continuing to test splitting combined samplers

Use this to test:

./build.sh -W debug matc tint spirv-opt spirv_test && \
    ./out/cmake-debug/tools/matc/matc -R -a vulkan -o out.cmat ./filament/src/materials/dof/dof.mat && \
    glslc dof_0x00.frag -o dof_0x00.spv && \
    ./out/cmake-debug/third_party/spirv-tools/tools/spirv-opt --split-combined-image-sampler dof_0x00.spv -o split.spv && \
    ./out/cmake-debug/samples/spirv_test split.spv offset.spv && \
    ./out/cmake-debug/third_party/dawn/tint --format wgsl offset.spv > out.wgsl

README.md

@@ -31,7 +31,7 @@ repositories {
 }
 
 dependencies {
-    implementation 'com.google.android.filament:filament-android:1.58.2'
+    implementation 'com.google.android.filament:filament-android:1.58.0'
 }
 ```
 
@@ -51,7 +51,7 @@ Here are all the libraries available in the group `com.google.android.filament`:
 iOS projects can use CocoaPods to install the latest release:
 
 ```shell
-pod 'Filament', '~> 1.58.2'
+pod 'Filament', '~> 1.58.0'
 ```
 
 ## Documentation

RELEASE_NOTES.md

@@ -7,10 +7,6 @@ A new header is inserted each time a *tag* is created.
 Instead, if you are authoring a PR for the main branch, add your release note to
 [NEW_RELEASE_NOTES.md](./NEW_RELEASE_NOTES.md).
 
-## v1.58.2
-
-- engine: Generate 1D instead of 3D LUTs for color grading whenever possible.
-
 ## v1.58.1
 
 

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

@@ -46,8 +46,9 @@ import static com.google.android.filament.Asserts.assertFloat4In;
  *
  * <h1>Performance</h1>
  *
- * Creating a new ColorGrading object may be more expensive than other Filament objects as a LUT may
- * need to be generated. The generation of this LUT, if necessary, may happen on the CPU.
+ * Creating a new ColorGrading object may be more expensive than other Filament objects as a
+ * 3D LUT may need to be generated. The generation of a 3D LUT, if necessary, may happen on
+ * the CPU.
  *
  * <h1>Ordering</h1>
  *
@@ -159,10 +160,6 @@ public class ColorGrading {
          * quality level will use a 32x32x32 10 bit LUT, a high quality will use a 32x32x32 16 bit
          * LUT, and a ultra quality will use a 64x64x64 16 bit LUT.
          *
-         * This setting has no effect if generating a 1D LUT.
-         *
-         * This overrides the values set by format() and dimensions().
-         *
          * The default quality is {@link QualityLevel#MEDIUM}.
          *
          * @param qualityLevel The desired quality of the color grading process
@@ -178,8 +175,6 @@ public class ColorGrading {
          * When color grading is implemented using a 3D LUT, this sets the texture format of
          * of the LUT. This overrides the value set by quality().
          *
-         * This setting has no effect if generating a 1D LUT.
-         *
          * The default is INTEGER
          *
          * @param format The desired format of the 3D LUT.
@@ -195,8 +190,6 @@ public class ColorGrading {
          * When color grading is implemented using a 3D LUT, this sets the dimension of the LUT.
          * This overrides the value set by quality().
          *
-         * This setting has no effect if generating a 1D LUT.
-         *
          * The default is 32
          *
          * @param dim The desired dimension of the LUT. Between 16 and 64.
@@ -623,3 +616,4 @@ public class ColorGrading {
 
     private static native long nBuilderBuild(long nativeBuilder, long nativeEngine);
 }
+

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

@@ -31,6 +31,8 @@ package com.google.android.filament;
  *   <li>DisplayRangeToneMapper</li>
  * </ul>
  * </ul>
+ *
+ * You can create custom tone mapping operators by subclassing ToneMapper.
  */
 public class ToneMapper {
     private final long mNativeObject;

android/gradle.properties

@@ -1,5 +1,5 @@
 GROUP=com.google.android.filament
-VERSION_NAME=1.58.2
+VERSION_NAME=1.58.0
 
 POM_DESCRIPTION=Real-time physically based rendering engine for Android.
 

filament/backend/include/backend/DriverEnums.h

@@ -256,18 +256,16 @@ struct DescriptorSetLayoutBinding {
     DescriptorFlags flags = DescriptorFlags::NONE;
     uint16_t count = 0;
 
-    //  TODO: uncomment when needed.  Note that this class is used as hash key.  We need to ensure
-    //  no uninitialized padding bytes.
-    //    uint8_t externalSamplerDataIndex = EXTERNAL_SAMPLER_DATA_INDEX_UNUSED;
+    uint8_t externalSamplerDataIndex = EXTERNAL_SAMPLER_DATA_INDEX_UNUSED;
 
-    friend inline bool operator==(DescriptorSetLayoutBinding const& lhs,
+    friend inline bool operator==(
+            DescriptorSetLayoutBinding const& lhs,
             DescriptorSetLayoutBinding const& rhs) noexcept {
         return lhs.type == rhs.type &&
                lhs.flags == rhs.flags &&
                lhs.count == rhs.count &&
-               lhs.stageFlags == rhs.stageFlags;
-//               lhs.stageFlags == rhs.stageFlags &&
-//               lhs.externalSamplerDataIndex == rhs.externalSamplerDataIndex;
+               lhs.stageFlags == rhs.stageFlags &&
+               lhs.externalSamplerDataIndex == rhs.externalSamplerDataIndex;
     }
 };
 
@@ -1140,9 +1138,7 @@ static_assert(sizeof(ExternalSamplerDatum) == 12);
 
 struct DescriptorSetLayout {
     utils::FixedCapacityVector<DescriptorSetLayoutBinding> bindings;
-
-//  TODO: uncomment when needed
-//    utils::FixedCapacityVector<ExternalSamplerDatum> externalSamplerData;
+    utils::FixedCapacityVector<ExternalSamplerDatum> externalSamplerData;
 };
 
 //! blending equation function

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/PlatformEGLAndroid.h

@@ -23,7 +23,6 @@
 #include <backend/platforms/PlatformEGL.h>
 
 #include <utils/android/PerformanceHintManager.h>
-#include <utils/compiler.h>
 
 #include <chrono>
 
@@ -44,21 +43,8 @@ public:
     PlatformEGLAndroid() noexcept;
     ~PlatformEGLAndroid() noexcept override;
 
-    /**
-     * Creates an ExternalImage from a EGLImageKHR
-     */
-    ExternalImageHandle createExternalImage(AHardwareBuffer const* buffer, bool sRGB) noexcept;
-
-    struct ExternalImageDescAndroid {
-        uint32_t width;      // Texture width
-        uint32_t height;     // Texture height
-        TextureFormat format;// Texture format
-        TextureUsage usage;  // Texture usage flags
-    };
-
-    ExternalImageDescAndroid getExternalImageDesc(ExternalImageHandle externalImage) noexcept;
-
 protected:
+
     // --------------------------------------------------------------------------------------------
     // Platform Interface
 
@@ -71,6 +57,11 @@ protected:
     Driver* createDriver(void* sharedContext,
             const Platform::DriverConfig& driverConfig) noexcept override;
 
+    /**
+     * Creates an ExternalImage from a EGLImageKHR
+     */
+    ExternalImageHandle createExternalImage(AHardwareBuffer const *buffer, bool sRGB) noexcept;
+
     // --------------------------------------------------------------------------------------------
     // OpenGLPlatform Interface
 
@@ -104,26 +95,15 @@ protected:
      */
     AcquiredImage transformAcquiredImage(AcquiredImage source) noexcept override;
 
-    OpenGLPlatform::ExternalTexture* createExternalImageTexture() noexcept override;
-    void destroyExternalImageTexture(ExternalTexture* texture) noexcept override;
+    bool setExternalImage(ExternalImageHandleRef externalImage, ExternalTexture* texture) noexcept override;
 
     struct ExternalImageEGLAndroid : public ExternalImageEGL {
         AHardwareBuffer* aHardwareBuffer = nullptr;
-        uint32_t width;      // Texture width
-        uint32_t height;     // Texture height
-        TextureFormat format;// Texture format
-        TextureUsage usage;  // Texture usage flags
         bool sRGB = false;
-
     protected:
         ~ExternalImageEGLAndroid() override;
     };
 
-    bool setExternalImage(ExternalImageHandleRef externalImage,
-            ExternalTexture* texture) noexcept override;
-    bool setImage(ExternalImageEGLAndroid const* eglExternalImage,
-            ExternalTexture* texture) noexcept;
-
 protected:
     bool makeCurrent(ContextType type,
             SwapChain* drawSwapChain,

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

@@ -368,16 +368,8 @@ public:
             uint32_t internalFormat, VkImage image, VkImageSubresourceRange range,
             VkImageViewType viewType, VkComponentMapping swizzle);
 
-protected:
-    virtual ExtensionSet getSwapchainInstanceExtensions() const;
-
-    using SurfaceBundle = std::tuple<VkSurfaceKHR, VkExtent2D>;
-    virtual SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
-            uint64_t flags) const noexcept;
-
 private:
-    // Platform dependent helper methods
-    static ExtensionSet getSwapchainInstanceExtensionsImpl();
+    static ExtensionSet getSwapchainInstanceExtensions();
 
     static ExternalImageMetadata getExternalImageMetadataImpl(ExternalImageHandleRef externalImage,
             VkDevice device);
@@ -394,7 +386,8 @@ private:
             VkComponentMapping swizzle);
 
     // Platform dependent helper methods
-    static SurfaceBundle createVkSurfaceKHRImpl(void* nativeWindow, VkInstance instance,
+    using SurfaceBundle = std::tuple<VkSurfaceKHR, VkExtent2D>;
+    static SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
             uint64_t flags) noexcept;
 
     friend struct VulkanPlatformPrivate;

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

@@ -17,45 +17,27 @@
 #ifndef TNT_FILAMENT_BACKEND_PLATFORMS_VULKAN_PLATFORM_ANDROID_H
 #define TNT_FILAMENT_BACKEND_PLATFORMS_VULKAN_PLATFORM_ANDROID_H
 
-#include <backend/DriverEnums.h>
-#include <backend/platforms/VulkanPlatform.h>
+#include <backend/Platform.h>
 
 #include <android/hardware_buffer.h>
 
-namespace filament::backend {
+namespace filament::backend::fvkandroid {
 
-class VulkanPlatformAndroid : public VulkanPlatform {
-public:
-    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;
-    };
-
-    Platform::ExternalImageHandle createExternalImage(AHardwareBuffer const* buffer,
-            bool sRGB) noexcept;
-
-    virtual ExternalImageMetadata getExternalImageMetadata(ExternalImageHandleRef externalImage);
-
-    using ImageData = VulkanPlatform::ImageData;
-    virtual ImageData createExternalImageData(ExternalImageHandleRef externalImage,
-            const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex,
-            VkImageUsageFlags usage);
+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:
-    virtual ExtensionSet getSwapchainInstanceExtensions() const;
-
-    using SurfaceBundle = VulkanPlatform::SurfaceBundle;
-    virtual SurfaceBundle createVkSurfaceKHR(void* nativeWindow, VkInstance instance,
-            uint64_t flags) const noexcept;
+    ~ExternalImageVulkanAndroid() override;
 };
 
-}// namespace filament::backend
+Platform::ExternalImageHandle createExternalImage(AHardwareBuffer const* buffer,
+        bool sRGB) noexcept;
 
-#endif// TNT_FILAMENT_BACKEND_PLATFORMS_VULKAN_PLATFORM_ANDROID_H
+} // namespace filament::backend::fvkandroid
+
+#endif // TNT_FILAMENT_BACKEND_PLATFORMS_VULKAN_PLATFORM_ANDROID_H

filament/backend/src/PlatformFactory.cpp

@@ -55,11 +55,7 @@
 #endif
 
 #if defined(FILAMENT_DRIVER_SUPPORTS_VULKAN)
-    #if defined(__ANDROID__)
-        #include "backend/platforms/VulkanPlatformAndroid.h"
-    #else
-        #include "backend/platforms/VulkanPlatform.h"
-    #endif
+    #include "backend/platforms/VulkanPlatform.h"
 #endif
 
 #if defined (FILAMENT_SUPPORTS_METAL)
@@ -108,11 +104,7 @@ Platform* PlatformFactory::create(Backend* backend) noexcept {
     }
     if (*backend == Backend::VULKAN) {
         #if defined(FILAMENT_DRIVER_SUPPORTS_VULKAN)
-            #if defined(__ANDROID__)
-                return new VulkanPlatformAndroid();
-            #else
-                return new VulkanPlatform();
-            #endif
+            return new VulkanPlatform();
         #else
             return nullptr;
         #endif

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/noop/NoopDriver.cpp

@@ -241,7 +241,7 @@ size_t NoopDriver::getMaxUniformBufferSize() {
 }
 
 size_t NoopDriver::getMaxTextureSize(SamplerType target) {
-    return 4096u;
+    return 2048u;
 }
 
 size_t NoopDriver::getMaxArrayTextureLayers() {

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

@@ -19,7 +19,6 @@
 #include <backend/platforms/PlatformEGL.h>
 #include <backend/platforms/PlatformEGLAndroid.h>
 
-#include <private/backend/BackendUtilsAndroid.h>
 #include <private/backend/VirtualMachineEnv.h>
 
 #include "opengl/GLUtils.h"
@@ -35,8 +34,6 @@
 #include <utils/ostream.h>
 #include <utils/Panic.h>
 #include <utils/Log.h>
-#include <utils/compiler.h>
-#include <utils/ostream.h>
 
 #include <EGL/egl.h>
 #include <EGL/eglext.h>
@@ -234,133 +231,26 @@ Driver* PlatformEGLAndroid::createDriver(void* sharedContext,
     return driver;
 }
 
-PlatformEGLAndroid::ExternalImageEGLAndroid::~ExternalImageEGLAndroid() {
-    if (__builtin_available(android 26, *)) {
-        if (aHardwareBuffer) {
-            AHardwareBuffer_release(aHardwareBuffer);
-        }
-    }
-}
+PlatformEGLAndroid::ExternalImageEGLAndroid::~ExternalImageEGLAndroid() = default;
 
-Platform::ExternalImageHandle PlatformEGLAndroid::createExternalImage(AHardwareBuffer const* buffer,
-        bool sRGB) noexcept {
-    if (__builtin_available(android 26, *)) {
-        auto* const p = new (std::nothrow) ExternalImageEGLAndroid;
-        auto hardwareBuffer = const_cast<AHardwareBuffer*>(buffer);
-        AHardwareBuffer_acquire(hardwareBuffer);
-        p->aHardwareBuffer = hardwareBuffer;
-        p->sRGB = sRGB;
-        AHardwareBuffer_Desc hardwareBufferDescription = {};
-        AHardwareBuffer_describe(hardwareBuffer, &hardwareBufferDescription);
-        p->height = hardwareBufferDescription.height;
-        p->width = hardwareBufferDescription.width;
-        auto textureFormat = mapToFilamentFormat(hardwareBufferDescription.format, sRGB);
-        p->usage = mapToFilamentUsage(hardwareBufferDescription.usage, textureFormat);
-        return ExternalImageHandle{ p };
-    }
-
-    return Platform::ExternalImageHandle{};
-}
-
-PlatformEGLAndroid::ExternalImageDescAndroid PlatformEGLAndroid::getExternalImageDesc(
-        ExternalImageHandle externalImage) noexcept {
-    auto const* const eglExternalImage =
-            static_cast<ExternalImageEGLAndroid const*>(externalImage.get());
-    ExternalImageDescAndroid metadata = {};
-    if (!eglExternalImage) {
-        return metadata;
-    }
-    metadata.height = eglExternalImage->height;
-    metadata.width = eglExternalImage->width;
-    metadata.format = eglExternalImage->format;
-    metadata.usage = eglExternalImage->usage;
-    return metadata;
+Platform::ExternalImageHandle PlatformEGLAndroid::createExternalImage(AHardwareBuffer const* buffer, bool sRGB) noexcept {
+    auto* const p = new(std::nothrow) ExternalImageEGLAndroid;
+    p->aHardwareBuffer = const_cast<AHardwareBuffer*>(buffer);
+    p->sRGB = sRGB;
+    return ExternalImageHandle{ p };
 }
 
 bool PlatformEGLAndroid::setExternalImage(ExternalImageHandleRef externalImage,
         UTILS_UNUSED_IN_RELEASE ExternalTexture* texture) noexcept {
-    auto const* const eglExternalImage =
-            static_cast<ExternalImageEGLAndroid const*>(externalImage.get());
+    auto const* const eglExternalImage = static_cast<ExternalImageEGLAndroid const*>(externalImage.get());
     if (eglExternalImage->aHardwareBuffer) {
-        return PlatformEGLAndroid::setImage(eglExternalImage, texture);
+        // TODO: implement PlatformEGLAndroid::setExternalImage w/ AHardwareBuffer
+        return true;
     }
     // not a AHardwareBuffer, fallback to the inherited version
     return PlatformEGL::setExternalImage(externalImage, texture);
 }
 
-OpenGLPlatform::ExternalTexture* PlatformEGLAndroid::createExternalImageTexture() noexcept {
-    ExternalTexture* outTexture = new (std::nothrow) ExternalTexture{};
-    glGenTextures(1, &outTexture->id);
-    return outTexture;
-}
-
-void PlatformEGLAndroid::destroyExternalImageTexture(ExternalTexture* texture) noexcept {
-    glDeleteTextures(1, &texture->id);
-    delete texture;
-}
-
-bool PlatformEGLAndroid::setImage(ExternalImageEGLAndroid const* eglExternalImage,
-        UTILS_UNUSED_IN_RELEASE ExternalTexture* texture) noexcept {
-    AHardwareBuffer* hardwareBuffer = eglExternalImage->aHardwareBuffer;
-
-    // Get the EGL client buffer from AHardwareBuffer
-    EGLClientBuffer clientBuffer = eglGetNativeClientBufferANDROID(hardwareBuffer);
-    EGLint imageAttrs[] = {
-        EGL_IMAGE_PRESERVED_KHR, EGL_TRUE,
-        EGL_NONE, EGL_NONE,  // Reserve space
-        EGL_NONE, EGL_NONE,  // Reserve space
-        EGL_NONE             // Ensure the list always ends with EGL_NONE
-    };
-    int attrIndex = 2;
-    if (eglExternalImage->sRGB) {
-        imageAttrs[attrIndex++] = EGL_GL_COLORSPACE;
-        imageAttrs[attrIndex++] = EGL_GL_COLORSPACE_SRGB;
-    }
-
-    if (static_cast<bool>(eglExternalImage->usage & TextureUsage::PROTECTED)) {
-        imageAttrs[attrIndex++] = EGL_PROTECTED_CONTENT_EXT;
-        imageAttrs[attrIndex++] = EGL_TRUE;
-    }
-    // Create an EGLImage from the client buffer
-    EGLImageKHR eglImage = eglCreateImageKHR(eglGetCurrentDisplay(), EGL_NO_CONTEXT,
-            EGL_NATIVE_BUFFER_ANDROID, clientBuffer, imageAttrs);
-    if (eglImage == EGL_NO_IMAGE_KHR) {
-        // Handle error
-        slog.e << "Failed to create EGL image" << io::endl;
-        glDeleteTextures(1, &texture->id);
-        return false;
-    }
-
-    // Create and bind the OpenGL texture
-    GLint prevActiveTexture, prevTexture;
-    glGetIntegerv(GL_ACTIVE_TEXTURE, &prevActiveTexture);
-    glGetIntegerv(GL_TEXTURE_BINDING_2D, &prevTexture);
-    glActiveTexture(GL_TEXTURE0);
-    glBindTexture(texture->target, texture->id);
-    GLenum error = glGetError();
-    if (UTILS_UNLIKELY(error != GL_NO_ERROR)) {
-        slog.e << "Error after glBindTexture: " << error << io::endl;
-        glDeleteTextures(1, &texture->id);
-        eglDestroyImageKHR(eglGetCurrentDisplay(), eglImage);
-        glActiveTexture(prevActiveTexture);
-        glBindTexture(GL_TEXTURE_2D, prevTexture);
-        return false;
-    }
-    glEGLImageTargetTexture2DOES(texture->target, static_cast<GLeglImageOES>(eglImage));
-    error = glGetError();
-    if (UTILS_UNLIKELY(error != GL_NO_ERROR)) {
-        slog.e << "Error after glEGLImageTargetTexture2DOES: " << error << io::endl;
-        glDeleteTextures(1, &texture->id);
-        eglDestroyImageKHR(eglGetCurrentDisplay(), eglImage);
-        glActiveTexture(prevActiveTexture);
-        glBindTexture(GL_TEXTURE_2D, prevTexture);
-        return false;
-    }
-    glActiveTexture(prevActiveTexture);
-    glBindTexture(GL_TEXTURE_2D, prevTexture);
-    return true;
-}
-
 void PlatformEGLAndroid::setPresentationTime(int64_t presentationTimeInNanosecond) noexcept {
     EGLSurface currentDrawSurface = eglGetCurrentSurface(EGL_DRAW);
     if (currentDrawSurface != EGL_NO_SURFACE) {

filament/backend/src/vulkan/VulkanDriver.cpp

@@ -1149,7 +1149,7 @@ size_t VulkanDriver::getMaxUniformBufferSize() {
 
 size_t VulkanDriver::getMaxTextureSize(SamplerType) {
     // TODO: return the actual size instead of hardcoded value
-    return 4096;
+    return 2048;
 }
 
 size_t VulkanDriver::getMaxArrayTextureLayers() {

filament/backend/src/vulkan/VulkanTexture.cpp

@@ -55,61 +55,63 @@ VkComponentMapping composeSwizzle(VkComponentMapping const& prev, VkComponentMap
         VK_COMPONENT_SWIZZLE_A,
     };
 
-    auto const compose = [](VkComponentMapping const& prev,
-                                 VkComponentMapping const& next) -> VkComponentMapping {
-        VkComponentSwizzle vals[4] = { next.r, next.g, next.b, next.a };
-        for (auto& out: vals) {
-            switch (out) {
-                case VK_COMPONENT_SWIZZLE_R:
-                    out = prev.r;
-                    break;
-                case VK_COMPONENT_SWIZZLE_G:
-                    out = prev.g;
-                    break;
-                case VK_COMPONENT_SWIZZLE_B:
-                    out = prev.b;
-                    break;
-                case VK_COMPONENT_SWIZZLE_A:
-                    out = prev.a;
-                    break;
-                // Do not modify the value
-                case VK_COMPONENT_SWIZZLE_IDENTITY:
-                case VK_COMPONENT_SWIZZLE_ZERO:
-                case VK_COMPONENT_SWIZZLE_ONE:
-                // Below is not exposed in Vulkan's API, but needs to be there for compilation.
-                case VK_COMPONENT_SWIZZLE_MAX_ENUM:
-                    break;
-            }
+    auto const compose = [](VkComponentSwizzle out, VkComponentMapping const& prev,
+                                 uint8_t channelIndex) {
+        // We need to first change all identities to its equivalent channel.
+        if (out == VK_COMPONENT_SWIZZLE_IDENTITY) {
+            out = IDENTITY[channelIndex];
         }
-        return { vals[0], vals[1], vals[2], vals[3] };
+        switch (out) {
+            case VK_COMPONENT_SWIZZLE_R:
+                out = prev.r;
+                break;
+            case VK_COMPONENT_SWIZZLE_G:
+                out = prev.g;
+                break;
+            case VK_COMPONENT_SWIZZLE_B:
+                out = prev.b;
+                break;
+            case VK_COMPONENT_SWIZZLE_A:
+                out = prev.a;
+                break;
+            case VK_COMPONENT_SWIZZLE_IDENTITY:
+            case VK_COMPONENT_SWIZZLE_ZERO:
+            case VK_COMPONENT_SWIZZLE_ONE:
+                return out;
+            // Below is not exposed in Vulkan's API, but needs to be there for compilation.
+            case VK_COMPONENT_SWIZZLE_MAX_ENUM:
+                break;
+        }
+        // If the result correctly corresponds to the identity, just return identity.
+        if (IDENTITY[channelIndex] == out) {
+            return VK_COMPONENT_SWIZZLE_IDENTITY;
+        }
+        return out;
     };
 
-    auto const identityToChannel = [](VkComponentMapping const& mapping) -> VkComponentMapping {
-        VkComponentSwizzle vals[4] = { mapping.r, mapping.g, mapping.b, mapping.a };
-        for (uint8_t i = 0; i < 4; i++) {
-            if (vals[i] != VK_COMPONENT_SWIZZLE_IDENTITY) {
-                continue;
-            }
-            vals[i] = IDENTITY[i];
+    auto const identityToChannel = [](VkComponentSwizzle val, uint8_t channelIndex) {
+        if (val != VK_COMPONENT_SWIZZLE_IDENTITY) {
+            return val;
         }
-        return { vals[0], vals[1], vals[2], vals[3] };
-    };
-    auto const channelToIdentity = [](VkComponentMapping const& mapping) -> VkComponentMapping {
-        VkComponentSwizzle vals[4] = { mapping.r, mapping.g, mapping.b, mapping.a };
-        for (uint8_t i = 0; i < 4; i++) {
-            if (IDENTITY[i] != vals[i]) {
-                continue;
-            }
-            vals[i] = VK_COMPONENT_SWIZZLE_IDENTITY;
-        }
-        return { vals[0], vals[1], vals[2], vals[3] };
+        return IDENTITY[channelIndex];
     };
 
     // We make sure all all identities are mapped into respective channels so that actual channel
     // mapping will be passed onto the output.
