIKELY(DEBUG_NO_EXTERNAL_STREAM_COPY ||
-                       gl.bugs.disable_shared_context_draws || !mOpenGLBlitter)) {
-        driver->queueCommand([this, t, s]() {
-            // the stream may have been destroyed since we enqueued the command
-            // also make sure that this texture is still associated with the same stream
-            auto& streams = mExternalStreams;
-            if (UTILS_LIKELY(std::find(streams.begin(), streams.end(), t) != streams.end()) &&
-                    (t->hwStream == s)) {
-                t->gl.id = s->gl.externalTextureId;
-            }
-        });
-    } else {
-        s->user_thread.cur = uint8_t(
-                (s->user_thread.cur + 1) % GLStream::ROUND_ROBIN_TEXTURE_COUNT);
-        GLuint writeTexture = s->user_thread.write[s->user_thread.cur];
-        GLuint readTexture = s->user_thread.read[s->user_thread.cur];
-
-        // Make sure we're using the proper size
-        GLStream::Info& info = s->user_thread.infos[s->user_thread.cur];
-        if (UTILS_UNLIKELY(info.width != s->width || info.height != s->height)) {
-
-            // nothing guarantees that this buffer is free (i.e. has been consumed by the
-            // GL thread), so we could potentially cause a glitch by reallocating the
-            // texture here. This should be very rare though.
-            // This could be fixed by always using a new temporary texture here, and
-            // replacing it in the queueCommand() below. imho, not worth it.
-
-            info.width = s->width;
-            info.height = s->height;
-
-            Platform::ExternalTexture* ets = s->user_thread.infos[s->user_thread.cur].ets;
-            mPlatform.reallocateExternalStorage(ets, info.width, info.height, TextureFormat::RGB8);
-
-            glActiveTexture(GL_TEXTURE0);
-            glBindTexture(GL_TEXTURE_2D, writeTexture);
-            glBindTexture(GL_TEXTURE_EXTERNAL_OES, readTexture);
-#ifdef GL_OES_EGL_image
-            glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, (GLeglImageOES)ets->image);
-            glEGLImageTargetTexture2DOES(GL_TEXTURE_EXTERNAL_OES, (GLeglImageOES)ets->image);
-#endif
-        }
-
-        // copy the texture...
-#ifndef NDEBUG
-        if (t->gl.fence) {
-            // we're about to overwrite a buffer that hasn't been consumed
-            slog.d << "OpenGLDriver::updateStream(): about to overwrite buffer " <<
-                   int(s->user_thread.cur) << " of Texture at " << t << " of Stream at " << s
-                   << io::endl;
-        }
-#endif
-        mOpenGLBlitter->blit(s->gl.externalTextureId, writeTexture, s->width, s->height);
-
-        // We need a fence to guarantee that this copy has happened when we need the texture
-        // in OpenGLProgram::updateSamplers(), i.e. when we bind textures just before use.
-        GLsync fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
-        // Per https://www.khronos.org/opengl/wiki/Sync_Object, flush to make sure that the
-        // sync object is in the driver's command queue.
-        glFlush();
-
-        // Update the stream timestamp. It's not clear to me that this is correct; which
-        // timestamp do we really want? Here we use "now" because we have nothing else we
-        // can use.
-        s->user_thread.timestamp = std::chrono::steady_clock::now().time_since_epoch().count();
-
-        driver->queueCommand([this, t, s, fence, readTexture, writeTexture]() {
-            // the stream may have been destroyed since we enqueued the command
-            // also make sure that this texture is still associated with the same stream
-            auto& streams = mExternalStreams;
-            if (UTILS_LIKELY(std::find(streams.begin(), streams.end(), t) != streams.end()) &&
-                (t->hwStream == s)) {
-                if (UTILS_UNLIKELY(t->gl.fence)) {
-                    // if the texture still has a fence set, destroy it now, so it's not leaked.
-                    glDeleteSync(t->gl.fence);
-                }
-                t->gl.id = readTexture;
-                t->gl.fence = fence;
-                s->gl.externalTexture2DId = writeTexture;
-            } else {
-                glDeleteSync(fence);
-            }
-        });
-    }
-}
-
-void OpenGLDriver::readStreamPixels(Handle<HwStream> sh,
-        uint32_t x, uint32_t y, uint32_t width, uint32_t height,
-        PixelBufferDescriptor&& p) {
-    DEBUG_MARKER()
-    auto& gl = mContext;
-
-    GLStream* s = handle_cast<GLStream*>(sh);
-
-    if (UTILS_UNLIKELY(s->streamType == StreamType::ACQUIRED)) {
-        PANIC_LOG("readStreamPixels with ACQUIRED streams is not yet implemented.");
-        return;
-    }
-
-    if (UTILS_LIKELY(s->streamType == StreamType::TEXTURE_ID)) {
-        GLuint tid = s->gl.externalTexture2DId;
-        if (tid == 0) {
-            return;
-        }
-
-        GLenum glFormat = getFormat(p.format);
-        GLenum glType = getType(p.type);
-
-        gl.pixelStore(GL_PACK_ROW_LENGTH, p.stride);
-        gl.pixelStore(GL_PACK_ALIGNMENT, p.alignment);
-        gl.pixelStore(GL_PACK_SKIP_PIXELS, p.left);
-        gl.pixelStore(GL_PACK_SKIP_ROWS, p.top);
-
-        if (s->gl.fbo == 0) {
-            glGenFramebuffers(1, &s->gl.fbo);
-        }
-        gl.bindFramebuffer(GL_FRAMEBUFFER, s->gl.fbo);
-
-        glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, tid, 0);
-        CHECK_GL_ERROR(utils::slog.e)
-
-        /*
-         * It looks like glReadPixels() behaves differently, or even wrongly,
-         * when the FBO is backed by an EXTERNAL texture...
-         *
-         *
-         *  External texture FBO           User buffer
-         *
-         *  O---+----------------+
-         *  |   |                |                .stride         .alignment
-         *  |   |                |         ----------------------->-->
-         *  |   | y              |         O----------------------+--+   low adresses
-         *  |   |                |         |          |           |  |
-         *  |   |         w      |         |          | .bottom   |  |
-         *  |   V   <--------->  |         |          V           |  |
-         *  |       +---------+  |         |     +---------+      |  |
-         *  |       |     ^   |  | ======> |     |         |      |  |
-         *  |   x   |    h|   |  |         |.left|         |      |  |
-         *  +------>|     v   |  |         +---->|         |      |  |
-         *  |       +.........+  |         |     +.........+      |  |
-         *  |                    |         |                      |  |
-         *  |                    |         +----------------------+--+  high adresses
-         *  +--------------------+
-         *
-         *  Origin is at the                The image is NOT y-reversed
-         *  top-left corner                 and bottom is counted from
-         *                                  the top! "bottom" is in fact treated
-         *                                  as "top".
-         */
-
-        // The filament API provides yoffset as the "bottom" offset, therefore it needs to
-        // be corrected to match glReadPixels()'s behavior.
-        y = (s->height - height) - y;
-
-        // TODO: we could use a PBO to make this asynchronous
-        glReadPixels(GLint(x), GLint(y), GLint(width), GLint(height), glFormat, glType, p.buffer);
-        CHECK_GL_ERROR(utils::slog.e)
-
-        gl.bindFramebuffer(GL_FRAMEBUFFER, 0);
-        scheduleDestroy(std::move(p));
-    }
-}
-
 // ------------------------------------------------------------------------------------------------
 // Setting rendering state
 // ------------------------------------------------------------------------------------------------
@@ -3223,7 +2986,11 @@ void OpenGLDriver::draw(PipelineState state, Handle<HwRenderPrimitive> rph, uint
                instanceCount);
    }

+#ifdef FILAMENT_ENABLE_MATDBG
+    CHECK_GL_ERROR_NON_FATAL(utils::slog.e)
+#else
    CHECK_GL_ERROR(utils::slog.e)
+#endif
 }

 // explicit instantiation of the Dispatcher
@@ -145,7 +145,6 @@ public:
    };

    struct GLStream : public HwStream {
-        static constexpr size_t ROUND_ROBIN_TEXTURE_COUNT = 3;      // 3 maximum
        using HwStream::HwStream;
        struct Info {
            // storage for the read/write textures below
@@ -153,29 +152,11 @@ public:
            GLuint width = 0;
            GLuint height = 0;
        };
-        struct {
-            // id of the texture where the external frames are streamed (i.e. the texture
-            // used by SurfaceTexture on Android)
-            GLuint externalTextureId = 0;
-
-            /*
-             * This is for making a cpu copy of the camera frame
-             */
-            GLuint externalTexture2DId = 0;
-            GLuint fbo = 0;
-        } gl;   // 20 bytes
-
-
        /*
         * The fields below are accessed from the main application thread
         * (not the GL thread)
         */
        struct {
-            // texture id used to texture from, always used in the GL thread
-            GLuint read[ROUND_ROBIN_TEXTURE_COUNT];     // 12 bytes
-            // texture id to write into, always used from the user thread
-            GLuint write[ROUND_ROBIN_TEXTURE_COUNT];    // 12 bytes
-            Info infos[ROUND_ROBIN_TEXTURE_COUNT];      // 48 bytes
            int64_t timestamp = 0;
            uint8_t cur = 0;
            AcquiredImage acquired;
@@ -364,7 +345,7 @@ private:
    void setViewportScissor(Viewport const& viewportScissor) noexcept;

    // sampler buffer binding points (nullptr if not used)
-    std::array<HwSamplerGroup*, Program::BINDING_COUNT> mSamplerBindings = {};   // 8 pointers
+    std::array<HwSamplerGroup*, Program::BINDING_COUNT> mSamplerBindings = {};   // 12 pointers

    mutable tsl::robin_map<uint32_t, GLuint> mSamplerMap;
    mutable std::vector<GLTexture*> mExternalStreams;
@@ -375,8 +356,6 @@ private:

    OpenGLPlatform& mPlatform;

-    OpenGLBlitter* mOpenGLBlitter = nullptr;
-    void updateStreamTexId(GLTexture* t, DriverApi* driver) noexcept;
    void updateStreamAcquired(GLTexture* t, DriverApi* driver) noexcept;
    void updateBuffer(GLBufferObject* buffer, BufferDescriptor const& p,
            uint32_t byteOffset, uint32_t alignment = 16) noexcept;
@@ -264,7 +264,7 @@ void OpenGLProgram::initialize(OpenGLContext& context) {
    mValid = OpenGLProgram::checkProgramStatus(name.c_str_safe(),
            gl.program, gl.shaders, initializationData->shaderSourceCode);

-    if (mValid) {
+    if (UTILS_LIKELY(mValid)) {
        initializeProgramState(context, gl.program,
                initializationData->uniformBlockInfo,
                initializationData->samplerGroupInfo);
@@ -60,7 +60,7 @@ public:

            // turns out the former might be relatively cheap to check, the later requires
            // a bit less. Compared to what updateSamplers() actually does, which is
-            // pretty little, I'm not sure we'll get ahead.
+            // pretty little, I'm not sure if we'll get ahead.

            updateSamplers(gld);
        }
@@ -113,7 +113,7 @@ private:
    union {
        // when mInitialized == true:
        // information about each USED sampler buffer per binding (no gaps)
-        std::array<uint8_t, Program::BINDING_COUNT> mUsedBindingPoints;   // 8 bytes
+        std::array<uint8_t, Program::BINDING_COUNT> mUsedBindingPoints;   // 12 bytes
        // when mInitialized == false:
        // lazy initialization data pointer
        LazyInitializationData* mLazyInitializationData;
@@ -55,11 +55,6 @@ public:
    void detach(Stream* stream) noexcept final {}
    void updateTexImage(Stream* stream, int64_t* timestamp) noexcept final {}

-    ExternalTexture* createExternalTextureStorage() noexcept final { return nullptr; }
-    void reallocateExternalStorage(ExternalTexture* ets,
-            uint32_t w, uint32_t h, TextureFormat format) noexcept final { }
-    void destroyExternalTextureStorage(ExternalTexture* ets) noexcept final { }
-
    int getOSVersion() const noexcept final { return 0; }

    bool pumpEvents() noexcept override;
@@ -56,11 +56,6 @@ public:
    void detach(Stream* stream) noexcept final {}
    void updateTexImage(Stream* stream, int64_t* timestamp) noexcept final {}

-    ExternalTexture* createExternalTextureStorage() noexcept final { return nullptr; }
-    void reallocateExternalStorage(ExternalTexture* ets,
-            uint32_t w, uint32_t h, TextureFormat format) noexcept final { }
-    void destroyExternalTextureStorage(ExternalTexture* ets) noexcept final { }
-
    int getOSVersion() const noexcept final { return 0; }

    bool setExternalImage(void* externalImage, void* texture) noexcept final;
@@ -51,11 +51,6 @@ public:
    void detach(Stream* stream) noexcept final override {}
    void updateTexImage(Stream* stream, int64_t* timestamp) noexcept final override {}

-    ExternalTexture* createExternalTextureStorage() noexcept final override { return nullptr; }
-    void reallocateExternalStorage(ExternalTexture* ets,
-            uint32_t w, uint32_t h, TextureFormat format) noexcept final override { }
-    void destroyExternalTextureStorage(ExternalTexture* ets) noexcept final override { }
-
    int getOSVersion() const noexcept final override { return 0; }
 };

@@ -62,11 +62,6 @@ public:
    void detach(Stream* stream) noexcept override {}
    void updateTexImage(Stream* stream, int64_t* timestamp) noexcept override {}

-    ExternalTexture* createExternalTextureStorage() noexcept override { return nullptr; }
-    void reallocateExternalStorage(ExternalTexture* ets,
-            uint32_t w, uint32_t h, TextureFormat format) noexcept override {}
-    void destroyExternalTextureStorage(ExternalTexture* ets) noexcept override {}
-
 protected:
    static void logEglError(const char* name) noexcept;

@@ -19,7 +19,6 @@
 #include "opengl/OpenGLDriver.h"
 #include "opengl/OpenGLContext.h"

-#include "android/ExternalTextureManagerAndroid.h"
 #include "ExternalStreamManagerAndroid.h"
 #include "private/backend/VirtualMachineEnv.h"

@@ -73,8 +72,7 @@ using EGLStream = Platform::Stream;

 PlatformEGLAndroid::PlatformEGLAndroid() noexcept
        : PlatformEGL(),
-          mExternalStreamManager(ExternalStreamManagerAndroid::create()),
-          mExternalTextureManager(ExternalTextureManagerAndroid::create()) {
+          mExternalStreamManager(ExternalStreamManagerAndroid::create()) {

    char scratch[PROP_VALUE_MAX + 1];
    int length = __system_property_get("ro.build.version.release", scratch);
@@ -85,6 +83,18 @@ PlatformEGLAndroid::PlatformEGLAndroid() noexcept
        length = __system_property_get("ro.build.version.sdk", scratch);
        mOSVersion = length >= 0 ? atoi(scratch) : 1;
    }
+
+    // This disables an ANGLE optimization on ARM, which turns out to be more costly for us
+    // see b/229017581
+    // We need to do this before we create the GL context.
+    // An alternative solution is use a system property:
+    //            __system_property_set(
+    //                    "debug.angle.feature_overrides_disabled",
+    //                    "preferSubmitAtFBOBoundary");
+    // but that would outlive this process, so the environment variable is better.
+    // We also make sure to not update the variable if it already exists.
+    // There is no harm setting this if we're not on ANGLE or ARM.
+    setenv("ANGLE_FEATURE_OVERRIDES_DISABLED", "preferSubmitAtFBOBoundary", false);
 }

 PlatformEGLAndroid::~PlatformEGLAndroid() noexcept = default;
@@ -92,7 +102,6 @@ PlatformEGLAndroid::~PlatformEGLAndroid() noexcept = default;

 void PlatformEGLAndroid::terminate() noexcept {
    ExternalStreamManagerAndroid::destroy(&mExternalStreamManager);
-    ExternalTextureManagerAndroid::destroy(&mExternalTextureManager);
    PlatformEGL::terminate();
 }

@@ -155,58 +164,6 @@ void PlatformEGLAndroid::updateTexImage(Stream* stream, int64_t* timestamp) noex
    mExternalStreamManager.updateTexImage(stream, timestamp);
 }

-Platform::ExternalTexture* PlatformEGLAndroid::createExternalTextureStorage() noexcept {
-    return mExternalTextureManager.createExternalTexture();
-}
-
-void PlatformEGLAndroid::reallocateExternalStorage(
-        Platform::ExternalTexture* externalTexture,
-        uint32_t w, uint32_t h, TextureFormat format) noexcept {
-    if (externalTexture) {
-        if ((EGLImageKHR)externalTexture->image != EGL_NO_IMAGE_KHR) {
-            eglDestroyImageKHR(mEGLDisplay, (EGLImageKHR)externalTexture->image);
-            externalTexture->image = (uintptr_t)EGL_NO_IMAGE_KHR;
-        }
-
-        mExternalTextureManager.reallocate(externalTexture, w, h, format,
-                AHARDWAREBUFFER_USAGE_GPU_COLOR_OUTPUT | AHARDWAREBUFFER_USAGE_GPU_SAMPLED_IMAGE);
-
-        auto ets = (ExternalTextureManagerAndroid::ExternalTexture*)externalTexture;
-        EGLClientBuffer clientBuffer;
-        if (ets->hardwareBuffer) {
-            clientBuffer = eglGetNativeClientBufferANDROID(ets->hardwareBuffer);
-            if (UTILS_UNLIKELY(!clientBuffer)) {
-                logEglError("eglGetNativeClientBufferANDROID");
-                return;
-            }
-        } else if (ets->clientBuffer) {
-            clientBuffer = (EGLClientBuffer)ets->clientBuffer;
-        } else {
-            // woops, reallocate failed.
-            return;
-        }
-
-        const EGLint attr[] = { EGL_NONE };
-        EGLImageKHR image = eglCreateImageKHR(mEGLDisplay,
-                EGL_NO_CONTEXT, EGL_NATIVE_BUFFER_ANDROID, clientBuffer, attr);
-        if (UTILS_UNLIKELY(!image)) {
-            logEglError("eglCreateImageKHR");
-        }
-        ets->image = (uintptr_t)image;
-    }
-}
-
-void PlatformEGLAndroid::destroyExternalTextureStorage(
-        Platform::ExternalTexture* externalTexture) noexcept {
-    if (externalTexture) {
-        mExternalTextureManager.destroy(externalTexture);
-        if ((EGLImageKHR)externalTexture->image != EGL_NO_IMAGE_KHR) {
-            eglDestroyImageKHR(mEGLDisplay, (EGLImageKHR)externalTexture->image);
-            externalTexture->image = (uintptr_t)EGL_NO_IMAGE_KHR;
-        }
-    }
-}
-
 int PlatformEGLAndroid::getOSVersion() const noexcept {
    return mOSVersion;
 }
@@ -22,7 +22,6 @@
 namespace filament::backend {

 class ExternalStreamManagerAndroid;
-class ExternalTextureManagerAndroid;

 class PlatformEGLAndroid final : public PlatformEGL {
 public:
@@ -44,17 +43,11 @@ public:
    void detach(Stream* stream) noexcept final;
    void updateTexImage(Stream* stream, int64_t* timestamp) noexcept final;

-    ExternalTexture* createExternalTextureStorage() noexcept final;
-    void reallocateExternalStorage(ExternalTexture* ets,
-            uint32_t w, uint32_t h, TextureFormat format) noexcept final;
-    void destroyExternalTextureStorage(ExternalTexture* ets) noexcept final;
-
    AcquiredImage transformAcquiredImage(AcquiredImage source) noexcept final;

 private:
    int mOSVersion;
    ExternalStreamManagerAndroid& mExternalStreamManager;
-    ExternalTextureManagerAndroid& mExternalTextureManager;
 };

 } // namespace filament::backend
@@ -55,11 +55,6 @@ public:
    void detach(Stream* stream) noexcept final override {}
    void updateTexImage(Stream* stream, int64_t* timestamp) noexcept final override {}

-    ExternalTexture* createExternalTextureStorage() noexcept final override { return nullptr; }
-    void reallocateExternalStorage(ExternalTexture* ets,
-            uint32_t w, uint32_t h, TextureFormat format) noexcept final override { }
-    void destroyExternalTextureStorage(ExternalTexture* ets) noexcept final override { }
-
    int getOSVersion() const noexcept final override { return 0; }

 private:
@@ -51,11 +51,6 @@ public:
    void detach(Stream* stream) noexcept final override {}
    void updateTexImage(Stream* stream, int64_t* timestamp) noexcept final override {}

-    ExternalTexture* createExternalTextureStorage() noexcept final override { return nullptr; }
-    void reallocateExternalStorage(ExternalTexture* ets,
-            uint32_t w, uint32_t h, TextureFormat format) noexcept final override { }
-    void destroyExternalTextureStorage(ExternalTexture* ets) noexcept final override { }
-
    int getOSVersion() const noexcept final override { return 0; }

 private:
@@ -52,11 +52,6 @@ public:
    void detach(Stream* stream) noexcept final override {}
    void updateTexImage(Stream* stream, int64_t* timestamp) noexcept final override {}

-    ExternalTexture* createExternalTextureStorage() noexcept final override { return nullptr; }
-    void reallocateExternalStorage(ExternalTexture* ets,
-            uint32_t w, uint32_t h, TextureFormat format) noexcept final override { }
-    void destroyExternalTextureStorage(ExternalTexture* ets) noexcept final override { }
-
    int getOSVersion() const noexcept final override { return 0; }
 };

@@ -608,10 +608,6 @@ void VulkanDriver::createSwapChainHeadlessR(Handle<HwSwapChain> sch,
    construct<VulkanSwapChain>(sch, mContext, mStagePool, width, height);
 }

-void VulkanDriver::createStreamFromTextureIdR(Handle<HwStream> sh, intptr_t externalTextureId,
-        uint32_t width, uint32_t height) {
-}
-
 void VulkanDriver::createTimerQueryR(Handle<HwTimerQuery> tqh, int) {
    // nothing to do, timer query was constructed in createTimerQueryS
 }
@@ -678,10 +674,6 @@ Handle<HwSwapChain> VulkanDriver::createSwapChainHeadlessS() noexcept {
    return allocHandle<VulkanSwapChain>();
 }

-Handle<HwStream> VulkanDriver::createStreamFromTextureIdS() noexcept {
-    return {};
-}
-
 Handle<HwTimerQuery> VulkanDriver::createTimerQueryS() noexcept {
    // The handle must be constructed here, as a synchronous call to getTimerQueryValue might happen
    // before createTimerQueryR is executed.
@@ -1608,14 +1600,8 @@ void VulkanDriver::readPixels(Handle<HwRenderTarget> src, uint32_t x, uint32_t y
    vkMapMemory(device, stagingMemory, 0, VK_WHOLE_SIZE, 0, (void**) &srcPixels);
    srcPixels += subResourceLayout.offset;

-    // NOTE: the reasons for this are unclear, but issuing ReadPixels on a VkImage that has been
-    // extracted from a swap chain does not need a Y flip, but explicitly created VkImages do. (The
-    // former can be tested with "Export Screenshots" in gltf_viewer, the latter can be tested with
-    // test_ReadPixels.cpp). We've seen this behavior with both SwiftShader and MoltenVK.
-    const bool flipY = !srcTarget->isSwapChain();
-
    if (!DataReshaper::reshapeImage(&pbd, getComponentType(srcFormat), srcPixels,
-            subResourceLayout.rowPitch, width, height, swizzle, flipY)) {
+            subResourceLayout.rowPitch, width, height, swizzle, false)) {
        utils::slog.e << "Unsupported PixelDataFormat or PixelDataType" << utils::io::endl;
    }

@@ -1629,11 +1615,6 @@ void VulkanDriver::readPixels(Handle<HwRenderTarget> src, uint32_t x, uint32_t y
    scheduleDestroy(std::move(pbd));
 }

-void VulkanDriver::readStreamPixels(Handle<HwStream> sh, uint32_t x, uint32_t y, uint32_t width,
-        uint32_t height, PixelBufferDescriptor&& p) {
-    scheduleDestroy(std::move(p));
-}
-
 void VulkanDriver::blit(TargetBufferFlags buffers, Handle<HwRenderTarget> dst, Viewport dstRect,
        Handle<HwRenderTarget> src, Viewport srcRect, SamplerMagFilter filter) {
    assert_invariant(mContext.currentRenderPass.renderPass == VK_NULL_HANDLE);
@@ -289,18 +289,17 @@ private:

    // Represents all the Vulkan state that comprises a bound descriptor set.
    struct DescriptorKey {
-        VkBuffer uniformBuffers[UBUFFER_BINDING_COUNT];             // 0
-        DescriptorImageInfo samplers[SAMPLER_BINDING_COUNT];        // 64
-        DescriptorImageInfo inputAttachments[TARGET_BINDING_COUNT]; // 832
-        uint32_t uniformBufferOffsets[UBUFFER_BINDING_COUNT];       // 1024
-        uint32_t uniformBufferSizes[UBUFFER_BINDING_COUNT];         // 1056
+        VkBuffer uniformBuffers[UBUFFER_BINDING_COUNT];             //   96     0
+        DescriptorImageInfo samplers[SAMPLER_BINDING_COUNT];        //  768    96
+        DescriptorImageInfo inputAttachments[TARGET_BINDING_COUNT]; //  192   864
+        uint32_t uniformBufferOffsets[UBUFFER_BINDING_COUNT];       //   48  1056
+        uint32_t uniformBufferSizes[UBUFFER_BINDING_COUNT];         //   48  1104
    };
-
-    static_assert(offsetof(DescriptorKey, samplers)              == 64);
-    static_assert(offsetof(DescriptorKey, inputAttachments)      == 832);
-    static_assert(offsetof(DescriptorKey, uniformBufferOffsets)  == 1024);
-    static_assert(offsetof(DescriptorKey, uniformBufferSizes)    == 1056);
-    static_assert(sizeof(DescriptorKey) == 1088, "DescriptorKey must not have implicit padding.");
+    static_assert(offsetof(DescriptorKey, samplers)              == 96);
+    static_assert(offsetof(DescriptorKey, inputAttachments)      == 864);
+    static_assert(offsetof(DescriptorKey, uniformBufferOffsets)  == 1056);
+    static_assert(offsetof(DescriptorKey, uniformBufferSizes)    == 1104);
+    static_assert(sizeof(DescriptorKey) == 1152, "DescriptorKey must not have implicit padding.");

    using DescHashFn = utils::hash::MurmurHashFn<DescriptorKey>;

