Include webgpu matc arg in android builds when appropriate.

.github/workflows/presubmit.yml

@@ -19,8 +19,6 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
-        with:
-          fetch-depth: 0
       - name: Run build script
         run: |
           WORKFLOW_OS=`echo \`uname\` | sed "s/Darwin/mac/" | tr [:upper:] [:lower:]`
@@ -46,8 +44,6 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
-        with:
-          fetch-depth: 0
       - uses: actions/setup-java@v3
         with:
           distribution: 'temurin'
@@ -64,8 +60,6 @@ jobs:
 
     steps:
       - uses: actions/checkout@v4.1.6
-        with:
-          fetch-depth: 0
       - name: Run build script
         run: |
           cd build/ios && printf "y" | ./build.sh presubmit
@@ -128,31 +122,3 @@ jobs:
         run: source ./build/linux/ci-common.sh && ./build.sh -W debug test_filamat filament
       - name: Run test
         run: ./out/cmake-debug/libs/filamat/test_filamat --gtest_filter=MaterialCompiler.Wgsl*
-
-  code-correcteness:
-    name: code-correctness
-    runs-on: 'macos-14-xlarge'
-    steps:
-      - uses: actions/checkout@v4.1.6
-        with:
-          fetch-depth: 0
-      - name: Set up Homebrew
-        id: set-up-homebrew
-        uses: Homebrew/actions/setup-homebrew@master
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: '3.x'
-      - name: Install prerequisites
-        run: |
-          pip install pyyaml
-          brew install llvm
-          sudo ln -s "$(brew --prefix llvm)/bin/clang-tidy" "/usr/local/bin/clang-tidy"
-      - name: Run build script
-        # We need to build before clang-tidy can run analysis
-        run: |
-          # This will build for all three desktop backends on mac
-          ./build.sh -p desktop debug gltf_viewer
-      - name: Run test
-        run: |
-          bash test/code-correctness/test.sh

android/buildSrc/src/main/groovy/FilamentPlugin.groovy

@@ -142,13 +142,16 @@ abstract class MaterialCompiler extends TaskWithBinary {
                 if (!exclude_vulkan) {
                     matcArgs += ['-a', 'vulkan']
                 }
+
                 def include_webgpu = providers
                         .gradleProperty("com.google.android.filament.include-webgpu")
                         .forUseAtConfigurationTime().present
-                if (include_webgpu) {
-                    matcArgs += ['-a', 'webgpu', '--variant-filter=skinning,stereo']
+
+                if (!include_webgpu) {
+                    matcArgs += ['-a', 'webgpu']
                 }
 
+
                 def mat_no_opt = providers
                         .gradleProperty("com.google.android.filament.matnopt")
                         .forUseAtConfigurationTime().present

filament/backend/CMakeLists.txt

@@ -177,6 +177,29 @@ endif()
 if (FILAMENT_SUPPORTS_VULKAN)
     list(APPEND SRCS
             include/backend/platforms/VulkanPlatform.h
+            src/vulkan/VulkanDescriptorSetCache.cpp
+            src/vulkan/VulkanDescriptorSetCache.h
+            src/vulkan/VulkanDescriptorSetLayoutCache.cpp
+            src/vulkan/VulkanDescriptorSetLayoutCache.h
+            src/vulkan/VulkanPipelineLayoutCache.cpp
+            src/vulkan/VulkanPipelineLayoutCache.h
+            src/vulkan/memory/ResourceManager.cpp
+            src/vulkan/memory/ResourceManager.h
+            src/vulkan/memory/ResourcePointer.h
+            src/vulkan/memory/Resource.cpp
+            src/vulkan/memory/Resource.h
+            src/vulkan/platform/VulkanPlatform.cpp
+            src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp
+            src/vulkan/platform/VulkanPlatformSwapChainImpl.h
+            src/vulkan/utils/Conversion.cpp
+            src/vulkan/utils/Conversion.h
+            src/vulkan/utils/Definitions.h
+            src/vulkan/utils/Helper.h
+            src/vulkan/utils/Image.h
+            src/vulkan/utils/Image.cpp
+            src/vulkan/utils/Spirv.h
+            src/vulkan/utils/Spirv.cpp
+            src/vulkan/utils/StaticVector.h
             src/vulkan/VulkanAsyncHandles.h
             src/vulkan/VulkanBlitter.cpp
             src/vulkan/VulkanBlitter.h
@@ -187,56 +210,31 @@ if (FILAMENT_SUPPORTS_VULKAN)
             src/vulkan/VulkanConstants.h
             src/vulkan/VulkanContext.cpp
             src/vulkan/VulkanContext.h
-            src/vulkan/VulkanDescriptorSetCache.cpp
-            src/vulkan/VulkanDescriptorSetCache.h
-            src/vulkan/VulkanDescriptorSetLayoutCache.cpp
-            src/vulkan/VulkanDescriptorSetLayoutCache.h
             src/vulkan/VulkanDriver.cpp
             src/vulkan/VulkanDriver.h
             src/vulkan/VulkanDriverFactory.h
-            src/vulkan/VulkanExternalImageManager.cpp
-            src/vulkan/VulkanExternalImageManager.h
             src/vulkan/VulkanFboCache.cpp
             src/vulkan/VulkanFboCache.h
             src/vulkan/VulkanHandles.cpp
             src/vulkan/VulkanHandles.h
-            src/vulkan/VulkanMemory.cpp
             src/vulkan/VulkanMemory.h
+            src/vulkan/VulkanMemory.cpp
             src/vulkan/VulkanPipelineCache.cpp
             src/vulkan/VulkanPipelineCache.h
-            src/vulkan/VulkanPipelineLayoutCache.cpp
-            src/vulkan/VulkanPipelineLayoutCache.h
             src/vulkan/VulkanQueryManager.cpp
             src/vulkan/VulkanQueryManager.h
-            src/vulkan/VulkanReadPixels.cpp
-            src/vulkan/VulkanReadPixels.h
             src/vulkan/VulkanSamplerCache.cpp
             src/vulkan/VulkanSamplerCache.h
             src/vulkan/VulkanStagePool.cpp
             src/vulkan/VulkanStagePool.h
             src/vulkan/VulkanSwapChain.cpp
             src/vulkan/VulkanSwapChain.h
+            src/vulkan/VulkanReadPixels.cpp
+            src/vulkan/VulkanReadPixels.h
             src/vulkan/VulkanTexture.cpp
             src/vulkan/VulkanTexture.h
             src/vulkan/VulkanYcbcrConversionCache.cpp
             src/vulkan/VulkanYcbcrConversionCache.h
-            src/vulkan/memory/Resource.cpp
-            src/vulkan/memory/Resource.h
-            src/vulkan/memory/ResourceManager.cpp
-            src/vulkan/memory/ResourceManager.h
-            src/vulkan/memory/ResourcePointer.h
-            src/vulkan/platform/VulkanPlatform.cpp
-            src/vulkan/platform/VulkanPlatformSwapChainImpl.cpp
-            src/vulkan/platform/VulkanPlatformSwapChainImpl.h
-            src/vulkan/utils/Conversion.cpp
-            src/vulkan/utils/Conversion.h
-            src/vulkan/utils/Definitions.h
-            src/vulkan/utils/Helper.h
-            src/vulkan/utils/Image.cpp
-            src/vulkan/utils/Image.h
-            src/vulkan/utils/Spirv.cpp
-            src/vulkan/utils/Spirv.h
-            src/vulkan/utils/StaticVector.h
     )
     if (LINUX OR WIN32)
         list(APPEND SRCS src/vulkan/platform/VulkanPlatformLinuxWindows.cpp)
@@ -259,6 +257,8 @@ if (FILAMENT_SUPPORTS_WEBGPU)
             src/webgpu/WebGPUDriver.h
             src/webgpu/WebGPUSwapChain.cpp
             src/webgpu/WebGPUSwapChain.h
+            src/webgpu/WebGPUHandles.cpp
+            src/webgpu/WebGPUHandles.h
     )
     if (WIN32)
         list(APPEND SRCS src/webgpu/platform/WebGPUPlatformWindows.cpp)
@@ -508,7 +508,6 @@ if (APPLE OR LINUX)
         test/ImageExpectations.cpp
         test/Lifetimes.cpp
         test/Shader.cpp
-        test/SharedShaders.cpp
         test/test_FeedbackLoops.cpp
         test/test_Blit.cpp
         test/test_MissingRequiredAttributes.cpp

filament/backend/include/backend/Platform.h

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

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

@@ -104,6 +104,10 @@ public:
 
         // Semaphore to be signaled once the image is available.
         VkSemaphore imageReadySemaphore = VK_NULL_HANDLE;
+
+        // A function called right before vkQueueSubmit. After this call, the image must be 
+        // available. This pointer can be null if imageReadySemaphore is not VK_NULL_HANDLE.
+        std::function<void(SwapChainPtr handle)> explicitImageReadyWait = nullptr;
     };
 
     VulkanPlatform();
@@ -294,16 +298,6 @@ public:
     VkQueue getProtectedGraphicsQueue() const noexcept;
 
     struct ExternalImageMetadata {
-        /**
-         * The Filament texture format.
-         */
-        TextureFormat filamentFormat;
-
-        /**
-         * The Filament texture usage.
-         */
-        TextureUsage filamentUsage;
-
         /**
          * The width of the external image
          */
@@ -314,6 +308,11 @@ public:
          */
         uint32_t height;
 
+        /**
+         * The layerCount of the external image
+         */
+        uint32_t layerCount;
+
         /**
          * The layer count of the external image
          */
@@ -329,6 +328,11 @@ public:
          */
         VkFormat format;
 
+        /**
+         * An external buffer can be protected. This tells you if it is.
+         */
+        bool isProtected;
+
         /**
          * The type of external format (opaque int) if used.
          */
@@ -348,44 +352,20 @@ public:
          * Heap information
          */
         uint32_t memoryTypeBits;
-
-        /**
-         * Ycbcr conversion components
-         */
-        VkComponentMapping ycbcrConversionComponents;
-
-        /**
-         * Ycbcr model
-         */
-        VkSamplerYcbcrModelConversion ycbcrModel;
-
-        /**
-         * Ycbcr range
-         */
-        VkSamplerYcbcrRange ycbcrRange;
-
-        /**
-         * Ycbcr x chroma offset
-         */
-        VkChromaLocation xChromaOffset;
-
-        /**
-         * Ycbcr y chroma offset
-         */
-        VkChromaLocation yChromaOffset;
     };
-
-
-    // Note that the image metadata might change per-frame, hence we need a method for extracting
-    // it.
-    virtual ExternalImageMetadata extractExternalImageMetadata(ExternalImageHandleRef image) const {
-        return {};
-    }
+    virtual ExternalImageMetadata getExternalImageMetadata(ExternalImageHandleRef externalImage);
 
     using ImageData = std::pair<VkImage, VkDeviceMemory>;
-    virtual ImageData createVkImageFromExternal(ExternalImageHandleRef image) const {
-        return { VK_NULL_HANDLE, VK_NULL_HANDLE };
-    }
+    virtual ImageData createExternalImageData(ExternalImageHandleRef externalImage,
+            const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex,
+            VkImageUsageFlags usage);
+
+    virtual VkSampler createExternalSampler(SamplerYcbcrConversion chroma,
+            SamplerParams sampler, uint32_t internalFormat);
+
+    virtual VkImageView createExternalImageView(SamplerYcbcrConversion chroma,
+            uint32_t internalFormat, VkImage image, VkImageSubresourceRange range,
+            VkImageViewType viewType, VkComponentMapping swizzle);
 
 protected:
     virtual ExtensionSet getSwapchainInstanceExtensions() const;
@@ -398,6 +378,20 @@ private:
     // Platform dependent helper methods
     static ExtensionSet getSwapchainInstanceExtensionsImpl();
 
+    static ExternalImageMetadata getExternalImageMetadataImpl(ExternalImageHandleRef externalImage,
+            VkDevice device);
+
+    static ImageData createExternalImageDataImpl(ExternalImageHandleRef externalImage,
+            VkDevice device, const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex,
+            VkImageUsageFlags usage);
+    static VkSampler createExternalSamplerImpl(VkDevice device,
+            SamplerYcbcrConversion chroma, SamplerParams sampler,
+            uint32_t internalFormat);
+    static VkImageView createExternalImageViewImpl(VkDevice device,
+            SamplerYcbcrConversion chroma, uint32_t internalFormat, VkImage image,
+            VkImageSubresourceRange range, VkImageViewType viewType,
+            VkComponentMapping swizzle);
+
     // Platform dependent helper methods
     static SurfaceBundle createVkSurfaceKHRImpl(void* nativeWindow, VkInstance instance,
             uint64_t flags) noexcept;

filament/backend/include/backend/platforms/VulkanPlatformAndroid.h

@@ -26,7 +26,7 @@ namespace filament::backend {
 
 class VulkanPlatformAndroid : public VulkanPlatform {
 public:
-    ExternalImageHandle UTILS_PUBLIC createExternalImage(AHardwareBuffer const* buffer,
+    Platform::ExternalImageHandle UTILS_PUBLIC createExternalImage(AHardwareBuffer const* buffer,
             bool sRGB) noexcept;
 
     struct UTILS_PUBLIC ExternalImageDescAndroid {
@@ -39,26 +39,31 @@ public:
     ExternalImageDescAndroid UTILS_PUBLIC getExternalImageDesc(
             ExternalImageHandleRef externalImage) const noexcept;
 
-    virtual ExternalImageMetadata extractExternalImageMetadata(
-            ExternalImageHandleRef image) const override;
-
-    virtual ImageData createVkImageFromExternal(ExternalImageHandleRef image) const override;
-
 protected:
-    virtual ExtensionSet getSwapchainInstanceExtensions() const override;
-
-    using SurfaceBundle = VulkanPlatform::SurfaceBundle;
-    virtual SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
-            uint64_t flags) const noexcept override;
-
-private:
     struct ExternalImageVulkanAndroid : public Platform::ExternalImage {
         AHardwareBuffer* aHardwareBuffer = nullptr;
         bool sRGB = false;
+        unsigned int width;  // Texture width
+        unsigned int height; // Texture height
+        TextureFormat format;// Texture format
+        TextureUsage usage;  // Texture usage flags
 
     protected:
         ~ExternalImageVulkanAndroid() override;
     };
+
+    virtual ExternalImageMetadata getExternalImageMetadata(ExternalImageHandleRef externalImage);
+
+    using ImageData = VulkanPlatform::ImageData;
+    virtual ImageData createExternalImageData(ExternalImageHandleRef externalImage,
+            const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex,
+            VkImageUsageFlags usage);
+
+    virtual ExtensionSet getSwapchainInstanceExtensions() const;
+
+    using SurfaceBundle = VulkanPlatform::SurfaceBundle;
+    virtual SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
+            uint64_t flags) const noexcept;
 };
 
 }// namespace filament::backend

filament/backend/src/metal/PlatformMetal.mm

@@ -45,9 +45,6 @@ PlatformMetal::~PlatformMetal() noexcept {
 }
 
 Driver* PlatformMetal::createDriver(void* /*sharedContext*/, const Platform::DriverConfig& driverConfig) noexcept {
-    pImpl->mDrawableFailureBehavior = driverConfig.metalDisablePanicOnDrawableFailure
-            ? DrawableFailureBehavior::ABORT_FRAME
-            : DrawableFailureBehavior::PANIC;
     return MetalDriverFactory::create(this, driverConfig);
 }
 

filament/backend/src/vulkan/VulkanConstants.h

@@ -93,7 +93,7 @@
 #endif
 
 #ifndef NDEBUG
-#define FVK_DEBUG_FLAGS (FVK_DEBUG_PERFORMANCE | FVK_DEBUG_FORWARDED_FLAG | FVK_DEBUG_VALIDATION)
+#define FVK_DEBUG_FLAGS (FVK_DEBUG_PERFORMANCE | FVK_DEBUG_FORWARDED_FLAG)
 #else
 #define FVK_DEBUG_FLAGS 0
 #endif

filament/backend/src/vulkan/VulkanContext.h

@@ -70,11 +70,13 @@ struct VulkanRenderPass {
 // context are stored in VulkanPlatform.
 struct VulkanContext {
 public:
-    static uint32_t selectMemoryType(VkPhysicalDeviceMemoryProperties const& memoryProperties,
-            uint32_t flags, VkFlags reqs) {
+    inline uint32_t selectMemoryType(uint32_t flags, VkFlags reqs) const {
+        if ((reqs & VK_MEMORY_PROPERTY_PROTECTED_BIT) != 0) {
+            assert_invariant(isProtectedMemorySupported() == true);
+        }
         for (uint32_t i = 0; i < VK_MAX_MEMORY_TYPES; i++) {
             if (flags & 1) {
-                if ((memoryProperties.memoryTypes[i].propertyFlags & reqs) == reqs) {
+                if ((mMemoryProperties.memoryTypes[i].propertyFlags & reqs) == reqs) {
                     return i;
                 }
             }
@@ -83,13 +85,6 @@ public:
         return (uint32_t) VK_MAX_MEMORY_TYPES;
     }
 
-    inline uint32_t selectMemoryType(uint32_t flags, VkFlags reqs) const {
-        if ((reqs & VK_MEMORY_PROPERTY_PROTECTED_BIT) != 0) {
-            assert_invariant(isProtectedMemorySupported());
-        }
-        return selectMemoryType(mMemoryProperties, flags, reqs);
-    }
-
     inline fvkutils::VkFormatList const& getAttachmentDepthStencilFormats() const {
         return mDepthStencilFormats;
     }
@@ -123,7 +118,7 @@ public:
     }
 
     inline bool isMultiviewEnabled() const noexcept {
-        return mPhysicalDeviceVk11Features.multiview == VK_TRUE;
+        return mMultiviewEnabled;
     }
 
     inline bool isClipDistanceSupported() const noexcept {
@@ -147,9 +142,6 @@ private:
     VkPhysicalDeviceProperties2 mPhysicalDeviceProperties = {
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2,
     };
-    VkPhysicalDeviceVulkan11Features mPhysicalDeviceVk11Features = {
-        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES,
-    };
     VkPhysicalDeviceFeatures2 mPhysicalDeviceFeatures = {
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2,
     };
@@ -162,6 +154,7 @@ private:
     };
     bool mDebugMarkersSupported = false;
     bool mDebugUtilsSupported = false;
+    bool mMultiviewEnabled = false;
     bool mLazilyAllocatedMemorySupported = false;
     bool mProtectedMemorySupported = false;
 

filament/backend/src/vulkan/VulkanDescriptorSetCache.cpp

@@ -282,25 +282,18 @@ void VulkanDescriptorSetCache::commit(VulkanCommandBuffer* commands,
     fvkutils::DescriptorSetMask curMask = setMask;
 
     auto& updateSets = mStashedSets;
-    bool const pipelineLayoutIsSame = mLastBoundInfo.pipelineLayout == pipelineLayout;
+    auto& lastBoundSets = mLastBoundInfo.boundSets;
 
-    if (pipelineLayoutIsSame) {
-        auto& lastBoundSets = mLastBoundInfo.boundSets;
-        setMask.forEachSetBit([&](size_t index) {
-            if (!updateSets[index] || updateSets[index] == lastBoundSets[index]) {
-                curMask.unset(index);
-            }
-        });
-        if (curMask.none() &&
-                mLastBoundInfo.setMask == setMask && mLastBoundInfo.boundSets == updateSets) {
-            return;
+    setMask.forEachSetBit([&](size_t index) {
+        if (!updateSets[index] || updateSets[index] == lastBoundSets[index]) {
+            curMask.unset(index);
         }
-    } else {
-        setMask.forEachSetBit([&](size_t index) {
-            if (!updateSets[index]) {
-                curMask.unset(index);
-            }
-        });
+    });
+
+    if (curMask.none() &&
+            (mLastBoundInfo.pipelineLayout == pipelineLayout && mLastBoundInfo.setMask == setMask &&
+                    mLastBoundInfo.boundSets == updateSets)) {
+        return;
     }
 
     curMask.forEachSetBit([&updateSets, commands, pipelineLayout](size_t index) {
@@ -308,7 +301,7 @@ void VulkanDescriptorSetCache::commit(VulkanCommandBuffer* commands,
         auto set = updateSets[index];
         VkCommandBuffer const cmdbuffer = commands->buffer();
         vkCmdBindDescriptorSets(cmdbuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, index,
-                1, &set->getVkSet(), set->uniqueDynamicUboCount, set->getOffsets()->data());
+                1, &set->vkSet, set->uniqueDynamicUboCount, set->getOffsets()->data());
         commands->acquire(set);
     });
 
@@ -336,7 +329,7 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
     }
     VkWriteDescriptorSet const descriptorWrite = {
         .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
-        .dstSet = set->getVkSet(),
+        .dstSet = set->vkSet,
         .dstBinding = binding,
         .descriptorCount = 1,
         .descriptorType = type,
@@ -349,6 +342,10 @@ void VulkanDescriptorSetCache::updateBuffer(fvkmemory::resource_ptr<VulkanDescri
 void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
         uint8_t binding, fvkmemory::resource_ptr<VulkanTexture> texture,
         VkSampler sampler) noexcept {
+    VkDescriptorImageInfo info{
+        .sampler = sampler,
+        .imageLayout = fvkutils::getVkLayout(texture->getDefaultLayout()),
+    };
     VkImageSubresourceRange range = texture->getPrimaryViewRange();
     VkImageViewType const expectedType = texture->getViewType();
     if (any(texture->usage & TextureUsage::DEPTH_ATTACHMENT) &&
@@ -358,16 +355,12 @@ void VulkanDescriptorSetCache::updateSampler(fvkmemory::resource_ptr<VulkanDescr
         range.levelCount = 1;
         range.layerCount = 1;
     }
-    VkDescriptorImageInfo info{
-        .sampler = sampler,
-        .imageView = texture->getView(range),
-        .imageLayout = fvkutils::getVkLayout(texture->getDefaultLayout()),
-    };
+    info.imageView = texture->getView(range);
 
     VkWriteDescriptorSet const descriptorWrite = {
         .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
         .pNext = nullptr,
-        .dstSet = set->getVkSet(),
+        .dstSet = set->vkSet,
         .dstBinding = binding,
         .descriptorCount = 1,
         .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
@@ -385,10 +378,10 @@ void VulkanDescriptorSetCache::updateInputAttachment(
 
 fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetCache::createSet(
         Handle<HwDescriptorSet> handle, fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-    auto const vkSet = getVkSet(layout);
+    auto const vkSet = mDescriptorPool->obtainSet(layout);
     auto const& count = layout->count;
     auto const vklayout = layout->getVkLayout();
-    auto set = fvkmemory::resource_ptr<VulkanDescriptorSet>::make(mResourceManager, handle,
+    return fvkmemory::resource_ptr<VulkanDescriptorSet>::make(mResourceManager, handle, vkSet,
             layout->bitmask.dynamicUbo, layout->count.dynamicUbo,
             [vkSet, count, vklayout, this](VulkanDescriptorSet*) {
                 // Note that mDescriptorPool could be gone due to terminate (when the backend shuts
@@ -397,20 +390,10 @@ fvkmemory::resource_ptr<VulkanDescriptorSet> VulkanDescriptorSetCache::createSet
                     mDescriptorPool->recycle(count, vklayout, vkSet);
                 }
             });
-    set->setVkSet(vkSet);
-    return set;
 }
 
-VkDescriptorSet VulkanDescriptorSetCache::getVkSet(
-        fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-    return mDescriptorPool->obtainSet(layout);
+void VulkanDescriptorSetCache::gc() {
+    mStashedSets = {};
 }
 
-void VulkanDescriptorSetCache::manualRecyle(VulkanDescriptorSetLayout::Count const& count,
-        VkDescriptorSetLayout vklayout, VkDescriptorSet vkSet) {
-    mDescriptorPool->recycle(count, vklayout, vkSet);
-}
-
-void VulkanDescriptorSetCache::gc() { mStashedSets = {}; }
-
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanDescriptorSetCache.h

@@ -41,8 +41,6 @@ public:
             VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT;
 
     using DescriptorSetLayoutArray = VulkanDescriptorSetLayout::DescriptorSetLayoutArray;
-    using DescriptorSetArray =
-            std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>, UNIQUE_DESCRIPTOR_SET_COUNT>;
 
     VulkanDescriptorSetCache(VkDevice device, fvkmemory::ResourceManager* resourceManager);
     ~VulkanDescriptorSetCache();
@@ -70,21 +68,14 @@ public:
     fvkmemory::resource_ptr<VulkanDescriptorSet> createSet(Handle<HwDescriptorSet> handle,
             fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
 
-    // This method is only meant to be used with external samplers (or internally within this
-    // class).
-    VkDescriptorSet getVkSet(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
-
-    // This method is only meant to be used with external samplers.
-    void manualRecyle(VulkanDescriptorSetLayout::Count const& count, VkDescriptorSetLayout vklayout,
-            VkDescriptorSet vkSet);
-
-    DescriptorSetArray const& getBoundSets() const { return mStashedSets; }
-
     void gc();
 
 private:
     class DescriptorInfinitePool;
 
+    using DescriptorSetArray =
+            std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>, UNIQUE_DESCRIPTOR_SET_COUNT>;
+
     VkDevice mDevice;
     fvkmemory::ResourceManager* mResourceManager;
     std::unique_ptr<DescriptorInfinitePool> mDescriptorPool;

filament/backend/src/vulkan/VulkanDescriptorSetLayoutCache.cpp

@@ -18,8 +18,6 @@
 
 #include "VulkanHandles.h"
 
-#include <utils/Hash.h>
-
 namespace filament::backend {
 
 namespace {
@@ -60,56 +58,6 @@ uint32_t appendBindings(VkDescriptorSetLayoutBinding* toBind, VkDescriptorType t
     return count;
 }
 
-uint32_t appendSamplerBindings(VkDescriptorSetLayoutBinding* toBind,
-        fvkutils::SamplerBitmask const& mask, fvkutils::SamplerBitmask const& external,
-        utils::FixedCapacityVector<VkSampler> const& immutableSamplers) {
-    using Bitmask = fvkutils::SamplerBitmask;
-    uint32_t count = 0;
-    Bitmask alreadySeen;
-    uint8_t immutableIndex = 0;
-    size_t const immutableSamplerCount = immutableSamplers.size();
-    mask.forEachSetBit([&](size_t index) {
-        VkShaderStageFlags stages = 0;
-        uint32_t binding = 0;
-        if (index < fvkutils::getFragmentStageShift<Bitmask>()) {
-            binding = (uint32_t) index;
-            stages |= VK_SHADER_STAGE_VERTEX_BIT;
-            auto fragIndex = index + fvkutils::getFragmentStageShift<Bitmask>();
-            if (mask.test(fragIndex)) {
-                stages |= VK_SHADER_STAGE_FRAGMENT_BIT;
-                alreadySeen.set(fragIndex);
-            }
-        } else if (!alreadySeen.test(index)) {
-            // We are in fragment stage bits
-            binding = (uint32_t) (index - fvkutils::getFragmentStageShift<Bitmask>());
-            stages |= VK_SHADER_STAGE_FRAGMENT_BIT;
-        }
-
-        if (stages) {
-            toBind[count++] = {
-                .binding = binding,
-                .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
-                .descriptorCount = 1,
-                .stageFlags = stages,
-                .pImmutableSamplers = external[index] && immutableSamplerCount > immutableIndex
-                                              ? &immutableSamplers[immutableIndex++]
-                                              : nullptr,
-            };
-        }
-    });
-    return count;
-}
-
-uint64_t computeImmutableSamplerHash(utils::FixedCapacityVector<VkSampler> const& samplers) {
-    size_t const size = samplers.size();
-    if (size == 0) {
-        return 0;
-    } else if (size == 1) {
-        return (uint64_t) samplers[0];
-    }
-    return utils::hash::murmur3((uint32_t*) samplers.data(), samplers.size() * 2, 0);
-}
-
 } // anonymous namespace
 
 VulkanDescriptorSetLayoutCache::VulkanDescriptorSetLayoutCache(VkDevice device,
@@ -125,44 +73,37 @@ void VulkanDescriptorSetLayoutCache::terminate() noexcept {
     }
 }
 
-VkDescriptorSetLayout VulkanDescriptorSetLayoutCache::getVkLayout(
-        VulkanDescriptorSetLayout::Bitmask const& bitmasks,
-        utils::FixedCapacityVector<VkSampler> immutableSamplers) {
-    LayoutKey key = {
-        .bitmask = bitmasks,
-        .immutableSamplerHash = computeImmutableSamplerHash(immutableSamplers),
-    };
-    if (auto itr = mVkLayouts.find(key); itr != mVkLayouts.end()) {
-        return itr->second;
-    }
-
-    VkDescriptorSetLayoutBinding toBind[VulkanDescriptorSetLayout::MAX_BINDINGS];
-    uint32_t count = 0;
-    count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC,
-            bitmasks.dynamicUbo);
-    count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, bitmasks.ubo);
-    count += appendSamplerBindings(&toBind[count], bitmasks.sampler, bitmasks.externalSampler,
-            immutableSamplers);
-    count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
-            bitmasks.inputAttachment);
-
-    assert_invariant(count != 0 && "Need at least one binding for descriptor set layout.");
-    VkDescriptorSetLayoutCreateInfo dlinfo = {
-        .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
-        .bindingCount = count,
-        .pBindings = toBind,
-    };
-    VkDescriptorSetLayout vklayout;
-    vkCreateDescriptorSetLayout(mDevice, &dlinfo, VKALLOC, &vklayout);
-    mVkLayouts[key] = vklayout;
-    return vklayout;
-}
-
 fvkmemory::resource_ptr<VulkanDescriptorSetLayout> VulkanDescriptorSetLayoutCache::createLayout(
         Handle<HwDescriptorSetLayout> handle, backend::DescriptorSetLayout&& info) {
     auto layout = fvkmemory::resource_ptr<VulkanDescriptorSetLayout>::make(mResourceManager, handle,
             info);
-    layout->setVkLayout(getVkLayout(layout->bitmask));
+    VkDescriptorSetLayout vklayout = VK_NULL_HANDLE;
+    auto const& bitmasks = layout->bitmask;
+    if (auto itr = mVkLayouts.find(bitmasks); itr != mVkLayouts.end()) {
+        vklayout = itr->second;
+    } else {
+        VkDescriptorSetLayoutBinding toBind[VulkanDescriptorSetLayout::MAX_BINDINGS];
+        uint32_t count = 0;
+        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC,
+                bitmasks.dynamicUbo);
+        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, bitmasks.ubo);
+        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+                bitmasks.sampler);
+        count += appendBindings(&toBind[count], VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT,
+                bitmasks.inputAttachment);
+
+        assert_invariant(count != 0 && "Need at least one binding for descriptor set layout.");
+        VkDescriptorSetLayoutCreateInfo dlinfo = {
+            .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
+            .pNext = nullptr,
+            .bindingCount = count,
+            .pBindings = toBind,
+        };
+
+        vkCreateDescriptorSetLayout(mDevice, &dlinfo, VKALLOC, &vklayout);
+        mVkLayouts[bitmasks] = vklayout;
+    }
+    layout->setVkLayout(vklayout);
     return layout;
 }
 

filament/backend/src/vulkan/VulkanDescriptorSetLayoutCache.h

@@ -26,11 +26,12 @@
 #include <backend/TargetBufferInfo.h>
 
 #include <utils/bitset.h>
-#include <utils/FixedCapacityVector.h>
 
 #include <bluevk/BlueVK.h>
 #include <tsl/robin_map.h>
 
+#include <memory>
+
 namespace filament::backend {
 
 class VulkanDescriptorSetLayoutCache {
@@ -40,34 +41,21 @@ public:
 
     void terminate() noexcept;
 