-    VkComponentMapping const prevExplicit = identityToChannel(prev);
-    VkComponentMapping const nextExplicit = identityToChannel(next);
-    return channelToIdentity(compose(prevExplicit, nextExplicit));
+    VkComponentMapping const prevExplicit = {
+            identityToChannel(prev.r, 0),
+            identityToChannel(prev.g, 1),
+            identityToChannel(prev.b, 2),
+            identityToChannel(prev.a, 3),
+    };
+
+    // Note that the channel index corresponds to the VkComponentMapping struct layout.
+    return {
+        compose(next.r, prevExplicit, 0),
+        compose(next.g, prevExplicit, 1),
+        compose(next.b, prevExplicit, 2),
+        compose(next.a, prevExplicit, 3),
+    };
 }
 
 inline VulkanLayout getDefaultLayoutImpl(TextureUsage usage) {
@@ -151,6 +153,7 @@ uint8_t getLayerCountFromDepth(uint32_t const depth) {
     return getLayerCount(getSamplerTypeFromDepth(depth), depth);
 }
 
+
 } // anonymous namespace
 
 VulkanTextureState::VulkanTextureState(VkDevice device, VmaAllocator allocator,
@@ -246,19 +249,12 @@ VulkanTexture::VulkanTexture(VkDevice device, VkPhysicalDevice physicalDevice,
 
     // Determine if we can use the transient usage flag combined with lazily allocated memory.
     const bool useTransientAttachment =
-            // Lazily allocated memory is available.
-            context.isLazilyAllocatedMemorySupported() &&
-            // Usage consists of attachment flags only.
-            none(tusage & ~TextureUsage::ALL_ATTACHMENTS) &&
-            // Usage contains at least one attachment flag.
-            any(tusage & TextureUsage::ALL_ATTACHMENTS) &&
-            // Depth resolve cannot use transient attachment because it uses a custom shader.
-            // TODO: see VulkanDriver::isDepthStencilResolveSupported() to know when to remove this
-            // restriction.
-            // Note that the custom shader does not resolve stencil. We do need to move to vk 1.2
-            // and above to be able to support stencil resolve (along with depth).
-            !(any(usage & TextureUsage::DEPTH_ATTACHMENT) && samples > 1);
-
+        // Lazily allocated memory is available.
+        context.isLazilyAllocatedMemorySupported() &&
+        // Usage consists of attachment flags only.
+        none(tusage & ~TextureUsage::ALL_ATTACHMENTS) &&
+        // Usage contains at least one attachment flag.
+        any(tusage & TextureUsage::ALL_ATTACHMENTS);
     mState->mIsTransientAttachment = useTransientAttachment;
 
     const VkImageUsageFlags transientFlag =

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

@@ -1000,15 +1000,6 @@ VkImageView VulkanPlatform::createExternalImageView(SamplerYcbcrConversion chrom
     return createExternalImageViewImpl(mImpl->mDevice, chroma, internalFormat, image, range,
             viewType, swizzle);
 }
-
-ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() const {
-    return getSwapchainInstanceExtensionsImpl();
-}
-
-VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHR(void* nativeWindow,
-        VkInstance instance, uint64_t flags) const noexcept {
-    return createVkSurfaceKHRImpl(nativeWindow, instance, flags);
-}
 #undef SWAPCHAIN_RET_FUNC
 
 }// namespace filament::backend

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

@@ -13,9 +13,10 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#include <backend/platforms/VulkanPlatformAndroid.h>
+#include <backend/platforms/VulkanPlatform.h>
 
 #include <backend/DriverEnums.h>
+#include <backend/platforms/VulkanPlatformAndroid.h>
 #include <private/backend/BackendUtilsAndroid.h>
 
 #include "vulkan/VulkanConstants.h"
@@ -156,7 +157,7 @@ VulkanPlatform::ImageData allocateExternalImage(AHardwareBuffer* buffer, VkDevic
         .handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID,
     };
 
-    VkImageCreateInfo imageInfo{ .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO };
+    VkImageCreateInfo imageInfo{.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO};
     imageInfo.pNext = &externalCreateInfo;
     imageInfo.format = metadata.format;
     imageInfo.imageType = VK_IMAGE_TYPE_2D;
@@ -199,12 +200,14 @@ VulkanPlatform::ImageData allocateExternalImage(AHardwareBuffer* buffer, VkDevic
     return data;
 }
 
-}// namespace
+} // namespace
 
-VulkanPlatformAndroid::ExternalImageVulkanAndroid::~ExternalImageVulkanAndroid() = default;
+namespace fvkandroid {
 
-Platform::ExternalImageHandle VulkanPlatformAndroid::createExternalImage(
-        AHardwareBuffer const* buffer, bool sRGB) noexcept {
+ExternalImageVulkanAndroid::~ExternalImageVulkanAndroid() = default;
+
+Platform::ExternalImageHandle createExternalImage(AHardwareBuffer const* buffer,
+        bool sRGB) noexcept {
     if (__builtin_available(android 26, *)) {
         AHardwareBuffer_Desc hardwareBufferDescription = {};
         AHardwareBuffer_describe(buffer, &hardwareBufferDescription);
@@ -223,10 +226,12 @@ Platform::ExternalImageHandle VulkanPlatformAndroid::createExternalImage(
     return Platform::ExternalImageHandle{};
 }
 
-VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::getExternalImageMetadata(
-        ExternalImageHandleRef externalImage) {
+} // namespace fvkandroid
+
+VulkanPlatform::ExternalImageMetadata VulkanPlatform::getExternalImageMetadataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device) {
     auto const* fvkExternalImage =
-            static_cast<ExternalImageVulkanAndroid const*>(externalImage.get());
+            static_cast<fvkandroid::ExternalImageVulkanAndroid const*>(externalImage.get());
 
     ExternalImageMetadata metadata;
     AHardwareBuffer* buffer = fvkExternalImage->aHardwareBuffer;
@@ -240,7 +245,7 @@ VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::getExternalImageMet
         std::tie(metadata.format, metadata.usage) =
                 getVKFormatAndUsage(bufferDesc, fvkExternalImage->sRGB);
     }
-
+    
     metadata.samples = VK_SAMPLE_COUNT_1_BIT;
 
     VkAndroidHardwareBufferFormatPropertiesANDROID formatInfo = {
@@ -251,11 +256,11 @@ VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::getExternalImageMet
         .sType = VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_PROPERTIES_ANDROID,
         .pNext = &formatInfo,
     };
-    VkResult result = vkGetAndroidHardwareBufferPropertiesANDROID(getDevice(), buffer, &properties);
+    VkResult result = vkGetAndroidHardwareBufferPropertiesANDROID(device, buffer, &properties);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
-            << "vkGetAndroidHardwareBufferProperties failed with error="
-            << static_cast<int32_t>(result);
-
+        << "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("
@@ -266,14 +271,14 @@ VulkanPlatform::ExternalImageMetadata VulkanPlatformAndroid::getExternalImageMet
     return metadata;
 }
 
-VulkanPlatformAndroid::ImageData VulkanPlatformAndroid::createExternalImageData(
-        ExternalImageHandleRef externalImage, const ExternalImageMetadata& metadata,
-        uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
+VulkanPlatform::ImageData VulkanPlatform::createExternalImageDataImpl(
+        ExternalImageHandleRef externalImage, VkDevice device,
+        const ExternalImageMetadata& metadata, uint32_t memoryTypeIndex, VkImageUsageFlags usage) {
     auto const* fvkExternalImage =
-            static_cast<ExternalImageVulkanAndroid const*>(externalImage.get());
-    ImageData data = allocateExternalImage(fvkExternalImage->aHardwareBuffer, getDevice(), metadata,
+            static_cast<fvkandroid::ExternalImageVulkanAndroid const*>(externalImage.get());
+    ImageData data = allocateExternalImage(fvkExternalImage->aHardwareBuffer, device, metadata,
             memoryTypeIndex, usage);
-    VkResult result = vkBindImageMemory(getDevice(), data.first, data.second, 0);
+    VkResult result = vkBindImageMemory(device, data.first, data.second, 0);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
         << "vkBindImageMemory error=" << static_cast<int32_t>(result);
     return data;
@@ -392,14 +397,14 @@ VkSampler VulkanPlatform::createExternalSamplerImpl(
     return sampler;
 }
 
-VulkanPlatform::ExtensionSet VulkanPlatformAndroid::getSwapchainInstanceExtensions() const {
+VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() {
     return {
         VK_KHR_ANDROID_SURFACE_EXTENSION_NAME,
     };
 }
 
-VulkanPlatform::SurfaceBundle VulkanPlatformAndroid::createVkSurfaceKHR(void* nativeWindow,
-        VkInstance instance, uint64_t flags) const noexcept {
+VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHR(void* nativeWindow,
+    VkInstance instance, uint64_t flags) noexcept {
     VkSurfaceKHR surface;
     VkExtent2D extent;
 
@@ -411,26 +416,6 @@ VulkanPlatform::SurfaceBundle VulkanPlatformAndroid::createVkSurfaceKHR(void* na
             vkCreateAndroidSurfaceKHR(instance, &createInfo, VKALLOC, (VkSurfaceKHR*) &surface);
     FILAMENT_CHECK_POSTCONDITION(result == VK_SUCCESS)
             << "vkCreateAndroidSurfaceKHR with error=" << static_cast<int32_t>(result);
-    return { surface, extent };
+    return {surface, extent};
 }
-
-// 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

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

@@ -52,7 +52,7 @@ using namespace bluevk;
 
 namespace filament::backend {
 
-VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl() {
+VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() {
     ExtensionSet const ret = {
     #if defined(__APPLE__)
         VK_MVK_MACOS_SURFACE_EXTENSION_NAME,  // TODO: replace with VK_EXT_metal_surface
@@ -87,7 +87,7 @@ VkImageView VulkanPlatform::createExternalImageViewImpl(VkDevice device,
     return VK_NULL_HANDLE;
 }
 
-VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHRImpl(void* nativeWindow,
+VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHR(void* nativeWindow,
         VkInstance instance, uint64_t flags) noexcept {
     VkSurfaceKHR surface;
     #if defined(__APPLE__)

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

@@ -108,7 +108,7 @@ VkImageView VulkanPlatform::createExternalImageViewImpl(VkDevice device,
     return VK_NULL_HANDLE;
 }
 
-VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl() {
+VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensions() {
     VulkanPlatform::ExtensionSet const ret = {
 #if defined(__linux__) && defined(FILAMENT_SUPPORTS_WAYLAND)
         VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME,
@@ -126,7 +126,7 @@ VulkanPlatform::ExtensionSet VulkanPlatform::getSwapchainInstanceExtensionsImpl(
     return ret;
 }
 
-VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHRImpl(void* nativeWindow,
+VulkanPlatform::SurfaceBundle VulkanPlatform::createVkSurfaceKHR(void* nativeWindow,
         VkInstance instance, uint64_t flags) noexcept {
     VkSurfaceKHR surface;
 

filament/backend/src/webgpu/WebGPUDriver.cpp

@@ -651,11 +651,11 @@ uint8_t WebGPUDriver::getMaxDrawBuffers() {
     return MRT::MAX_SUPPORTED_RENDER_TARGET_COUNT;
 }
 
-size_t WebGPUDriver::getMaxUniformBufferSize() {
+size_t WebGPUDriver::getMaxUniformBufferSize(SamplerType) {
     return 16384u;
 }
 
-size_t WebGPUDriver::getMaxTextureSize(SamplerType target) {
+size_t WebGPUDriver::getMaxTextureSize() {
     return 2048u;
 }
 

filament/include/filament/ColorGrading.h

@@ -64,8 +64,9 @@ class ColorSpace;
  * Performance
  * ===========
  *
- * Creating a new ColorGrading object may be more expensive than other Filament objects as a LUT may
- * need to be generated. The generation of this LUT, if necessary, may happen on the CPU.
+ * Creating a new ColorGrading object may be more expensive than other Filament objects as a
+ * 3D LUT may need to be generated. The generation of a 3D LUT, if necessary, may happen on
+ * the CPU.
  *
  * Ordering
  * ========
@@ -154,9 +155,6 @@ public:
          * 3D texture. For instance, a low quality level will use a 16x16x16 10 bit LUT, a medium
          * quality level will use a 32x32x32 10 bit LUT, a high quality will use a 32x32x32 16 bit
          * LUT, and a ultra quality will use a 64x64x64 16 bit LUT.
-         *
-         * This setting has no effect if generating a 1D LUT.
-         *
          * This overrides the values set by format() and dimensions().
          *
          * The default quality is medium.
@@ -171,8 +169,6 @@ public:
          * When color grading is implemented using a 3D LUT, this sets the texture format of
          * of the LUT. This overrides the value set by quality().
          *
-         * This setting has no effect if generating a 1D LUT.
-         *
          * The default is INTEGER
          *
          * @param format The desired format of the 3D LUT.
@@ -185,8 +181,6 @@ public:
          * When color grading is implemented using a 3D LUT, this sets the dimension of the LUT.
          * This overrides the value set by quality().
          *
-         * This setting has no effect if generating a 1D LUT.
-         *
          * The default is 32
          *
          * @param dim The desired dimension of the LUT. Between 16 and 64.

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/include/filament/ToneMapper.h

@@ -68,21 +68,6 @@ struct UTILS_PUBLIC ToneMapper {
      *         function applied ("linear")
      */
     virtual math::float3 operator()(math::float3 c) const noexcept = 0;
-
-    /**
-     * If true, then this function holds that f(x) = vec3(f(x.r), f(x.g), f(x.b))
-     *
-     * This may be used to indicate that the color grading's LUT only requires a 1D texture instead
-     * of a 3D texture, potentially saving a significant amount of memory and generation time.
-     */
-    virtual bool isOneDimensional() const noexcept { return false; }
-
-    /**
-     * True if this tonemapper only works in low-dynamic-range.
-     *
-     * This may be used to indicate that the color grading's LUT doesn't need to be log encoded.
-     */
-    virtual bool isLDR() const noexcept { return false; }
 };
 
 /**
@@ -94,8 +79,6 @@ struct UTILS_PUBLIC LinearToneMapper final : public ToneMapper {
     ~LinearToneMapper() noexcept final;
 
     math::float3 operator()(math::float3 c) const noexcept override;
-    bool isOneDimensional() const noexcept override { return true; }
-    bool isLDR() const noexcept override { return true; }
 };
 
 /**
@@ -108,8 +91,6 @@ struct UTILS_PUBLIC ACESToneMapper final : public ToneMapper {
     ~ACESToneMapper() noexcept final;
 
     math::float3 operator()(math::float3 c) const noexcept override;
-    bool isOneDimensional() const noexcept override { return false; }
-    bool isLDR() const noexcept override { return false; }
 };
 
 /**
@@ -123,8 +104,6 @@ struct UTILS_PUBLIC ACESLegacyToneMapper final : public ToneMapper {
     ~ACESLegacyToneMapper() noexcept final;
 
     math::float3 operator()(math::float3 c) const noexcept override;
-    bool isOneDimensional() const noexcept override { return false; }
-    bool isLDR() const noexcept override { return false; }
 };
 
 /**
@@ -138,8 +117,6 @@ struct UTILS_PUBLIC FilmicToneMapper final : public ToneMapper {
     ~FilmicToneMapper() noexcept final;
 
     math::float3 operator()(math::float3 x) const noexcept override;
-    bool isOneDimensional() const noexcept override { return true; }
-    bool isLDR() const noexcept override { return false; }
 };
 
 /**
@@ -152,8 +129,6 @@ struct UTILS_PUBLIC PBRNeutralToneMapper final : public ToneMapper {
     ~PBRNeutralToneMapper() noexcept final;
 
     math::float3 operator()(math::float3 x) const noexcept override;
-    bool isOneDimensional() const noexcept override { return false; }
-    bool isLDR() const noexcept override { return false; }
 };
 
 /**
@@ -175,8 +150,6 @@ struct UTILS_PUBLIC AgxToneMapper final : public ToneMapper {
     ~AgxToneMapper() noexcept final;
 
     math::float3 operator()(math::float3 x) const noexcept override;
-    bool isOneDimensional() const noexcept override { return false; }
-    bool isLDR() const noexcept override { return false; }
 
     AgxLook look;
 };
@@ -221,8 +194,6 @@ struct UTILS_PUBLIC GenericToneMapper final : public ToneMapper {
     GenericToneMapper& operator=(GenericToneMapper&& rhs) noexcept;
 
     math::float3 operator()(math::float3 x) const noexcept override;
-    bool isOneDimensional() const noexcept override { return true; }
-    bool isLDR() const noexcept override { return false; }
 
     /** Returns the contrast of the curve as a strictly positive value. */
     float getContrast() const noexcept;
@@ -285,8 +256,6 @@ struct UTILS_PUBLIC DisplayRangeToneMapper final : public ToneMapper {
     ~DisplayRangeToneMapper() noexcept override;
 
     math::float3 operator()(math::float3 c) const noexcept override;
-    bool isOneDimensional() const noexcept override { return false; }
-    bool isLDR() const noexcept override { return false; }
 };
 
 } // namespace filament

filament/src/PostProcessManager.cpp

@@ -2306,11 +2306,36 @@ void PostProcessManager::colorGradingPrepareSubpass(DriverApi& driver,
         const FColorGrading* colorGrading, ColorGradingConfig const& colorGradingConfig,
         VignetteOptions const& vignetteOptions, uint32_t const width, uint32_t const height) noexcept {
 
-    auto& material = getPostProcessMaterial("colorGradingAsSubpass");
-    FMaterialInstance* const mi =
-            configureColorGradingMaterial(material, colorGrading, colorGradingConfig,
-                    vignetteOptions, width, height);
+    float4 const vignetteParameters = getVignetteParameters(vignetteOptions, width, height);
+
+    auto const& material = getPostProcessMaterial("colorGradingAsSubpass");
+    FMaterialInstance* const mi = PostProcessMaterial::getMaterialInstance(mEngine, material);
+
+    mi->setParameter("lut", colorGrading->getHwHandle(), {
+            .filterMag = SamplerMagFilter::LINEAR,
+            .filterMin = SamplerMinFilter::LINEAR,
+            .wrapS = SamplerWrapMode::CLAMP_TO_EDGE,
+            .wrapT = SamplerWrapMode::CLAMP_TO_EDGE,
+            .wrapR = SamplerWrapMode::CLAMP_TO_EDGE,
+            .anisotropyLog2 = 0
+    });
+    const float lutDimension = float(colorGrading->getDimension());
+    mi->setParameter("lutSize", float2{
+        0.5f / lutDimension, (lutDimension - 1.0f) / lutDimension,
+    });
+
+    const float temporalNoise = mUniformDistribution(mEngine.getRandomEngine());
+
+    mi->setParameter("vignette", vignetteParameters);
+    mi->setParameter("vignetteColor", vignetteOptions.color);
+    mi->setParameter("dithering", colorGradingConfig.dithering);
+    mi->setParameter("outputLuminance", colorGradingConfig.outputLuminance);
+    mi->setParameter("temporalNoise", temporalNoise);
     mi->commit(driver);
+
+    // load both variants
+    material.getMaterial(mEngine, PostProcessVariant::OPAQUE);
+    material.getMaterial(mEngine, PostProcessVariant::TRANSLUCENT);
 }
 
 void PostProcessManager::colorGradingSubpass(DriverApi& driver,
@@ -2462,14 +2487,22 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::colorGrading(FrameGraph& fg,
 
                 auto const& out = resources.getRenderPassInfo();
 
-                auto const& input = resources.getDescriptor(data.input);
-                auto const& output = resources.getDescriptor(data.output);
+                auto const& material = getPostProcessMaterial("colorGrading");
+
+                PostProcessVariant const variant = colorGradingConfig.translucent ?
+                        PostProcessVariant::TRANSLUCENT : PostProcessVariant::OPAQUE;
 
-                auto& material = getPostProcessMaterial("colorGrading");
                 FMaterialInstance* const mi =
-                        configureColorGradingMaterial(material, colorGrading, colorGradingConfig,
-                                vignetteOptions, output.width, output.height);
+                        PostProcessMaterial::getMaterialInstance(mEngine, material, variant);
 
+                mi->setParameter("lut", colorGrading->getHwHandle(), {
+                        .filterMag = SamplerMagFilter::LINEAR,
+                        .filterMin = SamplerMinFilter::LINEAR
+                });
+                const float lutDimension = float(colorGrading->getDimension());
+                mi->setParameter("lutSize", float2{
+                        0.5f / lutDimension, (lutDimension - 1.0f) / lutDimension,
+                });
                 mi->setParameter("colorBuffer", colorTexture, { /* shader uses texelFetch */ });
                 mi->setParameter("bloomBuffer", bloomTexture, {
                         .filterMag = SamplerMagFilter::LINEAR,
@@ -2501,7 +2534,19 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::colorGrading(FrameGraph& fg,
                     bloomParameters.y = 1.0f - bloomParameters.x;
                 }
 
+                auto const& input = resources.getDescriptor(data.input);
+                auto const& output = resources.getDescriptor(data.output);
+                float4 const vignetteParameters = getVignetteParameters(
+                        vignetteOptions, output.width, output.height);
+
+                const float temporalNoise = mUniformDistribution(mEngine.getRandomEngine());
+
+                mi->setParameter("dithering", colorGradingConfig.dithering);
                 mi->setParameter("bloom", bloomParameters);
+                mi->setParameter("vignette", vignetteParameters);
+                mi->setParameter("vignetteColor", vignetteOptions.color);
+                mi->setParameter("outputLuminance", colorGradingConfig.outputLuminance);
+                mi->setParameter("temporalNoise", temporalNoise);
                 mi->setParameter("viewport", float4{
                         float(vp.left)   / input.width,
                         float(vp.bottom) / input.height,
@@ -2509,9 +2554,7 @@ FrameGraphId<FrameGraphTexture> PostProcessManager::colorGrading(FrameGraph& fg,
                         float(vp.height) / input.height
                 });
 
-                commitAndRenderFullScreenQuad(driver, out, mi,
-                        colorGradingConfig.translucent
-                        ? PostProcessVariant::TRANSLUCENT : PostProcessVariant::OPAQUE);
+                commitAndRenderFullScreenQuad(driver, out, mi, variant);
             }
     );
 
@@ -2649,70 +2692,11 @@ void PostProcessManager::configureTemporalAntiAliasingMaterial(
     setConstantParameter(ma, "varianceGamma", taaOptions.varianceGamma);
     if (dirty) {
         ma->invalidate();
-        // TODO: call Material::compile(), we can't do that now because it works only
+        // TODO: call Material::compile(), we can't si that now because it works only
         //       with surface materials
     }
 }
 
-FMaterialInstance* PostProcessManager::configureColorGradingMaterial(
-        PostProcessMaterial& material, FColorGrading const* colorGrading,
-        ColorGradingConfig const& colorGradingConfig, VignetteOptions const& vignetteOptions,
-        uint32_t const width, uint32_t const height) noexcept {
-    FMaterial* const ma = material.getMaterial(mEngine);
-    bool dirty = false;
-
-    auto setConstantParameter =
-            [&dirty](FMaterial* const material, std::string_view const name, auto value) noexcept {
-        auto id = material->getSpecializationConstantId(name);
-        if (id.has_value()) {
-            if (material->setConstant(id.value(), value)) {
-                dirty = true;
-            }
-        }
-    };
-
-    setConstantParameter(ma, "isOneDimensional", colorGrading->isOneDimensional());
-    setConstantParameter(ma, "isLDR", colorGrading->isLDR());
-
-    if (dirty) {
-        ma->invalidate();
-        // TODO: call Material::compile(), we can't do that now because it works only
-        //       with surface materials
-    }
-
-    PostProcessVariant const variant = colorGradingConfig.translucent ?
-            PostProcessVariant::TRANSLUCENT : PostProcessVariant::OPAQUE;
-    FMaterialInstance* const mi =
-            PostProcessMaterial::getMaterialInstance(mEngine, material, variant);
-
-    const SamplerParams params = {
-            .filterMag = SamplerMagFilter::LINEAR,
-            .filterMin = SamplerMinFilter::LINEAR,
-            .wrapS = SamplerWrapMode::CLAMP_TO_EDGE,
-            .wrapT = SamplerWrapMode::CLAMP_TO_EDGE,
-            .wrapR = SamplerWrapMode::CLAMP_TO_EDGE,
-            .anisotropyLog2 = 0
-    };
-
-    mi->setParameter("lut", colorGrading->getHwHandle(), params);
-
-    const float lutDimension = float(colorGrading->getDimension());
-    mi->setParameter("lutSize", float2{
-        0.5f / lutDimension, (lutDimension - 1.0f) / lutDimension,
-    });
-
-    const float temporalNoise = mUniformDistribution(mEngine.getRandomEngine());
-
-    float4 const vignetteParameters = getVignetteParameters(vignetteOptions, width, height);
-    mi->setParameter("vignette", vignetteParameters);
-    mi->setParameter("vignetteColor", vignetteOptions.color);
-    mi->setParameter("dithering", colorGradingConfig.dithering);
-    mi->setParameter("outputLuminance", colorGradingConfig.outputLuminance);
-    mi->setParameter("temporalNoise", temporalNoise);
-
-    return mi;
-}
-
 FrameGraphId<FrameGraphTexture> PostProcessManager::taa(FrameGraph& fg,
         FrameGraphId<FrameGraphTexture> input,
         FrameGraphId<FrameGraphTexture> const depth,

filament/src/PostProcessManager.h

@@ -401,13 +401,6 @@ public:
             FMaterialInstance const* mi,
             PostProcessVariant variant = PostProcessVariant::OPAQUE) const noexcept;
 
-    // Sets the necessary spec constants and uniforms common to both colorGrading.mat and
-    // colorGradingAsSubpass.mat.
