wip: shader compilation benchmark

When parsing a lexeme, we use one less byte than it's intended to be for
comparing the current string.

This results in a success in cases like:
- true and truX
- false and falsX
- null and nulX
where X means an arbitrary character.

Fix this by the full intended length.

CMakeLists.txt

@@ -41,6 +41,8 @@ option(FILAMENT_LINUX_IS_MOBILE "Treat Linux as Mobile" OFF)
 
 option(FILAMENT_ENABLE_ASAN_UBSAN "Enable Address and Undefined Behavior Sanitizers" OFF)
 
+option(FILAMENT_ENABLE_TSAN "Enable Thread Sanitizer" OFF)
+
 option(FILAMENT_ENABLE_FEATURE_LEVEL_0 "Enable Feature Level 0" ON)
 
 set(FILAMENT_NDK_VERSION "" CACHE STRING
@@ -426,7 +428,13 @@ endif()
 if (FILAMENT_ENABLE_ASAN_UBSAN)
    set(EXTRA_SANITIZE_OPTIONS "-fsanitize=address -fsanitize=undefined")
 endif()
-
+if (FILAMENT_ENABLE_TSAN)
+   set(EXTRA_SANITIZE_OPTIONS "-fsanitize=thread")
+endif()
+if (ANDROID)
+    # keep STL debug infos (mimics what the NDK does)
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fno-limit-debug-info")
+endif()
 if (NOT MSVC AND NOT WEBGL)
     set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fstack-protector")
 endif()

NEW_RELEASE_NOTES.md

@@ -7,3 +7,5 @@ for next branch cut* header.
 appropriate header in [RELEASE_NOTES.md](./RELEASE_NOTES.md).
 
 ## Release notes for next branch cut
+
+- Metal: fix some shader artifacts by disabling fast math optimizations.

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.49.1'
+    implementation 'com.google.android.filament:filament-android:1.49.3'
 }
 ```
 
@@ -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:
 
 ```shell
-pod 'Filament', '~> 1.49.1'
+pod 'Filament', '~> 1.49.3'
 ```
 
 ### Snapshots

RELEASE_NOTES.md

@@ -7,6 +7,16 @@ A new header is inserted each time a *tag* is created.
 Instead, if you are authoring a PR for the main branch, add your release note to
 [NEW_RELEASE_NOTES.md](./NEW_RELEASE_NOTES.md).
 
+## v1.50.0
+- engine: TAA now supports 4x upscaling [BETA] [⚠️ **New Material Version**]
+
+## v1.49.3
+
+- matc: Generate stereo variants for FL0 materials [⚠️ **Recompile materials**]
+
+## v1.49.2
+
+
 ## v1.49.1
 
 

android/build.gradle

@@ -81,14 +81,14 @@ buildscript {
     ext.versions = [
         'jdk': 17,
         'minSdk': 19,
-        'targetSdk': 33,
-        'compileSdk': 33,
-        'kotlin': '1.9.0',
-        'kotlin_coroutines': '1.7.2',
+        'targetSdk': 34,
+        'compileSdk': 34,
+        'kotlin': '1.9.21',
+        'kotlin_coroutines': '1.7.3',
         'buildTools': '34.0.0',
-        'ndk': '25.1.8937393',
-        'androidx_core': '1.10.1',
-        'androidx_annotations': '1.6.0'
+        'ndk': '26.1.10909125',
+        'androidx_core': '1.12.0',
+        'androidx_annotations': '1.7.0'
     ]
 
     ext.deps = [
@@ -104,7 +104,7 @@ buildscript {
     ]
 
     dependencies {
-        classpath 'com.android.tools.build:gradle:8.1.0'
+        classpath 'com.android.tools.build:gradle:8.2.0'
         classpath "org.jetbrains.kotlin:kotlin-gradle-plugin:${versions.kotlin}"
     }
 

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

@@ -1637,6 +1637,10 @@ public class View {
          * circle of confusion scale factor (amount of blur)
          */
         public float cocScale = 1.0f;
+        /**
+         * width/height aspect ratio of the circle of confusion (simulate anamorphic lenses)
+         */
+        public float cocAspectRatio = 1.0f;
         /**
          * maximum aperture diameter in meters (zero to disable rotation)
          */
@@ -1852,7 +1856,7 @@ public class View {
     }
 
     /**
-     * Options for Temporal Multi-Sample Anti-aliasing (MSAA)
+     * Options for Multi-Sample Anti-aliasing (MSAA)
      * @see setMultiSampleAntiAliasingOptions()
      */
     public static class MultiSampleAntiAliasingOptions {
@@ -1876,21 +1880,106 @@ public class View {
 
     /**
      * Options for Temporal Anti-aliasing (TAA)
+     * Most TAA parameters are extremely costly to change, as they will trigger the TAA post-process
+     * shaders to be recompiled. These options should be changed or set during initialization.
+     * `filterWidth`, `feedback` and `jitterPattern`, however, could be changed at any time.
+     *
+     * `feedback` of 0.1 effectively accumulates a maximum of 19 samples in steady state.
+     * see "A Survey of Temporal Antialiasing Techniques" by Lei Yang and all for more information.
+     *
      * @see setTemporalAntiAliasingOptions()
      */
     public static class TemporalAntiAliasingOptions {
+        public enum BoxType {
+            /**
+             * use an AABB neighborhood
+             */
+            AABB,
+            /**
+             * use the variance of the neighborhood (not recommended)
+             */
+            VARIANCE,
+            /**
+             * use both AABB and variance
+             */
+            AABB_VARIANCE,
+        }
+
+        public enum BoxClipping {
+            /**
+             * Accurate box clipping
+             */
+            ACCURATE,
+            /**
+             * clamping
+             */
+            CLAMP,
+            /**
+             * no rejections (use for debugging)
+             */
+            NONE,
+        }
+
+        public enum JitterPattern {
+            RGSS_X4,
+            UNIFORM_HELIX_X4,
+            HALTON_23_X8,
+            HALTON_23_X16,
+        }
+
         /**
-         * reconstruction filter width typically between 0 (sharper, aliased) and 1 (smoother)
+         * reconstruction filter width typically between 0.2 (sharper, aliased) and 1.5 (smoother)
          */
         public float filterWidth = 1.0f;
         /**
          * history feedback, between 0 (maximum temporal AA) and 1 (no temporal AA).
          */
-        public float feedback = 0.04f;
+        public float feedback = 0.12f;
+        /**
+         * texturing lod bias (typically -1 or -2)
+         */
+        public float lodBias = -1.0f;
         /**
          * enables or disables temporal anti-aliasing
          */
         public boolean enabled = false;
+        /**
+         * 4x TAA upscaling. Disables Dynamic Resolution. [BETA]
+         */
+        public boolean upscaling = false;
+        /**
+         * whether to filter the history buffer
+         */
+        public boolean filterHistory = true;
+        /**
+         * whether to apply the reconstruction filter to the input
+         */
+        public boolean filterInput = true;
+        /**
+         * whether to use the YcoCg color-space for history rejection
+         */
+        public boolean useYCoCg = false;
+        /**
+         * type of color gamut box
+         */
+        @NonNull
+        public TemporalAntiAliasingOptions.BoxType boxType = TemporalAntiAliasingOptions.BoxType.AABB;
+        /**
+         * clipping algorithm
+         */
+        @NonNull
+        public TemporalAntiAliasingOptions.BoxClipping boxClipping = TemporalAntiAliasingOptions.BoxClipping.ACCURATE;
+        @NonNull
+        public TemporalAntiAliasingOptions.JitterPattern jitterPattern = TemporalAntiAliasingOptions.JitterPattern.HALTON_23_X16;
+        public float varianceGamma = 1.0f;
+        /**
+         * adjust the feedback dynamically to reduce flickering
+         */
+        public boolean preventFlickering = false;
+        /**
+         * whether to apply history reprojection (debug option)
+         */
+        public boolean historyReprojection = true;
     }
 
     /**

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

@@ -329,7 +329,6 @@ public class UiHelper {
         @Override
         public void detach() {
             mTextureView.setSurfaceTextureListener(null);
-            setSurface(null);
         }
 
 
@@ -382,6 +381,7 @@ public class UiHelper {
         @Override
         public boolean onSurfaceTextureDestroyed(@NonNull SurfaceTexture surfaceTexture) {
             if (LOGGING) Log.d(LOG_TAG, "onSurfaceTextureDestroyed()");
+            setSurface(null);
             destroySwapChain();
             return true;
         }
@@ -445,6 +445,9 @@ public class UiHelper {
      * {@link #attachTo(TextureView)}, or {@link #attachTo(SurfaceHolder)}.
      */
     public void detach() {
+        if (mRenderSurface != null) {
+            mRenderSurface.detach();
+        }
         destroySwapChain();
         mNativeWindow = null;
         mRenderSurface = null;
@@ -590,6 +593,10 @@ public class UiHelper {
                 // nothing to do
                 return false;
             }
+            if (mRenderSurface != null) {
+                mRenderSurface.detach();
+                mRenderSurface = null;
+            }
             destroySwapChain();
         }
         mNativeWindow = nativeWindow;
@@ -604,9 +611,6 @@ public class UiHelper {
     }
 
     private void destroySwapChain() {
-        if (mRenderSurface != null) {
-            mRenderSurface.detach();
-        }
         if (mRenderCallback != null) {
             mRenderCallback.onDetachedFromSurface();
         }

android/gltfio-android/src/main/java/com/google/android/filament/gltfio/FilamentInstance.java

@@ -144,7 +144,7 @@ public class FilamentInstance {
      *
      * Ignored if variantIndex is out of bounds.
      */
-    void applyMaterialVariant(@IntRange(from = 0) int variantIndex) {
+    public void applyMaterialVariant(@IntRange(from = 0) int variantIndex) {
         nApplyMaterialVariant(mNativeObject, variantIndex);
     }
 

android/gradle.properties

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.49.1
+VERSION_NAME=1.49.3
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 

android/gradle/wrapper/gradle-wrapper.properties

@@ -1,6 +1,6 @@
 #Wed Nov 17 10:40:18 PST 2021
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-8.0-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-8.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

docs/webgl/reference.html

@@ -587,6 +587,7 @@
 <li><em>overrides</em> Dictionary with one or more of the following properties: mapSize, shadowCascades, constantBias, normalBias, shadowFar, shadowNearHint, shadowFarHint, stable, polygonOffsetConstant, polygonOffsetSlope, \</li>
 </ul>
 </li>
+<li><strong>engine.setStereoscopicOptions()</strong></li>
 <li><strong>engine.setTemporalAntiAliasingOptions()</strong></li>
 <li><strong>engine.setVignetteOptions()</strong></li>
 </ul>
@@ -1462,6 +1463,8 @@ This defines the following functions:</p>
 <li>STENCIL_ATTACHMENT</li>
 <li>UPLOADABLE</li>
 <li>SAMPLEABLE</li>
+<li>BLIT_SRC</li>
+<li>BLIT_DST</li>
 <li>SUBPASS_INPUT</li>
 </ul>
 </div>

filament/CMakeLists.txt

@@ -209,10 +209,13 @@ set(PRIVATE_HDRS
 )
 
 set(MATERIAL_SRCS
-        src/materials/fsr/fsr_easu.mat
-        src/materials/fsr/fsr_easu_mobile.mat
-        src/materials/fsr/fsr_easu_mobileF.mat
-        src/materials/fsr/fsr_rcas.mat
+        src/materials/antiAliasing/fxaa.mat
+        src/materials/antiAliasing/taa.mat
+        src/materials/blitLow.mat
+        src/materials/bloom/bloomDownsample.mat
+        src/materials/bloom/bloomDownsample2x.mat
+        src/materials/bloom/bloomDownsample9.mat
+        src/materials/bloom/bloomUpsample.mat
         src/materials/colorGrading/colorGrading.mat
         src/materials/colorGrading/colorGradingAsSubpass.mat
         src/materials/colorGrading/customResolveAsSubpass.mat
@@ -220,29 +223,27 @@ set(MATERIAL_SRCS
         src/materials/defaultMaterial.mat
         src/materials/dof/dof.mat
         src/materials/dof/dofCoc.mat
-        src/materials/dof/dofDownsample.mat
         src/materials/dof/dofCombine.mat
+        src/materials/dof/dofDilate.mat
+        src/materials/dof/dofDownsample.mat
+        src/materials/dof/dofMedian.mat
+        src/materials/dof/dofMipmap.mat
         src/materials/dof/dofTiles.mat
         src/materials/dof/dofTilesSwizzle.mat
-        src/materials/dof/dofDilate.mat
-        src/materials/dof/dofMipmap.mat
-        src/materials/dof/dofMedian.mat
         src/materials/flare/flare.mat
-        src/materials/bloom/bloomDownsample.mat
-        src/materials/bloom/bloomDownsample2x.mat
-        src/materials/bloom/bloomDownsample9.mat
-        src/materials/bloom/bloomUpsample.mat
+        src/materials/fsr/fsr_easu.mat
+        src/materials/fsr/fsr_easu_mobile.mat
+        src/materials/fsr/fsr_easu_mobileF.mat
+        src/materials/fsr/fsr_rcas.mat
+        src/materials/resolveDepth.mat
+        src/materials/separableGaussianBlur.mat
+        src/materials/skybox.mat
         src/materials/ssao/bilateralBlur.mat
         src/materials/ssao/bilateralBlurBentNormals.mat
         src/materials/ssao/mipmapDepth.mat
-        src/materials/skybox.mat
         src/materials/ssao/sao.mat
         src/materials/ssao/saoBentNormals.mat
-        src/materials/separableGaussianBlur.mat
-        src/materials/antiAliasing/fxaa.mat
-        src/materials/antiAliasing/taa.mat
         src/materials/vsmMipmap.mat
-        src/materials/blitLow.mat
 )
 
 set(MATERIAL_FL0_SRCS
@@ -320,7 +321,7 @@ foreach (mat_src ${MATERIAL_SRCS})
     list(APPEND MATERIAL_BINS ${output_path})
 endforeach()
 
-if (FILAMENT_ENABLE_FEATURE_LEVEL_0 AND FILAMENT_SUPPORTS_OPENGL)
+if (FILAMENT_ENABLE_FEATURE_LEVEL_0)
     foreach (mat_src ${MATERIAL_FL0_SRCS})
         get_filename_component(localname "${mat_src}" NAME_WE)
         get_filename_component(fullname "${mat_src}" ABSOLUTE)
@@ -328,7 +329,7 @@ if (FILAMENT_ENABLE_FEATURE_LEVEL_0 AND FILAMENT_SUPPORTS_OPENGL)
 
         add_custom_command(
                 OUTPUT ${output_path}
-                COMMAND matc -a opengl -p ${MATC_TARGET} ${MATC_OPT_FLAGS} -o ${output_path} ${fullname}
+                COMMAND matc ${MATC_BASE_FLAGS} -o ${output_path} ${fullname}
                 MAIN_DEPENDENCY ${fullname}
                 DEPENDS matc
                 COMMENT "Compiling material ${mat_src} to ${output_path}"

filament/backend/include/private/backend/DriverAPI.inc

@@ -300,6 +300,7 @@ DECL_DRIVER_API_SYNCHRONOUS_0(bool, isAutoDepthResolveSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isSRGBSwapChainSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isStereoSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(bool, isParallelShaderCompileSupported)
+DECL_DRIVER_API_SYNCHRONOUS_0(bool, isDepthStencilResolveSupported)
 DECL_DRIVER_API_SYNCHRONOUS_0(uint8_t, getMaxDrawBuffers)
 DECL_DRIVER_API_SYNCHRONOUS_0(size_t, getMaxUniformBufferSize)
 DECL_DRIVER_API_SYNCHRONOUS_0(math::float2, getClipSpaceParams)

filament/backend/src/metal/MetalBlitter.h

@@ -34,24 +34,12 @@ public:
 
     explicit MetalBlitter(MetalContext& context) noexcept;
 
-    enum class FormatType { NONE, COLOR, DEPTH, STENCIL, DEPTH_STENCIL };
-    static FormatType getFormatType(TextureFormat format) noexcept {
-        bool const depth = isDepthFormat(format);
-        bool const stencil = isStencilFormat(format);
-        if (depth && stencil) return FormatType::DEPTH_STENCIL;
-        if (depth) return FormatType::DEPTH;
-        if (stencil) return FormatType::STENCIL;
-        return FormatType::COLOR;
-    }
-
     struct BlitArgs {
         struct Attachment {
-            using Type = FormatType;
             id<MTLTexture> texture = nil;
             MTLRegion region = {};
             uint8_t level = 0;
             uint32_t slice = 0;      // must be 0 on source attachment
-            Type type = Type::NONE;
         };
 
         // Valid source formats:       2D, 2DArray, 2DMultisample, 3D
@@ -60,16 +48,6 @@ public:
         Attachment destination;
         SamplerMagFilter filter;
 
-        bool blitColor() const {
-            return source.type == Attachment::Type::COLOR &&
-                   destination.type == Attachment::Type::COLOR;
-        }
-
-        bool blitDepth() const {
-            return source.type == Attachment::Type::DEPTH &&
-                   destination.type == Attachment::Type::DEPTH;
-        }
-
         bool destinationIsFullAttachment() const {
             return destination.texture.width == destination.region.size.width &&
                    destination.texture.height == destination.region.size.height;
@@ -90,31 +68,20 @@ private:
             const BlitArgs& args, uint32_t depthPlane);
 
     struct BlitFunctionKey {
-        bool blitColor;
-        bool blitDepth;
-        bool msaaColorSource;
-        bool msaaDepthSource;
-        bool sources3D;
-
-        char padding[3];
+        bool msaaColorSource{};
+        bool sources3D{};
+        char padding[2]{};
 
         bool isValid() const noexcept {
             // MSAA 3D textures do not exist.
-            bool const hasMsaa = msaaColorSource || msaaDepthSource;
+            bool const hasMsaa = msaaColorSource;
             return !(hasMsaa && sources3D);
         }
 
         bool operator==(const BlitFunctionKey& rhs) const noexcept {
-            return blitColor == rhs.blitColor &&
-                   blitDepth == rhs.blitDepth &&
-                   msaaColorSource == rhs.msaaColorSource &&
-                   msaaDepthSource == rhs.msaaDepthSource &&
+            return msaaColorSource == rhs.msaaColorSource &&
                    sources3D == rhs.sources3D;
         }
-
-        BlitFunctionKey() {
-            std::memset(this, 0, sizeof(BlitFunctionKey));
-        }
     };
 
     static bool blitFastPath(id<MTLCommandBuffer> cmdBuffer,
@@ -123,9 +90,8 @@ private:
     void blitSlowPath(id<MTLCommandBuffer> cmdBuffer,
             const BlitArgs& args, const char* label);
 
-    void blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, bool blitColor, bool blitDepth,
-            const BlitArgs& args, uint32_t depthPlaneSource, uint32_t depthPlaneDest,
-            const char* label);
+    void blitDepthPlane(id <MTLCommandBuffer> cmdBuffer, const BlitArgs& args,
+            uint32_t depthPlaneSource, uint32_t depthPlaneDest, const char* label);
 
     id<MTLFunction> compileFragmentFunction(BlitFunctionKey key) const;
     id<MTLFunction> getBlitVertexFunction();

filament/backend/src/metal/MetalBlitter.mm

@@ -36,13 +36,7 @@ struct VertexOut
 
 struct FragmentOut
 {
-#ifdef BLIT_COLOR
     float4 color [[color(0)]];
-#endif
-
-#ifdef BLIT_DEPTH
-    float depth [[depth(any)]];
-#endif
 };
 
 vertex VertexOut
@@ -66,7 +60,6 @@ fragment FragmentOut
 blitterFrag(VertexOut in [[stage_in]],
             sampler sourceSampler [[sampler(0)]],
 
-#ifdef BLIT_COLOR
 #ifdef MSAA_COLOR_SOURCE
             texture2d_ms<float, access::read> sourceColor [[texture(0)]],
 #elif SOURCES_3D
@@ -74,32 +67,18 @@ blitterFrag(VertexOut in [[stage_in]],
 #else
             texture2d<float, access::sample> sourceColor [[texture(0)]],
 #endif  // MSAA_COLOR_SOURCE
-#endif  // BLIT_COLOR
-
-#ifdef BLIT_DEPTH
-#ifdef MSAA_DEPTH_SOURCE
-            texture2d_ms<float, access::read> sourceDepth [[texture(1)]],
-#elif SOURCES_3D
-            texture3d<float, access::sample> sourceDepth [[texture(1)]],
-#else
-            texture2d<float, access::sample> sourceDepth [[texture(1)]],
-#endif  // MSAA_DEPTH_SOURCE
-#endif  // BLIT_DEPTH
 
             constant FragmentArgs* args [[buffer(0)]])
 {
     FragmentOut out = {};
 
-#if defined(BLIT_COLOR) || defined(BLIT_DEPTH)
     // These coordinates match the Vulkan vkCmdBlitImage spec:
     // https://www.khronos.org/registry/vulkan/specs/1.2-extensions/man/html/vkCmdBlitImage.html
     float2 uvbase = in.position.xy; // unnormalized coordinates at center of texel: (1.5, 2.5, etc)
     float2 uvoffset = uvbase - args->dstOffset;
     float2 uvscaled = uvoffset * args->scale;
     float2 uv = uvscaled + args->srcOffset;
-#endif
 
-#ifdef BLIT_COLOR
 #ifdef MSAA_COLOR_SOURCE
     out.color = float4(0.0);
     for (uint s = 0; s < sourceColor.get_num_samples(); s++) {
@@ -115,25 +94,7 @@ blitterFrag(VertexOut in [[stage_in]],
     float2 uvnorm = uv / float2(sourceColor.get_width(args->lod), sourceColor.get_height(args->lod));
     out.color += sourceColor.sample(sourceSampler, uvnorm, level(args->lod));
 #endif  // MSAA_COLOR_SOURCE
-#endif  // BLIT_COLOR
 
-#ifdef BLIT_DEPTH
-#ifdef MSAA_DEPTH_SOURCE
-    out.depth = 0.0;
-    for (uint s = 0; s < sourceDepth.get_num_samples(); s++) {
-        out.depth += sourceDepth.read(static_cast<uint2>(uv), s).r;
-    }
-    out.depth /= sourceDepth.get_num_samples();
-#elif SOURCES_3D
-    float2 uvnormd = uv / float2(sourceDepth.get_width(args->lod), sourceDepth.get_height(args->lod));
-    float3 coords = float3(uvnormd, (static_cast<float>(args->depthPlane) + 0.5) /
-            sourceDepth.get_depth(args->lod));
-    out.depth = sourceDepth.sample(sourceSampler, coords, level(args->lod)).r;
-#else
-    float2 uvnormd = uv / float2(sourceDepth.get_width(args->lod), sourceDepth.get_height(args->lod));
-    out.depth = sourceDepth.sample(sourceSampler, uvnormd, level(args->lod)).r;
-#endif  // MSAA_DEPTH_SOURCE
-#endif  // BLIT_DEPTH
     return out;
 }
 )";
@@ -158,9 +119,12 @@ void MetalBlitter::blit(id<MTLCommandBuffer> cmdBuffer, const BlitArgs& args, co
         return;
     }
 
-    // If the destination is MSAA, and we weren't able to use the fast path (meaning formats
-    // didn't match or sample count didn't match), report an error, as blitting to a MSAA texture
-    // isn't supported through the "slow path" yet.
+    // If we end-up here, it means that either:
+    // - we're resolving a color texture
+    // - src/dest formats didn't match, or we're scaling -- this can only happen with the legacy
+    //   blit() path and implies that the format is not depth.
+    // note: in the future we will support a fast-path resolve
+
     UTILS_UNUSED_IN_RELEASE
     const bool destinationIsMultisample =
             args.destination.texture.textureType == MTLTextureType2DMultisample;
@@ -197,49 +161,31 @@ bool MetalBlitter::blitFastPath(id<MTLCommandBuffer> cmdBuffer,
 
 void MetalBlitter::blitSlowPath(id<MTLCommandBuffer> cmdBuffer,
         const BlitArgs& args, const char* label) {
-    bool const blitColor = args.blitColor();
-    bool const blitDepth = args.blitDepth();
     // scaling in any dimension is not allowed
     assert_invariant(args.source.region.size.depth == args.destination.region.size.depth);
     // we're always blitting a single plane
-    assert_invariant(blitColor != blitDepth);
     uint32_t depthPlaneSource = args.source.region.origin.z;
     uint32_t depthPlaneDest = args.destination.region.origin.z;
     for (NSUInteger d = 0; d < args.source.region.size.depth; d++) {
-        blitDepthPlane(cmdBuffer, blitColor, blitDepth, args,
+        blitDepthPlane(cmdBuffer, args,
                 depthPlaneSource++, depthPlaneDest++, label);
     }
 }
 
-void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, bool blitColor, bool blitDepth,
-        const BlitArgs& args, uint32_t depthPlaneSource, uint32_t depthPlaneDest,
-        const char* label) {
-
-    assert_invariant(blitColor != blitDepth);
+void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, const BlitArgs& args,
+        uint32_t depthPlaneSource, uint32_t depthPlaneDest, const char* label) {
 
     MTLRenderPassDescriptor* const descriptor = [MTLRenderPassDescriptor renderPassDescriptor];
 
-    if (blitColor) {
-        setupAttachment(descriptor.colorAttachments[0], args, depthPlaneDest);
-    }
-
-    if (blitDepth) {
-        setupAttachment(descriptor.depthAttachment, args, depthPlaneDest);
-    }
+    setupAttachment(descriptor.colorAttachments[0], args, depthPlaneDest);
 
     id<MTLRenderCommandEncoder> const encoder =
             [cmdBuffer renderCommandEncoderWithDescriptor:descriptor];
     encoder.label = @(label);
 
     BlitFunctionKey key;
-    key.blitColor = blitColor;
-    key.blitDepth = blitDepth;
-    key.msaaColorSource = blitColor && args.source.texture.textureType == MTLTextureType2DMultisample;
-    key.msaaDepthSource = blitDepth && args.source.texture.textureType == MTLTextureType2DMultisample;
-    key.sources3D       = blitColor && args.source.texture.textureType == MTLTextureType3D;
-    if (key.sources3D && blitDepth) {
-        assert_invariant(args.source.texture.textureType == MTLTextureType3D);
-    }
+    key.msaaColorSource = args.source.texture.textureType == MTLTextureType2DMultisample;
+    key.sources3D       = args.source.texture.textureType == MTLTextureType3D;
     id<MTLFunction> const fragmentFunction = getBlitFragmentFunction(key);
 
     MetalPipelineState const pipelineState {
@@ -247,7 +193,7 @@ void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, bool blitColor
         .fragmentFunction = fragmentFunction,
         .vertexDescription = {},
         .colorAttachmentPixelFormat = {
-            blitColor ? args.destination.texture.pixelFormat : MTLPixelFormatInvalid,
+            args.destination.texture.pixelFormat,
             MTLPixelFormatInvalid,
             MTLPixelFormatInvalid,
             MTLPixelFormatInvalid,
@@ -256,8 +202,7 @@ void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, bool blitColor
             MTLPixelFormatInvalid,
             MTLPixelFormatInvalid
         },
-        .depthAttachmentPixelFormat =
-                blitDepth ? args.destination.texture.pixelFormat : MTLPixelFormatInvalid,
+        .depthAttachmentPixelFormat = MTLPixelFormatInvalid,
         .sampleCount = 1,
         .blendState = {}
     };
@@ -267,21 +212,11 @@ void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, bool blitColor
 
     // For texture arrays, create a view of the texture at the given slice (layer).
 
-    if (blitColor) {
-        id<MTLTexture> srcTextureColor = args.source.texture;
-        if (srcTextureColor && srcTextureColor.textureType == MTLTextureType2DArray) {
-            srcTextureColor = createTextureViewWithSingleSlice(srcTextureColor, args.source.slice);
-        }
-        [encoder setFragmentTexture:srcTextureColor atIndex:0];
-    }
-
-    if (blitDepth) {
-        id<MTLTexture> srcTextureDepth = args.source.texture;
-        if (srcTextureDepth && srcTextureDepth.textureType == MTLTextureType2DArray) {
-            srcTextureDepth = createTextureViewWithSingleSlice(srcTextureDepth, args.source.slice);
-        }
-        [encoder setFragmentTexture:srcTextureDepth atIndex:1];
+    id<MTLTexture> srcTextureColor = args.source.texture;
+    if (srcTextureColor && srcTextureColor.textureType == MTLTextureType2DArray) {
+        srcTextureColor = createTextureViewWithSingleSlice(srcTextureColor, args.source.slice);
     }
+    [encoder setFragmentTexture:srcTextureColor atIndex:0];
 
     SamplerMinFilter filterMin = SamplerMinFilter::NEAREST_MIPMAP_NEAREST;
     if (args.filter == SamplerMagFilter::NEAREST) {
@@ -310,7 +245,7 @@ void MetalBlitter::blitDepthPlane(id<MTLCommandBuffer> cmdBuffer, bool blitColor
 
     DepthStencilState const depthStencilState {
         .depthCompare = MTLCompareFunctionAlways,
-        .depthWriteEnabled = blitDepth
+        .depthWriteEnabled = false
     };
     id<MTLDepthStencilState> const depthStencil =
             mContext.depthStencilStateCache.getOrCreateState(depthStencilState);
@@ -370,18 +305,9 @@ void MetalBlitter::setupAttachment(MTLRenderPassAttachmentDescriptor* descriptor
 id<MTLFunction> MetalBlitter::compileFragmentFunction(BlitFunctionKey key) const {
     MTLCompileOptions* const options = [MTLCompileOptions new];
     NSMutableDictionary* const macros = [NSMutableDictionary dictionary];
-    if (key.blitColor) {
-        macros[@"BLIT_COLOR"] = @"1";
-    }
-    if (key.blitDepth) {
-        macros[@"BLIT_DEPTH"] = @"1";
-    }
     if (key.msaaColorSource) {
         macros[@"MSAA_COLOR_SOURCE"] = @"1";
     }
-    if (key.msaaDepthSource) {
-        macros[@"MSAA_DEPTH_SOURCE"] = @"1";
-    }
     if (key.sources3D) {
         macros[@"SOURCES_3D"] = @"1";
     }

filament/backend/src/metal/MetalBufferPool.h

@@ -73,7 +73,10 @@ private:
     std::unordered_set<MetalBufferPoolEntry const*> mUsedStages;
 
     // Store the current "time" (really just a frame count) and LRU eviction parameters.