-    // Just a wrapper around getVkLayout()
     fvkmemory::resource_ptr<VulkanDescriptorSetLayout> createLayout(
             Handle<HwDescriptorSetLayout> handle, backend::DescriptorSetLayout&& info);
 
-    // This method is meant to be used with external samplers
-    VkDescriptorSetLayout getVkLayout(VulkanDescriptorSetLayout::Bitmask const& bitmasks,
-            utils::FixedCapacityVector<VkSampler> immutableSamplers = {});
-
 private:
     VkDevice mDevice;
     fvkmemory::ResourceManager* mResourceManager;
 
-    struct LayoutKey {
-        // this describes the layout using bitset.
-        VulkanDescriptorSetLayout::Bitmask bitmask = {};
-        // number of immutable samplers can be arbitrary; so we hash them into 64-bit.
-        uint64_t immutableSamplerHash = 0;
-    };
-    static_assert(sizeof(LayoutKey) == 48);
-
-    using LayoutKeyHashFn = utils::hash::MurmurHashFn<LayoutKey>;
-    struct LayoutKeyEqual {
-        bool operator()(LayoutKey const& k1, LayoutKey const& k2) const {
-            return k1.bitmask == k2.bitmask && k1.immutableSamplerHash == k2.immutableSamplerHash;
-        }
+    using BitmaskGroup = VulkanDescriptorSetLayout::Bitmask;
+    using BitmaskGroupHashFn = utils::hash::MurmurHashFn<BitmaskGroup>;
+    struct BitmaskGroupEqual {
+        bool operator()(BitmaskGroup const& k1, BitmaskGroup const& k2) const { return k1 == k2; }
     };
 
-    tsl::robin_map<LayoutKey, VkDescriptorSetLayout, LayoutKeyHashFn, LayoutKeyEqual> mVkLayouts;
+    tsl::robin_map<BitmaskGroup, VkDescriptorSetLayout, BitmaskGroupHashFn, BitmaskGroupEqual>
+            mVkLayouts;
 };
 
 } // namespace filament::backend

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -200,7 +200,9 @@ Dispatcher VulkanDriver::getDispatcher() const noexcept {
 VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext const& context,
         Platform::DriverConfig const& driverConfig) noexcept
     : mPlatform(platform),
-      mResourceManager(driverConfig.handleArenaSize, driverConfig.disableHandleUseAfterFreeCheck,
+      mResourceManager(
+              driverConfig.handleArenaSize,
+              driverConfig.disableHandleUseAfterFreeCheck,
               driverConfig.disableHeapHandleTags),
       mAllocator(createAllocator(mPlatform->getInstance(), mPlatform->getPhysicalDevice(),
               mPlatform->getDevice())),
@@ -219,8 +221,6 @@ VulkanDriver::VulkanDriver(VulkanPlatform* platform, VulkanContext const& contex
       mDescriptorSetLayoutCache(mPlatform->getDevice(), &mResourceManager),
       mDescriptorSetCache(mPlatform->getDevice(), &mResourceManager),
       mQueryManager(mPlatform->getDevice()),
-      mExternalImageManager(platform, &mSamplerCache, &mYcbcrConversionCache, &mDescriptorSetCache,
-              &mDescriptorSetLayoutCache),
       mIsSRGBSwapChainSupported(mPlatform->getCustomization().isSRGBSwapChainSupported),
       mStereoscopicType(driverConfig.stereoscopicType) {
 
@@ -313,7 +313,7 @@ void VulkanDriver::terminate() {
 
     mCurrentSwapChain = {};
     mDefaultRenderTarget = {};
-    mPipelineState = {};
+    mBoundPipeline = {};
 
     mQueryManager.terminate();
 
@@ -325,10 +325,6 @@ void VulkanDriver::terminate() {
 
     mCommands.terminate();
 
-    // Must come before samplerCache, ycbcrConversionCache, descriptorSetCache,
-    // descriptorSetLayoutCache
-    mExternalImageManager.terminate();
-
     mStagePool.terminate();
     mPipelineCache.terminate();
     mFramebufferCache.reset();
@@ -385,10 +381,6 @@ void VulkanDriver::beginFrame(int64_t monotonic_clock_ns,
         int64_t refreshIntervalNs, uint32_t frameId) {
     FVK_PROFILE_MARKER(PROFILE_NAME_BEGINFRAME);
     // Do nothing.
-
-    if (mAppState.hasExternalSamplers()) {
-        mExternalImageManager.onBeginFrame();
-    }
 }
 
 void VulkanDriver::setFrameScheduledCallback(Handle<HwSwapChain> sch, CallbackHandler* handler,
@@ -429,16 +421,12 @@ void VulkanDriver::updateDescriptorSetTexture(
     auto set = resource_ptr<VulkanDescriptorSet>::cast(&mResourceManager, dsh);
     auto texture = resource_ptr<VulkanTexture>::cast(&mResourceManager, th);
 
-    if (mExternalImageManager.isExternallySampledTexture(texture)) {
-        mExternalImageManager.bindExternallySampledTexture(set, binding, texture, params);
-        mAppState.hasBoundExternalImages = true;
-    } else {
-        VulkanSamplerCache::Params cacheParams = {
-            .sampler = params,
-        };
-        VkSampler const vksampler = mSamplerCache.getSampler(cacheParams);
-        mDescriptorSetCache.updateSampler(set, binding, texture, vksampler);
-    }
+    // TODO: YcbcrConversion?
+    VulkanSamplerCache::Params cacheParams = {
+        .sampler = params,
+    };
+    VkSampler const vksampler = mSamplerCache.getSampler(cacheParams);
+    mDescriptorSetCache.updateSampler(set, binding, texture, vksampler);
 }
 
 void VulkanDriver::flush(int) {
@@ -575,35 +563,41 @@ void VulkanDriver::createTextureExternalImage2R(Handle<HwTexture> th, backend::S
         backend::TextureFormat format, uint32_t width, uint32_t height, backend::TextureUsage usage,
         Platform::ExternalImageHandleRef externalImage) {
     FVK_SYSTRACE_SCOPE();
-    auto const& metadata = mPlatform->extractExternalImageMetadata(externalImage);
+    auto const& metadata = mPlatform->getExternalImageMetadata(externalImage);
+    if (metadata.isProtected) {
+        usage |= backend::TextureUsage::PROTECTED;
+    }
 
-    // In theory the following are reasonable expectations, but in practice it's hard for client's
-    // to match up the dimensions of the texture with that of the AHB.
-    // assert_invariant(width == metadata.width);
-    // assert_invariant(height == metadata.height);
-    // assert_invariant(format == metadata.filamentFormat);
-    // assert_invariant(fvkutils::getVkFormat(format) == metadata.format);
+    VkImageUsageFlags vkUsage = metadata.usage;
+    if (any(usage & TextureUsage::BLIT_SRC)) {
+        vkUsage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
+    }
+
+    if (any(usage & (TextureUsage::BLIT_DST | TextureUsage::UPLOADABLE))) {
+        vkUsage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT;
+    }
+
+    assert_invariant(width == metadata.width);
+    assert_invariant(height == metadata.height);
+    assert_invariant(fvkutils::getVkFormat(format) == metadata.format);
+
+    VkMemoryPropertyFlags const requiredMemoryFlags = any(usage & TextureUsage::UPLOADABLE)
+                                                              ? VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
+                                                              : VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+    uint32_t const memoryTypeIndex =
+            mContext.selectMemoryType(metadata.memoryTypeBits, requiredMemoryFlags);
+    FILAMENT_CHECK_POSTCONDITION(memoryTypeIndex != VK_MAX_MEMORY_TYPES)
+            << "failed to find a valid memory type for external image memory.";
 
     VkImage vkimg;
     VkDeviceMemory deviceMemory;
-    std::tie(vkimg, deviceMemory) = mPlatform->createVkImageFromExternal(externalImage);
-
-    VkSamplerYcbcrConversion conversion =
-            mExternalImageManager.getVkSamplerYcbcrConversion(metadata);
+    std::tie(vkimg, deviceMemory) =
+            mPlatform->createExternalImageData(externalImage, metadata, memoryTypeIndex, vkUsage);
 
     auto texture = resource_ptr<VulkanTexture>::make(&mResourceManager, th, mContext,
             mPlatform->getDevice(), mAllocator, &mResourceManager, &mCommands, vkimg, deviceMemory,
-            metadata.format, conversion, metadata.samples, metadata.width, metadata.height,
-            metadata.layers, usage, mStagePool);
-
-    if (conversion != VK_NULL_HANDLE) {
-        mExternalImageManager.addExternallySampledTexture(texture, externalImage);
-    }
-
-    // Unlike uploaded textures or swapchains, we need to explicit transition this
-    // texture into the read layout.
-    auto& commands = mCommands.get();
-    texture->transitionLayout(&commands, texture->getPrimaryViewRange(), VulkanLayout::READ_ONLY);
+            metadata.format, VK_NULL_HANDLE, metadata.samples, metadata.width, metadata.height,
+            metadata.layerCount, usage, mStagePool);
 
     texture.inc();
 }
@@ -636,8 +630,6 @@ void VulkanDriver::destroyTexture(Handle<HwTexture> th) {
     }
     auto texture = resource_ptr<VulkanTexture>::cast(&mResourceManager, th);
     texture.dec();
-
-    mExternalImageManager.removeExternallySampledTexture(texture);
 }
 
 void VulkanDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {
@@ -806,11 +798,6 @@ void VulkanDriver::createDescriptorSetR(Handle<HwDescriptorSet> dsh,
             fvkmemory::resource_ptr<VulkanDescriptorSetLayout>::cast(&mResourceManager, dslh);
     auto set = mDescriptorSetCache.createSet(dsh, layout);
     set.inc();
-
-    if (layout->hasExternalSamplers()) {
-        mAppState.hasExternalSamplerLayouts = true;
-        mExternalImageManager.addDescriptorSet(layout, set);
-    }
 }
 
 Handle<HwVertexBufferInfo> VulkanDriver::createVertexBufferInfoS() noexcept {
@@ -930,19 +917,11 @@ void VulkanDriver::destroyTimerQuery(Handle<HwTimerQuery> tqh) {
 void VulkanDriver::destroyDescriptorSetLayout(Handle<HwDescriptorSetLayout> dslh) {
     auto layout = resource_ptr<VulkanDescriptorSetLayout>::cast(&mResourceManager, dslh);
     layout.dec();
-
-    if (layout->hasExternalSamplers()) {
-        mExternalImageManager.removeDescriptorSetLayout(layout);
-    }
 }
 
 void VulkanDriver::destroyDescriptorSet(Handle<HwDescriptorSet> dsh) {
     auto set = resource_ptr<VulkanDescriptorSet>::cast(&mResourceManager, dsh);
     set.dec();
-
-    if (mAppState.hasExternalSamplers()) {
-        mExternalImageManager.removeDescriptorSet(set);
-    }
 }
 
 Handle<HwStream> VulkanDriver::createStreamNative(void* nativeStream) {
@@ -1503,6 +1482,7 @@ void VulkanDriver::endRenderPass(int) {
     // pipeline barrier between framebuffer writes and shader reads.
     rt->emitBarriersEndRenderPass(*mCurrentRenderPass.commandBuffer);
 
+    mRenderPassFboInfo = {};
     mCurrentRenderPass.renderTarget = {};
     mCurrentRenderPass.renderPass = VK_NULL_HANDLE;
 
@@ -1562,10 +1542,10 @@ void VulkanDriver::commit(Handle<HwSwapChain> sch) {
 
 void VulkanDriver::setPushConstant(backend::ShaderStage stage, uint8_t index,
         backend::PushConstantVariant value) {
-    assert_invariant(mPipelineState.program && "Expect a program when writing to push constants");
+    assert_invariant(mBoundPipeline.program && "Expect a program when writing to push constants");
     assert_invariant(mCurrentRenderPass.commandBuffer && "Should be called within a renderpass");
-    mPipelineState.program->writePushConstant(mCurrentRenderPass.commandBuffer->buffer(),
-            mPipelineState.pipelineLayout, stage, index, value);
+    mBoundPipeline.program->writePushConstant(mCurrentRenderPass.commandBuffer->buffer(),
+            mBoundPipeline.pipelineLayout, stage, index, value);
 }
 
 void VulkanDriver::insertEventMarker(char const* string) {
@@ -1744,27 +1724,6 @@ void VulkanDriver::blitDEPRECATED(TargetBufferFlags buffers,
 }
 
 void VulkanDriver::bindPipeline(PipelineState const& pipelineState) {
-    // We need to determine whether to delay bindning until draw().
-    mPipelineState.bindInDraw.first = false;
-    if (mAppState.hasExternalSamplers()) {
-        auto& layouts = pipelineState.pipelineLayout.setLayout;
-        auto haveExternalSamplers = [&](auto hwHandle) {
-            if (!hwHandle) {
-                return false;
-            }
-            auto layout =
-                    resource_ptr<VulkanDescriptorSetLayout>::cast(&mResourceManager, hwHandle);
-            return layout->hasExternalSamplers();
-        };
-        if (std::any_of(layouts.begin(), layouts.end(), haveExternalSamplers)) {
-            mPipelineState.bindInDraw = { true, pipelineState };
-            return;
-        }
-    }
-    bindPipelineImpl(pipelineState);
-}
-
-void VulkanDriver::bindPipelineImpl(PipelineState const& pipelineState) {
     FVK_SYSTRACE_SCOPE();
     auto commands = mCurrentRenderPass.commandBuffer;
     auto vbi = resource_ptr<VulkanVertexBufferInfo>::cast(&mResourceManager,
@@ -1833,11 +1792,10 @@ void VulkanDriver::bindPipelineImpl(PipelineState const& pipelineState) {
 
     constexpr uint8_t descriptorSetMaskTable[4] = {0x1, 0x3, 0x7, 0xF};
 
-    mPipelineState = {
+    mBoundPipeline = {
         .program = program,
         .pipelineLayout = pipelineLayout,
         .descriptorSetMask = fvkutils::DescriptorSetMask(descriptorSetMaskTable[layoutCount]),
-        .bindInDraw = {false, {}},
     };
 
     mPipelineCache.bindLayout(pipelineLayout);
@@ -1884,24 +1842,14 @@ void VulkanDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t ins
     FVK_SYSTRACE_SCOPE();
     VkCommandBuffer cmdbuffer = mCurrentRenderPass.commandBuffer->buffer();
 
-    if (mAppState.hasExternalSamplers()) {
-        auto const& [bindInDraw, pipelineSt] = mPipelineState.bindInDraw;
-        bool const hasUpdated =
-                mExternalImageManager.prepareBindSets(mDescriptorSetCache.getBoundSets());
-        if (bindInDraw || hasUpdated) {
-            bindPipelineImpl(pipelineSt);
-        }
-        mPipelineState.bindInDraw.first = false;
-    }
-
     mDescriptorSetCache.commit(mCurrentRenderPass.commandBuffer,
-            mPipelineState.pipelineLayout,
-            mPipelineState.descriptorSetMask);
+            mBoundPipeline.pipelineLayout,
+            mBoundPipeline.descriptorSetMask);
 
     // Finally, make the actual draw call. TODO: support subranges
-    uint32_t const firstIndex = indexOffset;
-    constexpr int32_t vertexOffset = 0;
-    constexpr uint32_t firstInstId = 0;
+    const uint32_t firstIndex = indexOffset;
+    const int32_t vertexOffset = 0;
+    const uint32_t firstInstId = 0;
 
     vkCmdDrawIndexed(cmdbuffer, indexCount, instanceCount, firstIndex, vertexOffset, firstInstId);
 }

filament/backend/src/vulkan/VulkanDriver.h

@@ -30,7 +30,6 @@
 #include "VulkanYcbcrConversionCache.h"
 #include "vulkan/VulkanDescriptorSetCache.h"
 #include "vulkan/VulkanDescriptorSetLayoutCache.h"
-#include "vulkan/VulkanExternalImageManager.h"
 #include "vulkan/VulkanPipelineLayoutCache.h"
 #include "vulkan/memory/ResourceManager.h"
 #include "vulkan/memory/ResourcePointer.h"
@@ -120,7 +119,6 @@ private:
 
 private:
     void collectGarbage();
-    void bindPipelineImpl(PipelineState const& pipelineState);
 
     VulkanPlatform* mPlatform = nullptr;
     fvkmemory::ResourceManager mResourceManager;
@@ -145,29 +143,21 @@ private:
     VulkanDescriptorSetLayoutCache mDescriptorSetLayoutCache;
     VulkanDescriptorSetCache mDescriptorSetCache;
     VulkanQueryManager mQueryManager;
-    VulkanExternalImageManager mExternalImageManager;
 
     // This is necessary for us to write to push constants after binding a pipeline.
     struct {
-        // For push constant
         resource_ptr<VulkanProgram> program;
-        // For push commiting dynamic ubos in draw()
         VkPipelineLayout pipelineLayout;
         fvkutils::DescriptorSetMask descriptorSetMask;
+    } mBoundPipeline = {};
 
-        std::pair<bool, PipelineState> bindInDraw = {false, {}};
-    } mPipelineState = {};
-
+    // We need to store information about a render pass to enable better barriers at the end of a
+    // renderpass.
     struct {
-        // This tracks whether the app has seen external samplers bound to a the descriptor set.
-        // This will force bindPipeline to take a slow path.
-        bool hasExternalSamplerLayouts = false;
-        bool hasBoundExternalImages = false;
-
-        bool hasExternalSamplers() const noexcept {
-            return hasExternalSamplerLayouts && hasBoundExternalImages;
-        }
-    } mAppState;
+        using AttachmentArray =
+                fvkutils::StaticVector<VulkanAttachment, MAX_RENDERTARGET_ATTACHMENT_TEXTURES>;
+        AttachmentArray attachments;
+    } mRenderPassFboInfo = {};
 
     bool const mIsSRGBSwapChainSupported;
     backend::StereoscopicType const mStereoscopicType;

filament/backend/src/vulkan/VulkanExternalImageManager.cpp

@@ -1,280 +0,0 @@
-/*
- * Copyright (C) 2025 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include "VulkanExternalImageManager.h"
-
-#include "VulkanDescriptorSetCache.h"
-#include "VulkanDescriptorSetLayoutCache.h"
-#include "VulkanSamplerCache.h"
-#include "VulkanYcbcrConversionCache.h"
-#include "vulkan/memory/ResourcePointer.h"
-#include "vulkan/utils/Conversion.h"
-
-#include <backend/platforms/VulkanPlatform.h>
-
-#include <algorithm>
-
-namespace filament::backend {
-
-namespace {
-
-template<typename T>
-void erase(std::vector<T>& v, std::function<bool(T const&)> f) {
-    auto newEnd = std::remove_if(v.begin(), v.end(), f);
-    v.erase(newEnd, v.end());
-}
-
-} // anonymous
-
-VulkanExternalImageManager::VulkanExternalImageManager(VulkanPlatform* platform,
-        VulkanSamplerCache* samplerCache, VulkanYcbcrConversionCache* ycbcrConversionCache,
-        VulkanDescriptorSetCache* setCache, VulkanDescriptorSetLayoutCache* layoutCache)
-    : mPlatform(platform),
-      mSamplerCache(samplerCache),
-      mYcbcrConversionCache(ycbcrConversionCache),
-      mDescriptorSetCache(setCache),
-      mDescriptorSetLayoutCache(layoutCache) {
-}
-
-VulkanExternalImageManager::~VulkanExternalImageManager() = default;
-
-void VulkanExternalImageManager::terminate() {
-    mSetAndLayouts.clear();
-    mSetBindings.clear();
-    mImages.clear();
-}
-
-void VulkanExternalImageManager::onBeginFrame() {
-    std::for_each(mImages.begin(), mImages.end(), [](ImageData& image) {
-        image.hasBeenValidated = false;
-    });
-}
-
-bool VulkanExternalImageManager::prepareBindSets(SetArray const& sets) {
-    bool hasUpdated = false;
-    for (auto set: sets) {
-        if (!set) {
-            continue;
-        }
-        if (auto itr = std::find_if(mSetAndLayouts.begin(), mSetAndLayouts.end(),
-                    [&](auto const& setAndLayout) { return setAndLayout.first == set; });
-                itr != mSetAndLayouts.end()) {
-            hasUpdated = updateSetAndLayout(itr->first, itr->second) || hasUpdated;
-        }
-    }
-    return hasUpdated;
-}
-
-bool VulkanExternalImageManager::updateSetAndLayout(
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-        fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout) {
-    auto findImage = [&](fvkmemory::resource_ptr<VulkanTexture> texture) -> ImageData* {
-        auto itr = std::find_if(mImages.begin(), mImages.end(), [&](ImageData const& data) {
-            return data.ptr == texture;
-        });
-        assert_invariant(itr != mImages.end());
-        return &(*itr);
-    };
-
-    //std::vector<std::pair<uint8_t, ImageData*>> externalImages;
-    utils::FixedCapacityVector<std::pair<uint8_t, VkSampler>> samplerAndBindings;
-    samplerAndBindings.reserve(MAX_SAMPLER_COUNT);
-
-    bool hasImageUpdates = false;
-    for (auto& bindingInfo : mSetBindings) {
-        if (bindingInfo.set != set) {
-            continue;
-        }
-        auto imageData = findImage(bindingInfo.image);
-        hasImageUpdates = updateImage(imageData) || hasImageUpdates;
-
-        auto samplerParams = bindingInfo.samplerParams;
-        // according to spec, these must match chromaFilter
-        // https://registry.khronos.org/vulkan/specs/latest/man/html/VkSamplerCreateInfo.html#VUID-VkSamplerCreateInfo-minFilter-01645
-        samplerParams.filterMag = SamplerMagFilter::NEAREST;
-        samplerParams.filterMin = SamplerMinFilter::NEAREST;
-
-        auto sampler = mSamplerCache->getSampler({
-            .sampler = samplerParams,
-            .conversion = imageData->conversion,
-        });
-        samplerAndBindings.push_back({ bindingInfo.binding, sampler });
-    }
-
-    // We need to sort by binding number
-    std::sort(samplerAndBindings.begin(), samplerAndBindings.end());
-
-    utils::FixedCapacityVector<VkSampler> outSamplers;
-    outSamplers.reserve(MAX_SAMPLER_COUNT);
-    std::for_each(samplerAndBindings.begin(), samplerAndBindings.end(),
-            [&](auto const& b) { outSamplers.push_back(b.second); });
-
-    VkDescriptorSetLayout const oldLayout = layout->getVkLayout();
-    VkDescriptorSetLayout const newLayout =
-            mDescriptorSetLayoutCache->getVkLayout(layout->bitmask, outSamplers);
-    bool const hasLayoutUpdate = oldLayout != newLayout;
-    layout->setVkLayout(newLayout);
-
-    assert_invariant(
-            (!hasImageUpdates && !hasLayoutUpdate) ||
-            (hasImageUpdates && hasLayoutUpdate));
-
-    if (!hasLayoutUpdate) {
-        return false;
-    }
-
-    auto foldBitsInHalf = [](auto bitset) {
-        constexpr size_t BITMASK_LOWER_BITS_LEN = sizeof(bitset) * 4;
-        decltype(bitset) outBitset;
-        bitset.forEachSetBit([&](size_t index) { outBitset.set(index % BITMASK_LOWER_BITS_LEN); });
-        return outBitset;
-    };
-    // We need to build a new descriptor set from the new layout
-    VkDescriptorSet oldSet = set->getVkSet();
-    VkDescriptorSet newSet = mDescriptorSetCache->getVkSet(layout);
-
-    using Bitmask = fvkutils::UniformBufferBitmask;
-    static_assert(sizeof(Bitmask) * 8 == fvkutils::MAX_DESCRIPTOR_SET_BITMASK_BITS);
-
-    auto const ubo = layout->bitmask.ubo | layout->bitmask.dynamicUbo;
-    auto const samplers = layout->bitmask.sampler & (~layout->bitmask.externalSampler);
-
-    // each bitmask denotes a binding index, and separated into two stages - vertex and buffer
-    // We fold the two stages into just the lower half of the bits to denote a combined set of
-    // bindings.
-    Bitmask const copyBindings = foldBitsInHalf(ubo | samplers);
-
-    // TODO: fix the size for better memory
-    std::vector<VkCopyDescriptorSet> copies;
-    copyBindings.forEachSetBit([&](size_t index) {
-        copies.push_back({
-            .sType = VK_STRUCTURE_TYPE_COPY_DESCRIPTOR_SET,
-            .srcSet = oldSet,
-            .srcBinding = (uint32_t) index,
-            .dstSet = newSet,
-            .dstBinding = (uint32_t) index,
-            .descriptorCount = 1,
-        });
-    });
-    vkUpdateDescriptorSets(mPlatform->getDevice(), 0, nullptr, copies.size(), copies.data());
-
-    set->setVkSet(newSet);
-
-    // We need to release the vkset, which is no longer used, back into the pool.
-    mDescriptorSetCache->manualRecyle(layout->count, oldLayout, oldSet);
-
-    // We need to update the external samplers in the set
-    for (auto& bindingInfo: mSetBindings) {
-        if (bindingInfo.set != set) {
-            continue;
-        }
-        mDescriptorSetCache->updateSampler(set, bindingInfo.binding, bindingInfo.image,
-                VK_NULL_HANDLE);
-    }
-    return true;
-}
-
-VkSamplerYcbcrConversion VulkanExternalImageManager::getVkSamplerYcbcrConversion(
-        VulkanPlatform::ExternalImageMetadata const& metadata) {
-    // This external image does not require external sampler (YUV conversion).
-    if (metadata.externalFormat == 0) {
-        return VK_NULL_HANDLE;
-    }
-    VulkanYcbcrConversionCache::Params ycbcrParams = {
-        .conversion = {
-            .ycbcrModel = fvkutils::getYcbcrModelConversionFilament(metadata.ycbcrModel),
-            .r = fvkutils::getSwizzleFilament(metadata.ycbcrConversionComponents.r, 0),
-            .g = fvkutils::getSwizzleFilament(metadata.ycbcrConversionComponents.g, 1),
-            .b = fvkutils::getSwizzleFilament(metadata.ycbcrConversionComponents.b, 2),
-            .a = fvkutils::getSwizzleFilament(metadata.ycbcrConversionComponents.a, 3),
-            .ycbcrRange = fvkutils::getYcbcrRangeFilament(metadata.ycbcrRange),
-            .xChromaOffset = fvkutils::getChromaLocationFilament(metadata.xChromaOffset),
-            .yChromaOffset = fvkutils::getChromaLocationFilament(metadata.yChromaOffset),
-
-            // Unclear where to get the chromaFilter, we just assume it's nearest.
-            .chromaFilter = SamplerMagFilter::NEAREST,
-        },
-        .format = metadata.filamentFormat,
-        .externalFormat = metadata.externalFormat,
-    };
-    return mYcbcrConversionCache->getConversion(ycbcrParams);
-}
-
-bool VulkanExternalImageManager::updateImage(ImageData* image) {
-    if (image->hasBeenValidated) {
-        return false;
-    }
-    image->hasBeenValidated = true;
-
-    auto metadata = mPlatform->extractExternalImageMetadata(image->platformHandle);
-    auto vkYcbcr = getVkSamplerYcbcrConversion(metadata);
-    if (vkYcbcr == image->conversion) {
-        return false;
-    }
-
-    image->ptr->setYcbcrConversion(vkYcbcr, metadata.externalFormat != 0);
-    image->conversion = vkYcbcr;
-    return true;
-}
-
-void VulkanExternalImageManager::addDescriptorSet(
-        fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout,
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set) {
-    mSetAndLayouts.push_back({set, layout});
-}
-
-void VulkanExternalImageManager::removeDescriptorSet(
-        fvkmemory::resource_ptr<VulkanDescriptorSet> inSet) {
-    erase<SetAndLayout>(mSetAndLayouts,
-            [&](auto const& setLayout) { return (setLayout.first == inSet); });
-    erase<SetBindingInfo>(mSetBindings,
-            [&](auto const& bindingInfo) { return (bindingInfo.set == inSet); });
-}
-
-void VulkanExternalImageManager::removeDescriptorSetLayout(
-        fvkmemory::resource_ptr<VulkanDescriptorSetLayout> inLayout) {
-    erase<SetAndLayout>(mSetAndLayouts,
-            [&](auto const& setLayout) { return (setLayout.second == inLayout); });
-}
-
-void VulkanExternalImageManager::bindExternallySampledTexture(
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set, uint8_t bindingPoint,
-        fvkmemory::resource_ptr<VulkanTexture> image, SamplerParams samplerParams) {
-    // Should we do duplicate validation here?
-    mSetBindings.push_back({ bindingPoint, image, set, samplerParams });
-}
-
-void VulkanExternalImageManager::addExternallySampledTexture(
-        fvkmemory::resource_ptr<VulkanTexture> image,
-        Platform::ExternalImageHandleRef platformHandleRef) {
-    mImages.push_back({ image, platformHandleRef, false });
-}
-
-void VulkanExternalImageManager::removeExternallySampledTexture(
-        fvkmemory::resource_ptr<VulkanTexture> image) {
-    erase<SetBindingInfo>(mSetBindings,
-            [&](auto const& bindingInfo) { return (bindingInfo.image == image); });
-    erase<ImageData>(mImages, [&](auto const& imageData) { return imageData.ptr == image; });
-}
-
-bool VulkanExternalImageManager::isExternallySampledTexture(
-        fvkmemory::resource_ptr<VulkanTexture> image) const {
-    return std::find_if(mImages.begin(), mImages.end(),
-                   [&](auto const& imageData) { return imageData.ptr == image; }) != mImages.end();
-}
-
-
-} // namesapce filament::backend