-    FMaterialInstance* configureColorGradingMaterial(
-            PostProcessMaterial& material, FColorGrading const* colorGrading,
-            ColorGradingConfig const& colorGradingConfig, VignetteOptions const& vignetteOptions,
-            uint32_t const width, uint32_t const height) noexcept;
-
 private:
     backend::RenderPrimitiveHandle mFullScreenQuadRph;
     backend::VertexBufferInfoHandle mFullScreenQuadVbih;

filament/src/details/ColorGrading.cpp

@@ -35,6 +35,7 @@
 
 #include <cmath>
 #include <cstdlib>
+#include <mutex>
 #include <tuple>
 
 namespace filament {
@@ -580,11 +581,8 @@ static float3 luminanceScaling(float3 x,
 // Quality
 //------------------------------------------------------------------------------
 
-static std::tuple<TextureFormat, PixelDataFormat, PixelDataType> selectLutTextureParams(
-        ColorGrading::LutFormat const lutFormat, const bool isOneDimensional) noexcept {
-    if (isOneDimensional) {
-        return { TextureFormat::R16F, PixelDataFormat::R, PixelDataType::HALF };
-    }
+static std::tuple<TextureFormat, PixelDataFormat, PixelDataType>
+        selectLutTextureParams(ColorGrading::LutFormat const lutFormat) noexcept {
     // We use RGBA16F for high quality modes instead of RGB16F because RGB16F
     // is not supported everywhere
     switch (lutFormat) {
@@ -660,44 +658,27 @@ FColorGrading::FColorGrading(FEngine& engine, const Builder& builder) {
 
     DriverApi& driver = engine.getDriverApi();
 
-    // XXX: The following two conditions also only hold true as long as the input and output color
-    // spaces are the same, but we currently don't check that. We must revise these conditions if we
-    // ever handle this case.
-    mIsOneDimensional = !builder->hasAdjustments && !builder->luminanceScaling
-            && builder->toneMapper->isOneDimensional()
-            && engine.features.engine.color_grading.use_1d_lut;
-    mIsLDR = mIsOneDimensional && builder->toneMapper->isLDR();
-
-    const Config config = {
-            mIsOneDimensional ? 512u : builder->dimension,
-            adaptationTransform(builder->whiteBalance),
-            selectColorGradingTransformIn(builder->toneMapping),
-            selectColorGradingTransformOut(builder->toneMapping),
-            selectColorGradingLuminance(builder->toneMapping),
-            selectOETF(builder->outputColorSpace),
-    };
-
-    mDimension = config.lutDimension;
-
-    uint32_t width;
-    uint32_t height;
-    uint32_t depth;
-    if (mIsOneDimensional) {
-        width = config.lutDimension;
-        height = 1;
-        depth = 1;
-    } else {
-        width = config.lutDimension;
-        height = config.lutDimension;
-        depth = config.lutDimension;
+    Config c;
+    // This lock protects the data inside Config, which is written to by the Filament thread,
+    // and read from multiple Job threads.
+    Mutex configLock;
+    {
+        std::lock_guard<Mutex> const lock(configLock);
+        c.lutDimension          = builder->dimension;
+        c.adaptationTransform   = adaptationTransform(builder->whiteBalance);
+        c.colorGradingIn        = selectColorGradingTransformIn(builder->toneMapping);
+        c.colorGradingOut       = selectColorGradingTransformOut(builder->toneMapping);
+        c.colorGradingLuminance = selectColorGradingLuminance(builder->toneMapping);
+        c.oetf                  = selectOETF(builder->outputColorSpace);
     }
 
-    size_t lutElementCount = width * height * depth;
-    size_t elementSize = mIsOneDimensional ? sizeof(half) : sizeof(half4);
+    mDimension = c.lutDimension;
+
+    size_t lutElementCount = c.lutDimension * c.lutDimension * c.lutDimension;
+    size_t elementSize = sizeof(half4);
     void* data = malloc(lutElementCount * elementSize);
 
-    auto [textureFormat, format, type] =
-            selectLutTextureParams(builder->format, mIsOneDimensional);
+    auto [textureFormat, format, type] = selectLutTextureParams(builder->format);
     assert_invariant(FTexture::isTextureFormatSupported(engine, textureFormat));
     assert_invariant(FTexture::validatePixelFormatAndType(textureFormat, format, type));
 
@@ -707,185 +688,170 @@ FColorGrading::FColorGrading(FEngine& engine, const Builder& builder) {
         converted = malloc(lutElementCount * sizeof(uint32_t));
     }
 
-    auto hdrColorAt = [builder, config](size_t r, size_t g, size_t b) {
-        float3 v = float3{r, g, b} * (1.0f / float(config.lutDimension - 1u));
-
-        // LogC encoding
-        v = LogC_to_linear(v);
-
-        // Kill negative values near 0.0f due to imprecision in the log conversion
-        v = max(v, 0.0f);
-
-        if (builder->hasAdjustments) {
-            // Exposure
-            v = adjustExposure(v, builder->exposure);
-
-            // Purkinje shift ("low-light" vision)
-            v = scotopicAdaptation(v, builder->nightAdaptation);
-        }
-
-        // Move to color grading color space
-        v = config.colorGradingIn * v;
-
-        if (builder->hasAdjustments) {
-            // White balance
-            v = chromaticAdaptation(v, config.adaptationTransform);
-
-            // Kill negative values before the next transforms
-            v = max(v, 0.0f);
-
-            // Channel mixer
-            v = channelMixer(v, builder->outRed, builder->outGreen, builder->outBlue);
-
-            // Shadows/mid-tones/highlights
-            v = tonalRanges(v, config.colorGradingLuminance,
-                    builder->shadows, builder->midtones, builder->highlights,
-                    builder->tonalRanges);
-
-            // The adjustments below behave better in log space
-            v = linear_to_LogC(v);
-
-            // ASC CDL
-            v = colorDecisionList(v, builder->slope, builder->offset, builder->power);
-
-            // Contrast in log space
-            v = contrast(v, builder->contrast);
-
-            // Back to linear space
-            v = LogC_to_linear(v);
-
-            // Vibrance in linear space
-            v = vibrance(v, config.colorGradingLuminance, builder->vibrance);
-
-            // Saturation in linear space
-            v = saturation(v, config.colorGradingLuminance, builder->saturation);
-
-            // Kill negative values before curves
-            v = max(v, 0.0f);
-
-            // RGB curves
-            v = curves(v,
-                    builder->shadowGamma, builder->midPoint, builder->highlightScale);
-        }
-
-        // Tone mapping
-        if (builder->luminanceScaling) {
-            v = luminanceScaling(v, *builder->toneMapper, config.colorGradingLuminance);
-        } else {
-            v = (*builder->toneMapper)(v);
-        }
-
-        // Go back to display color space
-        v = config.colorGradingOut * v;
-
-        // Apply gamut mapping
-        if (builder->gamutMapping) {
-            // TODO: This should depend on the output color space
-            v = gamutMapping_sRGB(v);
-        }
-
-        // TODO: We should convert to the output color space if we use a working
-        //       color space that's not sRGB
-        // TODO: Allow the user to customize the output color space
-
-        // We need to clamp for the output transfer function
-        v = saturate(v);
-
-        // Apply OETF
-        v = config.oetf(v);
-
-        return v;
-    };
-
     //auto now = std::chrono::steady_clock::now();
 
-    if (mIsOneDimensional) {
-        half* UTILS_RESTRICT p = (half*) data;
-        if (mIsLDR) {
-            for (size_t rgb = 0; rgb < config.lutDimension; rgb++) {
-                float3 v = float3(rgb) * (1.0f / float(config.lutDimension - 1u));
-
-                v = (*builder->toneMapper)(float3(v));
-
-                // We need to clamp for the output transfer function
-                v = saturate(v);
-
-                // Apply OETF
-                v = config.oetf(v);
-
-                *p++ = half(v.r);
+    // Multithreadedly generate the tone mapping 3D look-up table using 32 jobs
+    // Slices are 8 KiB (128 cache lines) apart.
+    // This takes about 3-6ms on Android in Release
+    JobSystem& js = engine.getJobSystem();
+    auto *slices = js.createJob();
+    for (size_t b = 0; b < c.lutDimension; b++) {
+        auto *job = js.createJob(slices,
+                [data, converted, b, &c, &configLock, builder](JobSystem&, JobSystem::Job*) {
+            Config config;
+            {
+                std::lock_guard<Mutex> lock(configLock);
+                config = c;
             }
-        } else {
-            for (size_t rgb = 0; rgb < config.lutDimension; rgb++) {
-                *p++ = half(hdrColorAt(rgb, rgb, rgb).r);
-            }
-        }
-    } else {
-        // Multithreadedly generate the tone mapping 3D look-up table using 32 jobs
-        // Slices are 8 KiB (128 cache lines) apart.
-        // This takes about 3-6ms on Android in Release
-        JobSystem& js = engine.getJobSystem();
-        auto *slices = js.createJob();
-        for (size_t b = 0; b < config.lutDimension; b++) {
-            auto* job = js.createJob(slices,
-                    [data, converted, b, &config, builder, &hdrColorAt](
-                            JobSystem&, JobSystem::Job*) {
-                half4* UTILS_RESTRICT p =
-                        (half4*) data + b * config.lutDimension * config.lutDimension;
-                for (size_t g = 0; g < config.lutDimension; g++) {
-                    for (size_t r = 0; r < config.lutDimension; r++) {
-                        *p++ = half4{hdrColorAt(r, g, b), 0.0f};
+            half4* UTILS_RESTRICT p = (half4*) data + b * config.lutDimension * config.lutDimension;
+            for (size_t g = 0; g < config.lutDimension; g++) {
+                for (size_t r = 0; r < config.lutDimension; r++) {
+                    float3 v = float3{r, g, b} * (1.0f / float(config.lutDimension - 1u));
+
+                    // LogC encoding
+                    v = LogC_to_linear(v);
+
+                    // Kill negative values near 0.0f due to imprecision in the log conversion
+                    v = max(v, 0.0f);
+
+                    if (builder->hasAdjustments) {
+                        // Exposure
+                        v = adjustExposure(v, builder->exposure);
+
+                        // Purkinje shift ("low-light" vision)
+                        v = scotopicAdaptation(v, builder->nightAdaptation);
                     }
-                }
 
-                if (converted) {
-                    uint32_t* const UTILS_RESTRICT dst = (uint32_t*) converted +
-                            b * config.lutDimension * config.lutDimension;
-                    half4* UTILS_RESTRICT src = (half4*) data +
-                            b * config.lutDimension * config.lutDimension;
-                    // we use a vectorize width of 8 because, on ARMv8 it allows the compiler to
-                    // write eight 32-bits results in one go.
-                    const size_t count = (config.lutDimension * config.lutDimension) & ~0x7u; // tell the compiler that we're a multiple of 8
+                    // Move to color grading color space
+                    v = c.colorGradingIn * v;
+
+                    if (builder->hasAdjustments) {
+                        // White balance
+                        v = chromaticAdaptation(v, config.adaptationTransform);
+
+                        // Kill negative values before the next transforms
+                        v = max(v, 0.0f);
+
+                        // Channel mixer
+                        v = channelMixer(v, builder->outRed, builder->outGreen, builder->outBlue);
+
+                        // Shadows/mid-tones/highlights
+                        v = tonalRanges(v, c.colorGradingLuminance,
+                                builder->shadows, builder->midtones, builder->highlights,
+                                builder->tonalRanges);
+
+                        // The adjustments below behave better in log space
+                        v = linear_to_LogC(v);
+
+                        // ASC CDL
+                        v = colorDecisionList(v, builder->slope, builder->offset, builder->power);
+
+                        // Contrast in log space
+                        v = contrast(v, builder->contrast);
+
+                        // Back to linear space
+                        v = LogC_to_linear(v);
+
+                        // Vibrance in linear space
+                        v = vibrance(v, c.colorGradingLuminance, builder->vibrance);
+
+                        // Saturation in linear space
+                        v = saturation(v, c.colorGradingLuminance, builder->saturation);
+
+                        // Kill negative values before curves
+                        v = max(v, 0.0f);
+
+                        // RGB curves
+                        v = curves(v,
+                                builder->shadowGamma, builder->midPoint, builder->highlightScale);
+                    }
+
+                    // Tone mapping
+                    if (builder->luminanceScaling) {
+                        v = luminanceScaling(v, *builder->toneMapper, c.colorGradingLuminance);
+                    } else {
+                        v = (*builder->toneMapper)(v);
+                    }
+
+                    // Go back to display color space
+                    v = c.colorGradingOut * v;
+
+                    // Apply gamut mapping
+                    if (builder->gamutMapping) {
+                        // TODO: This should depend on the output color space
+                        v = gamutMapping_sRGB(v);
+                    }
+
+                    // TODO: We should convert to the output color space if we use a working
+                    //       color space that's not sRGB
+                    // TODO: Allow the user to customize the output color space
+
+                    // We need to clamp for the output transfer function
+                    v = saturate(v);
+
+                    // Apply OETF
+                    v = c.oetf(v);
+
+                    *p++ = half4{v, 0.0f};
+                }
+            }
+
+            if (converted) {
+                uint32_t* const UTILS_RESTRICT dst = (uint32_t*) converted +
+                        b * config.lutDimension * config.lutDimension;
+                half4* UTILS_RESTRICT src = (half4*) data +
+                        b * config.lutDimension * config.lutDimension;
+                // we use a vectorize width of 8 because, on ARMv8 it allows the compiler to write eight
+                // 32-bits results in one go.
+                const size_t count = (config.lutDimension * config.lutDimension) & ~0x7u; // tell the compiler that we're a multiple of 8
 #if defined(__clang__)
-#pragma clang loop vectorize_width(8)
+                #pragma clang loop vectorize_width(8)
 #endif
-                    for (size_t i = 0; i < count; ++i) {
-                        float4 v{src[i]};
-                        uint32_t pr = uint32_t(std::floor(v.x * 1023.0f + 0.5f));
-                        uint32_t pg = uint32_t(std::floor(v.y * 1023.0f + 0.5f));
-                        uint32_t pb = uint32_t(std::floor(v.z * 1023.0f + 0.5f));
-                        dst[i] = (pb << 20u) | (pg << 10u) | pr;
-                    }
+                for (size_t i = 0; i < count; ++i) {
+                    float4 v{src[i]};
+                    uint32_t pr = uint32_t(std::floor(v.x * 1023.0f + 0.5f));
+                    uint32_t pg = uint32_t(std::floor(v.y * 1023.0f + 0.5f));
+                    uint32_t pb = uint32_t(std::floor(v.z * 1023.0f + 0.5f));
+                    dst[i] = (pb << 20u) | (pg << 10u) | pr;
                 }
-            });
-            js.run(job);
-        }
+            }
 
-        // TODO: Should we do a runAndRetain() and defer the wait() + texture creation until
-        //       getHwHandle() is invoked?
-        js.runAndWait(slices);
+        });
+        js.run(job);
     }
 
+    // TODO: Should we do a runAndRetain() and defer the wait() + texture creation until
+    //       getHwHandle() is invoked?
+    js.runAndWait(slices);
+
     //std::chrono::duration<float, std::milli> duration = std::chrono::steady_clock::now() - now;
     //slog.d << "LUT generation time: " << duration.count() << " ms" << io::endl;
 
+    mLutHandle = driver.createTexture(
+            SamplerType::SAMPLER_3D,
+            1,
+            textureFormat,
+            1,
+            c.lutDimension,
+            c.lutDimension,
+            c.lutDimension,
+            TextureUsage::DEFAULT
+    );
+
     if (converted) {
         free(data);
         data = converted;
         elementSize = sizeof(uint32_t);
     }
 
-    // Create texture.
-    mLutHandle = driver.createTexture(SamplerType::SAMPLER_3D, 1, textureFormat, 1,
-            width, height, depth, TextureUsage::DEFAULT);
-
     driver.update3DImage(mLutHandle, 0,
             0, 0, 0,
-            width, height, depth,
+            c.lutDimension, c.lutDimension, c.lutDimension,
             PixelBufferDescriptor{
-                data, lutElementCount * elementSize, format, type,
-                        [](void* buffer, size_t, void*) { free(buffer); }
-                        });
+                    data, lutElementCount * elementSize,format, type,
+                    [](void* buffer, size_t, void*) { free(buffer); }
+            }
+    );
 }
 
 FColorGrading::~FColorGrading() noexcept = default;

filament/src/details/ColorGrading.h

@@ -42,15 +42,12 @@ public:
     void terminate(FEngine& engine);
 
     backend::TextureHandle getHwHandle() const noexcept { return mLutHandle; }
+
     uint32_t getDimension() const noexcept { return mDimension; }
-    bool isOneDimensional() const noexcept { return mIsOneDimensional; }
-    bool isLDR() const noexcept { return mIsLDR; }
 
 private:
     backend::TextureHandle mLutHandle;
     uint32_t mDimension;
-    bool mIsOneDimensional;
-    bool mIsLDR;
 };
 
 FILAMENT_DOWNCAST(ColorGrading)

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);
 

filament/src/details/Engine.h

@@ -687,9 +687,6 @@ public:
 
     struct {
         struct {
-            struct {
-                bool use_1d_lut = false;
-            } color_grading;
             struct {
                 bool use_shadow_atlas = false;
             } shadows;
@@ -722,9 +719,6 @@ public:
             { "backend.opengl.assert_native_window_is_valid",
               "Asserts that the ANativeWindow is valid when rendering starts.",
               &features.backend.opengl.assert_native_window_is_valid, true },
-            { "engine.color_grading.use_1d_lut",
-              "Uses a 1D LUT for color grading.",
-              &features.engine.color_grading.use_1d_lut, false },
             { "engine.shadows.use_shadow_atlas",
               "Uses an array of atlases to store shadow maps.",
               &features.engine.shadows.use_shadow_atlas, false },

filament/src/details/Texture.cpp

@@ -153,10 +153,9 @@ Texture::Builder& Texture::Builder::name(const char* name, size_t const len) noe
 }
 
 Texture* Texture::Builder::build(Engine& engine) {
-    if (mImpl->mTarget != SamplerType::SAMPLER_EXTERNAL) {
-        FILAMENT_CHECK_PRECONDITION(Texture::isTextureFormatSupported(engine, mImpl->mFormat))
-                << "Texture format " << uint16_t(mImpl->mFormat) << " not supported on this platform";
-    }
+    FILAMENT_CHECK_PRECONDITION(Texture::isTextureFormatSupported(engine, mImpl->mFormat))
+            << "Texture format " << uint16_t(mImpl->mFormat) << " not supported on this platform";
+
     const bool isProtectedTexturesSupported =
             downcast(engine).getDriverApi().isProtectedTexturesSupported();
     const bool useProtectedMemory = bool(mImpl->mUsage & TextureUsage::PROTECTED);

filament/src/materials/colorGrading/colorGrading.fs

@@ -1,29 +1,13 @@
-vec3 LogC_to_linear(const vec3 x) {
+vec3 colorGrade(mediump sampler3D lut, const vec3 x) {
     // Alexa LogC EI 1000
     const float a = 5.555556;
     const float b = 0.047996;
     const float c = 0.244161 / log2(10.0);
     const float d = 0.386036;
-    return c * log2(a * x + b) + d;
-}
+    vec3 logc = c * log2(a * x + b) + d;
 
-vec3 colorGrade3D(mediump sampler3D lut, const vec3 v) {
-    return textureLod(lut, v, 0.0).rgb;
-}
+    // Remap to sample pixel centers
+    logc = materialParams.lutSize.x + logc * materialParams.lutSize.y;
 
-vec3 colorGrade1D(mediump sampler3D lut, const vec3 v) {
-    return vec3(
-        textureLod(lut, vec3(v.r, 0.5, 0.5), 0.0).r,
-        textureLod(lut, vec3(v.g, 0.5, 0.5), 0.0).r,
-        textureLod(lut, vec3(v.b, 0.5, 0.5), 0.0).r);
-}
-
-vec3 colorGrade(mediump sampler3D lut, vec3 v) {
-    if (!materialConstants_isLDR) {
-        v = LogC_to_linear(v);
-    }
-    // Remap to sample pixel centers.
-    v = materialParams.lutSize.x + v * materialParams.lutSize.y;
-    return materialConstants_isOneDimensional
-            ? colorGrade1D(lut, v) : colorGrade3D(lut, v);
+    return textureLod(lut, logc, 0.0).rgb;
 }

filament/src/materials/colorGrading/colorGrading.mat

@@ -66,16 +66,6 @@ material {
             precision : high
         }
     ],
-    constants : [
-        {
-            type : bool,
-            name : isOneDimensional
-        },
-        {
-            type : bool,
-            name : isLDR
-        }
-    ],
     variables : [
         vertex
     ],
@@ -164,7 +154,7 @@ void postProcess(inout PostProcessInputs postProcess) {
     }
 
     // Color grading
-    color.rgb = colorGrade(materialParams_lut, color.rgb);
+    color.rgb  = colorGrade(materialParams_lut, color.rgb);
 
     // output in premultiplied alpha
 #if !POST_PROCESS_OPAQUE

filament/src/materials/colorGrading/colorGradingAsSubpass.mat

@@ -32,16 +32,6 @@ material {
             name : vignetteColor
         }
     ],
-    constants : [
-        {
-            type : bool,
-            name : isOneDimensional
-        },
-        {
-            type : bool,
-            name : isLDR
-        }
-    ],
     subpasses : [
         {
             type : subpassInput,
@@ -113,7 +103,7 @@ void postProcess(inout PostProcessInputs postProcess) {
     }
 
     // Color grading
-    color.rgb = colorGrade(materialParams_lut, color.rgb);
+    color.rgb  = colorGrade(materialParams_lut, color.rgb);
 
     // output in premultiplied alpha
 #if !POST_PROCESS_OPAQUE

ios/CocoaPods/Filament.podspec

@@ -1,12 +1,12 @@
 Pod::Spec.new do |spec|
   spec.name = "Filament"
-  spec.version = "1.58.2"
+  spec.version = "1.58.0"
   spec.license = { :type => "Apache 2.0", :file => "LICENSE" }
   spec.homepage = "https://google.github.io/filament"
   spec.authors = "Google LLC."
   spec.summary = "Filament is a real-time physically based rendering engine for Android, iOS, Windows, Linux, macOS, and WASM/WebGL."
   spec.platform = :ios, "11.0"
-  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.58.2/filament-v1.58.2-ios.tgz" }
+  spec.source = { :http => "https://github.com/google/filament/releases/download/v1.58.0/filament-v1.58.0-ios.tgz" }
 
   # Fix linking error with Xcode 12; we do not yet support the simulator on Apple silicon.
   spec.pod_target_xcconfig = {

libs/fgviewer/README.md

@@ -12,7 +12,7 @@
 
 ## Capabilities
 
-fgviewer is a library and web application for real-time visualization of the frame graph in Filament.
+fgviewer is a library and web application for real-time visualization of the frame graph in Filament.  
 It displays active passes and resource usage, providing insights into the rendering pipeline.
 
 ## Setup for Desktop

libs/fgviewer/include/fgviewer/FrameGraphInfo.h

@@ -58,7 +58,7 @@ public:
             utils::CString name;
             utils::CString value;
         };
-
+        
         Resource(ResourceId id, utils::CString name,
                  std::vector<Property> properties);
 

libs/fgviewer/src/ApiHandler.cpp

@@ -17,22 +17,14 @@
 
 #include "ApiHandler.h"
 
-#include <__chrono/duration.h>
 #include <fgviewer/DebugServer.h>
 #include <fgviewer/JsonWriter.h>
 
 #include <utils/FixedCapacityVector.h>
 #include <utils/Log.h>
-#include <utils/ostream.h>
 
 #include <CivetServer.h>
 
-#include <cstdint>
-#include <cstdio>
-#include <cstring>
-#include <mutex>
-#include <string>
-
 namespace filament::fgviewer {
 
 using namespace std::chrono_literals;

libs/fgviewer/src/DebugServer.cpp

@@ -21,45 +21,36 @@
 
 #include <CivetServer.h>
 
+#include <utils/FixedCapacityVector.h>
+#include <utils/Hash.h>
 #include <utils/Log.h>
-#include <utils/Mutex.h>
-#include <utils/ostream.h>
 
-#include <mutex>
 #include <string>
 #include <string_view>
 
+namespace filament::fgviewer {
 
 // If set to 0, this serves HTML from a resgen resource. Use 1 only during local development, which
 // serves files directly from the source code tree.