@@ -26,6 +26,14 @@ namespace test {
 /**
 * A wrapper class that manages the necessary resources for a simple triangle renderable used for
 * test cases.
+ *
+ * The default vertices orient the triangle as such:
+ * o
+ * . .
+ * .   .
+ * .     .
+ * .       .
+ * o . . . . o
 */
 class TrianglePrimitive {
 public:
@@ -41,6 +41,10 @@ layout(location = 0) in vec4 mesh_position;
 uniform Params { highp vec4 color; highp vec4 scale; } params;
 void main() {
    gl_Position = vec4((mesh_position.xy + 0.5) * params.scale.xy, params.scale.z, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
 })";

 static const char* const triangleFs = R"(#version 450 core
@@ -106,7 +110,7 @@ static uint32_t toUintColor(float4 color) {
 }

 static void createBitmap(DriverApi& dapi, Handle<HwTexture> texture, int baseWidth, int baseHeight,
-        int level, int layer, float3 color) {
+        int level, int layer, float3 color, bool flipY) {
    auto cb = [](void* buffer, size_t size, void* user) { free(buffer); };
    const int width = baseWidth >> level;
    const int height = baseHeight >> level;
@@ -116,17 +120,35 @@ static void createBitmap(DriverApi& dapi, Handle<HwTexture> texture, int baseWid

    const float3 foreground = color;
    const float3 background = float3(1, 1, 0);
-    const float radius = 0.25f;

-    // Draw a circle on a yellow background.
+    // Draw a triangle on a yellow background.
+    //
+    // The triangle is oriented as such:
+    // low addresses
+    // |      .
+    // |     ...
+    // |    .....
+    // v   ........
+    // high addresses
+    //
+    // If flipY is specified (e.g., for OpenGL) we flip the image:
+    // high addresses
+    // ^      .
+    // |     ...
+    // |    .....
+    // |   ........
+    // low addresses
+    // This is because OpenGL automatically flips image data when uploading into the texture.
    uint32_t* texels = (uint32_t*) buffer0;
    for (int row = 0; row < height; row++) {
        for (int col = 0; col < width; col++) {
            float2 uv = { (col - width / 2.0f) / width, (row - height / 2.0f) / height };
-            const float d = distance(uv, float2(0));
-            const float t = d < radius ? 1.0 : 0.0;
+            const float d = abs(uv.x);
+            const float triangleWidth = uv.y >= -.3 && uv.y <= .3 ? (.4f / .6f * uv.y + .2f) : 0;
+            const float t = d < triangleWidth ? 1.0 : 0.0;
            const float3 color = mix(foreground, background, t);
-            texels[row * width + col] = toUintColor(float4(color, 1.0f));
+            int rowFlipped = flipY ? (height - 1) - row : row;
+            texels[rowFlipped * width + col] = toUintColor(float4(color, 1.0f));
        }
    }

@@ -135,8 +157,8 @@ static void createBitmap(DriverApi& dapi, Handle<HwTexture> texture, int baseWid
 }

 static void createBitmap(DriverApi& dapi, Handle<HwTexture> texture, int baseWidth, int baseHeight,
-        int level, float3 color) {
-    createBitmap(dapi, texture, baseWidth, baseHeight, level, 0u, color);
+        int level, float3 color, bool flipY) {
+    createBitmap(dapi, texture, baseWidth, baseHeight, level, 0u, color, flipY);
 }

 static void createFaces(DriverApi& dapi, Handle<HwTexture> texture, int baseWidth, int baseHeight,
@@ -246,8 +268,9 @@ TEST_F(BackendTest, ColorMagnify) {
    Handle<HwTexture> srcTexture = api.createTexture(
        SamplerType::SAMPLER_2D, kNumLevels, kSrcTexFormat, 1, kSrcTexWidth, kSrcTexHeight, 1,
        TextureUsage::SAMPLEABLE | TextureUsage::UPLOADABLE | TextureUsage::COLOR_ATTACHMENT);
-    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 0, float3(0.5, 0, 0));
-    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 1, float3(0, 0, 0.5));
+    const bool flipY = sBackend == Backend::OPENGL;
+    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 0, float3(0.5, 0, 0), flipY);
+    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 1, float3(0, 0, 0.5), flipY);

    // Create a destination texture.
    Handle<HwTexture> dstTexture = api.createTexture(
@@ -288,7 +311,7 @@ TEST_F(BackendTest, ColorMagnify) {
    getDriver().purge();

    // Check if the image matches perfectly to our golden run.
-    const uint32_t expected = 0xb830a36a;
+    const uint32_t expected = 0x410bdd31;
    printf("Computed hash is 0x%8.8x, Expected 0x%8.8x\n", params.pixelHashResult, expected);
    EXPECT_TRUE(params.pixelHashResult == expected);

@@ -320,8 +343,9 @@ TEST_F(BackendTest, ColorMinify) {
    Handle<HwTexture> srcTexture = api.createTexture(
        SamplerType::SAMPLER_2D, kNumLevels, kSrcTexFormat, 1, kSrcTexWidth, kSrcTexHeight, 1,
        TextureUsage::SAMPLEABLE | TextureUsage::UPLOADABLE | TextureUsage::COLOR_ATTACHMENT);
-    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 0, float3(0.5, 0, 0));
-    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 1, float3(0, 0, 0.5));
+    const bool flipY = sBackend == Backend::OPENGL;
+    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 0, float3(0.5, 0, 0), flipY);
+    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 1, float3(0, 0, 0.5), flipY);

    // Create a destination texture.
    Handle<HwTexture> dstTexture = api.createTexture(
@@ -362,7 +386,7 @@ TEST_F(BackendTest, ColorMinify) {
    getDriver().purge();

    // Check if the image matches perfectly to our golden run.
-    const uint32_t expected = 0xe2353ca6;
+    const uint32_t expected = 0x7739bef5;
    printf("Computed hash is 0x%8.8x, Expected 0x%8.8x\n", params.pixelHashResult, expected);
    EXPECT_TRUE(params.pixelHashResult == expected);

@@ -788,7 +812,8 @@ TEST_F(BackendTest, Blit2DTextureArray) {
    Handle<HwTexture> srcTexture = api.createTexture(
            SamplerType::SAMPLER_2D_ARRAY, kNumLevels, kSrcTexFormat, 1, kSrcTexWidth, kSrcTexHeight, kSrcTexDepth,
            TextureUsage::SAMPLEABLE | TextureUsage::UPLOADABLE | TextureUsage::COLOR_ATTACHMENT);
-    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 0, kSrcTexLayer, float3(0.5, 0, 0));
+    const bool flipY = sBackend == Backend::OPENGL;
+    createBitmap(api, srcTexture, kSrcTexWidth, kSrcTexHeight, 0, kSrcTexLayer, float3(0.5, 0, 0), flipY);

    // Create a destination texture.
    Handle<HwTexture> dstTexture = api.createTexture(
@@ -826,7 +851,7 @@ TEST_F(BackendTest, Blit2DTextureArray) {
    getDriver().purge();

    // Check if the image matches perfectly to our golden run.
-    const uint32_t expected = 0xe734bc44;
+    const uint32_t expected = 0x8de7d55b;
    printf("Computed hash is 0x%8.8x, Expected 0x%8.8x\n", params.pixelHashResult, expected);
    EXPECT_TRUE(params.pixelHashResult == expected);

@@ -33,6 +33,10 @@ layout(location = 0) out uvec4 indices;

 void main() {
    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
 }
 )");

@@ -152,16 +156,6 @@ struct MaterialParams {
    float blue;
 };

-static void uploadUniforms(DriverApi& dapi, Handle<HwBufferObject> ubh, MaterialParams params) {
-    MaterialParams* tmp = new MaterialParams(params);
-    auto cb = [](void* buffer, size_t size, void* user) {
-        MaterialParams* sp = (MaterialParams*) buffer;
-        delete sp;
-    };
-    BufferDescriptor bd(tmp, sizeof(MaterialParams), cb);
-    dapi.updateBufferObject(ubh, std::move(bd), 0);
-}
-
 TEST_F(BackendTest, BufferObjectUpdateWithOffset) {
    // Create a platform-specific SwapChain and make it current.
    auto swapChain = createSwapChain();
@@ -115,6 +115,9 @@ static void dumpScreenshot(DriverApi& dapi, Handle<HwRenderTarget> rt) {
    dapi.readPixels(rt, 0, 0, kTexWidth, kTexHeight, std::move(pb));
 }

+// TODO: This test needs work to get Metal and OpenGL to agree on results.
+// The problems are caused by both uploading and rendering into the same texture, since the OpenGL
+// backend's readPixels does not work correctly with textures that have image data uploaded.
 TEST_F(BackendTest, FeedbackLoops) {
    auto& api = getDriverApi();

@@ -24,11 +24,6 @@
 #include <fstream>
 #include <vector>

-#ifndef IOS
-#include <imageio/BlockCompression.h>
-using namespace image;
-#endif
-
 using namespace filament;
 using namespace filament::backend;

@@ -45,7 +40,7 @@ layout(location = 0) in vec4 mesh_position;
 void main() {
    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
 #if defined(TARGET_VULKAN_ENVIRONMENT)
-    //gl_Position.y = 1.0f - gl_Position.y;
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
    gl_Position.y *= -1.0f;
 #endif
 }
@@ -137,28 +132,6 @@ static void fillCheckerboard(void* buffer, size_t size, size_t stride, size_t co
    }
 }

-#ifndef IOS
-static PixelBufferDescriptor compressedCheckerboardPixelBuffer(size_t size) {
-    LinearImage uncompressed(size, size, 4);
-    fillCheckerboard<float>(uncompressed.getPixelRef(), size, size, 4, 1.0f);
-
-    S3tcConfig config {
-        .format = CompressedFormat::RGBA_S3TC_DXT1,
-        .srgb = false
-    };
-    CompressedTexture compressed = s3tcCompress(uncompressed, config);
-
-    void* buffer = malloc(compressed.size);
-    memcpy(buffer, compressed.data.get(), compressed.size);
-
-    PixelBufferDescriptor descriptor(buffer, compressed.size, CompressedPixelDataType::DXT1_RGBA,
-            compressed.size, [](void* buffer, size_t size, void* user) {
-                free(buffer);
-            }, nullptr);
-    return descriptor;
-}
-#endif
-
 static PixelBufferDescriptor checkerboardPixelBuffer(PixelDataFormat format, PixelDataType type,
        size_t size, size_t bufferPadding = 0) {
    size_t components; int bpp;
@@ -355,11 +328,6 @@ TEST_F(BackendTest, UpdateImage2D) {
    testCases.emplace_back("RGBA, UBYTE -> RGBA8 (subregions, buffer padding)", PixelDataFormat::RGBA, PixelDataType::UBYTE, TextureFormat::RGBA8, 64u, true);
    testCases.emplace_back("RGB, FLOAT -> RGB32F (subregions, buffer padding)", PixelDataFormat::RGB, PixelDataType::FLOAT, TextureFormat::RGB32F, 64u, true);

-    // Test compresseed format upload.
-#ifndef IOS
-    testCases.emplace_back("RGBA, DXT1_RGBA -> DXT1_RGBA", PixelDataFormat::RGBA, CompressedPixelDataType::DXT1_RGBA, TextureFormat::DXT1_RGBA);
-#endif
-
    auto& api = getDriverApi();

    api.startCapture();
@@ -388,31 +356,21 @@ TEST_F(BackendTest, UpdateImage2D) {
                t.textureFormat, 1, 512, 512, 1u, usage);

        // Upload some pixel data.
-        if (t.compressed) {
-#ifdef IOS
-            assert_invariant(false);
-#else
-            assert_invariant(!t.uploadSubregions);
-            PixelBufferDescriptor descriptor = compressedCheckerboardPixelBuffer(512);
-            api.update2DImage(texture, 0, 0, 0, 512, 512, std::move(descriptor));
-#endif
+        if (t.uploadSubregions) {
+            const auto& pf = t.pixelFormat;
+            const auto& pt = t.pixelType;
+            PixelBufferDescriptor subregion1 = checkerboardPixelBuffer(pf, pt, 256, t.bufferPadding);
+            PixelBufferDescriptor subregion2 = checkerboardPixelBuffer(pf, pt, 256, t.bufferPadding);
+            PixelBufferDescriptor subregion3 = checkerboardPixelBuffer(pf, pt, 256, t.bufferPadding);
+            PixelBufferDescriptor subregion4 = checkerboardPixelBuffer(pf, pt, 256, t.bufferPadding);
+            api.update2DImage(texture, 0, 0, 0, 256, 256, std::move(subregion1));
+            api.update2DImage(texture, 0, 256, 0, 256, 256, std::move(subregion2));
+            api.update2DImage(texture, 0, 0, 256, 256, 256, std::move(subregion3));
+            api.update2DImage(texture, 0, 256, 256, 256, 256, std::move(subregion4));
        } else {
-            if (t.uploadSubregions) {
-                const auto& pf = t.pixelFormat;
-                const auto& pt = t.pixelType;
-                PixelBufferDescriptor subregion1 = checkerboardPixelBuffer(pf, pt, 256, t.bufferPadding);
-                PixelBufferDescriptor subregion2 = checkerboardPixelBuffer(pf, pt, 256, t.bufferPadding);
-                PixelBufferDescriptor subregion3 = checkerboardPixelBuffer(pf, pt, 256, t.bufferPadding);
-                PixelBufferDescriptor subregion4 = checkerboardPixelBuffer(pf, pt, 256, t.bufferPadding);
-                api.update2DImage(texture, 0, 0, 0, 256, 256, std::move(subregion1));
-                api.update2DImage(texture, 0, 256, 0, 256, 256, std::move(subregion2));
-                api.update2DImage(texture, 0, 0, 256, 256, 256, std::move(subregion3));
-                api.update2DImage(texture, 0, 256, 256, 256, 256, std::move(subregion4));
-            } else {
-                PixelBufferDescriptor descriptor
-                    = checkerboardPixelBuffer(t.pixelFormat, t.pixelType, 512, t.bufferPadding);
-                api.update2DImage(texture, 0, 0, 0, 512, 512, std::move(descriptor));
-            }
+            PixelBufferDescriptor descriptor
+                = checkerboardPixelBuffer(t.pixelFormat, t.pixelType, 512, t.bufferPadding);
+            api.update2DImage(texture, 0, 0, 0, 512, 512, std::move(descriptor));
        }

        SamplerGroup samplers(1);
@@ -33,6 +33,10 @@ layout(location = 0) out uvec4 indices;

 void main() {
    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
 }
 )");

@@ -45,6 +45,10 @@ layout(location = 0) in vec4 mesh_position;

 void main() {
    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
 }
 )");

@@ -123,6 +127,9 @@ TEST_F(ReadPixelsTest, ReadPixels) {
        // The offset and stride set on the pixel buffer.
        size_t left = 0, top = 0, alignment = 1;

+        // If true, use the default RT to render into the SwapChain as opposed to a texture.
+        bool useDefaultRT = false;
+
        size_t getPixelBufferStride() const {
            return bufferDimension;
        }
@@ -219,7 +226,13 @@ TEST_F(ReadPixelsTest, ReadPixels) {
    t7.textureFormat = TextureFormat::RG16F;
    t7.hash = 3726805703;

-    TestCase testCases[] = { t, t2, t3, t4, t5, t6, t7 };
+    // Check that readPixels works when rendering into the SwapChain.
+    // This requires that the test runner's native window size is 512x512.
+    TestCase t8;
+    t8.testName = "readPixels_swapchain";
+    t8.useDefaultRT = true;
+
+    TestCase testCases[] = { t, t2, t3, t4, t5, t6, t7, t8 };

    // Create programs.
    Handle<HwProgram> programFloat, programUint;
@@ -234,11 +247,18 @@ TEST_F(ReadPixelsTest, ReadPixels) {
        programUint = getDriverApi().createProgram(std::move(p));
    }

+
    for (const auto& t : testCases)
    {
        // Create a platform-specific SwapChain and make it current.
-        auto swapChain = getDriverApi().createSwapChainHeadless(t.getRenderTargetSize(),
-                t.getRenderTargetSize(), 0);
+        Handle<HwSwapChain> swapChain;
+        if (t.useDefaultRT) {
+            swapChain = createSwapChain();
+        } else {
+            swapChain = getDriverApi().createSwapChainHeadless(t.getRenderTargetSize(),
+                    t.getRenderTargetSize(), 0);
+        }
+
        getDriverApi().makeCurrent(swapChain, swapChain);

        // Create a Texture and RenderTarget to render into.
@@ -247,12 +267,19 @@ TEST_F(ReadPixelsTest, ReadPixels) {
                t.mipLevels, t.textureFormat, 1, renderTargetBaseSize, renderTargetBaseSize, 1,
                usage);

-        // The width and height must match the width and height of the respective mip
-        // level (at least for OpenGL).
-        Handle<HwRenderTarget> renderTarget = getDriverApi().createRenderTarget(
-                TargetBufferFlags::COLOR, t.getRenderTargetSize(),
-                t.getRenderTargetSize(), t.samples, {{ texture, uint8_t(t.mipLevel) }}, {},
-                {});
+        Handle<HwRenderTarget> renderTarget;
+        if (t.useDefaultRT) {
+            // The width and height must match the width and height of the respective mip
+            // level (at least for OpenGL).
+            renderTarget = getDriverApi().createDefaultRenderTarget();
+        } else {
+            // The width and height must match the width and height of the respective mip
+            // level (at least for OpenGL).
+            renderTarget = getDriverApi().createRenderTarget(
+                    TargetBufferFlags::COLOR, t.getRenderTargetSize(),
+                    t.getRenderTargetSize(), t.samples, {{ texture, uint8_t(t.mipLevel) }}, {},
+                    {});
+        }

        TrianglePrimitive triangle(getDriverApi());

@@ -260,7 +260,7 @@ public:
     * range can be adjusted with this method.
     *
     * @param intensity  Scale factor applied to the environment and irradiance such that
-     *                   the result is in lux, or <i>lumen/m^2(default = 30000)
+     *                   the result is in lux, or <i>lumen/m^2</i> (default = 30000)
     */
    void setIntensity(float intensity) noexcept;

@@ -444,6 +444,10 @@ public:
     *
     * It is also possible to use a Fence to wait for the read-back.
     *
+     * OpenGL only: if issuing a readPixels on a RenderTarget backed by a Texture that had data
+     * uploaded to it via setImage, the data returned from readPixels will be y-flipped with respect
+     * to the setImage call.
+     *
     * @remark
     * readPixels() is intended for debugging and testing. It will impact performance significantly.
     *
@@ -124,7 +124,7 @@ public:
         *
         * Ignored if an environment is set.
         *
-         * @param color
+         * @param color the constant color
         *
         * @return This Builder, for chaining calls.
         */
@@ -35,10 +35,9 @@ class Engine;
 /**
 * Stream is used to attach a video stream to a Filament `Texture`.
 *
- * Note that the `Stream` class is fairly Android centric. It supports three different
+ * Note that the `Stream` class is fairly Android centric. It supports two different
 * configurations:
 *
- *   - TEXTURE_ID...takes an OpenGL texture ID and incurs a copy
 *   - ACQUIRED.....connects to an Android AHardwareBuffer
 *   - NATIVE.......connects to an Android SurfaceTexture
 *
@@ -67,10 +66,6 @@ class Engine;
 * - Filament invokes low-level graphics commands on the \em{driver thread}.
 * - The thread that calls `beginFrame` is called the \em{main thread}.
 *
- * The TEXTURE_ID configuration achieves synchronization automatically. In this mode, Filament
- * performs a copy on the main thread during `beginFrame` by blitting the external image into
- * an internal round-robin queue of images. This copy has a run-time cost.
- *
 * For ACQUIRED streams, there is no need to perform the copy because Filament explictly acquires
 * the stream, then releases it later via a callback function. This configuration is especially
 * useful when the Vulkan backend is enabled.
@@ -97,7 +92,7 @@ public:
     * By default, Stream objects are ACQUIRED and must have external images pushed to them via
     * <pre>Stream::setAcquiredImage</pre>.
     *
-     * To create a NATIVE or TEXTURE_ID stream, call one of the <pre>stream</pre> methods
+     * To create a NATIVE stream, call one of the <pre>stream</pre> methods
     * on the builder.
     */
    class Builder : public BuilderBase<BuilderDetails> {
@@ -122,23 +117,6 @@ public:
         */
        Builder& stream(void* stream) noexcept;

-        /**
-         * Creates a TEXTURE_ID stream. This will sample data from the supplied
-         * external texture and copy it into an internal private texture.
-         *
-         * @param externalTextureId An opaque texture id (typically a GLuint created with glGenTextures)
-         *                          In a context shared with filament. In that case this texture's
-         *                          target must be GL_TEXTURE_EXTERNAL_OES and the wrap mode must
-         *                          be CLAMP_TO_EDGE.
-         *
-         * @return This Builder, for chaining calls.
-         *
-         * @see Texture::setExternalStream()
-         * @deprecated this method existed only for ARCore which doesn't need this anymore, use Texture::import() instead.
-         */
-        UTILS_DEPRECATED
-        Builder& stream(intptr_t externalTextureId) noexcept;
-
        /**
         *
         * @param width initial width of the incoming stream. Whether this value is used is
@@ -173,7 +151,7 @@ public:
    };

    /**
-     * Indicates whether this stream is a NATIVE stream, TEXTURE_ID stream, or ACQUIRED stream.
+     * Indicates whether this stream is a NATIVE stream or ACQUIRED stream.
     */
    StreamType getStreamType() const noexcept;

@@ -190,7 +168,7 @@ public:
     * also where the callback is invoked. This method can only be used for streams that were
     * constructed without calling the `stream` method on the builder.
     *
-     * @see Stream for more information about NATIVE, TEXTURE_ID, and ACQUIRED configurations.
+     * @see Stream for more information about NATIVE and ACQUIRED configurations.
     *
     * @param image      Pointer to AHardwareBuffer, casted to void* since this is a public header.
     * @param callback   This is triggered by Filament when it wishes to release the image.
@@ -222,63 +200,6 @@ public:
     */
    void setDimensions(uint32_t width, uint32_t height) noexcept;

-    /**
-     * Read-back the content of the last frame of a Stream since the last call to
-     * Renderer.beginFrame().
-     *
-     * The Stream must be of type externalTextureId. This function is a no-op otherwise.
-     *
-     * @param xoffset   Left offset of the sub-region to read back.
-     * @param yoffset   Bottom offset of the sub-region to read back.
-     * @param width     Width of the sub-region to read back.
-     * @param height    Height of the sub-region to read back.
-     * @param buffer    Client-side buffer where the read-back will be written.
-     *
-     *                  The following format are always supported:
-     *                      - PixelBufferDescriptor::PixelDataFormat::RGBA
-     *                      - PixelBufferDescriptor::PixelDataFormat::RGBA_INTEGER
-     *
-     *                  The following types are always supported:
-     *                      - PixelBufferDescriptor::PixelDataType::UBYTE
-     *                      - PixelBufferDescriptor::PixelDataType::UINT
-     *                      - PixelBufferDescriptor::PixelDataType::INT
-     *                      - PixelBufferDescriptor::PixelDataType::FLOAT
-     *
-     *                  Other combination of format/type may be supported. If a combination is
-     *                  not supported, this operation may fail silently. Use a DEBUG build
-     *                  to get some logs about the failure.
-     *
-     *  Stream buffer                  User buffer (PixelBufferDescriptor&)
-     *  +--------------------+
-     *  |                    |                .stride         .alignment
-     *  |                    |         ----------------------->-->
-     *  |                    |         O----------------------+--+   low addresses
-     *  |                    |         |          |           |  |
-     *  |             w      |         |          | .top      |  |
-     *  |       <--------->  |         |          V           |  |
-     *  |       +---------+  |         |     +---------+      |  |
-     *  |       |     ^   |  | ======> |     |         |      |  |
-     *  |   x   |    h|   |  |         |.left|         |      |  |
-     *  +------>|     v   |  |         +---->|         |      |  |
-     *  |       +.........+  |         |     +.........+      |  |
-     *  |            ^       |         |                      |  |
-     *  |          y |       |         +----------------------+--+  high addresses
-     *  O------------+-------+
-     *
-     * Typically readPixels() will be called after Renderer.beginFrame().
-     *
-     * After issuing this method, the callback associated with `buffer` will be invoked on the
-     * main thread, indicating that the read-back has completed. Typically, this will happen
-     * after multiple calls to beginFrame(), render(), endFrame().
-     *
-     * It is also possible to use a Fence to wait for the read-back.
-     *
-     * @remark
-     * readPixels() is intended for debugging and testing. It will impact performance significantly.
-     */
-    void readPixels(uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
-            backend::PixelBufferDescriptor&& buffer) noexcept;
-
    /**
     * Returns the presentation time of the currently displayed frame in nanosecond.
     *
@@ -134,12 +134,12 @@ struct UTILS_PUBLIC GenericToneMapper final : public ToneMapper {
     * constructor parameters approximate an ACES tone mapping curve
     * and the maximum input value is set to 10.0.
     *
-     * @param contrast: controls the contrast of the curve, must be > 0.0, values
-     *                  in the range 0.5..2.0 are recommended.
-     * @param midGrayIn: sets the input middle gray, between 0.0 and 1.0.
-     * @param midGrayOut: sets the output middle gray, between 0.0 and 1.0.
-     * @param hdrMax: defines the maximum input value that will be mapped to
-     *                output white. Must be >= 1.0.
+     * @param contrast controls the contrast of the curve, must be > 0.0, values
+     *                 in the range 0.5..2.0 are recommended.
+     * @param midGrayIn sets the input middle gray, between 0.0 and 1.0.
+     * @param midGrayOut sets the output middle gray, between 0.0 and 1.0.
+     * @param hdrMax defines the maximum input value that will be mapped to
+     *               output white. Must be >= 1.0.
     */
    explicit GenericToneMapper(
            float contrast = 1.55f,
@@ -25,7 +25,7 @@ const uint16_t DFG::DFG_LUT[] = {
 #include "generated/data/dfg.inc"
 };

-DFG::DFG(FEngine& engine) noexcept : mEngine(engine) {
+void DFG::init(FEngine& engine) noexcept {
    constexpr size_t fp16Count = DFG_LUT_SIZE * DFG_LUT_SIZE * 3;
    constexpr size_t byteCount = fp16Count * sizeof(uint16_t);

@@ -35,19 +35,19 @@ DFG::DFG(FEngine& engine) noexcept : mEngine(engine) {
            .width(DFG_LUT_SIZE)
            .height(DFG_LUT_SIZE)
            .format(backend::TextureFormat::RGB16F)
-            .build(mEngine);
+            .build(engine);

    Texture::PixelBufferDescriptor buffer(DFG_LUT, byteCount,
            Texture::Format::RGB, Texture::Type::HALF);

-    lut->setImage(mEngine, 0, std::move(buffer));
+    lut->setImage(engine, 0, std::move(buffer));

    mLUT = upcast(lut);
 }

-void DFG::terminate() {
+void DFG::terminate(FEngine& engine) noexcept {
    if (mLUT) {
-        mEngine.destroy(mLUT);
+        engine.destroy(mLUT);
    }
 }

@@ -33,7 +33,14 @@ class FEngine;

 class DFG {
 public:
-    explicit DFG(FEngine& engine) noexcept;
+    explicit DFG() noexcept = default;
+
+    DFG(DFG const& rhs) = delete;
+    DFG(DFG&& rhs) = delete;
+    DFG& operator=(DFG const& rhs) = delete;
+    DFG& operator=(DFG&& rhs) = delete;
+
+    void init(FEngine& engine) noexcept;

    size_t getLutSize() const noexcept {
        return DFG_LUT_SIZE;
@@ -47,15 +54,9 @@ public:
        return mLUT->getHwHandle();
    }