filament/backend/src/vulkan/VulkanExternalImageManager.h

@@ -1,115 +0,0 @@
-/*
- * Copyright (C) 2025 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef TNT_FILAMENT_BACKEND_CACHING_VULKANEXTERNALIMAGEMANAGER_H
-#define TNT_FILAMENT_BACKEND_CACHING_VULKANEXTERNALIMAGEMANAGER_H
-
-#include "VulkanHandles.h"
-
-#include <backend/DriverEnums.h>
-
-#include <array>
-#include <vector>
-
-namespace filament::backend {
-
-class VulkanYcbcrConversionCache;
-class VulkanSamplerCache;
-class VulkanDescriptorSetLayoutCache;
-class VulkanDescriptorSetCache;
-
-// Manages the logic of external images and their quirks wrt Vulikan.
-class VulkanExternalImageManager {
-public:
-
-    VulkanExternalImageManager(
-            VulkanPlatform* platform,
-            VulkanSamplerCache* samplerCache,
-            VulkanYcbcrConversionCache* ycbcrConversionCache,
-            VulkanDescriptorSetCache* setCache,
-            VulkanDescriptorSetLayoutCache* layoutCache);
-
-    ~VulkanExternalImageManager();
-
-    void terminate();
-
-    void onBeginFrame();
-
-    using SetArray = std::array<fvkmemory::resource_ptr<VulkanDescriptorSet>,
-            VulkanDescriptorSetLayout::UNIQUE_DESCRIPTOR_SET_COUNT>;
-
-    // This sets the currently bound layouts objects for the pipeline
-    bool prepareBindSets(SetArray const& layouts);
-
-    void addDescriptorSet(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout,
-            fvkmemory::resource_ptr<VulkanDescriptorSet> set);
-
-    void removeDescriptorSetLayout(fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
-
-    void removeDescriptorSet(fvkmemory::resource_ptr<VulkanDescriptorSet> set);
-
-    void bindExternallySampledTexture(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-            uint8_t bindingPoint, fvkmemory::resource_ptr<VulkanTexture> image,
-            SamplerParams samplerParams);
-
-    void addExternallySampledTexture(fvkmemory::resource_ptr<VulkanTexture> image,
-            Platform::ExternalImageHandleRef platformHandleRef);
-
-    void removeExternallySampledTexture(fvkmemory::resource_ptr<VulkanTexture> image);
-
-    bool isExternallySampledTexture(fvkmemory::resource_ptr<VulkanTexture> image) const;
-
-    VkSamplerYcbcrConversion getVkSamplerYcbcrConversion(
-            VulkanPlatform::ExternalImageMetadata const& metadata);
-
-private:
-
-    struct ImageData {
-        fvkmemory::resource_ptr<VulkanTexture> ptr;
-        Platform::ExternalImageHandle platformHandle;
-        bool hasBeenValidated = false; // indicates whether the image has been validated *this frame*
-        VkSamplerYcbcrConversion conversion = VK_NULL_HANDLE;
-    };
-
-    bool updateSetAndLayout(fvkmemory::resource_ptr<VulkanDescriptorSet> set,
-            fvkmemory::resource_ptr<VulkanDescriptorSetLayout> layout);
-    bool updateImage(ImageData* imageData);
-
-    VulkanPlatform* mPlatform;
-    VulkanSamplerCache* mSamplerCache;
-    VulkanYcbcrConversionCache* mYcbcrConversionCache;
-    VulkanDescriptorSetCache* mDescriptorSetCache;
-    VulkanDescriptorSetLayoutCache* mDescriptorSetLayoutCache;
-
-    using SetAndLayout = std::pair<fvkmemory::resource_ptr<VulkanDescriptorSet>,
-            fvkmemory::resource_ptr<VulkanDescriptorSetLayout>>;
-
-    struct SetBindingInfo {
-        uint8_t binding = 0;
-        fvkmemory::resource_ptr<VulkanTexture> image;
-        fvkmemory::resource_ptr<VulkanDescriptorSet> set;
-        SamplerParams samplerParams;
-    };
-
-    // Use vectors instead of hash maps because we only expect small number of entries.
-    std::vector<SetAndLayout> mSetAndLayouts;
-    std::vector<SetBindingInfo> mSetBindings;
-    std::vector<ImageData> mImages;
-};
-
-} // filament::backend
-
-#endif // TNT_FILAMENT_BACKEND_CACHING_VULKANEXTERNALIMAGEMANAGER_H

filament/backend/src/vulkan/VulkanHandles.cpp

@@ -30,7 +30,6 @@
 
 #include <backend/platforms/VulkanPlatform.h>
 
-#include <utils/compiler.h> // UTILS_FALLTHROUGH
 #include <utils/Panic.h>    // ASSERT_POSTCONDITION
 
 using namespace bluevk;
@@ -90,9 +89,8 @@ BitmaskGroup fromBackendLayout(DescriptorSetLayout const& layout) {
                 }
                 break;
             }
+            // TODO: properly handle external sampler
             case DescriptorType::SAMPLER_EXTERNAL:
-                fromStageFlags(binding.stageFlags, binding.binding, mask.externalSampler);
-                UTILS_FALLTHROUGH;
             case DescriptorType::SAMPLER: {
                 fromStageFlags(binding.stageFlags, binding.binding, mask.sampler);
                 break;

filament/backend/src/vulkan/VulkanHandles.h

@@ -66,21 +66,18 @@ struct VulkanDescriptorSetLayout : public HwDescriptorSetLayout, fvkmemory::Reso
 
     // The bitmask representation of a set layout.
     struct Bitmask {
+        // TODO: better utiltize the space below and use bitset instead.
         fvkutils::UniformBufferBitmask ubo;         // 8 bytes
         fvkutils::UniformBufferBitmask dynamicUbo;  // 8 bytes
         fvkutils::SamplerBitmask sampler;           // 8 bytes
         fvkutils::InputAttachmentBitmask inputAttachment; // 8 bytes
 
-        // This is a subset of the bitmask.sampler field.
-        fvkutils::SamplerBitmask externalSampler; // 8 bytes
-
         bool operator==(Bitmask const& right) const {
             return ubo == right.ubo && dynamicUbo == right.dynamicUbo && sampler == right.sampler &&
-                   inputAttachment == right.inputAttachment &&
-                   externalSampler == right.externalSampler;
+                   inputAttachment == right.inputAttachment;
         }
     };
-    static_assert(sizeof(Bitmask) == 40);
+    static_assert(sizeof(Bitmask) == 32);
 
     // This is a convenience struct to quickly check layout compatibility in terms of descriptor set
     // pools.
@@ -122,16 +119,10 @@ struct VulkanDescriptorSetLayout : public HwDescriptorSetLayout, fvkmemory::Reso
 
     VulkanDescriptorSetLayout(DescriptorSetLayout const& layout);
 
-    // Note that we don't destroy the vklayout. This is done by the layout cache.
     ~VulkanDescriptorSetLayout() = default;
 
-    VkDescriptorSetLayout const& getVkLayout() const noexcept { return mVkLayout; }
-
-    // It is possible to have the layout switch out due to AHardwarebuffer (external image) format
-    // changes.
-    void setVkLayout(VkDescriptorSetLayout vklayout) noexcept { mVkLayout = vklayout; }
-
-    bool hasExternalSamplers() const noexcept { return bitmask.externalSampler.count() > 0; }
+    VkDescriptorSetLayout getVkLayout() const { return mVkLayout; }
+    void setVkLayout(VkDescriptorSetLayout vklayout) { mVkLayout = vklayout; }
 
     Bitmask const bitmask;
     Count const count;
@@ -146,11 +137,12 @@ public:
     // can use to repackage the vk handle.
     using OnRecycle = std::function<void(VulkanDescriptorSet*)>;
 
-    VulkanDescriptorSet(
+    VulkanDescriptorSet(VkDescriptorSet rawSet,
             fvkutils::UniformBufferBitmask const& dynamicUboMask,
             uint8_t uniqueDynamicUboCount,
             OnRecycle&& onRecycleFn)
-        : dynamicUboMask(dynamicUboMask),
+        : vkSet(rawSet),
+          dynamicUboMask(dynamicUboMask),
           uniqueDynamicUboCount(uniqueDynamicUboCount),
           mOnRecycleFn(std::move(onRecycleFn)) {}
 
@@ -160,13 +152,6 @@ public:
         }
     }
 
-    VkDescriptorSet const& getVkSet() const noexcept {
-        return mVkSet;
-    }
-
-    // Note that the only case where you'd set it more than once is with external images/samplers.
-    void setVkSet(VkDescriptorSet vkset) noexcept { mVkSet = vkset; }
-
     void setOffsets(backend::DescriptorSetOffsetArray&& offsets) noexcept {
         mOffsets = std::move(offsets);
     }
@@ -178,11 +163,11 @@ public:
     void acquire(fvkmemory::resource_ptr<VulkanTexture> texture);
     void acquire(fvkmemory::resource_ptr<VulkanBufferObject> buffer);
 
+    VkDescriptorSet const vkSet;
     fvkutils::UniformBufferBitmask const dynamicUboMask;
     uint8_t const uniqueDynamicUboCount;
 
 private:
-    VkDescriptorSet mVkSet = VK_NULL_HANDLE;
     backend::DescriptorSetOffsetArray mOffsets;
     std::vector<fvkmemory::resource_ptr<fvkmemory::Resource>> mResources;
     OnRecycle mOnRecycleFn;

filament/backend/src/vulkan/VulkanSamplerCache.cpp

@@ -18,7 +18,6 @@
 
 #include "VulkanConstants.h"
 #include "vulkan/utils/Conversion.h"
-#include "vulkan/vulkan_core.h"
 
 #include <utils/Panic.h>
 
@@ -34,15 +33,8 @@ VkSampler VulkanSamplerCache::getSampler(Params params) noexcept {
     if (UTILS_LIKELY(iter != mCache.end())) {
         return iter->second;
     }
-    VkSamplerYcbcrConversionInfo ycbcrConversion = {
-        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO,
-        .conversion = params.conversion,
-    };
-
     auto const& samplerParams = params.sampler;
-    VkSamplerCreateInfo samplerInfo{
-        .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
-        .pNext = params.conversion != VK_NULL_HANDLE ? &ycbcrConversion : VK_NULL_HANDLE,
+    VkSamplerCreateInfo samplerInfo{ .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
         .magFilter = fvkutils::getFilter(samplerParams.filterMag),
         .minFilter = fvkutils::getFilter(samplerParams.filterMin),
         .mipmapMode = fvkutils::getMipmapMode(samplerParams.filterMin),
@@ -56,8 +48,7 @@ VkSampler VulkanSamplerCache::getSampler(Params params) noexcept {
         .minLod = 0.0f,
         .maxLod = fvkutils::getMaxLod(samplerParams.filterMin),
         .borderColor = VK_BORDER_COLOR_INT_OPAQUE_BLACK,
-        .unnormalizedCoordinates = VK_FALSE,
-    };
+        .unnormalizedCoordinates = VK_FALSE };
     VkSampler sampler;
     VkResult result = vkCreateSampler(mDevice, &samplerInfo, VKALLOC, &sampler);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS) << "Unable to create sampler."

filament/backend/src/vulkan/VulkanSwapChain.cpp

@@ -108,6 +108,11 @@ void VulkanSwapChain::present() {
 
     mCommands->flush();
 
+    // call the image ready wait function
+    if (mExplicitImageReadyWait != nullptr) {
+        mExplicitImageReadyWait(swapChain);
+    }
+
     // We only present if it is not headless. No-op for headless.
     if (!mHeadless) {
         VkSemaphore const finishedDrawing = mCommands->acquireFinishedSignal();
@@ -141,6 +146,7 @@ void VulkanSwapChain::acquire(bool& resized) {
     VulkanPlatform::ImageSyncData imageSyncData;
     VkResult const result = mPlatform->acquire(swapChain, &imageSyncData);
     mCurrentSwapIndex = imageSyncData.imageIndex;
+    mExplicitImageReadyWait = imageSyncData.explicitImageReadyWait;
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS || result == VK_SUBOPTIMAL_KHR)
             << "Cannot acquire in swapchain. error=" << static_cast<int32_t>(result);
     if (imageSyncData.imageReadySemaphore != VK_NULL_HANDLE) {

filament/backend/src/vulkan/VulkanSwapChain.h

@@ -49,9 +49,7 @@ struct VulkanSwapChain : public HwSwapChain, fvkmemory::Resource {
 
     void present();
 
-    // Acquire a new image from the swapchain. If the image is not available it would wait until it
-    // is.
-    void acquire(bool& resized);
+    void acquire(bool& reized);
 
     fvkmemory::resource_ptr<VulkanTexture> getCurrentColor() const noexcept {
         uint32_t const imageIndex = mCurrentSwapIndex;
@@ -101,6 +99,7 @@ private:
     VkExtent2D mExtent;
     uint32_t mLayerCount;
     uint32_t mCurrentSwapIndex;
+    std::function<void(Platform::SwapChain* handle)> mExplicitImageReadyWait = nullptr;
     bool mAcquired;
     bool mIsFirstRenderPass;
 };

filament/backend/src/vulkan/VulkanYcbcrConversionCache.cpp

@@ -43,7 +43,7 @@ VkSamplerYcbcrConversion VulkanYcbcrConversionCache::getConversion(
     TextureSwizzle const swizzleArray[] = { chroma.r, chroma.g, chroma.b, chroma.a };
     VkSamplerYcbcrConversionCreateInfo conversionInfo = {
         .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_CREATE_INFO,
-        .format = fvkutils::getVkFormat(params.format),
+        .format = VK_FORMAT_UNDEFINED,
         .ycbcrModel = fvkutils::getYcbcrModelConversion(chroma.ycbcrModel),
         .ycbcrRange = fvkutils::getYcbcrRange(chroma.ycbcrRange),
         .components = fvkutils::getSwizzleMap(swizzleArray),
@@ -52,7 +52,6 @@ VkSamplerYcbcrConversion VulkanYcbcrConversionCache::getConversion(
         .chromaFilter = fvkutils::getFilter(chroma.chromaFilter),
     };
 
-    // We could put this in the platform class, but that seems like a bit of an overkill
 #if defined(__ANDROID__)
     VkExternalFormatANDROID externalFormat = {
         .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
@@ -60,7 +59,6 @@ VkSamplerYcbcrConversion VulkanYcbcrConversionCache::getConversion(
     };
     if (params.externalFormat) {
         conversionInfo.pNext = &externalFormat;
-        conversionInfo.format = VK_FORMAT_UNDEFINED;
     }
 #endif
 

filament/backend/src/vulkan/VulkanYcbcrConversionCache.h

@@ -31,8 +31,7 @@ class VulkanYcbcrConversionCache {
 public:
     struct Params {
         SamplerYcbcrConversion conversion = {};
-        TextureFormat format = {};
-        uint16_t padding = 0;
+        uint32_t padding = 0;
         uint64_t externalFormat = 0;
     };
     static_assert(sizeof(Params) == 16);

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

@@ -132,10 +132,6 @@ public:
         return id() == other.id() && type() == other.type();
     }
 
-    inline bool operator!=(resource_ptr<D> const& other) const {
-        return !((*this) == other);
-    }
-
     inline explicit operator bool() const {
         return bool(mRef);
     }

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

@@ -211,7 +211,6 @@ ExtensionSet getDeviceExtensions(VkPhysicalDevice device) {
         VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME,
         VK_KHR_SAMPLER_YCBCR_CONVERSION_EXTENSION_NAME,
         VK_ANDROID_EXTERNAL_MEMORY_ANDROID_HARDWARE_BUFFER_EXTENSION_NAME,
-        VK_EXT_QUEUE_FAMILY_FOREIGN_EXTENSION_NAME,
 #endif
         // MoltenVk is the only non-conformant implementation we're interested in.
 #if defined(__APPLE__)
@@ -327,9 +326,7 @@ VkInstance createInstance(ExtensionSet const& requiredExts) {
 }
 
 VkDevice createLogicalDevice(VkPhysicalDevice physicalDevice,
-        VkPhysicalDeviceFeatures2 const& features,
-        VkPhysicalDeviceVulkan11Features const& vk11Features,
-        uint32_t graphicsQueueFamilyIndex,
+        VkPhysicalDeviceFeatures2 const& features, uint32_t graphicsQueueFamilyIndex,
         uint32_t protectedGraphicsQueueFamilyIndex, ExtensionSet const& deviceExtensions,
         bool requestImageView2DOn3DImage) {
     VkDevice device;
@@ -362,28 +359,14 @@ VkDevice createLogicalDevice(VkPhysicalDevice physicalDevice,
 
     // We could simply enable all supported features, but since that may have performance
     // consequences let's just enable the features we need.
-    VkPhysicalDeviceFeatures enabledFeatures = {
+    VkPhysicalDeviceFeatures enabledFeatures{
         .depthClamp = features.features.depthClamp,
         .samplerAnisotropy = features.features.samplerAnisotropy,
         .textureCompressionETC2 = features.features.textureCompressionETC2,
         .textureCompressionBC = features.features.textureCompressionBC,
         .shaderClipDistance = features.features.shaderClipDistance,
     };
-
-    VkPhysicalDeviceFeatures2 enabledFeatures2 = {
-        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2,
-        .features = enabledFeatures,
-    };
-    chainStruct(&deviceCreateInfo, &enabledFeatures2);
-
-    VkPhysicalDeviceVulkan11Features enabledVk11Features = {
-        .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES,
-        .multiview = vk11Features.multiview,
-#if defined(__ANDROID__)
-        .samplerYcbcrConversion = vk11Features.samplerYcbcrConversion,
-#endif
-    };
-    chainStruct(&deviceCreateInfo, &enabledVk11Features);
+    deviceCreateInfo.pEnabledFeatures = &enabledFeatures;
 
     deviceCreateInfo.enabledExtensionCount = (uint32_t) requestExtensions.size();
     deviceCreateInfo.ppEnabledExtensionNames = requestExtensions.data();
@@ -400,7 +383,7 @@ VkDevice createLogicalDevice(VkPhysicalDevice physicalDevice,
 
     VkPhysicalDeviceMultiviewFeaturesKHR multiview = {
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES_KHR,
-        .multiview = vk11Features.multiview,
+        .multiview = VK_TRUE,
         .multiviewGeometryShader = VK_FALSE,
         .multiviewTessellationShader = VK_FALSE,
     };
@@ -751,7 +734,6 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
         .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_PROPERTIES,
     };
     chainStruct(&context.mPhysicalDeviceFeatures, &queryProtectedMemoryFeatures);
-    chainStruct(&context.mPhysicalDeviceFeatures, &context.mPhysicalDeviceVk11Features);
     chainStruct(&context.mPhysicalDeviceProperties, &protectedMemoryProperties);
 
     // Initialize the following fields: physicalDeviceProperties, memoryProperties,
@@ -813,10 +795,10 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
     }
 
     if (mImpl->mDevice == VK_NULL_HANDLE) {
-        mImpl->mDevice = createLogicalDevice(mImpl->mPhysicalDevice,
-                context.mPhysicalDeviceFeatures, context.mPhysicalDeviceVk11Features,
-                mImpl->mGraphicsQueueFamilyIndex, mImpl->mProtectedGraphicsQueueFamilyIndex,
-                deviceExts, requestPortabilitySubsetImageView2DOn3DImage);
+        mImpl->mDevice =
+                createLogicalDevice(mImpl->mPhysicalDevice, context.mPhysicalDeviceFeatures,
+                        mImpl->mGraphicsQueueFamilyIndex, mImpl->mProtectedGraphicsQueueFamilyIndex,
+                        deviceExts, requestPortabilitySubsetImageView2DOn3DImage);
     }
 
     assert_invariant(mImpl->mDevice != VK_NULL_HANDLE);
@@ -844,10 +826,12 @@ Driver* VulkanPlatform::createDriver(void* sharedContext,
     if (!mImpl->mSharedContext) {
         context.mDebugUtilsSupported = setContains(instExts, VK_EXT_DEBUG_UTILS_EXTENSION_NAME);
         context.mDebugMarkersSupported = setContains(deviceExts, VK_EXT_DEBUG_MARKER_EXTENSION_NAME);
+        context.mMultiviewEnabled = setContains(deviceExts, VK_KHR_MULTIVIEW_EXTENSION_NAME);
     } else {
         VulkanSharedContext const* scontext = (VulkanSharedContext const*) sharedContext;
         context.mDebugUtilsSupported = scontext->debugUtilsSupported;
         context.mDebugMarkersSupported = scontext->debugMarkersSupported;
+        context.mMultiviewEnabled = scontext->multiviewSupported;
     }
 
     // Check the availability of lazily allocated memory
@@ -993,6 +977,30 @@ VkQueue VulkanPlatform::getProtectedGraphicsQueue() const noexcept {
     return mImpl->mProtectedGraphicsQueue;
 }
 
+VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadata(
+        ExternalImageHandleRef externalImage) {
+    return getExternalImageMetadataImpl(externalImage, mImpl->mDevice);
+}
+
+VulkanPlatform::ImageData VulkanPlatform::createExternalImageData(
+        ExternalImageHandleRef externalImage, const ExternalImageMetadata& metadata,
+        uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
+    return createExternalImageDataImpl(externalImage, mImpl->mDevice, metadata, memoryTypeIndex,
+            usage);
+}
+
+VkSampler VulkanPlatform::createExternalSampler(SamplerYcbcrConversion chroma,
+        SamplerParams sampler, uint32_t internalFormat) {
+    return createExternalSamplerImpl(mImpl->mDevice, chroma, sampler, internalFormat);
+}
+
+VkImageView VulkanPlatform::createExternalImageView(SamplerYcbcrConversion chroma,
+        uint32_t internalFormat, VkImage image, VkImageSubresourceRange range,
+        VkImageViewType viewType, VkComponentMapping swizzle) {
+    return createExternalImageViewImpl(mImpl->mDevice, chroma, internalFormat, image, range,
+            viewType, swizzle);
+}
+
 ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() const {
     return getSwapchainInstanceExtensionsImpl();
 }

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

@@ -15,14 +15,14 @@
  */
 #include <backend/platforms/VulkanPlatformAndroid.h>
 
-#include "vulkan/VulkanConstants.h"
-#include "vulkan/VulkanContext.h"
-#include "vulkan/vulkan_core.h"
-
 #include <backend/DriverEnums.h>
 #include <private/backend/BackendUtilsAndroid.h>
 
+#include "vulkan/VulkanConstants.h"
+
 #include <utils/Panic.h>
+#include "vulkan/utils/Image.h"
+#include "vulkan/utils/Conversion.h"
 
 #include <bluevk/BlueVK.h>
 
@@ -57,7 +57,7 @@ VkFormat transformVkFormat(VkFormat format, bool sRGB) {
 }
 
 bool isProtectedFromUsage(uint64_t usage) {
-    return usage & AHARDWAREBUFFER_USAGE_PROTECTED_CONTENT;
+    return (usage & AHARDWAREBUFFER_USAGE_PROTECTED_CONTENT) ? true : false;
 }
 
 std::pair<VkFormat, VkImageUsageFlags> getVKFormatAndUsage(const AHardwareBuffer_Desc& desc,
@@ -120,9 +120,7 @@ std::pair<VkFormat, VkImageUsageFlags> getVKFormatAndUsage(const AHardwareBuffer
     usage = 0;
     if (desc.usage & AHARDWAREBUFFER_USAGE_GPU_SAMPLED_IMAGE) {
         usage |= VK_IMAGE_USAGE_SAMPLED_BIT;
-
-        // We shouldn't be using external samplers as input attachments
-        // usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT;
+        usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT;
     }
     if (desc.usage & AHARDWAREBUFFER_USAGE_GPU_FRAMEBUFFER) {
         if (isDepthFormat) {
@@ -138,38 +136,87 @@ std::pair<VkFormat, VkImageUsageFlags> getVKFormatAndUsage(const AHardwareBuffer
     return { format, usage };
 }
 
-std::pair<TextureFormat, TextureUsage> getFilamentFormatAndUsage(const AHardwareBuffer_Desc& desc,
-        bool sRGB) {
-    auto const format = mapToFilamentFormat(desc.format, sRGB);
-    return {
-        format,
-        mapToFilamentUsage(desc.usage, format),
-    };
-}
+VulkanPlatform::ImageData allocateExternalImage(AHardwareBuffer* buffer, VkDevice device,
+        VulkanPlatform::ExternalImageMetadata const& metadata, uint32_t memoryTypeIndex,
+        VkImageUsageFlags usage) {
+    VulkanPlatform::ImageData data;
 
+    // if external format we need to specifiy it in the allocation
+    const bool useExternalFormat = metadata.format == VK_FORMAT_UNDEFINED;
+
+    const VkExternalFormatANDROID externalFormat = {
+        .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
+        .pNext = nullptr,
+        // pass down the format (external means we don't have it VK defined)
+        .externalFormat = metadata.externalFormat,
+    };
+    const VkExternalMemoryImageCreateInfo externalCreateInfo = {
+        .sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO,
+        .pNext = useExternalFormat ? &externalFormat : nullptr,
+        .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID,
+    };
+
+    VkImageCreateInfo imageInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO };
+    imageInfo.pNext = &externalCreateInfo;
+    imageInfo.format = metadata.format;
+    imageInfo.imageType = VK_IMAGE_TYPE_2D;
+    imageInfo.extent = {
+        metadata.width,
+        metadata.height,
+        1u,
+    };
+    imageInfo.mipLevels = 1;
+    imageInfo.arrayLayers = metadata.layers;
+    imageInfo.samples = metadata.samples;
+    imageInfo.usage = usage;
+
+    VkResult result = vkCreateImage(device, &imageInfo, VKALLOC, &data.first);
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+            << "vkCreateImage failed with error=" << static_cast<int32_t>(result);
+
+    // Allocate the memory
+    VkImportAndroidHardwareBufferInfoANDROID androidHardwareBufferInfo = {
+        .sType = VK_STRUCTURE_TYPE_IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID,
+        .pNext = nullptr,
+        .buffer = buffer,
+    };
+    VkMemoryDedicatedAllocateInfo memoryDedicatedAllocateInfo = {
+        .sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO,
+        .pNext = &androidHardwareBufferInfo,
+        .image = data.first,
+        .buffer = VK_NULL_HANDLE,
+    };
+    VkMemoryAllocateInfo allocInfo = {
+        .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
+        .pNext = &memoryDedicatedAllocateInfo,
+        .allocationSize = metadata.allocationSize,
+        .memoryTypeIndex = memoryTypeIndex,
+    };
+    result = vkAllocateMemory(device, &allocInfo, VKALLOC, &data.second);
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+            << "vkAllocateMemory failed with error=" << static_cast<int32_t>(result);
+
+    return data;
+}
 
 }// namespace
 
-VulkanPlatformAndroid::ExternalImageVulkanAndroid::~ExternalImageVulkanAndroid() {
-    if (__builtin_available(android 26, *)) {
-        if (aHardwareBuffer) {
-            AHardwareBuffer_release(aHardwareBuffer);
-        }
-    }
-}
+VulkanPlatformAndroid::ExternalImageVulkanAndroid::~ExternalImageVulkanAndroid() = default;
 
 Platform::ExternalImageHandle VulkanPlatformAndroid::createExternalImage(
         AHardwareBuffer const* buffer, bool sRGB) noexcept {
     if (__builtin_available(android 26, *)) {
-        auto bufferImpl = const_cast<AHardwareBuffer*>(buffer);
-        AHardwareBuffer_acquire(bufferImpl);
-
         AHardwareBuffer_Desc hardwareBufferDescription = {};
         AHardwareBuffer_describe(buffer, &hardwareBufferDescription);
 
         auto* const p = new (std::nothrow) ExternalImageVulkanAndroid;
         p->aHardwareBuffer = const_cast<AHardwareBuffer*>(buffer);
         p->sRGB = sRGB;
+        p->height = hardwareBufferDescription.height;
+        p->width = hardwareBufferDescription.width;
+        TextureFormat textureFormat = mapToFilamentFormat(hardwareBufferDescription.format, sRGB);
+        p->format = textureFormat;
+        p->usage = mapToFilamentUsage(hardwareBufferDescription.usage, textureFormat);
         return Platform::ExternalImageHandle{ p };
     }
 
@@ -178,20 +225,23 @@ Platform::ExternalImageHandle VulkanPlatformAndroid::createExternalImage(
 
 VulkanPlatformAndroid::ExternalImageDescAndroid VulkanPlatformAndroid::getExternalImageDesc(
         ExternalImageHandleRef externalImage) const noexcept {
-    auto metadata = extractExternalImageMetadata(externalImage);
+    auto const* fvkExternalImage =
+            static_cast<ExternalImageVulkanAndroid const*>(externalImage.get());
+
     return {
-        .width = metadata.width,
-        .height = metadata.height,
-        .format = metadata.filamentFormat,
-        .usage = metadata.filamentUsage,
+        .width = fvkExternalImage->width,
+        .height = fvkExternalImage->height,
+        .format = fvkExternalImage->format,
+        .usage = fvkExternalImage->usage,
     };
 }
 
-VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::extractExternalImageMetadata(
-        ExternalImageHandleRef image) const {
-    auto const* fvkExternalImage = static_cast<ExternalImageVulkanAndroid const*>(image.get());
+VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::getExternalImageMetadata(
+        ExternalImageHandleRef externalImage) {
+    auto const* fvkExternalImage =
+            static_cast<ExternalImageVulkanAndroid const*>(externalImage.get());
 
-    ExternalImageMetadata metadata = {};
+    ExternalImageMetadata metadata;
     AHardwareBuffer* buffer = fvkExternalImage->aHardwareBuffer;
     if (__builtin_available(android 26, *)) {
         AHardwareBuffer_Desc bufferDesc;
@@ -199,30 +249,16 @@ VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::extractExternalImag
         metadata.width = bufferDesc.width;
         metadata.height = bufferDesc.height;
         metadata.layers = bufferDesc.layers;
+        metadata.isProtected = isProtectedFromUsage(bufferDesc.usage);
         std::tie(metadata.format, metadata.usage) =
                 getVKFormatAndUsage(bufferDesc, fvkExternalImage->sRGB);
-        std::tie(metadata.filamentFormat, metadata.filamentUsage) =
-                getFilamentFormatAndUsage(bufferDesc, fvkExternalImage->sRGB);
-
-        if (isProtectedFromUsage(bufferDesc.usage)) {
-            metadata.filamentUsage |= TextureUsage::PROTECTED;
-        }
-
-        // TODO: The following seems unnecessary. we should be able to discern directly from the
-        // bufferDesc.
-        if (any(metadata.filamentUsage & TextureUsage::BLIT_SRC)) {
-            metadata.usage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
-        }
-
-        if (any(metadata.filamentUsage & (TextureUsage::BLIT_DST | TextureUsage::UPLOADABLE))) {
-            metadata.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT;
-        }
     }
-    metadata.samples = VK_SAMPLE_COUNT_1_BIT;
 