 #define SERVE_FROM_SOURCE_TREE 0
 
 #if SERVE_FROM_SOURCE_TREE
-
 namespace {
 std::string const BASE_URL = "libs/fgviewer/web";
 } // anonymous
-
 #else
-
 #include "fgviewer_resources.h"
 #include <unordered_map>
 
 namespace {
-
 struct Asset {
     std::string_view mime;
     std::string_view data;
 };
-
 std::unordered_map<std::string_view, Asset> ASSET_MAP;
-
 } // anonymous
-
 #endif // SERVE_FROM_SOURCE_TREE
 
-namespace filament::fgviewer {
-
 using namespace utils;
 
 std::string_view const DebugServer::kSuccessHeader =

libs/fgviewer/src/FrameGraphInfo.cpp

@@ -16,11 +16,6 @@
 
 #include <fgviewer/FrameGraphInfo.h>
 
-#include <utils/CString.h>
-
-#include <unordered_map>
-#include <vector>
-
 namespace filament::fgviewer {
 
 FrameGraphInfo::FrameGraphInfo(utils::CString viewName):

libs/fgviewer/src/JsonWriter.cpp

@@ -14,15 +14,11 @@
  * limitations under the License.
  */
 
-#include <fgviewer/FrameGraphInfo.h>
 #include <fgviewer/JsonWriter.h>
+#include <fgviewer/FrameGraphInfo.h>
 
-#include <utils/CString.h>
-
-#include <cstddef>
-#include <ostream>
+#include <iomanip>
 #include <sstream>
-#include <vector>
 
 namespace filament::fgviewer {
 

libs/fgviewer/web/app.js

@@ -223,8 +223,8 @@ class FrameGraphSidePanel extends LitElement {
             return html`
                 <menu-section title="${title}">
                     ${this.framegraphs.map(({ fgid, name }) => html`
-                        <div class="framegraph"
-                            @click="${() => this._handleFrameGraphClick(fgid)}"
+                        <div class="framegraph" 
+                            @click="${() => this._handleFrameGraphClick(fgid)}" 
                             data-id="${fgid}">
                             ${fgid === this.selectedFrameGraph ? '● ' : ''}${name}
                         </div>
@@ -440,15 +440,15 @@ class FrameGraphTable extends LitElement {
         return html`
             <tr id="resource-${resourceIndex}">
                 <th class="sticky-col resource ${selectedStyle}" @click="${onClickResource}">
-                    ${hasSubresources
+                    ${hasSubresources 
                         ? html`
                             <span class="toggle-icon"
                                   @click="${(e) => { e.stopPropagation(); this._toggleCollapse(resourceIndex); }}">
                               ${isExpanded ? '▼' : '▶'}
-                            </span>`
+                            </span>` 
                         : nothing}
                     ${resource.name}
-                    ${hasSubresources && !isExpanded ? html`(${subresourceIds.length})` : nothing}
+                    ${hasSubresources && isExpanded ? nothing : html`(${subresourceIds.length})`}
                 </th>
                 ${this._renderResourceUsage(allPasses, resourceIds, DEFAULT_COLOR)}
             </tr>
@@ -592,12 +592,12 @@ class FrameGraphViewer extends LitElement {
         return html`
             <framegraph-sidepanel id="sidepanel"
                 ?connected="${this.connected}"
-                selected-framegraph="${this.selectedFrameGraph}"
+                selected-framegraph="${this.selectedFrameGraph}" 
                 selected-resource="${this.selectedResourceId}">
             </framegraph-sidepanel>
-            <framegraph-table id="table"
+            <framegraph-table id="table" 
                 ?connected="${this.connected}"
-                selected-framegraph="${this.selectedFrameGraph}"
+                selected-framegraph="${this.selectedFrameGraph}" 
                 selected-resource="${this.selectedResourceId}">
             </framegraph-table>
         `;

libs/filabridge/include/filament/MaterialEnums.h

@@ -28,7 +28,7 @@
 namespace filament {
 
 // update this when a new version of filament wouldn't work with older materials
-static constexpr size_t MATERIAL_VERSION = 58;
+static constexpr size_t MATERIAL_VERSION = 57;
 
 /**
  * Supported shading models

libs/filamat/src/GLSLPostProcessor.cpp

@@ -535,6 +535,7 @@ bool GLSLPostProcessor::spirvToWgsl(SpirvBlob *spirv, std::string *outWsl) {
         //TODO remove this if block once combined image sampler conversion works.
         if (tintRead.Diagnostics().Str().rfind("error: WGSL does not support combined image-samplers") != std::string::npos) {
             slog.w << "This tint reader error is currently ignored during WebGPU bringup: " << tintRead.Diagnostics().Str() << io::endl;
+            return false;
         }
         else {
             slog.e << "Tint Reader Error: " << tintRead.Diagnostics().Str() << io::endl;
@@ -919,6 +920,10 @@ std::shared_ptr<spvtools::Optimizer> GLSLPostProcessor::createOptimizer(
         optimizer->RegisterPass(CreateRedundancyEliminationPass());
         optimizer->RegisterPass(CreateAggressiveDCEPass());
     }
+    if (config.targetApi == MaterialBuilder::TargetApi::WEBGPU) {
+        optimizer->RegisterPass(CreateSplitCombinedImageSamplerPass());
+        optimizer->RegisterPass(CreateResolveBindingConflictsPass());
+    }
 
     return optimizer;
 }

libs/filamat/src/MaterialBuilder.cpp

@@ -1049,6 +1049,7 @@ bool MaterialBuilder::generateShaders(JobSystem& jobSystem, const std::vector<Va
 
 
                 switch (targetApi) {
+                    // TODO: Handle webgpu here
                     case TargetApi::WEBGPU:
                         assert(!spirv.empty());
                         assert(wgsl.length() > 0);

samples/CMakeLists.txt

@@ -245,6 +245,20 @@ function(add_demo NAME)
     endif()
 endfunction()
 
+function(add_demo_plain NAME)
+    include_directories(${GENERATION_ROOT})
+    add_executable(${NAME} ${NAME}.cpp)
+    target_compile_options(${NAME} PRIVATE ${COMPILER_FLAGS})
+    set_target_properties(${NAME} PROPERTIES FOLDER Samples)
+
+    # This is needed after XCode 15.3
+    if (APPLE)
+        set_target_properties(${NAME} PROPERTIES BUILD_WITH_INSTALL_RPATH TRUE)
+        set_target_properties(${NAME} PROPERTIES INSTALL_RPATH /usr/local/lib)
+    endif()
+endfunction()
+
+
 if (NOT ANDROID)
     add_demo(animation)
     add_demo(depthtesting)
@@ -280,6 +294,9 @@ if (NOT ANDROID)
     add_demo(vbotest)
     add_demo(viewtest)
 
+    add_demo_plain(spirv_test)
+    target_link_libraries(spirv_test PRIVATE SPIRV-Headers utils)
+
     # Sample app specific
     target_link_libraries(frame_generator PRIVATE imageio)
     target_link_libraries(gltf_viewer PRIVATE gltf-demo-resources uberarchive gltfio viewer)

samples/spirv_test.cpp

@@ -0,0 +1,145 @@
+#include "spirv/unified1/spirv.hpp"
+#include <utils/Log.h>
+
+#include <functional>
+#include <fstream>
+#include <vector>
+#include <unordered_map>
+
+size_t getFileSize(const std::string& filename) {
+    std::ifstream file(filename, std::ios::binary | std::ios::ate);
+    if (!file.is_open()) {
+        return -1;
+    }
+    size_t size = file.tellg();
+    file.close();
+    return size;
+}
+
+std::vector<uint32_t> transform(std::vector<uint32_t> const& spirv) {
+    constexpr size_t HEADER_SIZE = 5;
+    size_t const dataSize = spirv.size();
+    std::vector<uint32_t> output;
+    output.resize(dataSize);
+    uint32_t const* data = (uint32_t*) spirv.data();
+    uint32_t* outputData = output.data();
+    memcpy(&outputData[0], &data[0], HEADER_SIZE * 4);
+
+    struct Descriptor {
+        uint32_t binding = 0xFFFFFFFF;
+        uint32_t set = 0xFFFFFFFF;
+        bool fixed = false;
+    };
+
+    std::unordered_map<uint32_t, Descriptor> descriptors;
+    auto getDesc = [&](uint32_t var) -> Descriptor& {
+        using itertype = std::unordered_map<uint32_t, Descriptor>::iterator;
+        itertype itr;
+        if (itr = descriptors.find(var); itr == descriptors.end()) {
+            itr = descriptors.insert({var, {}}).first;
+        }
+        return itr->second;
+    };
+
+    auto pass = [&](uint32_t targetOp, std::function<void(uint32_t, uint32_t)> f) {
+        for (uint32_t cursor = HEADER_SIZE, cursorEnd = dataSize; cursor < cursorEnd;) {
+            uint32_t const firstWord = data[cursor];
+            uint32_t const wordCount = firstWord >> 16;
+            uint32_t const op = firstWord & 0x0000FFFF;
+            if (targetOp == op) {
+                f(wordCount-1, cursor+1);
+            }
+            cursor += wordCount;
+        }
+    };
+
+    uint32_t samplerTypeId = 0;
+    uint32_t imageTypeId = 0;
+
+    // First we find the ID used to identify the sampler type and the image type.
+    pass(spv::Op::OpTypeSampler, [&](uint32_t count, uint32_t pos) {
+        samplerTypeId = data[pos];
+    });
+
+    pass(spv::Op::OpTypeImage, [&](uint32_t count, uint32_t pos) {
+        imageTypeId = data[pos];
+    });
+
+    // Next we identify all the descriptors and record their set id and binding.
+    pass(spv::Op::OpDecorate, [&](uint32_t count, uint32_t pos) {
+        uint32_t const type = data[pos + 1];
+        if (type == spv::Decoration::DecorationBinding) {
+            uint32_t const targetVar = data[pos];
+            uint32_t const binding = data[pos + 2];
+            auto& desc = getDesc(targetVar);
+            desc.binding = binding;
+            // Note these decorations do not need to be written to the output.
+        } else if (type == spv::Decoration::DecorationDescriptorSet) {
+            uint32_t const targetVar = data[pos];
+            uint32_t const set = data[pos + 2];
+            auto& desc = getDesc(targetVar);
+            desc.set = set;
+        }
+    });
+
+    // If any descriptor is loaded to be a sampler, we offset it by a fixed amount. Note that it
+    // could be "loaded" multiple times, so we use a boolean to skip if it's already offsetted.
+    constexpr uint32_t SAMPLER_BINDING_OFFSET = 100;
+    pass(spv::Op::OpLoad, [&](uint32_t count, uint32_t pos) {
+        uint32_t const retType = data[pos];
+        uint32_t const targetVar = data[pos + 2];
+        if (retType == samplerTypeId) {
+            auto& desc = getDesc(targetVar);
+            if (!desc.fixed) {
+                desc.binding += SAMPLER_BINDING_OFFSET;
+                desc.fixed = true;
+            }
+        }
+    });
+
+    // Write out the offsetted bindings and copy the rest.
+    for (uint32_t cursor = HEADER_SIZE, cursorEnd = dataSize; cursor < cursorEnd;) {
+        uint32_t const firstWord = data[cursor];
+        uint32_t const wordCount = firstWord >> 16;
+        uint32_t const op = firstWord & 0x0000FFFF;
+
+        uint32_t* out = &outputData[cursor];
+        std::memcpy(out, &data[cursor], wordCount * 4);
+
+        if (op == spv::Op::OpDecorate && data[cursor + 2] == spv::Decoration::DecorationBinding) {
+            uint32_t const targetVar = data[cursor + 1];
+            auto const& modified = getDesc(targetVar);
+            out[3] = modified.binding;
+        }
+        cursor += wordCount;
+    }
+
+    // Note that you should be able to do this in-place (without making another vector).
+    return output;
+}
+
+int main(int argc, char* argv[]) {
+    if (argc < 3) {
+        utils::slog.e <<"need input and output spirv filenames" << utils::io::endl;
+        return 1;
+    }
+
+    std::string infname {argv[1]};
+    size_t const fsize = getFileSize(infname);
+    size_t const wordCount = fsize / 4;
+
+    std::ifstream inputFile(infname, std::ios::binary);
+
+    std::vector<uint32_t> words;
+    words.resize(wordCount);
+    inputFile.read((char*) words.data(), fsize);
+    inputFile.close();
+
+    std::string outfname {argv[2]};
+    std::vector<uint32_t> output = transform(words);
+    std::ofstream outputFile(outfname, std::ios::binary);
+    outputFile.write((char const*) output.data(), output.size() * 4);
+    outputFile.close();
+
+    return 0;
+}

third_party/dawn/tnt/CMakeLists.txt

@@ -14,7 +14,7 @@ set(DAWN_SPIRV_HEADERS_DIR ${EXTERNAL}/spirv-headers)
 set(DAWN_GLSLANG_DIR ${EXTERNAL}/glslang/)
 set(DAWN_DXC_ENABLE_ASSERTS_IN_NDEBUG OFF)
 
-set(TINT_BUILD_CMD_TOOLS OFF)
+#set(TINT_BUILD_CMD_TOOLS OFF)
 
 if (FILAMENT_BUILD_FILAMAT)
     set(TINT_BUILD_SPV_READER ON)

third_party/spirv-tools/Android.mk

@@ -172,11 +172,13 @@ SPVTOOLS_OPT_SRC_FILES := \
 		source/opt/remove_unused_interface_variables_pass.cpp \
 		source/opt/replace_desc_array_access_using_var_index.cpp \
 		source/opt/replace_invalid_opc.cpp \
+		source/opt/resolve_binding_conflicts_pass.cpp \
 		source/opt/scalar_analysis.cpp \
 		source/opt/scalar_analysis_simplification.cpp \
 		source/opt/scalar_replacement_pass.cpp \
 		source/opt/set_spec_constant_default_value_pass.cpp \
 		source/opt/simplification_pass.cpp \
+		source/opt/split_combined_image_sampler_pass.cpp \
 		source/opt/spread_volatile_semantics.cpp \
 		source/opt/ssa_rewrite_pass.cpp \
 		source/opt/strength_reduction_pass.cpp \

third_party/spirv-tools/BUILD.bazel

@@ -201,6 +201,7 @@ cc_library(
     deps = [
         ":spirv_tools_internal",
         "@spirv_headers//:spirv_common_headers",
+        "@spirv_headers//:spirv_c_headers",
     ],
 )
 

third_party/spirv-tools/BUILD.gn

@@ -771,6 +771,8 @@ static_library("spvtools_opt") {
     "source/opt/replace_desc_array_access_using_var_index.h",
     "source/opt/replace_invalid_opc.cpp",
     "source/opt/replace_invalid_opc.h",
+    "source/opt/resolve_binding_conflicts_pass.cpp",
+    "source/opt/resolve_binding_conflicts_pass.h",
     "source/opt/scalar_analysis.cpp",
     "source/opt/scalar_analysis.h",
     "source/opt/scalar_analysis_nodes.h",
@@ -781,6 +783,8 @@ static_library("spvtools_opt") {
     "source/opt/set_spec_constant_default_value_pass.h",
     "source/opt/simplification_pass.cpp",
     "source/opt/simplification_pass.h",
+    "source/opt/split_combined_image_sampler_pass.cpp",
+    "source/opt/split_combined_image_sampler_pass.h",
     "source/opt/spread_volatile_semantics.cpp",
     "source/opt/spread_volatile_semantics.h",
     "source/opt/ssa_rewrite_pass.cpp",
@@ -1382,6 +1386,7 @@ if (build_with_chromium && spvtools_build_executables) {
       "test/libspirv_macros_test.cpp",
       "test/name_mapper_test.cpp",
       "test/named_id_test.cpp",
+      "test/op_unknown_test.cpp",
       "test/opcode_make_test.cpp",
       "test/opcode_require_capabilities_test.cpp",
       "test/opcode_split_test.cpp",

third_party/spirv-tools/CHANGES

@@ -1,5 +1,40 @@
 Revision history for SPIRV-Tools
 
+v2025.1 2025-02-28
+  - General
+    - diff: Fix crash in OpString matching (#5988)
+    - Add SPV_AMDX_shader_enqueue version 2 support (#5838)
+    - add support for SPV_INTEL_subgroup_matrix_multiply_accumulate (#5928)
+    - update cmake_minimum_required to 3.22.1 (#5925)
+    - Add OpImageSampleFootprintNV to IsAllowedSampledImageOperand (#5914)
+    - assembler: ensure progress when seeking the version string (#5910)
+  - Optimizer
+    - opt: keep all OpSource instructions (#5901)
+    - [opt] Fix bug opt::InstructionBuilder::AddVariable (#6007)
+    - [OPT] Add SPV_KHR_ray_tracing to allow list (#5941)
+    - opt: keep all OpSource instructions (#5901)
+  - Validator
+    - spirv-val: Add AllowVulkan32BitBitwise option (#6001)
+    - Fix untyped pointer comparison validation (#6004)
+    - spirv-val: Update VUIDs for 308 header (#5990)
+    - spirv-val: fix env parsing for vk1.1spv1.4 (#5985)
+    - Add validation for SPV_NV_linear_swept_spheres. (#5975)
+    - Add validation SPV_NV_cluster_acceleration_structure. (#5974)
+    - Improve the instruction diagnostic for some access chain errors (#5978)
+    - Update physical storage buffer validation to match SPIR-V 1.6.5 (#5977)
+    - Validate SPV_NV_cooperative_vector (#5972)
+    - Fix layout checks with untyped pointers (#5970)
+    - spirv-val: Update mislabeled VUIDs (#5964)
+    - More explicit layout validation (#5958)
+    - spirv-val: Add VK_KHR_maintenance8 support (#5951)
+    - Add EXT_mesh_shader validation support (#5640)
+    - spirv-val: Remove OpenCL ivec3 req (#5940)
+    - spirv-val: Validate zero product workgroup size (#5407)
+    - Relax DebugLine validation (#5916)
+  - Linker
+    - linker: remove LinkOnceODR decorations when linking executables (#5979)
+    - fix: handle LinkOnceODR correctly (#5938)
+
 v2024.4 2024-12-04
   - General
     - Add FPEncoding operand type. (#5726)

third_party/spirv-tools/DEPS

@@ -3,18 +3,18 @@ use_relative_paths = True
 vars = {
   'github': 'https://github.com',
 
-  'abseil_revision': '69b58f9227bf91a9315475579df15fbb7c5c612e',
+  'abseil_revision': '310e6f4f0f202da13720fdd6cb0095e10a98fe1c',
 
-  'effcee_revision': '12241cbc30f20730b656db7fd5a3fa36cd420843',
+  'effcee_revision': '874b47102c57a8979c0f154cf8e0eab53c0a0502',
 
-  'googletest_revision': 'e235eb34c6c4fed790ccdad4b16394301360dcd4',
+  'googletest_revision': '4902ea2d7c6faed89b6facee00baa34bb108fc0d',
 
   # Use protobufs before they gained the dependency on abseil
   'protobuf_revision': 'v21.12',
 
-  're2_revision': '6dcd83d60f7944926bfd308cc13979fc53dd69ca',
+  're2_revision': 'c84a140c93352cdabbfb547c531be34515b12228',
 
-  'spirv_headers_revision': 'e7294a8ebed84f8c5bd3686c68dbe12a4e65b644',
+  'spirv_headers_revision': 'ddd2c099be25e3fec7cdc14106c17aca99512082',
 }
 
 deps = {

third_party/spirv-tools/docs/syntax.md

@@ -215,6 +215,50 @@ Note that this has some interesting consequences, including:
   that named ID being output.  This may be valid SPIR-V, contrary to the
   presumed intention of the writer.
 
+## OpUnknown
+<a name="op-unknown"></a>
+
+HLSL has a feature that allows users to specify an exact SPIR-V type using [the
+`SpirvType` template](https://github.com/microsoft/hlsl-specs/blob/main/proposals/0011-inline-spirv.md#types).
+This feature allows the user to specify a type opcode that the compiler does
+not know. In this case, it will be unable to generate assembly using the
+correct mnemonic.
+
+In order to represent unknown opcodes in assembly format, the `OpUnknown`
+pseudo-instruction may be used. The syntax is:
+
+```
+OpUnknown(<enumerant>, <WordCount>) <operand 1> ...
+```
+
+`enumerant` is the opcode enumerant, and `WordCount` is the number of words in
+the instruction. These will be assembled into a single word representing the
+opcode. Operands will be parsed according to the alternate parsing mode
+described in [Arbitrary Integers](#op-unknown). Named enumerated values cannot
+be handled by this mode and must be represented using the arbitrary integer
+syntax.
+
+It must be used at the beginning of a new instruction, and if there is a result
+ID it must explicitly be passed in as an operand. This is because the physical
+layout of a SPIR-V instruction may include a result type operand before the
+result ID operand, but it depends on the opcode and cannot be inferred for an
+unknown operand.
+
+For example, a 32-bit signed integer type could be represented like this:
+
+```
+OpUnknown(21, 4) %int_t 32 1
+```
+
+An OpStore instruction could be represented as:
+
+```
+OpUnknown(62, 3) %9 %12
+```
+
+The enumerant and word count must be decimal integers.
+
+
 ## Notes
 
 * Some enumerants cannot be used by name, because the target instruction

third_party/spirv-tools/include/spirv-tools/libspirv.h

@@ -740,6 +740,10 @@ SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetAllowLocalSizeId(
 SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetAllowOffsetTextureOperand(
     spv_validator_options options, bool val);
 
+// Allow base operands of some bit operations to be non-32-bit wide.
+SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetAllowVulkan32BitBitwise(
+    spv_validator_options options, bool val);
+
 // Whether friendly names should be used in validation error messages.
 SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetFriendlyNames(
     spv_validator_options options, bool val);

third_party/spirv-tools/include/spirv-tools/libspirv.hpp

@@ -132,6 +132,11 @@ class SPIRV_TOOLS_EXPORT ValidatorOptions {
     spvValidatorOptionsSetAllowOffsetTextureOperand(options_, val);
   }
 
+  // Allow base operands of some bit operations to be non-32-bit wide.
+  void SetAllowVulkan32BitBitwise(bool val) {
+    spvValidatorOptionsSetAllowVulkan32BitBitwise(options_, val);
+  }
+
   // Records whether or not the validator should relax the rules on pointer
   // usage in logical addressing mode.
   //

third_party/spirv-tools/include/spirv-tools/optimizer.hpp

@@ -240,7 +240,7 @@ class SPIRV_TOOLS_EXPORT Optimizer {
 
  private:
   struct SPIRV_TOOLS_LOCAL Impl;  // Opaque struct for holding internal data.
-  std::unique_ptr<Impl> impl_;  // Unique pointer to internal data.
+  std::unique_ptr<Impl> impl_;    // Unique pointer to internal data.
 };
 
 // Creates a null pass.
@@ -655,7 +655,7 @@ Optimizer::PassToken CreateRedundancyEliminationPass();
 // element if those elements are accessed individually.  The parameter is a
 // limit on the number of members in the composite variable that the pass will
 // consider replacing.
-Optimizer::PassToken CreateScalarReplacementPass(uint32_t size_limit = 100);
+Optimizer::PassToken CreateScalarReplacementPass(uint32_t size_limit = 0);
 
 // Create a private to local pass.
 // This pass looks for variables declared in the private storage class that are
@@ -968,6 +968,60 @@ Optimizer::PassToken CreateInvocationInterlockPlacementPass();
 // Creates a pass to add/remove maximal reconvergence execution mode.
 // This pass either adds or removes maximal reconvergence from all entry points.
 Optimizer::PassToken CreateModifyMaximalReconvergencePass(bool add);
+
+// Creates a pass to split combined image+sampler variables and function
+// parameters into separate image and sampler parts. Binding numbers and
+// other decorations are copied.
+Optimizer::PassToken CreateSplitCombinedImageSamplerPass();
+
+// Creates a pass to remap bindings to avoid conflicts, assuming the module
+// is valid for Vulkan.  A conflict exits when an entry point uses two distinct
+// variables with the same descriptor set and binding.  Vulkan allows one kind
+// of conflict: when one varible is an image (or array of images), and the
+// other is a sampler (or an array of samplers).
+
+// Conflicts are eliminated by incrementing the binding number of the sampler
+// part, and then propagating that increment through variables with
+// higher-numbered bindings until no conflict remains. This handles the case
+// when multiple shaders may share the same resource variables; this can
+// introduce holes in binding slots.
+//
+// Here's an example where shaders Alpha, Beta, Gamma, Delta collectively use
+// resource variables %100, %101, %102, %103, %104 all with the same
+// DescriptorSet and with Bindings as in the following table:
+//
+// Before:
+//
+//   Binding:   0          1        2        3
+//   Alpha:    %100,%101
+//   Beta:     %100       %102
+//   Gamma:               %102     %103
+//   Delta:                        %103     %104
+//
+// The Alpha shader has a conflict where variables %100, %101 have the same
+// descriptor set and binding.  If %100 is a sampler resource variable, then
+// the conflict is resolved by incrementing the binding number on %100 from 0
+// to 1.  But this causes a new confict for shader Beta because it now uses
+// both %100 and %102 with binding number 1.  That conflict is resolved by
+// incrementing the binding number on its variable that originally appeared
+// second (i.e. %102), so %102 gets binding 2. This now produces a conflict
+// for Gamma between %102 and %103 using binding number 2. Since %103 originally
+// appeared second (in the view from Gamma), the algorithm bumps %103 to binding
+// number %103. Now Delta has a conflict between %103 and %104, resulting in
+// %104 getting the next binding number, 4.  The picture afterward is:
+//
+// After:
+//
+//   Binding:   0          1        2        3        4
+//   Alpha:    %101       %100
+//   Beta:                %100     %102
+//   Gamma:                        %102     %103
+//   Delta:                                 %103     %104
+//
+//
+// This pass assumes binding numbers are not applid via decoration groups
+// (OpDecorationGroup).