-    void terminate();
-
-    DFG(DFG const& rhs) = delete;
-    DFG(DFG&& rhs) = delete;
-    DFG& operator=(DFG const& rhs) = delete;
-    DFG& operator=(DFG&& rhs) = delete;
+    void terminate(FEngine& engine) noexcept;

 private:
-    FEngine& mEngine;
    FTexture* mLUT = nullptr;

    // make sure to use the right size here
@@ -244,7 +244,6 @@ bool Froxelizer::update() noexcept {
        mClipToFroxelX = (0.5f * viewport.width)  / froxelDimension.x;
        mClipToFroxelY = (0.5f * viewport.height) / froxelDimension.y;

-        mOneOverDimension = 1.0f / float2(froxelDimension);
        uniformsNeedUpdating = true;

 #ifndef NDEBUG
@@ -513,10 +512,10 @@ void Froxelizer::commit(backend::DriverApi& driverApi) {
 }

 void Froxelizer::froxelizeLights(FEngine& engine,
-        CameraInfo const& UTILS_RESTRICT camera,
+        mat4f const& UTILS_RESTRICT viewMatrix,
        const FScene::LightSoa& UTILS_RESTRICT lightData) noexcept {
    // note: this is called asynchronously
-    froxelizeLoop(engine, camera, lightData);
+    froxelizeLoop(engine, viewMatrix, lightData);
    froxelizeAssignRecordsCompress();

 #ifndef NDEBUG
@@ -544,7 +543,7 @@ void Froxelizer::froxelizeLights(FEngine& engine,
 }

 void Froxelizer::froxelizeLoop(FEngine& engine,
-        const CameraInfo& UTILS_RESTRICT camera,
+        const mat4f& UTILS_RESTRICT viewMatrix,
        const FScene::LightSoa& UTILS_RESTRICT lightData) noexcept {
    SYSTRACE_CALL();

@@ -557,11 +556,11 @@ void Froxelizer::froxelizeLoop(FEngine& engine,
    auto const* UTILS_RESTRICT instances    = lightData.data<FScene::LIGHT_INSTANCE>();

    auto process = [ this, &froxelThreadData,
-                     spheres, directions, instances, &camera, &lcm ]
+                     spheres, directions, instances, &viewMatrix, &lcm ]
            (size_t count, size_t offset, size_t stride) {

        const mat4f& projection = mProjection;
-        const mat3f& vn = camera.view.upperLeft();
+        const mat3f& vn = viewMatrix.upperLeft();

        // We use minimum cone angle of 0.5 degrees because too small angles cause issues in the
        // sphere/cone intersection test, due to floating-point precision.
@@ -572,7 +571,7 @@ void Froxelizer::froxelizeLoop(FEngine& engine,
            const size_t j = i + FScene::DIRECTIONAL_LIGHTS_COUNT;
            FLightManager::Instance li = instances[j];
            LightParams light = {
-                    .position = (camera.view * float4{ spheres[j].xyz, 1 }).xyz, // to view-space
+                    .position = (viewMatrix * float4{ spheres[j].xyz, 1 }).xyz,     // to view-space
                    .cosSqr = std::min(maxCosSquared, lcm.getCosOuterSquared(li)),  // spot only
                    .axis = vn * directions[j],                                     // spot only
                    .invSin = lcm.getSinInverse(li),                                // spot only
@@ -120,15 +120,13 @@ public:
    float getLightFar() const noexcept { return mZLightFar; }

    // update Records and Froxels texture with lights data. this is thread-safe.
-    void froxelizeLights(FEngine& engine, CameraInfo const& camera,
+    void froxelizeLights(FEngine& engine, math::mat4f const& viewMatrix,
            const FScene::LightSoa& lightData) noexcept;

    void updateUniforms(PerViewUib& s) {
        s.zParams = mParamsZ;
-        s.fParams = mParamsF.yz;
-        s.fParamsX = mParamsF.x;
-        s.oneOverFroxelDimensionX = mOneOverDimension.x;
-        s.oneOverFroxelDimensionY = mOneOverDimension.y;
+        s.fParams = mParamsF;
+        s.froxelCountXY = math::float2{ mViewport.width, mViewport.height } / mFroxelDimension;
    }

    // send froxel data to GPU
@@ -194,7 +192,7 @@ private:
    bool update() noexcept;

    void froxelizeLoop(FEngine& engine,
-            const CameraInfo& camera, const FScene::LightSoa& lightData) noexcept;
+            math::mat4f const& viewMatrix, const FScene::LightSoa& lightData) noexcept;

    void froxelizeAssignRecordsCompress() noexcept;

@@ -242,7 +240,6 @@ private:
    float mLinearizer = 0.0f;
    float mClipToFroxelX = 0.0f;
    float mClipToFroxelY = 0.0f;
-    math::float2 mOneOverDimension = {};
    backend::BufferObjectHandle mRecordsBuffer;
    backend::Handle<backend::HwTexture> mFroxelTexture;

@@ -52,10 +52,10 @@ PerViewUniforms::PerViewUniforms(FEngine& engine) noexcept
    // with a clip-space of [0,  w] ==> z' = (w - z)/2
    mClipControl = driver.getClipSpaceParams();

-    if (engine.getDFG()->isValid()) {
+    if (engine.getDFG().isValid()) {
        TextureSampler sampler(TextureSampler::MagFilter::LINEAR);
        mSamplers.setSampler(PerViewSib::IBL_DFG_LUT,
-                engine.getDFG()->getTexture(), sampler.getSamplerParams());
+                engine.getDFG().getTexture(), sampler.getSamplerParams());
    }
 }

@@ -81,7 +81,7 @@ void PerViewUniforms::prepareCamera(const CameraInfo& camera) noexcept {
    s.clipFromWorldMatrix = clipFromWorld;    // projection * view
    s.worldFromClipMatrix = worldFromClip;    // 1/(projection * view)
    s.cameraPosition = float3{ camera.getPosition() };
-    s.worldOffset = camera.worldOffset;
+    s.worldOffset = camera.getWorldOffset();
    s.cameraFar = camera.zf;
    s.oneOverFarMinusNear = 1.0f / (camera.zf - camera.zn);
    s.nearOverFarMinusNear = camera.zn / (camera.zf - camera.zn);
@@ -105,12 +105,14 @@ void PerViewUniforms::prepareExposure(float ev100) noexcept {
    s.ev100 = ev100;
 }

-void PerViewUniforms::prepareViewport(const filament::Viewport &viewport) noexcept {
+void PerViewUniforms::prepareViewport(const filament::Viewport& viewport,
+        uint32_t xoffset, uint32_t yoffset) noexcept {
    const float w = float(viewport.width);
    const float h = float(viewport.height);
    auto& s = mUniforms.edit();
    s.resolution = float4{ w, h, 1.0f / w, 1.0f / h };
    s.origin = float2{ viewport.left, viewport.bottom };
+    s.offset = float2{ xoffset, yoffset };
 }

 void PerViewUniforms::prepareTime(math::float4 const& userTime) noexcept {
@@ -127,7 +129,7 @@ void PerViewUniforms::prepareTemporalNoise(TemporalAntiAliasingOptions const& op
    s.temporalNoise = options.enabled ? temporalNoise : 0.0f;
 }

-void PerViewUniforms::prepareFog(const CameraInfo& camera, FogOptions const& options) noexcept {
+void PerViewUniforms::prepareFog(float3 const& cameraPosition, FogOptions const& options) noexcept {
    // this can't be too high because we need density / heightFalloff to produce something
    // close to fogOptions.density in the fragment shader which use 16-bits floats.
    constexpr float epsilon = 0.001f;
@@ -135,7 +137,7 @@ void PerViewUniforms::prepareFog(const CameraInfo& camera, FogOptions const& opt

    // precalculate the constant part of density  integral and correct for exp2() in the shader
    const float density = ((options.density / heightFalloff) *
-            std::exp(-heightFalloff * (camera.getPosition().y - options.height)))
+            std::exp(-heightFalloff * (cameraPosition.y - options.height)))
                    * float(1.0f / F_LN2);

    auto& s = mUniforms.edit();
@@ -283,23 +285,24 @@ void PerViewUniforms::prepareDynamicLights(Froxelizer& froxelizer) noexcept {
    s.lightFarAttenuationParams = 0.5f * float2{ 10.0f, 10.0f / (f * f) };
 }

-void PerViewUniforms::prepareShadowMapping(ShadowMappingUniforms const& shadowMappingUniforms,
-        VsmShadowOptions const& options) noexcept {
+void PerViewUniforms::prepareShadowMapping() noexcept {
    auto& s = mUniforms.edit();
    s.vsmExponent = 5.54f;  // fp16: max 5.54f, fp32: max 42.0
-    s.vsmDepthScale = options.minVarianceScale * 0.01f * s.vsmExponent;
-    s.vsmLightBleedReduction = options.lightBleedReduction;
+}

-    s.lightFromWorldMatrix = shadowMappingUniforms.lightFromWorldMatrix;
-    s.cascadeSplits = shadowMappingUniforms.cascadeSplits;
-    s.shadowBulbRadiusLs = shadowMappingUniforms.shadowBulbRadiusLs;
-    s.shadowBias = shadowMappingUniforms.shadowBias;
-    s.ssContactShadowDistance = shadowMappingUniforms.ssContactShadowDistance;
-    s.directionalShadows = shadowMappingUniforms.directionalShadows;
-    s.cascades = shadowMappingUniforms.cascades;
+void PerViewUniforms::prepareShadowSampling(PerViewUib& uniforms,
+        ShadowMappingUniforms const& shadowMappingUniforms) noexcept {
+    uniforms.lightFromWorldMatrix       = shadowMappingUniforms.lightFromWorldMatrix;
+    uniforms.cascadeSplits              = shadowMappingUniforms.cascadeSplits;
+    uniforms.shadowBulbRadiusLs         = shadowMappingUniforms.shadowBulbRadiusLs;
+    uniforms.shadowBias                 = shadowMappingUniforms.shadowBias;
+    uniforms.ssContactShadowDistance    = shadowMappingUniforms.ssContactShadowDistance;
+    uniforms.directionalShadows         = shadowMappingUniforms.directionalShadows;
+    uniforms.cascades                   = shadowMappingUniforms.cascades;
 }

 void PerViewUniforms::prepareShadowVSM(Handle<HwTexture> texture,
+        ShadowMappingUniforms const& shadowMappingUniforms,
        VsmShadowOptions const& options) noexcept {
    SamplerMinFilter filterMin = SamplerMinFilter::LINEAR;
    if (options.anisotropy > 0 || options.mipmapping) {
@@ -313,9 +316,14 @@ void PerViewUniforms::prepareShadowVSM(Handle<HwTexture> texture,
            }});
    auto& s = mUniforms.edit();
    s.shadowSamplingType = SHADOW_SAMPLING_RUNTIME_VSM;
+    s.vsmExponent = 5.54f;  // fp16: max 5.54f, fp32: max 42.0
+    s.vsmDepthScale = options.minVarianceScale * 0.01f * s.vsmExponent;
+    s.vsmLightBleedReduction = options.lightBleedReduction;
+    PerViewUniforms::prepareShadowSampling(s, shadowMappingUniforms);
 }

-void PerViewUniforms::prepareShadowPCF(Handle<HwTexture> texture) noexcept {
+void PerViewUniforms::prepareShadowPCF(Handle<HwTexture> texture,
+        ShadowMappingUniforms const& shadowMappingUniforms) noexcept {
    mSamplers.setSampler(PerViewSib::SHADOW_MAP, {
            texture, {
                    .filterMag = SamplerMagFilter::LINEAR,
@@ -325,22 +333,27 @@ void PerViewUniforms::prepareShadowPCF(Handle<HwTexture> texture) noexcept {
            }});
    auto& s = mUniforms.edit();
    s.shadowSamplingType = SHADOW_SAMPLING_RUNTIME_PCF;
+    PerViewUniforms::prepareShadowSampling(s, shadowMappingUniforms);
 }

-void PerViewUniforms::prepareShadowDPCF(Handle <HwTexture> texture,
+void PerViewUniforms::prepareShadowDPCF(Handle<HwTexture> texture,
+        ShadowMappingUniforms const& shadowMappingUniforms,
        SoftShadowOptions const& options) noexcept {
    mSamplers.setSampler(PerViewSib::SHADOW_MAP, { texture, {}});
    auto& s = mUniforms.edit();
    s.shadowSamplingType = SHADOW_SAMPLING_RUNTIME_DPCF;
    s.shadowPenumbraRatioScale = options.penumbraRatioScale;
+    PerViewUniforms::prepareShadowSampling(s, shadowMappingUniforms);
 }

-void PerViewUniforms::prepareShadowPCSS(Handle <HwTexture> texture,
+void PerViewUniforms::prepareShadowPCSS(Handle<HwTexture> texture,
+        ShadowMappingUniforms const& shadowMappingUniforms,
        SoftShadowOptions const& options) noexcept {
    mSamplers.setSampler(PerViewSib::SHADOW_MAP, { texture, {}});
    auto& s = mUniforms.edit();
    s.shadowSamplingType = SHADOW_SAMPLING_RUNTIME_PCSS;
    s.shadowPenumbraRatioScale = options.penumbraRatioScale;
+    PerViewUniforms::prepareShadowSampling(s, shadowMappingUniforms);
 }

 void PerViewUniforms::commit(backend::DriverApi& driver) noexcept {
@@ -65,11 +65,20 @@ public:

    void prepareCamera(const CameraInfo& camera) noexcept;
    void prepareUpscaler(math::float2 scale, DynamicResolutionOptions const& options) noexcept;
-    void prepareViewport(const filament::Viewport& viewport) noexcept;
+
+    /*
+     * @param viewport  viewport (should be same as RenderPassParams::viewport)
+     * @param xoffset   horizontal rendering offset *within* the viewport.
+     *                  Non-zero when we have guard bands.
+     * @param yoffset   vertical rendering offset *within* the viewport.
+     *                  Non-zero when we have guard bands.
+     */
+    void prepareViewport(const filament::Viewport& viewport, uint32_t xoffset, uint32_t yoffset) noexcept;
+
    void prepareTime(math::float4 const& userTime) noexcept;
    void prepareTemporalNoise(TemporalAntiAliasingOptions const& options) noexcept;
    void prepareExposure(float ev100) noexcept;
-    void prepareFog(const CameraInfo& camera, FogOptions const& options) noexcept;
+    void prepareFog(math::float3 const& cameraPosition, FogOptions const& options) noexcept;
    void prepareStructure(TextureHandle structure) noexcept;
    void prepareSSAO(TextureHandle ssao, AmbientOcclusionOptions const& options) noexcept;
    void prepareBlending(bool needsAlphaChannel) noexcept;
@@ -84,8 +93,7 @@ public:
            math::mat4f const& uvFromViewMatrix,
            ScreenSpaceReflectionsOptions const& ssrOptions) noexcept;

-    void prepareShadowMapping(ShadowMappingUniforms const& shadowMappingUniforms,
-            VsmShadowOptions const& options) noexcept;
+    void prepareShadowMapping() noexcept;

    void prepareDirectionalLight(float exposure,
            math::float3 const& sceneSpaceDirection, LightManagerInstance instance) noexcept;
@@ -94,11 +102,20 @@ public:

    void prepareDynamicLights(Froxelizer& froxelizer) noexcept;

-    // maybe these should have their own UBO, they're needed only when GENERATING the shadowmaps
-    void prepareShadowVSM(TextureHandle texture, VsmShadowOptions const& options) noexcept;
-    void prepareShadowPCF(TextureHandle texture) noexcept;
-    void prepareShadowDPCF(TextureHandle texture, SoftShadowOptions const& options) noexcept;
-    void prepareShadowPCSS(TextureHandle texture, SoftShadowOptions const& options) noexcept;
+    void prepareShadowVSM(TextureHandle texture,
+            ShadowMappingUniforms const& shadowMappingUniforms,
+            VsmShadowOptions const& options) noexcept;
+
+    void prepareShadowPCF(TextureHandle texture,
+            ShadowMappingUniforms const& shadowMappingUniforms) noexcept;
+
+    void prepareShadowDPCF(TextureHandle texture,
+            ShadowMappingUniforms const& shadowMappingUniforms,
+            SoftShadowOptions const& options) noexcept;
+
+    void prepareShadowPCSS(TextureHandle texture,
+            ShadowMappingUniforms const& shadowMappingUniforms,
+            SoftShadowOptions const& options) noexcept;

    // update local data into GPU UBO
    void commit(backend::DriverApi& driver) noexcept;
@@ -116,6 +133,8 @@ private:
    backend::Handle<backend::HwBufferObject> mUniformBufferHandle;
    backend::Handle<backend::HwSamplerGroup> mSamplerGroupHandle;
    std::uniform_real_distribution<float> mUniformDistribution{ 0.0f, 1.0f };
+    static void prepareShadowSampling(PerViewUib& uniforms,
+            ShadowMappingUniforms const& shadowMappingUniforms) noexcept;
 };

 } // namespace filament
@@ -531,22 +531,23 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::ssr(FrameGraph& fg,
                    data.history = builder.sample(history);
                }
            },
-            [=, &uniforms, renderPass = pass](FrameGraphResources const& resources,
-                    auto const& data, DriverApi& driver) mutable {
-
+            [this, projection = cameraInfo.projection,
+                    userViewMatrix = cameraInfo.getUserViewMatrix(), uvFromClipMatrix, historyProjection,
+                    options, &uniforms, renderPass = pass]
+            (FrameGraphResources const& resources, auto const& data, DriverApi& driver) mutable {
                // set structure sampler
                uniforms.prepareStructure(data.structure ?
                        resources.getTexture(data.structure) : getOneTexture());

                // set screen-space reflections and screen-space refractions
-                mat4f uvFromViewMatrix = uvFromClipMatrix * cameraInfo.projection;
-                mat4f reprojection = uvFromClipMatrix * historyProjection
-                                     * inverse(cameraInfo.view * cameraInfo.worldOrigin);
+                mat4f uvFromViewMatrix = uvFromClipMatrix * projection;
+                mat4f reprojection = mat4f{ uvFromClipMatrix * historyProjection
+                        * inverse(userViewMatrix) };

                // the history sampler is a regular texture2D
                TextureHandle history = data.history ?
                        resources.getTexture(data.history) : getZeroTexture();
-                uniforms.prepareHistorySSR(history, reprojection, uvFromViewMatrix,options);
+                uniforms.prepareHistorySSR(history, reprojection, uvFromViewMatrix, options);

                uniforms.commit(driver);

@@ -804,11 +805,9 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::screenSpaceAmbientOcclusion(
                mi->setParameter("ssctConeAngleTangeant", std::tan(options.ssct.lightConeRad * 0.5f));
                mi->setParameter("ssctContactDistanceMaxInv", 1.0f / options.ssct.contactDistanceMax);
                // light direction in view space
-                // (note: this is actually equivalent to using the camera view matrix -- before the
-                // world matrix is accounted for)
-                auto m = cameraInfo.view * cameraInfo.worldOrigin;
+                const mat4f view{ cameraInfo.getUserViewMatrix() };
                const float3 l = normalize(
-                        mat3f::getTransformForNormals(m.upperLeft())
+                        mat3f::getTransformForNormals(view.upperLeft())
                                * options.ssct.lightDirection);
                mi->setParameter("ssctIntensity",
                        options.ssct.enabled ? options.ssct.intensity : 0.0f);
@@ -972,8 +971,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::generateGaussianMipmap(Frame
    auto from = input;
    for (size_t i = 0; i < levels - 1; i++) {
        auto output = mipmapPass->out[i];
-        gaussianBlurPass(fg, from, output, reinhard, kernelWidth, sigma);
-        from = output;
+        from = gaussianBlurPass(fg, from, output, reinhard, kernelWidth, sigma);
        reinhard = false; // only do the reinhard filtering on the first level
    }

@@ -1034,7 +1032,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::gaussianBlurPass(FrameGraph&
    // and because it's a separable filter, the effective 2D filter kernel size is 17*17
    // The total number of samples needed over the two passes is 18.

-    fg.addPass<BlurPassData>("Gaussian Blur Pass (separable)",
+    auto& blurPass = fg.addPass<BlurPassData>("Gaussian Blur Pass (separable)",
            [&](FrameGraph::Builder& builder, auto& data) {
                auto inDesc = builder.getDescriptor(input);

@@ -1142,7 +1140,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::gaussianBlurPass(FrameGraph&
                render(hwOutRT, separableGaussianBlur.getPipelineState(mEngine), driver);
            });

-    return output;
+    return blurPass->out;
 }

 PostProcessManager::ScreenSpaceRefConfig PostProcessManager::prepareMipmapSSR(FrameGraph& fg,
@@ -1341,7 +1339,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::generateMipmapSSR(
            input = ppm.resolveBaseLevel(fg, "ssr", input);
            // then blit into an appropriate texture
            // this handles scaling, format conversion and mipmaping
-            input = ppm.opaqueBlit(fg, input, outDesc);
+            input = ppm.opaqueBlit(fg, input, { 0, 0, desc.width, desc.height }, outDesc);
        }
    }

@@ -1465,18 +1463,21 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::dof(FrameGraph& fg,
    };

    /*
-     * dofResolution is used (at compile time for now) to chose between full- or quarter-resolution
+     * dofResolution is used to chose between full- or quarter-resolution
     * for the DoF calculations. Set to [1] for full resolution or [2] for quarter-resolution.
     */
    const uint32_t dofResolution = dofOptions.nativeResolution ? 1u : 2u;

-    constexpr const uint32_t maxMipLevels = 4u; // mip levels at full-resolution
-    constexpr const uint32_t maxMipLevelsMask = (1u << maxMipLevels) - 1u;
    auto const& colorDesc = fg.getDescriptor(input);
-    const uint32_t width  = ((colorDesc.width  + maxMipLevelsMask) & ~maxMipLevelsMask) / dofResolution;
-    const uint32_t height = ((colorDesc.height + maxMipLevelsMask) & ~maxMipLevelsMask) / dofResolution;
-    const uint8_t maxLevelCount = FTexture::maxLevelCount(width, height);
-    uint8_t mipmapCount = min(maxLevelCount, uint8_t(maxMipLevels));
+    const uint32_t width  = colorDesc.width  / dofResolution;
+    const uint32_t height = colorDesc.height / dofResolution;
+
+    // at full resolution, 4 "safe" levels are guaranteed
+    constexpr const uint32_t maxMipLevels = 4u;
+
+    // compute numbers of "safe" levels (should be 4, but can be 3 at half res)
+    const uint8_t mipmapCount = std::min(maxMipLevels, ctz(width | height));
+    assert_invariant(mipmapCount == maxMipLevels || mipmapCount == maxMipLevels-1);

    /*
     * Setup:
@@ -1894,22 +1895,14 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloom(FrameGraph& fg,
 PostProcessManager::BloomPassOutput PostProcessManager::bloomPass(FrameGraph& fg,
        FrameGraphId<FrameGraphTexture> input, TextureFormat outFormat,
        BloomOptions& inoutBloomOptions, float2 scale) noexcept {
-    // Chrome does not support feedback loops in WebGL 2.0. See also:
-    // https://bugs.chromium.org/p/chromium/issues/detail?id=1066201
-#if defined(__EMSCRIPTEN__)
-    constexpr bool isWebGL = true;
-#else
-    constexpr bool isWebGL = false;
-#endif
-
    // Figure out a good size for the bloom buffer.
    auto const& desc = fg.getDescriptor(input);

    // width and height after dynamic resolution upscaling
-    const float aspect = (desc.width * scale.y) / (desc.height * scale.x);
+    const float aspect = (float(desc.width) * scale.y) / (float(desc.height) * scale.x);

    // compute the desired bloom buffer size
-    float bloomHeight = inoutBloomOptions.resolution;
+    float bloomHeight = float(inoutBloomOptions.resolution);
    float bloomWidth  = bloomHeight * aspect;

    // Anamorphic bloom by always scaling down one of the dimension -- we do this (as opposed
@@ -1923,11 +1916,11 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloomPass(FrameGraph& fg
    }

    // convert back to integer width/height
-    const uint32_t width  = std::max(1.0f, std::floor(bloomWidth));
-    const uint32_t height = std::max(1.0f, std::floor(bloomHeight));
+    const uint32_t width  = std::max(1u, uint32_t(std::floor(bloomWidth)));
+    const uint32_t height = std::max(1u, uint32_t(std::floor(bloomHeight)));

    // we might need to adjust the max # of levels
-    const uint32_t major = std::max(bloomWidth,  bloomHeight);
+    const uint32_t major = uint32_t(std::max(bloomWidth,  bloomHeight));
    const uint8_t maxLevels = FTexture::maxLevelCount(major);
    inoutBloomOptions.levels = std::min(inoutBloomOptions.levels, maxLevels);
    inoutBloomOptions.levels = std::min(inoutBloomOptions.levels, kMaxBloomLevels);
@@ -1935,295 +1928,161 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloomPass(FrameGraph& fg
    if (2 * width < desc.width || 2 * height < desc.height) {
        // if we're scaling down by more than 2x, prescale the image with a blit to improve
        // performance. This is important on mobile/tilers.
-        input = opaqueBlit(fg, input, {
+        input = opaqueBlit(fg, input, { 0, 0, desc.width, desc.height }, {
                .width = std::max(1u, desc.width / 2),
                .height = std::max(1u, desc.height / 2),
                .format = outFormat
        });
    }

-    if constexpr(!isWebGL) {
+    struct BloomPassData {
+        FrameGraphId<FrameGraphTexture> in;
+        FrameGraphId<FrameGraphTexture> out;
+        FrameGraphId<FrameGraphTexture> stage;
+        uint32_t outRT[kMaxBloomLevels];
+        uint32_t stageRT[kMaxBloomLevels];
+    };

-        struct BloomPassData {
-            FrameGraphId<FrameGraphTexture> in;
-            FrameGraphId<FrameGraphTexture> out;
-        };
+    // downsample phase
+    auto& bloomDownsamplePass = fg.addPass<BloomPassData>("Bloom Downsample",
+            [&](FrameGraph::Builder& builder, auto& data) {
+                data.in = builder.sample(input);
+                data.out = builder.createTexture("Bloom Out Texture", {
+                        .width = width,
+                        .height = height,
+                        .levels = inoutBloomOptions.levels,
+                        .format = outFormat
+                });
+                data.out = builder.sample(data.out);

-        // downsample phase
-        auto& bloomDownsamplePass = fg.addPass<BloomPassData>("Bloom Downsample",
-                [&](FrameGraph::Builder& builder, auto& data) {
-                    data.in = builder.sample(input);
-                    data.out = builder.createTexture("Bloom Texture", {
-                            .width = width,
-                            .height = height,
-                            .levels = inoutBloomOptions.levels,
-                            .format = outFormat
-                    });
+                data.stage = builder.createTexture("Bloom Stage Texture", {
+                        .width = width,
+                        .height = height,
+                        .levels = inoutBloomOptions.levels,
+                        .format = outFormat
+                });
+                data.stage = builder.sample(data.stage);