+    metadata.samples = VK_SAMPLE_COUNT_1_BIT;
 
     VkAndroidHardwareBufferFormatPropertiesANDROID formatInfo = {
         .sType = VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_FORMAT_PROPERTIES_ANDROID,
+        .pNext = nullptr,
     };
     VkAndroidHardwareBufferPropertiesANDROID properties = {
         .sType = VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_PROPERTIES_ANDROID,
@@ -232,116 +268,138 @@ VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::extractExternalImag
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
             << "vkGetAndroidHardwareBufferProperties failed with error="
             << static_cast<int32_t>(result);
+
     VkFormat bufferPropertiesFormat = transformVkFormat(formatInfo.format, fvkExternalImage->sRGB);
     FILAMENT_CHECK_POSTCONDITION(metadata.format == bufferPropertiesFormat)
             << "mismatched image format( " << metadata.format << ") and queried format("
             << bufferPropertiesFormat << ") for external image (AHB)";
     metadata.externalFormat = formatInfo.externalFormat;
-
-    // Choose either externalFormat > 0 or metadata.format and prefer the latter.
-    if (metadata.externalFormat > 0 &&  metadata.format != VK_FORMAT_UNDEFINED) {
-        // See VUID-VkImageCreateInfo-pNext-09457
-        metadata.usage = VK_IMAGE_USAGE_SAMPLED_BIT;
-        metadata.externalFormat = 0;
-    }
-
     metadata.allocationSize = properties.allocationSize;
     metadata.memoryTypeBits = properties.memoryTypeBits;
-
-    metadata.ycbcrConversionComponents = formatInfo.samplerYcbcrConversionComponents;
-    metadata.ycbcrModel = formatInfo.suggestedYcbcrModel;
-    metadata.ycbcrRange = formatInfo.suggestedYcbcrRange;
-    metadata.xChromaOffset = formatInfo.suggestedXChromaOffset;
-    metadata.yChromaOffset = formatInfo.suggestedYChromaOffset;
-
     return metadata;
 }
 
-VulkanPlatformAndroid::ImageData VulkanPlatformAndroid::createVkImageFromExternal(
-        ExternalImageHandleRef externalImage) const {
-    auto const& metadata = extractExternalImageMetadata(externalImage);
-
+VulkanPlatformAndroid::ImageData VulkanPlatformAndroid::createExternalImageData(
+        ExternalImageHandleRef externalImage, const ExternalImageMetadata& metadata,
+        uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
     auto const* fvkExternalImage =
             static_cast<ExternalImageVulkanAndroid const*>(externalImage.get());
-    AHardwareBuffer* buffer = fvkExternalImage->aHardwareBuffer;
+    ImageData data = allocateExternalImage(fvkExternalImage->aHardwareBuffer, getDevice(), metadata,
+            memoryTypeIndex, usage);
+    VkResult result = vkBindImageMemory(getDevice(), data.first, data.second, 0);
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+        << "vkBindImageMemory error=" << static_cast<int32_t>(result);
+    return data;
+}
 
-    // if external format we need to specifiy it in the allocation
-    bool const useExternalFormat = metadata.format == VK_FORMAT_UNDEFINED;
-
-    VkExternalFormatANDROID const externalFormat = {
+VkImageView VulkanPlatform::createExternalImageViewImpl(VkDevice device, SamplerYcbcrConversion chroma,
+            uint32_t internalFormat, VkImage image, VkImageSubresourceRange range,
+            VkImageViewType viewType, VkComponentMapping swizzle){
+    VkExternalFormatANDROID externalFormat = {
         .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
-        .pNext = nullptr,
-        .externalFormat = metadata.externalFormat,
-    };
-    VkExternalMemoryImageCreateInfo const externalCreateInfo = {
-        .sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO,
-        .pNext = useExternalFormat ? &externalFormat : nullptr,
-        .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID,
+        .externalFormat = internalFormat,
     };
 
-    VkImageCreateInfo const imageInfo{
-        .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
-        .pNext = &externalCreateInfo,
-        .flags = useExternalFormat ? VK_IMAGE_CREATE_ALIAS_BIT : 0u,
-        .imageType = VK_IMAGE_TYPE_2D,
-        .format = metadata.format,
-        .extent = {
-            metadata.width,
-            metadata.height,
-            1u,
-        },
-        .mipLevels = 1,
-        .arrayLayers = metadata.layers,
-        .samples = metadata.samples,
-        .usage = metadata.usage,
+    TextureSwizzle const swizzleArray[] = {chroma.r, chroma.g, chroma.b, chroma.a};
+    VkSamplerYcbcrConversionCreateInfo conversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_CREATE_INFO,
+        .pNext = &externalFormat,
+        .format = VK_FORMAT_UNDEFINED,
+        .ycbcrModel = fvkutils::getYcbcrModelConversion(chroma.ycbcrModel),
+        .ycbcrRange = fvkutils::getYcbcrRange(chroma.ycbcrRange),
+        .components = fvkutils::getSwizzleMap(swizzleArray),
+        .xChromaOffset = fvkutils::getChromaLocation(chroma.xChromaOffset),
+        .yChromaOffset = fvkutils::getChromaLocation(chroma.yChromaOffset),
+        .chromaFilter = fvkutils::getFilter(chroma.chromaFilter),
     };
-
-    VkDevice const device = getDevice();
-
-    VkImage image;
-    VkResult result = vkCreateImage(device, &imageInfo, VKALLOC, &image);
+    VkSamplerYcbcrConversion conversion = VK_NULL_HANDLE;
+    VkResult result = vkCreateSamplerYcbcrConversion(device, &conversionInfo,
+                                                     nullptr, &conversion);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-            << "vkCreateImage failed with error=" << static_cast<int32_t>(result);
+        << "Unable to create Ycbcr Conversion."
+        << " error=" << static_cast<int32_t>(result);
 
-    // Allocate the memory
-    VkImportAndroidHardwareBufferInfoANDROID const androidHardwareBufferInfo = {
-        .sType = VK_STRUCTURE_TYPE_IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID,
-        .buffer = buffer,
+    VkSamplerYcbcrConversionInfo samplerYcbcrConversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO,
+        .conversion = conversion,
     };
-    VkMemoryDedicatedAllocateInfo const memoryDedicatedAllocateInfo = {
-        .sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO,
-        .pNext = &androidHardwareBufferInfo,
+
+    VkImageViewCreateInfo viewInfo = {
+        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+        .pNext = &samplerYcbcrConversionInfo,
+        .flags = 0,
         .image = image,
-        .buffer = VK_NULL_HANDLE,
+        .viewType = viewType,
+        .format = VK_FORMAT_UNDEFINED,
+        .components = swizzle,
+        .subresourceRange = range,
+    };
+    VkImageView imageView;
+    result = vkCreateImageView(device, &viewInfo, VKALLOC, &imageView);
+    FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
+        << "Unable to create VkImageView."
+        << " error=" << static_cast<int32_t>(result);
+
+    return imageView;
+}
+
+VkSampler VulkanPlatform::createExternalSamplerImpl(
+        VkDevice device, SamplerYcbcrConversion chroma, SamplerParams params,
+        uint32_t internalFormat) {
+    VkExternalFormatANDROID externalFormat = {
+        .sType = VK_STRUCTURE_TYPE_EXTERNAL_FORMAT_ANDROID,
+        .externalFormat = internalFormat,
     };
 
-    VkPhysicalDeviceMemoryProperties memoryProperties;
-    vkGetPhysicalDeviceMemoryProperties(getPhysicalDevice(), &memoryProperties);
-
-
-    VkMemoryPropertyFlags const requiredMemoryFlags =
-            any(metadata.filamentUsage & TextureUsage::UPLOADABLE)
-                    ? VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
-                    : VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
-
-    uint32_t const memoryTypeIndex = VulkanContext::selectMemoryType(memoryProperties,
-            metadata.memoryTypeBits, requiredMemoryFlags);
-
-    VkMemoryAllocateInfo const allocInfo = {
-        .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
-        .pNext = &memoryDedicatedAllocateInfo,
-        .allocationSize = metadata.allocationSize,
-        .memoryTypeIndex = memoryTypeIndex,
+    TextureSwizzle const swizzleArray[] = {chroma.r, chroma.g, chroma.b, chroma.a};
+    VkSamplerYcbcrConversionCreateInfo conversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_CREATE_INFO,
+        .pNext = &externalFormat,
+        .format = VK_FORMAT_UNDEFINED,
+        .ycbcrModel = fvkutils::getYcbcrModelConversion(chroma.ycbcrModel),
+        .ycbcrRange = fvkutils::getYcbcrRange(chroma.ycbcrRange),
+        .components = fvkutils::getSwizzleMap(swizzleArray),
+        .xChromaOffset = fvkutils::getChromaLocation(chroma.xChromaOffset),
+        .yChromaOffset = fvkutils::getChromaLocation(chroma.yChromaOffset),
+        .chromaFilter = fvkutils::getFilter(chroma.chromaFilter),
     };
-    VkDeviceMemory memory;
-    result = vkAllocateMemory(device, &allocInfo, VKALLOC, &memory);
+    VkSamplerYcbcrConversion conversion = VK_NULL_HANDLE;
+    VkResult result = vkCreateSamplerYcbcrConversion(device, &conversionInfo,
+                                                     nullptr, &conversion);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-            << "vkAllocateMemory failed with error=" << static_cast<int32_t>(result);
+        << "Unable to create Ycbcr Conversion."
+        << " error=" << static_cast<int32_t>(result);
 
-    result = vkBindImageMemory(getDevice(), image, memory, 0);
+    VkSamplerYcbcrConversionInfo samplerYcbcrConversionInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO,
+        .pNext = nullptr,
+        .conversion = conversion,
+    };
+
+    VkSamplerCreateInfo samplerInfo = {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
+        .pNext = &samplerYcbcrConversionInfo,
+        .magFilter = fvkutils::getFilter(params.filterMag),
+        .minFilter = fvkutils::getFilter(params.filterMin),
+        .mipmapMode = fvkutils::getMipmapMode(params.filterMin),
+        .addressModeU = fvkutils::getWrapMode(params.wrapS),
+        .addressModeV = fvkutils::getWrapMode(params.wrapT),
+        .addressModeW = fvkutils::getWrapMode(params.wrapR),
+        .anisotropyEnable = params.anisotropyLog2 == 0 ? VK_FALSE : VK_TRUE,
+        .maxAnisotropy = (float)(1u << params.anisotropyLog2),
+        .compareEnable = fvkutils::getCompareEnable(params.compareMode),
+        .compareOp = fvkutils::getCompareOp(params.compareFunc),
+        .minLod = 0.0f,
+        .maxLod = fvkutils::getMaxLod(params.filterMin),
+        .borderColor = VK_BORDER_COLOR_INT_OPAQUE_BLACK,
+        .unnormalizedCoordinates = VK_FALSE,
+    };
+    VkSampler sampler;
+    result = vkCreateSampler(device, &samplerInfo, VKALLOC, &sampler);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-            << "vkBindImageMemory error=" << static_cast<int32_t>(result);
-    return { image, memory };
+        << "Unable to create sampler."
+        << " error=" << static_cast<int32_t>(result);
+    return sampler;
 }
 
 VulkanPlatform::ExtensionSet VulkanPlatformAndroid::getSwapchainInstanceExtensions() const {
@@ -369,9 +427,20 @@ VulkanPlatform::SurfaceBundle VulkanPlatformAndroid::createVkSurfaceKHR(void* na
 // Deprecated platform dependent helper methods
 VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl() { return {}; }
 
+VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device) {
+    return ExternalImageMetadata{};
+}
+
+VulkanPlatform::ImageData VulkanPlatform::createExternalImageDataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device,
+        const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
+    return ImageData{};
+}
+
 VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHRImpl(void* nativeWindow,
         VkInstance instance, uint64_t flags) noexcept {
     return SurfaceBundle{};
 }
 
-} // namespace filament::backend
+}// namespace filament::backend

filament/backend/src/vulkan/platform/VulkanPlatformApple.mm

@@ -63,6 +63,30 @@ VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl(
     return ret;
 }
 
+VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device) {
+    return {};
+}
+
+VulkanPlatform::ImageData VulkanPlatform::createExternalImageDataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device,
+        const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
+    return {};
+}
+
+VkSampler VulkanPlatform::createExternalSamplerImpl(VkDevice device,
+    SamplerYcbcrConversion chroma,
+    SamplerParams sampler,
+    uint32_t internalFormat) {
+    return VK_NULL_HANDLE;
+}
+
+VkImageView VulkanPlatform::createExternalImageViewImpl(VkDevice device,
+        SamplerYcbcrConversion chroma, uint32_t internalFormat, VkImage image,
+        VkImageSubresourceRange range, VkImageViewType viewType, VkComponentMapping swizzle) {
+    return VK_NULL_HANDLE;
+}
+
 VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHRImpl(void* nativeWindow,
         VkInstance instance, uint64_t flags) noexcept {
     VkSurfaceKHR surface;

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

@@ -84,6 +84,30 @@ using namespace bluevk;
 
 namespace filament::backend {
 
+VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device) {
+    return {};
+}
+
+VulkanPlatform::ImageData VulkanPlatform::createExternalImageDataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device,
+        const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
+    return {};
+}
+
+VkSampler VulkanPlatform::createExternalSamplerImpl(VkDevice device,
+    SamplerYcbcrConversion chroma,
+    SamplerParams sampler,
+    uint32_t internalFormat) {
+    return VK_NULL_HANDLE;
+}
+
+VkImageView VulkanPlatform::createExternalImageViewImpl(VkDevice device,
+        SamplerYcbcrConversion chroma, uint32_t internalFormat, VkImage image,
+        VkImageSubresourceRange range, VkImageViewType viewType, VkComponentMapping swizzle) {
+    return VK_NULL_HANDLE;
+}
+
 VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl() {
     VulkanPlatform::ExtensionSet const ret = {
 #if defined(__linux__) && defined(FILAMENT_SUPPORTS_WAYLAND)

filament/backend/src/vulkan/utils/Conversion.cpp

@@ -668,111 +668,4 @@ VkShaderStageFlags getShaderStageFlags(ShaderStageFlags stageFlags) {
     return flags;
 }
 
-VkSamplerYcbcrModelConversion getYcbcrModelConversion(
-    SamplerYcbcrModelConversion model) {
-    switch (model) {
-    case SamplerYcbcrModelConversion::RGB_IDENTITY:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY;
-    case SamplerYcbcrModelConversion::YCBCR_IDENTITY:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY;
-    case SamplerYcbcrModelConversion::YCBCR_709:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709;
-    case SamplerYcbcrModelConversion::YCBCR_601:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601;
-    case SamplerYcbcrModelConversion::YCBCR_2020:
-        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020;
-    default:
-        assert_invariant(false &&
-            "Unknown data type, conversion is not supported.");
-    }
-}
-
-VkSamplerYcbcrRange getYcbcrRange(SamplerYcbcrRange range) {
-    switch (range) {
-    case SamplerYcbcrRange::ITU_FULL:
-        return VK_SAMPLER_YCBCR_RANGE_ITU_FULL;
-    case SamplerYcbcrRange::ITU_NARROW:
-        return VK_SAMPLER_YCBCR_RANGE_ITU_NARROW;
-    default:
-        assert_invariant(false &&
-            "Unknown data type, conversion is not supported.");
-    }
-}
-
-VkChromaLocation getChromaLocation(ChromaLocation loc) {
-    switch (loc) {
-    case ChromaLocation::COSITED_EVEN:
-        return VK_CHROMA_LOCATION_COSITED_EVEN;
-    case ChromaLocation::MIDPOINT:
-        return VK_CHROMA_LOCATION_MIDPOINT;
-    default:
-        assert_invariant(false &&
-            "Unknown data type, conversion is not supported.");
-    }
-}
-
-SamplerYcbcrModelConversion getYcbcrModelConversionFilament(VkSamplerYcbcrModelConversion model) {
-    switch (model) {
-        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY:
-            return SamplerYcbcrModelConversion::RGB_IDENTITY;
-        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY:
-            return SamplerYcbcrModelConversion::YCBCR_IDENTITY;
-        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709:
-            return SamplerYcbcrModelConversion::YCBCR_709;
-        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601:
-            return SamplerYcbcrModelConversion::YCBCR_601;
-        case VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020:
-            return SamplerYcbcrModelConversion::YCBCR_2020;
-        default:
-            assert_invariant(false && "Unknown data type, conversion is not supported.");
-            return {};
-    }
-}
-
-SamplerYcbcrRange getYcbcrRangeFilament(VkSamplerYcbcrRange range) {
-    switch (range) {
-        case VK_SAMPLER_YCBCR_RANGE_ITU_FULL:
-            return SamplerYcbcrRange::ITU_FULL;
-        case VK_SAMPLER_YCBCR_RANGE_ITU_NARROW:
-            return SamplerYcbcrRange::ITU_NARROW;
-        default:
-            assert_invariant(false && "Unknown data type, conversion is not supported.");
-            return {};
-    }
-}
-
-ChromaLocation getChromaLocationFilament(VkChromaLocation loc) {
-    switch (loc) {
-        case VK_CHROMA_LOCATION_COSITED_EVEN:
-            return ChromaLocation::COSITED_EVEN;
-        case VK_CHROMA_LOCATION_MIDPOINT:
-            return ChromaLocation::MIDPOINT;
-        default:
-            assert_invariant(false && "Unknown data type, conversion is not supported.");
-            return {};
-    }
-}
-
-TextureSwizzle getSwizzleFilament(VkComponentSwizzle c, uint8_t rgbaIndex) {
-    switch (c) {
-        case VK_COMPONENT_SWIZZLE_ZERO:
-            return TextureSwizzle::SUBSTITUTE_ZERO;
-        case VK_COMPONENT_SWIZZLE_ONE:
-            return TextureSwizzle::SUBSTITUTE_ONE;
-        case VK_COMPONENT_SWIZZLE_IDENTITY:
-            return (TextureSwizzle) (((uint8_t) TextureSwizzle::CHANNEL_0) + rgbaIndex);
-        case VK_COMPONENT_SWIZZLE_R:
-            return TextureSwizzle::CHANNEL_0;
-        case VK_COMPONENT_SWIZZLE_G:
-            return TextureSwizzle::CHANNEL_1;
-        case VK_COMPONENT_SWIZZLE_B:
-            return TextureSwizzle::CHANNEL_2;
-        case VK_COMPONENT_SWIZZLE_A:
-            return TextureSwizzle::CHANNEL_3;
-        default:
-            assert_invariant(false && "Unknown data type, conversion is not supported.");
-            return {};
-    }
-}
-
 } // namespace filament::backend::fvkutils

filament/backend/src/vulkan/utils/Conversion.h

@@ -69,12 +69,6 @@ VkSamplerYcbcrModelConversion getYcbcrModelConversion(SamplerYcbcrModelConversio
 VkSamplerYcbcrRange getYcbcrRange(SamplerYcbcrRange range);
 VkChromaLocation getChromaLocation(ChromaLocation loc);
 
-// Ycbcr related functions
-SamplerYcbcrModelConversion getYcbcrModelConversionFilament(VkSamplerYcbcrModelConversion model);
-SamplerYcbcrRange getYcbcrRangeFilament(VkSamplerYcbcrRange range);
-ChromaLocation getChromaLocationFilament(VkChromaLocation loc);
-TextureSwizzle getSwizzleFilament(VkComponentSwizzle c, uint8_t rgbaIndex);
-
 inline VkImageViewType getViewType(SamplerType target) {
     switch (target) {
         case SamplerType::SAMPLER_CUBEMAP:

filament/backend/src/vulkan/utils/Definitions.h

@@ -348,10 +348,6 @@ using SamplerBitmask = utils::bitset64;
 // general.
 using InputAttachmentBitmask = utils::bitset64;
 
-constexpr uint8_t MAX_DESCRIPTOR_SET_BITMASK_BITS =
-        std::max(std::max(sizeof(UniformBufferBitmask), sizeof(SamplerBitmask)),
-                sizeof(InputAttachmentBitmask)) * 8;
-
 template<typename Bitmask>
 static constexpr uint8_t getVertexStageShift() noexcept {
     // We assume the bottom half of bits are for vertex stages.

filament/backend/src/vulkan/utils/Image.cpp

@@ -193,6 +193,49 @@ uint8_t reduceSampleCount(uint8_t sampleCount, VkSampleCountFlags mask) {
     return mostSignificantBit((sampleCount - 1) & mask);
 }
 
+VkSamplerYcbcrModelConversion getYcbcrModelConversion(
+    SamplerYcbcrModelConversion model) {
+    switch (model) {
+    case SamplerYcbcrModelConversion::RGB_IDENTITY:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_RGB_IDENTITY;
+    case SamplerYcbcrModelConversion::YCBCR_IDENTITY:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_IDENTITY;
+    case SamplerYcbcrModelConversion::YCBCR_709:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709;
+    case SamplerYcbcrModelConversion::YCBCR_601:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_601;
+    case SamplerYcbcrModelConversion::YCBCR_2020:
+        return VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_2020;
+    default:
+        assert_invariant(false &&
+            "Unknown data type, conversion is not supported.");
+    }
+}
+
+VkSamplerYcbcrRange getYcbcrRange(SamplerYcbcrRange range) {
+    switch (range) {
+    case SamplerYcbcrRange::ITU_FULL:
+        return VK_SAMPLER_YCBCR_RANGE_ITU_FULL;
+    case SamplerYcbcrRange::ITU_NARROW:
+        return VK_SAMPLER_YCBCR_RANGE_ITU_NARROW;
+    default:
+        assert_invariant(false &&
+            "Unknown data type, conversion is not supported.");
+    }
+}
+
+VkChromaLocation getChromaLocation(ChromaLocation loc) {
+    switch (loc) {
+    case ChromaLocation::COSITED_EVEN:
+        return VK_CHROMA_LOCATION_COSITED_EVEN;
+    case ChromaLocation::MIDPOINT:
+        return VK_CHROMA_LOCATION_MIDPOINT;
+    default:
+        assert_invariant(false &&
+            "Unknown data type, conversion is not supported.");
+    }
+}
+
 } // namespace filament::backend::fvkutils
 
 bool operator<(const VkImageSubresourceRange& a, const VkImageSubresourceRange& b) {

filament/backend/src/webgpu/WebGPUDriver.cpp

@@ -22,6 +22,7 @@
 #include "CommandStreamDispatcher.h"
 #include "DriverBase.h"
 #include "private/backend/Dispatcher.h"
+
 #include <backend/DriverEnums.h>
 #include <backend/Handle.h>
 
@@ -256,10 +257,14 @@ void WebGPUDriver::terminate() {
 }
 
 void WebGPUDriver::tick(int) {
+    mDevice.Tick();
 }
 
 void WebGPUDriver::beginFrame(int64_t monotonic_clock_ns,
         int64_t refreshIntervalNs, uint32_t frameId) {
+    wgpu::CommandEncoderDescriptor commandEncoderDescriptor{};
+    commandEncoderDescriptor.nextInChain = nullptr;
+    mCommandEncoder = mDevice.CreateCommandEncoder(&commandEncoderDescriptor);
 }
 
 void WebGPUDriver::setFrameScheduledCallback(Handle<HwSwapChain> sch,
@@ -276,6 +281,12 @@ void WebGPUDriver::setPresentationTime(int64_t monotonic_clock_ns) {
 }
 
 void WebGPUDriver::endFrame(uint32_t frameId) {
+    // FWGPU_LOGW << __FUNCTION__<< "\n";
+    mQueue.Submit(1, &mCommandBuffer);
+    mCommandEncoder = nullptr;
+    mCommandBuffer = nullptr;
+    mTextureView = nullptr;
+    mSwapChain->Present();
 }
 
 void WebGPUDriver::flush(int) {
@@ -342,7 +353,7 @@ Handle<HwSwapChain> WebGPUDriver::createSwapChainHeadlessS() noexcept {
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureS() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUTexture>();
 }
 
 Handle<HwTexture> WebGPUDriver::importTextureS() noexcept {
@@ -366,19 +377,19 @@ Handle<HwIndexBuffer> WebGPUDriver::createIndexBufferS() noexcept {
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureViewS() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUTexture>();
 }
 
 Handle<HwBufferObject> WebGPUDriver::createBufferObjectS() noexcept {
-    return Handle<HwBufferObject>((Handle<HwBufferObject>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUBufferObject>();
 }
 
 Handle<HwRenderTarget> WebGPUDriver::createRenderTargetS() noexcept {
-    return Handle<HwRenderTarget>((Handle<HwRenderTarget>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPURenderTarget>();
 }
 
 Handle<HwVertexBuffer> WebGPUDriver::createVertexBufferS() noexcept {
-    return Handle<HwVertexBuffer>((Handle<HwVertexBuffer>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUVertexBuffer>();
 }
 
 Handle<HwDescriptorSet> WebGPUDriver::createDescriptorSetS() noexcept {
@@ -390,15 +401,15 @@ Handle<HwRenderPrimitive> WebGPUDriver::createRenderPrimitiveS() noexcept {
 }
 
 Handle<HwVertexBufferInfo> WebGPUDriver::createVertexBufferInfoS() noexcept {
-    return Handle<HwVertexBufferInfo>((Handle<HwVertexBufferInfo>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUVertexBufferInfo>();
 }
 
 Handle<HwTexture> WebGPUDriver::createTextureViewSwizzleS() noexcept {
-    return Handle<HwTexture>((Handle<HwTexture>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPUTexture>();
 }
 
 Handle<HwRenderTarget> WebGPUDriver::createDefaultRenderTargetS() noexcept {
-    return Handle<HwRenderTarget>((Handle<HwRenderTarget>::HandleId) mNextFakeHandle++);
+    return allocHandle<WGPURenderTarget>();
 }
 
 Handle<HwDescriptorSetLayout> WebGPUDriver::createDescriptorSetLayoutS() noexcept {
@@ -422,7 +433,7 @@ void WebGPUDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow,
     // TODO:  use webgpu handle allocator from.
     //        https://github.com/google/filament/pull/8566
     // HwSwapChain* hwSwapChain = handleCast<HwSwapChain*>(sch);
-    mSwapChain = nullptr;
+    assert_invariant(!mSwapChain);
     wgpu::Surface surface = mPlatform.createSurface(nativeWindow, flags);
     mAdapter = mPlatform.requestAdapter(surface);
 #if FWGPU_ENABLED(FWGPU_PRINT_SYSTEM)
@@ -434,6 +445,28 @@ void WebGPUDriver::createSwapChainR(Handle<HwSwapChain> sch, void* nativeWindow,
 #endif
     mQueue = mDevice.GetQueue();
     mSwapChain = std::make_unique<WebGPUSwapChain>(std::move(surface), mAdapter, mDevice, flags);
+    // TODO configure the surface (maybe before or after creating the swapchain?
+    //                             how do we get the surface extent?)
+    // TODO actually create the swapchain
+    // auto onQueueWorkDone = [](wgpu::QueueWorkDoneStatus status, void* /* pUserData */) {
+        // FWGPU_LOGW << "Queued work finished with status: " << status << std::endl;
+    // };
+    // mQueue.OnSubmittedWorkDone(wgpu::CallbackMode::WaitAnyOnly, [](wgpu::QueueWorkDoneStatus status, void* pUserdata) {
+    //     FWGPU_LOGW << "Queued work finished with status:\n";
+    // }, nullptr /* pUserData */);
+    void* userDataPtr = nullptr;
+    mQueue.OnSubmittedWorkDone(
+        wgpu::CallbackMode::AllowProcessEvents,
+        [](wgpu::QueueWorkDoneStatus status, void* pUserData) {
+            FWGPU_LOGW << "Queued work finished with status: " << static_cast<int>(status) << "\n";
+            if (pUserData == nullptr) {
+                // Expected case
+            } else {
+                FWGPU_LOGW << "Unexpected non-null pUserData received.\n";
+            }
+        },
+        userDataPtr /* pUserData */
+    );
     FWGPU_LOGW << "WebGPU support is still essentially a no-op at this point in development (only "
                   "background components have been instantiated/selected, such as surface/screen, "
                   "graphics device/GPU, etc.), thus nothing is being drawn to the screen."
@@ -471,7 +504,13 @@ void WebGPUDriver::createTextureR(Handle<HwTexture> th, SamplerType target, uint
         TextureUsage usage) {}
 
 void WebGPUDriver::createTextureViewR(Handle<HwTexture> th, Handle<HwTexture> srch,
-        uint8_t baseLevel, uint8_t levelCount) {}
+        uint8_t baseLevel, uint8_t levelCount) {
+    // FWGPU_LOGW << __FUNCTION__<< "\n";
+    WGPUTexture const* src = handleCast<WGPUTexture>(srch);
+    (void) src;
+    // textures.insert(
+    //         constructHandle<WGPUTexture>(th, src, baseLevel, levelCount));
+}
 
 void WebGPUDriver::createTextureViewSwizzleR(Handle<HwTexture> th, Handle<HwTexture> srch,
         backend::TextureSwizzle r, backend::TextureSwizzle g, backend::TextureSwizzle b,
@@ -498,7 +537,11 @@ void WebGPUDriver::createRenderPrimitiveR(Handle<HwRenderPrimitive> rph, Handle<
 
 void WebGPUDriver::createProgramR(Handle<HwProgram> ph, Program&& program) {}
 
-void WebGPUDriver::createDefaultRenderTargetR(Handle<HwRenderTarget> rth, int) {}
+void WebGPUDriver::createDefaultRenderTargetR(Handle<HwRenderTarget> rth, int) {
+    assert_invariant(!mDefaultRenderTarget);
+    mDefaultRenderTarget = constructHandle<WGPURenderTarget>(rth);
+    assert_invariant(mDefaultRenderTarget);
+}
 
 void WebGPUDriver::createRenderTargetR(Handle<HwRenderTarget> rth, TargetBufferFlags targets,
         uint32_t width, uint32_t height, uint8_t samples, uint8_t layerCount, MRT color,
@@ -659,6 +702,11 @@ void WebGPUDriver::resetBufferObject(Handle<HwBufferObject> boh) {
 
 void WebGPUDriver::setVertexBufferObject(Handle<HwVertexBuffer> vbh, uint32_t index,
         Handle<HwBufferObject> boh) {
+    auto* vertexBuffer = handleCast<WGPUVertexBuffer>(vbh);
+    auto* bufferObject = handleCast<WGPUBufferObject>(boh);
+    assert_invariant(index < vertexBuffer->mBuffers.size());
+    vertexBuffer->setBuffer(bufferObject, index);
+
 }
 
 void WebGPUDriver::update3DImage(Handle<HwTexture> th,
@@ -691,9 +739,33 @@ void WebGPUDriver::compilePrograms(CompilerPriorityQueue priority,
 }
 
 void WebGPUDriver::beginRenderPass(Handle<HwRenderTarget> rth, const RenderPassParams& params) {
+    // FWGPU_LOGW << __FUNCTION__<< "\n";
+    mTextureView = mSwapChain->GetNextSurfaceTextureView(params.viewport.width, params.viewport.height);
+    wgpu::RenderPassColorAttachment renderPassColorAttachment = {};
+    renderPassColorAttachment.view = mTextureView;
+    renderPassColorAttachment.resolveTarget = nullptr;
+    renderPassColorAttachment.loadOp = wgpu::LoadOp::Clear;
+    renderPassColorAttachment.storeOp = wgpu::StoreOp::Store;
+    renderPassColorAttachment.clearValue = wgpu::Color{1, 0 , 0 , 1};
+    renderPassColorAttachment.depthSlice = WGPU_DEPTH_SLICE_UNDEFINED;
+
+    wgpu::RenderPassDescriptor renderPassDescriptor = {};
+    renderPassDescriptor.nextInChain = nullptr;
+    renderPassDescriptor.colorAttachmentCount = 1;
+    renderPassDescriptor.colorAttachments = &renderPassColorAttachment;
+    renderPassDescriptor.depthStencilAttachment = nullptr;
+    renderPassDescriptor.timestampWrites = nullptr;
+
+    mRenderPassEncoder = mCommandEncoder.BeginRenderPass(&renderPassDescriptor);
+    mRenderPassEncoder.SetViewport((float)params.viewport.left, (float)params.viewport.bottom,
+            (float) params.viewport.width, (float) params.viewport.height, params.depthRange.near, params.depthRange.far);
+            // (float) 1024/*params.viewport.width*/, (float) 640/*params.viewport.height*/, params.depthRange.near, params.depthRange.far);
 }
 
 void WebGPUDriver::endRenderPass(int) {
+    // FWGPU_LOGW << __FUNCTION__<< "\n";
+    mRenderPassEncoder.End();
+    mRenderPassEncoder = nullptr;
 }
 
 void WebGPUDriver::nextSubpass(int) {
@@ -703,6 +775,12 @@ void WebGPUDriver::makeCurrent(Handle<HwSwapChain> drawSch, Handle<HwSwapChain>
 }
 
 void WebGPUDriver::commit(Handle<HwSwapChain> sch) {
+    wgpu::CommandBufferDescriptor commandBufferDescriptor = {};
+    commandBufferDescriptor.nextInChain = nullptr;
+    mCommandBuffer = mCommandEncoder.Finish(&commandBufferDescriptor);
+
+    // mQueue.Submit(1, &mCommandBuffer);
+    // mCommandBuffer = nullptr;
 }
 
 void WebGPUDriver::setPushConstant(backend::ShaderStage stage, uint8_t index,
@@ -759,6 +837,8 @@ void WebGPUDriver::bindRenderPrimitive(Handle<HwRenderPrimitive> rph) {
 }
 
 void WebGPUDriver::draw2(uint32_t indexOffset, uint32_t indexCount, uint32_t instanceCount) {
+    // mRenderPassEncoder.DrawIndexed(indexCount, instanceCount, indexOffset, 0, 0);
+
 }
 
 void WebGPUDriver::draw(PipelineState pipelineState, Handle<HwRenderPrimitive> rph,

filament/backend/src/webgpu/WebGPUDriver.h

@@ -30,6 +30,8 @@
 
 #include <webgpu/webgpu_cpp.h>
 
+#include "WebGPUHandles.h"
+
 #include <cstdint>
 #include <memory>
 
@@ -63,7 +65,11 @@ private:
     // TODO consider moving to handle allocator when ready
     std::unique_ptr<WebGPUSwapChain> mSwapChain = nullptr;
     uint64_t mNextFakeHandle = 1;
-
+    wgpu::CommandEncoder mCommandEncoder = nullptr;
+    wgpu::TextureView mTextureView = nullptr;
+    wgpu::RenderPassEncoder mRenderPassEncoder = nullptr;
+    wgpu::CommandBuffer mCommandBuffer = nullptr;
+    WGPURenderTarget* mDefaultRenderTarget = nullptr;
     /*
      * Driver interface
      */
@@ -93,6 +99,15 @@ private:
         return mHandleAllocator.allocate<D>();
     }
 