-    uint64_t mCurrentFrame = 0;
+    // An atomic is necessary as mCurrentFrame is incremented in gc() (called on
+    // the driver thread) and read from acquireBuffer() and releaseBuffer(),
+    // which may be called on non-driver threads.
+    std::atomic<uint64_t> mCurrentFrame = 0;
     static constexpr uint32_t TIME_BEFORE_EVICTION = 10;
 };
 

filament/backend/src/metal/MetalContext.h

@@ -142,7 +142,10 @@ struct MetalContext {
     // Fences, only supported on macOS 10.14 and iOS 12 and above.
     API_AVAILABLE(macos(10.14), ios(12.0))
     MTLSharedEventListener* eventListener = nil;
-    uint64_t signalId = 1;
+    // signalId is incremented in the MetalFence constructor, which is called on
+    // both the driver (MetalTimerQueryFence::beginTimeElapsedQuery) and main
+    // threads (in createFenceS), so an atomic is necessary.
+    std::atomic<uint64_t> signalId = 1;
 
     MetalTimerQueryInterface* timerQueryImpl;
 

filament/backend/src/metal/MetalDriver.h

@@ -28,6 +28,10 @@
 #include <utils/Log.h>
 #include <utils/debug.h>
 
+#include <functional>
+#include <mutex>
+#include <vector>
+
 namespace filament {
 namespace backend {
 
@@ -52,13 +56,13 @@ class MetalDriver final : public DriverBase {
 
 public:
     static Driver* create(MetalPlatform* platform, const Platform::DriverConfig& driverConfig);
+    void runAtNextTick(const std::function<void()>& fn) noexcept;
 
 private:
 
     friend class MetalSwapChain;
 
     MetalPlatform& mPlatform;
-
     MetalContext* mContext;
 
     ShaderModel getShaderModel() const noexcept final;
@@ -66,6 +70,13 @@ private:
     // Overrides the default implementation by wrapping the call to fn in an @autoreleasepool block.
     void execute(std::function<void(void)> const& fn) noexcept final;
 
+    /*
+     * Tasks run regularly on the driver thread.
+     */
+    void executeTickOps() noexcept;
+    std::vector<std::function<void()>> mTickOps;
+    std::mutex mTickOpsLock;
+
     /*
      * Driver interface
      */

filament/backend/src/metal/MetalDriver.mm

@@ -166,6 +166,7 @@ MetalDriver::~MetalDriver() noexcept {
 }
 
 void MetalDriver::tick(int) {
+    executeTickOps();
 }
 
 void MetalDriver::beginFrame(int64_t monotonic_clock_ns, uint32_t frameId) {
@@ -583,6 +584,8 @@ void MetalDriver::terminate() {
     // This must be done before calling bufferPool->reset() to ensure no buffers are in flight.
     finish();
 
+    executeTickOps();
+
     mContext->bufferPool->reset();
     mContext->commandQueue = nil;
 
@@ -727,6 +730,10 @@ bool MetalDriver::isParallelShaderCompileSupported() {
     return true;
 }
 
+bool MetalDriver::isDepthStencilResolveSupported() {
+    return false;
+}
+
 bool MetalDriver::isWorkaroundNeeded(Workaround workaround) {
     switch (workaround) {
         case Workaround::SPLIT_EASU:
@@ -1257,11 +1264,9 @@ void MetalDriver::readPixels(Handle<HwRenderTarget> src, uint32_t x, uint32_t y,
     args.source.level = miplevel;
     args.source.region = MTLRegionMake2D(0, 0, srcTexture.width >> miplevel, srcTexture.height >> miplevel);
     args.source.texture = srcTexture;
-    args.source.type = MetalBlitter::BlitArgs::Attachment::Type::COLOR;
     args.destination.level = 0;
     args.destination.region = MTLRegionMake2D(0, 0, readPixelsTexture.width, readPixelsTexture.height);
     args.destination.texture = readPixelsTexture;
-    args.destination.type = MetalBlitter::BlitArgs::Attachment::Type::COLOR;
 
     mContext->blitter->blit(getPendingCommandBuffer(mContext), args, "readPixels blit");
 
@@ -1319,6 +1324,12 @@ void MetalDriver::resolve(
     ASSERT_PRECONDITION(srcTexture->format == dstTexture->format,
             "src and dst texture format don't match");
 
+    ASSERT_PRECONDITION(!isDepthFormat(srcTexture->format),
+            "can't resolve depth formats");
+
+    ASSERT_PRECONDITION(!isStencilFormat(srcTexture->format),
+            "can't resolve stencil formats");
+
     ASSERT_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST),
             "texture doesn't have BLIT_DST");
 
@@ -1378,9 +1389,7 @@ void MetalDriver::blit(
     // FIXME: we shouldn't need to know the type here. This is an artifact of using the old API.
 
     args.source.texture = srcTexture->getMtlTextureForRead();
-    args.source.type = MetalBlitter::getFormatType(srcTexture->format);
     args.destination.texture = dstTexture->getMtlTextureForWrite();
-    args.destination.type = MetalBlitter::getFormatType(dstTexture->format);
     args.source.level = srcLevel;
     args.source.slice = srcLayer;
     args.destination.level = dstLevel;
@@ -1435,13 +1444,11 @@ void MetalDriver::blitDEPRECATED(TargetBufferFlags buffers,
         args.source.texture = srcColorAttachment.getTexture();
         args.source.level = srcColorAttachment.level;
         args.source.slice = srcColorAttachment.layer;
-        args.source.type = MetalBlitter::FormatType::COLOR;
 
         args.destination.region = dstTarget->getRegionFromClientRect(dstRect);
         args.destination.texture = dstColorAttachment.getTexture();
         args.destination.level = dstColorAttachment.level;
         args.destination.slice = dstColorAttachment.layer;
-        args.destination.type = MetalBlitter::FormatType::COLOR;
 
         mContext->blitter->blit(getPendingCommandBuffer(mContext), args, "blitDEPRECATED");
     }
@@ -1581,7 +1588,7 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph, uint32_t
         stencilPixelFormat = stencilAttachment.getPixelFormat();
         assert_invariant(isMetalFormatStencil(stencilPixelFormat));
     }
-    MetalPipelineState pipelineState {
+    MetalPipelineState const pipelineState {
         .vertexFunction = vertex,
         .fragmentFunction = fragment,
         .vertexDescription = primitive->vertexDescription,
@@ -1599,13 +1606,13 @@ void MetalDriver::draw(PipelineState ps, Handle<HwRenderPrimitive> rph, uint32_t
         .stencilAttachmentPixelFormat = stencilPixelFormat,
         .sampleCount = mContext->currentRenderTarget->getSamples(),
         .blendState = BlendState {
-            .blendingEnabled = rs.hasBlending(),
-            .rgbBlendOperation = getMetalBlendOperation(rs.blendEquationRGB),
             .alphaBlendOperation = getMetalBlendOperation(rs.blendEquationAlpha),
-            .sourceRGBBlendFactor = getMetalBlendFactor(rs.blendFunctionSrcRGB),
-            .sourceAlphaBlendFactor = getMetalBlendFactor(rs.blendFunctionSrcAlpha),
+            .rgbBlendOperation = getMetalBlendOperation(rs.blendEquationRGB),
+            .destinationAlphaBlendFactor = getMetalBlendFactor(rs.blendFunctionDstAlpha),
             .destinationRGBBlendFactor = getMetalBlendFactor(rs.blendFunctionDstRGB),
-            .destinationAlphaBlendFactor = getMetalBlendFactor(rs.blendFunctionDstAlpha)
+            .sourceAlphaBlendFactor = getMetalBlendFactor(rs.blendFunctionSrcAlpha),
+            .sourceRGBBlendFactor = getMetalBlendFactor(rs.blendFunctionSrcRGB),
+            .blendingEnabled = rs.hasBlending(),
         },
         .colorWrite = rs.colorWrite
     };
@@ -1904,6 +1911,21 @@ void MetalDriver::enumerateBoundBuffers(BufferObjectBinding bindingType,
 void MetalDriver::resetState(int) {
 }
 
+void MetalDriver::runAtNextTick(const std::function<void()>& fn) noexcept {
+    std::lock_guard<std::mutex> const lock(mTickOpsLock);
+    mTickOps.push_back(fn);
+}
+
+void MetalDriver::executeTickOps() noexcept {
+    std::vector<std::function<void()>> ops;
+    mTickOpsLock.lock();
+    std::swap(ops, mTickOps);
+    mTickOpsLock.unlock();
+    for (const auto& f : ops) {
+        f();
+    }
+}
+
 // explicit instantiation of the Dispatcher
 template class ConcreteDispatcher<MetalDriver>;
 

filament/backend/src/metal/MetalHandles.mm

@@ -245,13 +245,30 @@ void MetalSwapChain::present() {
     }
 }
 
+struct PresentDrawableData {
+    void* drawable = nullptr;
+    MetalDriver* driver = nullptr;
+};
+
 void presentDrawable(bool presentFrame, void* user) {
-    // CFBridgingRelease here is used to balance the CFBridgingRetain inside of acquireDrawable.
-    id<CAMetalDrawable> drawable = (id<CAMetalDrawable>) CFBridgingRelease(user);
+    auto* presentDrawableData = static_cast<PresentDrawableData*>(user);
+
+    // CFBridgingRelease here is used to balance the CFBridgingRetain inside acquireDrawable.
+    id<CAMetalDrawable> drawable =
+            (id<CAMetalDrawable>)CFBridgingRelease(presentDrawableData->drawable);
     if (presentFrame) {
         [drawable present];
     }
-    // The drawable will be released here when the "drawable" variable goes out of scope.
+
+    // Schedule the drawable destruction on the driver thread.
+    void* voidDrawable = (void*) CFBridgingRetain(drawable);
+    MetalDriver* driver = presentDrawableData->driver;
+    driver->runAtNextTick([voidDrawable]() {
+        // The drawable is released here.
+        CFBridgingRelease(voidDrawable);
+    });
+
+    delete presentDrawableData;
 }
 
 void MetalSwapChain::scheduleFrameScheduledCallback() {
@@ -266,13 +283,15 @@ void MetalSwapChain::scheduleFrameScheduledCallback() {
     // capture the _this_ pointer (MetalSwapChain*) instead of the drawable.
     id<CAMetalDrawable> d = drawable;
     void* userData = frameScheduledUserData;
+    MetalDriver* driver = context.driver;
     [getPendingCommandBuffer(&context) addScheduledHandler:^(id<MTLCommandBuffer> cb) {
         // CFBridgingRetain is used here to give the drawable a +1 retain count before
         // casting it to a void*.
-        PresentCallable callable(presentDrawable, (void*) CFBridgingRetain(d));
-        dispatch_async(dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT, 0), ^{
-            callback(callable, userData);
-        });
+        auto* presentDrawableData = new PresentDrawableData;
+        presentDrawableData->drawable = (void*) CFBridgingRetain(d);
+        presentDrawableData->driver = driver;
+        PresentCallable callable(presentDrawable, (void*) presentDrawableData);
+        callback(callable, userData);
     }];
 }
 
@@ -801,12 +820,10 @@ void MetalTexture::loadWithBlit(uint32_t level, uint32_t slice, MTLRegion region
     args.source.slice = 0;
     args.source.region = sourceRegion;
     args.source.texture = stagingTexture;
-    args.source.type = MetalBlitter::FormatType::COLOR;
     args.destination.level = level;
     args.destination.slice = slice;
     args.destination.region = region;
     args.destination.texture = destinationTexture;
-    args.destination.type = MetalBlitter::FormatType::COLOR;
     context.blitter->blit(getPendingCommandBuffer(&context), args, "Texture upload blit");
 }
 

filament/backend/src/metal/MetalShaderCompiler.mm

@@ -103,10 +103,19 @@ void MetalShaderCompiler::terminate() noexcept {
         NSString* objcSource = [[NSString alloc] initWithBytes:source.data()
                                                         length:source.size() - 1
                                                       encoding:NSUTF8StringEncoding];
+
+        // By default, Metal uses the most recent language version.
+        MTLCompileOptions* options = [MTLCompileOptions new];
+
+        // Disable Fast Math optimizations.
+        // This ensures that operations adhere to IEEE standards for floating-point arithmetic,
+        // which is crucial for half precision floats in scenarios where fast math optimizations
+        // lead to inaccuracies, such as in handling special values like NaN or Infinity.
+        options.fastMathEnabled = NO;
+
         NSError* error = nil;
-        // When options is nil, Metal uses the most recent language version available.
         id<MTLLibrary> library = [device newLibraryWithSource:objcSource
-                                                      options:nil
+                                                      options:options
                                                         error:&error];
         if (library == nil) {
             if (error) {

filament/backend/src/noop/NoopDriver.cpp

@@ -182,6 +182,10 @@ bool NoopDriver::isParallelShaderCompileSupported() {
     return false;
 }
 
+bool NoopDriver::isDepthStencilResolveSupported() {
+    return true;
+}
+
 bool NoopDriver::isWorkaroundNeeded(Workaround) {
     return false;
 }

filament/backend/src/opengl/OpenGLContext.cpp

@@ -274,6 +274,7 @@ void OpenGLContext::setDefaultState() noexcept {
     glHint(GL_FRAGMENT_SHADER_DERIVATIVE_HINT, GL_NICEST);
 #endif
 
+#if !defined(__EMSCRIPTEN__)
     if (ext.EXT_clip_control) {
 #if defined(BACKEND_OPENGL_VERSION_GL)
         glClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE);
@@ -281,6 +282,7 @@ void OpenGLContext::setDefaultState() noexcept {
         glClipControlEXT(GL_LOWER_LEFT_EXT, GL_ZERO_TO_ONE_EXT);
 #endif
     }
+#endif
 
     if (ext.EXT_clip_cull_distance) {
         glEnable(GL_CLIP_DISTANCE0);
@@ -309,7 +311,9 @@ void OpenGLContext::initProcs(Procs* procs,
 #   ifdef BACKEND_OPENGL_VERSION_GL
     procs->getQueryObjectui64v = glGetQueryObjectui64v; // only core in GL
 #   elif defined(GL_EXT_disjoint_timer_query)
-    procs->getQueryObjectui64v = glGetQueryObjectui64vEXT;
+#       ifndef __EMSCRIPTEN__
+            procs->getQueryObjectui64v = glGetQueryObjectui64vEXT;
+#       endif
 #   endif // BACKEND_OPENGL_VERSION_GL
 
     // core in ES 3.0 and GL 4.3
@@ -321,6 +325,7 @@ void OpenGLContext::initProcs(Procs* procs,
 
 #ifdef BACKEND_OPENGL_VERSION_GLES
 #   ifndef IOS // IOS is guaranteed to have ES3.x
+#       ifndef __EMSCRIPTEN__
     if (UTILS_UNLIKELY(major == 2)) {
         // Runtime OpenGL version is ES 2.x
         if (UTILS_LIKELY(ext.OES_vertex_array_object)) {
@@ -348,6 +353,7 @@ void OpenGLContext::initProcs(Procs* procs,
 
         procs->maxShaderCompilerThreadsKHR = glMaxShaderCompilerThreadsKHR;
     }
+#       endif // __EMSCRIPTEN__
 #   endif // IOS
 #else
     procs->maxShaderCompilerThreadsKHR = glMaxShaderCompilerThreadsARB;
@@ -580,17 +586,23 @@ void OpenGLContext::initExtensionsGLES(Extensions* ext, GLint major, GLint minor
     // figure out and initialize the extensions we need
     using namespace std::literals;
     ext->APPLE_color_buffer_packed_float = exts.has("GL_APPLE_color_buffer_packed_float"sv);
+#ifndef __EMSCRIPTEN__
     ext->EXT_clip_control = exts.has("GL_EXT_clip_control"sv);
+#endif
     ext->EXT_clip_cull_distance = exts.has("GL_EXT_clip_cull_distance"sv);
     ext->EXT_color_buffer_float = exts.has("GL_EXT_color_buffer_float"sv);
     ext->EXT_color_buffer_half_float = exts.has("GL_EXT_color_buffer_half_float"sv);
+#ifndef __EMSCRIPTEN__
     ext->EXT_debug_marker = exts.has("GL_EXT_debug_marker"sv);
+#endif
     ext->EXT_discard_framebuffer = exts.has("GL_EXT_discard_framebuffer"sv);
+#ifndef __EMSCRIPTEN__
     ext->EXT_disjoint_timer_query = exts.has("GL_EXT_disjoint_timer_query"sv);
     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);
+#endif
     ext->EXT_shader_framebuffer_fetch = exts.has("GL_EXT_shader_framebuffer_fetch"sv);
-#if !defined(__EMSCRIPTEN__)
+#ifndef __EMSCRIPTEN__
     ext->EXT_texture_compression_etc2 = true;
 #endif
     ext->EXT_texture_compression_s3tc = exts.has("GL_EXT_texture_compression_s3tc"sv);

filament/backend/src/opengl/OpenGLDriver.cpp

@@ -537,7 +537,7 @@ void OpenGLDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {
 
 
     if (UTILS_UNLIKELY(mContext.isES2())) {
-        // Here we patch the specification constants to enable or not the rec709 output
+        // Here we patch the specialization constants to enable or not the rec709 output
         // color space emulation in this program. Obviously, the backend shouldn't know about
         // specific spec-constants, so we need to handle failures gracefully. This cannot be
         // done at Material creation time because only the backend has access to
@@ -1064,6 +1064,7 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
         }
         CHECK_GL_ERROR(utils::slog.e)
     } else
+#ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_multisampled_render_to_texture
         // EXT_multisampled_render_to_texture only support GL_COLOR_ATTACHMENT0
     if (!attachmentTypeNotSupportedByMSRTT && (t->depth <= 1)
@@ -1076,7 +1077,7 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
         // This extension only exists on OpenGL ES.
         gl.bindFramebuffer(GL_FRAMEBUFFER, rt->gl.fbo);
         if (any(t->usage & TextureUsage::SAMPLEABLE)) {
-            glext::glFramebufferTexture2DMultisampleEXT(GL_FRAMEBUFFER,
+            glFramebufferTexture2DMultisampleEXT(GL_FRAMEBUFFER,
                     attachment, target, t->gl.id, binfo.level, rt->gl.samples);
         } else {
             glFramebufferRenderbuffer(GL_FRAMEBUFFER, attachment,
@@ -1084,7 +1085,8 @@ void OpenGLDriver::framebufferTexture(TargetBufferInfo const& binfo,
         }
         CHECK_GL_ERROR(utils::slog.e)
     } else
-#endif
+#endif // GL_EXT_multisampled_render_to_texture
+#endif // __EMSCRIPTEN__
     if (!any(t->usage & TextureUsage::SAMPLEABLE) && t->samples > 1) {
         assert_invariant(rt->gl.samples > 1);
         assert_invariant(glIsRenderbuffer(t->gl.id));
@@ -1183,6 +1185,7 @@ void OpenGLDriver::renderBufferStorage(GLuint rbo, GLenum internalformat, uint32
         uint32_t height, uint8_t samples) const noexcept {
     glBindRenderbuffer(GL_RENDERBUFFER, rbo);
     if (samples > 1) {
+#ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_multisampled_render_to_texture
         auto& gl = mContext;
         if (gl.ext.EXT_multisampled_render_to_texture ||
@@ -1190,7 +1193,8 @@ void OpenGLDriver::renderBufferStorage(GLuint rbo, GLenum internalformat, uint32
             glext::glRenderbufferStorageMultisampleEXT(GL_RENDERBUFFER,
                     samples, internalformat, width, height);
         } else
-#endif
+#endif // GL_EXT_multisampled_render_to_texture
+#endif // __EMSCRIPTEN__
         {
 #ifndef FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2
             glRenderbufferStorageMultisample(GL_RENDERBUFFER,
@@ -1906,6 +1910,10 @@ bool OpenGLDriver::isParallelShaderCompileSupported() {
     return mShaderCompilerService.isParallelShaderCompileSupported();
 }
 
+bool OpenGLDriver::isDepthStencilResolveSupported() {
+    return true;
+}
+
 bool OpenGLDriver::isWorkaroundNeeded(Workaround workaround) {
     switch (workaround) {
         case Workaround::SPLIT_EASU:
@@ -2283,6 +2291,7 @@ void OpenGLDriver::setTextureData(GLTexture* t, uint32_t level,
         PixelBufferDescriptor&& p) {
     auto& gl = mContext;
 
+    assert_invariant(t != nullptr);
     assert_invariant(xoffset + width <= std::max(1u, t->width >> level));
     assert_invariant(yoffset + height <= std::max(1u, t->height >> level));
     assert_invariant(t->samples <= 1);
@@ -2981,16 +2990,18 @@ GLuint OpenGLDriver::getSamplerSlow(SamplerParams params) const noexcept {
 #endif
 
 void OpenGLDriver::insertEventMarker(char const* string, uint32_t len) {
+#ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_debug_marker
     auto& gl = mContext;
     if (gl.ext.EXT_debug_marker) {
         glInsertEventMarkerEXT(GLsizei(len ? len : strlen(string)), string);
     }
 #endif
+#endif
 }
 
 void OpenGLDriver::pushGroupMarker(char const* string, uint32_t len) {
-
+#ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_debug_marker
 #if DEBUG_MARKER_LEVEL & DEBUG_MARKER_OPENGL
     if (UTILS_LIKELY(mContext.ext.EXT_debug_marker)) {
@@ -3003,9 +3014,11 @@ void OpenGLDriver::pushGroupMarker(char const* string, uint32_t len) {
     SYSTRACE_CONTEXT();
     SYSTRACE_NAME_BEGIN(string);
 #endif
+#endif
 }
 
 void OpenGLDriver::popGroupMarker(int) {
+#ifndef __EMSCRIPTEN__
 #ifdef GL_EXT_debug_marker
 #if DEBUG_MARKER_LEVEL & DEBUG_MARKER_OPENGL
     if (UTILS_LIKELY(mContext.ext.EXT_debug_marker)) {
@@ -3018,6 +3031,7 @@ void OpenGLDriver::popGroupMarker(int) {
     SYSTRACE_CONTEXT();
     SYSTRACE_NAME_END();
 #endif
+#endif
 }
 
 void OpenGLDriver::startCapture(int) {

filament/backend/src/opengl/ShaderCompilerService.cpp

@@ -547,7 +547,8 @@ void ShaderCompilerService::compileShaders(OpenGLContext& context,
             // add support for ARB_shading_language_packing if needed
             auto const packingFunctions = process_ARB_shading_language_packing(context);
 
-            // split shader source, so we can insert the specification constants and the packing functions
+            // split shader source, so we can insert the specialization constants and the packing
+            // functions
             auto const [prolog, body] = splitShaderSource(source);
 
             const std::array<const char*, 4> sources = {

filament/backend/src/opengl/gl_headers.cpp

@@ -28,6 +28,7 @@ static void getProcAddress(T& pfn, const char* name) noexcept {
 }
 
 namespace glext {
+#ifndef __EMSCRIPTEN__
 #ifdef GL_OES_EGL_image
 PFNGLEGLIMAGETARGETTEXTURE2DOESPROC glEGLImageTargetTexture2DOES;
 #endif
@@ -72,10 +73,11 @@ PFNGLMAXSHADERCOMPILERTHREADSKHRPROC glMaxShaderCompilerThreadsKHR;
 // use getProcAddress for ES3.1 and above entry points.
 PFNGLDISPATCHCOMPUTEPROC glDispatchCompute;
 #endif
-
 static std::once_flag sGlExtInitialized;
+#endif // __EMSCRIPTEN__
 
 void importGLESExtensionsEntryPoints() {
+#ifndef __EMSCRIPTEN__
     std::call_once(sGlExtInitialized, +[]() {
 #ifdef GL_OES_EGL_image
     getProcAddress(glEGLImageTargetTexture2DOES, "glEGLImageTargetTexture2DOES");
@@ -119,6 +121,7 @@ void importGLESExtensionsEntryPoints() {
         getProcAddress(glDispatchCompute, "glDispatchCompute");
 #endif
     });
+#endif // __EMSCRIPTEN__
 }
 
 } // namespace glext

filament/backend/src/opengl/gl_headers.h

@@ -112,6 +112,7 @@ namespace glext {
 // it is currently called from PlatformEGL.
 void importGLESExtensionsEntryPoints();
 
+#ifndef __EMSCRIPTEN__
 #ifdef GL_OES_EGL_image
 extern PFNGLEGLIMAGETARGETTEXTURE2DOESPROC glEGLImageTargetTexture2DOES;
 #endif
@@ -153,6 +154,7 @@ extern PFNGLMAXSHADERCOMPILERTHREADSKHRPROC glMaxShaderCompilerThreadsKHR;
 #if defined(__ANDROID__) && !defined(FILAMENT_SILENCE_NOT_SUPPORTED_BY_ES2)
 extern PFNGLDISPATCHCOMPUTEPROC glDispatchCompute;
 #endif
+#endif // __EMSCRIPTEN__
 } // namespace glext
 
 using namespace glext;

filament/backend/src/ostream.cpp

@@ -455,7 +455,7 @@ io::ostream& operator<<(io::ostream& out, filament::backend::Viewport const& vie
 
 io::ostream& operator<<(io::ostream& out, TargetBufferFlags flags) {
     // TODO: implement decoding of enum
-    out << uint8_t(flags);
+    out << uint32_t(flags);
     return out;
 }
 

filament/backend/src/vulkan/VulkanBlitter.cpp

@@ -154,18 +154,14 @@ struct BlitterUniforms {
 
 }// anonymous namespace
 
-VulkanBlitter::VulkanBlitter(VulkanStagePool& stagePool, VulkanPipelineCache& pipelineCache,
-        VulkanFboCache& fboCache, VulkanSamplerCache& samplerCache) noexcept
-    : mStagePool(stagePool), mPipelineCache(pipelineCache), mFramebufferCache(fboCache),
-      mSamplerCache(samplerCache) {}
+VulkanBlitter::VulkanBlitter() noexcept = default;
 
 void VulkanBlitter::initialize(VkPhysicalDevice physicalDevice, VkDevice device,
-        VmaAllocator allocator, VulkanCommands* commands, VulkanTexture* emptyTexture) noexcept {
+        VmaAllocator allocator, VulkanCommands* commands) noexcept {
     mPhysicalDevice = physicalDevice;
     mDevice = device;
     mAllocator = allocator;
     mCommands = commands;
-    mEmptyTexture = emptyTexture;
 }
 
 void VulkanBlitter::resolve(VulkanAttachment dst, VulkanAttachment src) {
@@ -173,14 +169,7 @@ void VulkanBlitter::resolve(VulkanAttachment dst, VulkanAttachment src) {
     // src and dst should have the same aspect here
     VkImageAspectFlags const aspect = src.texture->getImageAspect();
 
-    if (aspect & VK_IMAGE_ASPECT_DEPTH_BIT &&
-        src.texture->samples > 1 && dst.texture->samples == 1) {
-        // we're resolving the depth here, which is not supported by the blitter in vulkan
-        // I think, eventually we want to do that inside filament, instead.
-        // FIXME: remove this once filament does it
-        resolveSlowDepth(src, dst);
-        return;
-    }
+    assert_invariant(!(aspect & VK_IMAGE_ASPECT_DEPTH_BIT));
 
 #if FVK_ENABLED(FVK_DEBUG_BLIT_FORMAT)
     VkPhysicalDevice const gpu = mPhysicalDevice;
@@ -231,272 +220,6 @@ void VulkanBlitter::blit(VkFilter filter,
 }
 
 void VulkanBlitter::terminate() noexcept {
-    if (mDepthResolveProgram) {
-        delete mDepthResolveProgram;
-        mDepthResolveProgram = nullptr;
-    }
-
-    if (mTriangleBuffer) {
-        delete mTriangleBuffer;
-        mTriangleBuffer = nullptr;
-    }
-
-    if (mParamsBuffer) {
-        delete mParamsBuffer;
-        mParamsBuffer = nullptr;
-    }
-}
-
-// If we created these shader modules in the constructor, the device might not be ready yet.