-                    data.out = builder.sample(data.out);
-                    for (size_t i = 0; i < inoutBloomOptions.levels; i++) {
-                        auto out = builder.createSubresource(data.out, "Bloom Texture mip",
-                                { .level = uint8_t(i) });
-                        builder.declareRenderPass(out);
-                    }
-                },
-                [=](FrameGraphResources const& resources,
-                        auto const& data, DriverApi& driver) {
+                for (size_t i = 0; i < inoutBloomOptions.levels; i++) {
+                    auto out = builder.createSubresource(data.out, "Bloom Out Texture mip",
+                            { .level = uint8_t(i) });
+                    auto stage = builder.createSubresource(data.stage,
+                            "Bloom Stage Texture mip", { .level = uint8_t(i) });
+                    builder.declareRenderPass(out, &data.outRT[i]);
+                    builder.declareRenderPass(stage, &data.stageRT[i]);
+                }
+            },
+            [=](FrameGraphResources const& resources,
+                    auto const& data, DriverApi& driver) {

-                    auto const& material = getPostProcessMaterial("bloomDownsample");
-                    FMaterialInstance* mi = material.getMaterialInstance(mEngine);
+                auto const& material = getPostProcessMaterial("bloomDownsample");
+                FMaterialInstance* mi = material.getMaterialInstance(mEngine);

-                    const PipelineState pipeline(material.getPipelineState(mEngine));
+                const PipelineState pipeline(material.getPipelineState(mEngine));

-                    auto hwIn = resources.getTexture(data.in);
-                    auto hwOut = resources.getTexture(data.out);
-                    auto const& outDesc = resources.getDescriptor(data.out);
+                auto hwIn = resources.getTexture(data.in);
+                auto hwOut = resources.getTexture(data.out);
+                auto hwStage = resources.getTexture(data.stage);

-                    mi->use(driver);
-                    mi->setParameter("source", hwIn, {
+                mi->use(driver);
+                mi->setParameter("source", hwIn, {
+                        .filterMag = SamplerMagFilter::LINEAR,
+                        .filterMin = SamplerMinFilter::LINEAR /* level is always 0 */
+                });
+                mi->setParameter("level", 0.0f);
+                mi->setParameter("threshold", inoutBloomOptions.threshold ? 1.0f : 0.0f);
+                mi->setParameter("invHighlight",
+                        std::isinf(inoutBloomOptions.highlight) ? 0.0f : 1.0f
+                                                                            / inoutBloomOptions.highlight);
+
+                for (size_t i = 0; i < inoutBloomOptions.levels; i++) {
+                    const bool parity = (i % 2) == 0;
+                    auto hwDstRT = resources.getRenderPassInfo(
+                            parity ? data.outRT[i] : data.stageRT[i]);
+
+                    mi->commit(driver);
+
+                    hwDstRT.params.flags.discardStart = TargetBufferFlags::COLOR;
+                    hwDstRT.params.flags.discardEnd = TargetBufferFlags::NONE;
+                    render(hwDstRT, pipeline, driver);
+
+                    // prepare the next level
+                    mi->setParameter("source", parity ? hwOut : hwStage, {
                            .filterMag = SamplerMagFilter::LINEAR,
-                            .filterMin = SamplerMinFilter::LINEAR /* level is always 0 */
+                            .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
                    });
-                    mi->setParameter("level", 0.0f);
-                    mi->setParameter("threshold", inoutBloomOptions.threshold ? 1.0f : 0.0f);
-                    mi->setParameter("invHighlight",
-                            std::isinf(inoutBloomOptions.highlight) ? 0.0f : 1.0f
-                                                                             / inoutBloomOptions.highlight);
+                    mi->setParameter("level", float(i));
+                }
+            });

-                    for (size_t i = 0; i < inoutBloomOptions.levels; i++) {
-                        auto hwOutRT = resources.getRenderPassInfo(i);
+    FrameGraphId<FrameGraphTexture> output = bloomDownsamplePass->out;
+    FrameGraphId<FrameGraphTexture> stage = bloomDownsamplePass->stage;

-                        auto w = FTexture::valueForLevel(i, outDesc.width);
-                        auto h = FTexture::valueForLevel(i, outDesc.height);
-                        mi->setParameter("resolution", float4{ w, h, 1.0f / w, 1.0f / h });
-                        mi->commit(driver);
+    // flare pass
+    auto flare = flarePass(fg, bloomDownsamplePass->out, width, height, outFormat, inoutBloomOptions);

-                        hwOutRT.params.flags.discardStart = TargetBufferFlags::COLOR;
-                        hwOutRT.params.flags.discardEnd = TargetBufferFlags::NONE;
-                        render(hwOutRT, pipeline, driver);
+    // upsample phase
+    auto& bloomUpsamplePass = fg.addPass<BloomPassData>("Bloom Upsample",
+            [&](FrameGraph::Builder& builder, auto& data) {
+                data.out = builder.sample(output);
+                data.stage = builder.sample(stage);
+                for (size_t i = 0; i < inoutBloomOptions.levels; i++) {
+                    auto out = builder.createSubresource(data.out, "Bloom Out Texture mip",
+                            { .level = uint8_t(i) });
+                    auto stage = builder.createSubresource(data.stage,
+                            "Bloom Stage Texture mip", { .level = uint8_t(i) });
+                    builder.declareRenderPass(out, &data.outRT[i]);
+                    builder.declareRenderPass(stage, &data.stageRT[i]);
+                }
+            },
+            [=](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {

-                        // prepare the next level
-                        mi->setParameter("source", hwOut, {
-                                .filterMag = SamplerMagFilter::LINEAR,
-                                .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
-                        });
-                        mi->setParameter("level", float(i));
-                        driver.setMinMaxLevels(hwOut, i,
-                                i); // safe because we're using LINEAR_MIPMAP_NEAREST
-                    }
-                    driver.setMinMaxLevels(hwOut, 0, inoutBloomOptions.levels - 1);
-                });
+                auto hwOut = resources.getTexture(data.out);
+                auto hwStage = resources.getTexture(data.stage);
+                auto const& outDesc = resources.getDescriptor(data.out);

-        input = bloomDownsamplePass->out;
+                auto const& material = getPostProcessMaterial("bloomUpsample");
+                FMaterialInstance* mi = material.getMaterialInstance(mEngine);
+                PipelineState pipeline(material.getPipelineState(mEngine));
+                pipeline.rasterState.blendFunctionSrcRGB = BlendFunction::ONE;
+                pipeline.rasterState.blendFunctionDstRGB = BlendFunction::ONE;

-        // flare pass
-        auto flare = flarePass(fg, input, width, height, outFormat, inoutBloomOptions);
+                mi->use(driver);

-        // upsample phase
-        auto& bloomUpsamplePass = fg.addPass<BloomPassData>("Bloom Upsample",
-                [&](FrameGraph::Builder& builder, auto& data) {
-                    data.in = builder.sample(input);
-                    data.out = input;
-                    for (size_t i = 0; i < inoutBloomOptions.levels; i++) {
-                        auto out = builder.createSubresource(data.out, "Bloom Texture mip",
-                                { .level = uint8_t(i) });
-                        builder.declareRenderPass(out);
-                    }
-                },
-                [=](FrameGraphResources const& resources,
-                        auto const& data, DriverApi& driver) {
+                for (size_t j = inoutBloomOptions.levels, i = j - 1; i >= 1; i--, j++) {
+                    const bool parity = (j % 2) == 0;

-                    auto hwIn = resources.getTexture(data.in);
-                    auto const& outDesc = resources.getDescriptor(data.out);
+                    auto hwDstRT = resources.getRenderPassInfo(
+                            parity ? data.outRT[i - 1] : data.stageRT[i - 1]);
+                    hwDstRT.params
+                            .flags
+                            .discardStart = TargetBufferFlags::NONE; // because we'll blend
+                    hwDstRT.params.flags.discardEnd = TargetBufferFlags::NONE;

-                    auto const& material = getPostProcessMaterial("bloomUpsample");
-                    FMaterialInstance* mi = material.getMaterialInstance(mEngine);
-                    PipelineState pipeline(material.getPipelineState(mEngine));
-                    pipeline.rasterState.blendFunctionSrcRGB = BlendFunction::ONE;
-                    pipeline.rasterState.blendFunctionDstRGB = BlendFunction::ONE;
-
-                    mi->use(driver);
-
-                    for (size_t i = inoutBloomOptions.levels - 1; i >= 1; i--) {
-                        auto hwDstRT = resources.getRenderPassInfo(i - 1);
-                        hwDstRT.params
-                               .flags
-                               .discardStart = TargetBufferFlags::NONE; // because we'll blend
-                        hwDstRT.params.flags.discardEnd = TargetBufferFlags::NONE;
-
-                        auto w = FTexture::valueForLevel(i - 1, outDesc.width);
-                        auto h = FTexture::valueForLevel(i - 1, outDesc.height);
-                        mi->setParameter("resolution", float4{ w, h, 1.0f / w, 1.0f / h });
-                        mi->setParameter("source", hwIn, {
-                                .filterMag = SamplerMagFilter::LINEAR,
-                                .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
-                        });
-                        mi->setParameter("level", float(i));
-                        driver.setMinMaxLevels(hwIn, i, i);
-                        mi->commit(driver);
-                        render(hwDstRT, pipeline, driver);
-                    }
-
-                    driver.setMinMaxLevels(hwIn, 0, inoutBloomOptions.levels - 1);
-                });
-
-        return { bloomUpsamplePass->out, flare };
-
-    } else { // !isWebGL
-
-        struct BloomPassData {
-            FrameGraphId<FrameGraphTexture> in;
-            FrameGraphId<FrameGraphTexture> out;
-            FrameGraphId<FrameGraphTexture> stage;
-            uint32_t outRT[kMaxBloomLevels];
-            uint32_t stageRT[kMaxBloomLevels];
-        };
-
-        // downsample phase
-        auto& bloomDownsamplePass = fg.addPass<BloomPassData>("Bloom Downsample",
-                [&](FrameGraph::Builder& builder, auto& data) {
-                    data.in = builder.sample(input);
-                    data.out = builder.createTexture("Bloom Out Texture", {
-                            .width = width,
-                            .height = height,
-                            .levels = inoutBloomOptions.levels,
-                            .format = outFormat
-                    });
-                    data.out = builder.sample(data.out);
-
-                    data.stage = builder.createTexture("Bloom Stage Texture", {
-                            .width = width,
-                            .height = height,
-                            .levels = inoutBloomOptions.levels,
-                            .format = outFormat
-                    });
-                    data.stage = builder.sample(data.stage);
-
-                    for (size_t i = 0; i < inoutBloomOptions.levels; i++) {
-                        auto out = builder.createSubresource(data.out, "Bloom Out Texture mip",
-                                { .level = uint8_t(i) });
-                        auto stage = builder.createSubresource(data.stage,
-                                "Bloom Stage Texture mip", { .level = uint8_t(i) });
-                        builder.declareRenderPass(out, &data.outRT[i]);
-                        builder.declareRenderPass(stage, &data.stageRT[i]);
-                    }
-                },
-                [=](FrameGraphResources const& resources,
-                        auto const& data, DriverApi& driver) {
-
-                    auto const& material = getPostProcessMaterial("bloomDownsample");
-                    FMaterialInstance* mi = material.getMaterialInstance(mEngine);
-
-                    const PipelineState pipeline(material.getPipelineState(mEngine));
-
-                    auto hwIn = resources.getTexture(data.in);
-                    auto hwOut = resources.getTexture(data.out);
-                    auto hwStage = resources.getTexture(data.stage);
-                    auto const& outDesc = resources.getDescriptor(data.out);
-
-                    mi->use(driver);
-                    mi->setParameter("source", hwIn, {
+                    auto w = FTexture::valueForLevel(i - 1, outDesc.width);
+                    auto h = FTexture::valueForLevel(i - 1, outDesc.height);
+                    mi->setParameter("resolution", float4{ w, h, 1.0f / w, 1.0f / h });
+                    mi->setParameter("source", parity ? hwStage : hwOut, {
                            .filterMag = SamplerMagFilter::LINEAR,
-                            .filterMin = SamplerMinFilter::LINEAR /* level is always 0 */
+                            .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
                    });
-                    mi->setParameter("level", 0.0f);
-                    mi->setParameter("threshold", inoutBloomOptions.threshold ? 1.0f : 0.0f);
-                    mi->setParameter("invHighlight",
-                            std::isinf(inoutBloomOptions.highlight) ? 0.0f : 1.0f
-                                                                             / inoutBloomOptions.highlight);
+                    mi->setParameter("level", float(i));
+                    mi->commit(driver);
+                    render(hwDstRT, pipeline, driver);
+                }

-                    for (size_t i = 0; i < inoutBloomOptions.levels; i++) {
-                        const bool parity = (i % 2) == 0;
-                        auto hwDstRT = resources.getRenderPassInfo(
-                                parity ? data.outRT[i] : data.stageRT[i]);
+                // Every other level is missing from the out texture, so we need to do
+                // blits to complete the chain.
+                const SamplerMagFilter filter = SamplerMagFilter::NEAREST;
+                for (size_t i = 1; i < inoutBloomOptions.levels; i += 2) {
+                    auto in = resources.getRenderPassInfo(data.stageRT[i]);
+                    auto out = resources.getRenderPassInfo(data.outRT[i]);
+                    driver.blit(TargetBufferFlags::COLOR, out.target, out.params.viewport,
+                            in.target, in.params.viewport, filter);
+                }
+            });

-                        auto w = FTexture::valueForLevel(i, outDesc.width);
-                        auto h = FTexture::valueForLevel(i, outDesc.height);
-                        mi->setParameter("resolution", float4{ w, h, 1.0f / w, 1.0f / h });
-                        mi->commit(driver);
-
-                        hwDstRT.params.flags.discardStart = TargetBufferFlags::COLOR;
-                        hwDstRT.params.flags.discardEnd = TargetBufferFlags::NONE;
-                        render(hwDstRT, pipeline, driver);
-
-                        // prepare the next level
-                        mi->setParameter("source", parity ? hwOut : hwStage, {
-                                .filterMag = SamplerMagFilter::LINEAR,
-                                .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
-                        });
-                        mi->setParameter("level", float(i));
-                    }
-                });
-
-        FrameGraphId<FrameGraphTexture> output = bloomDownsamplePass->out;
-        FrameGraphId<FrameGraphTexture> stage = bloomDownsamplePass->stage;
-
-        // flare pass
-        auto flare = flarePass(fg, bloomDownsamplePass->out, width, height, outFormat, inoutBloomOptions);
-
-        // upsample phase
-        auto& bloomUpsamplePass = fg.addPass<BloomPassData>("Bloom Upsample",
-                [&](FrameGraph::Builder& builder, auto& data) {
-                    data.out = builder.sample(output);
-                    data.stage = builder.sample(stage);
-                    for (size_t i = 0; i < inoutBloomOptions.levels; i++) {
-                        auto out = builder.createSubresource(data.out, "Bloom Out Texture mip",
-                                { .level = uint8_t(i) });
-                        auto stage = builder.createSubresource(data.stage,
-                                "Bloom Stage Texture mip", { .level = uint8_t(i) });
-                        builder.declareRenderPass(out, &data.outRT[i]);
-                        builder.declareRenderPass(stage, &data.stageRT[i]);
-                    }
-                },
-                [=](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {
-
-                    auto hwOut = resources.getTexture(data.out);
-                    auto hwStage = resources.getTexture(data.stage);
-                    auto const& outDesc = resources.getDescriptor(data.out);
-
-                    auto const& material = getPostProcessMaterial("bloomUpsample");
-                    FMaterialInstance* mi = material.getMaterialInstance(mEngine);
-                    PipelineState pipeline(material.getPipelineState(mEngine));
-                    pipeline.rasterState.blendFunctionSrcRGB = BlendFunction::ONE;
-                    pipeline.rasterState.blendFunctionDstRGB = BlendFunction::ONE;
-
-                    mi->use(driver);
-
-                    for (size_t j = inoutBloomOptions.levels, i = j - 1; i >= 1; i--, j++) {
-                        const bool parity = (j % 2) == 0;
-
-                        auto hwDstRT = resources.getRenderPassInfo(
-                                parity ? data.outRT[i - 1] : data.stageRT[i - 1]);
-                        hwDstRT.params
-                               .flags
-                               .discardStart = TargetBufferFlags::NONE; // because we'll blend
-                        hwDstRT.params.flags.discardEnd = TargetBufferFlags::NONE;
-
-                        auto w = FTexture::valueForLevel(i - 1, outDesc.width);
-                        auto h = FTexture::valueForLevel(i - 1, outDesc.height);
-                        mi->setParameter("resolution", float4{ w, h, 1.0f / w, 1.0f / h });
-                        mi->setParameter("source", parity ? hwStage : hwOut, {
-                                .filterMag = SamplerMagFilter::LINEAR,
-                                .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
-                        });
-                        mi->setParameter("level", float(i));
-                        mi->commit(driver);
-                        render(hwDstRT, pipeline, driver);
-                    }
-
-                    // Every other level is missing from the out texture, so we need to do
-                    // blits to complete the chain.
-                    const SamplerMagFilter filter = SamplerMagFilter::NEAREST;
-                    for (size_t i = 1; i < inoutBloomOptions.levels; i += 2) {
-                        auto in = resources.getRenderPassInfo(data.stageRT[i]);
-                        auto out = resources.getRenderPassInfo(data.outRT[i]);
-                        driver.blit(TargetBufferFlags::COLOR, out.target, out.params.viewport,
-                                in.target, in.params.viewport, filter);
-                    }
-                });
-
-        return { bloomUpsamplePass->out, flare };
-    }
+    return { bloomUpsamplePass->out, flare };
 }

 UTILS_NOINLINE
@@ -2406,14 +2265,13 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::customResolveUncompressPass(
 }

 FrameGraphId<FrameGraphTexture> PostProcessManager::colorGrading(FrameGraph& fg,
-        FrameGraphId<FrameGraphTexture> input,
+        FrameGraphId<FrameGraphTexture> input, filament::Viewport const& vp,
        FrameGraphId<FrameGraphTexture> bloom,
        FrameGraphId<FrameGraphTexture> flare,
        FColorGrading const* colorGrading,
        ColorGradingConfig const& colorGradingConfig,
        BloomOptions const& bloomOptions,
-        VignetteOptions const& vignetteOptions,
-        float2 scale) noexcept
+        VignetteOptions const& vignetteOptions) noexcept
 {
    FrameGraphId<FrameGraphTexture> bloomDirt;
    FrameGraphId<FrameGraphTexture> starburst;
@@ -2450,11 +2308,10 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::colorGrading(FrameGraph& fg,

    auto& ppColorGrading = fg.addPass<PostProcessColorGrading>("colorGrading",
            [&](FrameGraph::Builder& builder, auto& data) {
-                auto const& inputDesc = fg.getDescriptor(input);
                data.input = builder.sample(input);
                data.output = builder.createTexture("colorGrading output", {
-                        .width = inputDesc.width,
-                        .height = inputDesc.height,
+                        .width = vp.width,
+                        .height = vp.height,
                        .format = colorGradingConfig.ldrFormat
                });
                data.output = builder.declareRenderPass(data.output);
@@ -2531,6 +2388,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::colorGrading(FrameGraph& fg,
                    bloomParameters.y = 1.0f - bloomParameters.x;
                }

+                auto const& input = resources.getDescriptor(data.input);
                auto const& output = resources.getDescriptor(data.output);
                float4 vignetteParameters = getVignetteParameters(
                        vignetteOptions, output.width, output.height);
@@ -2543,6 +2401,12 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::colorGrading(FrameGraph& fg,
                mi->setParameter("vignetteColor", vignetteOptions.color);
                mi->setParameter("fxaa", colorGradingConfig.fxaa);
                mi->setParameter("temporalNoise", temporalNoise);
+                mi->setParameter("viewport", float4{
+                        (float)vp.left   / input.width,
+                        (float)vp.bottom / input.height,
+                        (float)vp.width  / input.width,
+                        (float)vp.height / input.height
+                });

                const uint8_t variant = uint8_t(colorGradingConfig.translucent ?
                            PostProcessVariant::TRANSLUCENT : PostProcessVariant::OPAQUE);
@@ -2555,7 +2419,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::colorGrading(FrameGraph& fg,
 }

 FrameGraphId<FrameGraphTexture> PostProcessManager::fxaa(FrameGraph& fg,
-        FrameGraphId<FrameGraphTexture> input,
+        FrameGraphId<FrameGraphTexture> input, filament::Viewport const& vp,
        TextureFormat outFormat, bool translucent) noexcept {

    struct PostProcessFXAA {
@@ -2565,17 +2429,16 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::fxaa(FrameGraph& fg,

    auto& ppFXAA = fg.addPass<PostProcessFXAA>("fxaa",
            [&](FrameGraph::Builder& builder, auto& data) {
-                auto const& inputDesc = fg.getDescriptor(input);
                data.input = builder.sample(input);
                data.output = builder.createTexture("fxaa output", {
-                        .width = inputDesc.width,
-                        .height = inputDesc.height,
+                        .width = vp.width,
+                        .height = vp.height,
                        .format = outFormat
                });
                data.output = builder.declareRenderPass(data.output);
            },
-            [=](FrameGraphResources const& resources,
-                auto const& data, DriverApi& driver) {
+            [=](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {
+                auto const& inDesc = resources.getDescriptor(data.input);
                auto const& texture = resources.getTexture(data.input);
                auto const& out = resources.getRenderPassInfo();

@@ -2585,6 +2448,13 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::fxaa(FrameGraph& fg,
                        .filterMag = SamplerMagFilter::LINEAR,
                        .filterMin = SamplerMinFilter::LINEAR
                });
+                mi->setParameter("viewport", float4{
+                        (float)vp.left   / inDesc.width,
+                        (float)vp.bottom / inDesc.height,
+                        (float)vp.width  / inDesc.width,
+                        (float)vp.height / inDesc.height
+                });
+                mi->setParameter("texelSize", 1.0f / float2{ inDesc.width, inDesc.height });

                const uint8_t variant = uint8_t(translucent ?
                    PostProcessVariant::TRANSLUCENT : PostProcessVariant::OPAQUE);
@@ -2604,7 +2474,7 @@ void PostProcessManager::prepareTaa(FrameGraph& fg, filament::Viewport const& sv
    auto& current = frameHistory.getCurrent().*pTaa;

    // compute projection
-    current.projection = inoutCameraInfo->projection * (inoutCameraInfo->view * inoutCameraInfo->worldOrigin);
+    current.projection = mat4f{ inoutCameraInfo->projection * inoutCameraInfo->getUserViewMatrix() };
    current.frameId = previous.frameId + 1;

    // sample position within a pixel [-0.5, 0.5]
@@ -2701,7 +2571,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
                    float2 d = sampleOffsets[i] - current.jitter;
                    d *= 1.0f / taaOptions.filterWidth;
                    // this is a gaussian fit of a 3.3 Blackman Harris window
-                    // see: "High Quality Temporal Supersampling" by Bruan Karis
+                    // see: "High Quality Temporal Supersampling" by Brian Karis
                    weights[i] = std::exp2(-3.3f * (d.x * d.x + d.y * d.y));
                    sum += weights[i];
                }
@@ -2769,7 +2639,8 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
 }

 FrameGraphId<FrameGraphTexture> PostProcessManager::opaqueBlit(FrameGraph& fg,
-        FrameGraphId<FrameGraphTexture> input, FrameGraphTexture::Descriptor const& outDesc,
+        FrameGraphId<FrameGraphTexture> input, filament::Viewport const& vp,
+        FrameGraphTexture::Descriptor const& outDesc,
        SamplerMagFilter filter) noexcept {

    struct PostProcessScaling {
@@ -2795,7 +2666,9 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::opaqueBlit(FrameGraph& fg,
                // HwTexture handle. Using a RenderPass works because data.input will resolve
                // to the actual imported render target and will have the correct viewport.
                builder.declareRenderPass("opaque blit input", {
-                        .attachments = { .color = { data.input }}});
+                        .attachments = { .color = { data.input }},
+                        .viewport = vp
+                });
            },
            [=](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {
                auto out = resources.getRenderPassInfo(0);
@@ -2812,7 +2685,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::opaqueBlit(FrameGraph& fg,

 FrameGraphId<FrameGraphTexture> PostProcessManager::blendBlit(
        FrameGraph& fg, bool translucent, DynamicResolutionOptions dsrOptions,
-        FrameGraphId<FrameGraphTexture> input,
+        FrameGraphId<FrameGraphTexture> input, filament::Viewport const& vp,
        FrameGraphTexture::Descriptor const& outDesc) noexcept {

    Handle<HwRenderPrimitive> fullScreenRenderPrimitive = mEngine.getFullScreenRenderPrimitive();
@@ -2829,13 +2702,12 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::blendBlit(
                || dsrOptions.quality == QualityLevel::HIGH);

    // helper to set the EASU uniforms
-    auto setEasuUniforms = [](FMaterialInstance* mi,
+    auto setEasuUniforms = [vp](FMaterialInstance* mi,
            FrameGraphTexture::Descriptor const& inputDesc,
            FrameGraphTexture::Descriptor const& outputDesc) {
        FSRUniforms uniforms{};
        FSR_ScalingSetup(&uniforms, {
-                .viewportWidth = inputDesc.width,
-                .viewportHeight = inputDesc.height,
+                .input = vp,
                .inputWidth = inputDesc.width,
                .inputHeight = inputDesc.height,
                .outputWidth = outputDesc.width,
@@ -2908,6 +2780,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::blendBlit(
                            float4{ outputDesc.width, outputDesc.height,
                                    1.0f / outputDesc.width, 1.0f / outputDesc.height });
                    mi->commit(driver);
+                    mi->use(driver);
                }

                { // just a scope to not leak local variables
@@ -2926,7 +2799,14 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::blendBlit(
                    mi->setParameter("resolution",
                            float4{ outputDesc.width, outputDesc.height,
                                    1.0f / outputDesc.width, 1.0f / outputDesc.height });
+                    mi->setParameter("viewport", float4{
+                            (float)vp.left   / inputDesc.width,
+                            (float)vp.bottom / inputDesc.height,
+                            (float)vp.width  / inputDesc.width,
+                            (float)vp.height / inputDesc.height
+                    });
                    mi->commit(driver);
+                    mi->use(driver);
                }

                // --------------------------------------------------------------------------------
@@ -2934,37 +2814,27 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::blendBlit(

                auto out = resources.getRenderPassInfo();

-                if (twoPassesEASU) {
-                    auto* mi = splitEasuMaterial->getMaterialInstance(mEngine);
-                    mi->use(driver);
-                    PipelineState pipeline(splitEasuMaterial->getPipelineState(mEngine));
+                if (UTILS_UNLIKELY(twoPassesEASU)) {
+                    PipelineState pipeline0(splitEasuMaterial->getPipelineState(mEngine));
+                    PipelineState pipeline1(easuMaterial->getPipelineState(mEngine));
+                    pipeline1.rasterState.depthFunc = backend::SamplerCompareFunc::NE;
                    if (translucent) {
-                        enableTranslucentBlending(pipeline);
+                        enableTranslucentBlending(pipeline0);
+                        enableTranslucentBlending(pipeline1);
                    }
-
                    driver.beginRenderPass(out.target, out.params);
-                    driver.draw(pipeline, fullScreenRenderPrimitive, 1);
-                }
-
-                { // scope to not leak local variables
-                    auto* mi = easuMaterial->getMaterialInstance(mEngine);
-                    mi->use(driver);
-
+                    driver.draw(pipeline0, fullScreenRenderPrimitive, 1);
+                    driver.draw(pipeline1, fullScreenRenderPrimitive, 1);
+                    driver.endRenderPass();
+                } else {
                    PipelineState pipeline(easuMaterial->getPipelineState(mEngine));
                    if (translucent) {
                        enableTranslucentBlending(pipeline);
                    }
-                    if (twoPassesEASU) {
-                        pipeline.rasterState.depthFunc = backend::SamplerCompareFunc::NE;
-                    }
-
-                    if (!twoPassesEASU) {
-                        driver.beginRenderPass(out.target, out.params);
-                    }
+                    driver.beginRenderPass(out.target, out.params);
                    driver.draw(pipeline, fullScreenRenderPrimitive, 1);
+                    driver.endRenderPass();
                }
-
-                driver.endRenderPass();
            });

    auto output = ppQuadBlit->output;
@@ -162,14 +162,13 @@ public:

        // Color grading, tone mapping, dithering and bloom
    FrameGraphId<FrameGraphTexture> colorGrading(FrameGraph& fg,
-            FrameGraphId<FrameGraphTexture> input,
+            FrameGraphId<FrameGraphTexture> input, filament::Viewport const& vp,
            FrameGraphId<FrameGraphTexture> bloom,
            FrameGraphId<FrameGraphTexture> flare,
            const FColorGrading* colorGrading,
            ColorGradingConfig const& colorGradingConfig,
            BloomOptions const& bloomOptions,
-            VignetteOptions const& vignetteOptions,
-            math::float2 scale) noexcept;
+            VignetteOptions const& vignetteOptions) noexcept;

    void colorGradingPrepareSubpass(backend::DriverApi& driver, const FColorGrading* colorGrading,
            ColorGradingConfig const& colorGradingConfig,
@@ -188,8 +187,8 @@ public:

    // Anti-aliasing
    FrameGraphId<FrameGraphTexture> fxaa(FrameGraph& fg,
-            FrameGraphId<FrameGraphTexture> input, backend::TextureFormat outFormat,
-            bool translucent) noexcept;
+            FrameGraphId<FrameGraphTexture> input, filament::Viewport const& vp,
+            backend::TextureFormat outFormat, bool translucent) noexcept;

    // Temporal Anti-aliasing
    void prepareTaa(FrameGraph& fg, filament::Viewport const& svp,
@@ -208,12 +207,13 @@ public:

    // Blit/rescaling/resolves
    FrameGraphId<FrameGraphTexture> opaqueBlit(FrameGraph& fg,
-            FrameGraphId<FrameGraphTexture> input, FrameGraphTexture::Descriptor const& outDesc,
+            FrameGraphId<FrameGraphTexture> input, filament::Viewport const& vp,
+            FrameGraphTexture::Descriptor const& outDesc,
            backend::SamplerMagFilter filter = backend::SamplerMagFilter::LINEAR) noexcept;

    FrameGraphId<FrameGraphTexture> blendBlit(
            FrameGraph& fg, bool translucent, DynamicResolutionOptions dsrOptions,
-            FrameGraphId<FrameGraphTexture> input,
+            FrameGraphId<FrameGraphTexture> input, filament::Viewport const& vp,
            FrameGraphTexture::Descriptor const& outDesc) noexcept;

    // resolve base level of input and outputs a 1-level texture
@@ -74,6 +74,11 @@ void RenderPass::setGeometry(FScene::RenderableSoa const& soa, Range<uint32_t> v
    mUboHandle = uboHandle;
 }

+void RenderPass::setCamera(const CameraInfo& camera) noexcept {
+    mCameraPosition = camera.getPosition();
+    mCameraForwardVector = camera.getForwardVector();
+}
+
 void RenderPass::overridePolygonOffset(backend::PolygonOffset* polygonOffset) noexcept {
    if ((mPolygonOffsetOverride = (polygonOffset != nullptr))) {
        mPolygonOffset = *polygonOffset;
@@ -97,7 +102,6 @@ void RenderPass::appendCommands(CommandTypeFlags const commandTypeFlags) noexcep
    const RenderFlags renderFlags = mFlags;
    const Variant variant = mVariant;
    const FScene::VisibleMaskType visibilityMask = mVisibilityMask;
-    CameraInfo const& camera = mCamera;

    // up-to-date summed primitive counts needed for generateCommands()
    FScene::RenderableSoa const& soa = *mRenderableSoa;
@@ -112,9 +116,8 @@ void RenderPass::appendCommands(CommandTypeFlags const commandTypeFlags) noexcep
    commandCount += 1; // for the sentinel
    Command* const curr = append(commandCount);

-    // we extract camera position/forward outside of the loop, because these are not cheap.
-    const float3 cameraPosition(camera.getPosition());
-    const float3 cameraForwardVector(camera.getForwardVector());
+    const float3 cameraPosition(mCameraPosition);
+    const float3 cameraForwardVector(mCameraForwardVector);
    auto work = [commandTypeFlags, curr, &soa, variant, renderFlags, visibilityMask, cameraPosition,
                 cameraForwardVector]
            (uint32_t startIndex, uint32_t indexCount) {
@@ -135,6 +138,15 @@ void RenderPass::appendCommands(CommandTypeFlags const commandTypeFlags) noexcep
    // "eof" command. these commands are guaranteed to be sorted last in the
    // command buffer.
    curr[commandCount - 1].key = uint64_t(Pass::SENTINEL);
+
+    // Go over all the commands and call prepareProgram().
+    // This must be done from the main thread.
+    for (Command const* first = curr, *last = curr + commandCount ; first != last ; ++first) {
+        if (UTILS_LIKELY((first->key & CUSTOM_MASK) == uint64_t(CustomCommand::PASS))) {
+            auto ma = first->primitive.mi->getMaterial();
+            ma->prepareProgram(first->primitive.materialVariant);
+        }
+    }
 }

 void RenderPass::appendCustomCommand(Pass pass, CustomCommand custom, uint32_t order,
@@ -399,7 +411,6 @@ void RenderPass::generateCommandsImpl(uint32_t extraFlags,
            if constexpr (isColorPass) {
                cmdColor.primitive.primitiveHandle = primitive.getHwHandle();
                RenderPass::setupColorCommand(cmdColor, variant, mi, inverseFrontFaces);
-                ma->prepareProgram(cmdColor.primitive.materialVariant);

                cmdColor.primitive.morphWeightBuffer = morphing.handle;
                cmdColor.primitive.morphTargetBuffer = morphTargets.buffer->getHwHandle();
@@ -484,7 +495,6 @@ void RenderPass::generateCommandsImpl(uint32_t extraFlags,
            }

            if constexpr (isDepthPass) {
-                ma->prepareProgram(cmdDepth.primitive.materialVariant);
                const RasterState rs = ma->getRasterState();
                const TransparencyMode mode = mi->getTransparencyMode();
                const BlendingMode blendingMode = ma->getBlendingMode();
@@ -532,7 +542,7 @@ void RenderPass::updateSummedPrimitiveCounts(

 void RenderPass::Executor::execute(const char* name,
        backend::Handle<backend::HwRenderTarget> renderTarget,
-        backend::RenderPassParams params) const noexcept {
+        backend::RenderPassParams const& params) const noexcept {
    FEngine& engine = mEngine;
    DriverApi& driver = engine.getDriverApi();

@@ -272,7 +272,7 @@ public:
            backend::Handle<backend::HwBufferObject> uboHandle) noexcept;

    // specifies camera information (e.g. used for sorting commands)
-    void setCamera(const CameraInfo& camera) noexcept { mCamera = camera; }
+    void setCamera(const CameraInfo& camera) noexcept;

    //  flags controlling how commands are generated
    void setRenderFlags(RenderFlags flags) noexcept { mFlags = flags; }
@@ -333,7 +333,7 @@ public:
        ~Executor() noexcept;
        void execute(const char* name,
                backend::Handle<backend::HwRenderTarget> renderTarget,
-                backend::RenderPassParams params) const noexcept;
+                backend::RenderPassParams const& params) const noexcept;
    };

    // returns a new executor for this pass
@@ -406,7 +406,8 @@ private:
    backend::Handle<backend::HwBufferObject> mUboHandle;

    // info about the camera
-    CameraInfo mCamera;
+    math::float3 mCameraPosition{};
+    math::float3 mCameraForwardVector{};

    // info about the scene features (e.g.: has shadows, lighting, etc...)
    RenderFlags mFlags{};
@@ -0,0 +1,327 @@
+/*
+ * Copyright (C) 2022 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 "RendererUtils.h"
+
+#include "details/Engine.h"
+#include "details/View.h"
+
+#include "fg/FrameGraph.h"
+#include "fg/FrameGraphId.h"
+#include "fg/FrameGraphResources.h"
+
+#include <utils/compiler.h>
+#include <utils/debug.h>
+#include <utils/Panic.h>
+
+namespace filament {
+
+using namespace backend;
+using namespace math;
+
+FrameGraphId<FrameGraphTexture> RendererUtils::colorPass(
+        FrameGraph& fg, const char* name, FEngine& engine, FView const& view,
+        FrameGraphTexture::Descriptor const& colorBufferDesc,
+        ColorPassConfig const& config, PostProcessManager::ColorGradingConfig colorGradingConfig,
+        RenderPass::Executor const& passExecutor) noexcept {
+
+    struct ColorPassData {
+        FrameGraphId<FrameGraphTexture> shadows;
+        FrameGraphId<FrameGraphTexture> color;
+        FrameGraphId<FrameGraphTexture> output;
+        FrameGraphId<FrameGraphTexture> depth;
+        FrameGraphId<FrameGraphTexture> ssao;
+        FrameGraphId<FrameGraphTexture> ssr;    // either screen-space reflections or refractions
+        FrameGraphId<FrameGraphTexture> structure;
+        float4 clearColor{};
+        TargetBufferFlags clearFlags{};
+    };
+
+    Blackboard& blackboard = fg.getBlackboard();
+
+    auto& colorPass = fg.addPass<ColorPassData>(name,
+            [&](FrameGraph::Builder& builder, ColorPassData& data) {
+
+                TargetBufferFlags clearDepthFlags = config.clearFlags & TargetBufferFlags::DEPTH;
+                TargetBufferFlags clearColorFlags = config.clearFlags & TargetBufferFlags::COLOR;
+
+                data.shadows = blackboard.get<FrameGraphTexture>("shadows");
+                data.ssao = blackboard.get<FrameGraphTexture>("ssao");
+                data.color = blackboard.get<FrameGraphTexture>("color");
+                data.depth = blackboard.get<FrameGraphTexture>("depth");
+
+                // Screen-space reflection or refractions
+                data.ssr = blackboard.get<FrameGraphTexture>("ssr");
+                if (data.ssr) {
+                    data.ssr = builder.sample(data.ssr);
+                }
+
+                if (config.hasContactShadows) {
+                    data.structure = blackboard.get<FrameGraphTexture>("structure");
+                    assert_invariant(data.structure);
+                    data.structure = builder.sample(data.structure);
+                }
+
+                if (data.shadows) {
+                    data.shadows = builder.sample(data.shadows);
+                }
+
+                if (data.ssao) {
+                    data.ssao = builder.sample(data.ssao);
+                }
+
+                if (!data.color) {
+                    // FIXME: this works only when the viewport is full
+                    //  if (view.isSkyboxVisible()) {
+                    //      // if the skybox is visible, then we don't need to clear at all
+                    //      clearColorFlags &= ~TargetBufferFlags::COLOR;
+                    //  }
+                    data.color = builder.createTexture("Color Buffer", colorBufferDesc);
+                }
+
+                const bool canResolveDepth = engine.getDriverApi().isAutoDepthResolveSupported();
+
+                if (!data.depth) {
+                    // clear newly allocated depth buffers, regardless of given clear flags
+                    clearDepthFlags = TargetBufferFlags::DEPTH;
+                    data.depth = builder.createTexture("Depth Buffer", {
+                            .width = colorBufferDesc.width,
+                            .height = colorBufferDesc.height,
+                            // If the color attachment requested MS, we assume this means the MS buffer
+                            // must be kept, and for that reason we allocate the depth buffer with MS
+                            // as well. On the other hand, if the color attachment was allocated without
+                            // MS, no need to allocate the depth buffer with MS, if the RT is MS,
+                            // the tile depth buffer will be MS, but it'll be resolved to single
+                            // sample automatically -- which is what we want.
+                            .samples = canResolveDepth ? colorBufferDesc.samples : uint8_t(config.msaa),
+                            .format = TextureFormat::DEPTH32F,
+                    });
+                }
+
+                if (colorGradingConfig.asSubpass) {
+                    data.output = builder.createTexture("Tonemapped Buffer", {
+                            .width = colorBufferDesc.width,
+                            .height = colorBufferDesc.height,
+                            .format = colorGradingConfig.ldrFormat
+                    });
+                    data.color = builder.read(data.color, FrameGraphTexture::Usage::SUBPASS_INPUT);
+                    data.output = builder.write(data.output, FrameGraphTexture::Usage::COLOR_ATTACHMENT);
+                } else if (colorGradingConfig.customResolve) {
+                    data.color = builder.read(data.color, FrameGraphTexture::Usage::SUBPASS_INPUT);
+                }
+
+                // We set a "read" constraint on these attachments here because we need to preserve them
+                // when the color pass happens in several passes (e.g. with SSR)
+                data.color = builder.read(data.color, FrameGraphTexture::Usage::COLOR_ATTACHMENT);
+                data.depth = builder.read(data.depth, FrameGraphTexture::Usage::DEPTH_ATTACHMENT);
+
+                data.color = builder.write(data.color, FrameGraphTexture::Usage::COLOR_ATTACHMENT);
+                data.depth = builder.write(data.depth, FrameGraphTexture::Usage::DEPTH_ATTACHMENT);
+
+                /*
+                 * There is a bit of magic happening here regarding the viewport used.
+                 * We do not specify the viewport in declareRenderPass() below, so it will be
+                 * deduced automatically to be { 0, 0, w, h }, with w,h the min width/height of
+                 * all the attachments. This has the side effect of moving the viewport to the
+                 * origin and ignore the left/bottom of 'svp'. The attachment sizes are set from
+                 * svp's width/height, however.
+                 * But that's not all! When we're rendering directly into the swap-chain (by way
+                 * of calling forwardResource() later), the effective viewport comes from the
+                 * imported resource (i.e. the swap-chain) and is set to 'vp' which has its
+                 * left/bottom honored -- the view is therefore rendered directly where it should
+                 * be (the imported resource viewport is set to 'vp', see  how 'fgViewRenderTarget'
+                 * is initialized in this file).
+                 */
+                builder.declareRenderPass("Color Pass Target", {
+                        .attachments = { .color = { data.color, data.output }, .depth = data.depth },
+                        .samples = config.msaa,
+                        .clearFlags = clearColorFlags | clearDepthFlags });
+
+                data.clearColor = config.clearColor;
+                data.clearFlags = clearColorFlags | clearDepthFlags;
+
+                blackboard["depth"] = data.depth;
+            },
+            [=, &view, &engine](FrameGraphResources const& resources,
+                    ColorPassData const& data, DriverApi& driver) {
+                auto out = resources.getRenderPassInfo();
+
+                // set samplers and uniforms
+                view.prepareSSAO(data.ssao ?
+                                 resources.getTexture(data.ssao) : engine.getOneTextureArray());
+
+                view.prepareShadowMap();
+
+                // set shadow sampler
+                view.prepareShadow(data.shadows ?
+                                   resources.getTexture(data.shadows) : engine.getOneTextureArray());
+
+                // set structure sampler
+                view.prepareStructure(data.structure ?
+                                      resources.getTexture(data.structure) : engine.getOneTexture());
+
+                // set screen-space reflections and screen-space refractions
+                TextureHandle ssr = data.ssr ?
+                                    resources.getTexture(data.ssr) : engine.getOneTextureArray();
+
+                view.prepareSSR(ssr, config.ssrLodOffset,
+                        view.getScreenSpaceReflectionsOptions());
+
+                // Note: here we can't use data.color's descriptor for the viewport because
+                // the actual viewport might be offset when the target is the swapchain.
+                // However, the width/height should be the same.
+                assert_invariant(
+                        out.params.viewport.width == resources.getDescriptor(data.color).width);
+                assert_invariant(
+                        out.params.viewport.height == resources.getDescriptor(data.color).height);
+
+                view.prepareViewport(static_cast<filament::Viewport&>(out.params.viewport),
+                        config.xoffset, config.yoffset);
+                view.commitUniforms(driver);
+
+                out.params.clearColor = data.clearColor;
+                out.params.flags.clear = data.clearFlags;
+                if (view.getBlendMode() == BlendMode::TRANSLUCENT) {
+                    if (any(out.params.flags.discardStart & TargetBufferFlags::COLOR0)) {
+                        // if the buffer is discarded (e.g. it's new) and we're blending,
+                        // then clear it to transparent
+                        out.params.flags.clear |= TargetBufferFlags::COLOR;
+                        out.params.clearColor = {};
+                    }
+                }
+
+                if (colorGradingConfig.asSubpass || colorGradingConfig.customResolve) {
+                    out.params.subpassMask = 1;
+                }
+                passExecutor.execute(resources.getPassName(), out.target, out.params);
+
+                // color pass is typically heavy, and we don't have much CPU work left after
+                // this point, so flushing now allows us to start the GPU earlier and reduce
+                // latency, without creating bubbles.
+                driver.flush();
+            }
+    );
+
+    // when color grading is done as a subpass, the output of the color-pass is the ldr buffer
+    auto output = colorGradingConfig.asSubpass ? colorPass->output : colorPass->color;
+
+    blackboard["color"] = output;
+    return output;
+}
+
+FrameGraphId<FrameGraphTexture> RendererUtils::refractionPass(
+        FrameGraph& fg, FEngine& engine, FView const& view,
+        ColorPassConfig config,
+        PostProcessManager::ScreenSpaceRefConfig const& ssrConfig,
+        PostProcessManager::ColorGradingConfig colorGradingConfig,
+        RenderPass const& pass) noexcept {
+
+    auto& blackboard = fg.getBlackboard();
+    auto input = blackboard.get<FrameGraphTexture>("color");
+    FrameGraphId<FrameGraphTexture> output;
+
+    // find the first refractive object
+    RenderPass::Command const* const refraction = std::partition_point(pass.begin(), pass.end(),
+            [](auto const& command) {
+                return (command.key & RenderPass::PASS_MASK) < uint64_t(RenderPass::Pass::REFRACT);
+            });
+
+    const bool hasScreenSpaceRefraction =
+            (refraction->key & RenderPass::PASS_MASK) == uint64_t(RenderPass::Pass::REFRACT);
+
+    // if there wasn't any refractive object, just skip everything below.
+    if (UTILS_UNLIKELY(hasScreenSpaceRefraction)) {
+        PostProcessManager& ppm = engine.getPostProcessManager();
+
+        // clear the color/depth buffers, which will orphan (and cull) the color pass
+        input.clear();
+        blackboard.remove("color");
+        blackboard.remove("depth");
+
+        input = RendererUtils::colorPass(fg, "Color Pass (opaque)", engine, view, {
+                        // When rendering the opaques, we need to conserve the sample buffer,
+                        // so create config that specifies the sample count.
+                        .width = config.width,
+                        .height = config.height,
+                        .samples = config.msaa,
+                        .format = config.hdrFormat
+                }, config, { .asSubpass = false },
+                pass.getExecutor(pass.begin(), refraction));
+
+        // generate the mipmap chain
+        PostProcessManager::generateMipmapSSR(ppm, fg,
+                input, ssrConfig.refraction, true, ssrConfig);
+
+        // Now we're doing the refraction pass proper.
+        // This uses the same framebuffer (color and depth) used by the opaque pass. This happens
+        // automatically because these are set in the Blackboard (they were set by the opaque
+        // pass). For this reason, `desc` below is only used in colorPass() for the width and
+        // height.
+        config.clearFlags = TargetBufferFlags::NONE;
+        output = RendererUtils::colorPass(fg, "Color Pass (transparent)", engine, view,
+                { .width = config.width, .height = config.height },
+                config, colorGradingConfig, pass.getExecutor(refraction, pass.end()));
+
+        if (config.msaa > 1 && !colorGradingConfig.asSubpass) {
+            // We need to do a resolve here because later passes (such as color grading or DoF) will need
+            // to sample from 'output'. However, because we have MSAA, we know we're not sampleable.
+            // And this is because in the SSR case, we had to use a renderbuffer to conserve the
+            // multi-sample buffer.
+            output = ppm.resolveBaseLevel(fg, "Resolved Color Buffer", output);
+        }
+    } else {
+        output = input;
+    }
+    return output;
+}
+
+UTILS_NOINLINE
+void RendererUtils::readPixels(backend::DriverApi& driver, Handle<HwRenderTarget> renderTargetHandle,
+        uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
+        backend::PixelBufferDescriptor&& buffer) {
+    if (!ASSERT_POSTCONDITION_NON_FATAL(
+            buffer.type != PixelDataType::COMPRESSED,
+            "buffer.format cannot be COMPRESSED")) {
+        return;
+    }
+
+    if (!ASSERT_POSTCONDITION_NON_FATAL(
+            buffer.alignment > 0 && buffer.alignment <= 8 &&
+            !(buffer.alignment & (buffer.alignment - 1u)),
+            "buffer.alignment must be 1, 2, 4 or 8")) {
+        return;
+    }
+
+    // It's not really possible to know here which formats will be supported because
+    // it can vary depending on the RenderTarget, in GL the following are ALWAYS supported though:
+    // format: RGBA, RGBA_INTEGER
+    // type: UBYTE, UINT, INT, FLOAT
+
+    const size_t sizeNeeded = PixelBufferDescriptor::computeDataSize(
+            buffer.format, buffer.type,
+            buffer.stride ? buffer.stride : width,
+            buffer.top + height,
+            buffer.alignment);
+
+    if (!ASSERT_POSTCONDITION_NON_FATAL(buffer.size >= sizeNeeded,
+            "Pixel buffer too small: has %u bytes, needs %u bytes", buffer.size, sizeNeeded)) {
+        return;
+    }
+
+    driver.readPixels(renderTargetHandle, xoffset, yoffset, width, height, std::move(buffer));
+}
+
+} // namespace filament
@@ -0,0 +1,89 @@
+/*
+ * Copyright (C) 2022 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef TNT_FILAMENT_DETAILS_RENDERERUTILS_H
+#define TNT_FILAMENT_DETAILS_RENDERERUTILS_H
+
+#include "PostProcessManager.h"
+#include "RenderPass.h"
+
+#include "fg/FrameGraphId.h"
+
+#include <backend/DriverEnums.h>
+#include <backend/PixelBufferDescriptor.h>
+
+#include <math/vec4.h>
+
+#include <stdint.h>
+
+namespace filament {
+
+class FRenderTarget;
+class FrameGraph;
+class FrameGraph;
+class FView;
+
+class RendererUtils {
+public:
+
+    struct ColorPassConfig {
+        // Rendering viewport width (e.g. scaled down viewport from dynamic resolution)
+        uint32_t width;
+        // Rendering viewport height (e.g. scaled down viewport from dynamic resolution)
+        uint32_t height;
+        // Rendering offset within the viewport (e.g. non-zero when we have guard bands)
+        uint32_t xoffset;
+        // Rendering offset within the viewport (e.g. non-zero when we have guard bands)
+        uint32_t yoffset;
+        // dynamic resolution scale
+        math::float2 scale;
+        // HDR format
+        backend::TextureFormat hdrFormat;
+        // MSAA sample count
+        uint8_t msaa;
+        // Clear flags
+        backend::TargetBufferFlags clearFlags;
+        // Clear color
+        math::float4 clearColor = {};
+        // Lod offset for the SSR passes
+        float ssrLodOffset;
+        // Contact shadow enabled?
+        bool hasContactShadows;
+    };
+
+    static FrameGraphId<FrameGraphTexture> colorPass(
+            FrameGraph& fg, const char* name, FEngine& engine, FView const& view,
+            FrameGraphTexture::Descriptor const& colorBufferDesc,
+            ColorPassConfig const& config,
+            PostProcessManager::ColorGradingConfig colorGradingConfig,
+            RenderPass::Executor const& passExecutor) noexcept;
+
+    static FrameGraphId<FrameGraphTexture> refractionPass(
+            FrameGraph& fg, FEngine& engine, FView const& view,
+            ColorPassConfig config,
+            PostProcessManager::ScreenSpaceRefConfig const& ssrConfig,
+            PostProcessManager::ColorGradingConfig colorGradingConfig,
+            RenderPass const& pass) noexcept;
+
+    static void readPixels(backend::DriverApi& driver,
+            backend::Handle<backend::HwRenderTarget> renderTargetHandle,
+            uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
+            backend::PixelBufferDescriptor&& buffer);
+};
+
+} // namespace filament
+
+#endif // TNT_FILAMENT_DETAILS_RENDERERUTILS_H
@@ -60,6 +60,9 @@ ShadowMap::ShadowMap(FEngine& engine) noexcept :
    }
 }

+void ShadowMap::terminate(FEngine& engine) {
+}
+
 ShadowMap::~ShadowMap() {
    FEngine& engine = mEngine;
    Entity entities[] = { mCamera->getEntity(), mDebugCamera->getEntity() };
@@ -70,9 +73,7 @@ ShadowMap::~ShadowMap() {
 }

 void ShadowMap::render(FScene const& scene, utils::Range<uint32_t> range,
-        FScene::VisibleMaskType visibilityMask, filament::CameraInfo const& cameraInfo,
-        RenderPass* const pass) noexcept {
-    pass->setCamera(cameraInfo);
+        FScene::VisibleMaskType visibilityMask, RenderPass* const pass) noexcept {
    pass->setVisibilityMask(visibilityMask);
    pass->setGeometry(scene.getRenderableData(), range, scene.getRenderableUBO());
    pass->overridePolygonOffset(&mShadowMapInfo.polygonOffset);
@@ -320,7 +321,7 @@ void ShadowMap::updateDirectional(const FScene::LightSoa& lightData, size_t inde

        if (params.options.stable) {
            // Use the world origin as reference point, fixed w.r.t. the camera
-            snapLightFrustum(s, o, Mv, camera.worldOrigin[3].xyz,
+            snapLightFrustum(s, o, Mv, -camera.getWorldOffset(),
                    1.0f / mShadowMapInfo.shadowDimension);
        }

@@ -1048,13 +1049,12 @@ void ShadowMap::visitScene(const FScene& scene, uint32_t visibleLayers,
    }
 }

-void ShadowMap::initSceneInfo(FScene const& scene, filament::CameraInfo const& camera,
+void ShadowMap::initSceneInfo(FScene const& scene, mat4f const& viewMatrix,
        ShadowMap::SceneInfo& sceneInfo) {
    sceneInfo.vsNearFar = { std::numeric_limits<float>::lowest(), std::numeric_limits<float>::max() };

    // We assume the light is at the origin to compute the SceneInfo. This is consumed later by
    // computeShadowCameraDirectional() which takes this into account.
-    const mat4f V = camera.view;