+    template<typename D, typename B, typename ... ARGS>
+    D* constructHandle(Handle<B>& handle, ARGS&& ... args) noexcept {
+        return mHandleAllocator.construct<D>(handle, std::forward<ARGS>(args)...);
+    }
+    template<typename D, typename B>
+    D* handleCast(Handle<B> handle) noexcept {
+        return mHandleAllocator.handle_cast<D*>(handle);
+    }
+
 };
 
 }// namespace filament::backend

filament/backend/src/webgpu/WebGPUHandles.cpp

@@ -0,0 +1,19 @@
+//
+// Created by Idris Idris Shah on 3/21/25.
+//
+
+#include "WebGPUHandles.h"
+
+namespace filament::backend {
+
+WGPUVertexBuffer::WGPUVertexBuffer(uint32_t vextexCount, uint32_t bufferCount,
+        Handle<WGPUVertexBufferInfo> vbih)
+    : HwVertexBuffer(vextexCount),
+      vbih(vbih),
+      mBuffers(MAX_VERTEX_BUFFER_COUNT) {}
+void WGPUVertexBuffer::setBuffer(WGPUBufferObject* bufferObject, uint32_t index) {}
+
+WGPUBufferObject::WGPUBufferObject(BufferObjectBinding bindingType, uint32_t byteCount)
+    : HwBufferObject(byteCount),
+      mBindingType(bindingType) {}
+}// namespace filament::backend

filament/backend/src/webgpu/WebGPUHandles.h

@@ -0,0 +1,109 @@
+//
+// Created by Idris Idris Shah on 3/21/25.
+//
+
+#ifndef TNT_FILAMENT_BACKEND_WEBGPUHANDLES_H
+#define TNT_FILAMENT_BACKEND_WEBGPUHANDLES_H
+
+#include "DriverBase.h"
+
+#include <backend/DriverEnums.h>
+#include <backend/Handle.h>
+#include <utils/FixedCapacityVector.h>
+
+#include <webgpu/webgpu_cpp.h>
+
+namespace filament::backend {
+
+struct WGPUBufferObject;
+struct WGPUVertexBufferInfo : public HwVertexBufferInfo {
+    WGPUVertexBufferInfo(uint8_t bufferCount, uint8_t attributeCount,
+            AttributeArray const& attributes)
+        : HwVertexBufferInfo(bufferCount, attributeCount),
+          attributes(attributes) {}
+    AttributeArray attributes;
+};
+
+struct WGPUVertexBuffer : public HwVertexBuffer {
+    WGPUVertexBuffer(uint32_t vextexCount, uint32_t bufferCount, Handle<WGPUVertexBufferInfo> vbih);
+    void setBuffer(WGPUBufferObject* bufferObject, uint32_t index);
+
+    Handle<WGPUVertexBufferInfo> vbih;
+    utils::FixedCapacityVector<wgpu::Buffer> mBuffers;
+};
+
+struct WGPUIndexBuffer : public HwIndexBuffer {
+    WGPUIndexBuffer(BufferUsage usage, uint8_t elementSize, uint32_t indexCount);
+
+    wgpu::Buffer buffer;
+};
+
+struct WGPUBufferObject : HwBufferObject {
+    WGPUBufferObject(BufferObjectBinding bindingType, uint32_t byteCount);
+
+    wgpu::Buffer mBuffer;
+    const BufferObjectBinding mBindingType;
+};
+
+class WGPUTexture : public HwTexture {
+public:
+    WGPUTexture(SamplerType target, uint8_t levels, TextureFormat format, uint8_t samples,
+            uint32_t width, uint32_t height, uint32_t depth, TextureUsage usage) noexcept;
+
+    // constructors for creating texture views
+    WGPUTexture(WGPUTexture const* src, uint8_t baseLevel, uint8_t levelCount) noexcept;
+
+    wgpu::Texture texture = nullptr;
+};
+
+class WGPURenderTarget : public HwRenderTarget {
+public:
+    class Attachment {
+    public:
+        friend class WGPURenderTarget;
+
+        Attachment() = default;
+        Attachment(WGPUTexture* gpuTexture, uint8_t level = 0, uint16_t layer = 0)
+            : level(level),
+              layer(layer),
+              texture(gpuTexture->texture),
+              mWGPUTexture(gpuTexture) {}
+
+        uint8_t level = 0;
+        uint16_t layer = 0;
+
+    private:
+        wgpu::Texture texture = nullptr;
+        WGPUTexture* mWGPUTexture = nullptr;
+    };
+
+    WGPURenderTarget(uint32_t width, uint32_t height, uint8_t samples,
+            Attachment colorAttachments[MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT]);
+    WGPURenderTarget()
+        : HwRenderTarget(0, 0),
+          defaultRenderTarget(true) {}
+
+    void setUpRenderPassAttachments(wgpu::RenderPassDescriptor* descriptor,
+            const RenderPassParams& params);
+
+    math::uint2 getAttachmentSize() noexcept;
+
+    bool isDefaultRenderTarget() const { return defaultRenderTarget; }
+    uint8_t getSamples() const { return samples; }
+
+    Attachment getDrawColorAttachment(size_t index);
+    Attachment getReadColorAttachment(size_t index);
+
+private:
+    static wgpu::LoadOp getLoadAction(const RenderPassParams& params, TargetBufferFlags buffer);
+    static wgpu::LoadOp getStoreAction(const RenderPassParams& params, TargetBufferFlags buffer);
+
+    bool defaultRenderTarget = false;
+    uint8_t samples = 1;
+
+    Attachment color[MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT] = {};
+    math::uint2 attachmentSize = {};
+};
+
+}// namespace filament::backend
+#endif// TNT_FILAMENT_BACKEND_WEBGPUHANDLES_H

filament/backend/src/webgpu/WebGPUSwapChain.cpp

@@ -246,4 +246,32 @@ void WebGPUSwapChain::GetCurrentTexture(uint32_t width, uint32_t height, wgpu::S
     mSurface.GetCurrentTexture(texture);
 }
 
+wgpu::TextureView WebGPUSwapChain::GetNextSurfaceTextureView(uint32_t width, uint32_t height) {
+    wgpu::SurfaceTexture surfaceTexture;
+    GetCurrentTexture(width, height, &surfaceTexture);
+    if (surfaceTexture.status != wgpu::SurfaceGetCurrentTextureStatus::Success) {
+        return nullptr;
+    }
+
+    // Create a view for this surface texture
+    wgpu::TextureViewDescriptor textureViewDescriptor;
+    textureViewDescriptor.nextInChain = nullptr;
+    textureViewDescriptor.label = "Surface texture view";
+    textureViewDescriptor.format = surfaceTexture.texture.GetFormat();
+    textureViewDescriptor.dimension = wgpu::TextureViewDimension::e2D;
+    textureViewDescriptor.baseMipLevel = 0;
+    textureViewDescriptor.mipLevelCount = 1;
+    textureViewDescriptor.baseArrayLayer = 0;
+    textureViewDescriptor.arrayLayerCount = 1;
+    textureViewDescriptor.aspect = wgpu::TextureAspect::All;
+    wgpu::TextureView textureView = surfaceTexture.texture.CreateView(&textureViewDescriptor);
+
+    return textureView;
+}
+
+void WebGPUSwapChain::Present() {
+    assert_invariant(mSurface);
+    mSurface.Present();
+}
+
 }// namespace filament::backend

filament/backend/src/webgpu/WebGPUSwapChain.h

@@ -32,6 +32,8 @@ public:
     ~WebGPUSwapChain();
 
     void GetCurrentTexture(uint32_t width, uint32_t height, wgpu::SurfaceTexture*);
+    wgpu::TextureView GetNextSurfaceTextureView(uint32_t width, uint32_t height);
+    void Present();
 
 private:
     wgpu::Surface mSurface = {};

filament/backend/test/Shader.cpp

@@ -23,44 +23,30 @@ namespace test {
 
 using namespace filament::backend;
 
-Shader::Shader(DriverApi& api, Cleanup& cleanup, ShaderConfig config) : mCleanup(cleanup) {
-    utils::FixedCapacityVector<DescriptorSetLayoutBinding> kLayouts(config.uniforms.size());
-    for (unsigned char i = 0; i < config.uniforms.size(); ++i) {
-        kLayouts[i] = {
-                config.uniforms[i].type.value_or(DescriptorType::UNIFORM_BUFFER),
-                ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, i };
-    }
+Shader::Shader(DriverApi& api, Cleanup& cleanup, ShaderConfig config) {
+    utils::FixedCapacityVector<DescriptorSetLayoutBinding> kLayouts(config.uniformNames.size());
+    for (unsigned char i = 0; i < config.uniformNames.size(); ++i) {
+        kLayouts[i] =
+                { DescriptorType::UNIFORM_BUFFER, ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, i };
+    };
 
-    // This assumes that the uniforms will all be in a single descriptor set at index 1.
-    // If there are shaders with uniforms in other sets then ShaderConfig will need to be expanded
-    // to accommodate that.
-    size_t kDescriptorSetIndex = 1;
     filamat::DescriptorSets descriptors;
-    descriptors[kDescriptorSetIndex] = filamat::DescriptorSetInfo(config.uniforms.size());
-    for (unsigned char i = 0; i < config.uniforms.size(); ++i) {
-        descriptors[kDescriptorSetIndex][i] = {
-                config.uniforms[i].name, kLayouts[i], config.uniforms[i].samplerInfo };
+    for (unsigned char i = 0; i < config.uniformNames.size(); ++i) {
+        descriptors[i + 1] = {{ config.uniformNames[i], kLayouts[i], {}}};
     }
     ShaderGenerator shaderGen(
             std::move(config.vertexShader), std::move(config.fragmentShader), BackendTest::sBackend,
             BackendTest::sIsMobilePlatform, std::move(descriptors));
     Program prog = shaderGen.getProgram(api);
-
-    Program::DescriptorBindingsInfo bindingsInfo(config.uniforms.size());
-    for (unsigned char i = 0; i < config.uniforms.size(); ++i) {
-        bindingsInfo[i] = {
-                config.uniforms[i].name,
-                config.uniforms[i].type.value_or(DescriptorType::UNIFORM_BUFFER), i };
+    for (unsigned char i = 0; i < config.uniformNames.size(); ++i) {
+        prog.descriptorBindings(1, {{ config.uniformNames[i], DescriptorType::UNIFORM_BUFFER, i }});
     }
-    prog.descriptorBindings(1, bindingsInfo);
     mProgram = cleanup.add(api.createProgram(std::move(prog)));
 
     mDescriptorSetLayout = cleanup.add(
             api.createDescriptorSetLayout(DescriptorSetLayout{ kLayouts }));
-}
 
-filament::backend::DescriptorSetHandle Shader::createDescriptorSet(DriverApi& api) const {
-    return mCleanup.add(api.createDescriptorSet(mDescriptorSetLayout));
+    mDescriptorSet = cleanup.add(api.createDescriptorSet(mDescriptorSetLayout));
 }
 
 filament::backend::ProgramHandle Shader::getProgram() const {
@@ -71,4 +57,8 @@ filament::backend::DescriptorSetLayoutHandle Shader::getDescriptorSetLayout() co
     return mDescriptorSetLayout;
 }
 
+filament::backend::DescriptorSetHandle Shader::getDescriptorSet() const {
+    return mDescriptorSet;
+}
+
 } // namespace test

filament/backend/test/Shader.h

@@ -18,22 +18,14 @@
 #define TNT_SHADER_H
 
 #include "Lifetimes.h"
-#include "private/filament/SamplerInterfaceBlock.h"
 
 namespace test {
 
-struct UniformConfig {
-    utils::CString name;
-    // If not specified this will be DescriptorType::UNIFORM_BUFFER
-    std::optional<filament::backend::DescriptorType> type;
-    std::optional<filament::SamplerInterfaceBlock::SamplerInfo> samplerInfo;
-};
-
 // All describing a shader that should be created.
 struct ShaderConfig {
     std::string vertexShader;
     std::string fragmentShader;
-    std::vector<UniformConfig> uniforms;
+    std::vector<utils::CString> uniformNames;
 };
 
 // All values describing a uniform.
@@ -43,7 +35,6 @@ struct ResolvedUniformBindingConfig {
     uint32_t byteOffset;
     filament::backend::descriptor_set_t set;
     filament::backend::descriptor_binding_t binding;
-    std::optional<filament::backend::DescriptorSetHandle> descriptorSet;
 };
 
 // An equivalent to ResolvedUniformBindingConfig with all fields optional.
@@ -55,7 +46,6 @@ struct UniformBindingConfig {
     std::optional<uint32_t> byteOffset;
     std::optional<filament::backend::descriptor_set_t> set;
     std::optional<filament::backend::descriptor_binding_t> binding;
-    std::optional<filament::backend::DescriptorSetHandle> descriptorSet;
 
     template<typename UniformType>
     ResolvedUniformBindingConfig resolve();
@@ -89,14 +79,12 @@ public:
 
     filament::backend::ProgramHandle getProgram() const;
     filament::backend::DescriptorSetLayoutHandle getDescriptorSetLayout() const;
-
-    filament::backend::DescriptorSetHandle createDescriptorSet(
-            filament::backend::DriverApi& api) const;
+    filament::backend::DescriptorSetHandle getDescriptorSet() const;
 
 protected:
-    Cleanup& mCleanup;
     filament::backend::ProgramHandle mProgram;
     filament::backend::DescriptorSetLayoutHandle mDescriptorSetLayout;
+    filament::backend::DescriptorSetHandle mDescriptorSet;
 };
 
 template<typename UniformType>
@@ -107,8 +95,7 @@ ResolvedUniformBindingConfig UniformBindingConfig::resolve() {
             .bufferSize = bufferSize.value_or(resolvedDataSize),
             .byteOffset = byteOffset.value_or(0),
             .set = set.value_or(1),
-            .binding = binding.value_or(0),
-            .descriptorSet = descriptorSet
+            .binding = binding.value_or(0)
     };
 }
 
@@ -133,16 +120,9 @@ void Shader::bindUniform(filament::backend::DriverApi& api,
         UniformBindingConfig config) const {
     auto resolvedConfig = config.resolve<UniformType>();
 
-    filament::backend::DescriptorSetHandle descriptorSet;
-    if (resolvedConfig.descriptorSet.has_value()) {
-        descriptorSet = *resolvedConfig.descriptorSet;
-    } else {
-        descriptorSet = createDescriptorSet(api);
-    }
-
-    api.updateDescriptorSetBuffer(descriptorSet, resolvedConfig.binding, hwBuffer, 0,
+    api.updateDescriptorSetBuffer(getDescriptorSet(), resolvedConfig.binding, hwBuffer, 0,
             resolvedConfig.bufferSize);
-    api.bindDescriptorSet(descriptorSet, resolvedConfig.set, {});
+    api.bindDescriptorSet(getDescriptorSet(), resolvedConfig.set, {});
 }
 
 template<typename UniformType>

filament/backend/test/SharedShaders.cpp

@@ -1,253 +0,0 @@
-/*
- * Copyright (C) 2021 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#include "SharedShaders.h"
-
-#include "Shader.h"
-#include "absl/strings/str_format.h"
-#include "gtest/gtest.h"
-
-namespace test {
-
-using namespace filament::backend;
-
-namespace {
-
-// A shader stored in pieces so that uniform declarations can be injected.
-struct ShaderText {
-    std::string mPrefix;
-    std::string mBody;
-
-    std::string withUniform(const std::string& uniformText) const {
-        return absl::StrFormat("%s\n%s\n%s", mPrefix.c_str(), uniformText.c_str(), mBody.c_str());
-    }
-};
-
-std::optional<ShaderText> GetGlslVertexShader(VertexShaderType type) {
-    switch (type) {
-        case VertexShaderType::Noop: {
-            return ShaderText{
-                    R"(
-#version 450 core
-layout(location = 0) in vec4 mesh_position;
-)", R"(
-void main() {
-    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
-#if defined(TARGET_VULKAN_ENVIRONMENT)
-    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
-    gl_Position.y = -gl_Position.y;
-#endif
-})" };
-        }
-        case VertexShaderType::Simple: {
-            return ShaderText{
-                    R"(
-#version 450 core
-layout(location = 0) in vec4 mesh_position;
-)", R"(
-void main() {
-    gl_Position = vec4(
-            mesh_position.xy * (params.scaleMinusOne.xy + 1.0) + params.offset.xy,
-            params.scaleMinusOne.z + 1.0,
-            1.0);
-#if defined(TARGET_VULKAN_ENVIRONMENT)
-    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
-    gl_Position.y = -gl_Position.y;
-#endif
-})" };
-        }
-        case VertexShaderType::Textured: {
-            return ShaderText{
-                    R"(
-#version 450 core
-layout(location = 0) in vec4 mesh_position;
-layout(location = 0) out vec2 uv;
-)", R"(
-void main() {
-    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
-    uv = (mesh_position.xy * 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.
-    gl_Position.y = -gl_Position.y;
-#endif
-})" };
-        }
-        default:
-            return std::nullopt;
-    }
-}
-
-std::optional<ShaderText> GetGlslFragmentShader(FragmentShaderType type) {
-    switch (type) {
-        case FragmentShaderType::White: {
-            return ShaderText{
-                    R"(
-#version 450 core
-precision mediump int; precision highp float;
-layout(location = 0) out vec4 fragColor;
-)", R"(
-void main() {
-    fragColor = vec4(1.0, 1.0, 1.0, 1.0);
-})" };
-        }
-        case FragmentShaderType::SolidColored: {
-            return ShaderText{
-                    R"(
-#version 450 core
-precision mediump int; precision highp float;
-layout(location = 0) out vec4 fragColor;
-)", R"(
-void main() {
-    fragColor = params.color;
-})" };
-        }
-        case FragmentShaderType::Textured: {
-            return ShaderText{
-                    R"(
-#version 450 core
-precision mediump int; precision highp float;
-layout(location = 0) out vec4 fragColor;
-layout(location = 0) in vec2 uv;
-)", R"(
-void main() {
-    fragColor = texture(test_tex, uv);
-})" };
-        }
-        default:
-            return std::nullopt;
-    }
-}
-
-std::optional<std::string> GetGlslUniform(ShaderUniformType type) {
-    switch (type) {
-        case ShaderUniformType::None: {
-            return "";
-        }
-        case ShaderUniformType::Simple: {
-            return R"(
-layout(binding = 0, set = 1) uniform Params {
-    highp vec4 color;
-    // Use scaleMinusOne instead of scale so that a 0 initialized value is a good default
-    highp vec4 scaleMinusOne;
-    highp vec4 offset;
-} params;
-)";
-        }
-        case ShaderUniformType::SimpleWithPadding: {
-            return R"(
-layout(binding = 0, set = 1) uniform Params {
-    highp vec4 padding[4];  // offset of 64 bytes
-
-    highp vec4 color;
-    // Use scaleMinusOne instead of scale so that a 0 initialized value is a good default
-    highp vec4 scaleMinusOne;
-    highp vec4 offset;
-} params;
-)";
-        }
-        case ShaderUniformType::Sampler: {
-            return R"(
-layout(location = 0, set = 1) uniform sampler2D test_tex;
-)";
-        }
-        default:
-            return std::nullopt;
-    }
-}
-
-std::vector<UniformConfig> GetUniformConfig(ShaderUniformType type) {
-    switch (type) {
-        case ShaderUniformType::None: {
-            return {};
-        }
-        case ShaderUniformType::Simple: {
-            return {{ "Params" }};
-        }
-        case ShaderUniformType::SimpleWithPadding: {
-            return {{ "Params" }};
-        }
-        case ShaderUniformType::Sampler: {
-            filament::SamplerInterfaceBlock::SamplerInfo samplerInfo{
-                    "backend_test", "test_tex", 0,
-                    SamplerType::SAMPLER_2D, SamplerFormat::FLOAT, Precision::HIGH, false };
-            return {{
-                            "test_tex", DescriptorType::SAMPLER, samplerInfo
-                    }};
-        }
-        default:
-            abort();
-    }
-}
-
-ShaderLanguage getShaderLanguage(const Backend& backend) {
-    switch (backend) {
-        case Backend::METAL:
-            return ShaderLanguage::MSL;
-        case Backend::WEBGPU:
-            return ShaderLanguage::WGSL;
-        case Backend::VULKAN:
-        case Backend::NOOP:
-        case Backend::OPENGL:
-        default: {
-            return ShaderLanguage::GLSL;
-        }
-    }
-}
-
-} // namespace
-
-Shader SharedShaders::makeShader(filament::backend::DriverApi& api, Cleanup& cleanup,
-        ShaderRequest request) {
-    std::optional<ShaderText> vertex;
-    std::optional<ShaderText> fragment;
-    std::optional<std::string> uniform;
-    if (getShaderLanguage(BackendTest::sBackend) != ShaderLanguage::GLSL) {
-        // TODO: If any shaders need backend/shader language specific shaders rather than transpiled
-        // versions of the GLSL shader, check environment.
-    }
-    vertex = GetGlslVertexShader(request.mVertexType);
-    fragment = GetGlslFragmentShader(request.mFragmentType);
-    uniform = GetGlslUniform(request.mUniformType);
-    if (vertex.has_value() && fragment.has_value() && uniform.has_value()) {
-        return Shader(
-                api, cleanup, ShaderConfig{
-                        vertex->withUniform(*uniform), fragment->withUniform(*uniform),
-                        GetUniformConfig(request.mUniformType)}
-        );
-    }
-    abort();
-}
-
-std::string SharedShaders::getVertexShaderText(VertexShaderType vertex, ShaderUniformType uniform) {
-    std::optional<ShaderText> vertexText = GetGlslVertexShader(vertex);
-    std::optional<std::string> uniformText = GetGlslUniform(uniform);
-    if (!vertexText.has_value() || !uniformText.has_value()) {
-        abort();
-    }
-    return vertexText->withUniform(*uniformText);
-}
-
-std::string SharedShaders::getFragmentShaderText(FragmentShaderType fragment,
-        ShaderUniformType uniform) {
-    std::optional<ShaderText> fragmentText = GetGlslFragmentShader(fragment);
-    std::optional<std::string> uniformText = GetGlslUniform(uniform);
-    if (!fragmentText.has_value() || !uniformText.has_value()) {
-        abort();
-    }
-    return fragmentText->withUniform(*uniformText);
-}
-
-} // namespace test

filament/backend/test/SharedShaders.h

@@ -1,50 +0,0 @@
-/*
- * Copyright (C) 2021 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef TNT_SHAREDSHADERS_H
-#define TNT_SHAREDSHADERS_H
-
-#include "Shader.h"
-#include "SharedShadersConstants.h"
-#include "Lifetimes.h"
-#include "PlatformRunner.h"
-
-namespace test {
-
-enum class ShaderLanguage : uint8_t {
-    GLSL,
-    MSL,
-    WGSL
-};
-
-struct ShaderRequest {
-    VertexShaderType mVertexType;
-    FragmentShaderType mFragmentType;
-    ShaderUniformType mUniformType;
-};
-
-class SharedShaders {
-public:
-    static Shader makeShader(filament::backend::DriverApi& api, Cleanup& cleanup,
-            ShaderRequest request);
-    static std::string getVertexShaderText(VertexShaderType vertex, ShaderUniformType uniform);
-    static std::string getFragmentShaderText(FragmentShaderType fragment,
-            ShaderUniformType uniform);
-};
-
-} // namespace test
-
-#endif //TNT_SHAREDSHADERS_H

filament/backend/test/SharedShadersConstants.h

@@ -1,63 +0,0 @@
-/*
- * Copyright (C) 2021 The Android Open Source Project
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- *      http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifndef TNT_SHAREDSHADERSCONSTANTS_H
-#define TNT_SHAREDSHADERSCONSTANTS_H
-
-#include "math/mathfwd.h"
-
-enum class ShaderUniformType : uint8_t {
-    None,
-    Simple,
-    SimpleWithPadding,
-    Sampler,
-};
-
-struct SimpleMaterialParams {
-    filament::math::float4 color;
-    // 1.0 will be added to this value before use.
-    // The XY values are used to scale position inputs and the Z value is used to set the output
-    // position's Z value.
-    filament::math::float4 scaleMinusOne;
-    // Offset will be applied after scale
-    filament::math::float4 offset;
-};
-
-struct SimpleWithPaddingMaterialParams {
-    // The associated uniform structure in the shader will have 64 bytes of padding at the beginning
-    // So users of this struct will need to add 64 bytes to its size and offset all uniform writes.
-    filament::math::float4 color;
-    // 1.0 will be added to this value before use.
-    // The XY values are used to scale position inputs and the Z value is used to set the output
-    // position's Z value.
-    filament::math::float4 scaleMinusOne;
-    // Offset will be applied after scale
-    filament::math::float4 offset;
-};
-
-enum class VertexShaderType : uint8_t {
-    Noop,
-    Simple,
-    Textured
-};
-
-enum class FragmentShaderType : uint8_t {
-    White,
-    SolidColored,
-    Textured
-};
-
-#endif //TNT_SHAREDSHADERSCONSTANTS_H

filament/backend/test/test_Blit.cpp

@@ -19,7 +19,6 @@
 #include "ImageExpectations.h"
 #include "Lifetimes.h"
 #include "Shader.h"
-#include "SharedShaders.h"
 #include "TrianglePrimitive.h"
 
 #include <utils/Hash.h>
@@ -41,14 +40,48 @@ using namespace filament::backend;
 using namespace filament::math;
 using namespace utils;
 
+struct MaterialParams {
+    float4 color;
+    float4 scale;
+};
+
 class BlitTest : public BackendTest {
 public:
     BlitTest() : mCleanup(getDriverApi()) {}
 
 protected:
+    Shader createShader();
+
     Cleanup mCleanup;
 };
 
+static const char* const triangleVs = R"(#version 450 core
+layout(location = 0) in vec4 mesh_position;
+layout(binding = 0, set = 1) uniform Params { highp vec4 color; highp vec4 scale; } params;
+void main() {
+    gl_Position = vec4((mesh_position.xy + 0.5) * params.scale.xy, params.scale.z, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
+})";
+
+static const char* const triangleFs = R"(#version 450 core
+precision mediump int; precision highp float;
+layout(location = 0) out vec4 fragColor;
+layout(binding = 0, set = 1) uniform Params { highp vec4 color; highp vec4 scale; } params;
+void main() {
+    fragColor = params.color;
+})";
+
+Shader BlitTest::createShader() {
+    return Shader(getDriverApi(), mCleanup, ShaderConfig{
+            .vertexShader = triangleVs,
+            .fragmentShader = triangleFs,
+            .uniformNames = { "Params" },
+    });
+}
+
 static uint32_t toUintColor(float4 color) {
     color = saturate(color);
     uint32_t r = color.r * 255.0f;
@@ -280,11 +313,7 @@ TEST_F(BlitTest, ColorResolve) {
     constexpr auto kColorTexFormat = TextureFormat::RGBA8;
     constexpr int kSampleCount = 4;
 