-// It is easier to do lazy initialization, which can also improve load time.
-void VulkanBlitter::lazyInit() noexcept {
-    if (mDepthResolveProgram != nullptr) {
-        return;
-    }
-    assert_invariant(mDevice);
-
-    auto decode = [device = mDevice](const uint8_t* compressed, int compressedSize) {
-        const size_t spirvSize = smolv::GetDecodedBufferSize(compressed, compressedSize);
-        FixedCapacityVector<uint8_t> spirv(spirvSize);
-        smolv::Decode(compressed, compressedSize, spirv.data(), spirvSize);
-
-        VkShaderModuleCreateInfo moduleInfo = {};
-        moduleInfo.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
-        moduleInfo.codeSize = spirvSize;
-        moduleInfo.pCode = (uint32_t*) spirv.data();
-
-        VkShaderModule result = VK_NULL_HANDLE;
-        vkCreateShaderModule(device, &moduleInfo, VKALLOC, &result);
-        assert_invariant(result);
-
-        return result;
-    };
-
-    VkShaderModule vertexShader = decode(VKSHADERS_BLITDEPTHVS_DATA, VKSHADERS_BLITDEPTHVS_SIZE);
-    VkShaderModule fragmentShader = decode(VKSHADERS_BLITDEPTHFS_DATA, VKSHADERS_BLITDEPTHFS_SIZE);
-
-    // Allocate one anonymous sampler at slot 0.
-    VulkanProgram::CustomSamplerInfoList const samplers = {
-            { 0, 0, ShaderStageFlags::FRAGMENT },
-    };
-    mDepthResolveProgram = new(std::nothrow) VulkanProgram(mDevice, vertexShader, fragmentShader, samplers);
-
-#if FVK_ENABLED(FVK_DEBUG_BLITTER)
-    utils::slog.d << "Created Shader Module for VulkanBlitter "
-                  << "shaders = (" << vertexShader << ", " << fragmentShader << ")"
-                  << utils::io::endl;
-#endif
-
-    static const float kTriangleVertices[] = {
-        -1.0f, -1.0f,
-        +1.0f, -1.0f,
-        -1.0f, +1.0f,
-        +1.0f, +1.0f,
-    };
-
-    VulkanCommandBuffer const& commands = mCommands->get();
-    VkCommandBuffer const cmdbuffer = commands.buffer();
-
-    mTriangleBuffer = new(std::nothrow) VulkanBuffer(mAllocator, mStagePool, VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
-            sizeof(kTriangleVertices));
-
-    mTriangleBuffer->loadFromCpu(cmdbuffer, kTriangleVertices, 0, sizeof(kTriangleVertices));
-
-    mParamsBuffer = new(std::nothrow) VulkanBuffer(mAllocator, mStagePool, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT,
-            sizeof(BlitterUniforms));
-}
-
-// At a high level, the procedure for resolving depth looks like this:
-// 1. Load uniforms
-// 2. Begin render pass
-// 3. Draw a big triangle
-// 4. End render pass.
-void VulkanBlitter::resolveSlowDepth(VulkanAttachment src, VulkanAttachment dst) {
-    lazyInit();
-
-    const VkOffset3D dstRect[2] = {
-            { 0, 0, 0 },
-            { (int32_t)dst.texture->width, (int32_t)dst.texture->height, 1 }};
-
-    VulkanCommandBuffer* commands = &mCommands->get();
-    VkCommandBuffer const cmdbuffer = commands->buffer();
-    commands->acquire(src.texture);
-    commands->acquire(dst.texture);
-
-    BlitterUniforms const uniforms = {
-            .sampleCount = src.texture->samples,
-            .inverseSampleCount = 1.0f / float(src.texture->samples),
-    };
-    mParamsBuffer->loadFromCpu(cmdbuffer, &uniforms, 0, sizeof(uniforms));
-
-    VkImageAspectFlags const aspect = VK_IMAGE_ASPECT_DEPTH_BIT;
-
-    // We will transition the src into sampler layout and also keep it in sampler layout for
-    // consistency.
-    VulkanLayout const samplerLayout = VulkanLayout::DEPTH_SAMPLER;
-
-    // BEGIN RENDER PASS
-    // -----------------
-
-    const VulkanFboCache::RenderPassKey rpkey = {
-        .initialColorLayoutMask = 0,
-        .initialDepthLayout = VulkanLayout::UNDEFINED,
-        .renderPassDepthLayout = samplerLayout,
-        .finalDepthLayout = samplerLayout,
-        .depthFormat = dst.getFormat(),
-        .clear = {},
-        .discardStart = TargetBufferFlags::DEPTH,
-        .samples = 1,
-    };
-
-    const VkRenderPass renderPass = mFramebufferCache.getRenderPass(rpkey);
-    mPipelineCache.bindRenderPass(renderPass, 0);
-
-    const VulkanFboCache::FboKey fbkey {
-        .renderPass = renderPass,
-        .width = (uint16_t) (dst.texture->width >> dst.level),
-        .height = (uint16_t) (dst.texture->height >> dst.level),
-        .layers = 1,
-        .samples = rpkey.samples,
-        .color = {},
-        .resolve = {},
-        .depth = dst.getImageView(aspect),
-    };
-    const VkFramebuffer vkfb = mFramebufferCache.getFramebuffer(fbkey);
-
-    VkRenderPassBeginInfo renderPassInfo {
-        .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
-        .renderPass = renderPass,
-        .framebuffer = vkfb,
-    };
-    renderPassInfo.renderArea.offset.x = dstRect[0].x;
-    renderPassInfo.renderArea.offset.y = dstRect[0].y;
-    renderPassInfo.renderArea.extent.width = dstRect[1].x - dstRect[0].x;
-    renderPassInfo.renderArea.extent.height = dstRect[1].y - dstRect[0].y;
-
-    // We need to transition the source into a sampler since it'll be sampled in the shader.
-    const VkImageSubresourceRange srcRange = {
-            .aspectMask = aspect,
-            .baseMipLevel = src.level,
-            .levelCount = 1,
-            .baseArrayLayer = src.layer,
-            .layerCount = 1,
-    };
-    src.texture->transitionLayout(cmdbuffer, srcRange, samplerLayout);
-
-    vkCmdBeginRenderPass(cmdbuffer, &renderPassInfo, VK_SUBPASS_CONTENTS_INLINE);
-
-    VkViewport const viewport = {
-        .x = (float) dstRect[0].x,
-        .y = (float) dstRect[0].y,
-        .width = (float) renderPassInfo.renderArea.extent.width,
-        .height = (float) renderPassInfo.renderArea.extent.height,
-        .minDepth = 0.0f,
-        .maxDepth = 1.0f
-    };
-
-    vkCmdSetViewport(cmdbuffer, 0, 1, &viewport);
-
-    // DRAW THE TRIANGLE
-    // -----------------
-
-    mPipelineCache.bindProgram(mDepthResolveProgram);
-    mPipelineCache.bindPrimitiveTopology(VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP);
-
-    auto vkraster = mPipelineCache.getCurrentRasterState();
-    vkraster.depthWriteEnable = true;
-    vkraster.depthCompareOp = SamplerCompareFunc::A;
-    vkraster.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT,
-    vkraster.alphaToCoverageEnable = false,
-    vkraster.blendEnable = VK_FALSE,
-    vkraster.srcColorBlendFactor = VK_BLEND_FACTOR_ONE,
-    vkraster.dstColorBlendFactor = VK_BLEND_FACTOR_ZERO,
-    vkraster.colorBlendOp = BlendEquation::ADD,
-    vkraster.srcAlphaBlendFactor = VK_BLEND_FACTOR_ONE,
-    vkraster.dstAlphaBlendFactor = VK_BLEND_FACTOR_ZERO,
-    vkraster.alphaBlendOp = BlendEquation::ADD,
-    vkraster.colorWriteMask = (VkColorComponentFlags) 0,
-    vkraster.cullMode = VK_CULL_MODE_NONE;
-    vkraster.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE;
-    vkraster.depthBiasEnable = VK_FALSE;
-    vkraster.colorTargetCount = 0;
-    mPipelineCache.bindRasterState(vkraster);
-
-    VkBuffer buffers[1] = {};
-    VkDeviceSize offsets[1] = {};
-    buffers[0] = mTriangleBuffer->getGpuBuffer();
-    VkVertexInputAttributeDescription const attribDesc = {
-            .location = 0,
-            .binding = 0,
-            .format = VK_FORMAT_R32G32_SFLOAT,
-    };
-    VkVertexInputBindingDescription const bufferDesc = {
-            .binding = 0,
-            .stride = sizeof(float) * 2,
-    };
-    mPipelineCache.bindVertexArray(&attribDesc, &bufferDesc, 1);
-
-    // Select nearest filtering and clamp_to_edge.
-    VkSampler vksampler = mSamplerCache.getSampler({});
-
-    VkDescriptorImageInfo samplers[VulkanPipelineCache::SAMPLER_BINDING_COUNT];
-    VulkanTexture* textures[VulkanPipelineCache::SAMPLER_BINDING_COUNT] = {nullptr};
-    for (auto& sampler : samplers) {
-        sampler = {
-            .sampler = vksampler,
-            .imageView = mEmptyTexture->getPrimaryImageView(),
-            .imageLayout = ImgUtil::getVkLayout(VulkanLayout::READ_WRITE),
-        };
-    }
-
-    samplers[0] = {
-        .sampler = vksampler,
-        .imageView = src.getImageView(VK_IMAGE_ASPECT_DEPTH_BIT),
-        .imageLayout = ImgUtil::getVkLayout(samplerLayout),
-    };
-    textures[0] = src.texture;
-
-    mPipelineCache.bindSamplers(samplers, textures,
-            VulkanPipelineCache::getUsageFlags(0, ShaderStageFlags::FRAGMENT));
-
-    auto previousUbo = mPipelineCache.getUniformBufferBinding(0);
-    mPipelineCache.bindUniformBuffer(0, mParamsBuffer->getGpuBuffer());
-
-    if (!mPipelineCache.bindDescriptors(cmdbuffer)) {
-        assert_invariant(false);
-    }
-
-    const VkRect2D scissor = {
-        .offset = renderPassInfo.renderArea.offset,
-        .extent = renderPassInfo.renderArea.extent,
-    };
-
-    mPipelineCache.bindScissor(cmdbuffer, scissor);
-
-    if (!mPipelineCache.bindPipeline(commands)) {
-        assert_invariant(false);
-    }
-
-    vkCmdBindVertexBuffers(cmdbuffer, 0, 1, buffers, offsets);
-    vkCmdDraw(cmdbuffer, 4, 1, 0, 0);
-
-    // END RENDER PASS
-    // ---------------
-
-    vkCmdEndRenderPass(cmdbuffer);
-
-    VkMemoryBarrier const barrier {
-        .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
-        .srcAccessMask = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
-        .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
-    };
-    VkPipelineStageFlags const srcStageMask = VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-    vkCmdPipelineBarrier(cmdbuffer, srcStageMask,
-            VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT | // <== For Mali
-            VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
-            0, 1, &barrier, 0, nullptr, 0, nullptr);
-
-    mPipelineCache.bindUniformBuffer(0, previousUbo.buffer, previousUbo.offset, previousUbo.size);
 }
 
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanBlitter.h

@@ -32,11 +32,10 @@ struct VulkanProgram;
 
 class VulkanBlitter {
 public:
-    VulkanBlitter(VulkanStagePool& stagePool, VulkanPipelineCache& pipelineCache,
-            VulkanFboCache& fboCache, VulkanSamplerCache& samplerCache) noexcept;
+    VulkanBlitter() noexcept;
 
-    void initialize(VkPhysicalDevice physicalDevice, VkDevice device, VmaAllocator allocator,
-            VulkanCommands* commands, VulkanTexture* emptyTexture) noexcept;
+    void initialize(VkPhysicalDevice physicalDevice, VkDevice device,
+            VmaAllocator allocator, VulkanCommands* commands) noexcept;
 
     void blit(VkFilter filter,
             VulkanAttachment dst, const VkOffset3D* dstRectPair,
@@ -47,24 +46,10 @@ public:
     void terminate() noexcept;
 
 private:
-    void lazyInit() noexcept;
-
-    void resolveSlowDepth(VulkanAttachment src, VulkanAttachment dst);
-
-    VulkanBuffer* mTriangleBuffer = nullptr;
-    VulkanBuffer* mParamsBuffer = nullptr;
-    VulkanProgram* mDepthResolveProgram = nullptr;
-
     UTILS_UNUSED VkPhysicalDevice mPhysicalDevice;
     VkDevice mDevice;
     VmaAllocator mAllocator;
     VulkanCommands* mCommands;
-    VulkanTexture* mEmptyTexture;
-
-    VulkanStagePool& mStagePool;
-    VulkanPipelineCache& mPipelineCache;
-    VulkanFboCache& mFramebufferCache;
-    VulkanSamplerCache& mSamplerCache;
 };
 
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -156,7 +156,6 @@ VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext const& contex
       mResourceManager(&mResourceAllocator),
       mThreadSafeResourceManager(&mResourceAllocator),
       mPipelineCache(&mResourceAllocator),
-      mBlitter(mStagePool, mPipelineCache, mFramebufferCache, mSamplerCache),
       mReadPixels(mPlatform->getDevice()),
       mIsSRGBSwapChainSupported(mPlatform->getCustomization().isSRGBSwapChainSupported) {
 
@@ -206,7 +205,7 @@ VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext const& contex
 
     mPipelineCache.setDummyTexture(mEmptyTexture->getPrimaryImageView());
     mBlitter.initialize(mPlatform->getPhysicalDevice(), mPlatform->getDevice(), mAllocator,
-            mCommands.get(), mEmptyTexture.get());
+            mCommands.get());
 }
 
 VulkanDriver::~VulkanDriver() noexcept = default;
@@ -757,6 +756,7 @@ bool VulkanDriver::isFrameTimeSupported() {
 }
 
 bool VulkanDriver::isAutoDepthResolveSupported() {
+    // TODO: this could be supported with vk 1.2 or VK_KHR_depth_stencil_resolve
     return false;
 }
 
@@ -772,6 +772,11 @@ bool VulkanDriver::isParallelShaderCompileSupported() {
     return false;
 }
 
+bool VulkanDriver::isDepthStencilResolveSupported() {
+    // TODO: apparently it could be supported in core 1.2 and/or with VK_KHR_depth_stencil_resolve
+    return false;
+}
+
 bool VulkanDriver::isWorkaroundNeeded(Workaround workaround) {
     switch (workaround) {
         case Workaround::SPLIT_EASU: {
@@ -1497,6 +1502,12 @@ void VulkanDriver::resolve(
     ASSERT_PRECONDITION(srcTexture->format == dstTexture->format,
             "src and dst texture format don't match");
 
+    ASSERT_PRECONDITION(!isDepthFormat(srcTexture->format),
+            "can't resolve depth formats");
+
+    ASSERT_PRECONDITION(!isStencilFormat(srcTexture->format),
+            "can't resolve stencil formats");
+
     ASSERT_PRECONDITION(any(dstTexture->usage & TextureUsage::BLIT_DST),
             "texture doesn't have BLIT_DST");
 
@@ -1545,7 +1556,7 @@ void VulkanDriver::blit(
     VkOffset3D const srcOffsets[2] = { { srcLeft, srcTop, 0 }, { srcRight, srcBottom, 1 }};
     VkOffset3D const dstOffsets[2] = { { dstLeft, dstTop, 0 }, { dstRight, dstBottom, 1 }};
 
-    // no scallng guaranteed
+    // no scaling guaranteed
     mBlitter.blit(VK_FILTER_NEAREST,
             { .texture = dstTexture, .level = dstLevel, .layer = dstLayer }, dstOffsets,
             { .texture = srcTexture, .level = srcLevel, .layer = srcLayer }, srcOffsets);

filament/include/filament/Options.h

@@ -293,6 +293,7 @@ struct DepthOfFieldOptions {
         MEDIAN
     };
     float cocScale = 1.0f;              //!< circle of confusion scale factor (amount of blur)
+    float cocAspectRatio = 1.0f;        //!< width/height aspect ratio of the circle of confusion (simulate anamorphic lenses)
     float maxApertureDiameter = 0.01f;  //!< maximum aperture diameter in meters (zero to disable rotation)
     bool enabled = false;               //!< enable or disable depth of field effect
     Filter filter = Filter::MEDIAN;     //!< filter to use for filling gaps in the kernel
@@ -401,7 +402,7 @@ struct AmbientOcclusionOptions {
 };
 
 /**
- * Options for Temporal Multi-Sample Anti-aliasing (MSAA)
+ * Options for Multi-Sample Anti-aliasing (MSAA)
  * @see setMultiSampleAntiAliasingOptions()
  */
 struct MultiSampleAntiAliasingOptions {
@@ -424,12 +425,51 @@ struct MultiSampleAntiAliasingOptions {
 
 /**
  * Options for Temporal Anti-aliasing (TAA)
+ * Most TAA parameters are extremely costly to change, as they will trigger the TAA post-process
+ * shaders to be recompiled. These options should be changed or set during initialization.
+ * `filterWidth`, `feedback` and `jitterPattern`, however, can be changed at any time.
+ *
+ * `feedback` of 0.1 effectively accumulates a maximum of 19 samples in steady state.
+ * see "A Survey of Temporal Antialiasing Techniques" by Lei Yang and all for more information.
+ *
  * @see setTemporalAntiAliasingOptions()
  */
 struct TemporalAntiAliasingOptions {
-    float filterWidth = 1.0f;   //!< reconstruction filter width typically between 0 (sharper, aliased) and 1 (smoother)
-    float feedback = 0.04f;     //!< history feedback, between 0 (maximum temporal AA) and 1 (no temporal AA).
+    float filterWidth = 1.0f;   //!< reconstruction filter width typically between 0.2 (sharper, aliased) and 1.5 (smoother)
+    float feedback = 0.12f;     //!< history feedback, between 0 (maximum temporal AA) and 1 (no temporal AA).
+    float lodBias = -1.0f;      //!< texturing lod bias (typically -1 or -2)
     bool enabled = false;       //!< enables or disables temporal anti-aliasing
+    bool upscaling = false;     //!< 4x TAA upscaling. Disables Dynamic Resolution. [BETA]
+
+    enum class BoxType : uint8_t {
+        AABB,           //!< use an AABB neighborhood
+        VARIANCE,       //!< use the variance of the neighborhood (not recommended)
+        AABB_VARIANCE   //!< use both AABB and variance
+    };
+
+    enum class BoxClipping : uint8_t {
+        ACCURATE,       //!< Accurate box clipping
+        CLAMP,          //!< clamping
+        NONE            //!< no rejections (use for debugging)
+    };
+
+    enum class JitterPattern : uint8_t {
+        RGSS_X4,             //!  4-samples, rotated grid sampling
+        UNIFORM_HELIX_X4,    //!  4-samples, uniform grid in helix sequence
+        HALTON_23_X8,        //!  8-samples of halton 2,3
+        HALTON_23_X16        //! 16-samples of halton 2,3
+    };
+
+    bool filterHistory = true;      //!< whether to filter the history buffer
+    bool filterInput = true;        //!< whether to apply the reconstruction filter to the input
+    bool useYCoCg = false;          //!< whether to use the YcoCg color-space for history rejection
+    BoxType boxType = BoxType::AABB;            //!< type of color gamut box
+    BoxClipping boxClipping = BoxClipping::ACCURATE;     //!< clipping algorithm
+    JitterPattern jitterPattern = JitterPattern::HALTON_23_X16; //! Jitter Pattern
+    float varianceGamma = 1.0f; //! High values increases ghosting artefact, lower values increases jittering, range [0.75, 1.25]
+
+    bool preventFlickering = false;     //!< adjust the feedback dynamically to reduce flickering
+    bool historyReprojection = true;    //!< whether to apply history reprojection (debug option)
 };
 
 /**

filament/include/filament/SwapChain.h

@@ -252,6 +252,12 @@ public:
      * automatically schedule itself for presentation. Instead, the application must call the
      * PresentCallable passed to the FrameScheduledCallback.
      *
+     * If your application delays the call to the PresentCallable by, for example, calling it on a
+     * separate thread, you must ensure all PresentCallables have been called before shutting down
+     * the Filament Engine. You can do this by issuing an Engine::flushAndWait before calling
+     * Engine::shutdown. This is necessary to ensure the Filament Engine has had a chance to clean
+     * up all memory related to frame presentation.
+     *
      * @param callback    A callback, or nullptr to unset.
      * @param user        An optional pointer to user data passed to the callback function.
      *

filament/src/PerShadowMapUniforms.cpp

@@ -71,8 +71,7 @@ void PerShadowMapUniforms::prepareCamera(Transaction const& transaction,
     s.clipControl = engine.getDriverApi().getClipSpaceParams();
 }
 
-void PerShadowMapUniforms::prepareLodBias(Transaction const& transaction,
-        float bias) noexcept {
+void PerShadowMapUniforms::prepareLodBias(Transaction const& transaction, float bias) noexcept {
     auto& s = edit(transaction);
     s.lodBias = bias;
 }

filament/src/PerViewUniforms.cpp

@@ -50,7 +50,7 @@ PerViewUniforms::PerViewUniforms(FEngine& engine) noexcept
             BufferObjectBinding::UNIFORM, BufferUsage::DYNAMIC);
 
     if (engine.getDFG().isValid()) {
-        TextureSampler sampler(TextureSampler::MagFilter::LINEAR);
+        TextureSampler const sampler(TextureSampler::MagFilter::LINEAR);
         mSamplers.setSampler(PerViewSib::IBL_DFG_LUT,
                 { engine.getDFG().getTexture(), sampler.getSamplerParams() });
     }
@@ -96,9 +96,10 @@ void PerViewUniforms::prepareCamera(FEngine& engine, const CameraInfo& camera) n
     s.clipControl = engine.getDriverApi().getClipSpaceParams();
 }
 
-void PerViewUniforms::prepareLodBias(float bias) noexcept {
+void PerViewUniforms::prepareLodBias(float bias, float2 derivativesScale) noexcept {
     auto& s = mUniforms.edit();
     s.lodBias = bias;
+    s.derivativesScale = derivativesScale;
 }
 
 void PerViewUniforms::prepareExposure(float ev100) noexcept {

filament/src/PerViewUniforms.h

@@ -69,7 +69,7 @@ public:
     void terminate(backend::DriverApi& driver);
 
     void prepareCamera(FEngine& engine, const CameraInfo& camera) noexcept;
-    void prepareLodBias(float bias) noexcept;
+    void prepareLodBias(float bias, math::float2 derivativesScale) noexcept;
 
     /*
      * @param viewport  viewport (should be same as RenderPassParams::viewport)

filament/src/PostProcessManager.cpp

@@ -27,9 +27,12 @@
 #include "details/Engine.h"
 
 #include "fg/FrameGraph.h"
+#include "fg/FrameGraphId.h"
 #include "fg/FrameGraphResources.h"
+#include "fg/FrameGraphTexture.h"
 
 #include "fsr.h"
+#include "FrameHistory.h"
 #include "PerViewUniforms.h"
 #include "RenderPass.h"
 
@@ -41,13 +44,44 @@
 
 #include "generated/resources/materials.h"
 
+#include <filament/Material.h>
 #include <filament/MaterialEnums.h>
+#include <filament/Options.h>
+#include <filament/Viewport.h>
+
+#include <private/filament/EngineEnums.h>
+
+#include <backend/DriverEnums.h>
+#include <backend/DriverApiForward.h>
+#include <backend/Handle.h>
+#include <backend/PipelineState.h>
+#include <backend/PixelBufferDescriptor.h>
+
+#include <private/backend/BackendUtils.h>
 
 #include <math/half.h>
 #include <math/mat2.h>
+#include <math/mat3.h>
+#include <math/mat4.h>
+#include <math/scalar.h>
+#include <math/vec3.h>
+#include <math/vec4.h>
+
+#include <utils/algorithm.h>
+#include <utils/BitmaskEnum.h>
+#include <utils/debug.h>
+#include <utils/compiler.h>
+#include <utils/FixedCapacityVector.h>
 
 #include <algorithm>
+#include <cmath>
 #include <limits>
+#include <string_view>
+#include <variant>
+#include <utility>
+
+#include <stddef.h>
+#include <stdint.h>
 
 namespace filament {
 
@@ -167,7 +201,21 @@ FMaterialInstance* PostProcessManager::PostProcessMaterial::getMaterialInstance(
 
 // ------------------------------------------------------------------------------------------------
 
-const math::float2 PostProcessManager::sHaltonSamples[16] = {
+const PostProcessManager::JitterSequence<4> PostProcessManager::sRGSS4 = {{
+        { 0.625f, 0.125f },
+        { 0.125f, 0.375f },
+        { 0.875f, 0.625f },
+        { 0.375f, 0.875f }
+}};
+
+const PostProcessManager::JitterSequence<4> PostProcessManager::sUniformHelix4 = {{
+        { 0.25f, 0.25f },
+        { 0.75f, 0.75f },
+        { 0.25f, 0.75f },
+        { 0.75f, 0.25f },
+}};
+
+const PostProcessManager::JitterSequence<16> PostProcessManager::sHaltonSamples = {{
         { filament::halton( 0, 2), filament::halton( 0, 3) },
         { filament::halton( 1, 2), filament::halton( 1, 3) },
         { filament::halton( 2, 2), filament::halton( 2, 3) },
@@ -184,7 +232,7 @@ const math::float2 PostProcessManager::sHaltonSamples[16] = {
         { filament::halton(13, 2), filament::halton(13, 3) },
         { filament::halton(14, 2), filament::halton(14, 3) },
         { filament::halton(15, 2), filament::halton(15, 3) }
-};
+}};
 
 PostProcessManager::PostProcessManager(FEngine& engine) noexcept
         : mEngine(engine),
@@ -200,7 +248,8 @@ void PostProcessManager::registerPostProcessMaterial(std::string_view name, Mate
 }
 
 UTILS_NOINLINE
-PostProcessManager::PostProcessMaterial& PostProcessManager::getPostProcessMaterial(std::string_view name) noexcept {
+PostProcessManager::PostProcessMaterial& PostProcessManager::getPostProcessMaterial(
+        std::string_view name) noexcept {
     assert_invariant(mMaterialRegistry.find(name) != mMaterialRegistry.end());
     return mMaterialRegistry[name];
 }
@@ -258,6 +307,7 @@ static const PostProcessManager::MaterialInfo sMaterialList[] = {
         { "fsr_easu_mobileF",           MATERIAL(FSR_EASU_MOBILEF) },
         { "fsr_rcas",                   MATERIAL(FSR_RCAS) },
         { "debugShadowCascades",        MATERIAL(DEBUGSHADOWCASCADES) },
+        { "resolveDepth",               MATERIAL(RESOLVEDEPTH) },
 };
 
 void PostProcessManager::init() noexcept {
@@ -754,7 +804,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::screenSpaceAmbientOcclusion(
                 auto ssao = resources.getRenderPassInfo();
                 auto const& desc = resources.getDescriptor(data.depth);
 
-                // estimate of the size in pixel of a 1m tall/wide object viewed from 1m away (i.e. at z=-1)
+                // Estimate of the size in pixel units of a 1m tall/wide object viewed from 1m away (i.e. at z=-1)
                 const float projectionScale = std::min(
                         0.5f * cameraInfo.projection[0].x * desc.width,
                         0.5f * cameraInfo.projection[1].y * desc.height);
@@ -911,7 +961,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::bilateralBlurPass(FrameGraph
                 auto const& desc = resources.getDescriptor(data.blurred);
 
                 // unnormalized gaussian half-kernel of a given standard deviation
-                // returns number of samples stored in array (max 16)
+                // returns number of samples stored in the array (max 16)
                 constexpr size_t kernelArraySize = 16; // limited by bilateralBlur.mat
                 auto gaussianKernel =
                         [kernelArraySize](float* outKernel, size_t gaussianWidth, float stdDev) -> uint32_t {
@@ -955,7 +1005,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::generateGaussianMipmap(Frame
 
     auto const subResourceDesc = fg.getSubResourceDescriptor(input);
 
-    // create one subresource per level to be generated from the input. These will be our
+    // Create one subresource per level to be generated from the input. These will be our
     // destinations.
     struct MipmapPassData {
         FixedCapacityVector<FrameGraphId<FrameGraphTexture>> out;
@@ -1116,7 +1166,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::gaussianBlurPass(FrameGraph&
                 size_t const m = computeGaussianCoefficients(kernel,
                         std::min(sizeof(kernel) / sizeof(*kernel), kernelStorageSize));
 
-                std::string_view sourceParameterName = is2dArray ? "sourceArray"sv : "source"sv;
+                std::string_view const sourceParameterName = is2dArray ? "sourceArray"sv : "source"sv;
                 // horizontal pass
                 mi->setParameter(sourceParameterName, hwIn, {
                         .filterMag = SamplerMagFilter::LINEAR,
@@ -1162,7 +1212,7 @@ PostProcessManager::ScreenSpaceRefConfig PostProcessManager::prepareMipmapSSR(Fr
 
     // The kernel-size was determined empirically so that we don't get too many artifacts
     // due to the down-sampling with a box filter (which happens implicitly).