+Optimizer::PassToken CreateResolveBindingConflictsPass();
 }  // namespace spvtools
 
 #endif  // INCLUDE_SPIRV_TOOLS_OPTIMIZER_HPP_

third_party/spirv-tools/kokoro/check-format/build-docker.sh

@@ -0,0 +1,26 @@
+#!/bin/bash
+# Copyright (c) 2025 Google LLC
+#
+# 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.
+
+# Fail on any error.
+set -e
+
+# This is required to run any git command in the docker since owner will
+# have changed between the clone environment, and the docker container.
+# Marking the root of the repo as safe for ownership changes.
+git config --global --add safe.directory "$PWD"
+
+echo $(date): Check formatting...
+./utils/check_code_format.sh ${1:-main}
+echo $(date): check completed.

third_party/spirv-tools/kokoro/check-format/build.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-# Copyright (c) 2018 Google LLC.
+# Copyright (c) 2025 Google LLC
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -12,37 +12,16 @@
 # 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.
-#
-# Android Build Script.
 
 # Fail on any error.
 set -e
-# Display commands being run.
-set -x
 
-BUILD_ROOT=$PWD
-SRC=$PWD/github/SPIRV-Tools
+SCRIPT_DIR="$( cd "$(dirname "${BASH_SOURCE[0]}")" >/dev/null 2>&1 && pwd )"
+SRC_ROOT="$( cd "${SCRIPT_DIR}/../.." >/dev/null 2>&1 && pwd )"
+TARGET_BRANCH="${KOKORO_GITHUB_PULL_REQUEST_TARGET_BRANCH-main}"
 
-# This is required to run any git command in the docker since owner will
-# have changed between the clone environment, and the docker container.
-# Marking the root of the repo as safe for ownership changes.
-git config --global --add safe.directory $SRC
-
-# Get clang-format-5.0.0.
-# Once kokoro upgrades the Ubuntu VMs, we can use 'apt-get install clang-format'
-curl -L http://releases.llvm.org/5.0.0/clang+llvm-5.0.0-linux-x86_64-ubuntu14.04.tar.xz -o clang-llvm.tar.xz
-tar xf clang-llvm.tar.xz
-export PATH=$PWD/clang+llvm-5.0.0-linux-x86_64-ubuntu14.04/bin:$PATH
-
-cd $SRC
-git clone --depth=1 https://github.com/KhronosGroup/SPIRV-Headers external/spirv-headers
-git clone https://github.com/google/googletest          external/googletest
-cd external && cd googletest && git reset --hard 1fb1bb23bb8418dc73a5a9a82bbed31dc610fec7 && cd .. && cd ..
-git clone --depth=1 https://github.com/google/effcee              external/effcee
-git clone --depth=1 https://github.com/google/re2                 external/re2
-# The --fail flag causes the command to fail on HTTP error response codes, like 404.
-curl -L --fail https://raw.githubusercontent.com/llvm/llvm-project/main/clang/tools/clang-format/clang-format-diff.py -o utils/clang-format-diff.py
-
-echo $(date): Check formatting...
-./utils/check_code_format.sh;
-echo $(date): check completed.
+docker run --rm -i \
+  --volume "${SRC_ROOT}:${SRC_ROOT}" \
+  --workdir "${SRC_ROOT}" \
+  "us-east4-docker.pkg.dev/shaderc-build/radial-docker/ubuntu-24.04-amd64/formatter" \
+  "${SCRIPT_DIR}/build-docker.sh" "${TARGET_BRANCH}"

third_party/spirv-tools/kokoro/scripts/linux/build.sh

@@ -52,7 +52,7 @@ docker run --rm -i \
   --env KOKORO_ARTIFACTS_DIR="${KOKORO_ARTIFACTS_DIR}" \
   --env BUILD_SHA="${BUILD_SHA}" \
   --entrypoint "${SCRIPT_DIR}/build-docker.sh" \
-  "gcr.io/shaderc-build/radial-build:latest"
+  us-east4-docker.pkg.dev/shaderc-build/radial-docker/ubuntu-24.04-amd64/cpp-builder
 RESULT=$?
 
 # This is important. If the permissions are not fixed, kokoro will fail

third_party/spirv-tools/source/diff/diff.cpp

@@ -67,7 +67,9 @@ void CompactIds(std::vector<uint32_t>& ids) {
   ids.resize(write_index);
 }
 
-// A mapping between src and dst ids.
+// A mapping from ids in one module to ids in the other.
+//
+// Differ contains two of these, for src->dst and dst->src.
 class IdMap {
  public:
   IdMap(size_t id_bound) { id_map_.resize(id_bound, 0); }
@@ -190,6 +192,7 @@ class SrcDstIdMap {
   IdMap dst_to_src_;
 };
 
+// Mappings from ids to instructions and metadata, for a single module's ids.
 struct IdInstructions {
   IdInstructions(const opt::Module* module)
       : inst_map_(module->IdBound(), nullptr),
@@ -198,6 +201,10 @@ struct IdInstructions {
         forward_pointer_map_(module->IdBound()) {
     // Map ids from all sections to instructions that define them.
     MapIdsToInstruction(module->ext_inst_imports());
+    MapIdsToInstruction(module->debugs1());
+    MapIdsToInstruction(module->debugs2());
+    MapIdsToInstruction(module->debugs3());
+    MapIdsToInstruction(module->ext_inst_debuginfo());
     MapIdsToInstruction(module->types_values());
     for (const opt::Function& function : *module) {
       function.ForEachInst(
@@ -321,6 +328,8 @@ class Differ {
   // Get various properties from an id.  These Helper functions are passed to
   // `GroupIds` and `GroupIdsAndMatch` below (as the `get_group` argument).
   uint32_t GroupIdsHelperGetTypeId(const IdInstructions& id_to, uint32_t id);
+  uint32_t GroupIdsHelperGetFunctionTypeId(const IdInstructions& id_to,
+                                           uint32_t id);
   spv::StorageClass GroupIdsHelperGetTypePointerStorageClass(
       const IdInstructions& id_to, uint32_t id);
   spv::Op GroupIdsHelperGetTypePointerTypeOp(const IdInstructions& id_to,
@@ -883,6 +892,17 @@ uint32_t Differ::GroupIdsHelperGetTypeId(const IdInstructions& id_to,
   return GetInst(id_to, id)->type_id();
 }
 
+// Return an `OpFunction` instruction's full `OpTypeFunction` type,
+// which includes parameter types.
+//
+// `GroupIdsHelperGetTypeId` applied to an `OpFunction` only gets the
+// function's return type, so this is a slightly more precise way to
+// match up functions by signature.
+uint32_t Differ::GroupIdsHelperGetFunctionTypeId(const IdInstructions& id_to,
+                                                 uint32_t id) {
+  return GetInst(id_to, id)->GetSingleWordOperand(3);
+}
+
 spv::StorageClass Differ::GroupIdsHelperGetTypePointerStorageClass(
     const IdInstructions& id_to, uint32_t id) {
   const opt::Instruction* inst = GetInst(id_to, id);
@@ -902,6 +922,24 @@ spv::Op Differ::GroupIdsHelperGetTypePointerTypeOp(const IdInstructions& id_to,
   return type_inst->opcode();
 }
 
+// Group unmatched ids in `ids` according to some characteristic,
+// determined by `get_group`.
+//
+// Using `get_group` to compute some sort of key for each id, set
+// `groups` to map each key to all the ids that have that key.
+//
+// For example, to group ids by name, pass `Differ::GetName` as
+// `get_group`. This will fill `groups` with a map from each name to
+// all the ids with that name.
+//
+// Under the assumption that we're trying to establish new pairings,
+// ids that are already paired are omitted from `groups`.
+//
+// The `is_src` parameter indicates whether `ids` are drawn from the
+// source module or the destination module.
+//
+// The template parameter `T` is the key type, like `std::string` or
+// `uint32_t`.
 template <typename T>
 void Differ::GroupIds(const IdGroup& ids, bool is_src,
                       std::map<T, IdGroup>* groups,
@@ -924,6 +962,10 @@ void Differ::GroupIds(const IdGroup& ids, bool is_src,
   }
 }
 
+// Group `src_ids` and `dst_ids` according to `get_group`, and then use
+// `match_group` to pair up ids in corresponding groups.
+//
+// Don't try to pair ids in groups whose key is `invalid_group_key`.
 template <typename T>
 void Differ::GroupIdsAndMatch(
     const IdGroup& src_ids, const IdGroup& dst_ids, T invalid_group_key,
@@ -2483,7 +2525,7 @@ void Differ::MatchFunctions() {
 
         // If there are multiple functions with the same name, group them by
         // type, and match only if the types match (and are unique).
-        GroupIdsAndMatch<uint32_t>(src_group, dst_group, 0,
+        GroupIdsAndMatchByMappedId(src_group, dst_group,
                                    &Differ::GroupIdsHelperGetTypeId,
                                    [this](const IdGroup& src_group_by_type_id,
                                           const IdGroup& dst_group_by_type_id) {
@@ -2526,9 +2568,19 @@ void Differ::MatchFunctions() {
                              dst_match_result, 0);
   }
 
-  // Best effort match functions with matching type.
-  GroupIdsAndMatch<uint32_t>(
-      src_func_ids, dst_func_ids, 0, &Differ::GroupIdsHelperGetTypeId,
+  // Best effort match functions with matching return and argument types.
+  GroupIdsAndMatchByMappedId(
+      src_func_ids, dst_func_ids, &Differ::GroupIdsHelperGetFunctionTypeId,
+      [this](const IdGroup& src_group_by_func_type_id,
+             const IdGroup& dst_group_by_func_type_id) {
+        BestEffortMatchFunctions(src_group_by_func_type_id,
+                                 dst_group_by_func_type_id, src_func_insts_,
+                                 dst_func_insts_);
+      });
+
+  // Best effort match functions with matching return types.
+  GroupIdsAndMatchByMappedId(
+      src_func_ids, dst_func_ids, &Differ::GroupIdsHelperGetTypeId,
       [this](const IdGroup& src_group_by_type_id,
              const IdGroup& dst_group_by_type_id) {
         BestEffortMatchFunctions(src_group_by_type_id, dst_group_by_type_id,

third_party/spirv-tools/source/link/linker.cpp

@@ -676,8 +676,10 @@ spv_result_t RemoveLinkageSpecificInstructions(
       if (inst->opcode() == spv::Op::OpDecorate &&
           spv::Decoration(inst->GetSingleWordOperand(1u)) ==
               spv::Decoration::LinkageAttributes &&
-          spv::LinkageType(inst->GetSingleWordOperand(3u)) ==
-              spv::LinkageType::Export) {
+          (spv::LinkageType(inst->GetSingleWordOperand(3u)) ==
+               spv::LinkageType::Export ||
+           spv::LinkageType(inst->GetSingleWordOperand(3u)) ==
+               spv::LinkageType::LinkOnceODR)) {
         linked_context->KillInst(&*inst);
       }
     }

third_party/spirv-tools/source/opt/CMakeLists.txt

@@ -109,11 +109,13 @@ set(SPIRV_TOOLS_OPT_SOURCES
   remove_unused_interface_variables_pass.h
   replace_desc_array_access_using_var_index.h
   replace_invalid_opc.h
+  resolve_binding_conflicts_pass.h
   scalar_analysis.h
   scalar_analysis_nodes.h
   scalar_replacement_pass.h
   set_spec_constant_default_value_pass.h
   simplification_pass.h
+  split_combined_image_sampler_pass.h
   spread_volatile_semantics.h
   ssa_rewrite_pass.h
   strength_reduction_pass.h
@@ -225,11 +227,13 @@ set(SPIRV_TOOLS_OPT_SOURCES
   remove_unused_interface_variables_pass.cpp
   replace_desc_array_access_using_var_index.cpp
   replace_invalid_opc.cpp
+  resolve_binding_conflicts_pass.cpp
   scalar_analysis.cpp
   scalar_analysis_simplification.cpp
   scalar_replacement_pass.cpp
   set_spec_constant_default_value_pass.cpp
   simplification_pass.cpp
+  split_combined_image_sampler_pass.cpp
   spread_volatile_semantics.cpp
   ssa_rewrite_pass.cpp
   strength_reduction_pass.cpp

third_party/spirv-tools/source/opt/copy_prop_arrays.cpp

@@ -276,6 +276,7 @@ CopyPropagateArrays::GetSourceObjectIfAny(uint32_t result) {
     case spv::Op::OpCompositeConstruct:
       return BuildMemoryObjectFromCompositeConstruct(result_inst);
     case spv::Op::OpCopyObject:
+    case spv::Op::OpCopyLogical:
       return GetSourceObjectIfAny(result_inst->GetSingleWordInOperand(0));
     case spv::Op::OpCompositeInsert:
       return BuildMemoryObjectFromInsert(result_inst);

third_party/spirv-tools/source/opt/folding_rules.cpp

@@ -2454,6 +2454,31 @@ FoldingRule RedundantFDiv() {
     FloatConstantKind kind0 = getFloatConstantKind(constants[0]);
     FloatConstantKind kind1 = getFloatConstantKind(constants[1]);
 
+    if (kind0 == FloatConstantKind::Zero || kind1 == FloatConstantKind::One) {
+      inst->SetOpcode(spv::Op::OpCopyObject);
+      inst->SetInOperands(
+          {{SPV_OPERAND_TYPE_ID, {inst->GetSingleWordInOperand(0)}}});
+      return true;
+    }
+
+    return false;
+  };
+}
+
+FoldingRule RedundantFMod() {
+  return [](IRContext* context, Instruction* inst,
+            const std::vector<const analysis::Constant*>& constants) {
+    assert(inst->opcode() == spv::Op::OpFMod &&
+           "Wrong opcode.  Should be OpFMod.");
+    assert(constants.size() == 2);
+
+    if (!inst->IsFloatingPointFoldingAllowed()) {
+      return false;
+    }
+
+    FloatConstantKind kind0 = getFloatConstantKind(constants[0]);
+    FloatConstantKind kind1 = getFloatConstantKind(constants[1]);
+
     if (kind0 == FloatConstantKind::Zero) {
       inst->SetOpcode(spv::Op::OpCopyObject);
       inst->SetInOperands(
@@ -2462,9 +2487,15 @@ FoldingRule RedundantFDiv() {
     }
 
     if (kind1 == FloatConstantKind::One) {
+      auto type = context->get_type_mgr()->GetType(inst->type_id());
+      std::vector<uint32_t> zero_words;
+      zero_words.resize(ElementWidth(type) / 32);
+      auto const_mgr = context->get_constant_mgr();
+      auto zero = const_mgr->GetConstant(type, std::move(zero_words));
+      auto zero_id = const_mgr->GetDefiningInstruction(zero)->result_id();
+
       inst->SetOpcode(spv::Op::OpCopyObject);
-      inst->SetInOperands(
-          {{SPV_OPERAND_TYPE_ID, {inst->GetSingleWordInOperand(0)}}});
+      inst->SetInOperands({{SPV_OPERAND_TYPE_ID, {zero_id}}});
       return true;
     }
 
@@ -2507,24 +2538,18 @@ FoldingRule RedundantFMix() {
   };
 }
 
-// This rule handles addition of zero for integers.
-FoldingRule RedundantIAdd() {
-  return [](IRContext* context, Instruction* inst,
-            const std::vector<const analysis::Constant*>& constants) {
-    assert(inst->opcode() == spv::Op::OpIAdd &&
-           "Wrong opcode. Should be OpIAdd.");
+// Returns a folding rule that folds the instruction to operand |foldToArg|
+// (0 or 1) if operand |arg| (0 or 1) is a zero constant.
+FoldingRule RedundantBinaryOpWithZeroOperand(uint32_t arg, uint32_t foldToArg) {
+  return [arg, foldToArg](
+             IRContext* context, Instruction* inst,
+             const std::vector<const analysis::Constant*>& constants) {
+    assert(constants.size() == 2);
 
-    uint32_t operand = std::numeric_limits<uint32_t>::max();
-    const analysis::Type* operand_type = nullptr;
-    if (constants[0] && constants[0]->IsZero()) {
-      operand = inst->GetSingleWordInOperand(1);
-      operand_type = constants[0]->type();
-    } else if (constants[1] && constants[1]->IsZero()) {
-      operand = inst->GetSingleWordInOperand(0);
-      operand_type = constants[1]->type();
-    }
+    if (constants[arg] && constants[arg]->IsZero()) {
+      auto operand = inst->GetSingleWordInOperand(foldToArg);
+      auto operand_type = constants[arg]->type();
 
-    if (operand != std::numeric_limits<uint32_t>::max()) {
       const analysis::Type* inst_type =
           context->get_type_mgr()->GetType(inst->type_id());
       if (inst_type->IsSame(operand_type)) {
@@ -2539,6 +2564,117 @@ FoldingRule RedundantIAdd() {
   };
 }
 
+// This rule handles any of RedundantBinaryRhs0Ops with a 0 or vector 0 on the
+// right-hand side (a | 0 => a).
+static const constexpr spv::Op RedundantBinaryRhs0Ops[] = {
+    spv::Op::OpBitwiseOr,
+    spv::Op::OpBitwiseXor,
+    spv::Op::OpShiftRightLogical,
+    spv::Op::OpShiftRightArithmetic,
+    spv::Op::OpShiftLeftLogical,
+    spv::Op::OpIAdd,
+    spv::Op::OpISub};
+FoldingRule RedundantBinaryRhs0(spv::Op op) {
+  assert(std::find(std::begin(RedundantBinaryRhs0Ops),
+                   std::end(RedundantBinaryRhs0Ops),
+                   op) != std::end(RedundantBinaryRhs0Ops) &&
+         "Wrong opcode.");
+  (void)op;
+  return RedundantBinaryOpWithZeroOperand(1, 0);
+}
+
+// This rule handles any of RedundantBinaryLhs0Ops with a 0 or vector 0 on the
+// left-hand side (0 | a => a).
+static const constexpr spv::Op RedundantBinaryLhs0Ops[] = {
+    spv::Op::OpBitwiseOr, spv::Op::OpBitwiseXor, spv::Op::OpIAdd};
+FoldingRule RedundantBinaryLhs0(spv::Op op) {
+  assert(std::find(std::begin(RedundantBinaryLhs0Ops),
+                   std::end(RedundantBinaryLhs0Ops),
+                   op) != std::end(RedundantBinaryLhs0Ops) &&
+         "Wrong opcode.");
+  (void)op;
+  return RedundantBinaryOpWithZeroOperand(0, 1);
+}
+
+// This rule handles shifts and divisions of 0 or vector 0 by any amount
+// (0 >> a => 0).
+static const constexpr spv::Op RedundantBinaryLhs0To0Ops[] = {
+    spv::Op::OpShiftRightLogical,
+    spv::Op::OpShiftRightArithmetic,
+    spv::Op::OpShiftLeftLogical,
+    spv::Op::OpSDiv,
+    spv::Op::OpUDiv,
+    spv::Op::OpSMod,
+    spv::Op::OpUMod};
+FoldingRule RedundantBinaryLhs0To0(spv::Op op) {
+  assert(std::find(std::begin(RedundantBinaryLhs0To0Ops),
+                   std::end(RedundantBinaryLhs0To0Ops),
+                   op) != std::end(RedundantBinaryLhs0To0Ops) &&
+         "Wrong opcode.");
+  (void)op;
+  return RedundantBinaryOpWithZeroOperand(0, 0);
+}
+
+// Returns true if all elements in |c| are 1.
+bool IsAllInt1(const analysis::Constant* c) {
+  if (auto composite = c->AsCompositeConstant()) {
+    auto& components = composite->GetComponents();
+    return std::all_of(std::begin(components), std::end(components), IsAllInt1);
+  } else if (c->AsIntConstant()) {
+    return c->GetSignExtendedValue() == 1;
+  }
+
+  return false;
+}
+
+// This rule handles divisions by 1 or vector 1 (a / 1 => a).
+FoldingRule RedundantSUDiv() {
+  return [](IRContext* context, Instruction* inst,
+            const std::vector<const analysis::Constant*>& constants) {
+    assert(constants.size() == 2);
+    assert((inst->opcode() == spv::Op::OpUDiv ||
+            inst->opcode() == spv::Op::OpSDiv) &&
+           "Wrong opcode.");
+
+    if (constants[1] && IsAllInt1(constants[1])) {
+      auto operand = inst->GetSingleWordInOperand(0);
+      auto operand_type = constants[1]->type();
+
+      const analysis::Type* inst_type =
+          context->get_type_mgr()->GetType(inst->type_id());
+      if (inst_type->IsSame(operand_type)) {
+        inst->SetOpcode(spv::Op::OpCopyObject);
+      } else {
+        inst->SetOpcode(spv::Op::OpBitcast);
+      }
+      inst->SetInOperands({{SPV_OPERAND_TYPE_ID, {operand}}});
+      return true;
+    }
+    return false;
+  };
+}
+
+// This rule handles modulo from division by 1 or vector 1 (a % 1 => 0).
+FoldingRule RedundantSUMod() {
+  return [](IRContext* context, Instruction* inst,
+            const std::vector<const analysis::Constant*>& constants) {
+    assert(constants.size() == 2);
+    assert((inst->opcode() == spv::Op::OpUMod ||
+            inst->opcode() == spv::Op::OpSMod) &&
+           "Wrong opcode.");
+
+    if (constants[1] && IsAllInt1(constants[1])) {
+      auto type = context->get_type_mgr()->GetType(inst->type_id());
+      auto zero_id = context->get_constant_mgr()->GetNullConstId(type);
+
+      inst->SetOpcode(spv::Op::OpCopyObject);
+      inst->SetInOperands({{SPV_OPERAND_TYPE_ID, {zero_id}}});
+      return true;
+    }
+    return false;
+  };
+}
+
 // This rule look for a dot with a constant vector containing a single 1 and
 // the rest 0s.  This is the same as doing an extract.
 FoldingRule DotProductDoingExtract() {
@@ -2876,6 +3012,17 @@ void FoldingRules::AddFoldingRules() {
   // Note that the order in which rules are added to the list matters. If a rule
   // applies to the instruction, the rest of the rules will not be attempted.
   // Take that into consideration.
+  for (auto op : RedundantBinaryRhs0Ops)
+    rules_[op].push_back(RedundantBinaryRhs0(op));
+  for (auto op : RedundantBinaryLhs0Ops)
+    rules_[op].push_back(RedundantBinaryLhs0(op));
+  for (auto op : RedundantBinaryLhs0To0Ops)
+    rules_[op].push_back(RedundantBinaryLhs0To0(op));
+  rules_[spv::Op::OpSDiv].push_back(RedundantSUDiv());
+  rules_[spv::Op::OpUDiv].push_back(RedundantSUDiv());
+  rules_[spv::Op::OpSMod].push_back(RedundantSUMod());
+  rules_[spv::Op::OpUMod].push_back(RedundantSUMod());
+
   rules_[spv::Op::OpBitcast].push_back(BitCastScalarOrVector());
 
   rules_[spv::Op::OpCompositeConstruct].push_back(
@@ -2907,6 +3054,8 @@ void FoldingRules::AddFoldingRules() {
   rules_[spv::Op::OpFDiv].push_back(MergeDivMulArithmetic());
   rules_[spv::Op::OpFDiv].push_back(MergeDivNegateArithmetic());
 
+  rules_[spv::Op::OpFMod].push_back(RedundantFMod());
+
   rules_[spv::Op::OpFMul].push_back(RedundantFMul());
   rules_[spv::Op::OpFMul].push_back(MergeMulMulArithmetic());
   rules_[spv::Op::OpFMul].push_back(MergeMulDivArithmetic());
@@ -2921,7 +3070,6 @@ void FoldingRules::AddFoldingRules() {
   rules_[spv::Op::OpFSub].push_back(MergeSubAddArithmetic());
   rules_[spv::Op::OpFSub].push_back(MergeSubSubArithmetic());
 
-  rules_[spv::Op::OpIAdd].push_back(RedundantIAdd());
   rules_[spv::Op::OpIAdd].push_back(MergeAddNegateArithmetic());
   rules_[spv::Op::OpIAdd].push_back(MergeAddAddArithmetic());
   rules_[spv::Op::OpIAdd].push_back(MergeAddSubArithmetic());

third_party/spirv-tools/source/opt/function.h

@@ -17,6 +17,7 @@
 
 #include <algorithm>
 #include <functional>
+#include <iterator>
 #include <memory>
 #include <string>
 #include <unordered_set>
@@ -39,6 +40,7 @@ class Function {
  public:
   using iterator = UptrVectorIterator<BasicBlock>;
   using const_iterator = UptrVectorIterator<BasicBlock, true>;
+  using ParamList = std::vector<std::unique_ptr<Instruction>>;
 
   // Creates a function instance declared by the given OpFunction instruction
   // |def_inst|.
@@ -77,6 +79,23 @@ class Function {
   // Does nothing if the function doesn't have such a parameter.
   inline void RemoveParameter(uint32_t id);
 
+  // Rewrites the function parameters by calling a replacer callback.
+  // The replacer takes two parameters: an expiring unique pointer to a current
+  // instruction, and a back-inserter into a new list of unique pointers to
+  // instructions.  The replacer is called for each current parameter, in order.