-    Shader shader = SharedShaders::makeShader(api, mCleanup, ShaderRequest{
-            .mVertexType = VertexShaderType::Simple,
-            .mFragmentType = FragmentShaderType::SolidColored,
-            .mUniformType = ShaderUniformType::Simple,
-    });
+    Shader shader = createShader();
 
     // Create a VertexBuffer, IndexBuffer, and RenderPrimitive.
     TrianglePrimitive const triangle(api);
@@ -327,15 +356,14 @@ TEST_F(BlitTest, ColorResolve) {
     state.rasterState.depthFunc = RasterState::DepthFunc::A;
     state.rasterState.culling = CullingMode::NONE;
 
-    auto ubuffer = mCleanup.add(api.createBufferObject(sizeof(SimpleMaterialParams),
+    auto ubuffer = mCleanup.add(api.createBufferObject(sizeof(MaterialParams),
             BufferObjectBinding::UNIFORM, BufferUsage::STATIC));
     // Draw red triangle into srcRenderTarget.
-    shader.uploadUniform(api, ubuffer, SimpleMaterialParams{
-        .color = float4(1, 0, 0, 1),
-        .scaleMinusOne = float4(0, 0, -0.5, 0),
-        .offset = float4(0.5, 0.5, 0, 0),
+    shader.uploadUniform(api, ubuffer, MaterialParams{
+            .color = float4(1, 0, 0, 1),
+            .scale = float4(1, 1, 0.5, 0),
     });
-    shader.bindUniform<SimpleMaterialParams>(api, ubuffer);
+    shader.bindUniform<MaterialParams>(api, ubuffer);
 
     // FIXME: on Metal this triangle is not drawn. Can't understand why.
     {

filament/backend/test/test_BufferUpdates.cpp

@@ -18,18 +18,72 @@
 
 #include "Lifetimes.h"
 #include "Shader.h"
-#include "SharedShaders.h"
+#include "ShaderGenerator.h"
 #include "TrianglePrimitive.h"
 
+namespace {
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+// Shaders
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+std::string vertex (R"(#version 450 core
+
+layout(location = 0) in vec4 mesh_position;
+
+layout(location = 0) out uvec4 indices;
+
+layout(binding = 0, set = 1) uniform Params {
+    highp vec4 padding[4];  // offset of 64 bytes
+
+    highp vec4 color;
+    highp vec4 offset;
+} params;
+
+void main() {
+    gl_Position = vec4(mesh_position.xy + params.offset.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
+}
+)");
+
+std::string fragment (R"(#version 450 core
+
+layout(location = 0) out vec4 fragColor;
+
+layout(binding = 0, set = 1) uniform Params {
+    highp vec4 padding[4];  // offset of 64 bytes
+
+    highp vec4 color;
+    highp vec4 offset;
+} params;
+
+void main() {
+    fragColor = vec4(params.color.rgb, 1.0f);
+}
+
+)");
+
+}
+
 namespace test {
 
 using namespace filament;
 using namespace filament::backend;
 
+// In the shader, these MaterialParams are offset by 64 bytes into the uniform buffer to test buffer
+// updates with offset.
+struct MaterialParams {
+    math::float4 color;
+    math::float4 offset;
+};
+static_assert(sizeof(MaterialParams) == 8 * sizeof(float));
 // Uniform config for writing MaterialParams to the shader uniform with 64 bytes of padding.
 const UniformBindingConfig kBindingConfig = {
-        .dataSize = sizeof(SimpleMaterialParams),
-        .bufferSize = sizeof(SimpleMaterialParams) + 64,
+        .dataSize = sizeof(MaterialParams),
+        .bufferSize = sizeof(MaterialParams) + 64,
         .byteOffset = 64
 };
 
@@ -39,10 +93,8 @@ public:
 
 protected:
     Shader createShader() {
-        return SharedShaders::makeShader(getDriverApi(), mCleanup, ShaderRequest{
-                .mVertexType = VertexShaderType::Simple,
-                .mFragmentType = FragmentShaderType::SolidColored,
-                .mUniformType = ShaderUniformType::SimpleWithPadding
+        return Shader(getDriverApi(), mCleanup, ShaderConfig{
+           vertex, fragment, {"Params"}
         });
     }
 
@@ -77,7 +129,7 @@ TEST_F(BufferUpdatesTest, VertexBufferUpdate) {
         RenderPassParams params = {};
         fullViewport(params);
         params.flags.clear = TargetBufferFlags::COLOR;
-        params.clearColor = { 0.f, 1.f, 0.f, 1.f };
+        params.clearColor = {0.f, 1.f, 0.f, 1.f};
         params.flags.discardStart = TargetBufferFlags::ALL;
         params.flags.discardEnd = TargetBufferFlags::NONE;
 
@@ -92,18 +144,17 @@ TEST_F(BufferUpdatesTest, VertexBufferUpdate) {
         // Create a uniform buffer.
         // We use STATIC here, even though the buffer is updated, to force the Metal backend to use
         // a GPU buffer, which is more interesting to test.
-        auto ubuffer = cleanup.add(api.createBufferObject(sizeof(SimpleMaterialParams) + 64,
+        auto ubuffer = cleanup.add(api.createBufferObject(sizeof(MaterialParams) + 64,
                 BufferObjectBinding::UNIFORM, BufferUsage::STATIC));
 
-        shader.bindUniform<SimpleMaterialParams>(api, ubuffer, kBindingConfig);
+        shader.bindUniform<MaterialParams>(api, ubuffer, kBindingConfig);
 
         api.startCapture(0);
 
         // Upload the uniform, but with an offset to accommodate the padding in the shader's
         // uniform definition.
-        shader.uploadUniform(api, ubuffer, kBindingConfig, SimpleMaterialParams{
+        shader.uploadUniform(api, ubuffer, kBindingConfig, MaterialParams{
                 .color = { 1.0f, 1.0f, 1.0f, 1.0f },
-                .scaleMinusOne = { 0.0, 0.0, 0.0, 0.0 },
                 .offset = { 0.0f, 0.0f, 0.0f, 0.0f }
         });
 
@@ -114,21 +165,19 @@ TEST_F(BufferUpdatesTest, VertexBufferUpdate) {
         size_t triangleIndex = 0;
         for (float i = -1.0f; i < 1.0f; i += 0.2f) {
             const float low = i, high = i + 0.2;
-            const filament::math::float2 v[3]{{ low,  low },
-                                              { high, low },
-                                              { low,  high }};
+            const filament::math::float2 v[3] {{low, low}, {high, low}, {low, high}};
             triangle.updateVertices(v);
 
             if (updateIndices) {
                 if (triangleIndex % 2 == 0) {
                     // Upload each index separately, to test offsets.
-                    const TrianglePrimitive::index_type i[3]{ 0, 1, 2 };
+                    const TrianglePrimitive::index_type i[3] {0, 1, 2};
                     triangle.updateIndices(i + 0, 1, 0);
                     triangle.updateIndices(i + 1, 1, 1);
                     triangle.updateIndices(i + 2, 1, 2);
                 } else {
                     // This effectively hides this triangle.
-                    const TrianglePrimitive::index_type i[3]{ 0, 0, 0 };
+                    const TrianglePrimitive::index_type i[3] {0, 0, 0};
                     triangle.updateIndices(i);
                 }
             }
@@ -171,29 +220,28 @@ TEST_F(BufferUpdatesTest, BufferObjectUpdateWithOffset) {
     // Create a uniform buffer.
     // We use STATIC here, even though the buffer is updated, to force the Metal backend to use a
     // GPU buffer, which is more interesting to test.
-    auto ubuffer = cleanup.add(api.createBufferObject(sizeof(SimpleMaterialParams) + 64,
+    auto ubuffer = cleanup.add(api.createBufferObject(sizeof(MaterialParams) + 64,
             BufferObjectBinding::UNIFORM, BufferUsage::STATIC));
 
-    shader.bindUniform<SimpleMaterialParams>(api, ubuffer, kBindingConfig);
+    shader.bindUniform<MaterialParams>(api, ubuffer, kBindingConfig);
 
     // Create a render target.
     auto colorTexture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, 1,
             TextureFormat::RGBA8, 1, 512, 512, 1, TextureUsage::COLOR_ATTACHMENT));
     auto renderTarget = cleanup.add(api.createRenderTarget(
-            TargetBufferFlags::COLOR0, 512, 512, 1, 0, {{ colorTexture }}, {}, {}));
+            TargetBufferFlags::COLOR0, 512, 512, 1, 0, {{colorTexture}}, {}, {}));
 
     // Upload uniforms for the first triangle.
     // Upload the uniform, but with an offset to accommodate the padding in the shader's
     // uniform definition.
-    shader.uploadUniform(api, ubuffer, kBindingConfig, SimpleMaterialParams{
+    shader.uploadUniform(api, ubuffer, kBindingConfig, MaterialParams{
             .color = { 1.0f, 0.0f, 0.5f, 1.0f },
-            .scaleMinusOne = { 0.0f, 0.0f, 0.0f, 0.0f },
             .offset = { 0.0f, 0.0f, 0.0f, 0.0f }
     });
 
     RenderPassParams params = {};
     params.flags.clear = TargetBufferFlags::COLOR;
-    params.clearColor = { 0.f, 0.f, 1.f, 1.f };
+    params.clearColor = {0.f, 0.f, 1.f, 1.f};
     params.flags.discardStart = TargetBufferFlags::ALL;
     params.flags.discardEnd = TargetBufferFlags::NONE;
     params.viewport.height = 512;
@@ -202,18 +250,18 @@ TEST_F(BufferUpdatesTest, BufferObjectUpdateWithOffset) {
             renderTarget, swapChain, shader.getProgram(), params);
 
     // Upload uniforms for the second triangle. To test partial buffer updates, we'll only update
-    // color.b, color.a, scaleMinusOne, offset.x, and offset.y.
-    const UniformBindingConfig partialBindingConfig = {
-            .dataSize = sizeof(float) * 8,
-            .bufferSize = sizeof(SimpleMaterialParams) + 64,
-            .byteOffset = 64 + offsetof(SimpleMaterialParams, color.b)
-    };
-    shader.uploadUniform(api, ubuffer, partialBindingConfig,
-            std::array<float, 8>{
-                    1.0f, 1.0f, // color.b, color.a
-                    0.0f, 0.0f, 0.0f, 0.0f, // scale
-                    0.5f, 0.5f // offset.x, offset.y
-            });
+    // color.b, color.a, offset.x, and offset.y.
+    shader.uploadUniform(api, ubuffer, UniformBindingConfig{
+                    .dataSize = sizeof(std::array<float, 4>),
+                    .bufferSize = kBindingConfig.bufferSize,
+                    .byteOffset = *kBindingConfig.byteOffset + offsetof(MaterialParams, color.b),
+            },
+            std::array<float, 4>{
+                    // color.b, color.a
+                    1.0f, 1.0f,
+                    // offset.x, offset.y
+                    0.5f, 0.5f }
+    );
 
     params.flags.clear = TargetBufferFlags::NONE;
     params.flags.discardStart = TargetBufferFlags::NONE;

filament/backend/test/test_FeedbackLoops.cpp

@@ -17,7 +17,7 @@
 #include "BackendTest.h"
 
 #include "Lifetimes.h"
-#include "Shader.h"
+#include "ShaderGenerator.h"
 #include "TrianglePrimitive.h"
 
 #include <backend/DriverEnums.h>
@@ -96,6 +96,16 @@ struct MaterialParams {
     float unused;
 };
 
+static void uploadUniforms(DriverApi& dapi, Handle<HwBufferObject> ubh, MaterialParams params) {
+    MaterialParams* tmp = new MaterialParams(params);
+    auto cb = [](void* buffer, size_t size, void* user) {
+        MaterialParams* sp = (MaterialParams*) buffer;
+        delete sp;
+    };
+    BufferDescriptor bd(tmp, sizeof(MaterialParams), cb);
+    dapi.updateBufferObject(ubh, std::move(bd), 0);
+}
+
 static void dumpScreenshot(DriverApi& dapi, Handle<HwRenderTarget> rt) {
     const size_t size = kTexWidth * kTexHeight * 4;
     void* buffer = calloc(1, size);
@@ -130,13 +140,38 @@ TEST_F(BackendTest, FeedbackLoops) {
         api.makeCurrent(swapChain, swapChain);
 
         // Create a program.
-        filament::SamplerInterfaceBlock::SamplerInfo samplerInfo { "test", "tex", 0,
+        ProgramHandle program;
+        {
+            filament::SamplerInterfaceBlock::SamplerInfo samplerInfo { "test", "tex", 0,
                 SamplerType::SAMPLER_2D, SamplerFormat::FLOAT, Precision::HIGH, false };
-        Shader shader = Shader(api, cleanup, ShaderConfig {
-            .vertexShader = fullscreenVs,
-            .fragmentShader = fullscreenFs,
-            .uniforms = {{"test_tex", DescriptorType::SAMPLER, samplerInfo}, {"Params"}}
-        });
+            filamat::DescriptorSets descriptors;
+            descriptors[1] = {
+                { "test_tex", { DescriptorType::SAMPLER, ShaderStageFlags::FRAGMENT, 0 },
+                        samplerInfo },
+                { "Params", { DescriptorType::UNIFORM_BUFFER, ShaderStageFlags::FRAGMENT, 1 }, {} }
+            };
+            ShaderGenerator shaderGen(fullscreenVs, fullscreenFs, sBackend, sIsMobilePlatform,
+                    std::move(descriptors));
+            Program prog = shaderGen.getProgram(api);
+            prog.descriptorBindings(1, {
+                    { "test_tex", DescriptorType::SAMPLER, 0 },
+                    { "Params", DescriptorType::UNIFORM_BUFFER, 1 }
+            });
+            program = cleanup.add(api.createProgram(std::move(prog)));
+        }
+
+        DescriptorSetLayoutHandle descriptorSetLayout = cleanup.add(api.createDescriptorSetLayout({
+                {{
+                         DescriptorType::SAMPLER,
+                         ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, 0,
+                         DescriptorFlags::NONE, 0
+                 },
+                 {
+                         DescriptorType::UNIFORM_BUFFER,
+                         ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, 1,
+                         DescriptorFlags::NONE, 0
+                 }}}));
+
 
         TrianglePrimitive const triangle(getDriverApi());
 
@@ -183,8 +218,8 @@ TEST_F(BackendTest, FeedbackLoops) {
             state.rasterState.colorWrite = true;
             state.rasterState.depthWrite = false;
             state.rasterState.depthFunc = RasterState::DepthFunc::A;
-            state.program = shader.getProgram();
-            state.pipelineLayout.setLayout[1] = { shader.getDescriptorSetLayout() };
+            state.program = program;
+            state.pipelineLayout.setLayout[1] = { descriptorSetLayout };
 
             api.makeCurrent(swapChain, swapChain);
             api.beginFrame(0, 0, 0);
@@ -198,24 +233,20 @@ TEST_F(BackendTest, FeedbackLoops) {
                 params.viewport.width = kTexWidth >> targetLevel;
                 params.viewport.height = kTexHeight >> targetLevel;
 
-                auto descriptorSet = shader.createDescriptorSet(api);
                 auto textureView = passCleanup.add(api.createTextureView(texture, sourceLevel, 1));
+                DescriptorSetHandle descriptorSet = passCleanup.add(api.createDescriptorSet(descriptorSetLayout));
                 api.updateDescriptorSetTexture(descriptorSet, 0, textureView, {
                         .filterMag = SamplerMagFilter::LINEAR,
                         .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
                 });
+                api.updateDescriptorSetBuffer(descriptorSet, 1, ubuffer, 0, sizeof(MaterialParams));
+                api.bindDescriptorSet(descriptorSet, 1, {});
 
-                UniformBindingConfig uniformBinding{
-                    .binding = 1,
-                    .descriptorSet = descriptorSet
-                };
-                shader.bindUniform<MaterialParams>(api, ubuffer, uniformBinding);
-                shader.uploadUniform(api, ubuffer, uniformBinding, MaterialParams{
+                uploadUniforms(getDriverApi(), ubuffer, {
                     .fbWidth = float(params.viewport.width),
                     .fbHeight = float(params.viewport.height),
                     .sourceLevel = float(sourceLevel),
                 });
-
                 api.beginRenderPass(renderTargets[targetLevel], params);
                 api.draw(state, triangle.getRenderPrimitive(), 0, 3, 1);
                 api.endRenderPass();
@@ -231,24 +262,20 @@ TEST_F(BackendTest, FeedbackLoops) {
                 params.viewport.width = kTexWidth >> targetLevel;
                 params.viewport.height = kTexHeight >> targetLevel;
 
-                auto descriptorSet = shader.createDescriptorSet(api);
                 auto textureView = passCleanup.add(api.createTextureView(texture, sourceLevel, 1));
+                DescriptorSetHandle descriptorSet = passCleanup.add(api.createDescriptorSet(descriptorSetLayout));
                 api.updateDescriptorSetTexture(descriptorSet, 0, textureView, {
                         .filterMag = SamplerMagFilter::LINEAR,
                         .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
                 });
+                api.updateDescriptorSetBuffer(descriptorSet, 1, ubuffer, 0, sizeof(MaterialParams));
+                api.bindDescriptorSet(descriptorSet, 1, {});
 
-                UniformBindingConfig uniformBinding{
-                        .binding = 1,
-                        .descriptorSet = descriptorSet
-                };
-                shader.bindUniform<MaterialParams>(api, ubuffer, uniformBinding);
-                shader.uploadUniform(api, ubuffer, uniformBinding, MaterialParams{
-                        .fbWidth = float(params.viewport.width),
-                        .fbHeight = float(params.viewport.height),
-                        .sourceLevel = float(sourceLevel),
+                uploadUniforms(getDriverApi(), ubuffer, {
+                    .fbWidth = float(params.viewport.width),
+                    .fbHeight = float(params.viewport.height),
+                    .sourceLevel = float(sourceLevel),
                 });
-
                 api.beginRenderPass(renderTargets[targetLevel], params);
                 api.draw(state, triangle.getRenderPrimitive(), 0, 3, 1);
                 api.endRenderPass();

filament/backend/test/test_LoadImage.cpp

@@ -18,14 +18,16 @@
 
 #include "BackendTestUtils.h"
 #include "Lifetimes.h"
-#include "Shader.h"
-#include "SharedShaders.h"
+#include "ShaderGenerator.h"
+#include "TrianglePrimitive.h"
 
 #include <backend/DriverEnums.h>
 #include <backend/Handle.h>
 
 #include "private/filament/SamplerInterfaceBlock.h"
 
+#include <math/half.h>
+
 #include <vector>
 
 #include <stddef.h>
@@ -40,6 +42,19 @@ namespace {
 // Shaders
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
+std::string vertex (R"(#version 450 core
+
+layout(location = 0) in vec4 mesh_position;
+
+void main() {
+    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y *= -1.0f;
+#endif
+}
+)");
+
 std::string fragmentTemplate (R"(#version 450 core
 
 layout(location = 0) out vec4 fragColor;
@@ -104,15 +119,6 @@ namespace test {
 
 template<typename componentType> inline componentType getMaxValue();
 
-class LoadImageTest : public BackendTest {
-public:
-    LoadImageTest() {
-        mVertexShader = SharedShaders::getVertexShaderText(VertexShaderType::Noop,
-                ShaderUniformType::None);
-    }
-
-    std::string mVertexShader;
-};
 
 
 inline std::string stringReplace(const std::string& find, const std::string& replace,
@@ -210,7 +216,7 @@ static SamplerFormat getSamplerFormat(TextureFormat textureFormat) {
     }
 }
 
-TEST_F(LoadImageTest, UpdateImage2D) {
+TEST_F(BackendTest, UpdateImage2D) {
 
     // All of these test cases should result in the same rendered image, and thus the same hash.
     static const uint32_t expectedHash = 3644679986;
@@ -300,14 +306,28 @@ TEST_F(LoadImageTest, UpdateImage2D) {
         // Create a program.
         filament::SamplerInterfaceBlock::SamplerInfo samplerInfo { "test", "tex", 0,
             SamplerType::SAMPLER_2D, getSamplerFormat(t.textureFormat), Precision::HIGH, false };
+        filamat::DescriptorSets descriptors;
+        descriptors[1] = { { "test_tex", { DescriptorType::SAMPLER, ShaderStageFlags::FRAGMENT, 0 },
+                samplerInfo } };
 
         std::string const fragment = stringReplace("{samplerType}",
                 getSamplerTypeName(t.textureFormat), fragmentTemplate);
-        Shader shader(api, cleanup, ShaderConfig{
-           .vertexShader = mVertexShader,
-           .fragmentShader= fragment,
-           .uniforms = {{"test_tex", DescriptorType::SAMPLER, samplerInfo}}
-        });
+        ShaderGenerator shaderGen(
+                vertex, fragment, sBackend, sIsMobilePlatform, std::move(descriptors));
+
+        Program prog = shaderGen.getProgram(api);
+        prog.descriptorBindings(1, {{"test_tex", DescriptorType::SAMPLER, 0}});
+
+        ProgramHandle const program = cleanup.add(api.createProgram(std::move(prog)));
+
+        DescriptorSetLayoutHandle descriptorSetLayout = cleanup.add(api.createDescriptorSetLayout({
+                {{
+                         DescriptorType::SAMPLER,
+                         ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, 0,
+                         DescriptorFlags::NONE, 0
+                 }}}));
+
+        DescriptorSetHandle descriptorSet = cleanup.add(api.createDescriptorSet(descriptorSetLayout));
 
         // Create a Texture.
         auto usage = TextureUsage::SAMPLEABLE | TextureUsage::UPLOADABLE;
@@ -329,15 +349,14 @@ TEST_F(LoadImageTest, UpdateImage2D) {
                     checkerboardPixelBuffer(t.pixelFormat, t.pixelType, 512, t.bufferPadding));
         }
 
-        DescriptorSetHandle  descriptorSet = shader.createDescriptorSet(api);
         api.updateDescriptorSetTexture(descriptorSet, 0, texture, {
                 .filterMag = SamplerMagFilter::NEAREST,
                 .filterMin = SamplerMinFilter::NEAREST_MIPMAP_NEAREST });
 
         api.bindDescriptorSet(descriptorSet, 1, {});
 
-        renderTriangle({{ DescriptorSetLayoutHandle{}, shader.getDescriptorSetLayout() }},
-                defaultRenderTarget, swapChain, shader.getProgram());
+        renderTriangle({{ DescriptorSetLayoutHandle{}, descriptorSetLayout }},
+                defaultRenderTarget, swapChain, program);
 
         readPixelsAndAssertHash(t.name, 512, 512, defaultRenderTarget, expectedHash);
 
@@ -351,9 +370,8 @@ TEST_F(LoadImageTest, UpdateImage2D) {
     flushAndWait();
 }
 
-TEST_F(LoadImageTest, UpdateImageSRGB) {
+TEST_F(BackendTest, UpdateImageSRGB) {
     auto& api = getDriverApi();
-    Cleanup cleanup(api);
     api.startCapture();
 
     PixelDataFormat const pixelFormat = PixelDataFormat::RGBA;
@@ -361,23 +379,36 @@ TEST_F(LoadImageTest, UpdateImageSRGB) {
     TextureFormat const textureFormat = TextureFormat::SRGB8_A8;
 
     // Create a platform-specific SwapChain and make it current.
-    auto swapChain = cleanup.add(createSwapChain());
+    auto swapChain = createSwapChain();
     api.makeCurrent(swapChain, swapChain);
-    auto defaultRenderTarget = cleanup.add(api.createDefaultRenderTarget(0));
+    auto defaultRenderTarget = api.createDefaultRenderTarget(0);
 
     // Create a program.
     filament::SamplerInterfaceBlock::SamplerInfo samplerInfo { "test", "tex", 0,
         SamplerType::SAMPLER_2D, getSamplerFormat(textureFormat), Precision::HIGH, false };
+    filamat::DescriptorSets descriptors;
+    descriptors[1] = { { "test_tex", { DescriptorType::SAMPLER, ShaderStageFlags::FRAGMENT, 0 },
+            samplerInfo } };
+
     std::string const fragment = stringReplace("{samplerType}",
             getSamplerTypeName(textureFormat), fragmentTemplate);
-    Shader shader(api, cleanup, ShaderConfig{
-        .vertexShader = mVertexShader, .fragmentShader = fragment, .uniforms = {{
-            "text_tex", DescriptorType::SAMPLER, samplerInfo
-    }}});
+    ShaderGenerator shaderGen(
+            vertex, fragment, sBackend, sIsMobilePlatform, std::move(descriptors));
+    Program prog = shaderGen.getProgram(api);
+    prog.descriptorBindings(1, {{"test_tex", DescriptorType::SAMPLER, 0}});
+    ProgramHandle const program = api.createProgram(std::move(prog));
+    DescriptorSetLayoutHandle descriptorSetLayout = api.createDescriptorSetLayout({
+            {{
+                     DescriptorType::SAMPLER,
+                     ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, 0,
+                     DescriptorFlags::NONE, 0
+             }}});
+
+    DescriptorSetHandle descriptorSet = api.createDescriptorSet(descriptorSetLayout);
 
     // Create a texture.
-    Handle<HwTexture> const texture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, 1,
-            textureFormat, 1, 512, 512, 1, TextureUsage::SAMPLEABLE | TextureUsage::UPLOADABLE));
+    Handle<HwTexture> const texture = api.createTexture(SamplerType::SAMPLER_2D, 1,
+            textureFormat, 1, 512, 512, 1, TextureUsage::SAMPLEABLE | TextureUsage::UPLOADABLE);
 
     // Create image data.
     size_t components; int bpp;
@@ -405,7 +436,6 @@ TEST_F(LoadImageTest, UpdateImageSRGB) {
     api.beginFrame(0, 0, 0);
 
     // Update samplers.
-    DescriptorSetHandle descriptorSet = shader.createDescriptorSet(api);
     api.updateDescriptorSetTexture(descriptorSet, 0, texture, {
             .filterMag = SamplerMagFilter::LINEAR,
             .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
@@ -413,8 +443,8 @@ TEST_F(LoadImageTest, UpdateImageSRGB) {
 
     api.bindDescriptorSet(descriptorSet, 1, {});
 
-    renderTriangle({{ DescriptorSetLayoutHandle{}, shader.getDescriptorSetLayout() }},
-            defaultRenderTarget, swapChain, shader.getProgram());
+    renderTriangle({{ DescriptorSetLayoutHandle{}, descriptorSetLayout }},
+            defaultRenderTarget, swapChain, program);
 
     static const uint32_t expectedHash = 359858623;
     readPixelsAndAssertHash("UpdateImageSRGB", 512, 512, defaultRenderTarget, expectedHash);
@@ -423,14 +453,21 @@ TEST_F(LoadImageTest, UpdateImageSRGB) {
     api.commit(swapChain);
     api.endFrame(0);
 
+    api.destroyDescriptorSet(descriptorSet);
+    api.destroyDescriptorSetLayout(descriptorSetLayout);
+    api.destroyProgram(program);
+    api.destroySwapChain(swapChain);
+    api.destroyRenderTarget(defaultRenderTarget);
+
     // This ensures all driver commands have finished before exiting the test.
     api.finish();
     api.stopCapture();
+
+    flushAndWait();
 }
 
-TEST_F(LoadImageTest, UpdateImageMipLevel) {
+TEST_F(BackendTest, UpdateImageMipLevel) {
     auto& api = getDriverApi();
-    Cleanup cleanup(api);
     api.startCapture();
 
     PixelDataFormat pixelFormat = PixelDataFormat::RGBA;
@@ -438,27 +475,38 @@ TEST_F(LoadImageTest, UpdateImageMipLevel) {
     TextureFormat textureFormat = TextureFormat::RGBA32F;
 
     // Create a platform-specific SwapChain and make it current.
-    auto swapChain = cleanup.add(createSwapChain());
+    auto swapChain = createSwapChain();
     api.makeCurrent(swapChain, swapChain);
-    auto defaultRenderTarget = cleanup.add(api.createDefaultRenderTarget(0));
+    auto defaultRenderTarget = api.createDefaultRenderTarget(0);
 
     // Create a program.
     filament::SamplerInterfaceBlock::SamplerInfo samplerInfo { "test", "tex", 0,
         SamplerType::SAMPLER_2D, getSamplerFormat(textureFormat), Precision::HIGH, false };
+    filamat::DescriptorSets descriptors;
+    descriptors[1] = { { "test_tex", { DescriptorType::SAMPLER, ShaderStageFlags::FRAGMENT, 0 },
+            samplerInfo } };
     std::string const fragment = stringReplace("{samplerType}",
             getSamplerTypeName(textureFormat), fragmentUpdateImageMip);
-    Shader shader(api, cleanup, ShaderConfig {
-        .vertexShader = mVertexShader,
-        .fragmentShader = fragment,
-        .uniforms = {{"test_tex", DescriptorType::SAMPLER, samplerInfo}}
-    });
+    ShaderGenerator shaderGen(
+            vertex, fragment, sBackend, sIsMobilePlatform, std::move(descriptors));
+    Program prog = shaderGen.getProgram(api);
+    prog.descriptorBindings(1, {{"test_tex", DescriptorType::SAMPLER, 0}});
+    ProgramHandle const program = api.createProgram(std::move(prog));
+    DescriptorSetLayoutHandle descriptorSetLayout = api.createDescriptorSetLayout({
+            {{
+                     DescriptorType::SAMPLER,
+                     ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, 0,
+                     DescriptorFlags::NONE, 0
+             }}});
+
+    DescriptorSetHandle descriptorSet = api.createDescriptorSet(descriptorSetLayout);
 
     // Create a texture with 3 mip levels.
     // Base level: 1024
     // Level 1:     512     <-- upload data and sample from this level
     // Level 2:     256
-    Handle<HwTexture> texture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, 3,
-            textureFormat, 1, 1024, 1024, 1, TextureUsage::SAMPLEABLE | TextureUsage::UPLOADABLE));
+    Handle<HwTexture> texture = api.createTexture(SamplerType::SAMPLER_2D, 3,
+            textureFormat, 1, 1024, 1024, 1, TextureUsage::SAMPLEABLE | TextureUsage::UPLOADABLE);
 