-    // requires only 6 stored coefficients and 11 tap/pass
+    // Requires only 6 stored coefficients and 11 tap/pass
     // e.g.: size of 13 (4 stored coefficients)
     //      +-------+-------+-------*===*-------+-------+-------+
     //  ... | 6 | 5 | 4 | 3 | 2 | 1 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | ...
@@ -1326,10 +1376,10 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::generateMipmapSSR(
         FrameGraphId<FrameGraphTexture> output,
         bool needInputDuplication, ScreenSpaceRefConfig const& config) noexcept {
 
-    // descriptor of our actual input image (e.g. reflection buffer or refraction framebuffer)
+    // Descriptor of our actual input image (e.g. reflection buffer or refraction framebuffer)
     auto const& desc = fg.getDescriptor(input);
 
-    // descriptor of the destination. output is a subresource (i.e. a layer of a 2D array)
+    // Descriptor of the destination. `output` is a subresource (i.e. a layer of a 2D array)
     auto const& outDesc = fg.getDescriptor(output);
 
     /*
@@ -1378,7 +1428,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::dof(FrameGraph& fg,
         FrameGraphId<FrameGraphTexture> depth,
         const CameraInfo& cameraInfo,
         bool translucent,
-        float bokehAspectRatio,
+        float2 bokehScale,
         const DepthOfFieldOptions& dofOptions) noexcept {
 
     assert_invariant(depth);
@@ -1389,7 +1439,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::dof(FrameGraph& fg,
     const TextureFormat format = translucent ? TextureFormat::RGBA16F
                                              : TextureFormat::R11F_G11F_B10F;
 
-    // rotate the bokeh based on the aperture diameter (i.e. angle of the blades)
+    // Rotate the bokeh based on the aperture diameter (i.e. angle of the blades)
     float bokehAngle = f::PI / 6.0f;
     if (dofOptions.maxApertureDiameter > 0.0f) {
         bokehAngle += f::PI_2 * saturate(cameraInfo.A / dofOptions.maxApertureDiameter);
@@ -1768,8 +1818,8 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::dof(FrameGraph& fg,
                 mi->setParameter("tiles", tilesCocMinMax,
                         { .filterMin = SamplerMinFilter::NEAREST });
                 mi->setParameter("cocToTexelScale", float2{
-                        bokehAspectRatio / (inputDesc.width  * dofResolution),
-                                     1.0 / (inputDesc.height * dofResolution)
+                        bokehScale.x / (inputDesc.width * dofResolution),
+                        bokehScale.y / (inputDesc.height * dofResolution)
                 });
                 mi->setParameter("cocToPixelScale", (1.0f / float(dofResolution)));
                 mi->setParameter("ringCounts", float4{
@@ -1875,14 +1925,6 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::dof(FrameGraph& fg,
     return ppDoFCombine->output;
 }
 
-PostProcessManager::BloomPassOutput PostProcessManager::bloom(FrameGraph& fg,
-        FrameGraphId<FrameGraphTexture> input,
-        BloomOptions& inoutBloomOptions,
-        backend::TextureFormat outFormat,
-        math::float2 scale) noexcept {
-    return bloomPass(fg, input, outFormat, inoutBloomOptions, scale);
-}
-
 FrameGraphId<FrameGraphTexture> PostProcessManager::downscalePass(FrameGraph& fg,
         FrameGraphId<FrameGraphTexture> input,
         FrameGraphTexture::Descriptor const& outDesc,
@@ -1914,9 +1956,11 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::downscalePass(FrameGraph& fg
     return downsamplePass->output;
 }
 
-PostProcessManager::BloomPassOutput PostProcessManager::bloomPass(FrameGraph& fg,
+PostProcessManager::BloomPassOutput PostProcessManager::bloom(FrameGraph& fg,
         FrameGraphId<FrameGraphTexture> input, TextureFormat outFormat,
-        BloomOptions& inoutBloomOptions, float2 scale) noexcept {
+        BloomOptions& inoutBloomOptions,
+        TemporalAntiAliasingOptions const& taaOptions,
+        float2 scale) noexcept {
 
     // Figure out a good size for the bloom buffer. We must use a fixed bloom buffer size so
     // that the size/strength of the bloom doesn't vary much with the resolution, otherwise
@@ -1958,6 +2002,9 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloomPass(FrameGraph& fg
 
     bool threshold = inoutBloomOptions.threshold;
 
+    // we don't need to do the fireflies reduction if we have TAA (it already does it)
+    bool fireflies = threshold && !taaOptions.enabled;
+
     while (2 * bloomWidth < float(desc.width) || 2 * bloomHeight < float(desc.height)) {
         if (inoutBloomOptions.quality == QualityLevel::LOW ||
             inoutBloomOptions.quality == QualityLevel::MEDIUM) {
@@ -1966,8 +2013,9 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloomPass(FrameGraph& fg
                             .height = (desc.height = std::max(1u, desc.height / 2)),
                             .format = outFormat
                     },
-                    threshold, inoutBloomOptions.highlight, threshold);
+                    threshold, inoutBloomOptions.highlight, fireflies);
             threshold = false; // we do the thresholding only once during down sampling
+            fireflies = false; // we do the fireflies reduction only once during down sampling
         } else if (inoutBloomOptions.quality == QualityLevel::HIGH ||
                    inoutBloomOptions.quality == QualityLevel::ULTRA) {
             // In high quality mode, we increase the size of the bloom buffer such that the
@@ -1988,7 +2036,7 @@ PostProcessManager::BloomPassOutput PostProcessManager::bloomPass(FrameGraph& fg
 
     input = downscalePass(fg, input,
             { .width = width, .height = height, .format = outFormat },
-            threshold, inoutBloomOptions.highlight, threshold);
+            threshold, inoutBloomOptions.highlight, fireflies);
 
     struct BloomPassData {
         FrameGraphId<FrameGraphTexture> out;
@@ -2497,7 +2545,9 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::fxaa(FrameGraph& fg,
     return ppFXAA->output;
 }
 
-void PostProcessManager::prepareTaa(FrameGraph& fg, filament::Viewport const& svp,
+void PostProcessManager::prepareTaa(FrameGraph& fg,
+        filament::Viewport const& svp,
+        TemporalAntiAliasingOptions const& taaOptions,
         FrameHistory& frameHistory,
         FrameHistoryEntry::TemporalAA FrameHistoryEntry::*pTaa,
         CameraInfo* inoutCameraInfo,
@@ -2509,9 +2559,33 @@ void PostProcessManager::prepareTaa(FrameGraph& fg, filament::Viewport const& sv
     current.projection = inoutCameraInfo->projection * inoutCameraInfo->getUserViewMatrix();
     current.frameId = previous.frameId + 1;
 
+    auto jitterPosition = [pattern = taaOptions.jitterPattern](size_t frameIndex){
+        using JitterPattern = TemporalAntiAliasingOptions::JitterPattern;
+        switch (pattern) {
+            case JitterPattern::RGSS_X4:
+                return sRGSS4(frameIndex);
+            case JitterPattern::UNIFORM_HELIX_X4:
+                return sUniformHelix4(frameIndex);
+            case JitterPattern::HALTON_23_X8:
+                return sHaltonSamples(frameIndex % 8);
+            case JitterPattern::HALTON_23_X16:
+                return sHaltonSamples(frameIndex);
+        }
+    };
+
     // sample position within a pixel [-0.5, 0.5]
-    float2 const jitter = halton(previous.frameId) - 0.5f;
-    current.jitter = jitter;
+    // for metal/vulkan we need to reverse the y-offset
+    current.jitter = jitterPosition(previous.frameId);
+    float2 jitter = current.jitter;
+    switch (mEngine.getBackend()) {
+        case Backend::VULKAN:
+        case Backend::METAL:
+            jitter.y = -jitter.y;
+            UTILS_FALLTHROUGH;
+        case Backend::OPENGL:
+        default:
+            break;
+    }
 
     float2 const jitterInClipSpace = jitter * (2.0f / float2{ svp.width, svp.height });
 
@@ -2528,6 +2602,37 @@ void PostProcessManager::prepareTaa(FrameGraph& fg, filament::Viewport const& sv
     });
 }
 
+void PostProcessManager::configureTemporalAntiAliasingMaterial(
+        TemporalAntiAliasingOptions const& taaOptions) noexcept {
+
+    FMaterial* const ma = getPostProcessMaterial("taa").getMaterial(mEngine);
+    bool dirty = false;
+
+    auto setConstantParameter =
+            [&dirty](FMaterial* const ma, std::string_view name, auto value) noexcept {
+        auto id = ma->getSpecializationConstantId(name);
+        if (id.has_value()) {
+            if (ma->setConstant(id.value(), value)) {
+                dirty = true;
+            }
+        }
+    };
+
+    setConstantParameter(ma, "historyReprojection", taaOptions.historyReprojection);
+    setConstantParameter(ma, "filterHistory", taaOptions.filterHistory);
+    setConstantParameter(ma, "filterInput", taaOptions.filterInput);
+    setConstantParameter(ma, "useYCoCg", taaOptions.useYCoCg);
+    setConstantParameter(ma, "preventFlickering", taaOptions.preventFlickering);
+    setConstantParameter(ma, "boxType", (int32_t)taaOptions.boxType);
+    setConstantParameter(ma, "boxClipping", (int32_t)taaOptions.boxClipping);
+    setConstantParameter(ma, "varianceGamma", taaOptions.varianceGamma);
+    if (dirty) {
+        ma->invalidate();
+        // TODO: call Material::compile(), we can't si that now because it works only
+        //       with surface materials
+    }
+}
+
 FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
         FrameGraphId<FrameGraphTexture> input,
         FrameGraphId<FrameGraphTexture> depth,
@@ -2560,6 +2665,10 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
     auto& taaPass = fg.addPass<TAAData>("TAA",
             [&](FrameGraph::Builder& builder, auto& data) {
                 auto desc = fg.getDescriptor(input);
+                if (taaOptions.upscaling) {
+                    desc.width *= 2;
+                    desc.height *= 2;
+                }
                 data.color = builder.sample(input);
                 data.depth = builder.sample(depth);
                 data.history = builder.sample(colorHistory);
@@ -2594,18 +2703,25 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
                         { -1.0f,  1.0f }, {  0.0f,  1.0f }, {  1.0f,  1.0f },
                 };
 
-                float sum = 0.0;
-                float weights[9];
+                constexpr float2 subSampleOffsets[4] = {
+                        { -0.25f, 0.25f }, {  0.25f, 0.25f }, { 0.25f, -0.25f }, { -0.25f, -0.25f }
+                };
+
+                float4 sum = 0.0;
+                float4 weights[9];
 
                 // this doesn't get vectorized (probably because of exp()), so don't bother
                 // unrolling it.
                 #pragma nounroll
                 for (size_t i = 0; i < 9; i++) {
-                    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 Brian Karis
-                    weights[i] = std::exp2(-3.3f * (d.x * d.x + d.y * d.y));
+                    float2 const o = sampleOffsets[i];
+                    for (size_t j = 0; j < 4; j++) {
+                        float2 const s = taaOptions.upscaling ? subSampleOffsets[j] : float2{ 0 };
+                        float2 const d = (o - current.jitter - s) / taaOptions.filterWidth;
+                        // This is a gaussian fit of a 3.3-wide Blackman-Harris window
+                        // see: "High Quality Temporal Supersampling" by Brian Karis
+                        weights[i][j] = std::exp(-2.29f * (d.x * d.x + d.y * d.y));
+                    }
                     sum += weights[i];
                 }
                 for (auto& w : weights) {
@@ -2616,8 +2732,8 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
                 auto color = resources.getTexture(data.color);
                 auto depth = resources.getTexture(data.depth);
                 auto history = resources.getTexture(data.history);
-
                 auto const& material = getPostProcessMaterial("taa");
+
                 FMaterialInstance* mi = material.getMaterialInstance(mEngine);
                 mi->setParameter("color",  color, {});  // nearest
                 mi->setParameter("depth",  depth, {});  // nearest
@@ -2627,6 +2743,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
                         .filterMin = SamplerMinFilter::LINEAR
                 });
                 mi->setParameter("filterWeights",  weights, 9);
+                mi->setParameter("jitter",  current.jitter);
                 mi->setParameter("reprojection",
                         mat4f{ historyProjection * inverse(current.projection) } *
                         normalizedToClip);
@@ -2947,6 +3064,14 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::resolve(FrameGraph& fg,
         return input;
     }
 
+    // we currently don't support stencil resolve
+    assert_invariant(!isStencilFormat(inDesc.format));
+
+    // The Metal / Vulkan backends currently don't support depth/stencil resolve.
+    if (isDepthFormat(inDesc.format) && (!mEngine.getDriverApi().isDepthStencilResolveSupported())) {
+        return resolveDepth(fg, outputBufferName, input, outDesc);
+    }
+
     outDesc.width = inDesc.width;
     outDesc.height = inDesc.height;
     outDesc.format = inDesc.format;
@@ -2968,21 +3093,12 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::resolve(FrameGraph& fg,
                 auto const& dst = resources.getTexture(data.output);
                 auto const& srcSubDesc = resources.getSubResourceDescriptor(data.input);
                 auto const& dstSubDesc = resources.getSubResourceDescriptor(data.output);
-                auto const& srcDesc = resources.getDescriptor(data.input);
-                auto const& dstDesc = resources.getDescriptor(data.output);
+                UTILS_UNUSED_IN_RELEASE auto const& srcDesc = resources.getDescriptor(data.input);
+                UTILS_UNUSED_IN_RELEASE auto const& dstDesc = resources.getDescriptor(data.output);
                 assert_invariant(src);
                 assert_invariant(dst);
-
-                ASSERT_POSTCONDITION(
-                        srcDesc.format == dstDesc.format,
-                        "resolve: src (%d) and dst (%d) formats don't match.",
-                        srcDesc.format, dstDesc.format);
-
-                ASSERT_POSTCONDITION(
-                        srcDesc.width == dstDesc.width && srcDesc.height == dstDesc.height,
-                        "resolve: src [%u, %u] and dst [%u, %u] sizes don't match.",
-                        srcDesc.width, srcDesc.height, dstDesc.width, dstDesc.height);
-
+                assert_invariant(srcDesc.format == dstDesc.format);
+                assert_invariant(srcDesc.width == dstDesc.width && srcDesc.height == dstDesc.height);
                 driver.resolve(
                         dst, dstSubDesc.level, dstSubDesc.layer,
                         src, srcSubDesc.level, srcSubDesc.layer);
@@ -2991,6 +3107,51 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::resolve(FrameGraph& fg,
     return ppResolve->output;
 }
 
+FrameGraphId<FrameGraphTexture> PostProcessManager::resolveDepth(FrameGraph& fg,
+        const char* outputBufferName, FrameGraphId<FrameGraphTexture> input,
+        FrameGraphTexture::Descriptor outDesc) noexcept {
+
+    // Don't do anything if we're not a MSAA buffer
+    auto const& inDesc = fg.getDescriptor(input);
+    if (inDesc.samples <= 1) {
+        return input;
+    }
+
+    UTILS_UNUSED_IN_RELEASE auto const& inSubDesc = fg.getSubResourceDescriptor(input);
+    assert_invariant(isDepthFormat(inDesc.format));
+    assert_invariant(inSubDesc.layer == 0);
+    assert_invariant(inSubDesc.level == 0);
+
+    outDesc.width = inDesc.width;
+    outDesc.height = inDesc.height;
+    outDesc.format = inDesc.format;
+    outDesc.samples = 0;
+
+    struct ResolveData {
+        FrameGraphId<FrameGraphTexture> input;
+        FrameGraphId<FrameGraphTexture> output;
+    };
+
+    auto& ppResolve = fg.addPass<ResolveData>("resolveDepth",
+            [&](FrameGraph::Builder& builder, auto& data) {
+                data.input = builder.sample(input);
+                data.output = builder.createTexture(outputBufferName, outDesc);
+                data.output = builder.write(data.output, FrameGraphTexture::Usage::DEPTH_ATTACHMENT);
+                builder.declareRenderPass(builder.getName(data.output), {
+                        .attachments = { .depth = { data.output }},
+                        .clearFlags = TargetBufferFlags::DEPTH });
+            },
+            [=](FrameGraphResources const& resources, auto const& data, DriverApi& driver) {
+                auto const& input = resources.getTexture(data.input);
+                auto const& material = getPostProcessMaterial("resolveDepth");
+                auto* mi = material.getMaterialInstance(mEngine);
+                mi->setParameter("depth", input, {}); // NEAREST
+                commitAndRender(resources.getRenderPassInfo(), material, driver);
+            });
+
+    return ppResolve->output;
+}
+
 FrameGraphId<FrameGraphTexture> PostProcessManager::vsmMipmapPass(FrameGraph& fg,
         FrameGraphId<FrameGraphTexture> input, uint8_t layer, size_t level,
         math::float4 clearColor, bool finalize) noexcept {
@@ -3030,7 +3191,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::vsmMipmapPass(FrameGraph& fg
                 auto& material = getPostProcessMaterial("vsmMipmap");
 
                 // When generating shadow map mip levels, we want to preserve the 1 texel border.
-                // (note clearing never respects the scissor in filament)
+                // (note clearing never respects the scissor in Filament)
                 PipelineState pipeline(material.getPipelineState(mEngine));
                 pipeline.scissor = { 1u, 1u, dim - 2u, dim - 2u };
 

filament/src/PostProcessManager.h

@@ -86,6 +86,10 @@ public:
     void init() noexcept;
     void terminate(backend::DriverApi& driver) noexcept;
 
+
+    void configureTemporalAntiAliasingMaterial(
+            TemporalAntiAliasingOptions const& taaOptions) noexcept;
+
     // methods below are ordered relative to their position in the pipeline (as much as possible)
 
     // structure (depth) pass
@@ -158,7 +162,7 @@ public:
             FrameGraphId<FrameGraphTexture> depth,
             const CameraInfo& cameraInfo,
             bool translucent,
-            float bokehAspectRatio,
+            math::float2 bokehScale,
             const DepthOfFieldOptions& dofOptions) noexcept;
 
     // Bloom
@@ -167,7 +171,9 @@ public:
         FrameGraphId<FrameGraphTexture> flare;
     };
     BloomPassOutput bloom(FrameGraph& fg, FrameGraphId<FrameGraphTexture> input,
-            BloomOptions& inoutBloomOptions, backend::TextureFormat outFormat,
+            backend::TextureFormat outFormat,
+            BloomOptions& inoutBloomOptions,
+            TemporalAntiAliasingOptions const& taaOptions,
             math::float2 scale) noexcept;
 
     FrameGraphId<FrameGraphTexture> flarePass(FrameGraph& fg,
@@ -207,7 +213,9 @@ public:
             backend::TextureFormat outFormat, bool translucent) noexcept;
 
     // Temporal Anti-aliasing
-    void prepareTaa(FrameGraph& fg, filament::Viewport const& svp,
+    void prepareTaa(FrameGraph& fg,
+            filament::Viewport const& svp,
+            TemporalAntiAliasingOptions const& taaOptions,
             FrameHistory& frameHistory,
             FrameHistoryEntry::TemporalAA FrameHistoryEntry::*pTaa,
             CameraInfo* inoutCameraInfo,
@@ -246,6 +254,12 @@ public:
             const char* outputBufferName, FrameGraphId<FrameGraphTexture> input,
             FrameGraphTexture::Descriptor outDesc) noexcept;
 
+    // Resolves base level of input and outputs a texture from outDesc.
+    // outDesc with, height, format and samples will be overridden.
+    FrameGraphId<FrameGraphTexture> resolveDepth(FrameGraph& fg,
+            const char* outputBufferName, FrameGraphId<FrameGraphTexture> input,
+            FrameGraphTexture::Descriptor outDesc) noexcept;
+
     // VSM shadow mipmap pass
     FrameGraphId<FrameGraphTexture> vsmMipmapPass(FrameGraph& fg,
             FrameGraphId<FrameGraphTexture> input, uint8_t layer, size_t level,
@@ -265,48 +279,6 @@ public:
     backend::Handle<backend::HwTexture> getOneTextureArray() const;
     backend::Handle<backend::HwTexture> getZeroTextureArray() const;
 
-    math::float2 halton(size_t index) const noexcept {
-        return mHaltonSamples[index & 0xFu];
-    }
-
-private:
-    FEngine& mEngine;
-    class PostProcessMaterial;
-
-    struct BilateralPassConfig {
-        uint8_t kernelSize = 11;
-        bool bentNormals = false;
-        float standardDeviation = 1.0f;
-        float bilateralThreshold = 0.0625f;
-        float scale = 1.0f;
-    };
-
-    FrameGraphId<FrameGraphTexture> bilateralBlurPass(FrameGraph& fg,
-            FrameGraphId<FrameGraphTexture> input, FrameGraphId<FrameGraphTexture> depth,
-            math::int2 axis, float zf, backend::TextureFormat format,
-            BilateralPassConfig const& config) noexcept;
-
-    BloomPassOutput bloomPass(FrameGraph& fg,
-            FrameGraphId<FrameGraphTexture> input, backend::TextureFormat outFormat,
-            BloomOptions& inoutBloomOptions, math::float2 scale) noexcept;
-
-    FrameGraphId<FrameGraphTexture> downscalePass(FrameGraph& fg,
-            FrameGraphId<FrameGraphTexture> input,
-            FrameGraphTexture::Descriptor const& outDesc,
-            bool threshold, float highlight, bool fireflies) noexcept;
-
-    void commitAndRender(FrameGraphResources::RenderPassInfo const& out,
-            PostProcessMaterial const& material, uint8_t variant,
-            backend::DriverApi& driver) const noexcept;
-
-    void commitAndRender(FrameGraphResources::RenderPassInfo const& out,
-            PostProcessMaterial const& material,
-            backend::DriverApi& driver) const noexcept;
-
-    void render(FrameGraphResources::RenderPassInfo const& out,
-            backend::PipelineState const& pipeline,
-            backend::DriverApi& driver) const noexcept;
-
     class PostProcessMaterial {
     public:
         PostProcessMaterial() noexcept;
@@ -340,21 +312,62 @@ private:
         utils::FixedCapacityVector<ConstantInfo> mConstants{};
     };
 
+    void registerPostProcessMaterial(std::string_view name, MaterialInfo const& info);
+
+    PostProcessMaterial& getPostProcessMaterial(std::string_view name) noexcept;
+
+    void commitAndRender(FrameGraphResources::RenderPassInfo const& out,
+            PostProcessMaterial const& material, uint8_t variant,
+            backend::DriverApi& driver) const noexcept;
+
+    void commitAndRender(FrameGraphResources::RenderPassInfo const& out,
+            PostProcessMaterial const& material,
+            backend::DriverApi& driver) const noexcept;
+
+    void render(FrameGraphResources::RenderPassInfo const& out,
+            backend::PipelineState const& pipeline,
+            backend::DriverApi& driver) const noexcept;
+
+private:
+    FEngine& mEngine;
+
+    struct BilateralPassConfig {
+        uint8_t kernelSize = 11;
+        bool bentNormals = false;
+        float standardDeviation = 1.0f;
+        float bilateralThreshold = 0.0625f;
+        float scale = 1.0f;
+    };
+
+    FrameGraphId<FrameGraphTexture> bilateralBlurPass(FrameGraph& fg,
+            FrameGraphId<FrameGraphTexture> input, FrameGraphId<FrameGraphTexture> depth,
+            math::int2 axis, float zf, backend::TextureFormat format,
+            BilateralPassConfig const& config) noexcept;
+
+    FrameGraphId<FrameGraphTexture> downscalePass(FrameGraph& fg,
+            FrameGraphId<FrameGraphTexture> input,
+            FrameGraphTexture::Descriptor const& outDesc,
+            bool threshold, float highlight, bool fireflies) noexcept;
+
     using MaterialRegistryMap = tsl::robin_map<
             std::string_view,
             PostProcessMaterial>;
 
     MaterialRegistryMap mMaterialRegistry;
 
-    void registerPostProcessMaterial(std::string_view name, MaterialInfo const& info);
-    PostProcessMaterial& getPostProcessMaterial(std::string_view name) noexcept;
-
     backend::Handle<backend::HwTexture> mStarburstTexture;
 
     std::uniform_real_distribution<float> mUniformDistribution{0.0f, 1.0f};
 
-    static const math::float2 sHaltonSamples[16];
-    math::float2 const* mHaltonSamples = sHaltonSamples;
+    template<size_t SIZE>
+    struct JitterSequence {
+        auto operator()(size_t i) const noexcept { return positions[i % SIZE] - 0.5f; }
+        const math::float2 positions[SIZE];
+    };
+
+    static const JitterSequence<4> sRGSS4;
+    static const JitterSequence<4> sUniformHelix4;
+    static const JitterSequence<16> sHaltonSamples;
 
     bool mWorkaroundSplitEasu : 1;
     bool mWorkaroundAllowReadOnlyAncillaryFeedbackLoop : 1;

filament/src/ShadowMapManager.cpp

@@ -247,7 +247,7 @@ FrameGraphId<FrameGraphTexture> ShadowMapManager::render(FEngine& engine, FrameG
 
 
                         // cameraInfo only valid after calling update
-                        const CameraInfo cameraInfo{ shadowMap.getCamera() };
+                        const CameraInfo cameraInfo{ shadowMap.getCamera(), mainCameraInfo };
 
                         auto transaction = ShadowMap::open(driver);
                         ShadowMap::prepareCamera(transaction, engine, cameraInfo);
@@ -844,7 +844,7 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateSpotShadowMaps(FEngine
         FScene::LightSoa const& lightData) noexcept {
 
     // The const_cast here is a little ugly, but conceptually lightData should be const,
-    // it's just that we're using it to store some temporary data. with SoA we can't have
+    // it's just that we're using it to store some temporary data. With SoA we can't have
     // a `mutable` element, so that's a workaround.
     FScene::ShadowInfo* const shadowInfo = const_cast<FScene::ShadowInfo*>(
             lightData.data<FScene::SHADOW_INFO>());
@@ -855,7 +855,7 @@ ShadowMapManager::ShadowTechnique ShadowMapManager::updateSpotShadowMaps(FEngine
         shadowTechnique |= ShadowTechnique::SHADOW_MAP;
         for (ShadowMap const& shadowMap : spotShadowMaps) {
             const size_t lightIndex = shadowMap.getLightIndex();
-            // gather the per-light (not per shadow map) information. For point lights we will
+            // Gather the per-light (not per shadow map) information. For point lights we will
             // "see" 6 shadowmaps (one per face), we must use the first face one, the shader
             // knows how to find the entry for other faces (they're guaranteed to be sequential).
             if (shadowMap.getFace() == 0) {

filament/src/details/Camera.cpp

@@ -265,24 +265,22 @@ uint8_t FCamera::getStereoscopicEyeCount() const noexcept {
 
 // ------------------------------------------------------------------------------------------------
 
-CameraInfo::CameraInfo(FCamera const& camera) noexcept {
-    for (int i = 0; i < camera.getStereoscopicEyeCount(); i++) {
-        eyeProjection[i]   = mat4f{ camera.getProjectionMatrix(i) };
-    }
-    cullingProjection  = mat4f{ camera.getCullingProjectionMatrix() };
-    model              = mat4f{ camera.getModelMatrix() };
-    view               = mat4f{ camera.getViewMatrix() };
-    zn                 = (float)camera.getNear();
-    zf                 = (float)camera.getCullingFar();
-    ev100              = Exposure::ev100(camera);
-    f                  = (float)camera.getFocalLength();
-    A                  = f / camera.getAperture();
-    d                  = std::max(zn, camera.getFocusDistance());
+CameraInfo::CameraInfo(FCamera const& camera) noexcept
+        : CameraInfo(camera, {}, camera.getModelMatrix()) {
 }
 
-CameraInfo::CameraInfo(FCamera const& camera, math::mat4 const& inWorldTransform) noexcept {
-    const mat4 modelMatrix{ inWorldTransform * camera.getModelMatrix() };
-    for (int i = 0; i < camera.getStereoscopicEyeCount(); i++) {
+CameraInfo::CameraInfo(FCamera const& camera, math::mat4 const& inWorldTransform) noexcept
+        : CameraInfo(camera, inWorldTransform, inWorldTransform * camera.getModelMatrix()) {
+}
+
+CameraInfo::CameraInfo(FCamera const& camera, CameraInfo const& mainCameraInfo) noexcept
+        : CameraInfo(camera, mainCameraInfo.worldTransform, camera.getModelMatrix()) {
+}
+
+CameraInfo::CameraInfo(FCamera const& camera,
+        math::mat4 const& inWorldTransform,
+        math::mat4 const& modelMatrix) noexcept {
+    for (size_t i = 0; i < camera.getStereoscopicEyeCount(); i++) {
         eyeProjection[i]   = mat4f{ camera.getProjectionMatrix(i) };
         eyeFromView[i]     = mat4f{ camera.getEyeFromViewMatrix(i) };
     }

filament/src/details/Camera.h

@@ -220,9 +220,19 @@ private:
 
 struct CameraInfo {
     CameraInfo() noexcept {}
-    explicit CameraInfo(FCamera const& camera) noexcept;
+
+    // Creates a CameraInfo relative to inWorldTransform (i.e. it's model matrix is
+    // transformed by inWorldTransform and inWorldTransform is recorded).