    // Compute scene bounds in world space, as well as the light-space and view-space near/far planes
    sceneInfo.wsShadowCastersVolume = {};
@@ -1071,7 +1071,7 @@ void ShadowMap::initSceneInfo(FScene const& scene, filament::CameraInfo const& c
                        min(sceneInfo.wsShadowReceiversVolume.min, receiver.min);
                sceneInfo.wsShadowReceiversVolume.max =
                        max(sceneInfo.wsShadowReceiversVolume.max, receiver.max);
-                float2 nf = ShadowMap::computeNearFar(V, receiver);
+                float2 nf = ShadowMap::computeNearFar(viewMatrix, receiver);
                sceneInfo.vsNearFar.x = std::max(sceneInfo.vsNearFar.x, nf.x);
                sceneInfo.vsNearFar.y = std::min(sceneInfo.vsNearFar.y, nf.y);
            }
@@ -43,6 +43,8 @@ public:

    ~ShadowMap();

+    void terminate(FEngine& engine);
+
    struct ShadowMapInfo {
        // the dimension of the encompassing texture atlas
        uint16_t atlasDimension = 0;
@@ -105,8 +107,7 @@ public:
            SceneInfo& sceneInfo) noexcept;

    void render(FScene const& scene, utils::Range<uint32_t> range,
-            FScene::VisibleMaskType visibilityMask, filament::CameraInfo const& cameraInfo,
-            RenderPass* pass) noexcept;
+            FScene::VisibleMaskType visibilityMask, RenderPass* const pass) noexcept;

    // Do we have visible shadows. Valid after calling update().
    bool hasVisibleShadows() const noexcept { return mHasVisibleShadows; }
@@ -129,7 +130,7 @@ public:

    // Call once per frame to populate the SceneInfo struct, then pass to update().
    // This computes values constant across all shadow maps.
-    static void initSceneInfo(FScene const& scene, filament::CameraInfo const& camera,
+    static void initSceneInfo(FScene const& scene, math::mat4f const& viewMatrix,
            ShadowMap::SceneInfo& sceneInfo);

    // Update SceneInfo struct for a given light
@@ -40,6 +40,10 @@ ShadowMapManager::ShadowMapManager(FEngine& engine) { // NOLINT(cppcoreguideline
    for (auto& entry : mShadowMapCache) {
        new (&entry) ShadowMap(engine);
    }
+
+    mShadowUbh = engine.getDriverApi().createBufferObject(mShadowUb.getSize(),
+            BufferObjectBinding::UNIFORM, BufferUsage::DYNAMIC);
+
    FDebugRegistry& debugRegistry = engine.getDebugRegistry();
    debugRegistry.registerProperty("d.shadowmap.visualize_cascades",
            &engine.debug.shadowmap.visualize_cascades);
@@ -47,18 +51,26 @@ ShadowMapManager::ShadowMapManager(FEngine& engine) { // NOLINT(cppcoreguideline
            &engine.debug.shadowmap.tightly_bound_scene);
 }

+void ShadowMapManager::terminate(FEngine& engine) {
+    DriverApi& driver = engine.getDriverApi();
+    driver.destroyBufferObject(mShadowUbh);
+    UTILS_NOUNROLL
+    for (auto& entry : mShadowMapCache) {
+        std::launder(reinterpret_cast<ShadowMap*>(&entry))->terminate(engine);
+    }
+}
+
 ShadowMapManager::~ShadowMapManager() {
    // destroy the ShadowMap array in-place
    UTILS_NOUNROLL
    for (auto& entry : mShadowMapCache) {
-        std::launder(reinterpret_cast<ShadowMap*>(&entry))->~ShadowMap();
+        std::destroy_at(std::launder(reinterpret_cast<ShadowMap*>(&entry)));
    }
 }

-ShadowMapManager::ShadowTechnique ShadowMapManager::update(
-        FEngine& engine, FView& view,
-        TypedUniformBuffer<ShadowUib>& shadowUb, FScene::RenderableSoa& renderableData,
-        FScene::LightSoa& lightData) noexcept {
+ShadowMapManager::ShadowTechnique ShadowMapManager::update(FEngine& engine, FView& view,
+        CameraInfo const& cameraInfo,
+        FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData) noexcept {
    ShadowTechnique shadowTechnique = {};

    calculateTextureRequirements(engine, view, lightData);
@@ -66,15 +78,19 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::update(
    ShadowMap::SceneInfo sceneInfo(view.getVisibleLayers());

    // Compute scene-dependent values shared across all shadow maps
-    ShadowMap::initSceneInfo(
-            *view.getScene(), view.getCameraInfo(),
-            sceneInfo);
+    ShadowMap::initSceneInfo(*view.getScene(), cameraInfo.view, sceneInfo);

    shadowTechnique |= updateCascadeShadowMaps(
-            engine, view, renderableData, lightData, sceneInfo);
+            engine, view, cameraInfo, renderableData, lightData, sceneInfo);

    shadowTechnique |= updateSpotShadowMaps(
-            engine, view, renderableData, lightData, sceneInfo, shadowUb);
+            engine, view, cameraInfo, renderableData, lightData, sceneInfo);
+
+    if (mShadowUb.isDirty()) {
+        DriverApi& driver = engine.getDriverApi();
+        driver.updateBufferObject(mShadowUbh,
+                mShadowUb.toBufferDescriptor(driver), 0);
+    }

    return shadowTechnique;
 }
@@ -169,8 +185,10 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FrameGraph& fg,
                        .format = view.hasVSM() ? vsmTextureFormat : mTextureFormat
                });
            },
-            [=](FrameGraphResources const& resources, auto const& data, DriverApi& driver) { });
-
+            [&view](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {
+                // set uniforms needed to render this ShadowMap
+                view.prepareShadowMap();
+            });

    // -------------------------------------------------------------------------------------------

@@ -180,12 +198,11 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FrameGraph& fg,

    struct ShadowPassData {
        FrameGraphId<FrameGraphTexture> tempBlurSrc{};  // temporary shadowmap when blurring
+        FrameGraphId<FrameGraphTexture> output;
        uint32_t blurRt{};
        uint32_t shadowRt{};
    };

-    auto shadows = prepareShadowPass.getData().shadows;
-
    auto& ppm = engine.getPostProcessManager();

    for (auto const& entry : passList) {
@@ -198,7 +215,7 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FrameGraph& fg,

                    FrameGraphRenderPass::Descriptor renderTargetDesc{};

-                    auto attachment = builder.createSubresource(prepareShadowPass->shadows,
+                    data.output = builder.createSubresource(prepareShadowPass->shadows,
                            "Shadowmap Layer", { .layer = layer });

                    if (view.hasVSM()) {
@@ -223,10 +240,10 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FrameGraph& fg,
                        depth = builder.write(depth,
                                FrameGraphTexture::Usage::DEPTH_ATTACHMENT);

-                        attachment = builder.write(attachment,
+                        data.output = builder.write(data.output,
                                FrameGraphTexture::Usage::COLOR_ATTACHMENT);

-                        renderTargetDesc.attachments.color[0] = attachment;
+                        renderTargetDesc.attachments.color[0] = data.output;
                        renderTargetDesc.attachments.depth = depth;
                        renderTargetDesc.clearFlags =
                                TargetBufferFlags::COLOR | TargetBufferFlags::DEPTH;
@@ -250,9 +267,9 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FrameGraph& fg,
                        }
                    } else {
                        // the shadowmap layer
-                        attachment = builder.write(attachment,
+                        data.output = builder.write(data.output,
                                FrameGraphTexture::Usage::DEPTH_ATTACHMENT);
-                        renderTargetDesc.attachments.depth = attachment;
+                        renderTargetDesc.attachments.depth = data.output;
                        renderTargetDesc.clearFlags = TargetBufferFlags::DEPTH;
                    }

@@ -270,7 +287,7 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FrameGraph& fg,

                    // generate and sort the commands for rendering the shadow map
                    RenderPass entryPass(pass);
-                    shadowMap.render(*scene, entry.range, entry.visibilityMask, cameraInfo, &entryPass);
+                    shadowMap.render(*scene, entry.range, entry.visibilityMask, &entryPass);

                    const auto& executor = entryPass.getExecutor();
                    const bool blur = view.hasVSM() && options->vsm.blurWidth > 0.0f;
@@ -289,12 +306,7 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FrameGraph& fg,
                    // than clearing.
                    const uint32_t dim = options->mapSize;
                    filament::Viewport viewport{ 1, 1, dim - 2, dim - 2 };
-                    view.prepareViewport(viewport);
-
-                    // set uniforms needed to render this ShadowMap
-                    // Currently these uniforms are owned by View and are global, but eventually
-                    // this will set a separate per shadowmap UBO
-                    view.prepareShadowMap();
+                    view.prepareViewport(viewport, 0, 0);

                    view.commitUniforms(driver);

@@ -317,7 +329,7 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FrameGraph& fg,
                const float ratio = float(kernelWidth + 1) / sigma;
                ppm.gaussianBlurPass(fg,
                        shadowPass->tempBlurSrc,
-                        shadows,
+                        shadowPass->output,
                        false, kernelWidth, ratio);
            }

@@ -327,21 +339,20 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FrameGraph& fg,
            if (textureRequirements.levels > 1) {
                for (size_t level = 0; level < textureRequirements.levels - 1; level++) {
                    const bool finalize = level == textureRequirements.levels - 2;
-                    shadows = ppm.vsmMipmapPass(fg, shadows, layer, level,
+                    ppm.vsmMipmapPass(fg, prepareShadowPass->shadows, layer, level,
                            vsmClearColor, finalize);
                }
            }
        }
    }

-    return shadows;
+    return prepareShadowPass->shadows;
 }

 ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEngine& engine,
-        FView& view, FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData,
-        ShadowMap::SceneInfo& sceneInfo) noexcept {
+        FView& view, CameraInfo const& cameraInfo, FScene::RenderableSoa& renderableData,
+        FScene::LightSoa& lightData, ShadowMap::SceneInfo& sceneInfo) noexcept {
    FScene* scene = view.getScene();
-    const CameraInfo& viewingCameraInfo = view.getCameraInfo();
    auto& lcm = engine.getLightManager();

    FLightManager::Instance directionalLight = lightData.elementAt<FScene::LIGHT_INSTANCE>(0);
@@ -365,7 +376,7 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
        ShadowMapEntry& entry = mCascadeShadowMaps[0];
        ShadowMap& shadowMap = entry.getShadowMap();

-        shadowMap.updateDirectional(lightData, 0, viewingCameraInfo, shadowMapInfo, *scene, sceneInfo);
+        shadowMap.updateDirectional(lightData, 0, cameraInfo, shadowMapInfo, *scene, sceneInfo);

        Frustum const& frustum = shadowMap.getCamera().getCullingFrustum();
        FView::cullRenderables(engine.getJobSystem(), renderableData, frustum,
@@ -384,8 +395,8 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
    }

    // Adjust the near and far planes to tightly bound the scene.
-    float vsNear = -viewingCameraInfo.zn;
-    float vsFar = -viewingCameraInfo.zf;
+    float vsNear = -cameraInfo.zn;
+    float vsFar = -cameraInfo.zf;
    if (engine.debug.shadowmap.tightly_bound_scene) {
        vsNear = std::min(vsNear, sceneInfo.vsNearFar.x);
        vsFar = std::max(vsFar, sceneInfo.vsNearFar.y);
@@ -402,7 +413,7 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
    }

    const CascadeSplits::Params p {
-        .proj = viewingCameraInfo.cullingProjection,
+        .proj = cameraInfo.cullingProjection,
        .near = vsNear,
        .far = vsFar,
        .cascadeCount = cascadeCount,
@@ -440,7 +451,7 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
        sceneInfo.csNearFar = { csSplitPosition[i], csSplitPosition[i + 1] };

        shadowMap.updateDirectional(lightData, 0,
-                viewingCameraInfo, shadowMapInfo,
+                cameraInfo, shadowMapInfo,
                *scene, sceneInfo);

        if (shadowMap.hasVisibleShadows()) {
@@ -479,11 +490,10 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateCascadeShadowMaps(FEng
 }

 ShadowMapManager::ShadowTechnique ShadowMapManager::updateSpotShadowMaps(FEngine& engine,
-        FView& view, FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData,
-        ShadowMap::SceneInfo& sceneInfo, TypedUniformBuffer<ShadowUib>& shadowUb) noexcept {
+        FView& view, CameraInfo const& cameraInfo, FScene::RenderableSoa& renderableData,
+        FScene::LightSoa& lightData, ShadowMap::SceneInfo& sceneInfo) noexcept {

    auto& lcm = engine.getLightManager();
-    const CameraInfo& viewingCameraInfo = view.getCameraInfo();

    // shadow-map shadows for point/spotlights
    ShadowTechnique shadowTechnique{};
@@ -529,7 +539,7 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateSpotShadowMaps(FEngine
                VISIBLE_SPOT_SHADOW_RENDERABLE_N_BIT(i));

        shadowMap.updateSpot(lightData, lightIndex,
-                viewingCameraInfo, shadowMapInfo,
+                cameraInfo, shadowMapInfo,
                *view.getScene(), sceneInfo);

        if (shadowMap.hasVisibleShadows()) {
@@ -541,7 +551,7 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateSpotShadowMaps(FEngine
            // note: normalBias is set to zero for VSM
            const float normalBias = shadowMapInfo.vsm ? 0.0f : options->normalBias;

-            auto& s = shadowUb.edit();
+            auto& s = mShadowUb.edit();
            const float n = shadowMap.getCamera().getNear();
            const float f = shadowMap.getCamera().getCullingFar();
            s.shadows[i].lightFromWorldMatrix = shadowMap.getLightSpaceMatrix();
@@ -74,6 +74,8 @@ public:
    explicit ShadowMapManager(FEngine& engine);
    ~ShadowMapManager();

+    void terminate(FEngine& engine);
+
    // Reset shadow map layout.
    void reset() noexcept;

@@ -82,8 +84,8 @@ public:

    // Updates all of the shadow maps and performs culling.
    // Returns true if any of the shadow maps have visible shadows.
-    ShadowTechnique update(FEngine& engine, FView& view,
-            TypedUniformBuffer<ShadowUib>& shadowUb,
+    ShadowMapManager::ShadowTechnique update(FEngine& engine, FView& view,
+            CameraInfo const& cameraInfo,
            FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData) noexcept;

    // Renders all the shadow maps.
@@ -101,7 +103,8 @@ public:

    ShadowMap* getSpotShadowMap(size_t spot) noexcept {
        assert_invariant(spot < CONFIG_MAX_SHADOW_CASTING_SPOTS);
-        return std::launder(reinterpret_cast<ShadowMap*>(&mShadowMapCache[CONFIG_MAX_SHADOW_CASCADES + spot]));
+        return std::launder(reinterpret_cast<ShadowMap*>(
+                &mShadowMapCache[CONFIG_MAX_SHADOW_CASCADES + spot]));
    }

    ShadowMap const* getSpotShadowMap(size_t spot) const noexcept {
@@ -113,14 +116,18 @@ public:
        return mShadowMappingUniforms;
    }

-private:
-    ShadowTechnique updateCascadeShadowMaps(FEngine& engine,
-            FView& view, FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData,
-            ShadowMap::SceneInfo& sceneInfo) noexcept;
+    auto& getShadowUniforms() const { return mShadowUb; }

-    ShadowTechnique updateSpotShadowMaps(FEngine& engine,
-            FView& view, FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData,
-            ShadowMap::SceneInfo& sceneInfo, TypedUniformBuffer<ShadowUib>& shadowUb) noexcept;
+    auto& getShadowUniformsHandle() const { return mShadowUbh; }
+
+private:
+    ShadowMapManager::ShadowTechnique updateCascadeShadowMaps(FEngine& engine,
+            FView& view, CameraInfo const& cameraInfo, FScene::RenderableSoa& renderableData,
+            FScene::LightSoa& lightData, ShadowMap::SceneInfo& sceneInfo) noexcept;
+
+    ShadowMapManager::ShadowTechnique updateSpotShadowMaps(FEngine& engine,
+            FView& view, CameraInfo const& cameraInfo, FScene::RenderableSoa& renderableData,
+            FScene::LightSoa& lightData, ShadowMap::SceneInfo& sceneInfo) noexcept;

    void calculateTextureRequirements(FEngine& engine, FView& view, FScene::LightSoa& lightData) noexcept;

@@ -205,6 +212,9 @@ private:
    // TODO: iOS does not support the DEPTH16 texture format.
    backend::TextureFormat mTextureFormat = backend::TextureFormat::DEPTH16;

+    mutable TypedUniformBuffer<ShadowUib> mShadowUb;
+    backend::Handle<backend::HwBufferObject> mShadowUbh;
+
    ShadowMappingUniforms mShadowMappingUniforms;

    utils::FixedCapacityVector<ShadowMapEntry> mCascadeShadowMaps{
@@ -36,11 +36,6 @@ void Stream::setDimensions(uint32_t width, uint32_t height) noexcept {
    upcast(this)->setDimensions(width, height);
 }

-void Stream::readPixels(uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
-        backend::PixelBufferDescriptor&& buffer) noexcept {
-    upcast(this)->readPixels(xoffset, yoffset, width, height, std::move(buffer));
-}
-
 int64_t Stream::getTimestamp() const noexcept {
    return upcast(this)->getTimestamp();
 }
@@ -156,8 +156,8 @@ math::mat4 FCamera::getProjectionMatrix() const noexcept {
    // Note that this math ends up setting the projection matrix' p33 to 0, which is where we're
    // getting back a lot of precision in the depth buffer.
    const mat4 m{ mat4::row_major_init{
-            mScaling.x, 0.0, 0.0, mShiftCS.x,
-            0.0, mScaling.y, 0.0, mShiftCS.y,
+            mScalingCS.x, 0.0, 0.0, mShiftCS.x,
+            0.0, mScalingCS.y, 0.0, mShiftCS.y,
            0.0, 0.0, -0.5, 0.5,    // GL to inverted DX convention
            0.0, 0.0, 0.0, 1.0
    }};
@@ -167,8 +167,8 @@ math::mat4 FCamera::getProjectionMatrix() const noexcept {
 math::mat4 FCamera::getCullingProjectionMatrix() const noexcept {
    // The culling projection matrix stays in the GL convention
    const mat4 m{ mat4::row_major_init{
-            mScaling.x, 0.0, 0.0, mShiftCS.x,
-            0.0, mScaling.y, 0.0, mShiftCS.y,
+            mScalingCS.x, 0.0, 0.0, mShiftCS.x,
+            0.0, mScalingCS.y, 0.0, mShiftCS.y,
            0.0, 0.0, 1.0, 0.0,
            0.0, 0.0, 0.0, 1.0
    }};
@@ -275,7 +275,7 @@ CameraInfo::CameraInfo(FCamera const& camera) noexcept {
    zn                 = camera.getNear();
    zf                 = camera.getCullingFar();
    ev100              = Exposure::ev100(camera);
-    f                  = camera.getFocalLength();
+    f                  = (float)camera.getFocalLength();
    A                  = f / camera.getAperture();
    d                  = std::max(zn, camera.getFocusDistance());
 }
@@ -286,14 +286,13 @@ CameraInfo::CameraInfo(FCamera const& camera, const math::mat4& worldOriginCamer
    cullingProjection  = mat4f{ camera.getCullingProjectionMatrix() };
    model              = mat4f{ modelMatrix };
    view               = mat4f{ inverse(modelMatrix) };
+    worldOrigin        = worldOriginCamera;
    zn                 = camera.getNear();
    zf                 = camera.getCullingFar();
    ev100              = Exposure::ev100(camera);
-    f                  = camera.getFocalLength();
+    f                  = (float)camera.getFocalLength();
    A                  = f / camera.getAperture();
    d                  = std::max(zn, camera.getFocusDistance());
-    worldOffset        = camera.getPosition();
-    worldOrigin        = mat4f{ worldOriginCamera };
 }

 } // namespace filament
@@ -63,9 +63,9 @@ public:
    void setCustomProjection(math::mat4 const& projection,
            math::mat4 const& projectionForCulling, double near, double far) noexcept;

-    void setScaling(math::double2 scaling) noexcept { mScaling = scaling; }
+    void setScaling(math::double2 scaling) noexcept { mScalingCS = scaling; }

-    math::double4 getScaling() const noexcept { return math::double4{ mScaling, 1.0, 1.0 }; }
+    math::double4 getScaling() const noexcept { return math::double4{ mScalingCS, 1.0, 1.0 }; }

    void setShift(math::double2 shift) noexcept { mShiftCS = shift * 2.0; }

@@ -189,8 +189,8 @@ private:

    math::mat4 mProjection;                // projection matrix (infinite far)
    math::mat4 mProjectionForCulling;      // projection matrix (with far plane)
-    math::double2 mScaling = { 1.0f };  // additional scaling applied to projection
-    math::double2 mShiftCS = { 0.0f };  // additional translation applied to projection
+    math::double2 mScalingCS = { 1.0 };    // additional scaling applied to projection
+    math::double2 mShiftCS = { 0.0 };      // additional translation applied to projection

    float mNear{};
    float mFar{};
@@ -209,19 +209,18 @@ struct CameraInfo {
    math::mat4f projection;         // projection matrix for drawing (infinite zfar)
    math::mat4f cullingProjection;  // projection matrix for culling
    math::mat4f model;              // camera model matrix
-    math::mat4f view;               // camera view matrix
+    math::mat4f view;               // camera view matrix (inverse(model))
+    math::mat4 worldOrigin;         // world origin transform (already applied to model and view)
    float zn{};                     // distance (positive) to the near plane
    float zf{};                     // distance (positive) to the far plane
    float ev100{};                  // exposure
    float f{};                      // focal length [m]
    float A{};                      // f-number or f / aperture diameter [m]
    float d{};                      // focus distance [m]
-    math::float3 worldOffset{};     // world offset, API-level camera position
    math::float3 const& getPosition() const noexcept { return model[3].xyz; }
    math::float3 getForwardVector() const noexcept { return normalize(-model[2].xyz); }
-
-    // for debugging:
-    math::mat4f worldOrigin; // this is already applied to model and view
+    math::float3 getWorldOffset() const noexcept { return -worldOrigin[3].xyz; }
+    math::mat4 getUserViewMatrix() const noexcept { return view * worldOrigin; }
 };

 FILAMENT_UPCAST(Camera)
@@ -18,8 +18,6 @@

 #include "MaterialParser.h"
 #include "ResourceAllocator.h"
-
-#include "DFG.h"
 #include "RenderPrimitive.h"

 #include "details/BufferObject.h"
@@ -82,6 +80,7 @@ FEngine* FEngine::create(Backend backend, Platform* platform, void* sharedGLCont
        }
        if (platform == nullptr) {
            slog.e << "Selected backend not supported in this build." << io::endl;
+            delete instance;
            return nullptr;
        }
        instance->mDriver = platform->createDriver(sharedGLContext);
@@ -95,6 +94,7 @@ FEngine* FEngine::create(Backend backend, Platform* platform, void* sharedGLCont
        if (UTILS_UNLIKELY(!instance->mDriver)) {
            // something went horribly wrong during driver initialization
            instance->mDriverThread.join();
+            delete instance;
            return nullptr;
        }
    }
@@ -142,6 +142,7 @@ FEngine* FEngine::getEngine(void* token) {
        if (UTILS_UNLIKELY(!instance->mDriver)) {
            // something went horribly wrong during driver initialization
            instance->mDriverThread.join();
+            delete instance;
            return nullptr;
        }

@@ -333,19 +334,11 @@ void FEngine::init() {

    mPostProcessManager.init();
    mLightManager.init(*this);
-    mDFG = std::make_unique<DFG>(*this);
+    mDFG.init(*this);
 }

 FEngine::~FEngine() noexcept {
    SYSTRACE_CALL();
-
-    ASSERT_DESTRUCTOR(mTerminated, "Engine destroyed but not terminated!");
-
-    // we use mResourceAllocator as a proxy for knowing if init() was called
-    if (mResourceAllocator) {
-        std::destroy_at(std::launder(reinterpret_cast<DriverApi*>(&mDriverApiStorage)));
-    }
-
    delete mResourceAllocator;
    delete mDriver;
    if (mOwnPlatform) {
@@ -356,6 +349,9 @@ FEngine::~FEngine() noexcept {
 void FEngine::shutdown() {
    SYSTRACE_CALL();

+    // by construction this should never be nullptr
+    assert_invariant(mResourceAllocator);
+
    ASSERT_PRECONDITION(ThreadUtils::isThisThread(mMainThreadId),
            "Engine::shutdown() called from the wrong thread!");

@@ -375,7 +371,7 @@ void FEngine::shutdown() {

    mPostProcessManager.terminate(driver);  // free-up post-process manager resources
    mResourceAllocator->terminate();
-    mDFG->terminate();                      // free-up the DFG
+    mDFG.terminate(*this);                  // free-up the DFG
    mRenderableManager.terminate();         // free-up all renderables
    mLightManager.terminate();              // free-up all lights
    mCameraManager.terminate();             // free-up all cameras
@@ -452,14 +448,15 @@ void FEngine::shutdown() {
    // These callbacks CANNOT call driver APIs.
    getDriver().purge();

+    // and destroy the CommandStream
+    std::destroy_at(std::launder(reinterpret_cast<DriverApi*>(&mDriverApiStorage)));
+
    /*
     * Terminate the JobSystem...
     */

-    // detach this thread from the jobsystem
+    // detach this thread from the JobSystem
    mJobSystem.emancipate();
-
-    mTerminated = true;
 }

 void FEngine::prepare() {
@@ -698,7 +695,6 @@ FRenderer* FEngine::createRenderer() noexcept {
    FRenderer* p = mHeapAllocator.make<FRenderer>(*this);
    if (p) {
        mRenderers.insert(p);
-        p->init();
    }
    return p;
 }
@@ -20,6 +20,7 @@
 #include "upcast.h"

 #include "Allocators.h"
+#include "DFG.h"
 #include "PostProcessManager.h"
 #include "ResourceList.h"

@@ -97,7 +98,6 @@ class FScene;
 class FSwapChain;
 class FView;

-class DFG;
 class ResourceAllocator;

 /*
@@ -153,7 +153,7 @@ public:
        return *std::launder(reinterpret_cast<DriverApi*>(&mDriverApiStorage));
    }

-    DFG* getDFG() const noexcept { return mDFG.get(); }
+    DFG const& getDFG() const noexcept { return mDFG; }

    // the per-frame Area is used by all Renderer, so they must run in sequence and
    // have freed all allocated memory when done. If this needs to change in the future,
@@ -383,7 +383,6 @@ private:
    Platform* mPlatform = nullptr;
    bool mOwnPlatform = false;
    void* mSharedGLContext = nullptr;
-    bool mTerminated = false;
    backend::Handle<backend::HwRenderPrimitive> mFullScreenTriangleRph;
    FVertexBuffer* mFullScreenTriangleVb = nullptr;
    FIndexBuffer* mFullScreenTriangleIb = nullptr;
@@ -424,7 +423,7 @@ private:
    // FMaterialInstance are handled directly by FMaterial
    std::unordered_map<const FMaterial*, ResourceList<FMaterialInstance>> mMaterialInstances;

-    std::unique_ptr<DFG> mDFG;
+    DFG mDFG;

    std::thread mDriverThread;
    backend::CommandBufferQueue mCommandBufferQueue;
@@ -139,17 +139,32 @@ FRenderTarget::FRenderTarget(FEngine& engine, const RenderTarget::Builder& build
        }
    };