+  // Not valid to call while also iterating through the parameter list, e.g.
+  // within the ForEachParam method.
+  using RewriteParamFn = std::function<void(
+      std::unique_ptr<Instruction>&&, std::back_insert_iterator<ParamList>&)>;
+  void RewriteParams(RewriteParamFn& replacer) {
+    ParamList new_params;
+    auto appender = std::back_inserter(new_params);
+    for (auto& param : params_) {
+      replacer(std::move(param), appender);
+    }
+    params_ = std::move(new_params);
+  }
+
   // Saves the given function end instruction.
   inline void SetFunctionEnd(std::unique_ptr<Instruction> end_inst);
 
@@ -197,7 +216,7 @@ class Function {
   // The OpFunction instruction that begins the definition of this function.
   std::unique_ptr<Instruction> def_inst_;
   // All parameters to this function.
-  std::vector<std::unique_ptr<Instruction>> params_;
+  ParamList params_;
   // All debug instructions in this function's header.
   InstructionList debug_insts_in_header_;
   // All basic blocks inside this function in specification order

third_party/spirv-tools/source/opt/inline_exhaustive_pass.cpp

@@ -55,6 +55,11 @@ Pass::Status InlineExhaustivePass::InlineExhaustive(Function* func) {
       }
     }
   }
+
+  if (modified) {
+    FixDebugDeclares(func);
+  }
+
   return (modified ? Status::SuccessWithChange : Status::SuccessWithoutChange);
 }
 

third_party/spirv-tools/source/opt/inline_pass.cpp

@@ -30,6 +30,8 @@ namespace {
 constexpr int kSpvFunctionCallFunctionId = 2;
 constexpr int kSpvFunctionCallArgumentId = 3;
 constexpr int kSpvReturnValueId = 0;
+constexpr int kSpvDebugDeclareVarInIdx = 3;
+constexpr int kSpvAccessChainBaseInIdx = 0;
 }  // namespace
 
 uint32_t InlinePass::AddPointerToType(uint32_t type_id,
@@ -858,5 +860,68 @@ void InlinePass::InitializeInline() {
 
 InlinePass::InlinePass() {}
 
+void InlinePass::FixDebugDeclares(Function* func) {
+  std::map<uint32_t, Instruction*> access_chains;
+  std::vector<Instruction*> debug_declare_insts;
+
+  func->ForEachInst([&access_chains, &debug_declare_insts](Instruction* inst) {
+    if (inst->opcode() == spv::Op::OpAccessChain) {
+      access_chains[inst->result_id()] = inst;
+    }
+    if (inst->GetCommonDebugOpcode() == CommonDebugInfoDebugDeclare) {
+      debug_declare_insts.push_back(inst);
+    }
+  });
+
+  for (auto& inst : debug_declare_insts) {
+    FixDebugDeclare(inst, access_chains);
+  }
+}
+
+void InlinePass::FixDebugDeclare(
+    Instruction* dbg_declare_inst,
+    const std::map<uint32_t, Instruction*>& access_chains) {
+  do {
+    uint32_t var_id =
+        dbg_declare_inst->GetSingleWordInOperand(kSpvDebugDeclareVarInIdx);
+
+    // The def-use chains are not kept up to date while inlining, so we need to
+    // get the variable by traversing the functions.
+    auto it = access_chains.find(var_id);
+    if (it == access_chains.end()) {
+      return;
+    }
+    Instruction* access_chain = it->second;
+
+    // If the variable id in the debug declare is an access chain, it is
+    // invalid. it needs to be fixed up. The debug declare will be updated so
+    // that its Var operand becomes the base of the access chain. The indexes of
+    // the access chain are prepended before the indexes of the debug declare.
+
+    std::vector<Operand> operands;
+    for (int i = 0; i < kSpvDebugDeclareVarInIdx; i++) {
+      operands.push_back(dbg_declare_inst->GetInOperand(i));
+    }
+
+    uint32_t access_chain_base =
+        access_chain->GetSingleWordInOperand(kSpvAccessChainBaseInIdx);
+    operands.push_back(Operand(SPV_OPERAND_TYPE_ID, {access_chain_base}));
+    operands.push_back(
+        dbg_declare_inst->GetInOperand(kSpvDebugDeclareVarInIdx + 1));
+
+    for (uint32_t i = kSpvAccessChainBaseInIdx + 1;
+         i < access_chain->NumInOperands(); ++i) {
+      operands.push_back(access_chain->GetInOperand(i));
+    }
+
+    for (uint32_t i = kSpvDebugDeclareVarInIdx + 2;
+         i < dbg_declare_inst->NumInOperands(); ++i) {
+      operands.push_back(dbg_declare_inst->GetInOperand(i));
+    }
+
+    dbg_declare_inst->SetInOperands(std::move(operands));
+  } while (true);
+}
+
 }  // namespace opt
 }  // namespace spvtools

third_party/spirv-tools/source/opt/inline_pass.h

@@ -150,6 +150,12 @@ class InlinePass : public Pass {
   // Initialize state for optimization of |module|
   void InitializeInline();
 
+  // Fixes invalid debug declare functions in `func` that were caused by
+  // inlining. This function cannot be called while in the middle of inlining
+  // because it needs to be able to find the instructions that define an
+  // id.
+  void FixDebugDeclares(Function* func);
+
   // Map from function's result id to function.
   std::unordered_map<uint32_t, Function*> id2function_;
 
@@ -241,6 +247,11 @@ class InlinePass : public Pass {
   // structural dominance.
   void UpdateSingleBlockLoopContinueTarget(
       uint32_t new_id, std::vector<std::unique_ptr<BasicBlock>>* new_blocks);
+
+  // Replaces the `var` operand of `dbg_declare_inst` and updates the indexes
+  // accordingly, if it is the id of an access chain in `access_chains`.
+  void FixDebugDeclare(Instruction* dbg_declare_inst,
+                       const std::map<uint32_t, Instruction*>& access_chains);
 };
 
 }  // namespace opt

third_party/spirv-tools/source/opt/ir_builder.h

@@ -15,6 +15,7 @@
 #ifndef SOURCE_OPT_IR_BUILDER_H_
 #define SOURCE_OPT_IR_BUILDER_H_
 
+#include <cassert>
 #include <limits>
 #include <memory>
 #include <utility>
@@ -480,8 +481,11 @@ class InstructionBuilder {
     return AddInstruction(std::move(select));
   }
 
-  Instruction* AddAccessChain(uint32_t type_id, uint32_t base_ptr_id,
-                              std::vector<uint32_t> ids) {
+  Instruction* AddOpcodeAccessChain(spv::Op opcode, uint32_t type_id,
+                                    uint32_t base_ptr_id,
+                                    const std::vector<uint32_t>& ids) {
+    assert(opcode == spv::Op::OpAccessChain ||
+           opcode == spv::Op::OpInBoundsAccessChain);
     std::vector<Operand> operands;
     operands.push_back({SPV_OPERAND_TYPE_ID, {base_ptr_id}});
 
@@ -490,12 +494,22 @@ class InstructionBuilder {
     }
 
     // TODO(1841): Handle id overflow.
-    std::unique_ptr<Instruction> new_inst(
-        new Instruction(GetContext(), spv::Op::OpAccessChain, type_id,
-                        GetContext()->TakeNextId(), operands));
+    std::unique_ptr<Instruction> new_inst(new Instruction(
+        GetContext(), opcode, type_id, GetContext()->TakeNextId(), operands));
     return AddInstruction(std::move(new_inst));
   }
 
+  Instruction* AddAccessChain(uint32_t type_id, uint32_t base_ptr_id,
+                              const std::vector<uint32_t>& ids) {
+    return AddOpcodeAccessChain(spv::Op::OpAccessChain, type_id, base_ptr_id,
+                                ids);
+  }
+  Instruction* AddInBoundsAccessChain(uint32_t type_id, uint32_t base_ptr_id,
+                                      const std::vector<uint32_t>& ids) {
+    return AddOpcodeAccessChain(spv::Op::OpInBoundsAccessChain, type_id,
+                                base_ptr_id, ids);
+  }
+
   Instruction* AddLoad(uint32_t type_id, uint32_t base_ptr_id,
                        uint32_t alignment = 0) {
     std::vector<Operand> operands;
@@ -514,9 +528,19 @@ class InstructionBuilder {
     return AddInstruction(std::move(new_inst));
   }
 
+  Instruction* AddCopyObject(uint32_t type_id, uint32_t value_id) {
+    std::vector<Operand> operands{{SPV_OPERAND_TYPE_ID, {value_id}}};
+
+    // TODO(1841): Handle id overflow.
+    std::unique_ptr<Instruction> new_inst(
+        new Instruction(GetContext(), spv::Op::OpCopyObject, type_id,
+                        GetContext()->TakeNextId(), operands));
+    return AddInstruction(std::move(new_inst));
+  }
+
   Instruction* AddVariable(uint32_t type_id, uint32_t storage_class) {
     std::vector<Operand> operands;
-    operands.push_back({SPV_OPERAND_TYPE_ID, {storage_class}});
+    operands.push_back({SPV_OPERAND_TYPE_STORAGE_CLASS, {storage_class}});
     std::unique_ptr<Instruction> new_inst(
         new Instruction(GetContext(), spv::Op::OpVariable, type_id,
                         GetContext()->TakeNextId(), operands));
@@ -572,6 +596,26 @@ class InstructionBuilder {
     return AddInstruction(std::move(new_inst));
   }
 
+  Instruction* AddDecoration(uint32_t target_id, spv::Decoration d,
+                             const std::vector<uint32_t>& literals) {
+    std::vector<Operand> operands;
+    operands.push_back({SPV_OPERAND_TYPE_ID, {target_id}});
+    operands.push_back({SPV_OPERAND_TYPE_DECORATION, {uint32_t(d)}});
+    for (uint32_t literal : literals) {
+      operands.push_back({SPV_OPERAND_TYPE_LITERAL_INTEGER, {literal}});
+    }
+
+    std::unique_ptr<Instruction> new_inst(
+        new Instruction(GetContext(), spv::Op::OpDecorate, 0, 0, operands));
+    // Decorations are annotation instructions.  Add it via the IR context,
+    // so the decoration manager will be updated.
+    // Decorations don't belong to basic blocks, so there is no need
+    // to update the instruction to block mapping.
+    Instruction* result = new_inst.get();
+    GetContext()->AddAnnotationInst(std::move(new_inst));
+    return result;
+  }
+
   Instruction* AddNaryExtendedInstruction(
       uint32_t result_type, uint32_t set, uint32_t instruction,
       const std::vector<uint32_t>& ext_operands) {
@@ -593,6 +637,23 @@ class InstructionBuilder {
     return AddInstruction(std::move(new_inst));
   }
 
+  Instruction* AddSampledImage(uint32_t sampled_image_type_id,
+                               uint32_t image_id, uint32_t sampler_id) {
+    std::vector<Operand> operands;
+    operands.push_back({SPV_OPERAND_TYPE_ID, {image_id}});
+    operands.push_back({SPV_OPERAND_TYPE_ID, {sampler_id}});
+
+    uint32_t result_id = GetContext()->TakeNextId();
+    if (result_id == 0) {
+      return nullptr;
+    }
+
+    std::unique_ptr<Instruction> new_inst(
+        new Instruction(GetContext(), spv::Op::OpSampledImage,
+                        sampled_image_type_id, result_id, operands));
+    return AddInstruction(std::move(new_inst));
+  }
+
   // Inserts the new instruction before the insertion point.
   Instruction* AddInstruction(std::unique_ptr<Instruction>&& insn) {
     Instruction* insn_ptr = &*insert_before_.InsertBefore(std::move(insn));

third_party/spirv-tools/source/opt/ir_context.cpp

@@ -564,6 +564,7 @@ void IRContext::AddCombinatorsForCapability(uint32_t capability) {
          (uint32_t)spv::Op::OpCompositeConstruct,
          (uint32_t)spv::Op::OpCompositeExtract,
          (uint32_t)spv::Op::OpCompositeInsert,
+         (uint32_t)spv::Op::OpCopyLogical,
          (uint32_t)spv::Op::OpCopyObject,
          (uint32_t)spv::Op::OpTranspose,
          (uint32_t)spv::Op::OpSampledImage,

third_party/spirv-tools/source/opt/optimizer.cpp

@@ -189,7 +189,7 @@ Optimizer& Optimizer::RegisterPerformancePasses(bool preserve_interface) {
       .RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
       .RegisterPass(CreateLocalSingleStoreElimPass())
       .RegisterPass(CreateAggressiveDCEPass(preserve_interface))
-      .RegisterPass(CreateScalarReplacementPass())
+      .RegisterPass(CreateScalarReplacementPass(0))
       .RegisterPass(CreateLocalAccessChainConvertPass())
       .RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
       .RegisterPass(CreateLocalSingleStoreElimPass())
@@ -203,7 +203,7 @@ Optimizer& Optimizer::RegisterPerformancePasses(bool preserve_interface) {
       .RegisterPass(CreateRedundancyEliminationPass())
       .RegisterPass(CreateCombineAccessChainsPass())
       .RegisterPass(CreateSimplificationPass())
-      .RegisterPass(CreateScalarReplacementPass())
+      .RegisterPass(CreateScalarReplacementPass(0))
       .RegisterPass(CreateLocalAccessChainConvertPass())
       .RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
       .RegisterPass(CreateLocalSingleStoreElimPass())
@@ -401,7 +401,7 @@ bool Optimizer::RegisterPassFromFlag(const std::string& flag,
     RegisterPass(CreateLoopUnswitchPass());
   } else if (pass_name == "scalar-replacement") {
     if (pass_args.size() == 0) {
-      RegisterPass(CreateScalarReplacementPass());
+      RegisterPass(CreateScalarReplacementPass(0));
     } else {
       int limit = -1;
       if (pass_args.find_first_not_of("0123456789") == std::string::npos) {
@@ -637,6 +637,10 @@ bool Optimizer::RegisterPassFromFlag(const std::string& flag,
     }
   } else if (pass_name == "trim-capabilities") {
     RegisterPass(CreateTrimCapabilitiesPass());
+  } else if (pass_name == "split-combined-image-sampler") {
+    RegisterPass(CreateSplitCombinedImageSamplerPass());
+  } else if (pass_name == "resolve-binding-conflicts") {
+    RegisterPass(CreateResolveBindingConflictsPass());
   } else {
     Errorf(consumer(), nullptr, {},
            "Unknown flag '--%s'. Use --help for a list of valid flags",
@@ -1188,6 +1192,16 @@ Optimizer::PassToken CreateOpExtInstWithForwardReferenceFixupPass() {
       MakeUnique<opt::OpExtInstWithForwardReferenceFixupPass>());
 }
 
+Optimizer::PassToken CreateSplitCombinedImageSamplerPass() {
+  return MakeUnique<Optimizer::PassToken::Impl>(
+      MakeUnique<opt::SplitCombinedImageSamplerPass>());
+}
+
+Optimizer::PassToken CreateResolveBindingConflictsPass() {
+  return MakeUnique<Optimizer::PassToken::Impl>(
+      MakeUnique<opt::ResolveBindingConflictsPass>());
+}
+
 }  // namespace spvtools
 
 extern "C" {

third_party/spirv-tools/source/opt/passes.h

@@ -74,9 +74,11 @@
 #include "source/opt/remove_unused_interface_variables_pass.h"
 #include "source/opt/replace_desc_array_access_using_var_index.h"
 #include "source/opt/replace_invalid_opc.h"
+#include "source/opt/resolve_binding_conflicts_pass.h"
 #include "source/opt/scalar_replacement_pass.h"
 #include "source/opt/set_spec_constant_default_value_pass.h"
 #include "source/opt/simplification_pass.h"
+#include "source/opt/split_combined_image_sampler_pass.h"
 #include "source/opt/spread_volatile_semantics.h"
 #include "source/opt/ssa_rewrite_pass.h"
 #include "source/opt/strength_reduction_pass.h"

third_party/spirv-tools/source/opt/resolve_binding_conflicts_pass.cpp

@@ -0,0 +1,328 @@
+// Copyright (c) 2025 Google LLC
+//
+// 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 "source/opt/resolve_binding_conflicts_pass.h"
+
+#include <algorithm>
+#include <unordered_map>
+#include <unordered_set>
+#include <vector>
+
+#include "source/opt/decoration_manager.h"
+#include "source/opt/def_use_manager.h"
+#include "source/opt/instruction.h"
+#include "source/opt/ir_builder.h"
+#include "source/opt/ir_context.h"
+#include "spirv/unified1/spirv.h"
+
+namespace spvtools {
+namespace opt {
+
+// A VarBindingInfo contains the binding information for a single resource
+// variable.
+//
+// Exactly one such object is created per resource variable in the
+// module. In particular, when a resource variable is statically used by
+// more than one entry point, those entry points share the same VarBindingInfo
+// object for that variable.
+struct VarBindingInfo {
+  const Instruction* const var;
+  const uint32_t descriptor_set;
+  Instruction* const binding_decoration;
+
+  // Returns the binding number.
+  uint32_t binding() const {
+    return binding_decoration->GetSingleWordInOperand(2);
+  }
+  // Sets the binding number to 'b'.
+  void updateBinding(uint32_t b) { binding_decoration->SetOperand(2, {b}); }
+};
+
+// The bindings in the same descriptor set that are used by an entry point.
+using BindingList = std::vector<VarBindingInfo*>;
+// A map from descriptor set number to the list of bindings in that descriptor
+// set, as used by a particular entry point.
+using DescriptorSets = std::unordered_map<uint32_t, BindingList>;
+
+IRContext::Analysis ResolveBindingConflictsPass::GetPreservedAnalyses() {
+  // All analyses are kept up to date.
+  // At most this modifies the Binding numbers on variables.
+  return IRContext::kAnalysisDefUse | IRContext::kAnalysisInstrToBlockMapping |
+         IRContext::kAnalysisDecorations | IRContext::kAnalysisCombinators |
+         IRContext::kAnalysisCFG | IRContext::kAnalysisDominatorAnalysis |
+         IRContext::kAnalysisLoopAnalysis | IRContext::kAnalysisNameMap |
+         IRContext::kAnalysisScalarEvolution |
+         IRContext::kAnalysisRegisterPressure |
+         IRContext::kAnalysisValueNumberTable |
+         IRContext::kAnalysisStructuredCFG | IRContext::kAnalysisBuiltinVarId |
+         IRContext::kAnalysisIdToFuncMapping | IRContext::kAnalysisConstants |
+         IRContext::kAnalysisTypes | IRContext::kAnalysisDebugInfo |
+         IRContext::kAnalysisLiveness;
+}
+
+// Orders variable binding info objects.
+// * The binding number is most signficant;
+// * Then a sampler-like object compares greater than non-sampler like object.
+// * Otherwise compare based on variable ID.
+// This provides a total order among bindings in a descriptor set for a valid
+// Vulkan module.
+bool Less(const VarBindingInfo* const lhs, const VarBindingInfo* const rhs) {
+  if (lhs->binding() < rhs->binding()) return true;
+  if (lhs->binding() > rhs->binding()) return false;
+
+  // Examine types.
+  // In valid Vulkan the only conflict can occur between
+  // images and samplers.  We only care about a specific
+  // comparison when one is a image-like thing and the other
+  // is a sampler-like thing of the same shape.  So unwrap
+  // types until we hit one of those two.
+
+  auto* def_use_mgr = lhs->var->context()->get_def_use_mgr();
+
+  // Returns the type found by iteratively following pointer pointee type,
+  // or array element type.
+  auto unwrap = [&def_use_mgr](Instruction* ty) {
+    bool keep_going = true;
+    do {
+      switch (ty->opcode()) {
+        case spv::Op::OpTypePointer:
+          ty = def_use_mgr->GetDef(ty->GetSingleWordInOperand(1));
+          break;
+        case spv::Op::OpTypeArray:
+        case spv::Op::OpTypeRuntimeArray:
+          ty = def_use_mgr->GetDef(ty->GetSingleWordInOperand(0));
+          break;
+        default:
+          keep_going = false;
+          break;
+      }
+    } while (keep_going);
+    return ty;
+  };
+
+  auto* lhs_ty = unwrap(def_use_mgr->GetDef(lhs->var->type_id()));
+  auto* rhs_ty = unwrap(def_use_mgr->GetDef(rhs->var->type_id()));
+  if (lhs_ty->opcode() == rhs_ty->opcode()) {
+    // Pick based on variable ID.
+    return lhs->var->result_id() < rhs->var->result_id();
+  }
+  // A sampler is always greater than an image.
+  if (lhs_ty->opcode() == spv::Op::OpTypeSampler) {
+    return false;
+  }
+  if (rhs_ty->opcode() == spv::Op::OpTypeSampler) {
+    return true;
+  }
+  // Pick based on variable ID.
+  return lhs->var->result_id() < rhs->var->result_id();
+}
+
+// Summarizes the caller-callee relationships between functions in a module.
+class CallGraph {
+ public:
+  // Returns the list of all functions statically reachable from entry points,
+  // where callees precede callers.
+  const std::vector<uint32_t>& CalleesBeforeCallers() const {
+    return visit_order_;
+  }
+  // Returns the list functions called from a given function.
+  const std::unordered_set<uint32_t>& Callees(uint32_t caller) {
+    return calls_[caller];
+  }
+
+  CallGraph(IRContext& context) {
+    // Populate calls_.
+    std::queue<uint32_t> callee_queue;
+    for (const auto& fn : *context.module()) {
+      auto& callees = calls_[fn.result_id()];
+      context.AddCalls(&fn, &callee_queue);
+      while (!callee_queue.empty()) {
+        callees.insert(callee_queue.front());
+        callee_queue.pop();
+      }
+    }
+
+    // Perform depth-first search, starting from each entry point.
+    // Populates visit_order_.
+    for (const auto& ep : context.module()->entry_points()) {
+      Visit(ep.GetSingleWordInOperand(1));
+    }
+  }
+
+ private:
+  // Visits a function, recursively visiting its callees. Adds this ID
+  // to the visit_order after all callees have been visited.
+  void Visit(uint32_t func_id) {
+    if (visited_.count(func_id)) {
+      return;
+    }
+    visited_.insert(func_id);
+    for (auto callee_id : calls_[func_id]) {
+      Visit(callee_id);
+    }
+    visit_order_.push_back(func_id);
+  }
+
+  // Maps the ID of a function to the IDs of functions it calls.
+  std::unordered_map<uint32_t, std::unordered_set<uint32_t>> calls_;
+
+  // IDs of visited functions;
+  std::unordered_set<uint32_t> visited_;
+  // IDs of functions, where callees precede callers.
+  std::vector<uint32_t> visit_order_;
+};
+
+// Returns vector binding info for all resource variables in the module.
+auto GetVarBindings(IRContext& context) {
+  std::vector<VarBindingInfo> vars;
+  auto* deco_mgr = context.get_decoration_mgr();
+  for (auto& inst : context.module()->types_values()) {
+    if (inst.opcode() == spv::Op::OpVariable) {
+      Instruction* descriptor_set_deco = nullptr;
+      Instruction* binding_deco = nullptr;
+      for (auto* deco : deco_mgr->GetDecorationsFor(inst.result_id(), false)) {
+        switch (static_cast<spv::Decoration>(deco->GetSingleWordInOperand(1))) {
+          case spv::Decoration::DescriptorSet:
+            assert(!descriptor_set_deco);
+            descriptor_set_deco = deco;
+            break;
+          case spv::Decoration::Binding:
+            assert(!binding_deco);
+            binding_deco = deco;
+            break;
+          default:
+            break;
+        }
+      }
+      if (descriptor_set_deco && binding_deco) {
+        vars.push_back({&inst, descriptor_set_deco->GetSingleWordInOperand(2),
+                        binding_deco});
+      }
+    }
+  }
+  return vars;
+}
+
+// Merges the bindings from source into sink. Maintains order and uniqueness
+// within a list of bindings.
+void Merge(DescriptorSets& sink, const DescriptorSets& source) {
+  for (auto index_and_bindings : source) {
+    const uint32_t index = index_and_bindings.first;
+    const BindingList& src1 = index_and_bindings.second;
+    const BindingList& src2 = sink[index];
+    BindingList merged;
+    merged.resize(src1.size() + src2.size());
+    auto merged_end = std::merge(src1.begin(), src1.end(), src2.begin(),
+                                 src2.end(), merged.begin(), Less);
+    auto unique_end = std::unique(merged.begin(), merged_end);
+    merged.resize(unique_end - merged.begin());
+    sink[index] = std::move(merged);
+  }
+}
+
+// Resolves conflicts within this binding list, so the binding number on an
+// item is at least one more than the binding number on the previous item.
+// When this does not yet hold, increase the binding number on the second
+// item in the pair. Returns true if any changes were applied.
+bool ResolveConflicts(BindingList& bl) {
+  bool changed = false;
+  for (size_t i = 1; i < bl.size(); i++) {
+    const auto prev_num = bl[i - 1]->binding();
+    if (prev_num >= bl[i]->binding()) {
+      bl[i]->updateBinding(prev_num + 1);
+      changed = true;
+    }
+  }
+  return changed;
+}
+
+Pass::Status ResolveBindingConflictsPass::Process() {
+  // Assumes the descriptor set and binding decorations are not provided
+  // via decoration groups.  Decoration groups were deprecated in SPIR-V 1.3
+  // Revision 6.  I have not seen any compiler generate them. --dneto
+
+  auto vars = GetVarBindings(*context());
+
+  // Maps a function ID to the variables used directly or indirectly by the
+  // function, organized into descriptor sets. Each descriptor set
+  // consists of a BindingList of distinct variables.