+    // This is typically used for the color pass camera.
     CameraInfo(FCamera const& camera, math::mat4 const& inWorldTransform) noexcept;
 
+    // Creates a CameraInfo from a camera that is relative to mainCameraInfo.
+    // This is typically used for the shadow pass cameras.
+    CameraInfo(FCamera const& camera, CameraInfo const& mainCameraInfo) noexcept;
+
+    // Creates a CameraInfo from the FCamera
+    explicit CameraInfo(FCamera const& camera) noexcept;
+
     union {
         // projection matrix for drawing (infinite zfar)
         // for monoscopic rendering
@@ -249,6 +259,11 @@ struct CameraInfo {
     math::float3 const& getPosition() const noexcept { return model[3].xyz; }
     math::float3 getForwardVector() const noexcept { return normalize(-model[2].xyz); }
     math::mat4 getUserViewMatrix() const noexcept { return view * worldTransform; }
+
+private:
+    CameraInfo(FCamera const& camera,
+            math::mat4 const& inWorldTransform,
+            math::mat4 const& modelMatrix) noexcept;
 };
 
 FILAMENT_DOWNCAST(Camera)

filament/src/details/Engine.cpp

@@ -373,8 +373,13 @@ void FEngine::init() {
 
     mDebugRegistry.registerProperty("d.shadowmap.debug_directional_shadowmap",
             &debug.shadowmap.debug_directional_shadowmap, [this]() {
-                mMaterials.forEach([](FMaterial* material) {
+                mMaterials.forEach([this](FMaterial* material) {
                     if (material->getMaterialDomain() == MaterialDomain::SURFACE) {
+
+                        material->setConstant(
+                                +ReservedSpecializationConstants::CONFIG_DEBUG_DIRECTIONAL_SHADOWMAP,
+                                debug.shadowmap.debug_directional_shadowmap);
+
                         material->invalidate(
                                 Variant::DIR | Variant::SRE | Variant::DEP,
                                 Variant::DIR | Variant::SRE);
@@ -384,8 +389,13 @@ void FEngine::init() {
 
     mDebugRegistry.registerProperty("d.lighting.debug_froxel_visualization",
             &debug.lighting.debug_froxel_visualization, [this]() {
-                mMaterials.forEach([](FMaterial* material) {
+                mMaterials.forEach([this](FMaterial* material) {
                     if (material->getMaterialDomain() == MaterialDomain::SURFACE) {
+
+                        material->setConstant(
+                                +ReservedSpecializationConstants::CONFIG_DEBUG_FROXEL_VISUALIZATION,
+                                debug.lighting.debug_froxel_visualization);
+
                         material->invalidate(
                                 Variant::DYN | Variant::DEP,
                                 Variant::DYN);

filament/src/details/Material.cpp

@@ -79,7 +79,10 @@ struct Material::BuilderDetails {
     size_t mSize = 0;
     MaterialParser* mMaterialParser = nullptr;
     bool mDefaultMaterial = false;
-    std::unordered_map<std::string, std::variant<int32_t, float, bool>> mConstantSpecializations;
+    std::unordered_map<
+        utils::CString,
+        std::variant<int32_t, float, bool>,
+        CString::Hasher> mConstantSpecializations;
 };
 
 FMaterial::DefaultMaterialBuilder::DefaultMaterialBuilder() : Material::Builder() {
@@ -216,9 +219,14 @@ FMaterial::FMaterial(FEngine& engine, const Material::Builder& builder)
         mSubpassInfo.isValid = false;
     }
 
-    utils::FixedCapacityVector<MaterialConstant> constants;
     // Older materials won't have a constants chunk, but that's okay.
-    parser->getConstants(&constants);
+    parser->getConstants(&mMaterialConstants);
+    for (size_t i = 0, c = mMaterialConstants.size(); i < c; i++) {
+        auto& item = mMaterialConstants[i];
+        // the key can be a string_view because mMaterialConstant owns the CString
+        std::string_view const key{ item.name.data(), item.name.size() };
+        mSpecializationConstantsNameToIndex[key] = i;
+    }
 
     // Verify that all the constant specializations exist in the material and that their types match.
     // The first specialization constants are defined internally by Filament.
@@ -238,7 +246,7 @@ FMaterial::FMaterial(FEngine& engine, const Material::Builder& builder)
     bool const powerVrShaderWorkarounds =
             engine.getDriverApi().isWorkaroundNeeded(Workaround::POWER_VR_SHADER_WORKAROUNDS);
 
-    mSpecializationConstants.reserve(constants.size() + CONFIG_MAX_RESERVED_SPEC_CONSTANTS);
+    mSpecializationConstants.reserve(mMaterialConstants.size() + CONFIG_MAX_RESERVED_SPEC_CONSTANTS);
     mSpecializationConstants.push_back({
                     +ReservedSpecializationConstants::BACKEND_FEATURE_LEVEL,
                     (int)engine.getSupportedFeatureLevel() });
@@ -250,16 +258,16 @@ FMaterial::FMaterial(FEngine& engine, const Material::Builder& builder)
                     (int)maxFroxelBufferHeight });
     mSpecializationConstants.push_back({
                     +ReservedSpecializationConstants::CONFIG_DEBUG_DIRECTIONAL_SHADOWMAP,
-                    (bool)engine.debug.shadowmap.debug_directional_shadowmap });
+                    engine.debug.shadowmap.debug_directional_shadowmap });
     mSpecializationConstants.push_back({
                     +ReservedSpecializationConstants::CONFIG_DEBUG_FROXEL_VISUALIZATION,
-                    (bool)engine.debug.lighting.debug_froxel_visualization });
+                    engine.debug.lighting.debug_froxel_visualization });
     mSpecializationConstants.push_back({
                     +ReservedSpecializationConstants::CONFIG_STATIC_TEXTURE_TARGET_WORKAROUND,
-                    (bool)staticTextureWorkaround });
+                    staticTextureWorkaround });
     mSpecializationConstants.push_back({
                     +ReservedSpecializationConstants::CONFIG_POWER_VR_SHADER_WORKAROUNDS,
-                    (bool)powerVrShaderWorkarounds });
+                    powerVrShaderWorkarounds });
     mSpecializationConstants.push_back({
                     +ReservedSpecializationConstants::CONFIG_STEREO_EYE_COUNT,
                     (int)engine.getConfig().stereoscopicEyeCount });
@@ -270,19 +278,18 @@ FMaterial::FMaterial(FEngine& engine, const Material::Builder& builder)
             false});
     }
 
-    for (const auto& [name, value] : builder->mConstantSpecializations) {
-        auto found = std::find_if(
-                constants.begin(), constants.end(), [name = name](const auto& constant) {
-                    return strncmp(constant.name.data(), name.data(), name.length()) == 0;
-                });
-        ASSERT_PRECONDITION(found != constants.end(),
+    for (auto const& [name, value] : builder->mConstantSpecializations) {
+        std::string_view const key{ name.data(), name.size() };
+        auto pos = mSpecializationConstantsNameToIndex.find(key);
+        ASSERT_PRECONDITION(pos != mSpecializationConstantsNameToIndex.end(),
                 "The material %s does not have a constant parameter named %s.",
                 mName.c_str_safe(), name.c_str());
         const char* const types[3] = {"an int", "a float", "a bool"};
         const char* const errorMessage =
                 "The constant parameter %s on material %s is of type %s, but %s was "
                 "provided.";
-        switch (found->type) {
+        auto& constant = mMaterialConstants[pos->second];
+        switch (constant.type) {
             case ConstantType::INT:
                 ASSERT_PRECONDITION(std::holds_alternative<int32_t>(value), errorMessage,
                         name.c_str(), mName.c_str_safe(), "int", types[value.index()]);
@@ -296,8 +303,7 @@ FMaterial::FMaterial(FEngine& engine, const Material::Builder& builder)
                         name.c_str(), mName.c_str_safe(), "bool", types[value.index()]);
                 break;
         }
-        uint32_t const index =
-                std::distance(constants.begin(), found) + CONFIG_MAX_RESERVED_SPEC_CONSTANTS;
+        uint32_t const index = pos->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS;
         mSpecializationConstants.push_back({ index, value });
     }
 
@@ -462,16 +468,6 @@ FMaterial::~FMaterial() noexcept {
 }
 
 void FMaterial::invalidate(Variant::type_t variantMask, Variant::type_t variantValue) noexcept {
-    // update the spec constants that can change
-    // TODO: should we just always update all of them?
-    for (auto& item : mSpecializationConstants) {
-        if (item.id == ReservedSpecializationConstants::CONFIG_DEBUG_DIRECTIONAL_SHADOWMAP) {
-            item.value = mEngine.debug.shadowmap.debug_directional_shadowmap;
-        } else if (item.id == ReservedSpecializationConstants::CONFIG_DEBUG_FROXEL_VISUALIZATION) {
-            item.value = mEngine.debug.lighting.debug_froxel_visualization;
-        }
-    }
-
     if (mMaterialDomain == MaterialDomain::SURFACE) {
         DriverApi& driverApi = mEngine.getDriverApi();
         auto& cachedPrograms = mCachedPrograms;
@@ -841,4 +837,55 @@ void FMaterial::destroyPrograms(FEngine& engine) {
     }
 }
 
+std::optional<uint32_t> FMaterial::getSpecializationConstantId(std::string_view name) const noexcept {
+    auto pos = mSpecializationConstantsNameToIndex.find(name);
+    if (pos != mSpecializationConstantsNameToIndex.end()) {
+        return pos->second + CONFIG_MAX_RESERVED_SPEC_CONSTANTS;
+    }
+    return std::nullopt;
+}
+
+template<typename T, typename>
+bool FMaterial::setConstant(uint32_t id, T value) noexcept {
+    size_t const maxId = mMaterialConstants.size() + CONFIG_MAX_RESERVED_SPEC_CONSTANTS;
+    if (UTILS_LIKELY(id < maxId)) {
+        if (id >= CONFIG_MAX_RESERVED_SPEC_CONSTANTS) {
+            // Constant from the material itself (as opposed to the reserved ones)
+            auto& constant = mMaterialConstants[id - CONFIG_MAX_RESERVED_SPEC_CONSTANTS];
+            using ConstantType = backend::ConstantType;
+            switch (constant.type) {
+                case ConstantType::INT:
+                    if (!std::is_same_v<T, int32_t>) return false;
+                    break;
+                case ConstantType::FLOAT:
+                    if (!std::is_same_v<T, float>) return false;
+                    break;
+                case ConstantType::BOOL:
+                    if (!std::is_same_v<T, bool>) return false;
+                    break;
+            }
+        }
+
+        auto pos = std::find_if(
+                mSpecializationConstants.begin(), mSpecializationConstants.end(),
+                [id](backend::Program::SpecializationConstant const& specializationConstant) {
+                    return specializationConstant.id == id;
+                });
+        if (pos != mSpecializationConstants.end()) {
+            if (std::get<T>(pos->value) != value) {
+                pos->value = value;
+                return true;
+            }
+        } else {
+            mSpecializationConstants.push_back({ id, value });
+            return true;
+        }
+    }
+    return false;
+}
+
+template bool FMaterial::setConstant<int32_t>(uint32_t id, int32_t value) noexcept;
+template bool FMaterial::setConstant<float>(uint32_t id, float value) noexcept;
+template bool FMaterial::setConstant<bool>(uint32_t id, bool value) noexcept;
+
 } // namespace filament

filament/src/details/Material.h

@@ -165,6 +165,14 @@ public:
 
     void destroyPrograms(FEngine& engine);
 
+    // return the id of a specialization constant specified by name for this material
+    std::optional<uint32_t> getSpecializationConstantId(std::string_view name) const noexcept ;
+
+    // Sets a specialization constant by id. call is no-op if the id is invalid.
+    // Return true is the value was changed.
+    template<typename T, typename = Builder::is_supported_constant_parameter_t<T>>
+    bool setConstant(uint32_t id, T value) noexcept;
+
 #if FILAMENT_ENABLE_MATDBG
     void applyPendingEdits() noexcept;
 
@@ -261,6 +269,11 @@ private:
 
     SamplerGroupBindingInfoList mSamplerGroupBindingInfoList;
     SamplerBindingToNameMap mSamplerBindingToNameMap;
+    // Constants defined by this Material
+    utils::FixedCapacityVector<MaterialConstant> mMaterialConstants;
+    // A map from the Constant name to the mMaterialConstant index
+    std::unordered_map<std::string_view, uint32_t> mSpecializationConstantsNameToIndex;
+    // current specialization constants for the HwProgram
     utils::FixedCapacityVector<backend::Program::SpecializationConstant> mSpecializationConstants;
 
 #if FILAMENT_ENABLE_MATDBG

filament/src/details/MaterialInstance.cpp

@@ -117,7 +117,7 @@ void FMaterialInstance::initDefaultInstance(FEngine& engine, FMaterial const* ma
     if (!material->getSamplerInterfaceBlock().isEmpty()) {
         mSamplers = SamplerGroup(material->getSamplerInterfaceBlock().getSize());
         mSbHandle = driver.createSamplerGroup(
-                mSamplers.getSize(), utils::FixedSizeString<32>("Default material"));
+                mSamplers.getSize(), utils::FixedSizeString<32>(mMaterial->getName().c_str_safe()));
     }
 
     const RasterState& rasterState = material->getRasterState();

filament/src/details/Renderer.cpp

@@ -491,6 +491,19 @@ void FRenderer::renderJob(ArenaScope& arena, FView& view) {
         hasDithering = false;
         hasFXAA = false;
         scale = 1.0f;
+    } else {
+        // This configures post-process materials by setting constant parameters
+        if (taaOptions.enabled) {
+            ppm.configureTemporalAntiAliasingMaterial(taaOptions);
+            if (taaOptions.upscaling) {
+                // for now TAA upscaling is incompatible with regular dsr
+                dsrOptions.enabled = false;
+                // also, upscaling doesn't work well with quater-resolution SSAO
+                aoOptions.resolution = 1.0;
+                // Currently we only support a fixed TAA upscaling ratio
+                scale = 0.5f;
+            }
+        }
     }
 
     const bool blendModeTranslucent = view.getBlendMode() == BlendMode::TRANSLUCENT;
@@ -520,7 +533,7 @@ void FRenderer::renderJob(ArenaScope& arena, FView& view) {
     };
 
     // whether we're scaled at all
-    const bool scaled = any(notEqual(scale, float2(1.0f)));
+     bool scaled = any(notEqual(scale, float2(1.0f)));
 
     // vp is the user defined viewport within the View
     filament::Viewport const& vp = view.getViewport();
@@ -609,7 +622,7 @@ void FRenderer::renderJob(ArenaScope& arena, FView& view) {
 
     view.prepare(engine, driver, arena, svp, cameraInfo, getShaderUserTime(), needsAlphaChannel);
 
-    view.prepareUpscaler(scale);
+    view.prepareUpscaler(scale, taaOptions, dsrOptions);
 
     /*
      * Allocate command buffer
@@ -813,14 +826,14 @@ void FRenderer::renderJob(ArenaScope& arena, FView& view) {
                 [=, &view](FrameGraphResources const& resources,
                         auto const&, DriverApi& driver) mutable {
                     auto out = resources.getRenderPassInfo();
-                    view.executePickingQueries(driver, out.target, aoOptions.resolution);
+                    view.executePickingQueries(driver, out.target, scale * aoOptions.resolution);
                 });
     }
 
     // Store this frame's camera projection in the frame history.
     if (UTILS_UNLIKELY(taaOptions.enabled)) {
         // Apply the TAA jitter to everything after the structure pass, starting with the color pass.
-        ppm.prepareTaa(fg, svp, view.getFrameHistory(), &FrameHistoryEntry::taa,
+        ppm.prepareTaa(fg, svp, taaOptions, view.getFrameHistory(), &FrameHistoryEntry::taa,
                 &cameraInfo, view.getPerViewUniforms());
     }
 
@@ -1006,6 +1019,15 @@ void FRenderer::renderJob(ArenaScope& arena, FView& view) {
     if (taaOptions.enabled) {
         input = ppm.taa(fg, input, depth, view.getFrameHistory(), &FrameHistoryEntry::taa,
                 taaOptions, colorGradingConfig);
+        if (taaOptions.upscaling) {
+            scale = 1.0f;
+            scaled = false;
+            UTILS_UNUSED_IN_RELEASE auto const& inputDesc = fg.getDescriptor(input);
+            svp.width = inputDesc.width;
+            svp.height = inputDesc.height;
+            xvp.width *= 2;
+            xvp.height *= 2;
+        }
     }
 
     // --------------------------------------------------------------------------------------------
@@ -1021,16 +1043,21 @@ void FRenderer::renderJob(ArenaScope& arena, FView& view) {
             // The bokeh height is always correct regardless of the dynamic resolution scaling.
             // (because the CoC is calculated w.r.t. the height), so we only need to adjust
             // the width.
-            float const bokehAspectRatio = scale.x / scale.y;
+            float const aspect = (scale.x / scale.y) * dofOptions.cocAspectRatio;
+            float2 const bokehScale{
+                aspect < 1.0f ? aspect : 1.0f,
+                aspect > 1.0f ? 1.0f / aspect : 1.0f
+            };
             input = ppm.dof(fg, input, depth, cameraInfo, needsAlphaChannel,
-                    bokehAspectRatio, dofOptions);
+                    bokehScale, dofOptions);
         }
 
         FrameGraphId<FrameGraphTexture> bloom, flare;
         if (bloomOptions.enabled) {
-            // generate the bloom buffer, which is stored in the blackboard as "bloom". This is
+            // Generate the bloom buffer, which is stored in the blackboard as "bloom". This is
             // consumed by the colorGrading pass and will be culled if colorGrading is disabled.
-            auto [bloom_, flare_] = ppm.bloom(fg, input, bloomOptions, TextureFormat::R11F_G11F_B10F, scale);
+            auto [bloom_, flare_] = ppm.bloom(fg, input, TextureFormat::R11F_G11F_B10F,
+                    bloomOptions, taaOptions, scale);
             bloom = bloom_;
             flare = flare_;
         }

filament/src/details/View.cpp

@@ -313,6 +313,10 @@ void FView::prepareShadowing(FEngine& engine, FScene::RenderableSoa& renderableD
         }
 
         if (UTILS_LIKELY(!lcm.isShadowCaster(li))) {
+            // Because we early exit here, we need to make sure we mark the light as non-casting.
+            // See `ShadowMapManager::updateSpotShadowMaps` for const_cast<> justification.
+            const_cast<FScene::ShadowInfo&>(
+                    lightData.elementAt<FScene::SHADOW_INFO>(l)).castsShadows = false;
             continue; // doesn't cast shadows
         }
 
@@ -721,11 +725,22 @@ UTILS_NOINLINE
     });
 }
 
-void FView::prepareUpscaler(float2 scale) const noexcept {
+void FView::prepareUpscaler(float2 scale,
+        TemporalAntiAliasingOptions const& taaOptions,
+        DynamicResolutionOptions const& dsrOptions) const noexcept {
     SYSTRACE_CALL();
-    const float bias = (mDynamicResolution.quality >= QualityLevel::HIGH) ?
-            std::log2(std::min(scale.x, scale.y)) : 0.0f;
-    mPerViewUniforms.prepareLodBias(bias);
+    float bias = 0.0f;
+    float2 derivativesScale{ 1.0f };
+    if (dsrOptions.enabled && dsrOptions.quality >= QualityLevel::HIGH) {
+        bias = std::log2(std::min(scale.x, scale.y));
+    }
+    if (taaOptions.enabled) {
+        bias += taaOptions.lodBias;
+        if (taaOptions.upscaling) {
+            derivativesScale = 0.5f;
+        }
+    }
+    mPerViewUniforms.prepareLodBias(bias, derivativesScale);
 }
 
 void FView::prepareCamera(FEngine& engine, const CameraInfo& cameraInfo) const noexcept {
@@ -981,15 +996,15 @@ void FView::drainFrameHistory(FEngine& engine) noexcept {
 }
 
 void FView::executePickingQueries(backend::DriverApi& driver,
-        backend::RenderTargetHandle handle, float scale) noexcept {
+        backend::RenderTargetHandle handle, float2 scale) noexcept {
 
     while (mActivePickingQueriesList) {
         FPickingQuery* const pQuery = mActivePickingQueriesList;
         mActivePickingQueriesList = pQuery->next;
 
         // adjust for dynamic resolution and structure buffer scale
-        const uint32_t x = uint32_t(float(pQuery->x) * (scale * mScale.x));
-        const uint32_t y = uint32_t(float(pQuery->y) * (scale * mScale.y));
+        const uint32_t x = uint32_t(float(pQuery->x) * scale.x);
+        const uint32_t y = uint32_t(float(pQuery->y) * scale.y);
 
         if (UTILS_UNLIKELY(driver.getFeatureLevel() == FeatureLevel::FEATURE_LEVEL_0)) {
             driver.readPixels(handle, x, y, 1, 1, {

filament/src/details/View.h

@@ -133,7 +133,9 @@ public:
         return mName.c_str_safe();
     }
 
-    void prepareUpscaler(math::float2 scale) const noexcept;
+    void prepareUpscaler(math::float2 scale,
+            TemporalAntiAliasingOptions const& taaOptions,
+            DynamicResolutionOptions const& dsrOptions) const noexcept;
     void prepareCamera(FEngine& engine, const CameraInfo& cameraInfo) const noexcept;
 
     void prepareViewport(
@@ -417,7 +419,7 @@ public:
             View::PickingQueryResultCallback callback) noexcept;
 
     void executePickingQueries(backend::DriverApi& driver,
-            backend::RenderTargetHandle handle, float scale) noexcept;
+            backend::RenderTargetHandle handle, math::float2 scale) noexcept;
 
     void setMaterialGlobal(uint32_t index, math::float4 const& value);
 

filament/src/fg/PassNode.cpp

@@ -229,6 +229,11 @@ void RenderPassNode::resolve() noexcept {
             rt.descriptor.samples    = pImportedRenderTarget->importedDesc.samples;
             rt.backend.target        = pImportedRenderTarget->target;
 
+            // We could end-up here more than once, for instance if the rendertarget is used
+            // by multiple passes (this would imply a read-back, btw). In this case, we don't want
+            // to clear it the 2nd time, so we clear the imported pass's clear flags.