+    UTILS_NOUNROLL
    for (size_t i = 0; i < MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT; i++) {
-        if (mAttachments[i].texture) {
-            mAttachmentMask |= getTargetBufferFlagsAt(i);
+        Attachment const& attachment = mAttachments[i];
+        if (attachment.texture) {
+            TargetBufferFlags targetBufferBit = getTargetBufferFlagsAt(i);
+            mAttachmentMask |= targetBufferBit;
            setAttachment(mrt[i], (AttachmentPoint)i);
+            if (any(attachment.texture->getUsage() &
+                    (TextureUsage::SAMPLEABLE | Texture::Usage::SUBPASS_INPUT))) {
+                mSampleableAttachmentsMask |= targetBufferBit;
+            }
        }
    }
-    if (mAttachments[(size_t)AttachmentPoint::DEPTH].texture) {
+
+    Attachment const& depthAttachment = mAttachments[(size_t)AttachmentPoint::DEPTH];
+    if (depthAttachment.texture) {
        mAttachmentMask |= TargetBufferFlags::DEPTH;
        setAttachment(dinfo, AttachmentPoint::DEPTH);
+        if (any(depthAttachment.texture->getUsage() &
+                (TextureUsage::SAMPLEABLE | Texture::Usage::SUBPASS_INPUT))) {
+            mSampleableAttachmentsMask |= TargetBufferFlags::DEPTH;
+        }
    }

+    // TODO: add stencil here when we support it
+
    FEngine::DriverApi& driver = engine.getDriverApi();
    mHandle = driver.createRenderTarget(mAttachmentMask,
            builder.mImpl->mWidth, builder.mImpl->mHeight, builder.mImpl->mSamples, mrt, dinfo, {});
@@ -55,6 +55,10 @@ public:
        return mAttachmentMask;
    }

+    backend::TargetBufferFlags getSampleableAttachmentsMask() const noexcept {
+        return mSampleableAttachmentsMask;
+    }
+
    uint8_t getSupportedColorAttachmentsCount() const noexcept {
        return mSupportedColorAttachmentsCount;
    }
@@ -67,6 +71,7 @@ private:
    Attachment mAttachments[ATTACHMENT_COUNT];
    HwHandle mHandle{};
    backend::TargetBufferFlags mAttachmentMask = {};
+    backend::TargetBufferFlags mSampleableAttachmentsMask = {};
    const uint8_t mSupportedColorAttachmentsCount;
 };

@@ -33,17 +33,13 @@
 #include <fg/FrameGraphTexture.h>

 #include <filament/Renderer.h>
-#include <filament/View.h>
 #include <filament/Viewport.h>

 #include <backend/DriverEnums.h>
 #include <backend/Handle.h>
-#include <backend/PresentCallable.h>

 #include <utils/compiler.h>
 #include <utils/Allocator.h>
-#include <utils/JobSystem.h>
-#include <utils/Slice.h>

 #include <tsl/robin_set.h>

@@ -53,12 +49,9 @@ namespace backend {
 class Driver;
 } // namespace backend

-class View;
-
 class FEngine;
 class FRenderTarget;
 class FView;
-class ShadowMap;

 /*
 * A concrete implementation of the Renderer Interface.
@@ -68,39 +61,42 @@ class FRenderer : public Renderer {

 public:
    explicit FRenderer(FEngine& engine);
+
    ~FRenderer() noexcept;

-    void init() noexcept;
+    void terminate(FEngine& engine);

    FEngine& getEngine() const noexcept { return mEngine; }

    math::float4 getShaderUserTime() const { return mShaderUserTime; }

-    // do all the work here!
-    void renderJob(ArenaScope& arena, FView& view);
+    void resetUserTime();

+    // renders a single standalone view. The view must have a a custom rendertarget.
+    void renderStandaloneView(FView const* view);
+
+    // start a frame
    bool beginFrame(FSwapChain* swapChain, uint64_t vsyncSteadyClockTimeNano);

+    // end a frame
+    void endFrame();
+
+    // render a view. must be called between beginFrame/enfFrame.
    void render(FView const* view);

+    // read pixel from the current swapchain. must be called between beginFrame/enfFrame.
    void readPixels(uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
            backend::PixelBufferDescriptor&& buffer);

-    void copyFrame(FSwapChain* dstSwapChain, Viewport const& dstViewport,
-            Viewport const& srcViewport, CopyFrameFlag flags);
-
-    void endFrame();
-
-    void renderStandaloneView(FView const* view);
-
+    // read pixel from a rendertarget. must be called between beginFrame/enfFrame.
    void readPixels(FRenderTarget* renderTarget,
            uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
            backend::PixelBufferDescriptor&& buffer);

-    void resetUserTime();
+    // blits the current swapchain to another one
+    void copyFrame(FSwapChain* dstSwapChain, Viewport const& dstViewport,
+            Viewport const& srcViewport, CopyFrameFlag flags);

-    // Clean-up everything, this is typically called when the client calls Engine::destroyRenderer()
-    void terminate(FEngine& engine);

    void setDisplayInfo(DisplayInfo const& info) noexcept {
        mDisplayInfo = info;
@@ -131,83 +127,45 @@ public:
 private:
    friend class Renderer;
    using Command = RenderPass::Command;
+    using clock = std::chrono::steady_clock;
+    using Epoch = clock::time_point;
+    using duration = clock::duration;
+
+    void initializeClearFlags();
+
+    backend::TextureFormat getHdrFormat(const FView& view, bool translucent) const noexcept;
+    backend::TextureFormat getLdrFormat(bool translucent) const noexcept;
+
+    Epoch getUserEpoch() const { return mUserEpoch; }
+    duration getUserTime() const noexcept { return clock::now() - getUserEpoch(); }

    void getRenderTarget(FView const& view,
            backend::TargetBufferFlags& outAttachementMask,
            backend::Handle<backend::HwRenderTarget>& outTarget) const noexcept;

-    void readPixels(backend::Handle<backend::HwRenderTarget> renderTargetHandle,
-            uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
-            backend::PixelBufferDescriptor&& buffer);
-
-    void renderInternal(FView const* view);
-
-    struct ColorPassConfig {
-        // User Viewport
-        Viewport vp;
-        // Scaled (down) viewport from dynamic resolution
-        Viewport svp;
-        // dynamic resolution scale
-        math::float2 scale;
-        // HDR format
-        backend::TextureFormat hdrFormat;
-        // MSAA sample count
-        uint8_t msaa;
-        // Clear flags
-        backend::TargetBufferFlags clearFlags;
-        // Clear color
-        math::float4 clearColor = {};
-        // Lod offset for the SSR passes
-        float ssrLodOffset;
-        // Contact shadow enabled?
-        bool hasContactShadows;
-    };
-
-    FrameGraphId<FrameGraphTexture> colorPass(FrameGraph& fg, const char* name,
-            FrameGraphTexture::Descriptor const& colorBufferDesc,
-            ColorPassConfig const& config,
-            PostProcessManager::ColorGradingConfig colorGradingConfig,
-            RenderPass::Executor const& passExecutor, FView const& view) const noexcept;
-
-    FrameGraphId<FrameGraphTexture> refractionPass(FrameGraph& fg,
-            ColorPassConfig config,
-            PostProcessManager::ScreenSpaceRefConfig const& ssrConfig,
-            PostProcessManager::ColorGradingConfig colorGradingConfig,
-            RenderPass const& pass, FView const& view) const noexcept;
-
    void recordHighWatermark(size_t watermark) noexcept {
        mCommandsHighWatermark = std::max(mCommandsHighWatermark, watermark);
    }

    size_t getCommandsHighWatermark() const noexcept {
-        return mCommandsHighWatermark * sizeof(RenderPass::Command);
+        return mCommandsHighWatermark;
    }

-    backend::TextureFormat getHdrFormat(const View& view, bool translucent) const noexcept;
-    backend::TextureFormat getLdrFormat(bool translucent) const noexcept;
-
-    void initializeClearFlags();
-
-    using clock = std::chrono::steady_clock;
-    using Epoch = clock::time_point;
-    using duration = clock::duration;
-
-    Epoch getUserEpoch() const { return mUserEpoch; }
-    duration getUserTime() const noexcept {
-        return clock::now() - getUserEpoch();
-    }
+    void renderInternal(FView const* view);
+    void renderJob(ArenaScope& arena, FView& view);

    // keep a reference to our engine
    FEngine& mEngine;
    FrameSkipper mFrameSkipper;
-    backend::Handle<backend::HwRenderTarget> mRenderTarget;
+    backend::Handle<backend::HwRenderTarget> mRenderTargetHandle;
    FSwapChain* mSwapChain = nullptr;
    size_t mCommandsHighWatermark = 0;
    uint32_t mFrameId = 0;
+    uint32_t mViewRenderedCount = 0;
    FrameInfoManager mFrameInfoManager;
-    backend::TextureFormat mHdrTranslucent{};
-    backend::TextureFormat mHdrQualityMedium{};
-    backend::TextureFormat mHdrQualityHigh{};
+    backend::TextureFormat mHdrTranslucent;
+    backend::TextureFormat mHdrQualityMedium;
+    backend::TextureFormat mHdrQualityHigh;
    bool mIsRGB8Supported : 1;
    Epoch mUserEpoch;
    math::float4 mShaderUserTime{};
@@ -33,7 +33,6 @@ using namespace backend;

 struct Stream::BuilderDetails {
    void* mStream = nullptr;
-    intptr_t mExternalTextureId = 0;
    uint32_t mWidth = 0;
    uint32_t mHeight = 0;
 };
@@ -52,11 +51,6 @@ Stream::Builder& Stream::Builder::stream(void* stream) noexcept {
    return *this;
 }

-Stream::Builder& Stream::Builder::stream(intptr_t externalTextureId) noexcept {
-    mImpl->mExternalTextureId = externalTextureId;
-    return *this;
-}
-
 Stream::Builder& Stream::Builder::width(uint32_t width) noexcept {
    mImpl->mWidth = width;
    return *this;
@@ -68,11 +62,6 @@ Stream::Builder& Stream::Builder::height(uint32_t height) noexcept {
 }

 Stream* Stream::Builder::build(Engine& engine) {
-    if (!ASSERT_PRECONDITION_NON_FATAL(!mImpl->mStream || !mImpl->mExternalTextureId,
-            "One and only one of the stream or external texture can be specified")) {
-        return nullptr;
-    }
-
    return upcast(engine).createStream(*this);
 }

@@ -80,21 +69,14 @@ Stream* Stream::Builder::build(Engine& engine) {

 FStream::FStream(FEngine& engine, const Builder& builder) noexcept
        : mEngine(engine),
-          mStreamType(
-            builder->mExternalTextureId ? StreamType::TEXTURE_ID :
-            (builder->mStream ? StreamType::NATIVE : StreamType::ACQUIRED)
-          ),
+          mStreamType(builder->mStream ? StreamType::NATIVE : StreamType::ACQUIRED),
          mNativeStream(builder->mStream),
-          mExternalTextureId(builder->mExternalTextureId),
          mWidth(builder->mWidth),
          mHeight(builder->mHeight) {

    if (mNativeStream) {
        // Note: this is a synchronous call. On Android, this calls back into Java.
        mStreamHandle = engine.getDriverApi().createStreamNative(mNativeStream);
-    } else if (mExternalTextureId) {
-        mStreamHandle = engine.getDriverApi().createStreamFromTextureId(
-                mExternalTextureId, mWidth, mHeight);
    } else {
        mStreamHandle = engine.getDriverApi().createStreamAcquired();
    }
@@ -124,28 +106,6 @@ void FStream::setDimensions(uint32_t width, uint32_t height) noexcept {
    mEngine.getDriverApi().setStreamDimensions(mStreamHandle, mWidth, mHeight);
 }

-void FStream::readPixels(uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
-        backend::PixelBufferDescriptor&& buffer) noexcept {
-    if (getStreamType() == StreamType::TEXTURE_ID) {
-        // this works only on external texture id streams
-
-        const size_t sizeNeeded = PixelBufferDescriptor::computeDataSize(
-                buffer.format, buffer.type,
-                buffer.stride ? buffer.stride : width,
-                buffer.top + height,
-                buffer.alignment);
-
-        if (!ASSERT_POSTCONDITION_NON_FATAL(buffer.size >= sizeNeeded,
-                "buffer.size too small %u bytes, needed %u bytes", buffer.size, sizeNeeded)) {
-            return;
-        }
-
-        FEngine::DriverApi& driver = mEngine.getDriverApi();
-        driver.readStreamPixels(mStreamHandle,
-                xoffset, yoffset, width, height, std::move(buffer));
-    }
-}
-
 int64_t FStream::getTimestamp() const noexcept {
    FEngine::DriverApi& driver = mEngine.getDriverApi();
    return driver.getStreamTimestamp(mStreamHandle);
@@ -41,9 +41,6 @@ public:

    void setDimensions(uint32_t width, uint32_t height) noexcept;

-    void readPixels(uint32_t xoffset, uint32_t yoffset, uint32_t width, uint32_t height,
-            backend::PixelBufferDescriptor&& buffer) noexcept;
-
    StreamType getStreamType() const noexcept { return mStreamType; }

    uint32_t getWidth() const noexcept { return mWidth; }
@@ -57,7 +54,6 @@ private:
    const StreamType mStreamType;
    backend::Handle<backend::HwStream> mStreamHandle;
    void* mNativeStream = nullptr;
-    intptr_t mExternalTextureId;
    uint32_t mWidth;
    uint32_t mHeight;
 };
@@ -92,9 +92,6 @@ FView::FView(FEngine& engine)
    mLightUbh = driver.createBufferObject(CONFIG_MAX_LIGHT_COUNT * sizeof(LightsUib),
            BufferObjectBinding::UNIFORM, BufferUsage::DYNAMIC);

-    mShadowUbh = driver.createBufferObject(mShadowUb.getSize(),
-            BufferObjectBinding::UNIFORM, BufferUsage::DYNAMIC);
-
    mIsDynamicResolutionSupported = driver.isFrameTimeSupported();

    mDefaultColorGrading = mColorGrading = engine.getDefaultColorGrading();
@@ -114,9 +111,9 @@ void FView::terminate(FEngine& engine) {

    DriverApi& driver = engine.getDriverApi();
    driver.destroyBufferObject(mLightUbh);
-    driver.destroyBufferObject(mShadowUbh);
    driver.destroyBufferObject(mRenderableUbh);
    drainFrameHistory(engine);
+    mShadowMapManager.terminate(engine);
    mPerViewUniforms.terminate(driver);
    mFroxelizer.terminate(driver);
 }
@@ -160,10 +157,6 @@ float2 FView::updateScale(FEngine& engine,
        Renderer::FrameRateOptions const& frameRateOptions,
        Renderer::DisplayInfo const& displayInfo) noexcept {

-    // scale factor returned to the caller is modified so the scaled viewport is rounded to
-    // 8 pixels. The internal scale factor, mScale, doesn't have this rounding.
-    float2 roundedScale = mScale;
-
    DynamicResolutionOptions const& options = mDynamicResolution;
    if (options.enabled) {
        if (!UTILS_UNLIKELY(info.valid)) {
@@ -241,14 +234,8 @@ float2 FView::updateScale(FEngine& engine,
        // (i.e. we clamped). This help not to have to wait too long for the Integral term
        // to kick in after a clamping event.
        mPidController.setIntegralInhibitionEnabled(mScale != s);
-
-        // now tweak the scaling factor to get multiples of 8 (to help quad-shading)
-        // i.e. 8x8=64 fragments, to try to help with warp sizes.
-        roundedScale.x = mScale.x == 1.0f ? 1.0f : (std::floor(mScale.x * float{ w } / 8) * 8) / float{ w };
-        roundedScale.y = mScale.y == 1.0f ? 1.0f : (std::floor(mScale.y * float{ h } / 8) * 8) / float{ h };
    } else {
        mScale = 1.0f;
-        roundedScale = 1.0f;
    }

 #ifndef NDEBUG
@@ -270,7 +257,7 @@ float2 FView::updateScale(FEngine& engine,
    };
 #endif

-    return roundedScale;
+    return mScale;
 }

 void FView::setVisibleLayers(uint8_t select, uint8_t values) noexcept {
@@ -283,7 +270,8 @@ bool FView::isSkyboxVisible() const noexcept {
 }

 void FView::prepareShadowing(FEngine& engine, DriverApi& driver,
-        FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData) noexcept {
+        FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData,
+        CameraInfo const& cameraInfo) noexcept {
    SYSTRACE_CALL();

    mHasShadowing = false;
@@ -338,18 +326,17 @@ void FView::prepareShadowing(FEngine& engine, DriverApi& driver,
        }
    }

-    auto shadowTechnique = mShadowMapManager.update(engine, *this,
-            mShadowUb, renderableData, lightData);
+    auto shadowTechnique = mShadowMapManager.update(engine, *this, cameraInfo,
+            renderableData, lightData);

    mHasShadowing = any(shadowTechnique);
    mNeedsShadowMap = any(shadowTechnique & ShadowMapManager::ShadowTechnique::SHADOW_MAP);
 }

 void FView::prepareLighting(FEngine& engine, FEngine::DriverApi& driver, ArenaScope& arena,
-        filament::Viewport const& viewport) noexcept {
+        filament::Viewport const& viewport, CameraInfo const& cameraInfo) noexcept {
    SYSTRACE_CALL();

-    const CameraInfo& camera = mViewingCameraInfo;
    FScene* const scene = mScene;
    auto const& lightData = scene->getLightData();

@@ -359,9 +346,10 @@ void FView::prepareLighting(FEngine& engine, FEngine::DriverApi& driver, ArenaSc

    mHasDynamicLighting = scene->getLightData().size() > FScene::DIRECTIONAL_LIGHTS_COUNT;
    if (mHasDynamicLighting) {
-        scene->prepareDynamicLights(camera, arena, mLightUbh);
+        scene->prepareDynamicLights(cameraInfo, arena, mLightUbh);
        Froxelizer& froxelizer = mFroxelizer;
-        if (froxelizer.prepare(driver, arena, viewport, camera.projection, camera.zn, camera.zf)) {
+        if (froxelizer.prepare(driver, arena, viewport,
+                cameraInfo.projection, cameraInfo.zn, cameraInfo.zf)) {
            // update our uniform buffer if needed
            mPerViewUniforms.prepareDynamicLights(mFroxelizer);
        }
@@ -374,9 +362,9 @@ void FView::prepareLighting(FEngine& engine, FEngine::DriverApi& driver, ArenaSc
     * Exposure
     */

-    const float exposure = Exposure::exposure(camera.ev100);
+    const float exposure = Exposure::exposure(cameraInfo.ev100);

-    mPerViewUniforms.prepareExposure(camera.ev100);
+    mPerViewUniforms.prepareExposure(cameraInfo.ev100);

    /*
     * Indirect light (IBL)
@@ -406,16 +394,8 @@ void FView::prepareLighting(FEngine& engine, FEngine::DriverApi& driver, ArenaSc
    mHasDirectionalLight = directionalLight.isValid();
 }

-void FView::prepare(FEngine& engine, DriverApi& driver, ArenaScope& arena,
-        filament::Viewport const& viewport, float4 const& userTime, bool needsAlphaChannel) noexcept {
-    JobSystem& js = engine.getJobSystem();
-
-    /*
-     * Prepare the scene -- this is where we gather all the objects added to the scene,
-     * and in particular their world-space AABB.
-     */
-
-    FScene* const scene = getScene();
+CameraInfo FView::computeCameraInfo(FEngine& engine) const noexcept {
+    FScene const* const scene = getScene();

    /*
     * We apply a "world origin" to "everything" in order to implement the IBL rotation.
@@ -444,20 +424,45 @@ void FView::prepare(FEngine& engine, DriverApi& driver, ArenaScope& arena,
        worldOriginScene[3].xyz -= camera->getPosition();
    }

-    // Note: for debugging (i.e. visualize what the camera / objects are doing, using
-    // the viewing camera), we can set worldOriginScene to identity when mViewingCamera
-    // is set
-    mViewingCameraInfo = CameraInfo(*camera, worldOriginScene);
+    return { *camera, worldOriginScene };
+}

-    mCullingFrustum = Frustum(mat4f{
-                    mCullingCamera->getCullingProjectionMatrix() *
-                    inverse(worldOriginScene * mCullingCamera->getModelMatrix()) });
+void FView::prepare(FEngine& engine, DriverApi& driver, ArenaScope& arena,
+        filament::Viewport const& viewport, CameraInfo const& cameraInfo,
+        float4 const& userTime, bool needsAlphaChannel) noexcept {
+
+    JobSystem& js = engine.getJobSystem();
+
+    /*
+     * Prepare the scene -- this is where we gather all the objects added to the scene,
+     * and in particular their world-space AABB.
+     */
+
+    auto getFrustum = [this, &cameraInfo]() -> Frustum {
+        if (UTILS_LIKELY(mViewingCamera == nullptr)) {
+            // In the common case when we don't have a viewing camera, cameraInfo.view is
+            // already the culling view matrix
+            return Frustum{ mat4f{ highPrecisionMultiply(cameraInfo.projection, cameraInfo.view) }};
+        } else {
+            // Otherwise, we need to recalculate it from the culling camera.
+            // Note: it is correct to always do the math from mCullingCamera, but it hides the
+            // intent of the code, which is that we should only depend on CameraInfo here.
+            // This is an extremely uncommon case.
+            const mat4 projection = mCullingCamera->getCullingProjectionMatrix();
+            const mat4 view = inverse(cameraInfo.worldOrigin * mCullingCamera->getModelMatrix());
+            return Frustum{ mat4f{ projection * view }};
+        }
+    };
+
+    const Frustum cullingFrustum = getFrustum();
+
+    FScene* const scene = getScene();

    /*
     * Gather all information needed to render this scene. Apply the world origin to all
     * objects in the scene.
     */
-    scene->prepare(worldOriginScene, hasVSM());
+    scene->prepare(cameraInfo.worldOrigin, hasVSM());

    /*
     * Light culling: runs in parallel with Renderable culling (below)
@@ -466,9 +471,9 @@ void FView::prepare(FEngine& engine, DriverApi& driver, ArenaScope& arena,
    JobSystem::Job* prepareVisibleLightsJob = nullptr;
    if (scene->getLightData().size() > FScene::DIRECTIONAL_LIGHTS_COUNT) {
        prepareVisibleLightsJob = js.runAndRetain(js.createJob(nullptr,
-                [this, &engine, &arena, scene](JobSystem&, JobSystem::Job*) {
+                [&cullingFrustum, &engine, &arena, &cameraInfo, scene](JobSystem&, JobSystem::Job*) {
                    FView::prepareVisibleLights(engine.getLightManager(), arena,
-                            mViewingCameraInfo, mCullingFrustum, scene->getLightData());
+                            cameraInfo.view, cullingFrustum, scene->getLightData());
                }));
    }

@@ -485,7 +490,7 @@ void FView::prepare(FEngine& engine, DriverApi& driver, ArenaScope& arena,
         * (this will set the VISIBLE_RENDERABLE bit)
         */

-        prepareVisibleRenderables(js, mCullingFrustum, renderableData);
+        prepareVisibleRenderables(js, cullingFrustum, renderableData);


        /*
@@ -497,7 +502,7 @@ void FView::prepare(FEngine& engine, DriverApi& driver, ArenaScope& arena,
        if (prepareVisibleLightsJob) {
            js.waitAndRelease(prepareVisibleLightsJob);
        }
-        prepareShadowing(engine, driver, renderableData, scene->getLightData());
+        prepareShadowing(engine, driver, renderableData, scene->getLightData(), cameraInfo);

        /*
         * Partition the SoA so that renderables are partitioned w.r.t their visibility into the
@@ -567,14 +572,14 @@ void FView::prepare(FEngine& engine, DriverApi& driver, ArenaScope& arena,
     * Relies on FScene::prepare() and prepareVisibleLights()
     */

-    prepareLighting(engine, driver, arena, viewport);
+    prepareLighting(engine, driver, arena, viewport, cameraInfo);

    /*
     * Update driver state
     */

    mPerViewUniforms.prepareTime(userTime);
-    mPerViewUniforms.prepareFog(mViewingCameraInfo, mFogOptions);
+    mPerViewUniforms.prepareFog(cameraInfo.getPosition(), mFogOptions);
    mPerViewUniforms.prepareTemporalNoise(mTemporalAntiAliasingOptions);
    mPerViewUniforms.prepareBlending(needsAlphaChannel);

@@ -644,14 +649,17 @@ void FView::prepareUpscaler(float2 scale) const noexcept {
    mPerViewUniforms.prepareUpscaler(scale, mDynamicResolution);
 }

-void FView::prepareCamera(const CameraInfo& camera) const noexcept {
+void FView::prepareCamera(const CameraInfo& cameraInfo) const noexcept {
    SYSTRACE_CALL();
-    mPerViewUniforms.prepareCamera(camera);
+    mPerViewUniforms.prepareCamera(cameraInfo);
 }

-void FView::prepareViewport(const filament::Viewport& viewport) const noexcept {
+void FView::prepareViewport(const filament::Viewport& viewport,
+        uint32_t xoffset, uint32_t yoffset) const noexcept {
    SYSTRACE_CALL();
-    mPerViewUniforms.prepareViewport(viewport);
+    // TODO: we should pass viewport.{left|bottom} to the backend so it can offset the
+    //       scissor properly.
+    mPerViewUniforms.prepareViewport(viewport, xoffset, yoffset);
 }

 void FView::prepareSSAO(Handle<HwTexture> ssao) const noexcept {
@@ -676,42 +684,39 @@ void FView::prepareStructure(Handle<HwTexture> structure) const noexcept {
 }

 void FView::prepareShadow(Handle<HwTexture> texture) const noexcept {
+    const auto& uniforms = mShadowMapManager.getShadowMappingUniforms();
    switch (mShadowType) {
        case filament::ShadowType::PCF:
-            mPerViewUniforms.prepareShadowPCF(texture);
+            mPerViewUniforms.prepareShadowPCF(texture, uniforms);
            break;
        case filament::ShadowType::VSM:
-            mPerViewUniforms.prepareShadowVSM(texture, mVsmShadowOptions);
+            mPerViewUniforms.prepareShadowVSM(texture, uniforms, mVsmShadowOptions);
            break;
        case filament::ShadowType::DPCF:
-            mPerViewUniforms.prepareShadowDPCF(texture, mSoftShadowOptions);
+            mPerViewUniforms.prepareShadowDPCF(texture, uniforms, mSoftShadowOptions);
            break;
        case filament::ShadowType::PCSS:
-            mPerViewUniforms.prepareShadowPCSS(texture, mSoftShadowOptions);
+            mPerViewUniforms.prepareShadowPCSS(texture, uniforms, mSoftShadowOptions);
            break;
    }
 }

 void FView::prepareShadowMap() const noexcept {
-    mPerViewUniforms.prepareShadowMapping(
-            mShadowMapManager.getShadowMappingUniforms(), mVsmShadowOptions);
+    mPerViewUniforms.prepareShadowMapping();
 }

 void FView::cleanupRenderPasses() const noexcept {
    mPerViewUniforms.unbindSamplers();
 }