     // Create image data.
     PixelBufferDescriptor descriptor = checkerboardPixelBuffer(pixelFormat, pixelType, 512);
@@ -467,7 +515,6 @@ TEST_F(LoadImageTest, UpdateImageMipLevel) {
     api.beginFrame(0, 0, 0);
 
     // Update samplers.
-    DescriptorSetHandle descriptorSet = shader.createDescriptorSet(api);
     api.updateDescriptorSetTexture(descriptorSet, 0, texture, {
             .filterMag = SamplerMagFilter::LINEAR,
             .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
@@ -475,8 +522,8 @@ TEST_F(LoadImageTest, UpdateImageMipLevel) {
 
     api.bindDescriptorSet(descriptorSet, 1, {});
 
-    renderTriangle({{ DescriptorSetLayoutHandle{}, shader.getDescriptorSetLayout() }},
-            defaultRenderTarget, swapChain, shader.getProgram());
+    renderTriangle({{ DescriptorSetLayoutHandle{}, descriptorSetLayout }},
+            defaultRenderTarget, swapChain, program);
 
     static const uint32_t expectedHash = 3644679986;
     readPixelsAndAssertHash("UpdateImageMipLevel", 512, 512, defaultRenderTarget, expectedHash);
@@ -485,14 +532,21 @@ TEST_F(LoadImageTest, UpdateImageMipLevel) {
     api.commit(swapChain);
     api.endFrame(0);
 
+    api.destroyDescriptorSet(descriptorSet);
+    api.destroyDescriptorSetLayout(descriptorSetLayout);
+    api.destroyProgram(program);
+    api.destroySwapChain(swapChain);
+    api.destroyRenderTarget(defaultRenderTarget);
+
     // This ensures all driver commands have finished before exiting the test.
     api.finish();
     api.stopCapture();
+
+    flushAndWait();
 }
 
-TEST_F(LoadImageTest, UpdateImage3D) {
+TEST_F(BackendTest, UpdateImage3D) {
     auto& api = getDriverApi();
-    Cleanup cleanup(api);
     api.startCapture();
 
     PixelDataFormat pixelFormat = PixelDataFormat::RGBA;
@@ -502,24 +556,35 @@ TEST_F(LoadImageTest, UpdateImage3D) {
     TextureUsage usage = TextureUsage::SAMPLEABLE | TextureUsage::UPLOADABLE;
 
     // Create a platform-specific SwapChain and make it current.
-    auto swapChain = cleanup.add(createSwapChain());
+    auto swapChain = createSwapChain();
     api.makeCurrent(swapChain, swapChain);
-    auto defaultRenderTarget = cleanup.add(api.createDefaultRenderTarget(0));
+    auto defaultRenderTarget = api.createDefaultRenderTarget(0);
 
     // Create a program.
     filament::SamplerInterfaceBlock::SamplerInfo samplerInfo { "test", "tex", 0,
         SamplerType::SAMPLER_2D_ARRAY, getSamplerFormat(textureFormat), Precision::HIGH, false };
+    filamat::DescriptorSets descriptors;
+    descriptors[1] = { { "test_tex", { DescriptorType::SAMPLER, ShaderStageFlags::FRAGMENT, 0 },
+            samplerInfo } };
     std::string fragment = stringReplace("{samplerType}",
             getSamplerTypeName(samplerType), fragmentUpdateImage3DTemplate);
-    Shader shader(api, cleanup, ShaderConfig {
-        .vertexShader = mVertexShader,
-        .fragmentShader = fragment,
-        .uniforms = {{"test_tex", DescriptorType::SAMPLER, samplerInfo}}
-    });
+    ShaderGenerator shaderGen(
+            vertex, fragment, sBackend, sIsMobilePlatform, std::move(descriptors));
+    Program prog = shaderGen.getProgram(api);
+    prog.descriptorBindings(1, {{ "test_tex", DescriptorType::SAMPLER, 0 }});
+    ProgramHandle const program = api.createProgram(std::move(prog));
+    DescriptorSetLayoutHandle descriptorSetLayout = api.createDescriptorSetLayout({
+            {{
+                     DescriptorType::SAMPLER,
+                     ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, 0,
+                     DescriptorFlags::NONE, 0
+             }}});
+
+    DescriptorSetHandle descriptorSet = api.createDescriptorSet(descriptorSetLayout);
 
     // Create a texture.
-    Handle<HwTexture> texture = cleanup.add(api.createTexture(samplerType, 1,
-            textureFormat, 1, 512, 512, 4, usage));
+    Handle<HwTexture> texture = api.createTexture(samplerType, 1,
+            textureFormat, 1, 512, 512, 4, usage);
 
     // Create image data for all 4 layers.
     size_t components; int bpp;
@@ -541,7 +606,6 @@ TEST_F(LoadImageTest, UpdateImage3D) {
     api.beginFrame(0, 0, 0);
 
     // Update samplers.
-    DescriptorSetHandle  descriptorSet = shader.createDescriptorSet(api);
     api.updateDescriptorSetTexture(descriptorSet, 0, texture, {
             .filterMag = SamplerMagFilter::LINEAR,
             .filterMin = SamplerMinFilter::LINEAR_MIPMAP_NEAREST
@@ -549,8 +613,8 @@ TEST_F(LoadImageTest, UpdateImage3D) {
 
     api.bindDescriptorSet(descriptorSet, 1, {});
 
-    renderTriangle({{ DescriptorSetLayoutHandle{}, shader.getDescriptorSetLayout() }},
-            defaultRenderTarget, swapChain, shader.getProgram());
+    renderTriangle({{ DescriptorSetLayoutHandle{}, descriptorSetLayout }},
+            defaultRenderTarget, swapChain, program);
 
     static const uint32_t expectedHash = 3644679986;
     readPixelsAndAssertHash("UpdateImage3D", 512, 512, defaultRenderTarget, expectedHash);
@@ -559,9 +623,17 @@ TEST_F(LoadImageTest, UpdateImage3D) {
     api.commit(swapChain);
     api.endFrame(0);
 
+    api.destroyDescriptorSet(descriptorSet);
+    api.destroyDescriptorSetLayout(descriptorSetLayout);
+    api.destroyProgram(program);
+    api.destroySwapChain(swapChain);
+    api.destroyRenderTarget(defaultRenderTarget);
+
     // This ensures all driver commands have finished before exiting the test.
     api.finish();
     api.stopCapture();
+
+    flushAndWait();
 }
 
 } // namespace test

filament/backend/test/test_MRT.cpp

@@ -17,8 +17,7 @@
 #include "BackendTest.h"
 
 #include "Lifetimes.h"
-#include "Shader.h"
-#include "SharedShaders.h"
+#include "ShaderGenerator.h"
 #include "TrianglePrimitive.h"
 
 namespace {
@@ -27,6 +26,21 @@ namespace {
 // Shaders
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
+std::string vertex (R"(#version 450 core
+
+layout(location = 0) in vec4 mesh_position;
+
+layout(location = 0) out uvec4 indices;
+
+void main() {
+    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
+}
+)");
+
 std::string fragment (R"(#version 450 core
 
 layout(location = 0) out vec4 fragColor;
@@ -57,12 +71,10 @@ TEST_F(BackendTest, MRT) {
         auto swapChain = cleanup.add(createSwapChain());
         api.makeCurrent(swapChain, swapChain);
 
-        Shader shader(api, cleanup, ShaderConfig{
-                .vertexShader = SharedShaders::getVertexShaderText(VertexShaderType::Noop,
-                        ShaderUniformType::None),
-                .fragmentShader = fragment,
-                .uniforms = {}
-        });
+        // Create a program.
+        ShaderGenerator shaderGen(vertex, fragment, sBackend, sIsMobilePlatform);
+        Program p = shaderGen.getProgram(api);
+        auto program = cleanup.add(api.createProgram(std::move(p)));
 
         TrianglePrimitive triangle(api);
 
@@ -110,7 +122,7 @@ TEST_F(BackendTest, MRT) {
         params.flags.discardEnd = TargetBufferFlags::NONE;
 
         PipelineState state;
-        state.program = shader.getProgram();
+        state.program = program;
         state.rasterState.colorWrite = true;
         state.rasterState.depthWrite = false;
         state.rasterState.depthFunc = RasterState::DepthFunc::A;

filament/backend/test/test_MipLevels.cpp

@@ -19,8 +19,7 @@
 
 #include "BackendTestUtils.h"
 #include "Lifetimes.h"
-#include "Shader.h"
-#include "SharedShaders.h"
+#include "ShaderGenerator.h"
 #include "TrianglePrimitive.h"
 
 #include <backend/DriverEnums.h>
@@ -35,7 +34,18 @@ namespace {
 // Shaders
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
-std::string fragmentTexturedLod (R"(#version 450 core
+std::string vertex (R"(#version 450 core
+
+layout(location = 0) in vec4 mesh_position;
+layout(location = 0) out vec2 uv;
+
+void main() {
+    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+    uv = (mesh_position.xy * 0.5 + 0.5);
+}
+)");
+
+std::string fragment (R"(#version 450 core
 
 layout(location = 0) out vec4 fragColor;
 layout(location = 0) in vec2 uv;
@@ -47,6 +57,18 @@ void main() {
 }
 )");
 
+std::string whiteFragment (R"(#version 450 core
+
+layout(location = 0) out vec4 fragColor;
+layout(location = 0) in vec2 uv;
+
+layout(location = 0, set = 1) uniform sampler2D backend_test_sib_tex;
+
+void main() {
+    fragColor = vec4(1.0);
+}
+)");
+
 }
 
 namespace test {
@@ -66,25 +88,39 @@ TEST_F(BackendTest, TextureViewLod) {
         auto swapChain = cleanup.add(createSwapChain());
         api.makeCurrent(swapChain, swapChain);
 
-        Shader whiteShader = SharedShaders::makeShader(api, cleanup, ShaderRequest {
-            .mVertexType = VertexShaderType::Textured,
-            .mFragmentType = FragmentShaderType::White,
-            .mUniformType = ShaderUniformType::Sampler
-        });
+        // Create a program that draws only white.
+        Handle<HwProgram> whiteProgram;
+        {
+            ShaderGenerator shaderGen(vertex, whiteFragment, sBackend, sIsMobilePlatform);
+            Program p = shaderGen.getProgram(api);
+            p.descriptorBindings(0, {{"backend_test_sib_tex", DescriptorType::SAMPLER, 0}});
+            whiteProgram = cleanup.add(api.createProgram(std::move(p)));
+        }
 
         // Create a program that samples a texture.
-        std::string vertexShader = SharedShaders::getVertexShaderText(
-                VertexShaderType::Textured, ShaderUniformType::Sampler);
-        filament::SamplerInterfaceBlock::SamplerInfo samplerInfo {
-            "backend_test", "sib_tex", 0,
-            SamplerType::SAMPLER_2D, SamplerFormat::FLOAT, Precision::HIGH, false };
-        Shader texturedShader(api, cleanup, ShaderConfig {
-           .vertexShader = vertexShader,
-           .fragmentShader = fragmentTexturedLod,
-           .uniforms = {{
-               "backend_test_sib_tex", DescriptorType::SAMPLER, samplerInfo
-           }}
-        });
+        Handle<HwProgram> textureProgram;
+        {
+            filament::SamplerInterfaceBlock::SamplerInfo samplerInfo { "backend_test", "sib_tex", 0,
+                SamplerType::SAMPLER_2D, SamplerFormat::FLOAT, Precision::HIGH, false };
+            filamat::DescriptorSets descriptors;
+            descriptors[1] = { { "backend_test_sib_tex",
+                    { DescriptorType::SAMPLER, ShaderStageFlags::FRAGMENT, 0 }, samplerInfo } };
+            ShaderGenerator shaderGen(vertex, fragment, sBackend, sIsMobilePlatform, std::move(descriptors));
+            Program p = shaderGen.getProgram(api);
+            p.descriptorBindings(0, {{"backend_test_sib_tex", DescriptorType::SAMPLER, 0}});
+            textureProgram = cleanup.add(api.createProgram(std::move(p)));
+        }
+
+        DescriptorSetLayoutHandle descriptorSetLayout = cleanup.add(api.createDescriptorSetLayout({
+                {{
+                         DescriptorType::SAMPLER,
+                         ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, 0,
+                         DescriptorFlags::NONE, 0
+                 }}}));
+
+        DescriptorSetHandle descriptorSet[2];
+        descriptorSet[0] = cleanup.add(api.createDescriptorSet(descriptorSetLayout));
+        descriptorSet[1] = cleanup.add(api.createDescriptorSet(descriptorSetLayout));
 
         // Create a texture that has 4 mip levels. Each level is a different color.
         // Level 0: 128x128 (red)
@@ -93,10 +129,9 @@ TEST_F(BackendTest, TextureViewLod) {
         // Level 3:   16x16 (yellow)
         const size_t kTextureSize = 128;
         const size_t kMipLevels = 4;
-        Handle<HwTexture> texture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D,
-                kMipLevels, TextureFormat::RGBA8, 1, kTextureSize, kTextureSize, 1,
-                TextureUsage::SAMPLEABLE | TextureUsage::COLOR_ATTACHMENT
-                | TextureUsage::UPLOADABLE));
+        Handle<HwTexture> texture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, kMipLevels,
+                TextureFormat::RGBA8, 1, kTextureSize, kTextureSize, 1,
+                TextureUsage::SAMPLEABLE | TextureUsage::COLOR_ATTACHMENT | TextureUsage::UPLOADABLE));
 
         // Create image data.
         auto pixelFormat = PixelDataFormat::RGBA;
@@ -144,7 +179,7 @@ TEST_F(BackendTest, TextureViewLod) {
             params.flags.discardStart = TargetBufferFlags::NONE;
             params.flags.discardEnd = TargetBufferFlags::NONE;
             PipelineState ps = {};
-            ps.program = whiteShader.getProgram();
+            ps.program = whiteProgram;
             ps.rasterState.colorWrite = true;
             ps.rasterState.depthWrite = false;
             api.beginRenderPass(renderTarget, params);
@@ -152,8 +187,7 @@ TEST_F(BackendTest, TextureViewLod) {
             api.endRenderPass();
         }
 
-        backend::Handle<HwRenderTarget> defaultRenderTarget =
-                cleanup.add(api.createDefaultRenderTarget(0));
+        backend::Handle<HwRenderTarget> defaultRenderTarget = cleanup.add(api.createDefaultRenderTarget(0));
 
         RenderPassParams params = {};
         fullViewport(params);
@@ -163,19 +197,18 @@ TEST_F(BackendTest, TextureViewLod) {
         params.flags.discardEnd = TargetBufferFlags::NONE;
 
         PipelineState state;
-        state.program = texturedShader.getProgram();
-        state.pipelineLayout.setLayout = { texturedShader.getDescriptorSetLayout() };
+        state.program = textureProgram;
+        state.pipelineLayout.setLayout = { descriptorSetLayout };
         state.rasterState.colorWrite = true;
         state.rasterState.depthWrite = false;
         state.rasterState.depthFunc = SamplerCompareFunc::A;
         state.rasterState.culling = CullingMode::NONE;
 
-        DescriptorSetHandle descriptorSet13 = texturedShader.createDescriptorSet(api);
-        api.updateDescriptorSetTexture(descriptorSet13, 0, texture13, {
+        api.updateDescriptorSetTexture(descriptorSet[0], 0, texture13, {
                 .filterMag = SamplerMagFilter::NEAREST,
                 .filterMin = SamplerMinFilter::NEAREST_MIPMAP_NEAREST });
 
-        api.bindDescriptorSet(descriptorSet13, 0, {});
+        api.bindDescriptorSet(descriptorSet[0], 0, {});
 
         // Render a triangle to the screen, sampling from mip level 1.
         // Because the min level is 1, the result color should be the white triangle drawn in the
@@ -188,12 +221,11 @@ TEST_F(BackendTest, TextureViewLod) {
         // Adjust the base mip to 2.
         auto texture22 = cleanup.add(api.createTextureView(texture, 2, 2));
 
-        DescriptorSetHandle descriptorSet22 = texturedShader.createDescriptorSet(api);
-        api.updateDescriptorSetTexture(descriptorSet22, 0, texture22, {
+        api.updateDescriptorSetTexture(descriptorSet[1], 0, texture22, {
                 .filterMag = SamplerMagFilter::NEAREST,
                 .filterMin = SamplerMinFilter::NEAREST_MIPMAP_NEAREST });
 
-        api.bindDescriptorSet(descriptorSet22, 0, {});
+        api.bindDescriptorSet(descriptorSet[1], 0, {});
 
         // Render a second, smaller, triangle, again sampling from mip level 1.
         // This triangle should be yellow striped.

filament/backend/test/test_MissingRequiredAttributes.cpp

@@ -17,8 +17,7 @@
 #include "BackendTest.h"
 
 #include "Lifetimes.h"
-#include "Shader.h"
-#include "SharedShaders.h"
+#include "ShaderGenerator.h"
 #include "TrianglePrimitive.h"
 
 namespace {
@@ -27,7 +26,7 @@ namespace {
 // Shaders
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
-std::string vertex(R"(#version 450 core
+std::string vertex (R"(#version 450 core
 
 layout(location = 0) in vec4 mesh_position;
 
@@ -47,6 +46,16 @@ void main() {
 }
 )");
 
+std::string fragment (R"(#version 450 core
+
+layout(location = 0) out vec4 fragColor;
+
+void main() {
+    fragColor = vec4(1.0);
+}
+
+)");
+
 }
 
 namespace test {
@@ -69,11 +78,9 @@ TEST_F(BackendTest, MissingRequiredAttributes) {
         api.makeCurrent(swapChain, swapChain);
 
         // Create a program.
-        Shader shader(api, cleanup, ShaderConfig{
-                .vertexShader = vertex,
-                .fragmentShader = SharedShaders::getFragmentShaderText(FragmentShaderType::White,
-                        ShaderUniformType::None),
-        });
+        ShaderGenerator shaderGen(vertex, fragment, sBackend, sIsMobilePlatform);
+        Program p = shaderGen.getProgram(api);
+        auto program = cleanup.add(api.createProgram(std::move(p)));
 
         auto defaultRenderTarget = cleanup.add(api.createDefaultRenderTarget(0));
 
@@ -82,12 +89,12 @@ TEST_F(BackendTest, MissingRequiredAttributes) {
         RenderPassParams params = {};
         fullViewport(params);
         params.flags.clear = TargetBufferFlags::COLOR;
-        params.clearColor = { 0.f, 1.f, 0.f, 1.f };
+        params.clearColor = {0.f, 1.f, 0.f, 1.f};
         params.flags.discardStart = TargetBufferFlags::ALL;
         params.flags.discardEnd = TargetBufferFlags::NONE;
 
         PipelineState state;
-        state.program = shader.getProgram();
+        state.program = program;
         state.rasterState.colorWrite = true;
         state.rasterState.depthWrite = false;
         state.rasterState.depthFunc = RasterState::DepthFunc::A;

filament/backend/test/test_ReadPixels.cpp

@@ -18,8 +18,7 @@
 
 #include "BackendTestUtils.h"
 #include "Lifetimes.h"
-#include "Shader.h"
-#include "SharedShaders.h"
+#include "ShaderGenerator.h"
 #include "TrianglePrimitive.h"
 
 #include <utils/Hash.h>
@@ -31,7 +30,6 @@ using namespace filament;
 using namespace filament::backend;
 
 #ifndef FILAMENT_IOS
-
 #include <imageio/ImageEncoder.h>
 #include <image/ColorTransform.h>
 
@@ -44,7 +42,20 @@ namespace {
 // Shaders
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
-std::string fragmentFloat(R"(#version 450 core
+std::string vertex (R"(#version 450 core
+
+layout(location = 0) in vec4 mesh_position;
+
+void main() {
+    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
+}
+)");
+
+std::string fragmentFloat (R"(#version 450 core
 
 layout(location = 0) out vec4 fragColor;
 
@@ -54,7 +65,7 @@ void main() {
 
 )");
 
-std::string fragmentUint(R"(#version 450 core
+std::string fragmentUint (R"(#version 450 core
 
 layout(location = 0) out uvec4 fragColor;
 
@@ -97,7 +108,7 @@ TEST_F(ReadPixelsTest, ReadPixels) {
         size_t samples = 1;
 
         // The size of the actual render target, taking mip level into account;
-        size_t getRenderTargetSize() const {
+        size_t getRenderTargetSize () const {
             return std::max(size_t(1), renderTargetBaseSize >> mipLevel);
         }
 
@@ -127,11 +138,11 @@ TEST_F(ReadPixelsTest, ReadPixels) {
         }
 
         void exportScreenshot(void* pixelData) const {
-#ifndef FILAMENT_IOS
+            #ifndef FILAMENT_IOS
             const size_t width = readRect.width, height = readRect.height;
             LinearImage image(width, height, 4);
             if (format == PixelDataFormat::RGBA && type == PixelDataType::UBYTE) {
-                image = toLinearWithAlpha<uint8_t>(width, height, width * 4, (uint8_t*)pixelData);
+                image = toLinearWithAlpha<uint8_t>(width, height, width * 4, (uint8_t*) pixelData);
             }
             if (format == PixelDataFormat::RGBA && type == PixelDataType::FLOAT) {
                 memcpy(image.getPixelRef(), pixelData, width * height * sizeof(math::float4));
@@ -140,13 +151,13 @@ TEST_F(ReadPixelsTest, ReadPixels) {
             std::ofstream outputStream(png.c_str(), std::ios::binary | std::ios::trunc);
             ImageEncoder::encode(outputStream, ImageEncoder::Format::PNG, image, "",
                     png.c_str());
-#endif
+            #endif
         }
 
         void exportRawBytes(void* pixelData) const {
             std::string out = std::string(testName) + ".raw";
             std::ofstream outputStream(out.c_str(), std::ios::binary | std::ios::trunc);
-            outputStream.write((char*)pixelData, getBufferSizeBytes());
+            outputStream.write((char*) pixelData, getBufferSizeBytes());
             outputStream.close();
         }
 
@@ -229,20 +240,22 @@ TEST_F(ReadPixelsTest, ReadPixels) {
     DriverApi& api = getDriverApi();
     Cleanup cleanup(api);
 
-    std::string vertexShader = SharedShaders::getVertexShaderText(VertexShaderType::Noop,
-            ShaderUniformType::None);
-    Shader floatShader(api, cleanup, ShaderConfig{
-            .vertexShader = vertexShader,
-            .fragmentShader = fragmentFloat,
-            .uniforms = {}
-    });
-    Shader uintShader(api, cleanup, ShaderConfig{
-            .vertexShader = vertexShader,
-            .fragmentShader = fragmentUint,
-            .uniforms = {}
-    });
+    // Create programs.
+    Handle<HwProgram> programFloat, programUint;
+    {
+        ShaderGenerator shaderGen(vertex, fragmentFloat, sBackend, sIsMobilePlatform);
+        Program p = shaderGen.getProgram(api);
+        programFloat = cleanup.add(api.createProgram(std::move(p)));
+    }
+    {
+        ShaderGenerator shaderGen(vertex, fragmentUint, sBackend, sIsMobilePlatform);
+        Program p = shaderGen.getProgram(api);
+        programUint = cleanup.add(api.createProgram(std::move(p)));
+    }
 
-    for (const auto& t: testCases) {
+
+    for (const auto& t : testCases)
+    {
         // Create a platform-specific SwapChain and make it current.
         Handle<HwSwapChain> swapChain;
         if (t.useDefaultRT) {
@@ -279,7 +292,7 @@ TEST_F(ReadPixelsTest, ReadPixels) {
         RenderPassParams params = {};
         fullViewport(params);
         params.flags.clear = TargetBufferFlags::COLOR;
-        params.clearColor = { 0.f, 0.f, 1.f, 1.f };
+        params.clearColor = {0.f, 0.f, 1.f, 1.f};
         params.flags.discardStart = TargetBufferFlags::ALL;
         params.flags.discardEnd = TargetBufferFlags::NONE;
         params.viewport.height = t.getRenderTargetSize();
@@ -292,9 +305,9 @@ TEST_F(ReadPixelsTest, ReadPixels) {
         api.beginRenderPass(renderTarget, params);
 
         PipelineState state;
-        state.program = floatShader.getProgram();
+        state.program = programFloat;
         if (isUnsignedIntFormat(t.textureFormat)) {
-            state.program = uintShader.getProgram();
+            state.program = programUint;
         }
         state.rasterState.colorWrite = true;
         state.rasterState.depthWrite = false;
@@ -312,7 +325,7 @@ TEST_F(ReadPixelsTest, ReadPixels) {
             Handle<HwRenderTarget> mipLevelOneRT = localCleanup.add(api.createRenderTarget(
                     TargetBufferFlags::COLOR, renderTargetBaseSize, renderTargetBaseSize, 1, 0,
                     {{ texture }}, {}, {}));
-            p.clearColor = { 1.f, 0.f, 0.f, 1.f };
+            p.clearColor = {1.f, 0.f, 0.f, 1.f};
             api.beginRenderPass(mipLevelOneRT, p);
             api.endRenderPass();
         }
@@ -322,28 +335,28 @@ TEST_F(ReadPixelsTest, ReadPixels) {
 
         PixelBufferDescriptor descriptor(buffer, t.getBufferSizeBytes(), t.format, t.type,
                 t.alignment, t.left, t.top, t.getPixelBufferStride(), [](void* buffer, size_t size,
-                        void* user) {
-                    const auto* test = (const TestCase*)user;
+                    void* user) {
+                    const auto* test = (const TestCase*) user;
                     assert_invariant(test);
 
                     test->exportScreenshot(buffer);
                     //test->exportRawBytes(buffer);
 
                     // Hash the contents of the buffer and check that they match.
-                    uint32_t hash = utils::hash::murmur3((const uint32_t*)buffer, size / 4, 0);
+                    uint32_t hash = utils::hash::murmur3((const uint32_t*) buffer, size / 4, 0);
 
                     ASSERT_EQ(test->hash, hash) << test->testName <<
-                                                " failed: hashes do not match." << std::endl;
+                        " failed: hashes do not match." << std::endl;
 
                     free(buffer);
-                }, (void*)&t);
+                }, (void*) &t);
 
         api.readPixels(renderTarget, t.readRect.x, t.readRect.y, t.readRect.width,
                 t.readRect.height, std::move(descriptor));
 
         // Now render red over what was just rendered. This ensures that readPixels captures the
         // state of rendering between render passes.
-        params.clearColor = { 1.f, 0.f, 0.f, 1.f };
+        params.clearColor = {1.f, 0.f, 0.f, 1.f};
         api.beginRenderPass(renderTarget, params);
         api.endRenderPass();
 
@@ -363,27 +376,25 @@ TEST_F(ReadPixelsTest, ReadPixelsPerformance) {
     Cleanup cleanup(api);
 
     // Create a platform-specific SwapChain and make it current.
-    auto swapChain = cleanup.add(
-            api.createSwapChainHeadless(renderTargetSize, renderTargetSize, 0));
+    auto swapChain = cleanup.add(api.createSwapChainHeadless(renderTargetSize, renderTargetSize, 0));
     api.makeCurrent(swapChain, swapChain);
 
-    Shader shader = SharedShaders::makeShader(api, cleanup, ShaderRequest{
-            .mVertexType = VertexShaderType::Noop,
-            .mFragmentType = FragmentShaderType::White,
-            .mUniformType = ShaderUniformType::None
-    });
+    // Create a program.
+    ShaderGenerator shaderGen(vertex, fragmentFloat, sBackend, sIsMobilePlatform);
+    Program p = shaderGen.getProgram(api);
+    auto program = cleanup.add(api.createProgram(std::move(p)));
 
     // Create a Texture and RenderTarget to render into.
     auto usage = TextureUsage::COLOR_ATTACHMENT | TextureUsage::SAMPLEABLE;
     Handle<HwTexture> texture = cleanup.add(api.createTexture(
-            SamplerType::SAMPLER_2D,            // target
-            1,                                  // levels
-            TextureFormat::RGBA8,               // format
-            1,                                  // samples
-            renderTargetSize,                   // width
-            renderTargetSize,                   // height
-            1,                                  // depth
-            usage));                             // usage
+                SamplerType::SAMPLER_2D,            // target
+                1,                                  // levels
+                TextureFormat::RGBA8,               // format
+                1,                                  // samples
+                renderTargetSize,                   // width
+                renderTargetSize,                   // height
+                1,                                  // depth
+                usage));                             // usage
 
     Handle<HwRenderTarget> renderTarget = cleanup.add(api.createRenderTarget(
             TargetBufferFlags::COLOR,
@@ -400,7 +411,7 @@ TEST_F(ReadPixelsTest, ReadPixelsPerformance) {
     RenderPassParams params = {};
     fullViewport(params);
     params.flags.clear = TargetBufferFlags::COLOR;
-    params.clearColor = { 0.f, 0.f, 1.f, 1.f };
+    params.clearColor = {0.f, 0.f, 1.f, 1.f};
     params.flags.discardStart = TargetBufferFlags::ALL;
     params.flags.discardEnd = TargetBufferFlags::NONE;
     params.viewport.height = renderTargetSize;
@@ -409,7 +420,7 @@ TEST_F(ReadPixelsTest, ReadPixelsPerformance) {
     void* buffer = calloc(1, renderTargetSize * renderTargetSize * 4);
 
     PipelineState state;
-    state.program = shader.getProgram();
+    state.program = program;
     state.rasterState.colorWrite = true;
     state.rasterState.depthWrite = false;
     state.rasterState.depthFunc = RasterState::DepthFunc::A;
@@ -433,12 +444,11 @@ TEST_F(ReadPixelsTest, ReadPixelsPerformance) {
         PixelBufferDescriptor descriptor(buffer, renderTargetSize * renderTargetSize * 4,
                 PixelDataFormat::RGBA, PixelDataType::UBYTE, 1, 0, 0, renderTargetSize,
                 [](void* buffer, size_t size, void* user) {
-                    ReadPixelsTest* test = (ReadPixelsTest*)user;
+                    ReadPixelsTest* test = (ReadPixelsTest*) user;
                     test->readPixelsFinished = true;
                 }, this);
 
-        api.readPixels(renderTarget, 0, 0, renderTargetSize, renderTargetSize,
-                std::move(descriptor));
+        api.readPixels(renderTarget, 0, 0, renderTargetSize, renderTargetSize, std::move(descriptor));
         api.commit(swapChain);
         api.endFrame(0);
 

filament/backend/test/test_RenderExternalImage.cpp

@@ -17,8 +17,7 @@
 #include "BackendTest.h"
 
 #include "Lifetimes.h"
-#include "Shader.h"
-#include "SharedShaders.h"
+#include "ShaderGenerator.h"
 #include "TrianglePrimitive.h"
 