+  std::unordered_map<uint32_t, DescriptorSets> used_vars;
+
+  // Determine variables directly used by functions.
+  auto* def_use_mgr = context()->get_def_use_mgr();
+  for (auto& var : vars) {
+    std::unordered_set<uint32_t> visited_functions_for_var;
+    def_use_mgr->ForEachUser(var.var, [&](Instruction* user) {
+      if (auto* block = context()->get_instr_block(user)) {
+        auto* fn = block->GetParent();
+        assert(fn);
+        const auto fn_id = fn->result_id();
+        if (visited_functions_for_var.insert(fn_id).second) {
+          used_vars[fn_id][var.descriptor_set].push_back(&var);
+        }
+      }
+    });
+  }
+
+  // Sort within a descriptor set by binding number.
+  for (auto& sets_for_fn : used_vars) {
+    for (auto& ds : sets_for_fn.second) {
+      BindingList& bindings = ds.second;
+      std::stable_sort(bindings.begin(), bindings.end(), Less);
+    }
+  }
+
+  // Propagate from callees to callers.
+  CallGraph call_graph(*context());
+  for (const uint32_t caller : call_graph.CalleesBeforeCallers()) {
+    DescriptorSets& caller_ds = used_vars[caller];
+    for (const uint32_t callee : call_graph.Callees(caller)) {
+      Merge(caller_ds, used_vars[callee]);
+    }
+  }
+
+  // At this point, the descriptor sets associated with each entry point
+  // capture exactly the set of resource variables statically used
+  // by the static call tree of that entry point.
+
+  // Resolve conflicts.
+  // VarBindingInfo objects may be shared between the bindings lists.
+  // Updating a binding in one list can require updating another list later.
+  // So repeat updates until settling.
+
+  // The union of BindingLists across all entry points.
+  std::vector<BindingList*> ep_bindings;
+
+  for (auto& ep : context()->module()->entry_points()) {
+    for (auto& ds : used_vars[ep.GetSingleWordInOperand(1)]) {
+      BindingList& bindings = ds.second;
+      ep_bindings.push_back(&bindings);
+    }
+  }
+  bool modified = false;
+  bool found_conflict;
+  do {
+    found_conflict = false;
+    for (BindingList* bl : ep_bindings) {
+      found_conflict |= ResolveConflicts(*bl);
+    }
+    modified |= found_conflict;
+  } while (found_conflict);
+
+  return modified ? Pass::Status::SuccessWithChange
+                  : Pass::Status::SuccessWithoutChange;
+}
+
+}  // namespace opt
+}  // namespace spvtools

third_party/spirv-tools/source/opt/resolve_binding_conflicts_pass.h

@@ -0,0 +1,38 @@
+// Copyright (c) 2025 Google LLC
+//
+// 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 LIBSPIRV_OPT_RESOLVE_BINDING_CONFLICTS_PASS_H_
+#define LIBSPIRV_OPT_RESOLVE_BINDING_CONFLICTS_PASS_H_
+
+#include <unordered_map>
+#include <utility>
+#include <vector>
+
+#include "source/diagnostic.h"
+#include "source/opt/pass.h"
+#include "source/util/small_vector.h"
+
+namespace spvtools {
+namespace opt {
+class ResolveBindingConflictsPass : public Pass {
+ public:
+  virtual ~ResolveBindingConflictsPass() override = default;
+  const char* name() const override { return "resolve-binding-conflicts"; }
+  IRContext::Analysis GetPreservedAnalyses() override;
+  Status Process() override;
+};
+}  // namespace opt
+}  // namespace spvtools
+
+#endif  // LIBSPIRV_OPT_RESOLVE_BINDING_CONFLICTS_PASS_H_

third_party/spirv-tools/source/opt/scalar_replacement_pass.h

@@ -33,7 +33,7 @@ namespace opt {
 // Documented in optimizer.hpp
 class ScalarReplacementPass : public MemPass {
  private:
-  static constexpr uint32_t kDefaultLimit = 100;
+  static constexpr uint32_t kDefaultLimit = 0;
 
  public:
   ScalarReplacementPass(uint32_t limit = kDefaultLimit)

third_party/spirv-tools/source/opt/split_combined_image_sampler_pass.cpp

@@ -0,0 +1,648 @@
+// Copyright (c) 2025 Google LLC
+//
+// 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 "source/opt/split_combined_image_sampler_pass.h"
+
+#include <algorithm>
+#include <cassert>
+#include <memory>
+
+#include "source/opt/instruction.h"
+#include "source/opt/ir_builder.h"
+#include "source/opt/ir_context.h"
+#include "source/opt/type_manager.h"
+#include "source/opt/types.h"
+#include "source/util/make_unique.h"
+#include "source/util/string_utils.h"
+#include "spirv/unified1/spirv.h"
+
+namespace spvtools {
+namespace opt {
+
+#define CHECK(cond)                                          \
+  {                                                          \
+    if ((cond) != SPV_SUCCESS) return Pass::Status::Failure; \
+  }
+
+#define CHECK_STATUS(cond)                           \
+  {                                                  \
+    if (auto c = (cond); c != SPV_SUCCESS) return c; \
+  }
+
+IRContext::Analysis SplitCombinedImageSamplerPass::GetPreservedAnalyses() {
+  return
+      // def use manager is updated
+      IRContext::kAnalysisDefUse
+
+      // decorations are updated
+      | IRContext::kAnalysisDecorations
+
+      // control flow is not changed
+      | IRContext::kAnalysisCFG           //
+      | IRContext::kAnalysisLoopAnalysis  //
+      | IRContext::kAnalysisStructuredCFG
+
+      // type manager is updated
+      | IRContext::kAnalysisTypes;
+}
+
+Pass::Status SplitCombinedImageSamplerPass::Process() {
+  def_use_mgr_ = context()->get_def_use_mgr();
+  type_mgr_ = context()->get_type_mgr();
+
+  FindCombinedTextureSamplers();
+  if (combined_types_to_remove_.empty() && !sampled_image_used_as_param_) {
+    return Ok();
+  }
+
+  CHECK(RemapFunctions());
+  CHECK(RemapVars());
+  CHECK(RemoveDeadTypes());
+
+  def_use_mgr_ = nullptr;
+  type_mgr_ = nullptr;
+
+  return Ok();
+}
+
+spvtools::DiagnosticStream SplitCombinedImageSamplerPass::Fail() {
+  return std::move(
+      spvtools::DiagnosticStream({}, consumer(), "", SPV_ERROR_INVALID_BINARY)
+      << "split-combined-image-sampler: ");
+}
+
+void SplitCombinedImageSamplerPass::FindCombinedTextureSamplers() {
+  for (auto& inst : context()->types_values()) {
+    RegisterGlobal(inst.result_id());
+    switch (inst.opcode()) {
+      case spv::Op::OpTypeSampler:
+        // Modules can't have duplicate sampler types.
+        assert(!sampler_type_);
+        sampler_type_ = &inst;
+        break;
+
+      case spv::Op::OpTypeSampledImage:
+        if (!first_sampled_image_type_) {
+          first_sampled_image_type_ = &inst;
+        }
+        combined_types_.insert(inst.result_id());
+        def_use_mgr_->WhileEachUser(inst.result_id(), [&](Instruction* i) {
+          sampled_image_used_as_param_ |=
+              i->opcode() == spv::Op::OpTypeFunction;
+          return !sampled_image_used_as_param_;
+        });
+        break;
+
+      case spv::Op::OpTypeArray:
+      case spv::Op::OpTypeRuntimeArray: {
+        auto pointee_id = inst.GetSingleWordInOperand(0);
+        if (combined_types_.find(pointee_id) != combined_types_.end()) {
+          combined_types_.insert(inst.result_id());
+          combined_types_to_remove_.push_back(inst.result_id());
+        }
+      } break;
+
+      case spv::Op::OpTypePointer: {
+        auto sc =
+            static_cast<spv::StorageClass>(inst.GetSingleWordInOperand(0));
+        if (sc == spv::StorageClass::UniformConstant) {
+          auto pointee_id = inst.GetSingleWordInOperand(1);
+          if (combined_types_.find(pointee_id) != combined_types_.end()) {
+            combined_types_.insert(inst.result_id());
+            combined_types_to_remove_.push_back(inst.result_id());
+          }
+        }
+      } break;
+
+      case spv::Op::OpVariable:
+        if (combined_types_.find(inst.type_id()) != combined_types_.end()) {
+          ordered_vars_.push_back(&inst);
+        }
+        break;
+
+      default:
+        break;
+    }
+  }
+}
+
+Instruction* SplitCombinedImageSamplerPass::GetSamplerType() {
+  if (!sampler_type_) {
+    analysis::Sampler s;
+    uint32_t sampler_type_id = type_mgr_->GetTypeInstruction(&s);
+    sampler_type_ = def_use_mgr_->GetDef(sampler_type_id);
+    assert(first_sampled_image_type_);
+    sampler_type_->InsertBefore(first_sampled_image_type_);
+    RegisterNewGlobal(sampler_type_->result_id());
+  }
+  return sampler_type_;
+}
+
+spv_result_t SplitCombinedImageSamplerPass::RemapVars() {
+  for (Instruction* var : ordered_vars_) {
+    CHECK_STATUS(RemapVar(var));
+  }
+  return SPV_SUCCESS;
+}
+
+std::pair<Instruction*, Instruction*> SplitCombinedImageSamplerPass::SplitType(
+    Instruction& combined_kind_type) {
+  if (auto where = type_remap_.find(combined_kind_type.result_id());
+      where != type_remap_.end()) {
+    auto& type_remap = where->second;
+    return {type_remap.image_kind_type, type_remap.sampler_kind_type};
+  }
+
+  switch (combined_kind_type.opcode()) {
+    case spv::Op::OpTypeSampledImage: {
+      auto* image_type =
+          def_use_mgr_->GetDef(combined_kind_type.GetSingleWordInOperand(0));
+      auto* sampler_type = GetSamplerType();
+      type_remap_[combined_kind_type.result_id()] = {&combined_kind_type,
+                                                     image_type, sampler_type};
+      return {image_type, sampler_type};
+      break;
+    }
+    case spv::Op::OpTypePointer: {
+      auto sc = static_cast<spv::StorageClass>(
+          combined_kind_type.GetSingleWordInOperand(0));
+      if (sc == spv::StorageClass::UniformConstant) {
+        auto* pointee =
+            def_use_mgr_->GetDef(combined_kind_type.GetSingleWordInOperand(1));
+        auto [image_pointee, sampler_pointee] = SplitType(*pointee);
+        // These would be null if the pointee is an image type or a sampler
+        // type. Don't decompose them. Currently this method does not check the
+        // assumption that it is being only called on combined types. So code
+        // this defensively.
+        if (image_pointee && sampler_pointee) {
+          auto* ptr_image = MakeUniformConstantPointer(image_pointee);
+          auto* ptr_sampler = MakeUniformConstantPointer(sampler_pointee);
+          type_remap_[combined_kind_type.result_id()] = {
+              &combined_kind_type, ptr_image, ptr_sampler};
+          return {ptr_image, ptr_sampler};
+        }
+      }
+      break;
+    }
+    case spv::Op::OpTypeArray: {
+      const auto* array_ty =
+          type_mgr_->GetType(combined_kind_type.result_id())->AsArray();
+      assert(array_ty);
+      const auto* sampled_image_ty = array_ty->element_type()->AsSampledImage();
+      assert(sampled_image_ty);
+
+      const analysis::Type* image_ty = sampled_image_ty->image_type();
+      assert(image_ty);
+      analysis::Array array_image_ty(image_ty, array_ty->length_info());
+      const uint32_t array_image_ty_id =
+          type_mgr_->GetTypeInstruction(&array_image_ty);
+      auto* array_image_ty_inst = def_use_mgr_->GetDef(array_image_ty_id);
+      if (!IsKnownGlobal(array_image_ty_id)) {
+        array_image_ty_inst->InsertBefore(&combined_kind_type);
+        RegisterNewGlobal(array_image_ty_id);
+        // GetTypeInstruction also updated the def-use manager.
+      }
+
+      analysis::Array sampler_array_ty(
+          type_mgr_->GetType(GetSamplerType()->result_id()),
+          array_ty->length_info());
+      const uint32_t array_sampler_ty_id =
+          type_mgr_->GetTypeInstruction(&sampler_array_ty);
+      auto* array_sampler_ty_inst = def_use_mgr_->GetDef(array_sampler_ty_id);
+      if (!IsKnownGlobal(array_sampler_ty_id)) {
+        array_sampler_ty_inst->InsertBefore(&combined_kind_type);
+        RegisterNewGlobal(array_sampler_ty_id);
+        // GetTypeInstruction also updated the def-use manager.
+      }
+      return {array_image_ty_inst, array_sampler_ty_inst};
+    }
+    case spv::Op::OpTypeRuntimeArray: {
+      // This is like the sized-array case, but there is no length parameter.
+      auto* array_ty =
+          type_mgr_->GetType(combined_kind_type.result_id())->AsRuntimeArray();
+      assert(array_ty);
+      auto* sampled_image_ty = array_ty->element_type()->AsSampledImage();
+      assert(sampled_image_ty);
+
+      const analysis::Type* image_ty = sampled_image_ty->image_type();
+      assert(image_ty);
+      analysis::RuntimeArray array_image_ty(image_ty);
+      const uint32_t array_image_ty_id =
+          type_mgr_->GetTypeInstruction(&array_image_ty);
+      auto* array_image_ty_inst = def_use_mgr_->GetDef(array_image_ty_id);
+      if (!IsKnownGlobal(array_image_ty_id)) {
+        array_image_ty_inst->InsertBefore(&combined_kind_type);
+        RegisterNewGlobal(array_image_ty_id);
+        // GetTypeInstruction also updated the def-use manager.
+      }
+
+      analysis::RuntimeArray sampler_array_ty(
+          type_mgr_->GetType(GetSamplerType()->result_id()));
+      const uint32_t array_sampler_ty_id =
+          type_mgr_->GetTypeInstruction(&sampler_array_ty);
+      auto* array_sampler_ty_inst = def_use_mgr_->GetDef(array_sampler_ty_id);
+      if (!IsKnownGlobal(array_sampler_ty_id)) {
+        array_sampler_ty_inst->InsertBefore(&combined_kind_type);
+        RegisterNewGlobal(array_sampler_ty_id);
+        // GetTypeInstruction also updated the def-use manager.
+      }
+      return {array_image_ty_inst, array_sampler_ty_inst};
+    }
+    default:
+      break;
+  }
+  return {nullptr, nullptr};
+}
+
+spv_result_t SplitCombinedImageSamplerPass::RemapVar(
+    Instruction* combined_var) {
+  InstructionBuilder builder(context(), combined_var,
+                             IRContext::kAnalysisDefUse);
+
+  // Create an image variable, and a sampler variable.
+  auto* combined_var_type = def_use_mgr_->GetDef(combined_var->type_id());
+  auto [ptr_image_ty, ptr_sampler_ty] = SplitType(*combined_var_type);
+  assert(ptr_image_ty);
+  assert(ptr_sampler_ty);
+  Instruction* sampler_var = builder.AddVariable(
+      ptr_sampler_ty->result_id(), SpvStorageClassUniformConstant);
+  Instruction* image_var = builder.AddVariable(ptr_image_ty->result_id(),
+                                               SpvStorageClassUniformConstant);
+  modified_ = true;
+  return RemapUses(combined_var, image_var, sampler_var);
+}
+
+spv_result_t SplitCombinedImageSamplerPass::RemapUses(
+    Instruction* combined, Instruction* image_part, Instruction* sampler_part) {
+  // The instructions to delete.
+  std::unordered_set<Instruction*> dead_insts;
+  // The insertion point should be updated before using this builder.
+  // We needed *something* here.
+  InstructionBuilder builder(context(), combined, IRContext::kAnalysisDefUse);
+
+  // This code must maintain the SPIR-V "Data rule" about sampled image values:
+  //  > All OpSampledImage instructions, or instructions that load an image or
+  //  > sampler reference, must be in the same block in which their Result <id>
+  //  > are consumed.
+  //
+  // When the code below inserts OpSampledImage instructions, it is always
+  // either:
+  // - in the same block as the previous OpSampledImage instruction it is
+  //   replacing, or
+  // - in the same block as the instruction using sampled image value it is
+  //   replacing.
+  //
+  // Assuming that rule is already honoured by the module, these updates will
+  // continue to honour the rule.
+
+  // Represents a single use of a value to be remapped.
+  struct RemapUse {
+    uint32_t used_id;  // The ID that is being used.
+    Instruction* user;
+    uint32_t index;
+    Instruction* image_part;    // The image part of the replacement.
+    Instruction* sampler_part;  // The sampler part of the replacement.
+  };
+  // The work list of uses to be remapped.
+  std::vector<RemapUse> uses;
+
+  // Adds remap records for each use of a value to be remapped.
+  // Also schedules the original value for deletion.
+  auto add_remap = [this, &dead_insts, &uses](Instruction* combined_arg,
+                                              Instruction* image_part_arg,
+                                              Instruction* sampler_part_arg) {
+    const uint32_t used_combined_id = combined_arg->result_id();
+
+    def_use_mgr_->ForEachUse(
+        combined_arg, [&](Instruction* user, uint32_t use_index) {
+          uses.push_back({used_combined_id, user, use_index, image_part_arg,
+                          sampler_part_arg});
+        });
+    dead_insts.insert(combined_arg);
+  };
+
+  add_remap(combined, image_part, sampler_part);
+
+  // Use index-based iteration because we can add to the work list as we go
+  // along, and reallocation would invalidate ordinary iterators.
+  for (size_t use_index = 0; use_index < uses.size(); ++use_index) {
+    auto& use = uses[use_index];
+    switch (use.user->opcode()) {
+      case spv::Op::OpCopyObject: {
+        // Append the uses of this OpCopyObject to the work list.
+        add_remap(use.user, image_part, sampler_part);
+        break;
+      }
+      case spv::Op::OpLoad: {
+        assert(use.index == 2 && "variable used as non-pointer index on load");
+        Instruction* load = use.user;
+
+        // Assume the loaded value is a sampled image.
+        assert(def_use_mgr_->GetDef(load->type_id())->opcode() ==
+               spv::Op::OpTypeSampledImage);
+
+        // Create loads for the image part and sampler part.
+        builder.SetInsertPoint(load);
+        auto* image = builder.AddLoad(PointeeTypeId(use.image_part),
+                                      use.image_part->result_id());
+        auto* sampler = builder.AddLoad(PointeeTypeId(use.sampler_part),
+                                        use.sampler_part->result_id());
+
+        // Move decorations, such as RelaxedPrecision.
+        auto* deco_mgr = context()->get_decoration_mgr();
+        deco_mgr->CloneDecorations(load->result_id(), image->result_id());
+        deco_mgr->CloneDecorations(load->result_id(), sampler->result_id());
+        deco_mgr->RemoveDecorationsFrom(load->result_id());
+
+        // Create a sampled image from the loads of the two parts.
+        auto* sampled_image = builder.AddSampledImage(
+            load->type_id(), image->result_id(), sampler->result_id());
+        // Replace the original sampled image value with the new one.
+        std::unordered_set<Instruction*> users;
+        def_use_mgr_->ForEachUse(
+            load, [&users, sampled_image](Instruction* user, uint32_t index) {
+              user->SetOperand(index, {sampled_image->result_id()});
+              users.insert(user);
+            });
+        for (auto* user : users) {
+          def_use_mgr_->AnalyzeInstUse(user);
+        }
+        dead_insts.insert(load);
+        break;
+      }
+      case spv::Op::OpDecorate: {
+        assert(use.index == 0 && "variable used as non-target index");
+        builder.SetInsertPoint(use.user);
+        spv::Decoration deco{use.user->GetSingleWordInOperand(1)};
+        std::vector<uint32_t> literals;
+        for (uint32_t i = 2; i < use.user->NumInOperands(); i++) {
+          literals.push_back(use.user->GetSingleWordInOperand(i));
+        }
+        builder.AddDecoration(use.image_part->result_id(), deco, literals);
+        builder.AddDecoration(use.sampler_part->result_id(), deco, literals);
+        // KillInst will delete names and decorations, so don't schedule a
+        // deletion of this instruction.
+        break;
+      }
+      case spv::Op::OpEntryPoint: {
+        // The entry point lists variables in the shader interface, i.e.
+        // module-scope variables referenced by the static call tree rooted
+        // at the entry point. (It can be a proper superset).  Before SPIR-V
+        // 1.4, only Input and Output variables are listed; in 1.4 and later,
+        // module-scope variables in all storage classes are listed.
+        // If a combined image+sampler is listed by the entry point, then
+        // the separated image and sampler variables should be.
+        assert(use.index >= 3 &&
+               "variable used in OpEntryPoint but not as an interface ID");
+        use.user->SetOperand(use.index, {use.image_part->result_id()});
+        use.user->InsertOperand(
+            use.user->NumOperands(),
+            {SPV_OPERAND_TYPE_ID, {use.sampler_part->result_id()}});
+        def_use_mgr_->AnalyzeInstUse(use.user);
+        break;
+      }
+      case spv::Op::OpName: {
+        // Synthesize new names from the old.
+        const auto name = use.user->GetOperand(1).AsString();
+        AddOpName(use.image_part->result_id(), name + "_image");
+        AddOpName(use.sampler_part->result_id(), name + "_sampler");
+
+        // KillInst will delete names and decorations, so don't schedule a
+        // deletion of this instruction.
+        break;
+      }
+      case spv::Op::OpFunctionCall: {
+        // Replace each combined arg with two args: the image part, then the
+        // sampler part.
+        // The combined value could have been used twice in the argument list.
+        // Moving things around now will invalidate the 'use' list above.
+        // So don't trust the use index value.
+        auto& call = *use.user;
+        // The insert API only takes absolute arg IDs, not "in" arg IDs.
+        const auto first_arg_operand_index = 3;  // Skip the callee ID
+        for (uint32_t i = first_arg_operand_index; i < call.NumOperands();
+             ++i) {
+          if (use.used_id == call.GetSingleWordOperand(i)) {
+            call.SetOperand(i, {use.sampler_part->result_id()});
+            call.InsertOperand(
+                i, {SPV_OPERAND_TYPE_ID, {use.image_part->result_id()}});
+            ++i;
+          }
+        }
+        def_use_mgr_->AnalyzeInstUse(&call);
+        break;
+      }
+      case spv::Op::OpAccessChain:
+      case spv::Op::OpInBoundsAccessChain: {
+        auto* original_access_chain = use.user;
+        builder.SetInsertPoint(original_access_chain);
+        // It can only be the base pointer
+        assert(use.index == 2);
+
+        // Replace the original access chain with access chains for the image
+        // part and the sampler part.
+        std::vector<uint32_t> indices;
+        for (uint32_t i = 3; i < original_access_chain->NumOperands(); i++) {
+          indices.push_back(original_access_chain->GetSingleWordOperand(i));
+        }
+
+        auto [result_image_part_ty, result_sampler_part_ty] =
+            SplitType(*def_use_mgr_->GetDef(original_access_chain->type_id()));
+        auto* result_image_part = builder.AddOpcodeAccessChain(
+            use.user->opcode(), result_image_part_ty->result_id(),
+            use.image_part->result_id(), indices);
+        auto* result_sampler_part = builder.AddOpcodeAccessChain(
+            use.user->opcode(), result_sampler_part_ty->result_id(),
+            use.sampler_part->result_id(), indices);
+
+        // Remap uses of the original access chain.
+        add_remap(original_access_chain, result_image_part,
+                  result_sampler_part);
+        break;
+      }
+      default: {
+        uint32_t used_type_id = def_use_mgr_->GetDef(use.used_id)->type_id();
+        auto* used_type = def_use_mgr_->GetDef(used_type_id);
+        if (used_type->opcode() == spv::Op::OpTypeSampledImage) {
+          // This value being used is a sampled image value.  But it's
+          // being replaced, so recreate it here.
+          // Example: used by OpImage, OpImageSampleExplicitLod, etc.
+          builder.SetInsertPoint(use.user);
+          auto* sampled_image =
+              builder.AddSampledImage(used_type_id, use.image_part->result_id(),
+                                      use.sampler_part->result_id());
+          use.user->SetOperand(use.index, {sampled_image->result_id()});
+          def_use_mgr_->AnalyzeInstUse(use.user);
+          break;
+        }
+        return Fail() << "unhandled user: " << *use.user;
+      }
+    }
+  }
+
+  for (auto* inst : dead_insts) {
+    KillInst(inst);
+  }
+
+  return SPV_SUCCESS;
+}
+
+spv_result_t SplitCombinedImageSamplerPass::RemapFunctions() {
+  // Remap function types. A combined type can appear as a parameter, but not as
+  // the return type.
+  {
+    std::unordered_set<Instruction*> dead_insts;
+    for (auto& inst : context()->types_values()) {
+      if (inst.opcode() != spv::Op::OpTypeFunction) {
+        continue;
+      }
+      analysis::Function* f_ty =
+          type_mgr_->GetType(inst.result_id())->AsFunction();
+      std::vector<const analysis::Type*> new_params;
+      for (const auto* param_ty : f_ty->param_types()) {
+        const auto param_ty_id = type_mgr_->GetId(param_ty);
+        if (combined_types_.find(param_ty_id) != combined_types_.end()) {
+          auto* param_type = def_use_mgr_->GetDef(param_ty_id);
+          auto [image_type, sampler_type] = SplitType(*param_type);
+          assert(image_type);
+          assert(sampler_type);
+          // The image and sampler types must already exist, so there is no
+          // need to move them to the right spot.