+            pImportedRenderTarget->importedDesc.clearFlags = TargetBufferFlags::NONE;
+
             // but don't discard attachments the imported target tells us to keep
             rt.backend.params.flags.discardStart &= ~pImportedRenderTarget->importedDesc.keepOverrideStart;
             rt.backend.params.flags.discardEnd   &= ~pImportedRenderTarget->importedDesc.keepOverrideEnd;

filament/src/materials/antiAliasing/taa.mat

@@ -24,10 +24,57 @@ material {
             precision: high
         },
         {
-            type : float[9],
+            type : float2,
+            name : jitter,
+            precision: high
+        },
+        {
+            type : float4[9],
             name : filterWeights
         }
     ],
+    constants : [
+        {
+            type : bool,
+            name : historyReprojection,
+            default : true
+        },
+        {
+            type : bool,
+            name : filterHistory,
+            default : true
+        },
+        {
+            type : bool,
+            name : filterInput,
+            default : true
+        },
+        {
+            type : int,
+            name : boxClipping,
+            default : 0
+        },
+        {
+            type : int,
+            name : boxType,
+            default : 1
+        },
+        {
+            type : bool,
+            name : useYCoCg,
+            default : false
+        },
+        {
+            type : bool,
+            name : preventFlickering,
+            default : false
+        },
+        {
+            type : float,
+            name : varianceGamma,
+            default : 1.0
+        }
+    ],
     variables : [
         vertex
     ],
@@ -44,14 +91,6 @@ vertex {
 
 fragment {
 
-#if defined(TARGET_METAL_ENVIRONMENT) || defined(TARGET_VULKAN_ENVIRONMENT)
-#define TEXTURE_SPACE_UP    -1
-#define TEXTURE_SPACE_DN     1
-#else
-#define TEXTURE_SPACE_UP     1
-#define TEXTURE_SPACE_DN    -1
-#endif
-
 /* Clipping box type */
 
 // min/max neighborhood
@@ -61,9 +100,6 @@ fragment {
 // uses both min/max and variance
 #define BOX_TYPE_AABB_VARIANCE  2
 
-// High VARIANCE_GAMMA [0.75, 1.25] increases ghosting artefact, lower values increases jittering
-#define VARIANCE_GAMMA          1.0
-
 /* Clipping algorithm */
 
 // accurate box clipping
@@ -73,34 +109,12 @@ fragment {
 // no clipping (for debugging only)
 #define BOX_CLIPPING_NONE       2
 
-/* Configuration mobile/desktop */
-
-#if FILAMENT_QUALITY < FILAMENT_QUALITY_HIGH
-#define BOX_CLIPPING            BOX_CLIPPING_ACCURATE
-#define BOX_TYPE                BOX_TYPE_VARIANCE
-#define USE_YCoCg               0
-#define FILTER_INPUT            1
-#define FILTER_HISTORY          0
-#else
-#define BOX_CLIPPING            BOX_CLIPPING_ACCURATE
-#define BOX_TYPE                BOX_TYPE_AABB_VARIANCE
-#define USE_YCoCg               1
-#define FILTER_INPUT            1
-#define FILTER_HISTORY          1
-#endif
-
-/* debugging helper */
-
-#define HISTORY_REPROJECTION    1
-#define PREVENT_FLICKERING      0   // FIXME: thin lines disapear
-
-
 float luma(const vec3 color) {
-#if USE_YCoCg
-    return color.x;
-#else
-    return luminance(color);
-#endif
+    if (materialConstants_useYCoCg) {
+        return color.x;
+    } else {
+        return luminance(color);
+    }
 }
 
 vec3 RGB_YCoCg(const vec3 c) {
@@ -194,108 +208,136 @@ vec4 sampleTextureCatmullRom(const sampler2D tex, const highp vec2 uv, const hig
 
     result *= 1.0 / (k0 + k1 + k2 + k3 + k4);
 
+    // we could end-up with negative values
+    result = max(vec4(0), result);
+
     return result;
 }
 
 void postProcess(inout PostProcessInputs postProcess) {
     highp vec4 uv = variable_vertex.xyxy; // interpolated to pixel center
 
-    // read the depth buffer center sample for reprojection
-    float depth = textureLod(materialParams_depth, uv.xy, 0.0).r;
-
-#if HISTORY_REPROJECTION
-    uv.zw = uvToRenderTargetUV(uv.zw);
-    // reproject history to current frame
-    highp vec4 q = materialParams.reprojection * vec4(uv.zw, depth, 1.0);
-    uv.zw = (q.xy * (1.0 / q.w)) * 0.5 + 0.5;
-    uv.zw = uvToRenderTargetUV(uv.zw);
-#endif
+    if (materialConstants_historyReprojection) {
+        // read the depth buffer center sample for reprojection
+        highp float depth = textureLod(materialParams_depth, uv.xy, 0.0).r;
+        // reproject history to current frame
+        uv.zw = uvToRenderTargetUV(uv.zw);
+        highp vec4 q = materialParams.reprojection * vec4(uv.zw, depth, 1.0);
+        uv.zw = (q.xy * (1.0 / q.w)) * 0.5 + 0.5;
+        uv.zw = uvToRenderTargetUV(uv.zw);
+    }
 
     // read center color and history samples
-    vec4 color = textureLod(materialParams_color, uv.xy, 0.0);
-#if FILTER_HISTORY
-    vec4 history = sampleTextureCatmullRom(materialParams_history, uv.zw,
-            vec2(textureSize(materialParams_history, 0)));
-#else
-    vec4 history = textureLod(materialParams_history, uv.zw, 0.0);
-#endif
+    vec4 history;
+    if (materialConstants_filterHistory) {
+        history = sampleTextureCatmullRom(materialParams_history, uv.zw,
+                vec2(textureSize(materialParams_history, 0)));
+    } else {
+        history = textureLod(materialParams_history, uv.zw, 0.0);
+    }
 
-#if USE_YCoCg
-    history.rgb = RGB_YCoCg(history.rgb);
-#endif
+    if (materialConstants_useYCoCg) {
+        history.rgb = RGB_YCoCg(history.rgb);
+    }
+
+    highp vec2 size = vec2(textureSize(materialParams_color, 0));
+    highp vec2 p = (floor(uv.xy * size) + 0.5) / size;
+    vec4 color = textureLod(materialParams_color, p, 0.0);
+
+    vec3 s[9];
+    if (materialConstants_filterInput || materialConstants_boxClipping != BOX_CLIPPING_NONE) {
+        s[0] = textureLodOffset(materialParams_color, p, 0.0, ivec2(-1, -1)).rgb;
+        s[1] = textureLodOffset(materialParams_color, p, 0.0, ivec2( 0, -1)).rgb;
+        s[2] = textureLodOffset(materialParams_color, p, 0.0, ivec2( 1, -1)).rgb;
+        s[3] = textureLodOffset(materialParams_color, p, 0.0, ivec2(-1,  0)).rgb;
+        s[4] = color.rgb;
+        s[5] = textureLodOffset(materialParams_color, p, 0.0, ivec2( 1,  0)).rgb;
+        s[6] = textureLodOffset(materialParams_color, p, 0.0, ivec2(-1,  1)).rgb;
+        s[7] = textureLodOffset(materialParams_color, p, 0.0, ivec2( 0,  1)).rgb;
+        s[8] = textureLodOffset(materialParams_color, p, 0.0, ivec2( 1,  1)).rgb;
+    }
+
+    if (materialConstants_useYCoCg) {
+        color.rgb = RGB_YCoCg(color.rgb);
+        if (materialConstants_filterInput || materialConstants_boxClipping != BOX_CLIPPING_NONE) {
+            for (int i = 0; i < 9; i++) {
+                s[i] = RGB_YCoCg(s[i]);
+            }
+        }
+    }
+
+    vec2 subPixelOffset = p - uv.xy;  // +/- [0.25, 0.25]
+    float confidence = 0.0;
+
+    vec4 filtered = color;
+    if (materialConstants_filterInput) {
+        // unjitter/filter input
+        // figure out which set of coeficients to use
+        int jxp = subPixelOffset.y > 0.0 ? 3 : 0;
+        int jxn = subPixelOffset.y > 0.0 ? 2 : 1;
+        int j   = subPixelOffset.x > 0.0 ? jxp : jxn;
+        filtered = vec4(0, 0, 0, color.a);
+        for (int i = 0; i < 9; i++) {
+            float w = materialParams.filterWeights[i][j];
+            filtered.rgb += s[i] * w;
+            confidence = max(confidence, w);
+        }
+    } else {
+        confidence = float(materialParams.jitter.x * subPixelOffset.x > 0.0 &&
+                           materialParams.jitter.y * subPixelOffset.y > 0.0);
+    }
 
     // build the history clamping box
-    vec3 s[9];
-    s[0] = textureLodOffset(materialParams_color, uv.xy, 0.0, ivec2(-1, TEXTURE_SPACE_DN)).rgb;
-    s[1] = textureLodOffset(materialParams_color, uv.xy, 0.0, ivec2( 0, TEXTURE_SPACE_DN)).rgb;
-    s[2] = textureLodOffset(materialParams_color, uv.xy, 0.0, ivec2( 1, TEXTURE_SPACE_DN)).rgb;
-    s[3] = textureLodOffset(materialParams_color, uv.xy, 0.0, ivec2(-1, 0)).rgb;
-    s[4] = color.rgb;
-    s[5] = textureLodOffset(materialParams_color, uv.xy, 0.0, ivec2( 1, 0)).rgb;
-    s[6] = textureLodOffset(materialParams_color, uv.xy, 0.0, ivec2(-1, TEXTURE_SPACE_UP)).rgb;
-    s[7] = textureLodOffset(materialParams_color, uv.xy, 0.0, ivec2( 0, TEXTURE_SPACE_UP)).rgb;
-    s[8] = textureLodOffset(materialParams_color, uv.xy, 0.0, ivec2( 1, TEXTURE_SPACE_UP)).rgb;
+    if (materialConstants_boxClipping != BOX_CLIPPING_NONE) {
+        vec3 boxmin;
+        vec3 boxmax;
+        if (materialConstants_boxType == BOX_TYPE_AABB ||
+                materialConstants_boxType == BOX_TYPE_AABB_VARIANCE) {
+            boxmin = min(s[4], min(min(s[1], s[3]), min(s[5], s[7])));
+            boxmax = max(s[4], max(max(s[1], s[3]), max(s[5], s[7])));
+            vec3 box9min = min(boxmin, min(min(s[0], s[2]), min(s[6], s[8])));
+            vec3 box9max = max(boxmax, max(max(s[0], s[2]), max(s[6], s[8])));
+            // round the corners of the 3x3 box
+            boxmin = (boxmin + box9min) * 0.5;
+            boxmax = (boxmax + box9max) * 0.5;
+        }
 
-#if USE_YCoCg
-    for (int i = 0 ; i < 9 ; i++) {
-        s[i] = RGB_YCoCg(s[i]);
+        if (materialConstants_boxType == BOX_TYPE_VARIANCE ||
+                materialConstants_boxType == BOX_TYPE_AABB_VARIANCE) {
+            // "An Excursion in Temporal Supersampling" by Marco Salvi
+            highp vec3 m0 = s[4];// conversion to highp
+            highp vec3 m1 = m0 * m0;
+            // we use only 5 samples instead of all 9
+            for (int i = 1; i < 9; i+=2) {
+                highp vec3 c = s[i];// conversion to highp
+                m0 += c;
+                m1 += c * c;
+            }
+            highp vec3 a0 = m0 * (1.0 / 5.0);
+            highp vec3 a1 = m1 * (1.0 / 5.0);
+            highp vec3 sigma = sqrt(a1 - a0 * a0);
+            if (materialConstants_boxType == BOX_TYPE_VARIANCE) {
+                boxmin = a0 - materialConstants_varianceGamma * sigma;
+                boxmax = a0 + materialConstants_varianceGamma * sigma;
+            } else {
+                // intersect both bounding boxes
+                boxmin = max(boxmin, a0 - materialConstants_varianceGamma * sigma);
+                boxmax = min(boxmax, a0 + materialConstants_varianceGamma * sigma);
+            }
+        }
+        // history clamping
+        history = clipToBox(materialConstants_boxClipping, boxmin, boxmax, filtered, history);
     }
-    color.rgb = s[4].rgb;
-#endif
-
-#if FILTER_INPUT
-    // unjitter/filter input
-    vec4 filtered = vec4(0, 0, 0, color.a);
-    for (int i = 0 ; i < 9 ; i++) {
-        filtered.rgb += s[i] * materialParams.filterWeights[i];
-    }
-#else
-    vec4 filtered = color;
-#endif
-
-#if BOX_TYPE == BOX_TYPE_AABB || BOX_TYPE == BOX_TYPE_AABB_VARIANCE
-    vec3 boxmin = min(s[4], min(min(s[1], s[3]), min(s[5], s[7])));
-    vec3 boxmax = max(s[4], max(max(s[1], s[3]), max(s[5], s[7])));
-    vec3 box9min = min(boxmin, min(min(s[0], s[2]), min(s[6], s[8])));
-    vec3 box9max = max(boxmax, max(max(s[0], s[2]), max(s[6], s[8])));
-    // round the corners of the 3x3 box
-    boxmin = (boxmin + box9min) * 0.5;
-    boxmax = (boxmax + box9max) * 0.5;
-#endif
-
-#if BOX_TYPE == BOX_TYPE_VARIANCE || BOX_TYPE == BOX_TYPE_AABB_VARIANCE
-    // "An Excursion in Temporal Supersampling" by Marco Salvi
-    vec3 m0 = s[4];
-    vec3 m1 = s[4] * s[4];
-    // we use only 5 samples instead of all 9
-    for (int i = 1 ; i < 9 ; i+=2) {
-        m0 += s[i];
-        m1 += s[i] * s[i];
-    }
-    vec3 a0 = m0 * (1.0 / 5.0);
-    vec3 a1 = m1 * (1.0 / 5.0);
-    vec3 sigma = sqrt(a1 - a0 * a0);
-#if BOX_TYPE == BOX_TYPE_VARIANCE
-    vec3 boxmin = a0 - VARIANCE_GAMMA * sigma;
-    vec3 boxmax = a0 + VARIANCE_GAMMA * sigma;
-#else
-    boxmin = min(boxmin, a0 - VARIANCE_GAMMA * sigma);
-    boxmax = max(boxmax, a0 + VARIANCE_GAMMA * sigma);
-#endif
-#endif
-
-    // history clamping
-    history = clipToBox(BOX_CLIPPING, boxmin, boxmax, filtered, history);
 
     float lumaColor   = luma(filtered.rgb);
     float lumaHistory = luma(history.rgb);
 
-    float alpha = materialParams.alpha;
-#if PREVENT_FLICKERING
-    // [Lottes] prevents flickering by modulating the blend weight by the difference in luma
-    float diff = 1.0 - abs(lumaColor - lumaHistory) / (0.001 + max(lumaColor, lumaHistory));
-    alpha *= diff * diff;
-#endif
+    float alpha = materialParams.alpha * confidence;
+    if (materialConstants_preventFlickering) {
+        // [Lottes] prevents flickering by modulating the blend weight by the difference in luma
+        float diff = 1.0 - abs(lumaColor - lumaHistory) / (0.001 + max(lumaColor, lumaHistory));
+        alpha *= diff * diff;
+    }
 
     // tonemapping for handling HDR
     filtered.rgb *= 1.0 / (1.0 + lumaColor);
@@ -307,20 +349,16 @@ void postProcess(inout PostProcessInputs postProcess) {
     // untonemap result
     result.rgb *= 1.0 / (1.0 - luma(result.rgb));
 
-#if USE_YCoCg
-    result.rgb = YCoCg_RGB(result.rgb);
-#endif
-
-    // store result (which will becomes new history)
-    // we could end-up with negative values due to the bicubic filter, which never recover on
-    // their own.
-    result = max(vec4(0), result);
+    if (materialConstants_useYCoCg) {
+        result.rgb = YCoCg_RGB(result.rgb);
+    }
 
 #if POST_PROCESS_OPAQUE
     // kill the work performed above
     result.a = 1.0;
 #endif
 
+    // store result (which will becomes new history)
     postProcess.color = result;
 }
 

filament/src/materials/resolveDepth.mat

@@ -0,0 +1,33 @@
+material {
+    name : resolveDepth,
+    parameters : [
+        {
+            type : sampler2d,
+            name : depth,
+            precision: high,
+            multisample : true
+        }
+    ],
+    outputs : [
+        {
+            name : depth,
+            target : depth,
+            type : float
+        }
+    ],
+    domain : postprocess,
+    depthWrite : true,
+    depthCulling : false,
+    culling: none
+}
+
+fragment {
+    // resolve depth by keeping the first sample. This is similar to what glBlitFramebuffer() does
+    // this material is currently only used on Metal and Vulkan.
+    // Technically it should be feature-level 2 material, but that would restrict when we can use
+    // it because Vulkan can return feature-level 1 currently (we should fix that).
+    void postProcess(inout PostProcessInputs postProcess) {
+        highp ivec2 icoord = ivec2(gl_FragCoord.xy);
+        postProcess.depth = texelFetch(materialParams_depth, icoord, 0).r;
+    }
+}

ios/CocoaPods/Filament.podspec

@@ -1,12 +1,12 @@
 Pod::Spec.new do |spec|
   spec.name = "Filament"
-  spec.version = "1.49.1"
+  spec.version = "1.49.3"
   spec.license = { :type => "Apache 2.0", :file => "LICENSE" }
   spec.homepage = "https://google.github.io/filament"
   spec.authors = "Google LLC."
   spec.summary = "Filament is a real-time physically based rendering engine for Android, iOS, Windows, Linux, macOS, and WASM/WebGL."
   spec.platform = :ios, "11.0"
-  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.49.1/filament-v1.49.1-ios.tgz" }
+  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.49.3/filament-v1.49.3-ios.tgz" }
 
   # Fix linking error with Xcode 12; we do not yet support the simulator on Apple silicon.
   spec.pod_target_xcconfig = {

ios/samples/gltf-viewer/gltf-viewer/FILModelView.h

@@ -16,6 +16,8 @@
 
 #import <UIKit/UIKit.h>
 
+#include <utils/Entity.h>
+
 namespace filament {
 class Engine;
 class Scene;
@@ -76,6 +78,17 @@ typedef NSData* _Nonnull (^ResourceCallback)(NSString* _Nonnull);
 
 - (void)destroyModel;
 
+/**
+ * Issues a pick query at the given view coordinates.
+ * The coordinates should be in UIKit's coordinate system, with the origin at
+ * the top-left.
+ * The callback is triggered with entity of the picked object.
+ */
+typedef void (^PickCallback)(utils::Entity);
+- (void)issuePickQuery:(CGPoint)point callback:(PickCallback)callback;
+
+- (NSString* _Nullable)getEntityName:(utils::Entity)entity;
+
 /**
  * Sets up a root transform on the current model to make it fit into a unit cube.
  */

ios/samples/gltf-viewer/gltf-viewer/FILModelView.mm

@@ -127,13 +127,12 @@ const float kSensitivity = 100.0f;
 
     _swapChain = _engine->createSwapChain((__bridge void*)self.layer);
 
-    _materialProvider = createUbershaderProvider(_engine,
-            UBERARCHIVE_DEFAULT_DATA, UBERARCHIVE_DEFAULT_SIZE);
+    _materialProvider =
+            createUbershaderProvider(_engine, UBERARCHIVE_DEFAULT_DATA, UBERARCHIVE_DEFAULT_SIZE);
     EntityManager& em = EntityManager::get();
     NameComponentManager* ncm = new NameComponentManager(em);
     _assetLoader = AssetLoader::create({_engine, _materialProvider, ncm, &em});
-    _resourceLoader = new ResourceLoader(
-            {.engine = _engine, .normalizeSkinningWeights = true});
+    _resourceLoader = new ResourceLoader({.engine = _engine, .normalizeSkinningWeights = true});
     _stbDecoder = createStbProvider(_engine);
     _ktxDecoder = createKtx2Provider(_engine);
     _resourceLoader->addTextureProvider("image/png", _stbDecoder);
@@ -182,6 +181,26 @@ const float kSensitivity = 100.0f;
     _animator = nullptr;
 }
 
+- (void)issuePickQuery:(CGPoint)point callback:(PickCallback)callback {
+    CGPoint pointOriginBottomLeft = CGPointMake(point.x, self.bounds.size.height - point.y);
+    CGPoint pointScaled = CGPointMake(pointOriginBottomLeft.x * self.contentScaleFactor,
+            pointOriginBottomLeft.y * self.contentScaleFactor);
+    _view->pick(pointScaled.x, pointScaled.y,
+            [callback](View::PickingQueryResult const& result) { callback(result.renderable); });
+}
+
+- (NSString* _Nullable)getEntityName:(utils::Entity)entity {
+    if (!_asset) {
+        return nil;
+    }
+    NameComponentManager* ncm = _assetLoader->getNames();
+    NameComponentManager::Instance instance = ncm->getInstance(entity);
+    if (instance) {
+        return [NSString stringWithUTF8String:ncm->getName(ncm->getInstance(entity))];
+    }
+    return nil;
+}
+
 - (void)transformToUnitCube {
     if (!_asset) {
         return;

ios/samples/gltf-viewer/gltf-viewer/FILViewController.mm

@@ -15,6 +15,7 @@
  */
 
 #import "FILViewController.h"
+#include <UIKit/UIKit.h>
 
 #import "FILModelView.h"
 
@@ -35,6 +36,9 @@ using namespace filament;
 using namespace utils;
 using namespace ktxreader;
 
+const float kToastAnimationDuration = 0.25f;
+const float kToastDelayDuration = 2.0f;
+
 @interface FILViewController ()
 
 - (void)startDisplayLink;
@@ -43,6 +47,9 @@ using namespace ktxreader;
 - (void)createRenderables;
 - (void)createLights;
 
+- (void)appWillResignActive:(NSNotification*)notification;
+- (void)appDidBecomeActive:(NSNotification*)notification;
+
 @end
 
 @implementation FILViewController {
@@ -50,6 +57,7 @@ using namespace ktxreader;
     CFTimeInterval _startTime;
     viewer::RemoteServer* _server;
     viewer::AutomationEngine* _automation;
+    UILabel* _toastLabel;
 
     Texture* _skyboxTexture;
     Skybox* _skybox;
@@ -57,6 +65,7 @@ using namespace ktxreader;
     IndirectLight* _indirectLight;
     Entity _sun;
 
+    UITapGestureRecognizer* _singleTapRecognizer;
     UITapGestureRecognizer* _doubleTapRecognizer;
 }
 
@@ -67,6 +76,16 @@ using namespace ktxreader;
 
     self.title = @"https://google.github.io/filament/remote";
 
+    // Observe lifecycle notifications to prevent us from rendering in the background.
+    [NSNotificationCenter.defaultCenter addObserver:self
+                                           selector:@selector(appWillResignActive:)
+                                               name:UIApplicationWillResignActiveNotification
+                                             object:nil];
+    [NSNotificationCenter.defaultCenter addObserver:self
+                                           selector:@selector(appDidBecomeActive:)
+                                               name:UIApplicationDidBecomeActiveNotification
+                                             object:nil];
+
     // Arguments:
     // --model <path>
     //     path to glb or gltf file to load from documents directory
@@ -94,9 +113,37 @@ using namespace ktxreader;
     _server = new viewer::RemoteServer();
     _automation = viewer::AutomationEngine::createDefault();
 
-    _doubleTapRecognizer = [[UITapGestureRecognizer alloc] initWithTarget:self action:@selector(reloadModel)];
+    _doubleTapRecognizer = [[UITapGestureRecognizer alloc] initWithTarget:self
+                                                                   action:@selector(reloadModel)];
     _doubleTapRecognizer.numberOfTapsRequired = 2;
     [self.modelView addGestureRecognizer:_doubleTapRecognizer];
+    _singleTapRecognizer =
+            [[UITapGestureRecognizer alloc] initWithTarget:self
+                                                    action:@selector(issuePickingQuery)];
+    _singleTapRecognizer.numberOfTapsRequired = 1;
+    [self.modelView addGestureRecognizer:_singleTapRecognizer];
+
+    // Create a label at the top of the screen to toast messages to the user.
+    CGRect labelRect = self.view.bounds;
+    labelRect.size.height = 50;
+    _toastLabel = [[UILabel alloc] initWithFrame:labelRect];
+    _toastLabel.autoresizingMask = UIViewAutoresizingFlexibleWidth;
+    _toastLabel.textAlignment = NSTextAlignmentCenter;
+    _toastLabel.textColor = [UIColor whiteColor];
+    _toastLabel.backgroundColor = [UIColor colorWithWhite:0.0f alpha:0.5f];
+    _toastLabel.numberOfLines = 0;
+    _toastLabel.lineBreakMode = NSLineBreakByWordWrapping;
+    _toastLabel.text = @"";
+    _toastLabel.alpha = 0.0f;
+    [self.view addSubview:_toastLabel];
+}
+
+- (void)appWillResignActive:(NSNotification*)notification {
+    [self stopDisplayLink];
+}
+
+- (void)appDidBecomeActive:(NSNotification*)notification {
+    [self startDisplayLink];
 }
 
 - (void)viewWillAppear:(BOOL)animated {
@@ -223,7 +270,8 @@ using namespace ktxreader;
         .indirectLight = _indirectLight,
         .sunlight = _sun,
     };
-    _automation->applySettings(self.modelView.engine, message->buffer, message->bufferByteCount, content);
+    _automation->applySettings(
+            self.modelView.engine, message->buffer, message->bufferByteCount, content);
     ColorGrading* const colorGrading = _automation->getColorGrading(self.modelView.engine);
     self.modelView.view->setColorGrading(colorGrading);
     self.modelView.cameraFocalLength = _automation->getViewerOptions().cameraFocalLength;
@@ -268,7 +316,42 @@ using namespace ktxreader;
     [self createDefaultRenderables];
 }
 
+- (void)issuePickingQuery {
+    CGPoint tapLocation = [_singleTapRecognizer locationInView:self.modelView];
+    __weak typeof(self) weakSelf = self;
+    [self.modelView issuePickQuery:tapLocation
+                          callback:^(utils::Entity entity) {
+                              NSString* name = [self.modelView getEntityName:entity];
+                              if (!name) {
+                                  name = @"<unnamed>";
+                              }
+                              NSString* message = [NSString
+                                      stringWithFormat:@"Picked entity %d (%@) at (%d,%d)",
+                                      entity.getId(), name, int(tapLocation.x), int(tapLocation.y)];
+                              [weakSelf toastMessage:message];
+                          }];
+}
+
+- (void)toastMessage:(NSString*)message {
+    _toastLabel.text = message;
+    _toastLabel.alpha = 0.0f;
+    [UIView animateWithDuration:kToastAnimationDuration
+            animations:^{
+                _toastLabel.alpha = 1.0f;
+            }
+            completion:^(BOOL finished) {
+                [UIView animateWithDuration:kToastAnimationDuration
+                                      delay:kToastDelayDuration
+                                    options:UIViewAnimationOptionCurveEaseInOut
+                                 animations:^{
+                                     _toastLabel.alpha = 0.0f;
+                                 }
+                                 completion:nil];
+            }];
+}
+
 - (void)dealloc {
+    [NSNotificationCenter.defaultCenter removeObserver:self];
     delete _server;
     delete _automation;
     self.modelView.engine->destroy(_indirectLight);

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

@@ -26,7 +26,9 @@
 
 namespace filament {
 
-static constexpr size_t POST_PROCESS_VARIANT_COUNT = 2;
+static constexpr size_t POST_PROCESS_VARIANT_BITS = 1;
+static constexpr size_t POST_PROCESS_VARIANT_COUNT = (1u << POST_PROCESS_VARIANT_BITS);
+static constexpr size_t POST_PROCESS_VARIANT_MASK = POST_PROCESS_VARIANT_COUNT - 1;
 enum class PostProcessVariant : uint8_t {
     OPAQUE,
     TRANSLUCENT
@@ -135,11 +137,15 @@ template<>
 struct utils::EnableIntegerOperators<filament::SamplerBindingPoints> : public std::true_type {};
 template<>
 struct utils::EnableIntegerOperators<filament::ReservedSpecializationConstants> : public std::true_type {};
+template<>
+struct utils::EnableIntegerOperators<filament::PostProcessVariant> : public std::true_type {};
 
 template<>
 inline constexpr size_t utils::Enum::count<filament::UniformBindingPoints>() { return 9; }
 template<>
 inline constexpr size_t utils::Enum::count<filament::SamplerBindingPoints>() { return 4; }
+template<>
+inline constexpr size_t utils::Enum::count<filament::PostProcessVariant>() { return filament::POST_PROCESS_VARIANT_COUNT; }
 
 static_assert(utils::Enum::count<filament::UniformBindingPoints>() <= filament::backend::CONFIG_UNIFORM_BINDING_COUNT);
 static_assert(utils::Enum::count<filament::SamplerBindingPoints>() <= filament::backend::CONFIG_SAMPLER_BINDING_COUNT);

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

@@ -99,6 +99,7 @@ struct PerViewUib { // NOLINT(cppcoreguidelines-pro-type-member-init)
 
     float lodBias;                  // load bias to apply to user materials
     float refractionLodOffset;
+    math::float2 derivativesScale;
 
     // camera position in view space (when camera_at_origin is enabled), i.e. it's (0,0,0).
     float oneOverFarMinusNear;      // 1 / (f-n), always positive
@@ -111,8 +112,6 @@ struct PerViewUib { // NOLINT(cppcoreguidelines-pro-type-member-init)
     // AO
     float aoSamplingQualityAndEdgeDistance;     // <0: no AO, 0: bilinear, !0: bilateral edge distance
     float aoBentNormals;                        // 0: no AO bent normal, >0.0 AO bent normals
-    float aoReserved0;
-    float aoReserved1;
 
     // --------------------------------------------------------------------------------------------
     // Dynamic Lighting [variant: DYN]

libs/filamat/include/filamat/MaterialBuilder.h

@@ -311,12 +311,8 @@ public:
      */
     MaterialBuilder& parameter(const char* name, SamplerType samplerType,
             SamplerFormat format = SamplerFormat::FLOAT,
-            ParameterPrecision precision = ParameterPrecision::DEFAULT) noexcept;
-
-    /// @copydoc parameter(SamplerType, SamplerFormat, ParameterPrecision, const char*)
-    MaterialBuilder& parameter(const char* name, SamplerType samplerType,
-            ParameterPrecision precision) noexcept;
-
+            ParameterPrecision precision = ParameterPrecision::DEFAULT,
+            bool multisample = false) noexcept;
 
     MaterialBuilder& buffer(filament::BufferInterfaceBlock bib) noexcept;
 
@@ -622,8 +618,8 @@ public:
         Parameter() noexcept: parameterType(INVALID) {}
 
         // Sampler
-        Parameter(const char* paramName, SamplerType t, SamplerFormat f, ParameterPrecision p)
-                : name(paramName), size(1), precision(p), samplerType(t), format(f), parameterType(SAMPLER) { }
+        Parameter(const char* paramName, SamplerType t, SamplerFormat f, ParameterPrecision p, bool ms)
+                : name(paramName), size(1), precision(p), samplerType(t), format(f), parameterType(SAMPLER), multisample(ms) { }
 
         // Uniform
         Parameter(const char* paramName, UniformType t, size_t typeSize, ParameterPrecision p)
@@ -640,6 +636,7 @@ public:
         SamplerType samplerType;
         SubpassType subpassType;
         SamplerFormat format;
+        bool multisample;
         enum {
             INVALID,
             UNIFORM,

libs/filamat/src/GLSLPostProcessor.cpp

@@ -234,7 +234,7 @@ void GLSLPostProcessor::spirvToMsl(const SpirvBlob *spirv, std::string *outMsl,
         for (const auto& info: infoList) {
             const std::string name = info.uniformName.c_str();
             argBufferBuilder
-                    .texture(info.offset * 2, name, info.type, info.format)
+                    .texture(info.offset * 2, name, info.type, info.format, info.multisample)
                     .sampler(info.offset * 2 + 1, name + "Smplr");
         }
 

libs/filamat/src/MaterialBuilder.cpp

@@ -250,15 +250,18 @@ MaterialBuilder& MaterialBuilder::parameter(const char* name, UniformType type,
 
 
 MaterialBuilder& MaterialBuilder::parameter(const char* name, SamplerType samplerType,
-        SamplerFormat format, ParameterPrecision precision) noexcept {
-    ASSERT_POSTCONDITION(mParameterCount < MAX_PARAMETERS_COUNT, "Too many parameters");
-    mParameters[mParameterCount++] = { name, samplerType, format, precision };
-    return *this;
-}
+        SamplerFormat format, ParameterPrecision precision, bool multisample) noexcept {
 
-MaterialBuilder& MaterialBuilder::parameter(const char* name, SamplerType samplerType,
-        ParameterPrecision precision) noexcept {
-    return parameter(name, samplerType, SamplerFormat::FLOAT, precision);
+    ASSERT_PRECONDITION(
+            !multisample || (format != SamplerFormat::SHADOW && (
+                                    samplerType == SamplerType::SAMPLER_2D
+                                            || samplerType == SamplerType::SAMPLER_2D_ARRAY)),
+            "multisample samplers only possible with SAMPLER_2D or SAMPLER_2D_ARRAY,"
+            " as long as type is not SHADOW");
+
+    ASSERT_POSTCONDITION(mParameterCount < MAX_PARAMETERS_COUNT, "Too many parameters");
+    mParameters[mParameterCount++] = { name, samplerType, format, precision, multisample };
+    return *this;
 }
 
 template<typename T, typename>
@@ -556,7 +559,7 @@ void MaterialBuilder::prepareToBuild(MaterialInfo& info) noexcept {
         assert_invariant(!param.isSubpass());
         if (param.isSampler()) {
             sbb.add({ param.name.data(), param.name.size() },
-                    param.samplerType, param.format, param.precision);
+                    param.samplerType, param.format, param.precision, param.multisample);
         } else if (param.isUniform()) {
             ibb.add({{{ param.name.data(), param.name.size() },
                       uint32_t(param.size == 1u ? 0u : param.size), param.uniformType,
@@ -1165,7 +1168,6 @@ error:
         mVariantFilter |= uint32_t(UserVariantFilterBit::SHADOW_RECEIVER);
         mVariantFilter |= uint32_t(UserVariantFilterBit::VSM);
         mVariantFilter |= uint32_t(UserVariantFilterBit::SSR);
-        mVariantFilter |= uint32_t(UserVariantFilterBit::STE);
     }
 
     // Create chunk tree.

libs/filamat/src/MetalArgumentBuffer.cpp

@@ -29,11 +29,20 @@ MetalArgumentBuffer::Builder& filamat::MetalArgumentBuffer::Builder::name(
 
 MetalArgumentBuffer::Builder& MetalArgumentBuffer::Builder::texture(size_t index,
         const std::string& name, filament::backend::SamplerType type,
-        filament::backend::SamplerFormat format) noexcept {
+        filament::backend::SamplerFormat format,
+        bool multisample) noexcept {
+
+    using namespace filament::backend;
+
     // All combinations of SamplerType and SamplerFormat are valid except for SAMPLER_3D / SHADOW.