 #include <backend/DriverEnums.h>
@@ -29,19 +28,42 @@
 #include <stddef.h>
 #include <stdint.h>
 
+namespace {
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+// Shaders
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+std::string vertex (R"(#version 450 core
+
+layout(location = 0) in vec4 mesh_position;
+layout(location = 0) out vec2 uv;
+
+void main() {
+    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+    uv = (mesh_position.xy * 0.5 + 0.5);
+}
+)");
+
+std::string fragment (R"(#version 450 core
+
+layout(location = 0) out vec4 fragColor;
+layout(location = 0) in vec2 uv;
+
+layout(location = 0, set = 1) uniform sampler2D test_tex;
+
+void main() {
+    fragColor = texture(test_tex, uv);
+}
+)");
+
+}
+
 namespace test {
 
 using namespace filament;
 using namespace filament::backend;
 
-Shader createShader(DriverApi& api, Cleanup& cleanup, Backend backend) {
-    return SharedShaders::makeShader(api, cleanup, ShaderRequest{
-            .mVertexType = VertexShaderType::Textured,
-            .mFragmentType = FragmentShaderType::Textured,
-            .mUniformType = ShaderUniformType::Sampler
-    });
-}
-
 // Rendering an external image without setting any data should not crash.
 TEST_F(BackendTest, RenderExternalImageWithoutSet) {
     auto& api = getDriverApi();
@@ -51,41 +73,57 @@ TEST_F(BackendTest, RenderExternalImageWithoutSet) {
 
     auto swapChain = cleanup.add(createSwapChain());
 
-    Shader shader = createShader(api, cleanup, sBackend);
+    filament::SamplerInterfaceBlock::SamplerInfo samplerInfo { "test", "tex", 0,
+        SamplerType::SAMPLER_EXTERNAL, SamplerFormat::FLOAT, Precision::HIGH, false };
+    filamat::DescriptorSets descriptors;
+    descriptors[1] = { { "test_tex", { DescriptorType::SAMPLER, ShaderStageFlags::FRAGMENT, 0 },
+            samplerInfo } };
+    ShaderGenerator shaderGen(
+            vertex, fragment, sBackend, sIsMobilePlatform, std::move(descriptors));
 
-    backend::Handle<HwRenderTarget> defaultRenderTarget = cleanup.add(
-            api.createDefaultRenderTarget(0));
+    // Create a program that samples a texture.
+    Program p = shaderGen.getProgram(api);
+    p.descriptorBindings(1, {{"test_tex", DescriptorType::SAMPLER, 0}});
+    backend::Handle<HwProgram> program = cleanup.add(api.createProgram(std::move(p)));
+    DescriptorSetLayoutHandle descriptorSetLayout = cleanup.add(api.createDescriptorSetLayout({
+            {{
+                     DescriptorType::SAMPLER,
+                     ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, 0,
+                     DescriptorFlags::NONE, 0
+             }}}));
+
+    DescriptorSetHandle descriptorSet = cleanup.add(api.createDescriptorSet(descriptorSetLayout));
+
+    backend::Handle<HwRenderTarget> defaultRenderTarget = cleanup.add(api.createDefaultRenderTarget(0));
 
     // Create a texture that will be backed by an external image.
     auto usage = TextureUsage::COLOR_ATTACHMENT | TextureUsage::SAMPLEABLE;
     const NativeView& view = getNativeView();
     backend::Handle<HwTexture> texture = cleanup.add(api.createTexture(
-            SamplerType::SAMPLER_EXTERNAL,      // target
-            1,                                  // levels
-            TextureFormat::RGBA8,               // format
-            1,                                  // samples
-            view.width,                         // width
-            view.height,                        // height
-            1,                                  // depth
-            usage));                             // usage
+                SamplerType::SAMPLER_EXTERNAL,      // target
+                1,                                  // levels
+                TextureFormat::RGBA8,               // format
+                1,                                  // samples
+                view.width,                         // width
+                view.height,                        // height
+                1,                                  // depth
+                usage));                             // usage
 
     RenderPassParams params = {};
     fullViewport(params);
     params.flags.clear = TargetBufferFlags::COLOR;
-    params.clearColor = { 0.f, 1.f, 0.f, 1.f };
+    params.clearColor = {0.f, 1.f, 0.f, 1.f};
     params.flags.discardStart = TargetBufferFlags::ALL;
     params.flags.discardEnd = TargetBufferFlags::NONE;
 
     PipelineState state;
-    state.program = shader.getProgram();
-    state.pipelineLayout.setLayout[1] = { shader.getDescriptorSetLayout() };
+    state.program = program;
+    state.pipelineLayout.setLayout[1] = { descriptorSetLayout };
     state.rasterState.colorWrite = true;
     state.rasterState.depthWrite = false;
     state.rasterState.depthFunc = RasterState::DepthFunc::A;
     state.rasterState.culling = CullingMode::NONE;
 
-    DescriptorSetHandle descriptorSet = shader.createDescriptorSet(api);
-
     api.startCapture(0);
     api.makeCurrent(swapChain, swapChain);
     api.beginFrame(0, 0, 0);
@@ -117,11 +155,29 @@ TEST_F(BackendTest, RenderExternalImage) {
 
     auto swapChain = cleanup.add(createSwapChain());
 
-    Shader shader = createShader(api, cleanup, sBackend);
-    DescriptorSetHandle descriptorSet = shader.createDescriptorSet(api);
+    filament::SamplerInterfaceBlock::SamplerInfo samplerInfo { "test", "tex", 0,
+        SamplerType::SAMPLER_EXTERNAL, SamplerFormat::FLOAT, Precision::HIGH, false };
+    filamat::DescriptorSets descriptors;
+    descriptors[1] = { { "test_tex", { DescriptorType::SAMPLER, ShaderStageFlags::FRAGMENT, 0 },
+            samplerInfo } };
+    ShaderGenerator shaderGen(
+            vertex, fragment, sBackend, sIsMobilePlatform, std::move(descriptors));
 
-    backend::Handle<HwRenderTarget> defaultRenderTarget = cleanup.add(
-            api.createDefaultRenderTarget(0));
+    // Create a program that samples a texture.
+    Program p = shaderGen.getProgram(api);
+    p.descriptorBindings(1, {{"test_tex", DescriptorType::SAMPLER, 0}});
+    auto program = cleanup.add(api.createProgram(std::move(p)));
+
+    DescriptorSetLayoutHandle descriptorSetLayout = cleanup.add(api.createDescriptorSetLayout({
+            {{
+                     DescriptorType::SAMPLER,
+                     ShaderStageFlags::ALL_SHADER_STAGE_FLAGS, 0,
+                     DescriptorFlags::NONE, 0
+             }}}));
+
+    DescriptorSetHandle descriptorSet = cleanup.add(api.createDescriptorSet(descriptorSetLayout));
+
+    backend::Handle<HwRenderTarget> defaultRenderTarget = cleanup.add(api.createDefaultRenderTarget(0));
 
     // require users to create two Filament textures and have two material parameters
     // add a "plane" parameter to setExternalImage
@@ -142,12 +198,10 @@ TEST_F(BackendTest, RenderExternalImage) {
     values[1] = values[0];
     values[2] = values[0];
     values[3] = values[0];
-    CFDictionaryRef options = CFDictionaryCreate(kCFAllocatorDefault, (const void**)keys,
-            (const void**)values, 4, nullptr, nullptr);
+    CFDictionaryRef options = CFDictionaryCreate(kCFAllocatorDefault, (const void**) keys, (const void**) values, 4, nullptr, nullptr);
     CVPixelBufferRef pixBuffer = nullptr;
     CVReturn status =
-            CVPixelBufferCreate(kCFAllocatorDefault, 1024, 1024, kCVPixelFormatType_32BGRA, options,
-                    &pixBuffer);
+            CVPixelBufferCreate(kCFAllocatorDefault, 1024, 1024, kCVPixelFormatType_32BGRA, options, &pixBuffer);
     assert(status == kCVReturnSuccess);
 
     // Fill image with checker-pattern.
@@ -156,7 +210,7 @@ TEST_F(BackendTest, RenderExternalImage) {
     const uint32_t black = 0xFF000000;
     CVReturn lockStatus = CVPixelBufferLockBaseAddress(pixBuffer, 0);
     assert(lockStatus == kCVReturnSuccess);
-    uint32_t* pix = (uint32_t*)CVPixelBufferGetBaseAddressOfPlane(pixBuffer, 0);
+    uint32_t* pix = (uint32_t*) CVPixelBufferGetBaseAddressOfPlane(pixBuffer, 0);
     assert(pix);
     for (size_t r = 0; r < 1024; r++) {
         for (size_t c = 0; c < 1024; c++) {
@@ -176,13 +230,13 @@ TEST_F(BackendTest, RenderExternalImage) {
     RenderPassParams params = {};
     fullViewport(params);
     params.flags.clear = TargetBufferFlags::COLOR;
-    params.clearColor = { 0.f, 1.f, 0.f, 1.f };
+    params.clearColor = {0.f, 1.f, 0.f, 1.f};
     params.flags.discardStart = TargetBufferFlags::ALL;
     params.flags.discardEnd = TargetBufferFlags::NONE;
 
     PipelineState state;
-    state.program = shader.getProgram();
-    state.pipelineLayout.setLayout[1] = { shader.getDescriptorSetLayout() };
+    state.program = program;
+    state.pipelineLayout.setLayout[1] = { descriptorSetLayout };
     state.rasterState.colorWrite = true;
     state.rasterState.depthWrite = false;
     state.rasterState.depthFunc = RasterState::DepthFunc::A;
@@ -193,6 +247,7 @@ TEST_F(BackendTest, RenderExternalImage) {
     api.beginFrame(0, 0, 0);
 
     api.updateDescriptorSetTexture(descriptorSet, 0, texture, {});
+
     api.bindDescriptorSet(descriptorSet, 1, {});
 
     // Render a triangle.

filament/backend/test/test_Scissor.cpp

@@ -17,8 +17,7 @@
 #include "BackendTest.h"
 
 #include "Lifetimes.h"
-#include "Shader.h"
-#include "SharedShaders.h"
+#include "ShaderGenerator.h"
 #include "TrianglePrimitive.h"
 
 #include <utils/Hash.h>
@@ -28,6 +27,23 @@ namespace test {
 using namespace filament;
 using namespace filament::backend;
 
+static const char* const triangleVs = R"(#version 450 core
+layout(location = 0) in vec4 mesh_position;
+void main() {
+    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
+})";
+
+static const char* const triangleFs = R"(#version 450 core
+precision mediump int; precision highp float;
+layout(location = 0) out vec4 fragColor;
+void main() {
+    fragColor = vec4(1.0f);
+})";
+
 TEST_F(BackendTest, ScissorViewportRegion) {
     auto& api = getDriverApi();
 
@@ -69,11 +85,10 @@ TEST_F(BackendTest, ScissorViewportRegion) {
         auto swapChain = cleanup.add(api.createSwapChainHeadless(256, 256, 0));
         api.makeCurrent(swapChain, swapChain);
 
-        Shader shader = SharedShaders::makeShader(api, cleanup, ShaderRequest{
-            .mVertexType = VertexShaderType::Noop,
-            .mFragmentType = FragmentShaderType::White,
-            .mUniformType = ShaderUniformType::None,
-        });
+        // Create a program.
+        ShaderGenerator shaderGen(triangleVs, triangleFs, sBackend, sIsMobilePlatform);
+        Program p = shaderGen.getProgram(api);
+        ProgramHandle program = cleanup.add(api.createProgram(std::move(p)));
 
         // Create source color and depth textures.
         Handle<HwTexture> srcTexture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, kNumLevels,
@@ -118,7 +133,7 @@ TEST_F(BackendTest, ScissorViewportRegion) {
         params.flags.discardEnd = TargetBufferFlags::NONE;
 
         PipelineState ps = {};
-        ps.program = shader.getProgram();
+        ps.program = program;
         ps.rasterState.colorWrite = true;
         ps.rasterState.depthWrite = false;
 
@@ -160,11 +175,10 @@ TEST_F(BackendTest, ScissorViewportEdgeCases) {
         auto swapChain = cleanup.add(api.createSwapChainHeadless(256, 256, 0));
         api.makeCurrent(swapChain, swapChain);
 
-        Shader shader = SharedShaders::makeShader(api, cleanup, ShaderRequest{
-                .mVertexType = VertexShaderType::Noop,
-                .mFragmentType = FragmentShaderType::White,
-                .mUniformType = ShaderUniformType::None,
-        });
+        // Create a program.
+        ShaderGenerator shaderGen(triangleVs, triangleFs, sBackend, sIsMobilePlatform);
+        Program p = shaderGen.getProgram(api);
+        ProgramHandle program = cleanup.add(api.createProgram(std::move(p)));
 
         // Create a source color textures.
         Handle<HwTexture> srcTexture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, 1,
@@ -206,7 +220,7 @@ TEST_F(BackendTest, ScissorViewportEdgeCases) {
         params.flags.discardEnd = TargetBufferFlags::NONE;
 
         PipelineState ps = {};
-        ps.program = shader.getProgram();
+        ps.program = program;
         ps.rasterState.colorWrite = true;
         ps.rasterState.depthWrite = false;
 

filament/backend/test/test_StencilBuffer.cpp

@@ -17,13 +17,43 @@
 #include "BackendTest.h"
 
 #include "Lifetimes.h"
-#include "Shader.h"
-#include "SharedShaders.h"
+#include "ShaderGenerator.h"
 #include "TrianglePrimitive.h"
 
 using namespace filament;
 using namespace filament::backend;
 
+namespace {
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+// Shaders
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+std::string vertex (R"(#version 450 core
+
+layout(location = 0) in vec4 mesh_position;
+
+void main() {
+    gl_Position = vec4(mesh_position.xy, 0.0, 1.0);
+#if defined(TARGET_VULKAN_ENVIRONMENT)
+    // In Vulkan, clip space is Y-down. In OpenGL and Metal, clip space is Y-up.
+    gl_Position.y = -gl_Position.y;
+#endif
+}
+)");
+
+std::string fragment (R"(#version 450 core
+
+layout(location = 0) out vec4 fragColor;
+
+void main() {
+    fragColor = vec4(1.0);
+}
+
+)");
+
+}
+
 namespace test {
 
 // 1. Clear the stencil buffer to all zeroes.
@@ -34,7 +64,7 @@ namespace test {
 class BasicStencilBufferTest : public BackendTest {
 public:
 
-    Handle <HwSwapChain> mSwapChain;
+    Handle<HwSwapChain> mSwapChain;
     ProgramHandle mProgram;
     Cleanup mCleanup;
 
@@ -47,23 +77,21 @@ public:
         mSwapChain = mCleanup.add(createSwapChain());
         api.makeCurrent(mSwapChain, mSwapChain);
 
-        Shader shader = SharedShaders::makeShader(api, mCleanup, ShaderRequest{
-                .mVertexType = VertexShaderType::Noop,
-                .mFragmentType = FragmentShaderType::White,
-                .mUniformType = ShaderUniformType::None
-        });
-        mProgram = shader.getProgram();
+        // Create a program.
+        ShaderGenerator shaderGen(vertex, fragment, sBackend, sIsMobilePlatform);
+        Program p = shaderGen.getProgram(api);
+        mProgram = mCleanup.add(api.createProgram(std::move(p)));
     }
 
-    void RunTest(Handle <HwRenderTarget> renderTarget) {
+    void RunTest(Handle<HwRenderTarget> renderTarget) {
         auto& api = getDriverApi();
 
         // We'll be using a triangle as geometry.
         TrianglePrimitive smallTriangle(api);
         static filament::math::float2 vertices[3] = {
                 { -0.5, -0.5 },
-                { 0.5,  -0.5 },
-                { -0.5, 0.5 }
+                {  0.5, -0.5 },
+                { -0.5,  0.5 }
         };
         smallTriangle.updateVertices(vertices);
         TrianglePrimitive triangle(api);
@@ -72,7 +100,7 @@ public:
         // Render a small triangle only to the stencil buffer, increasing the stencil buffer to 1.
         RenderPassParams params = {};
         params.flags.clear = TargetBufferFlags::COLOR0 | TargetBufferFlags::STENCIL;
-        params.viewport = { 0, 0, 512, 512 };
+        params.viewport = {0, 0, 512, 512};
         params.clearColor = math::float4(0.0f, 0.0f, 1.0f, 1.0f);
         params.clearStencil = 0u;
         params.flags.discardStart = TargetBufferFlags::ALL;
@@ -125,7 +153,7 @@ TEST_F(BasicStencilBufferTest, StencilBuffer) {
             TextureFormat::STENCIL8, 1, 512, 512, 1, TextureUsage::STENCIL_ATTACHMENT));
     auto renderTarget = cleanup.add(api.createRenderTarget(
             TargetBufferFlags::COLOR0 | TargetBufferFlags::STENCIL, 512, 512, 1, 0,
-            {{ colorTexture }}, {}, {{ stencilTexture }}));
+            {{colorTexture}}, {}, {{stencilTexture}}));
 
     RunTest(renderTarget);
 
@@ -143,11 +171,10 @@ TEST_F(BasicStencilBufferTest, DepthAndStencilBuffer) {
     auto colorTexture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, 1,
             TextureFormat::RGBA8, 1, 512, 512, 1, TextureUsage::COLOR_ATTACHMENT));
     auto depthStencilTexture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, 1,
-            TextureFormat::DEPTH24_STENCIL8, 1, 512, 512, 1,
-            TextureUsage::STENCIL_ATTACHMENT | TextureUsage::DEPTH_ATTACHMENT));
+            TextureFormat::DEPTH24_STENCIL8, 1, 512, 512, 1, TextureUsage::STENCIL_ATTACHMENT | TextureUsage::DEPTH_ATTACHMENT));
     auto renderTarget = cleanup.add(api.createRenderTarget(
             TargetBufferFlags::COLOR0 | TargetBufferFlags::STENCIL, 512, 512, 1, 0,
-            {{ colorTexture }}, { depthStencilTexture }, {{ depthStencilTexture }}));
+            {{colorTexture}}, {depthStencilTexture}, {{depthStencilTexture}}));
 
     RunTest(renderTarget);
 
@@ -167,17 +194,15 @@ TEST_F(BasicStencilBufferTest, StencilBufferMSAA) {
     // Pass 1: Render a triangle into (an auto-created) MSAA color buffer using the stencil test.
     //         Performs an auto-resolve on the color.
     auto colorTexture = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, 1,
-            TextureFormat::RGBA8, 1, 512, 512, 1,
-            TextureUsage::COLOR_ATTACHMENT | TextureUsage::SAMPLEABLE));
+            TextureFormat::RGBA8, 1, 512, 512, 1, TextureUsage::COLOR_ATTACHMENT | TextureUsage::SAMPLEABLE));
     auto depthStencilTextureMSAA = cleanup.add(api.createTexture(SamplerType::SAMPLER_2D, 1,
-            TextureFormat::DEPTH24_STENCIL8, 4, 512, 512, 1,
-            TextureUsage::STENCIL_ATTACHMENT | TextureUsage::DEPTH_ATTACHMENT));
+            TextureFormat::DEPTH24_STENCIL8, 4, 512, 512, 1, TextureUsage::STENCIL_ATTACHMENT | TextureUsage::DEPTH_ATTACHMENT));
     auto renderTarget0 = cleanup.add(api.createRenderTarget(
             TargetBufferFlags::DEPTH_AND_STENCIL, 512, 512, 4, 0,
-            {{}}, { depthStencilTextureMSAA }, { depthStencilTextureMSAA }));
+            {{}}, {depthStencilTextureMSAA}, {depthStencilTextureMSAA}));
     auto renderTarget1 = cleanup.add(api.createRenderTarget(
             TargetBufferFlags::COLOR0 | TargetBufferFlags::DEPTH_AND_STENCIL, 512, 512, 4, 0,
-            {{ colorTexture }}, { depthStencilTextureMSAA }, { depthStencilTextureMSAA }));
+            {{colorTexture}}, {depthStencilTextureMSAA}, {depthStencilTextureMSAA}));
 
     api.startCapture(0);
 
@@ -185,8 +210,8 @@ TEST_F(BasicStencilBufferTest, StencilBufferMSAA) {
     TrianglePrimitive smallTriangle(api);
     static filament::math::float2 vertices[3] = {
             { -0.5, -0.5 },
-            { 0.5,  -0.5 },
-            { -0.5, 0.5 }
+            {  0.5, -0.5 },
+            { -0.5,  0.5 }
     };
     smallTriangle.updateVertices(vertices);
     TrianglePrimitive triangle(api);
@@ -195,7 +220,7 @@ TEST_F(BasicStencilBufferTest, StencilBufferMSAA) {
     // Render a small triangle only to the stencil buffer, increasing the stencil buffer to 1.
     RenderPassParams params = {};
     params.flags.clear = TargetBufferFlags::STENCIL;
-    params.viewport = { 0, 0, 512, 512 };
+    params.viewport = {0, 0, 512, 512};
     params.clearStencil = 0u;
     params.flags.discardStart = TargetBufferFlags::ALL;
     params.flags.discardEnd = TargetBufferFlags::NONE;

filament/include/filament/Engine.h

@@ -317,15 +317,6 @@ public:
          */
         size_t metalUploadBufferSizeBytes = 512 * 1024;
 
-        /**
-         * The action to take if a Drawable cannot be acquired.
-         *
-         * Each frame rendered requires a CAMetalDrawable texture, which is
-         * presented on-screen at the completion of each frame. These are
-         * limited and provided round-robin style by the system.
-         */
-        bool metalDisablePanicOnDrawableFailure = false;
-
         /**
          * Set to `true` to forcibly disable parallel shader compilation in the backend.
          * Currently only honored by the GL and Metal backends.

filament/src/details/Engine.cpp

@@ -138,7 +138,6 @@ Engine* FEngine::create(Builder const& builder) {
                 .forceGLES2Context = instance->getConfig().forceGLES2Context,
                 .stereoscopicType = instance->getConfig().stereoscopicType,
                 .assertNativeWindowIsValid = instance->features.backend.opengl.assert_native_window_is_valid,
-                .metalDisablePanicOnDrawableFailure = instance->getConfig().metalDisablePanicOnDrawableFailure,
         };
         instance->mDriver = platform->createDriver(sharedContext, driverConfig);
 
@@ -734,7 +733,6 @@ int FEngine::loop() {
             .forceGLES2Context = mConfig.forceGLES2Context,
             .stereoscopicType =  mConfig.stereoscopicType,
             .assertNativeWindowIsValid = features.backend.opengl.assert_native_window_is_valid,
-            .metalDisablePanicOnDrawableFailure = mConfig.metalDisablePanicOnDrawableFailure,
     };
     mDriver = mPlatform->createDriver(mSharedGLContext, driverConfig);
 

test/code-correctness/src/run.py

@@ -1,109 +0,0 @@
-# Copyright (C) 2025 The Android Open Source Project
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#      http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import os
-import glob
-import yaml
-import hashlib
-import concurrent.futures
-import re
-
-from utils import execute, ArgParseImpl
-
-def get_line(file_name, offset):
-  with open(f'{file_name}', 'rb') as file:
-    bytes = file.read()[0:offset]
-    f_str = bytes.decode('utf-8')
-    return len(f_str.split('\n'))
-  return -1
-
-def get_func_name(msg):
-  pattern = r"\'(.+)\'"
-  res = re.findall(pattern, msg)
-  if len(res) > 0:
-    return res[0].replace("'", '')
-  return msg
-
-def run_tidy(files):
-  files_str = ' '.join(files)
-  hid = hashlib.md5(files_str.encode('utf-8')).hexdigest()
-  _, _ = execute(f'clang-tidy --export-fixes=/tmp/{hid}.yaml --quiet --checks=-*,bugprone-exception-escape {files_str}')
-  results = []
-  with open(f'/tmp/{hid}.yaml', 'r') as file:
-    data = yaml.safe_load(file)
-    for d in data['Diagnostics']:
-      if d['DiagnosticName'] != 'bugprone-exception-escape':
-        continue
-      msg = d['DiagnosticMessage']
-      fpath = msg['FilePath']
-      offset = msg['FileOffset']
-      line_num = get_line(fpath, offset)
-      results.append((msg['FilePath'].replace(f'{os.getcwd()}/', ''), line_num, get_func_name(msg['Message'])))
-  return results
-
-def exception_escape_test():
-  files = glob.glob('filament/**/*.mm', recursive=True) + \
-      glob.glob('filament/**/*.cpp', recursive=True) + \
-      glob.glob('filament/**/*.h', recursive=True)
-
-  num_workers = 5  # Number of threads to spawn
-  part_len = len(files) // num_workers
-  workloads = []
-  for i in range(num_workers):
-    next = min(len(files), part_len)
-    workloads.append(files[0:next])
-    files = files[next:]
-
-  all_results = []
-  with concurrent.futures.ThreadPoolExecutor(max_workers=num_workers) as executor:
-    future_to_worker_id = {executor.submit(run_tidy, workloads[i]): i for i in range(num_workers)}
-
-    for future in concurrent.futures.as_completed(future_to_worker_id):
-      worker_id = future_to_worker_id[future]
-      try:
-        all_results.extend(future.result())
-      except Exception as exc:
-          print(f"Main: Worker {worker_id} generated an exception: {exc}")
-  test_name = 'code-correctness::exception-escape'
-  failure_str_lines = []
-  if len(all_results) > 0:
-    failure_str_lines.append(f'Number of failures: {len(all_results)}')
-    for fname, line_num, msg in all_results:
-      failure_str_lines.append(f'{fname}({line_num}): {msg}()')
-  return (len(all_results) == 0, failure_str_lines)
-
-TESTS = [
-  (exception_escape_test,
-   'exception-escape',
-   '\'an exception may be thrown in a function which should not throw exceptions\'')
-]
-
-if __name__ == "__main__":
-  has_failures = False
-  for test_func, test_name, test_desc in TESTS:
-    result, res_strs = test_func()
-    ss = ' ' * 4
-    if result:
-      print(f'[{test_name}] PASSED')
-    else:
-      has_failures = True
-      print(f'[{test_name}] FAILED')
-      print(f'{ss}Description: \'{test_desc}\'')
-      for s in res_strs:
-        print(f'{ss}{s}')
-  if has_failures:
-    # TODO: Enable this when we've fixed all the exception-escape errors
-    #exit(1)    
-    pass
-

test/code-correctness/src/utils.py

@@ -1,68 +0,0 @@
-# Copyright (C) 2024 The Android Open Source Project
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#      http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import subprocess
-import os
-import argparse
-import sys
-
-def execute(cmd,
-            cwd=None,
-            capture_output=True,
-            stdin=None,
-            env=None,
-            raise_errors=False):
-  in_env = os.environ
-  in_env.update(env if env else {})
-  home = os.environ['HOME']
-  if f'{home}/bin' not in in_env['PATH']:
-    in_env['PATH'] = in_env['PATH'] + f':{home}/bin'
-
-  stdout = subprocess.PIPE if capture_output else sys.stdout
-  stderr = subprocess.PIPE if capture_output else sys.stdout
-  output = ''
-  err_output = ''
-  return_code = -1
-  kwargs = {
-      'cwd': cwd,
-      'env': in_env,
-      'stdout': stdout,
-      'stderr': stderr,
-      'stdin': stdin,
-      'universal_newlines': True
-  }
-  if capture_output:
-    process = subprocess.Popen(cmd.split(' '), **kwargs)
-    output, err_output = process.communicate()
-    return_code = process.returncode
-  else:
-    return_code = subprocess.call(cmd.split(' '), **kwargs)
-
-  if return_code:
-    # Error
-    if raise_errors:
-      raise subprocess.CalledProcessError(return_code, cmd)
-  if output:
-    if type(output) != str:
-      try:
-        output = output.decode('utf-8').strip()
-      except UnicodeDecodeError as e:
-        print('cannot decode ', output, file=sys.stderr)
-  return return_code, (output if return_code == 0 else err_output)
-
-class ArgParseImpl(argparse.ArgumentParser):
-  def error(self, message):
-    sys.stderr.write('error: %s\n' % message)
-    self.print_help()
-    sys.exit(1)

test/code-correctness/test.sh

@@ -1,28 +0,0 @@
-# Copyright (C) 2025 The Android Open Source Project
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#      http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-#!/usr/bin/bash
-
-CODE_CORRECTNESS_TEST_DIR="$(pwd)/test/code-correctness"
-
-# Check if the clang-tidy command exists and is executable
-if ! command -v clang-tidy > /dev/null 2>&1; then
-  # If command -v fails (returns a non-zero exit status), clang-tidy is not found
-  echo "Error: clang-tidy command not found." >&2
-  echo "Please install clang-tidy or ensure it is in your system's PATH." >&2
-  exit 1 # Exit the script with an error code (conventionally non-zero for failure)
-fi
-
-set -e && set -x && \
-    python3 ${CODE_CORRECTNESS_TEST_DIR}/src/run.py

third_party/dawn/tnt/CMakeLists.txt

@@ -33,36 +33,22 @@ if (ANDROID AND FILAMENT_BUILD_FILAMAT)
         tint_lang_core
         tint_lang_core_constant
         tint_lang_core_intrinsic
-        tint_lang_core_ir
-        tint_lang_core_ir_transform
-        tint_lang_core_ir_type
         tint_lang_core_type
-        tint_lang_spirv
-        tint_lang_spirv_intrinsic
-        tint_lang_spirv_ir
         tint_lang_spirv_reader
         tint_lang_spirv_reader_ast_lower
         tint_lang_spirv_reader_ast_parser
         tint_lang_spirv_reader_common
-        tint_lang_spirv_reader_lower
-        tint_lang_spirv_reader_parser
-        tint_lang_spirv_type
-        tint_lang_spirv_validate
         tint_lang_spirv_writer
         tint_lang_wgsl
         tint_lang_wgsl_ast
         tint_lang_wgsl_ast_transform
-        tint_lang_wgsl_common
         tint_lang_wgsl_features
         tint_lang_wgsl_intrinsic
-        tint_lang_wgsl_ir
         tint_lang_wgsl_program
         tint_lang_wgsl_resolver
         tint_lang_wgsl_sem
         tint_lang_wgsl_writer
         tint_lang_wgsl_writer_ast_printer
-        tint_lang_wgsl_writer_ir_to_program
-        tint_lang_wgsl_writer_raise
         tint_results
         tint_utils
         tint_utils_diagnostic
@@ -72,7 +58,6 @@ if (ANDROID AND FILAMENT_BUILD_FILAMAT)
         tint_utils_strconv
         tint_utils_symbol
         tint_utils_text
-        tint_utils_text_generator
     )
 
     # Use the following no-op definitions to introduce a library and its dependency.