-void FView::froxelize(FEngine& engine) const noexcept {
+void FView::froxelize(FEngine& engine, mat4f const& viewMatrix) const noexcept {
    SYSTRACE_CALL();
    assert_invariant(mHasDynamicLighting);
-    mFroxelizer.froxelizeLights(engine, mViewingCameraInfo, mScene->getLightData());
+    mFroxelizer.froxelizeLights(engine, viewMatrix, mScene->getLightData());
 }

 void FView::commitUniforms(DriverApi& driver) const noexcept {
    mPerViewUniforms.commit(driver);
-    if (mShadowUb.isDirty()) {
-        driver.updateBufferObject(mShadowUbh, mShadowUb.toBufferDescriptor(driver), 0);
-    }
 }

 void FView::commitFroxels(DriverApi& driverApi) const noexcept {
@@ -759,7 +764,8 @@ void FView::cullRenderables(JobSystem& js,
 }

 void FView::prepareVisibleLights(FLightManager const& lcm, ArenaScope& rootArena,
-        const CameraInfo& camera, Frustum const& frustum, FScene::LightSoa& lightData) noexcept {
+        mat4f const& viewMatrix, Frustum const& frustum,
+        FScene::LightSoa& lightData) noexcept {
    SYSTRACE_CALL();
    assert_invariant(lightData.size() > FScene::DIRECTIONAL_LIGHTS_COUNT);

@@ -837,7 +843,7 @@ void FView::prepareVisibleLights(FLightManager const& lcm, ArenaScope& rootArena
        // pre-compute the lights' distance to the camera, for sorting below
        // - we don't skip the directional light, because we don't care, it's ignored during sorting
        float4 const* const UTILS_RESTRICT spheres = lightData.data<FScene::POSITION_RADIUS>();
-        computeLightCameraDistances(distances, camera, spheres, size);
+        computeLightCameraDistances(distances, viewMatrix, spheres, size);

        // skip directional light
        Zip2Iterator<FScene::LightSoa::iterator, float*> b = { lightData.begin(), distances };
@@ -855,15 +861,15 @@ void FView::prepareVisibleLights(FLightManager const& lcm, ArenaScope& rootArena
 UTILS_ALWAYS_INLINE
 inline void FView::computeLightCameraDistances(
        float* UTILS_RESTRICT const distances,
-        CameraInfo const& UTILS_RESTRICT camera,
-        float4 const* UTILS_RESTRICT const spheres, size_t count) noexcept {
+        mat4f const& UTILS_RESTRICT viewMatrix,
+        float4 const* UTILS_RESTRICT spheres, size_t count) noexcept {

    // without this, the vectorization is less efficient
    // we're guaranteed to have a multiple of 4 lights (at least)
    count = uint32_t(count + 3u) & ~3u;
    for (size_t i = 0 ; i < count; i++) {
        const float4 sphere = spheres[i];
-        const float4 center = camera.view * sphere.xyz; // camera points towards the -z axis
+        const float4 center = viewMatrix * sphere.xyz; // camera points towards the -z axis
        distances[i] = length(center);
    }
 }
@@ -51,6 +51,7 @@
 #include <utils/Slice.h>

 #include <math/scalar.h>
+#include <math/mat4.h>

 namespace utils {
 class JobSystem;
@@ -118,8 +119,11 @@ public:

    void terminate(FEngine& engine);

+    CameraInfo computeCameraInfo(FEngine& engine) const noexcept;
+
    void prepare(FEngine& engine, backend::DriverApi& driver, ArenaScope& arena,
-            Viewport const& viewport, math::float4 const& userTime, bool needsAlphaChannel) noexcept;
+            filament::Viewport const& viewport, CameraInfo const& cameraInfo,
+            math::float4 const& userTime, bool needsAlphaChannel) noexcept;

    void setScene(FScene* scene) { mScene = scene; }
    FScene const* getScene() const noexcept { return mScene; }
@@ -128,8 +132,6 @@ public:
    void setCullingCamera(FCamera* camera) noexcept { mCullingCamera = camera; }
    void setViewingCamera(FCamera* camera) noexcept { mViewingCamera = camera; }

-    CameraInfo const& getCameraInfo() const noexcept { return mViewingCameraInfo; }
-
    void setViewport(Viewport const& viewport) noexcept;
    Viewport const& getViewport() const noexcept {
        return mViewport;
@@ -159,16 +161,17 @@ public:

    // returns the view's name. The pointer is owned by View.
    const char* getName() const noexcept {
-        return mName.c_str();
+        return mName.c_str_safe();
    }

    void prepareUpscaler(math::float2 scale) const noexcept;
-    void prepareCamera(const CameraInfo& camera) const noexcept;
-    void prepareViewport(const Viewport& viewport) const noexcept;
+    void prepareCamera(const CameraInfo& cameraInfo) const noexcept;
+    void prepareViewport(const Viewport& viewport, uint32_t xoffset, uint32_t yoffset) const noexcept;
    void prepareShadowing(FEngine& engine, backend::DriverApi& driver,
-            FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData) noexcept;
-    void prepareLighting(FEngine& engine, FEngine::DriverApi& driver,
-            ArenaScope& arena, Viewport const& viewport) noexcept;
+            FScene::RenderableSoa& renderableData, FScene::LightSoa& lightData,
+            CameraInfo const& cameraInfo) noexcept;
+    void prepareLighting(FEngine& engine, FEngine::DriverApi& driver, ArenaScope& arena,
+            filament::Viewport const& viewport, CameraInfo const &cameraInfo) noexcept;

    void prepareSSAO(backend::Handle<backend::HwTexture> ssao) const noexcept;
    void prepareSSR(backend::Handle<backend::HwTexture> ssr, float refractionLodOffset,
@@ -181,7 +184,7 @@ public:
    void prepareShadowMap() const noexcept;

    void cleanupRenderPasses() const noexcept;
-    void froxelize(FEngine& engine) const noexcept;
+    void froxelize(FEngine& engine, math::mat4f const& viewMatrix) const noexcept;
    void commitUniforms(backend::DriverApi& driver) const noexcept;
    void commitFroxels(backend::DriverApi& driverApi) const noexcept;

@@ -407,8 +410,6 @@ public:
    static void cullRenderables(utils::JobSystem& js, FScene::RenderableSoa& renderableData,
            Frustum const& frustum, size_t bit) noexcept;

-    auto& getShadowUniforms() const { return mShadowUb; }
-
    PerViewUniforms const& getPerViewUniforms() const noexcept { return mPerViewUniforms; }
    PerViewUniforms& getPerViewUniforms() noexcept { return mPerViewUniforms; }

@@ -419,7 +420,7 @@ public:

    // Clean-up the oldest frame and save the current frame information.
    // This is typically called after all operations for this View's rendering are complete.
-    // (e.g.: after the FrameFraph execution).
+    // (e.g.: after the FrameGraph execution).
    void commitFrameHistory(FEngine& engine) noexcept;

    // create the picking query
@@ -461,11 +462,11 @@ private:
            Frustum const& frustum, FScene::RenderableSoa& renderableData) const noexcept;

    static void prepareVisibleLights(FLightManager const& lcm, ArenaScope& rootArena,
-            const CameraInfo& camera, Frustum const& frustum,
+            math::mat4f const& viewMatrix, Frustum const& frustum,
            FScene::LightSoa& lightData) noexcept;

    static inline void computeLightCameraDistances(float* distances,
-            const CameraInfo& camera, const math::float4* spheres, size_t count) noexcept;
+            math::mat4f const& viewMatrix, const math::float4* spheres, size_t count) noexcept;

    static void computeVisibilityMasks(
            uint8_t visibleLayers, uint8_t const* layers,
@@ -476,7 +477,7 @@ private:
    void bindPerViewUniformsAndSamplers(FEngine::DriverApi& driver) const noexcept {
        mPerViewUniforms.bind(driver);
        driver.bindUniformBuffer(BindingPoints::LIGHTS, mLightUbh);
-        driver.bindUniformBuffer(BindingPoints::SHADOW, mShadowUbh);
+        driver.bindUniformBuffer(BindingPoints::SHADOW, mShadowMapManager.getShadowUniformsHandle());
        driver.bindUniformBuffer(BindingPoints::FROXEL_RECORDS, mFroxelizer.getRecordBuffer());
    }

@@ -493,16 +494,14 @@ private:

    // these are accessed in the render loop, keep together
    backend::Handle<backend::HwBufferObject> mLightUbh;
-    backend::Handle<backend::HwBufferObject> mShadowUbh;
    backend::Handle<backend::HwBufferObject> mRenderableUbh;

    FScene* mScene = nullptr;
+    // The camera set by the user, used for culling and viewing
    FCamera* mCullingCamera = nullptr;
+    // The optional (debug) camera, used only for viewing
    FCamera* mViewingCamera = nullptr;

-    CameraInfo mViewingCameraInfo;
-    Frustum mCullingFrustum{};
-
    mutable Froxelizer mFroxelizer;

    Viewport mViewport;
@@ -540,7 +539,6 @@ private:
    RenderQuality mRenderQuality;

    mutable PerViewUniforms mPerViewUniforms;
-    mutable TypedUniformBuffer<ShadowUib> mShadowUb;

    mutable FrameHistory mFrameHistory{};

@@ -215,8 +215,6 @@ void FrameGraph::execute(backend::DriverApi& driver) noexcept {

        driver.popGroupMarker();
    }
-    // this is a good place to kick the GPU, since we've just done a bunch of work
-    driver.flush();
    driver.popGroupMarker();
 }

@@ -39,14 +39,19 @@ using namespace math;
 #pragma clang diagnostic pop

 void FSR_ScalingSetup(FSRUniforms* outUniforms, FSRScalingConfig config) noexcept {
-    FsrEasuCon( outUniforms->EasuCon0.v, outUniforms->EasuCon1.v,
+    FsrEasuConOffset( outUniforms->EasuCon0.v, outUniforms->EasuCon1.v,
                outUniforms->EasuCon2.v, outUniforms->EasuCon3.v,
            // Viewport size (top left aligned) in the input image which is to be scaled.
-            config.viewportWidth, config.viewportHeight,
+            config.input.width, config.input.height,
            // The size of the input image.
            config.inputWidth, config.inputHeight,
            // The output resolution.
-            config.outputWidth, config.outputHeight);
+            config.outputWidth, config.outputHeight,
+            // Input image offset
+            config.input.left, config.input.bottom);
+    // FIXME: FsrEasu api claims it needs the top,left offset, but when using that we get
+    //        some bad artifacts. To investigate. For reference:
+    //           top = config.inputHeight - (config.input.bottom + config.input.height)
 }

 void FSR_SharpeningSetup(FSRUniforms* outUniforms, FSRSharpeningConfig config) noexcept {
@@ -17,6 +17,8 @@
 #ifndef TNT_FILAMENT_FSR_H
 #define TNT_FILAMENT_FSR_H

+#include <filament/Viewport.h>
+
 #include <math/vec4.h>

 #include <stdint.h>
@@ -24,12 +26,11 @@
 namespace filament {

 struct FSRScalingConfig {
-    uint32_t viewportWidth;
-    uint32_t viewportHeight;
-    uint32_t inputWidth;
-    uint32_t inputHeight;
-    uint32_t outputWidth;
-    uint32_t outputHeight;
+    filament::Viewport input;   // region of source to be scaled
+    uint32_t inputWidth;        // width of source
+    uint32_t inputHeight;       // height of source
+    uint32_t outputWidth;       // width of destination
+    uint32_t outputHeight;      // height of destination
 };

 struct FSRSharpeningConfig {
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Benjamin Doherty	791f51dcb4	Update glslang to e3bca2a	2022-05-04 12:29:21 -07:00
Benjamin Doherty	4621f3ec5d	Filament specific changes	2022-05-04 12:06:26 -07:00
Benjamin Doherty	ab9c09d0d9	Update spirv-tools to 46492aa	2022-05-04 12:00:04 -07:00
Benjamin Doherty	dbf51c1c7d	Update SPIRV-Cross to 0c8a79b	2022-05-04 11:07:15 -07:00
Mathias Agopian	ebd5f150c1	fix a potential threading/memory corruption We can't call prepareProgram() from any thread, since it is calling into the backend. When we had more than a certain number of commands, we would use the jobsystem to generate them and so prepareProgram() could end up being called from a jobsystem thread. This is fixed by looping through all the commands once they're generated and calling prepareProgram() then.	2022-05-03 15:16:11 -07:00
Philip Rideout	156dcb54e8	gltf_viewer: fix accidental removal of camera selector.	2022-05-03 14:33:43 -07:00
Mathias Agopian	88ef76989c	Cleanup Renderer.cpp The main changes are: - ColorPass() and RefractionPass() are now static and moved into a new RendererUtils file. These methods don't need FRenderer and are more like a big "script". - Cleaned-up includes - Reformatted/reordered methods	2022-05-03 14:27:48 -07:00
Philip Rideout	fc284af684	Rename SimpleViewer to ViewerGui. The SimpleViewer C++ class was not viewer component like `ModelViewer` and `<filament-viewer>`. It was actually just the UI builder used in the `gltf_viewer` desktop app and the remote Android interface.	2022-05-03 08:10:53 -07:00
Philip Rideout	3a8685cd15	Upgrade emsdk to 3.1.9	2022-05-03 08:10:26 -07:00
Philip Rideout	f035655f5b	Add support for KHR_texture_basisu. This adds a new implementation of the TextureProvider interface called Ktx2Provider. Tested using the KTX2 variant of the StainedGlassLamp model in the Khronos samples repo. Tested on WebGL 2.0 (Chrome v100), Android (Pixel 6 Pro), and Desktop (Metal, OpenGL, and Vulkan via MoltenVK).	2022-05-02 13:15:56 -07:00
Ben Doherty	f498e08d1e	Update hello-pbr to work in all orientations (#5488 )	2022-05-02 14:41:32 -04:00
Philip Rideout	db3186a65f	Remove etc2comp and astcenc from the repo.	2022-05-02 11:29:19 -07:00
Philip Rideout	248815a3e0	Fix regression with <filament-viewer>.	2022-05-02 11:09:26 -07:00
Benjamin Doherty	ebb62b5d55	Release Filament 1.21.3	2022-05-02 13:17:33 -04:00
Philip Rideout	b9a5ad205a	Fix unit test failures. Basis does not support DXT3 so in the past this unit test passed only due to the enum translation bug that was recently fixed.	2022-05-02 09:48:36 -07:00
Philip Rideout	100986d8d9	Fix WebGL debug build (assert with 3-channel images)	2022-04-29 16:30:14 -07:00
Philip Rideout	b7c9a36d57	Ktx2Reader: Fix error in enum translation table. This caused DXT5 content to be interpreted as DXT3 by the GPU, which seemed to manifest as visual artifacts in the alpha channel.	2022-04-29 16:29:53 -07:00
Philip Rideout	d76551ab28	Add async interface to Ktx2Reader.	2022-04-29 14:20:57 -07:00
Mathias Agopian	92bc0752d1	repair VSM gaussian blur When there was more than one shadowmap we would hit an assert. This bug was introduced recently and was due to incorrect use of the framegraph.	2022-04-29 12:46:37 -07:00
Mathias Agopian	657c251463	Ensure we always have 4 "safe" mip levels A "safe" mip-level is one that has exactly half the dimensions of the current level. In other words, we guarantee that we can halve the dimensions 4 times without loss (i.e. odd dimension). This is achieved by effectively padding the viewport (i.e. making the viewport's dimensions multiple of 16). The post-processing passes eventually take care of cropping the extra padding so the final image is unchanged. This is a more generic solution to dynamic resolution padding, helps with memory allocations and helps all the algorithms that depend on mipmaping (e.g. DoF, SSAO, Bloom). One important improvement brought by this change is that all "final" post-processing effects are now able to handle a sub-region of the source, including: color-gradding, fxaa and upscaling.	2022-04-29 12:05:08 -07:00
Philip Rideout	b6b8fee483	WebGL: fix isTextureFormatSupported for ETC2 formats. It was assumed that ETC2 is supported on all ES3 devices, but WebGL on desktop machines does not support ETC2. Note that WebGL has its own special extension strings for compressed textures.	2022-04-29 12:04:59 -07:00
Mathias Agopian	be5b60f895	cleaner handling of shadow related uniforms Some uniforms are needed during shadow map generation, other during shadow sampling (color pass). We didn't have a clear separation between them and ended up setting the UBOs that were needed only for sampling multiple times during generation.	2022-04-29 12:04:47 -07:00
Mathias Agopian	681d31758f	move the ShadowUib out of View into ShadowMapManager	2022-04-29 12:04:47 -07:00
Ben Doherty	80dc2f5c9d	Move toCompressedFilamentEnum to Ktx1Reader header (#5493 )	2022-04-29 14:49:54 -04:00
Mathias Agopian	ce7dd7a1fb	reorganize again our main UBO - try to sort data by usage. - clarify what post-process materials have access to (not enforced yet)	2022-04-28 10:13:20 -07:00
Mathias Agopian	85617c96a2	shrink post-process materials a bit post-process materials don't need common_shadowing nor common_shading. also don't rely on MIN_ROUGHNESS in .mat files, because it's not public.	2022-04-28 10:13:20 -07:00
Mathias Agopian	58760adb6a	fix typo introduced in previous PR. fixes froxels.	2022-04-27 17:28:16 -07:00
Mathias Agopian	4ef0b72773	rearrange the per-view UBO (breaks materials) - rename common_math and common_shadowing from .fs to .glsl because they were both included in both vertex and fragment shaders - add getFragCoord() which always returns coordinates in the glsl convention. This is useful when used with RNGs to get consistant results accross backends.	2022-04-27 15:27:47 -07:00
Mathias Agopian	45631c3b81	update binding points from 8 to 12.	2022-04-27 15:27:03 -07:00
Mathias Agopian	6f6dde2579	Optimize custom rendertargets use in Renderer we just store the "sampleable" attachments as a bitmask in RenderTarget instead of computing it each time.	2022-04-27 15:26:45 -07:00
Mathias Agopian	e0acbe42cf	update material version	2022-04-27 10:14:38 -07:00
Mathias Agopian	4f137ccf89	don't rely on frameUniform.resolution for froxelization this is because this uniform is not well defined -- it's unclear if it is set to the viewport or the buffer size. also, it's not needed, and removes a couple multiplies from the shader.	2022-04-27 09:38:33 -07:00
Mathias Agopian	04010e7dd9	when matdbg is used don't terminate the program on errors	2022-04-27 08:20:00 -07:00
Mathias Agopian	f4f9f331c0	CameraInfo cleanups - add getUserViewMatrix() on CameraInfo, the "user" view matrix is the view matrix before we apply the world origin transform, it is needed in a few places, so we make it a method so that in the future we could precompute it if we wanted to. - remove worldOffset which is just the last column of worldOrigin	2022-04-26 11:32:41 -07:00
Mathias Agopian	dada291f6b	rename mScaling to mScalingCS	2022-04-26 11:32:41 -07:00
Mathias Agopian	18f2f8b6d4	remove unused parameters and state	2022-04-26 11:32:41 -07:00
Mathias Agopian	d4117ac968	use interleavedGradientNoise() instead of custom implementaion interleavedGradientNoise() is available in common_math, so use that instead of our custom gradient noise RNG.	2022-04-26 11:32:41 -07:00
Mathias Agopian	eccbc8c242	simplify FXAA setup We now do all the setup math on the CPU side, but we also remove some unnecessary computations based on assumptions that are no longer true.	2022-04-26 11:32:41 -07:00
Mathias Agopian	3cdf24f515	remove getUV() from post-process materials it made it actually more confusing to have two version of the uv (normalized and not). Also the unnormalized version was computed from frameUniforms which is not great in the case of post-processing (it just happened to work).	2022-04-26 11:30:35 -07:00
Mathias Agopian	d91faf9df8	avoid unnecessary parameter copies	2022-04-26 11:28:36 -07:00
Brian Stringham	d950efa6ea	Changed ALL_SHADER_STATE_FLAGS initializer to support c++17, which is required for C++/CLI support. (#5465 )	2022-04-26 08:59:44 -07:00
Benjamin Doherty	861d69ea7c	Release Filament 1.21.2	2022-04-25 16:57:00 -04:00
LaiJF	b2cc904fab	Fix typo	2022-04-25 09:06:12 -07:00
Mathias Agopian	3f87b520bb	ShadowMap::render() doesn't need CameraInfo anymore	2022-04-22 12:35:00 -07:00
LaiJF	135a362176	Fix typo (#5455 )	2022-04-22 10:46:04 -04:00
Philip Rideout	391ebc3702	gltfio: fix progress indicator when error occurs. The progress bar in `gltf_viewer` should not appear to hang if any of an asset's textures fail to load.	2022-04-21 17:41:32 -07:00
Philip Rideout	b4d1443720	Repair iOS build. I think.	2022-04-21 15:36:37 -07:00
Philip Rideout	c06737aca2	TextureProvider fixup in ResourceLoader JNI.	2022-04-21 13:44:44 -07:00
Philip Rideout	f847232296	gltfio: introduce TextureProvider API. This allows clients to provide their own asynchronous texture decoders for various mime-typed images. This is a plug-in component for gltfio, in some ways similar to MaterialProvider. There are two motivations for this: to decouple gltfio from STB and to make it easier to integrate support for BasisU textures. This also has the side effect of simplifying ResourceLoader, since the texture decoding jobs have been moved out. As part of this work, I made the "stb" CMake target into a traditional static library. Previously we had several files called `Image.cpp` whose sole purpose was to enable STB_IMAGE_IMPLEMENTATION.	2022-04-21 13:25:40 -07:00
LaiJF	89f06b8923	Fix typo (#5454 )	2022-04-21 09:52:04 -06:00
Philip Rideout	90a2b1f3c9	Remove WebGL workaround for bloom. Bloom now works fine (tested with Chrome 100 on a x86 MacBook) without this workaround for overly aggressive feedback loop detection.	2022-04-20 12:07:49 -07:00
Ben Doherty	059f51338d	Audit backends and test cases for correct Y orientation. (#5440 )	2022-04-20 11:01:31 -06:00
LaiJF	d3cea5121e	Fix typo	2022-04-20 09:58:46 -07:00
LaiJF	c37301d2dc	Fix typo.	2022-04-19 20:27:52 -07:00
Mathias Agopian	aa16778fbd	don't leak Engine if the driver fail to initialize If there is an error during the driver initialization, Engine::create() returns nullptr, but the FEngine was allocated and would then never be destroyed. mTerminated was useless because by construction it would always have the expected value.	2022-04-19 11:49:07 -07:00
Mathias Agopian	ee499ad2ca	don't use a unique_ptr to hold the DFG we can simply make it a member of FEngine.	2022-04-19 11:49:07 -07:00
Mathias Agopian	958dd5bb27	a lot of our workaround don't apply when using ANGLE	2022-04-18 20:19:09 -07:00
Mathias Agopian	f988b75999	fix the 2 passes version of FSR This version is used only on qualcomm devices. It's now forbidden to create a program inside a renderpass. Unfortunately there was still one codepath doing so.	2022-04-18 20:18:45 -07:00
Philip Rideout	e3ccb4bb3c	gltfio Animator: add resetBoneMatrices(). See #5436.	2022-04-18 15:07:35 -07:00
Philip Rideout	3f0dc6f7e6	Minor Ktx2Reader refactoring.	2022-04-18 15:02:08 -07:00
Mathias Agopian	c212dd5bc7	remove the CameraInfo state from FView this state only exists during Renderer::render(), having it outside of view is more flexible as it give us an opportunity to adjust camera parameter globally before drawing. Also greatly reduced the size of RenderPass (which is a Value object).	2022-04-18 12:04:18 -07:00
LaiJF	1eceab0040	Fix typo.	2022-04-18 10:46:31 -07:00
Philip Rideout	ab9d1e5956	Fix unit test.	2022-04-15 12:22:12 -07:00
Philip Rideout	c906290416	Add to release notes.	2022-04-15 10:54:27 -07:00
Philip Rideout	d52a24b6e9	Incorporate feedback from code review.	2022-04-15 10:48:48 -07:00
Philip Rideout	9aeec3a759	Add Ktx2Reader and BasisEncoder and use them in samples. mipgen can now emit basis-encoded KTX2 files. Both the desktop and web "suzanne" samples use this as a test for compressed textures. This PR does not add KTX2 support to glTF, but it's on the way. `BasisEncoder` has a builder style API that calls the basis encoder to create KTX2 files. This hides some low-level BasisU features that we are not using, like file I/O and mipmap generation. `Ktx2Reader` is an easy-to-use API for creating Filament textures from KTX2 files. Its API primarily consists of these two methods: bool requestFormat(Texture::InternalFormat format); Filament::Texture* load(const uint8_t* data, size_t size); The first method is used to build an ordered list of formats that are supported by your hardware. The second method consumes the contents of a basis-encoded KTX2 file and attempts to produce a Filament texture with a preferred format. IMPORTANT: Our tools still let you use KTX1 for non-compressed images because it is useful for HDR, but you can no longer use KTX1 for block-compressed data. Partial fix for #4771.	2022-04-15 10:48:48 -07:00
Mathias Agopian	550fcfe2e3	fix typos	2022-04-15 09:18:06 -07:00
Mathias Agopian	b2cdf9f2b4	don't issue a flush systematically after framegraph's execute This is only a good place for a flush if we're going to draw another view (so the GPU can start working). Otherwise, it can be costly on some drivers and unnecessary since swapbuffers will do the same.	2022-04-15 09:12:39 -07:00
Mathias Agopian	19a3fe3dce	allow radius parameter for upsample bloom stage The original article we used for our bloom (siggraph'14) called for a radius parameter for the upsample stage. We're always using 1.0 right now, but it is useful to keep this possibility for experimentations.	2022-04-14 14:04:32 -07:00
Mathias Agopian	754abbf78e	bloom pass cleanups	2022-04-14 14:04:32 -07:00
Mathias Agopian	9b855e9a03	fix bloom downsample a bug was introduced in 8/2020 where the downsampling filter was wrong, due to a mistake using textureLodOffset, which obviously uses offsets in source texel -- we were assuming destination.	2022-04-14 14:04:32 -07:00
Mathias Agopian	2534fdf92d	propagate the matdbg flag to graddle	2022-04-14 14:04:15 -07:00
Mathias Agopian	dab02e915e	Remove deprecated Stream APIs	2022-04-14 12:51:03 -07:00
Mathias Agopian	68da1a25eb	disable ANGLE optimization that hurts filament	2022-04-14 08:52:56 -07:00
Romain Guy	c67b363576	Fix compilation warnings	2022-04-13 15:15:16 -07:00
Philip Rideout	5c0e7d8211	sample-gltf-viewer: fix regression with zip files. Path resolution was broken with #5285, which only worked when the glTF happened to be at root level. This new fix has been tested against: - My Sponza zip file that hass all files in a "sponza" folder. - My Bistro zip file, which has resources in a child folder relative to the gltf file - A self-created flat zip using IridescentDishWithOlives, with everything at root level. Side note: Bistro still fails on a Pixel 6 Pro Mali but for an unrelated reason: (null deref in backend::TimerQueryNative::queryResultAvailable)	2022-04-13 12:18:47 -07:00