-    assert_invariant(type != filament::backend::SamplerType::SAMPLER_3D ||
-            format != filament::backend::SamplerFormat::SHADOW);
-    mArguments.emplace_back(TextureArgument { name, index, type, format });
+    assert_invariant(type != SamplerType::SAMPLER_3D || format != SamplerFormat::SHADOW);
+
+    // multisample textures have restrictions too
+    assert_invariant(!multisample || (
+            format != SamplerFormat::SHADOW && (
+                    type == SamplerType::SAMPLER_2D || type == SamplerType::SAMPLER_2D_ARRAY)));
+
+    mArguments.emplace_back(TextureArgument { name, index, type, format, multisample });
     return *this;
 }
 
@@ -79,6 +88,10 @@ std::ostream& MetalArgumentBuffer::Builder::TextureArgument::write(std::ostream&
             break;
     }
 
+    if (multisample) {
+        os << "_ms";
+    }
+
     switch (format) {
         case filament::backend::SamplerFormat::INT:
             os << "<int>";

libs/filamat/src/MetalArgumentBuffer.h

@@ -48,7 +48,8 @@ public:
          */
         Builder& texture(size_t index, const std::string& name,
                 filament::backend::SamplerType type,
-                filament::backend::SamplerFormat format) noexcept;
+                filament::backend::SamplerFormat format,
+                bool multisample) noexcept;
 
         /**
          * Add a sampler argument to the argument buffer structure.
@@ -69,6 +70,7 @@ public:
             size_t index;
             filament::backend::SamplerType type;
             filament::backend::SamplerFormat format;
+            bool multisample;
 
             std::ostream& write(std::ostream& os) const;
         };

libs/filamat/src/sca/GLSLTools.cpp

@@ -176,7 +176,7 @@ bool GLSLTools::analyzeComputeShader(const std::string& shaderCode,
     bool const ok = tShader.parse(&DefaultTBuiltInResource, version, false, msg);
     if (!ok) {
         utils::slog.e << "ERROR: Unable to parse compute shader:" << utils::io::endl;
-        utils::slog.e << tShader.getInfoLog() << utils::io::flush;
+        utils::slog.e << tShader.getInfoLog() << utils::io::endl;
         return false;
     }
 
@@ -214,7 +214,7 @@ std::optional<GLSLTools::FragmentShaderInfo> GLSLTools::analyzeFragmentShader(
     bool const ok = tShader.parse(&DefaultTBuiltInResource, version, false, msg);
     if (!ok) {
         utils::slog.e << "ERROR: Unable to parse fragment shader:" << utils::io::endl;
-        utils::slog.e << tShader.getInfoLog() << utils::io::flush;
+        utils::slog.e << tShader.getInfoLog() << utils::io::endl;
         return std::nullopt;
     }
 
@@ -294,7 +294,7 @@ bool GLSLTools::analyzeVertexShader(const std::string& shaderCode,
     bool const ok = tShader.parse(&DefaultTBuiltInResource, version, false, msg);
     if (!ok) {
         utils::slog.e << "ERROR: Unable to parse vertex shader" << utils::io::endl;
-        utils::slog.e << tShader.getInfoLog() << utils::io::flush;
+        utils::slog.e << tShader.getInfoLog() << utils::io::endl;
         return false;
     }
 

libs/filamat/src/shaders/CodeGenerator.cpp

@@ -1150,13 +1150,12 @@ char const* CodeGenerator::getOutputTypeName(MaterialBuilder::OutputType type) n
 
 char const* CodeGenerator::getSamplerTypeName(SamplerType type, SamplerFormat format,
         bool multisample) const noexcept {
-    assert(!multisample);   // multisample samplers not yet supported.
     switch (type) {
         case SamplerType::SAMPLER_2D:
             switch (format) {
-                case SamplerFormat::INT:    return "isampler2D";
-                case SamplerFormat::UINT:   return "usampler2D";
-                case SamplerFormat::FLOAT:  return "sampler2D";
+                case SamplerFormat::INT:    return multisample ? "isampler2DMS" : "isampler2D";
+                case SamplerFormat::UINT:   return multisample ? "usampler2DMS" : "usampler2D";
+                case SamplerFormat::FLOAT:  return multisample ? "sampler2DMS" : "sampler2D";
                 case SamplerFormat::SHADOW: return "sampler2DShadow";
             }
         case SamplerType::SAMPLER_3D:
@@ -1169,9 +1168,9 @@ char const* CodeGenerator::getSamplerTypeName(SamplerType type, SamplerFormat fo
             }
         case SamplerType::SAMPLER_2D_ARRAY:
             switch (format) {
-                case SamplerFormat::INT:    return "isampler2DArray";
-                case SamplerFormat::UINT:   return "usampler2DArray";
-                case SamplerFormat::FLOAT:  return "sampler2DArray";
+                case SamplerFormat::INT:    return multisample ? "isampler2DMSArray": "isampler2DArray";
+                case SamplerFormat::UINT:   return multisample ? "usampler2DMSArray": "usampler2DArray";
+                case SamplerFormat::FLOAT:  return multisample ? "sampler2DMSArray": "sampler2DArray";
                 case SamplerFormat::SHADOW: return "sampler2DArrayShadow";
             }
         case SamplerType::SAMPLER_CUBEMAP:

libs/filamat/src/shaders/ShaderGenerator.cpp

@@ -52,7 +52,7 @@ void ShaderGenerator::generateSurfaceMaterialVariantDefines(utils::io::sstream&
     CodeGenerator::generateDefine(out, "VARIANT_HAS_VSM",
             filament::Variant::isVSMVariant(variant));
     CodeGenerator::generateDefine(out, "VARIANT_HAS_INSTANCED_STEREO",
-            filament::Variant::isStereoVariant(variant));
+            hasInstancedStereo(variant, featureLevel));
 
     switch (stage) {
         case ShaderStage::VERTEX:
@@ -758,4 +758,12 @@ bool ShaderGenerator::hasSkinningOrMorphing(
             && featureLevel > MaterialBuilder::FeatureLevel::FEATURE_LEVEL_0;
 }
 
+bool ShaderGenerator::hasInstancedStereo(
+        filament::Variant variant, MaterialBuilder::FeatureLevel featureLevel) noexcept {
+    return variant.hasInstancedStereo()
+            // HACK(exv): Ignore stereo variant when targeting ESSL 1.0. We should properly build a
+            // system in matc which allows the set of included variants to differ per-feature level.
+            && featureLevel > MaterialBuilder::FeatureLevel::FEATURE_LEVEL_0;
+}
+
 } // namespace filament

libs/filamat/src/shaders/ShaderGenerator.h

@@ -115,6 +115,10 @@ private:
             filament::Variant variant,
             MaterialBuilder::FeatureLevel featureLevel) noexcept;
 
+    static bool hasInstancedStereo(
+            filament::Variant variant,
+            MaterialBuilder::FeatureLevel featureLevel) noexcept;
+
     MaterialBuilder::PropertyList mProperties;
     MaterialBuilder::VariableList mVariables;
     MaterialBuilder::OutputList mOutputs;

libs/filamat/src/shaders/UibGenerator.cpp

@@ -60,6 +60,7 @@ BufferInterfaceBlock const& UibGenerator::getPerViewUib() noexcept  {
 
             { "lodBias",                0, Type::FLOAT, Precision::DEFAULT, FeatureLevel::FEATURE_LEVEL_0 },
             { "refractionLodOffset",    0, Type::FLOAT, Precision::DEFAULT, FeatureLevel::FEATURE_LEVEL_0 },
+            { "derivativesScale",       0, Type::FLOAT2                  },
 
             { "oneOverFarMinusNear",    0, Type::FLOAT,  Precision::HIGH, FeatureLevel::FEATURE_LEVEL_0 },
             { "nearOverFarMinusNear",   0, Type::FLOAT,  Precision::HIGH, FeatureLevel::FEATURE_LEVEL_0 },
@@ -71,8 +72,6 @@ BufferInterfaceBlock const& UibGenerator::getPerViewUib() noexcept  {
             // AO
             { "aoSamplingQualityAndEdgeDistance", 0, Type::FLOAT         },
             { "aoBentNormals",          0, Type::FLOAT                   },
-            { "aoReserved0",            0, Type::FLOAT                   },
-            { "aoReserved1",            0, Type::FLOAT                   },
 
             // ------------------------------------------------------------------------------------
             // Dynamic Lighting [variant: DYN]

libs/filamat/tests/test_argBufferFixup.cpp

@@ -72,7 +72,7 @@ TEST(ArgBufferFixup, TextureAndSampler) {
     auto argBuffer =
             MetalArgumentBuffer::Builder()
                     .name("myArgumentBuffer")
-                    .texture(0, "textureA", SamplerType::SAMPLER_2D, SamplerFormat::FLOAT)
+                    .texture(0, "textureA", SamplerType::SAMPLER_2D, SamplerFormat::FLOAT, false)
                     .sampler(1, "samplerA")
                     .build();
     auto argBufferStr = argBuffer->getMsl();
@@ -88,13 +88,33 @@ TEST(ArgBufferFixup, TextureAndSampler) {
     MetalArgumentBuffer::destroy(&argBuffer);
 }
 
+TEST(ArgBufferFixup, TextureAndSamplerMS) {
+    auto argBuffer =
+            MetalArgumentBuffer::Builder()
+                    .name("myArgumentBuffer")
+                    .texture(0, "textureA", SamplerType::SAMPLER_2D, SamplerFormat::FLOAT, true)
+                    .sampler(1, "samplerA")
+                    .build();
+    auto argBufferStr = argBuffer->getMsl();
+
+    const std::string expected =
+            "struct myArgumentBuffer {\n"
+            "texture2d_ms<float> textureA [[id(0)]];\n"
+            "sampler samplerA [[id(1)]];\n"
+            "}";
+
+    EXPECT_EQ(argBuffer->getMsl(), expected);
+
+    MetalArgumentBuffer::destroy(&argBuffer);
+}
+
 TEST(ArgBufferFixup, Sorted) {
     auto argBuffer =
             MetalArgumentBuffer::Builder()
                     .name("myArgumentBuffer")
                     .sampler(3, "samplerB")
-                    .texture(0, "textureA", SamplerType::SAMPLER_2D, SamplerFormat::FLOAT)
-                    .texture(2, "textureB", SamplerType::SAMPLER_2D, SamplerFormat::FLOAT)
+                    .texture(0, "textureA", SamplerType::SAMPLER_2D, SamplerFormat::FLOAT, false)
+                    .texture(2, "textureB", SamplerType::SAMPLER_2D, SamplerFormat::FLOAT, false)
                     .sampler(1, "samplerA")
                     .build();
     auto argBufferStr = argBuffer->getMsl();
@@ -116,12 +136,12 @@ TEST(ArgBufferFixup, TextureTypes) {
     auto argBuffer =
             MetalArgumentBuffer::Builder()
                     .name("myArgumentBuffer")
-                    .texture(0, "textureA", SamplerType::SAMPLER_2D, SamplerFormat::INT)
-                    .texture(1, "textureB", SamplerType::SAMPLER_2D_ARRAY, SamplerFormat::UINT)
-                    .texture(2, "textureC", SamplerType::SAMPLER_CUBEMAP, SamplerFormat::FLOAT)
-                    .texture(3, "textureD", SamplerType::SAMPLER_EXTERNAL, SamplerFormat::FLOAT)
-                    .texture(4, "textureE", SamplerType::SAMPLER_3D, SamplerFormat::FLOAT)
-                    .texture(5, "textureF", SamplerType::SAMPLER_CUBEMAP_ARRAY, SamplerFormat::SHADOW)
+                    .texture(0, "textureA", SamplerType::SAMPLER_2D, SamplerFormat::INT, false)
+                    .texture(1, "textureB", SamplerType::SAMPLER_2D_ARRAY, SamplerFormat::UINT, false)
+                    .texture(2, "textureC", SamplerType::SAMPLER_CUBEMAP, SamplerFormat::FLOAT, false)
+                    .texture(3, "textureD", SamplerType::SAMPLER_EXTERNAL, SamplerFormat::FLOAT, false)
+                    .texture(4, "textureE", SamplerType::SAMPLER_3D, SamplerFormat::FLOAT, false)
+                    .texture(5, "textureF", SamplerType::SAMPLER_CUBEMAP_ARRAY, SamplerFormat::SHADOW, false)
                     .build();
     auto argBufferStr = argBuffer->getMsl();
 
@@ -147,7 +167,7 @@ TEST(ArgBufferFixup, InvalidType) {
         auto argBuffer =
                 MetalArgumentBuffer::Builder()
                         .name("myArgumentBuffer")
-                        .texture(0, "textureA", SamplerType::SAMPLER_3D, SamplerFormat::SHADOW)
+                        .texture(0, "textureA", SamplerType::SAMPLER_3D, SamplerFormat::SHADOW, false)
                         .build();
         MetalArgumentBuffer::destroy(&argBuffer);
     }, "failed assertion");

libs/matdbg/src/ApiHandler.cpp

@@ -215,7 +215,7 @@ bool ApiHandler::handleGetApiShader(struct mg_connection* conn,
 }
 
 void ApiHandler::addMaterial(MaterialRecord const* material) {
-    std::unique_lock<utils::Mutex> lock(mStatusMutex);
+    std::unique_lock const lock(mStatusMutex);
     snprintf(statusMaterialId, sizeof(statusMaterialId), "%8.8x", material->key);
     mStatusCondition.notify_all();
 }
@@ -229,7 +229,7 @@ bool ApiHandler::handleGetStatus(struct mg_connection* conn,
         return true;
     }
 
-    std::unique_lock<utils::Mutex> lock(mStatusMutex);
+    std::unique_lock<std::mutex> lock(mStatusMutex);
     uint64_t const currentStatusCount = mCurrentStatus;
     if (mStatusCondition.wait_for(lock, 10s,
                 [this, currentStatusCount] { return currentStatusCount < mCurrentStatus; })) {
@@ -258,13 +258,13 @@ bool ApiHandler::handlePost(CivetServer* server, struct mg_connection* conn) {
     //   [material id] [api index] [shader index] [shader source....]
     if (uri == "/api/edit") {
         struct mg_request_info const* req_info = mg_get_request_info(conn);
-        size_t msgLen = req_info->content_length;
+        size_t const msgLen = req_info->content_length;
 
         char buf[1024];
         size_t readLen = 0;
         std::stringstream sstream;
         while (readLen < msgLen) {
-            int res = mg_read(conn, buf, sizeof(buf));
+            int const res = mg_read(conn, buf, sizeof(buf));
             if (res < 0) {
                 utils::slog.e << "civet error parsing /api/edit body: " << res << utils::io::endl;
                 break;
@@ -279,7 +279,7 @@ bool ApiHandler::handlePost(CivetServer* server, struct mg_connection* conn) {
         int api;
         int shaderIndex;
         sstream >> std::hex >> matid >> std::dec >> api >> shaderIndex;
-        std::string shader = sstream.str().substr(sstream.tellg());
+        std::string const shader = sstream.str().substr(sstream.tellg());
 
         mServer->handleEditCommand(matid, backend::Backend(api), shaderIndex, shader.c_str(),
                 shader.size());
@@ -298,7 +298,7 @@ bool ApiHandler::handleGet(CivetServer* server, struct mg_connection* conn) {
         mServer->updateActiveVariants();
 
         // Careful not to lock the above line.
-        std::unique_lock<utils::Mutex> lock(mServer->mMaterialRecordsMutex);
+        std::unique_lock const lock(mServer->mMaterialRecordsMutex);
         mg_printf(conn, kSuccessHeader.data(), "application/json");
         mg_printf(conn, "{");
 
@@ -329,7 +329,7 @@ bool ApiHandler::handleGet(CivetServer* server, struct mg_connection* conn) {
     }
 
     if (uri == "/api/matids") {
-        std::unique_lock<utils::Mutex> lock(mServer->mMaterialRecordsMutex);
+        std::unique_lock const lock(mServer->mMaterialRecordsMutex);
         mg_printf(conn, kSuccessHeader.data(), "application/json");
         mg_printf(conn, "[");
         int index = 0;
@@ -342,7 +342,7 @@ bool ApiHandler::handleGet(CivetServer* server, struct mg_connection* conn) {
     }
 
     if (uri == "/api/materials") {
-        std::unique_lock<utils::Mutex> lock(mServer->mMaterialRecordsMutex);
+        std::unique_lock const lock(mServer->mMaterialRecordsMutex);
         mg_printf(conn, kSuccessHeader.data(), "application/json");
         mg_printf(conn, "[");
         int index = 0;

libs/matdbg/src/ApiHandler.h

@@ -17,7 +17,8 @@
 #ifndef MATDBG_APIHANDLER_H
 #define MATDBG_APIHANDLER_H
 
-#include <utils/Mutex.h>
+#include <mutex>
+#include <condition_variable>
 
 #include <CivetServer.h>
 
@@ -53,7 +54,7 @@ private:
 
     DebugServer* mServer;
 
-    utils::Mutex mStatusMutex;
+    std::mutex mStatusMutex;
     std::condition_variable mStatusCondition;
     char statusMaterialId[9] = {};
 

libs/matdbg/web/index.html

@@ -12,6 +12,11 @@
           margin: 0;
           font-family: "Open Sans";
       }
+
+      @font-face {
+        font-family: codicon;
+        src: url(https://cdnjs.cloudflare.com/ajax/libs/monaco-editor/0.25.2/min/vs/base/browser/ui/codicons/codicon/codicon.ttf);
+      }
     </style>
     <script src="api.js"></script>
     <script src="https://cdnjs.cloudflare.com/ajax/libs/monaco-editor/0.25.2/min/vs/loader.js"></script>

libs/utils/include/utils/CString.h

@@ -35,7 +35,7 @@ struct hashCStrings {
     typedef size_t result_type;
     result_type operator()(argument_type cstr) const noexcept {
         size_t hash = 5381;
-        while (int c = *cstr++) {
+        while (int const c = *cstr++) {
             hash = (hash * 33u) ^ size_t(c);
         }
         return hash;

libs/utils/include/utils/ostream.h

@@ -29,7 +29,7 @@ namespace utils::io {
 
 struct ostream_;
 
-class UTILS_PUBLIC  ostream : protected utils::PrivateImplementation<ostream_> {
+class UTILS_PUBLIC ostream : protected utils::PrivateImplementation<ostream_> {
     friend struct ostream_;
 
 public:
@@ -69,6 +69,13 @@ public:
     ostream& dec() noexcept;
     ostream& hex() noexcept;
 
+    /*! @cond PRIVATE */
+    // Sets a consumer of the log. The consumer is invoked on flush() and replaces the default.
+    // Thread safe and reentrant.
+    using ConsumerCallback = void(*)(void*, char const*);
+    void setConsumer(ConsumerCallback consumer, void* user) noexcept;
+    /*! @endcond */
+
 protected:
     ostream& print(const char* format, ...) noexcept;
 
@@ -104,7 +111,7 @@ private:
     friend ostream& hex(ostream& s) noexcept;
     friend ostream& dec(ostream& s) noexcept;
     friend ostream& endl(ostream& s) noexcept;
-    friend ostream& flush(ostream& s) noexcept;
+    UTILS_PUBLIC friend ostream& flush(ostream& s) noexcept;
 
     enum type {
         SHORT, USHORT, CHAR, UCHAR, INT, UINT, LONG, ULONG, LONG_LONG, ULONG_LONG, FLOAT, DOUBLE,
@@ -132,8 +139,7 @@ inline ostream& operator<<(ostream& stream, const VECTOR<T>& v) {
 
 inline ostream& hex(ostream& s) noexcept { return s.hex(); }
 inline ostream& dec(ostream& s) noexcept { return s.dec(); }
-inline ostream& endl(ostream& s) noexcept { s << '\n'; return s.flush(); }
-inline ostream& flush(ostream& s) noexcept { return s.flush(); }
+inline ostream& endl(ostream& s) noexcept { return flush(s << '\n'); }
 
 } // namespace utils::io
 

libs/utils/src/Log.cpp

@@ -31,6 +31,8 @@
 #include <emscripten/console.h>
 #endif
 
+#include <mutex>
+
 namespace utils {
 namespace io {
 
@@ -46,43 +48,41 @@ public:
     ostream& flush() noexcept override;
 
 private:
-    Priority mPriority;
+    const Priority mPriority;
 };
 
 ostream& LogStream::flush() noexcept {
-    std::lock_guard lock(mImpl->mLock);
+    std::lock_guard const lock(mImpl->mLock);
     Buffer& buf = getBuffer();
-#ifdef __ANDROID__
-    switch (mPriority) {
-        case LOG_DEBUG:
-            __android_log_write(ANDROID_LOG_DEBUG, UTILS_LOG_TAG, buf.get());
-            break;
-        case LOG_ERROR:
-            __android_log_write(ANDROID_LOG_ERROR, UTILS_LOG_TAG, buf.get());
-            break;
-        case LOG_WARNING:
-            __android_log_write(ANDROID_LOG_WARN, UTILS_LOG_TAG, buf.get());
-            break;
-        case LOG_INFO:
-            __android_log_write(ANDROID_LOG_INFO, UTILS_LOG_TAG, buf.get());
-            break;
-        case LOG_VERBOSE:
-            __android_log_write(ANDROID_LOG_VERBOSE, UTILS_LOG_TAG, buf.get());
-            break;
+    char const* const str = buf.get();
+    if (UTILS_UNLIKELY(!str)) {
+        // this can happen if the log hasn't been written ever
+        return *this;
     }
+
+#ifdef __ANDROID__
+    int prio = ANDROID_LOG_UNKNOWN;
+    switch (mPriority) {
+        case LOG_DEBUG:     prio = ANDROID_LOG_DEBUG;   break;
+        case LOG_ERROR:     prio = ANDROID_LOG_ERROR;   break;
+        case LOG_WARNING:   prio = ANDROID_LOG_WARN;    break;
+        case LOG_INFO:      prio = ANDROID_LOG_INFO;    break;
+        case LOG_VERBOSE:   prio = ANDROID_LOG_VERBOSE; break;
+    }
+    __android_log_write(prio, UTILS_LOG_TAG, str);
 #elif defined(__EMSCRIPTEN__)
     switch (mPriority) {
         case LOG_DEBUG:
         case LOG_WARNING:
         case LOG_INFO:
-            _emscripten_out(buf.get());
+            _emscripten_out(str);
             break;
         case LOG_ERROR:
-            _emscripten_err(buf.get());
+            _emscripten_err(str);
             break;
         case LOG_VERBOSE:
 #ifndef NFIL_DEBUG
-            _emscripten_out(buf.get());
+            _emscripten_out(str);
 #endif
             break;
     }
@@ -91,14 +91,14 @@ ostream& LogStream::flush() noexcept {
         case LOG_DEBUG:
         case LOG_WARNING:
         case LOG_INFO:
-            fprintf(stdout, "%s", buf.get());
+            fprintf(stdout, "%s", str);
             break;
         case LOG_ERROR:
-            fprintf(stderr, "%s", buf.get());
+            fprintf(stderr, "%s", str);
             break;
         case LOG_VERBOSE:
 #ifndef NDEBUG
-            fprintf(stdout, "%s", buf.get());
+            fprintf(stdout, "%s", str);
 #endif
             break;
     }

libs/utils/src/ostream.cpp

@@ -24,10 +24,14 @@
 #include <utils/PrivateImplementation-impl.h>
 
 #include <algorithm>
+#include <mutex>
 #include <string>
 #include <string_view>
+#include <utility>
 
+#include <stdlib.h>
 #include <stdarg.h>
+#include <string.h>
 
 template class utils::PrivateImplementation<utils::io::ostream_>;
 
@@ -35,6 +39,34 @@ namespace utils::io {
 
 ostream::~ostream() = default;
 
+void ostream::setConsumer(ConsumerCallback consumer, void* user) noexcept {
+    auto* const pImpl = mImpl;
+    std::lock_guard const lock(pImpl->mLock);
+    pImpl->mConsumer = { consumer, user };
+}
+
+ostream& flush(ostream& s) noexcept {
+    auto* const pImpl = s.mImpl;
+    pImpl->mLock.lock();
+    auto const callback = pImpl->mConsumer;
+    if (UTILS_UNLIKELY(callback.first)) {
+        auto& buf = s.getBuffer();
+        char const* const data = buf.get();
+        if (UTILS_LIKELY(data)) {
+            char* const str = strdup(data);
+            buf.reset();
+            pImpl->mLock.unlock();
+            // call ConsumerCallback without lock held
+            callback.first(callback.second, str);
+            ::free(str);
+            return s;
+        }
+    }
+    pImpl->mLock.unlock();
+
+    return s.flush();
+}
+
 ostream::Buffer& ostream::getBuffer() noexcept {
     return mImpl->mData;
 }
@@ -69,12 +101,12 @@ ostream& ostream::print(const char* format, ...) noexcept {
     // figure out how much size to we need
     va_start(args0, format);
     va_copy(args1, args0);
-    ssize_t s = vsnprintf(nullptr, 0, format, args0);
+    ssize_t const s = vsnprintf(nullptr, 0, format, args0);
     va_end(args0);
 
 
     { // scope for the lock
-        std::lock_guard lock(mImpl->mLock);
+        std::lock_guard const lock(mImpl->mLock);
 
         Buffer& buf = getBuffer();
 
@@ -205,14 +237,14 @@ ostream::Buffer::~Buffer() noexcept {
 
 void ostream::Buffer::advance(ssize_t n) noexcept {
     if (n > 0) {
-        size_t written = n < size ? size_t(n) : size;
+        size_t const written = n < size ? size_t(n) : size;
         curr += written;
         size -= written;
     }
 }
 
 void ostream::Buffer::reserve(size_t newSize) noexcept {
-    size_t offset = curr - buffer;
+    size_t const offset = curr - buffer;
     if (buffer == nullptr) {
         buffer = (char*)malloc(newSize);
     } else {
@@ -237,8 +269,8 @@ void ostream::Buffer::reset() noexcept {
 
 std::pair<char*, size_t> ostream::Buffer::grow(size_t s) noexcept {
     if (UTILS_UNLIKELY(size < s)) {
-        size_t used = curr - buffer;
-        size_t newCapacity = std::max(size_t(32), used + (s * 3 + 1) / 2); // 32 bytes minimum
+        size_t const used = curr - buffer;
+        size_t const newCapacity = std::max(size_t(32), used + (s * 3 + 1) / 2); // 32 bytes minimum
         reserve(newCapacity);
         assert(size >= s);
     }

libs/utils/src/ostream_.h

@@ -25,6 +25,7 @@ namespace utils::io {
 struct ostream_ {
     std::mutex mLock;
     ostream::Buffer mData;
+    std::pair<ostream::ConsumerCallback, void*> mConsumer{};
     bool mShowHex = false;
 };
 

libs/utils/test/test_sstream.cpp

@@ -18,8 +18,33 @@
 
 #include <utils/sstream.h>
 
+#include <utils/Log.h>
+
+using namespace utils;
 using namespace utils::io;
 
+TEST(ostream, setConsumer) {
+    slog.d.setConsumer(+[](void*, char const*) {
+        GTEST_FAIL();
+    }, nullptr);
+    // we test that we don't crash if the log is empty and that we don't call the consumer.
+    flush(slog.d);
+
+    slog.d.setConsumer(nullptr, nullptr);
+    slog.d << "hello world";
+    // we test that after resetting the consumer, it's not called on flush.
+    flush(slog.d);
+
+    const char* str = "hello world!";
+    slog.d.setConsumer(+[](void* user, char const* str) {
+        ASSERT_STREQ(str, (const char*)user);
+    }, (void*)str);
+    slog.d << str;
+    // we test that the comsumer is called with the right string
+    flush(slog.d);
+    slog.d.setConsumer(nullptr, nullptr);
+}
+
 TEST(sstream, EmptyString) {
     sstream ss;
     EXPECT_STREQ("", ss.c_str());

libs/viewer/src/Settings_generated.cpp

@@ -391,6 +391,8 @@ int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, DepthOfFieldOpt
         CHECK_KEY(tok);
         if (compare(tok, jsonChunk, "cocScale") == 0) {
             i = parse(tokens, i + 1, jsonChunk, &out->cocScale);
+        } else if (compare(tok, jsonChunk, "cocAspectRatio") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->cocAspectRatio);
         } else if (compare(tok, jsonChunk, "maxApertureDiameter") == 0) {
             i = parse(tokens, i + 1, jsonChunk, &out->maxApertureDiameter);
         } else if (compare(tok, jsonChunk, "enabled") == 0) {
@@ -424,6 +426,7 @@ int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, DepthOfFieldOpt
 std::ostream& operator<<(std::ostream& out, const DepthOfFieldOptions& in) {
     return out << "{\n"
         << "\"cocScale\": " << (in.cocScale) << ",\n"
+        << "\"cocAspectRatio\": " << (in.cocAspectRatio) << ",\n"
         << "\"maxApertureDiameter\": " << (in.maxApertureDiameter) << ",\n"
         << "\"enabled\": " << to_string(in.enabled) << ",\n"
         << "\"filter\": " << (in.filter) << ",\n"
@@ -647,6 +650,65 @@ std::ostream& operator<<(std::ostream& out, const MultiSampleAntiAliasingOptions
         << "}";
 }
 
+int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, TemporalAntiAliasingOptions::BoxType* out) {
+    if (0 == compare(tokens[i], jsonChunk, "AABB")) { *out = TemporalAntiAliasingOptions::BoxType::AABB; }
+    else if (0 == compare(tokens[i], jsonChunk, "VARIANCE")) { *out = TemporalAntiAliasingOptions::BoxType::VARIANCE; }
+    else if (0 == compare(tokens[i], jsonChunk, "AABB_VARIANCE")) { *out = TemporalAntiAliasingOptions::BoxType::AABB_VARIANCE; }
+    else {
+        slog.w << "Invalid TemporalAntiAliasingOptions::BoxType: '" << STR(tokens[i], jsonChunk) << "'" << io::endl;
+    }
+    return i + 1;
+}
+
+std::ostream& operator<<(std::ostream& out, TemporalAntiAliasingOptions::BoxType in) {
+    switch (in) {
+        case TemporalAntiAliasingOptions::BoxType::AABB: return out << "\"AABB\"";
+        case TemporalAntiAliasingOptions::BoxType::VARIANCE: return out << "\"VARIANCE\"";
+        case TemporalAntiAliasingOptions::BoxType::AABB_VARIANCE: return out << "\"AABB_VARIANCE\"";
+    }
+    return out << "\"INVALID\"";
+}
+
+int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, TemporalAntiAliasingOptions::BoxClipping* out) {
+    if (0 == compare(tokens[i], jsonChunk, "ACCURATE")) { *out = TemporalAntiAliasingOptions::BoxClipping::ACCURATE; }
+    else if (0 == compare(tokens[i], jsonChunk, "CLAMP")) { *out = TemporalAntiAliasingOptions::BoxClipping::CLAMP; }
+    else if (0 == compare(tokens[i], jsonChunk, "NONE")) { *out = TemporalAntiAliasingOptions::BoxClipping::NONE; }
+    else {
+        slog.w << "Invalid TemporalAntiAliasingOptions::BoxClipping: '" << STR(tokens[i], jsonChunk) << "'" << io::endl;
+    }
+    return i + 1;
+}
+
+std::ostream& operator<<(std::ostream& out, TemporalAntiAliasingOptions::BoxClipping in) {
+    switch (in) {
+        case TemporalAntiAliasingOptions::BoxClipping::ACCURATE: return out << "\"ACCURATE\"";
+        case TemporalAntiAliasingOptions::BoxClipping::CLAMP: return out << "\"CLAMP\"";
+        case TemporalAntiAliasingOptions::BoxClipping::NONE: return out << "\"NONE\"";
+    }
+    return out << "\"INVALID\"";
+}
+
+int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, TemporalAntiAliasingOptions::JitterPattern* out) {
+    if (0 == compare(tokens[i], jsonChunk, "RGSS_X4")) { *out = TemporalAntiAliasingOptions::JitterPattern::RGSS_X4; }
+    else if (0 == compare(tokens[i], jsonChunk, "UNIFORM_HELIX_X4")) { *out = TemporalAntiAliasingOptions::JitterPattern::UNIFORM_HELIX_X4; }
+    else if (0 == compare(tokens[i], jsonChunk, "HALTON_23_X8")) { *out = TemporalAntiAliasingOptions::JitterPattern::HALTON_23_X8; }
+    else if (0 == compare(tokens[i], jsonChunk, "HALTON_23_X16")) { *out = TemporalAntiAliasingOptions::JitterPattern::HALTON_23_X16; }
+    else {
+        slog.w << "Invalid TemporalAntiAliasingOptions::JitterPattern: '" << STR(tokens[i], jsonChunk) << "'" << io::endl;
+    }
+    return i + 1;
+}
+
+std::ostream& operator<<(std::ostream& out, TemporalAntiAliasingOptions::JitterPattern in) {
+    switch (in) {
+        case TemporalAntiAliasingOptions::JitterPattern::RGSS_X4: return out << "\"RGSS_X4\"";
+        case TemporalAntiAliasingOptions::JitterPattern::UNIFORM_HELIX_X4: return out << "\"UNIFORM_HELIX_X4\"";
+        case TemporalAntiAliasingOptions::JitterPattern::HALTON_23_X8: return out << "\"HALTON_23_X8\"";
+        case TemporalAntiAliasingOptions::JitterPattern::HALTON_23_X16: return out << "\"HALTON_23_X16\"";
+    }
+    return out << "\"INVALID\"";
+}
+
 int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, TemporalAntiAliasingOptions* out) {
     CHECK_TOKTYPE(tokens[i], JSMN_OBJECT);
     int size = tokens[i++].size;
@@ -657,8 +719,30 @@ int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, TemporalAntiAli
             i = parse(tokens, i + 1, jsonChunk, &out->filterWidth);
         } else if (compare(tok, jsonChunk, "feedback") == 0) {
             i = parse(tokens, i + 1, jsonChunk, &out->feedback);
+        } else if (compare(tok, jsonChunk, "lodBias") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->lodBias);
         } else if (compare(tok, jsonChunk, "enabled") == 0) {
             i = parse(tokens, i + 1, jsonChunk, &out->enabled);
+        } else if (compare(tok, jsonChunk, "upscaling") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->upscaling);
+        } else if (compare(tok, jsonChunk, "filterHistory") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->filterHistory);
+        } else if (compare(tok, jsonChunk, "filterInput") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->filterInput);
+        } else if (compare(tok, jsonChunk, "useYCoCg") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->useYCoCg);
+        } else if (compare(tok, jsonChunk, "boxType") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->boxType);
+        } else if (compare(tok, jsonChunk, "boxClipping") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->boxClipping);
+        } else if (compare(tok, jsonChunk, "jitterPattern") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->jitterPattern);
+        } else if (compare(tok, jsonChunk, "varianceGamma") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->varianceGamma);
+        } else if (compare(tok, jsonChunk, "preventFlickering") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->preventFlickering);
+        } else if (compare(tok, jsonChunk, "historyReprojection") == 0) {
+            i = parse(tokens, i + 1, jsonChunk, &out->historyReprojection);
         } else {
             slog.w << "Invalid TemporalAntiAliasingOptions key: '" << STR(tok, jsonChunk) << "'" << io::endl;
             i = parse(tokens, i + 1);
@@ -675,7 +759,18 @@ std::ostream& operator<<(std::ostream& out, const TemporalAntiAliasingOptions& i
     return out << "{\n"
         << "\"filterWidth\": " << (in.filterWidth) << ",\n"
         << "\"feedback\": " << (in.feedback) << ",\n"
-        << "\"enabled\": " << to_string(in.enabled) << "\n"
+        << "\"lodBias\": " << (in.lodBias) << ",\n"
+        << "\"enabled\": " << to_string(in.enabled) << ",\n"
+        << "\"upscaling\": " << to_string(in.upscaling) << ",\n"
+        << "\"filterHistory\": " << to_string(in.filterHistory) << ",\n"
+        << "\"filterInput\": " << to_string(in.filterInput) << ",\n"
+        << "\"useYCoCg\": " << to_string(in.useYCoCg) << ",\n"
+        << "\"boxType\": " << (in.boxType) << ",\n"
+        << "\"boxClipping\": " << (in.boxClipping) << ",\n"
+        << "\"jitterPattern\": " << (in.jitterPattern) << ",\n"
+        << "\"varianceGamma\": " << (in.varianceGamma) << ",\n"
+        << "\"preventFlickering\": " << to_string(in.preventFlickering) << ",\n"
+        << "\"historyReprojection\": " << to_string(in.historyReprojection) << "\n"
         << "}";
 }
 

libs/viewer/src/Settings_generated.h

@@ -69,6 +69,15 @@ std::ostream& operator<<(std::ostream& out, const AmbientOcclusionOptions& in);
 int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, MultiSampleAntiAliasingOptions* out);
 std::ostream& operator<<(std::ostream& out, const MultiSampleAntiAliasingOptions& in);
 
+int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, TemporalAntiAliasingOptions::BoxType* out);
+std::ostream& operator<<(std::ostream& out, TemporalAntiAliasingOptions::BoxType in);
+
+int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, TemporalAntiAliasingOptions::BoxClipping* out);
+std::ostream& operator<<(std::ostream& out, TemporalAntiAliasingOptions::BoxClipping in);
+
+int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, TemporalAntiAliasingOptions::JitterPattern* out);
+std::ostream& operator<<(std::ostream& out, TemporalAntiAliasingOptions::JitterPattern in);
+
 int parse(jsmntok_t const* tokens, int i, const char* jsonChunk, TemporalAntiAliasingOptions* out);
 std::ostream& operator<<(std::ostream& out, const TemporalAntiAliasingOptions& in);
 

libs/viewer/src/ViewerGui.cpp

@@ -792,6 +792,28 @@ void ViewerGui::updateUserInterface() {
         ImGui::Checkbox("Lens Flare", &mSettings.view.bloom.lensFlare);
     }
 
+    if (ImGui::CollapsingHeader("TAA Options")) {
+        ImGui::Checkbox("Upscaling", &mSettings.view.taa.upscaling);
+        ImGui::Checkbox("History Reprojection", &mSettings.view.taa.historyReprojection);
+        ImGui::SliderFloat("Feedback", &mSettings.view.taa.feedback, 0.0f, 1.0f);
+        ImGui::Checkbox("Filter History", &mSettings.view.taa.filterHistory);
+        ImGui::Checkbox("Filter Input", &mSettings.view.taa.filterInput);
+        ImGui::SliderFloat("FilterWidth", &mSettings.view.taa.filterWidth, 0.2f, 2.0f);
+        ImGui::SliderFloat("LOD bias", &mSettings.view.taa.lodBias, -8.0f, 0.0f);
+        ImGui::Checkbox("Use YCoCg", &mSettings.view.taa.useYCoCg);
+        ImGui::Checkbox("Prevent Flickering", &mSettings.view.taa.preventFlickering);
+        int jitterSequence = (int)mSettings.view.taa.jitterPattern;
+        int boxClipping = (int)mSettings.view.taa.boxClipping;
+        int boxType = (int)mSettings.view.taa.boxType;
+        ImGui::Combo("Jitter Pattern", &jitterSequence, "RGSS x4\0Uniform Helix x4\0Halton x8\0Halton x16\0\0");
+        ImGui::Combo("Box Clipping", &boxClipping, "Accurate\0Clamp\0None\0\0");
+        ImGui::Combo("Box Type", &boxType, "AABB\0Variance\0Both\0\0");
+        ImGui::SliderFloat("Variance Gamma", &mSettings.view.taa.varianceGamma, 0.75f, 1.25f);
+        mSettings.view.taa.boxClipping = (TemporalAntiAliasingOptions::BoxClipping)boxClipping;
+        mSettings.view.taa.boxType = (TemporalAntiAliasingOptions::BoxType)boxType;
+        mSettings.view.taa.jitterPattern = (TemporalAntiAliasingOptions::JitterPattern)jitterSequence;
+    }
+
     if (ImGui::CollapsingHeader("SSAO Options")) {
         auto& ssao = mSettings.view.ssao;
 
@@ -1014,6 +1036,7 @@ void ViewerGui::updateUserInterface() {
             ImGui::Checkbox("Enabled##dofEnabled", &mSettings.view.dof.enabled);
             ImGui::SliderFloat("Focus distance", &mSettings.viewer.cameraFocusDistance, 0.0f, 30.0f);
             ImGui::SliderFloat("Blur scale", &mSettings.view.dof.cocScale, 0.1f, 10.0f);
+            ImGui::SliderFloat("CoC aspect-ratio", &mSettings.view.dof.cocAspectRatio, 0.25f, 4.0f);
             ImGui::SliderInt("Ring count", &dofRingCount, 1, 17);
             ImGui::SliderInt("Max CoC", &dofMaxCoC, 1, 32);
             ImGui::Checkbox("Native Resolution", &mSettings.view.dof.nativeResolution);

shaders/src/post_process.vs

@@ -5,22 +5,27 @@ void main() {
 
     inputs.normalizedUV = position.xy * 0.5 + 0.5;
 
-    vec4 position = getPosition();
+    inputs.position = getPosition();
 
-    // GL convention to inverted DX convention
-    position.z = position.z * -0.5 + 0.5;
+    // Invoke user code
+    postProcessVertex(inputs);
+
+    // (vertex domain) GL convention to inverted DX convention
+    inputs.position.z = inputs.position.z * -0.5 + 0.5;
+
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    inputs.position.y = -inputs.position.y;
+#endif
 
     // Adjust clip-space
 #if !defined(TARGET_VULKAN_ENVIRONMENT) && !defined(TARGET_METAL_ENVIRONMENT)
     // This is not needed in Vulkan or Metal because clipControl is always (1, 0)
-    // (We don't use a dot() here because it workaround a spirv-opt optimization that in turn
+    // (We don't use a dot() here because it works around a spirv-opt optimization that in turn
     //  causes a crash on PowerVR, see #5118)
-    position.z = position.z * frameUniforms.clipControl.x + position.w * frameUniforms.clipControl.y;
+    inputs.position.z = inputs.position.z * frameUniforms.clipControl.x + inputs.position.w * frameUniforms.clipControl.y;
 #endif
 
-    // Invoke user code
-    postProcessVertex(inputs);
-
     // Handle user-defined interpolated attributes
 #if defined(VARIABLE_CUSTOM0)
     VARIABLE_CUSTOM_AT0 = inputs.VARIABLE_CUSTOM0;
@@ -36,5 +41,5 @@ void main() {
 #endif
 
     // some PowerVR drivers crash when gl_Position is written more than once
-    gl_Position = position;
+    gl_Position = inputs.position;
 }

shaders/src/post_process_getters.vs

@@ -1,11 +1,4 @@
 /** @public-api */
 vec4 getPosition() {
-    vec4 pos = position;
-
-// In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
-#if defined(TARGET_VULKAN_ENVIRONMENT)
-    pos.y = -pos.y;
-#endif
-
-    return pos;
+    return position;
 }

shaders/src/post_process_inputs.vs

@@ -5,6 +5,9 @@ struct PostProcessVertexInputs {
     // We provide normalized texture coordinates to custom vertex shaders.
     vec2 normalizedUV;
 
+    // vertex position, can be modified by the user code
+    vec4 position;
+
 #ifdef VARIABLE_CUSTOM0
     vec4 VARIABLE_CUSTOM0;
 #endif

shaders/src/shading_lit.fs

@@ -37,8 +37,11 @@ float normalFiltering(float perceptualRoughness, const vec3 worldNormal) {
     vec3 du = dFdx(worldNormal);
     vec3 dv = dFdy(worldNormal);
 
-    float variance = materialParams._specularAntiAliasingVariance * (dot(du, du) + dot(dv, dv));
+    // specular AA factor to correct for resolution scaling (DSR and TAAx4)
+    du *= frameUniforms.derivativesScale.x;
+    dv *= frameUniforms.derivativesScale.y;
 
+    float variance = materialParams._specularAntiAliasingVariance * (dot(du, du) + dot(dv, dv));
     float roughness = perceptualRoughnessToRoughness(perceptualRoughness);
     float kernelRoughness = min(2.0 * variance, materialParams._specularAntiAliasingThreshold);
     float squareRoughness = saturate(roughness * roughness + kernelRoughness);

tools/matc/src/matc/JsonishLexer.cpp

@@ -27,9 +27,7 @@ JsonType JsonishLexer::readIdentifier() noexcept {
         consume();
     }
 
-    const char* lexemeEnd = mCursor - 1;
-
-    size_t lexemeSize = lexemeEnd - lexemeStart;
+    size_t lexemeSize = mCursor - lexemeStart;
 
     // Check what kind of keyword we got here.
     if (strncmp("true", lexemeStart, lexemeSize) == 0) {

tools/matc/src/matc/MaterialCompiler.cpp

@@ -229,7 +229,9 @@ static bool reflectParameters(const MaterialBuilder& builder) {
             std::cout << R"(      "format": ")" <<
                       Enums::toString(parameter.format) << "\"," << std::endl;
             std::cout << R"(      "precision": ")" <<
-                      Enums::toString(parameter.precision) << "\"" << std::endl;
+                      Enums::toString(parameter.precision) << "\"," << std::endl;
+            std::cout << R"(      "multisample": ")" <<
+                      (parameter.multisample ? "true" : "false")<< "\"" << std::endl;
         } else if (parameter.isUniform()) {
             std::cout << R"(      "type": ")" <<
                       Enums::toString(parameter.uniformType) << "\"," << std::endl;

tools/matc/src/matc/ParametersProcessor.cpp

@@ -160,7 +160,6 @@ static bool processParameter(MaterialBuilder& builder, const JsonishObject& json
             std::cerr << "parameters: precision must be a STRING." << std::endl;
             return false;
         }
-
         auto precisionString = precisionValue->toJsonString();
         if (!Enums::isValid<ParameterPrecision>(precisionString->getString())){
             return logEnumIssue("parameters", *precisionString, Enums::map<ParameterPrecision>());
@@ -173,20 +172,27 @@ static bool processParameter(MaterialBuilder& builder, const JsonishObject& json
             std::cerr << "parameters: format must be a STRING." << std::endl;
             return false;
         }
-
         auto formatString = formatValue->toJsonString();
         if (!Enums::isValid<SamplerFormat>(formatString->getString())){
             return logEnumIssue("parameters", *formatString, Enums::map<SamplerFormat>());
         }
     }
 
+    const JsonishValue* multiSampleValue = jsonObject.getValue("multisample");
+    if (multiSampleValue) {
+        if (multiSampleValue->getType() != JsonishValue::BOOL) {
+            std::cerr << "parameters: multisample must be a BOOL." << std::endl;
+            return false;
+        }
+    }
+
     auto typeString = typeValue->toJsonString()->getString();
     auto nameString = nameValue->toJsonString()->getString();
 
-    size_t arraySize = extractArraySize(typeString);
+    size_t const arraySize = extractArraySize(typeString);
 
     if (Enums::isValid<UniformType>(typeString)) {
-        MaterialBuilder::UniformType type = Enums::toEnum<UniformType>(typeString);
+        MaterialBuilder::UniformType const type = Enums::toEnum<UniformType>(typeString);
         ParameterPrecision precision = ParameterPrecision::DEFAULT;
         if (precisionValue) {
             precision =
@@ -205,22 +211,18 @@ static bool processParameter(MaterialBuilder& builder, const JsonishObject& json
             return false;
         }
 
-        MaterialBuilder::SamplerType type = Enums::toEnum<SamplerType>(typeString);
-        if (precisionValue && formatValue) {
-            auto format = Enums::toEnum<SamplerFormat>(formatValue->toJsonString()->getString());
-            auto precision =
-                    Enums::toEnum<ParameterPrecision>(precisionValue->toJsonString()->getString());
-            builder.parameter(nameString.c_str(), type, format, precision);
-        } else if (formatValue) {
-            auto format = Enums::toEnum<SamplerFormat>(formatValue->toJsonString()->getString());
-            builder.parameter(nameString.c_str(), type, format);
-        } else if (precisionValue) {
-            auto precision =
-                    Enums::toEnum<ParameterPrecision>(precisionValue->toJsonString()->getString());
-            builder.parameter(nameString.c_str(), type, precision);
-        } else {
-            builder.parameter(nameString.c_str(), type);
-        }
+        MaterialBuilder::SamplerType const type = Enums::toEnum<SamplerType>(typeString);
+
+        auto format = formatValue ? Enums::toEnum<SamplerFormat>(
+                formatValue->toJsonString()->getString()) : SamplerFormat::FLOAT;
+
+        auto precision = precisionValue ? Enums::toEnum<ParameterPrecision>(
+                precisionValue->toJsonString()->getString()) : ParameterPrecision::DEFAULT;
+
+        auto multisample = multiSampleValue ? multiSampleValue->toJsonBool()->getBool() : false;
+
+        builder.parameter(nameString.c_str(), type, format, precision, multisample);
+
     } else {
         std::cerr << "parameters: the type '" << typeString
                << "' for parameter with name '" << nameString << "' is neither a valid uniform "

web/filament-js/extensions_generated.js

@@ -60,6 +60,7 @@ Filament.loadGeneratedExtensions = function() {
     Filament.View.prototype.setDepthOfFieldOptionsDefaults = function(overrides) {
         const options = {
             cocScale: 1.0,
+            cocAspectRatio: 1.0,
             maxApertureDiameter: 0.01,
             enabled: false,
             filter: Filament.View$DepthOfFieldOptions$Filter.MEDIAN,
@@ -138,8 +139,19 @@ Filament.loadGeneratedExtensions = function() {
     Filament.View.prototype.setTemporalAntiAliasingOptionsDefaults = function(overrides) {
         const options = {
             filterWidth: 1.0,
-            feedback: 0.04,
+            feedback: 0.12,
+            lodBias: -1.0,
             enabled: false,
+            upscaling: false,
+            filterHistory: true,
+            filterInput: true,
+            useYCoCg: false,
+            boxType: Filament.View$TemporalAntiAliasingOptions$BoxType.AABB,
+            boxClipping: Filament.View$TemporalAntiAliasingOptions$BoxClipping.ACCURATE,
+            jitterPattern: Filament.View$TemporalAntiAliasingOptions$JitterPattern.HALTON_23_X16,
+            varianceGamma: 1.0,
+            preventFlickering: false,
+            historyReprojection: true,
         };
         return Object.assign(options, overrides);
     };

web/filament-js/filament.d.ts

@@ -1409,6 +1409,10 @@ export interface View$DepthOfFieldOptions {
      * circle of confusion scale factor (amount of blur)
      */
     cocScale?: number;
+    /**
+     * width/height aspect ratio of the circle of confusion (simulate anamorphic lenses)
+     */
+    cocAspectRatio?: number;
     /**
      * maximum aperture diameter in meters (zero to disable rotation)
      */
@@ -1614,7 +1618,7 @@ export interface View$AmbientOcclusionOptions {
 }
 
 /**
- * Options for Temporal Multi-Sample Anti-aliasing (MSAA)
+ * Options for Multi-Sample Anti-aliasing (MSAA)
  * @see setMultiSampleAntiAliasingOptions()
  */
 export interface View$MultiSampleAntiAliasingOptions {
@@ -1636,23 +1640,87 @@ export interface View$MultiSampleAntiAliasingOptions {
     customResolve?: boolean;
 }
 
+export enum View$TemporalAntiAliasingOptions$BoxType {
+    AABB, // use an AABB neighborhood
+    VARIANCE, // use the variance of the neighborhood (not recommended)
+    AABB_VARIANCE, // use both AABB and variance
+}
+
+export enum View$TemporalAntiAliasingOptions$BoxClipping {
+    ACCURATE, // Accurate box clipping
+    CLAMP, // clamping
+    NONE, // no rejections (use for debugging)
+}
+
+export enum View$TemporalAntiAliasingOptions$JitterPattern {
+    RGSS_X4,
+    UNIFORM_HELIX_X4,
+    HALTON_23_X8,
+    HALTON_23_X16,
+}
+
 /**
  * Options for Temporal Anti-aliasing (TAA)
+ * Most TAA parameters are extremely costly to change, as they will trigger the TAA post-process
+ * shaders to be recompiled. These options should be changed or set during initialization.
+ * `filterWidth`, `feedback` and `jitterPattern`, however, could be changed at any time.
+ *
+ * `feedback` of 0.1 effectively accumulates a maximum of 19 samples in steady state.
+ * see "A Survey of Temporal Antialiasing Techniques" by Lei Yang and all for more information.
+ *
  * @see setTemporalAntiAliasingOptions()
  */
 export interface View$TemporalAntiAliasingOptions {
     /**
-     * reconstruction filter width typically between 0 (sharper, aliased) and 1 (smoother)
+     * reconstruction filter width typically between 0.2 (sharper, aliased) and 1.5 (smoother)
      */
     filterWidth?: number;
     /**
      * history feedback, between 0 (maximum temporal AA) and 1 (no temporal AA).
      */
     feedback?: number;
+    /**
+     * texturing lod bias (typically -1 or -2)
+     */
+    lodBias?: number;
     /**
      * enables or disables temporal anti-aliasing
      */
     enabled?: boolean;
+    /**
+     * 4x TAA upscaling. Disables Dynamic Resolution. [BETA]
+     */
+    upscaling?: boolean;
+    /**
+     * whether to filter the history buffer
+     */
+    filterHistory?: boolean;
+    /**
+     * whether to apply the reconstruction filter to the input
+     */
+    filterInput?: boolean;
+    /**
+     * whether to use the YcoCg color-space for history rejection
+     */
+    useYCoCg?: boolean;
+    /**
+     * type of color gamut box
+     */
+    boxType?: View$TemporalAntiAliasingOptions$BoxType;
+    /**
+     * clipping algorithm
+     */
+    boxClipping?: View$TemporalAntiAliasingOptions$BoxClipping;
+    jitterPattern?: View$TemporalAntiAliasingOptions$JitterPattern;
+    varianceGamma?: number;
+    /**
+     * adjust the feedback dynamically to reduce flickering
+     */
+    preventFlickering?: boolean;
+    /**
+     * whether to apply history reprojection (debug option)
+     */
+    historyReprojection?: boolean;
 }
 
 /**

web/filament-js/jsbindings_generated.cpp

@@ -58,6 +58,7 @@ value_object<View::FogOptions>("View$FogOptions")
 
 value_object<View::DepthOfFieldOptions>("View$DepthOfFieldOptions")
     .field("cocScale", &View::DepthOfFieldOptions::cocScale)
+    .field("cocAspectRatio", &View::DepthOfFieldOptions::cocAspectRatio)
     .field("maxApertureDiameter", &View::DepthOfFieldOptions::maxApertureDiameter)
     .field("enabled", &View::DepthOfFieldOptions::enabled)
     .field("filter", &View::DepthOfFieldOptions::filter)
@@ -119,7 +120,18 @@ value_object<View::MultiSampleAntiAliasingOptions>("View$MultiSampleAntiAliasing
 value_object<View::TemporalAntiAliasingOptions>("View$TemporalAntiAliasingOptions")
     .field("filterWidth", &View::TemporalAntiAliasingOptions::filterWidth)
     .field("feedback", &View::TemporalAntiAliasingOptions::feedback)
+    .field("lodBias", &View::TemporalAntiAliasingOptions::lodBias)
     .field("enabled", &View::TemporalAntiAliasingOptions::enabled)
+    .field("upscaling", &View::TemporalAntiAliasingOptions::upscaling)
+    .field("filterHistory", &View::TemporalAntiAliasingOptions::filterHistory)
+    .field("filterInput", &View::TemporalAntiAliasingOptions::filterInput)
+    .field("useYCoCg", &View::TemporalAntiAliasingOptions::useYCoCg)
+    .field("boxType", &View::TemporalAntiAliasingOptions::boxType)
+    .field("boxClipping", &View::TemporalAntiAliasingOptions::boxClipping)
+    .field("jitterPattern", &View::TemporalAntiAliasingOptions::jitterPattern)
+    .field("varianceGamma", &View::TemporalAntiAliasingOptions::varianceGamma)
+    .field("preventFlickering", &View::TemporalAntiAliasingOptions::preventFlickering)
+    .field("historyReprojection", &View::TemporalAntiAliasingOptions::historyReprojection)
     ;
 
 value_object<View::ScreenSpaceReflectionsOptions>("View$ScreenSpaceReflectionsOptions")

web/filament-js/jsenums_generated.cpp

@@ -33,6 +33,25 @@ enum_<View::DepthOfFieldOptions::Filter>("View$DepthOfFieldOptions$Filter")
     .value("MEDIAN", View::DepthOfFieldOptions::Filter::MEDIAN)
     ;
 
+enum_<View::TemporalAntiAliasingOptions::BoxType>("View$TemporalAntiAliasingOptions$BoxType")
+    .value("AABB", View::TemporalAntiAliasingOptions::BoxType::AABB)
+    .value("VARIANCE", View::TemporalAntiAliasingOptions::BoxType::VARIANCE)
+    .value("AABB_VARIANCE", View::TemporalAntiAliasingOptions::BoxType::AABB_VARIANCE)
+    ;
+
+enum_<View::TemporalAntiAliasingOptions::BoxClipping>("View$TemporalAntiAliasingOptions$BoxClipping")
+    .value("ACCURATE", View::TemporalAntiAliasingOptions::BoxClipping::ACCURATE)
+    .value("CLAMP", View::TemporalAntiAliasingOptions::BoxClipping::CLAMP)
+    .value("NONE", View::TemporalAntiAliasingOptions::BoxClipping::NONE)
+    ;
+
+enum_<View::TemporalAntiAliasingOptions::JitterPattern>("View$TemporalAntiAliasingOptions$JitterPattern")
+    .value("RGSS_X4", View::TemporalAntiAliasingOptions::JitterPattern::RGSS_X4)
+    .value("UNIFORM_HELIX_X4", View::TemporalAntiAliasingOptions::JitterPattern::UNIFORM_HELIX_X4)
+    .value("HALTON_23_X8", View::TemporalAntiAliasingOptions::JitterPattern::HALTON_23_X8)
+    .value("HALTON_23_X16", View::TemporalAntiAliasingOptions::JitterPattern::HALTON_23_X16)
+    ;
+
 enum_<View::AntiAliasing>("View$AntiAliasing")
     .value("NONE", View::AntiAliasing::NONE)
     .value("FXAA", View::AntiAliasing::FXAA)

web/filament-js/package.json

@@ -1,6 +1,6 @@
 {
   "name": "filament",
-  "version": "1.49.1",
+  "version": "1.49.3",
   "description": "Real-time physically based rendering engine",
   "main": "filament.js",
   "module": "filament.js",