+          new_params.push_back(type_mgr_->GetType(image_type->result_id()));
+          new_params.push_back(type_mgr_->GetType(sampler_type->result_id()));
+        } else {
+          new_params.push_back(param_ty);
+        }
+      }
+      if (new_params.size() != f_ty->param_types().size()) {
+        // Replace this type.
+        analysis::Function new_f_ty(f_ty->return_type(), new_params);
+        const uint32_t new_f_ty_id = type_mgr_->GetTypeInstruction(&new_f_ty);
+        std::unordered_set<Instruction*> users;
+        def_use_mgr_->ForEachUse(
+            &inst,
+            [&users, new_f_ty_id](Instruction* user, uint32_t use_index) {
+              user->SetOperand(use_index, {new_f_ty_id});
+              users.insert(user);
+            });
+        for (auto* user : users) {
+          def_use_mgr_->AnalyzeInstUse(user);
+        }
+        dead_insts.insert(&inst);
+      }
+    }
+    for (auto* inst : dead_insts) {
+      KillInst(inst);
+    }
+  }
+
+  // Rewite OpFunctionParameter in function definitions.
+  for (Function& fn : *context()->module()) {
+    // Rewrite the function parameters and record their replacements.
+    struct Replacement {
+      Instruction* combined;
+      Instruction* image;
+      Instruction* sampler;
+    };
+    std::vector<Replacement> replacements;
+
+    Function::RewriteParamFn rewriter =
+        [&](std::unique_ptr<Instruction>&& param,
+            std::back_insert_iterator<Function::ParamList>& appender) {
+          if (combined_types_.count(param->type_id()) == 0) {
+            appender = std::move(param);
+            return;
+          }
+
+          // Replace this parameter with two new parameters.
+          auto* combined_inst = param.release();
+          auto* combined_type = def_use_mgr_->GetDef(combined_inst->type_id());
+          auto [image_type, sampler_type] = SplitType(*combined_type);
+          auto image_param = MakeUnique<Instruction>(
+              context(), spv::Op::OpFunctionParameter, image_type->result_id(),
+              context()->TakeNextId(), Instruction::OperandList{});
+          auto sampler_param = MakeUnique<Instruction>(
+              context(), spv::Op::OpFunctionParameter,
+              sampler_type->result_id(), context()->TakeNextId(),
+              Instruction::OperandList{});
+          replacements.push_back(
+              {combined_inst, image_param.get(), sampler_param.get()});
+          appender = std::move(image_param);
+          appender = std::move(sampler_param);
+        };
+    fn.RewriteParams(rewriter);
+
+    for (auto& r : replacements) {
+      modified_ = true;
+      def_use_mgr_->AnalyzeInstDefUse(r.image);
+      def_use_mgr_->AnalyzeInstDefUse(r.sampler);
+      CHECK_STATUS(RemapUses(r.combined, r.image, r.sampler));
+    }
+  }
+  return SPV_SUCCESS;
+}
+
+Instruction* SplitCombinedImageSamplerPass::MakeUniformConstantPointer(
+    Instruction* pointee) {
+  uint32_t ptr_id = type_mgr_->FindPointerToType(
+      pointee->result_id(), spv::StorageClass::UniformConstant);
+  auto* ptr = def_use_mgr_->GetDef(ptr_id);
+  if (!IsKnownGlobal(ptr_id)) {
+    // The pointer type was created at the end. Put it right after the
+    // pointee.
+    ptr->InsertBefore(pointee);
+    pointee->InsertBefore(ptr);
+    RegisterNewGlobal(ptr_id);
+    // FindPointerToType also updated the def-use manager.
+  }
+  return ptr;
+}
+
+void SplitCombinedImageSamplerPass::AddOpName(uint32_t id,
+                                              const std::string& name) {
+  std::unique_ptr<Instruction> opname{new Instruction{
+      context(),
+      spv::Op::OpName,
+      0u,
+      0u,
+      {{SPV_OPERAND_TYPE_ID, {id}},
+       {SPV_OPERAND_TYPE_LITERAL_STRING,
+        utils::MakeVector<spvtools::opt::Operand::OperandData>(name)}}}};
+
+  context()->AddDebug2Inst(std::move(opname));
+}
+
+spv_result_t SplitCombinedImageSamplerPass::RemoveDeadTypes() {
+  for (auto dead_type_id : combined_types_to_remove_) {
+    if (auto* ty = def_use_mgr_->GetDef(dead_type_id)) {
+      KillInst(ty);
+    }
+  }
+  return SPV_SUCCESS;
+}
+
+void SplitCombinedImageSamplerPass::KillInst(Instruction* inst) {
+  // IRContext::KillInst will remove associated debug instructions and
+  // decorations. It will delete the object only if it is already in a list.
+  const bool was_in_list = inst->IsInAList();
+  context()->KillInst(inst);
+  if (!was_in_list) {
+    // Avoid leaking
+    delete inst;
+  }
+  modified_ = true;
+}
+
+}  // namespace opt
+}  // namespace spvtools

third_party/spirv-tools/source/opt/split_combined_image_sampler_pass.h

@@ -0,0 +1,169 @@
+// Copyright (c) 2025 Google LLC
+//
+// 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 LIBSPIRV_OPT_SPLIT_COMBINED_IMAGE_SAMPLER_PASS_H_
+#define LIBSPIRV_OPT_SPLIT_COMBINED_IMAGE_SAMPLER_PASS_H_
+
+#include <unordered_map>
+#include <utility>
+#include <vector>
+
+#include "source/diagnostic.h"
+#include "source/opt/decoration_manager.h"
+#include "source/opt/def_use_manager.h"
+#include "source/opt/pass.h"
+#include "source/opt/type_manager.h"
+
+namespace spvtools {
+namespace opt {
+
+// Replaces each combined-image sampler variable with an image variable
+// and a sampler variable. Similar for function parameters.
+//
+// Copy the descriptor set and binding number. Vulkan allows this, surprisingly.
+class SplitCombinedImageSamplerPass : public Pass {
+ public:
+  virtual ~SplitCombinedImageSamplerPass() override = default;
+  const char* name() const override { return "split-combined-image-sampler"; }
+  IRContext::Analysis GetPreservedAnalyses() override;
+  Status Process() override;
+
+ private:
+  // Records failure for the current module, and returns a stream
+  // that can be used to provide user error information to the message
+  // consumer.
+  spvtools::DiagnosticStream Fail();
+
+  // Find variables that contain combined texture-samplers, or arrays of them.
+  // Also populate known_globals_.
+  void FindCombinedTextureSamplers();
+
+  // Returns the sampler type. If it does not yet exist, then it is created
+  // and placed before the first sampled image type.
+  Instruction* GetSamplerType();
+
+  // Remaps function types and function declarations.  Each
+  // pointer-to-sampled-image-type operand is replaced with a pair of
+  // pointer-to-image-type and pointer-to-sampler-type pair.
+  // Updates the def-use manager and type manager.
+  spv_result_t RemapFunctions();
+  // Remap resource variables.
+  // Updates the def-use manager.
+  spv_result_t RemapVars();
+  // Remap a single resource variable for combined var.
+  // Updates the def-use manager and the decorations manager.
+  spv_result_t RemapVar(Instruction* combined_var);
+  // Transitively remaps uses of the combined object with uses of the
+  // decomposed image and sampler parts.  The combined object can be sampled
+  // image value, a pointer to one, an array of one, or a pointer to an array
+  // of one. The image and sampler parts have corresponding shapes.
+  // Updates the def-use manager and the decorations manager.
+  spv_result_t RemapUses(Instruction* combined, Instruction* image_part,
+                         Instruction* sampler_part);
+  // Removes types that are no longer referenced.
+  spv_result_t RemoveDeadTypes();
+
+  // Returns the type instruction for a UniformConstant pointer to the given
+  // pointee type. If it does not yet exist, the new type instruction is created
+  // and placed immediately after the pointee type instruction. Updates def-use
+  // and type managers, and the set of known globals.
+  Instruction* MakeUniformConstantPointer(Instruction* pointee);
+
+  // Returns the ID of the pointee type for a pointer value instruction.
+  uint32_t PointeeTypeId(Instruction* ptr_value) {
+    auto* ptr_ty = def_use_mgr_->GetDef(ptr_value->type_id());
+    assert(ptr_ty->opcode() == spv::Op::OpTypePointer);
+    return ptr_ty->GetSingleWordInOperand(1);
+  }
+
+  // Creates a new OpName instruction mapping the given name to the given
+  // string, and adds it to the module at the end of the OpName and OpMemberName
+  // section.
+  void AddOpName(uint32_t id, const std::string& name);
+
+  // Cached from the IRContext. Valid while Process() is running.
+  analysis::DefUseManager* def_use_mgr_ = nullptr;
+  // Cached from the IRContext. Valid while Process() is running.
+  analysis::TypeManager* type_mgr_ = nullptr;
+
+  // Did processing modify the module?
+  bool modified_ = false;
+  Pass::Status Ok() {
+    return modified_ ? Pass::Status::SuccessWithChange
+                     : Pass::Status::SuccessWithoutChange;
+  }
+
+  // The first OpTypeSampledImage instruction in the module, if one exists.
+  Instruction* first_sampled_image_type_ = nullptr;
+  // An OpTypeSampler instruction, if one existed already, or if we created one.
+  Instruction* sampler_type_ = nullptr;
+
+  // The known types and module-scope values.
+  // We use this to know when a new such value was created.
+  std::unordered_set<uint32_t> known_globals_;
+  bool IsKnownGlobal(uint32_t id) const {
+    return known_globals_.find(id) != known_globals_.end();
+  }
+  void RegisterGlobal(uint32_t id) { known_globals_.insert(id); }
+  void RegisterNewGlobal(uint32_t id) {
+    modified_ = true;
+    RegisterGlobal(id);
+  }
+
+  // Deletes an instruction and associated debug and decoration instructions.
+  // Updates the def-use manager.
+  void KillInst(Instruction* inst);
+
+  // Combined types.  The known combined sampled-image type,
+  // and recursively pointers or arrays of them.
+  std::unordered_set<uint32_t> combined_types_;
+  // The pre-existing types this pass should remove: pointer to
+  // combined type, array of combined type, pointer to array of combined type.
+  std::vector<uint32_t> combined_types_to_remove_;
+  // Is an OpTypeSampledImage used as a function parameter? Those should be
+  // transformed.
+  bool sampled_image_used_as_param_ = false;
+
+  // Remaps a combined-kind type to corresponding sampler-kind and image-kind
+  // of type.
+  struct TypeRemapInfo {
+    // The instruction for the combined type, pointer to combined type,
+    // or point to array of combined type.
+    Instruction* combined_kind_type;
+    // The corresponding image type, with the same shape of indirection as the
+    // combined_kind_type.
+    Instruction* image_kind_type;
+    // The corresponding sampler type, with the same shape of indirection as the
+    // combined_kind_type.
+    Instruction* sampler_kind_type;
+  };
+  // Maps the ID of a combined-image-sampler type kind to its corresponding
+  // split parts.
+  std::unordered_map<uint32_t, TypeRemapInfo> type_remap_;
+
+  // Returns the image-like and sampler-like types of the same indirection shape
+  // as the given combined-like type.  If combined_kind_type is not a combined
+  // type or a pointer to one, or an array of one or a pointer to an array of
+  // one, then returns a pair of null pointer. Either both components are
+  // non-null, or both components are null. Updates the def-use manager and the
+  // type manager if new instructions are created.
+  std::pair<Instruction*, Instruction*> SplitType(
+      Instruction& combined_kind_type);
+
+  // The combined-image-sampler variables to be replaced.
+  std::vector<Instruction*> ordered_vars_;
+};
+}  // namespace opt
+}  // namespace spvtools
+#endif  // LIBSPIRV_OPT_SPLIT_COMBINED_IMAGE_SAMPLER_PASS_H_

third_party/spirv-tools/source/opt/trim_capabilities_pass.cpp

@@ -241,6 +241,37 @@ Handler_OpTypePointer_StorageUniformBufferBlock16(
              : std::nullopt;
 }
 
+static std::optional<spv::Capability>
+Handler_OpTypePointer_StorageBuffer16BitAccess(const Instruction* instruction) {
+  assert(instruction->opcode() == spv::Op::OpTypePointer &&
+         "This handler only support OpTypePointer opcodes.");
+
+  // Requires StorageBuffer, ShaderRecordBufferKHR or PhysicalStorageBuffer
+  // storage classes.
+  spv::StorageClass storage_class = spv::StorageClass(
+      instruction->GetSingleWordInOperand(kOpTypePointerStorageClassIndex));
+  if (storage_class != spv::StorageClass::StorageBuffer &&
+      storage_class != spv::StorageClass::ShaderRecordBufferKHR &&
+      storage_class != spv::StorageClass::PhysicalStorageBuffer) {
+    return std::nullopt;
+  }
+
+  const auto* decoration_mgr = instruction->context()->get_decoration_mgr();
+  const bool matchesCondition =
+      AnyTypeOf(instruction, [decoration_mgr](const Instruction* item) {
+        if (!decoration_mgr->HasDecoration(item->result_id(),
+                                           spv::Decoration::Block)) {
+          return false;
+        }
+
+        return AnyTypeOf(item, is16bitType);
+      });
+
+  return matchesCondition
+             ? std::optional(spv::Capability::StorageBuffer16BitAccess)
+             : std::nullopt;
+}
+
 static std::optional<spv::Capability> Handler_OpTypePointer_StorageUniform16(
     const Instruction* instruction) {
   assert(instruction->opcode() == spv::Op::OpTypePointer &&
@@ -388,7 +419,7 @@ Handler_OpImageSparseRead_StorageImageReadWithoutFormat(
 }
 
 // Opcode of interest to determine capabilities requirements.
-constexpr std::array<std::pair<spv::Op, OpcodeHandler>, 13> kOpcodeHandlers{{
+constexpr std::array<std::pair<spv::Op, OpcodeHandler>, 14> kOpcodeHandlers{{
     // clang-format off
     {spv::Op::OpImageRead,         Handler_OpImageRead_StorageImageReadWithoutFormat},
     {spv::Op::OpImageWrite,        Handler_OpImageWrite_StorageImageWriteWithoutFormat},
@@ -403,6 +434,7 @@ constexpr std::array<std::pair<spv::Op, OpcodeHandler>, 13> kOpcodeHandlers{{
     {spv::Op::OpTypePointer,       Handler_OpTypePointer_StorageUniform16},
     {spv::Op::OpTypePointer,       Handler_OpTypePointer_StorageUniform16},
     {spv::Op::OpTypePointer,       Handler_OpTypePointer_StorageUniformBufferBlock16},
+    {spv::Op::OpTypePointer,       Handler_OpTypePointer_StorageBuffer16BitAccess},
     // clang-format on
 }};
 

third_party/spirv-tools/source/opt/trim_capabilities_pass.h

@@ -99,6 +99,7 @@ class TrimCapabilitiesPass : public Pass {
       spv::Capability::RayTraversalPrimitiveCullingKHR,
       spv::Capability::Shader,
       spv::Capability::ShaderClockKHR,
+      spv::Capability::StorageBuffer16BitAccess,
       spv::Capability::StorageImageReadWithoutFormat,
       spv::Capability::StorageImageWriteWithoutFormat,
       spv::Capability::StorageInputOutput16,

third_party/spirv-tools/source/opt/type_manager.cpp

@@ -495,6 +495,8 @@ uint32_t TypeManager::GetTypeInstruction(const Type* type) {
       break;
   }
   context()->AddType(std::move(typeInst));
+  // TODO(dneto): This next call to AnalyzeDefUse is redundant becaues
+  // IRContext::AddType already does it.
   context()->AnalyzeDefUse(&*--context()->types_values_end());
   AttachDecorations(id, type);
   return id;

third_party/spirv-tools/source/spirv_validator_options.cpp

@@ -131,6 +131,11 @@ void spvValidatorOptionsSetAllowOffsetTextureOperand(
   options->allow_offset_texture_operand = val;
 }
 
+void spvValidatorOptionsSetAllowVulkan32BitBitwise(
+    spv_validator_options options, bool val) {
+  options->allow_vulkan_32_bit_bitwise = val;
+}
+
 void spvValidatorOptionsSetFriendlyNames(spv_validator_options options,
                                          bool val) {
   options->use_friendly_names = val;

third_party/spirv-tools/source/spirv_validator_options.h

@@ -49,6 +49,7 @@ struct spv_validator_options_t {
         skip_block_layout(false),
         allow_localsizeid(false),
         allow_offset_texture_operand(false),
+        allow_vulkan_32_bit_bitwise(false),
         before_hlsl_legalization(false),
         use_friendly_names(true) {}
 
@@ -62,6 +63,7 @@ struct spv_validator_options_t {
   bool skip_block_layout;
   bool allow_localsizeid;
   bool allow_offset_texture_operand;
+  bool allow_vulkan_32_bit_bitwise;
   bool before_hlsl_legalization;
   bool use_friendly_names;
 };

third_party/spirv-tools/source/text.cpp

@@ -515,6 +515,124 @@ spv_result_t encodeInstructionStartingWithImmediate(
   return SPV_SUCCESS;
 }
 
+/// @brief Translate an instruction started by OpUnknown and the following
+/// operands to binary form
+///
+/// @param[in] grammar the grammar to use for compilation
+/// @param[in, out] context the dynamic compilation info
+/// @param[out] pInst returned binary Opcode
+///
+/// @return result code
+spv_result_t encodeInstructionStartingWithOpUnknown(
+    const spvtools::AssemblyGrammar& grammar,
+    spvtools::AssemblyContext* context, spv_instruction_t* pInst) {
+  spv_position_t nextPosition = {};
+
+  uint16_t opcode;
+  uint16_t wordCount;
+
+  // The '(' character.
+  if (context->advance())
+    return context->diagnostic() << "Expected '(', found end of stream.";
+  if ('(' != context->peek()) {
+    return context->diagnostic() << "'(' expected after OpUnknown but found '"
+                                 << context->peek() << "'.";
+  }
+  context->seekForward(1);
+
+  // The opcode enumerant.
+  if (context->advance())
+    return context->diagnostic()
+           << "Expected opcode enumerant, found end of stream.";
+  std::string opcodeString;
+  spv_result_t error = context->getWord(&opcodeString, &nextPosition);
+  if (error) return context->diagnostic(error) << "Internal Error";
+
+  if (!spvtools::utils::ParseNumber(opcodeString.c_str(), &opcode)) {
+    return context->diagnostic()
+           << "Invalid opcode enumerant: \"" << opcodeString << "\".";
+  }
+
+  context->setPosition(nextPosition);
+
+  // The ',' character.
+  if (context->advance())
+    return context->diagnostic() << "Expected ',', found end of stream.";
+  if (',' != context->peek()) {
+    return context->diagnostic()
+           << "',' expected after opcode enumerant but found '"
+           << context->peek() << "'.";
+  }
+  context->seekForward(1);
+
+  // The number of words.
+  if (context->advance())
+    return context->diagnostic()
+           << "Expected number of words, found end of stream.";
+  std::string wordCountString;
+  error = context->getWord(&wordCountString, &nextPosition);
+  if (error) return context->diagnostic(error) << "Internal Error";
+
+  if (!spvtools::utils::ParseNumber(wordCountString.c_str(), &wordCount)) {
+    return context->diagnostic()
+           << "Invalid number of words: \"" << wordCountString << "\".";
+  }
+
+  if (wordCount == 0) {
+    return context->diagnostic() << "Number of words (which includes the "
+                                    "opcode) must be greater than zero.";
+  }
+
+  context->setPosition(nextPosition);
+
+  // The ')' character.
+  if (context->advance())
+    return context->diagnostic() << "Expected ')', found end of stream.";
+  if (')' != context->peek()) {
+    return context->diagnostic()
+           << "')' expected after number of words but found '"
+           << context->peek() << "'.";
+  }
+  context->seekForward(1);
+
+  pInst->opcode = static_cast<spv::Op>(opcode);
+  context->binaryEncodeU32(spvOpcodeMake(wordCount, pInst->opcode), pInst);
+
+  wordCount--;  // Subtract the opcode from the number of words left to read.
+
+  while (wordCount-- > 0) {
+    if (context->advance() == SPV_END_OF_STREAM) {
+      return context->diagnostic() << "Expected " << wordCount + 1
+                                   << " more operands, found end of stream.";
+    }
+    if (context->isStartOfNewInst()) {
+      std::string invalid;
+      context->getWord(&invalid, &nextPosition);
+      return context->diagnostic()
+             << "Unexpected start of new instruction: \"" << invalid
+             << "\". Expected " << wordCount + 1 << " more operands";
+    }
+
+    std::string operandValue;
+    if ((error = context->getWord(&operandValue, &nextPosition)))
+      return context->diagnostic(error) << "Internal Error";
+
+    if (operandValue == "=")
+      return context->diagnostic() << "OpUnknown not allowed before =.";
+
+    // Needed to pass to spvTextEncodeOpcode(), but it shouldn't ever be
+    // expanded.
+    spv_operand_pattern_t dummyExpectedOperands;
+    error = spvTextEncodeOperand(
+        grammar, context, SPV_OPERAND_TYPE_OPTIONAL_CIV, operandValue.c_str(),
+        pInst, &dummyExpectedOperands);
+    if (error) return error;
+    context->setPosition(nextPosition);
+  }
+
+  return SPV_SUCCESS;
+}
+
 /// @brief Translate single Opcode and operands to binary form
 ///
 /// @param[in] grammar the grammar to use for compilation
@@ -574,6 +692,16 @@ spv_result_t spvTextEncodeOpcode(const spvtools::AssemblyGrammar& grammar,
     }
   }
 
+  if (opcodeName == "OpUnknown") {
+    if (!result_id.empty()) {
+      return context->diagnostic()
+             << "OpUnknown not allowed in assignment. Use an explicit result "
+                "id operand instead.";
+    }
+    context->setPosition(nextPosition);
+    return encodeInstructionStartingWithOpUnknown(grammar, context, pInst);
+  }
+
   // NOTE: The table contains Opcode names without the "Op" prefix.
   const char* pInstName = opcodeName.data() + 2;
 

third_party/spirv-tools/source/text_handler.cpp

@@ -118,6 +118,9 @@ spv_result_t getWord(spv_text text, spv_position position, std::string* word) {
           break;
         case ' ':
         case ';':
+        case ',':
+        case '(':
+        case ')':
         case '\t':
         case '\n':
         case '\r':

third_party/spirv-tools/source/util/small_vector.h

@@ -43,6 +43,7 @@ namespace utils {
 template <class T, size_t small_size>
 class SmallVector {
  public:
+  using value_type = T;
   using iterator = T*;
   using const_iterator = const T*;
 

third_party/spirv-tools/source/util/string_utils.h

@@ -17,13 +17,12 @@
 
 #include <assert.h>
 
+#include <cstdint>
 #include <cstring>
 #include <sstream>
 #include <string>
 #include <vector>
 
-#include "source/util/string_utils.h"
-
 namespace spvtools {
 namespace utils {
 
@@ -48,8 +47,9 @@ std::pair<std::string, std::string> SplitFlagArgs(const std::string& flag);
 
 // Encodes a string as a sequence of words, using the SPIR-V encoding, appending
 // to an existing vector.
-inline void AppendToVector(const std::string& input,
-                           std::vector<uint32_t>* result) {
+template <class VectorType = std::vector<uint32_t>>
+inline void AppendToVector(const std::string& input, VectorType* result) {
+  static_assert(std::is_same<uint32_t, typename VectorType::value_type>::value);
   uint32_t word = 0;
   size_t num_bytes = input.size();
   // SPIR-V strings are null-terminated.  The byte_index == num_bytes
@@ -70,8 +70,10 @@ inline void AppendToVector(const std::string& input,
 }
 
 // Encodes a string as a sequence of words, using the SPIR-V encoding.
-inline std::vector<uint32_t> MakeVector(const std::string& input) {
-  std::vector<uint32_t> result;
+template <class VectorType = std::vector<uint32_t>>
+inline VectorType MakeVector(const std::string& input) {
+  static_assert(std::is_same<uint32_t, typename VectorType::value_type>::value);
+  VectorType result;
   AppendToVector(input, &result);
   return result;
 }

third_party/spirv-tools/source/val/validate_bitwise.cpp

@@ -30,14 +30,14 @@ spv_result_t ValidateBaseType(ValidationState_t& _, const Instruction* inst,
 
   if (!_.IsIntScalarType(base_type) && !_.IsIntVectorType(base_type)) {
     return _.diag(SPV_ERROR_INVALID_DATA, inst)
-           << _.VkErrorID(4781)
            << "Expected int scalar or vector type for Base operand: "
            << spvOpcodeString(opcode);
   }
 
   // Vulkan has a restriction to 32 bit for base
   if (spvIsVulkanEnv(_.context()->target_env)) {
-    if (_.GetBitWidth(base_type) != 32) {
+    if (_.GetBitWidth(base_type) != 32 &&
+        !_.options()->allow_vulkan_32_bit_bitwise) {
       return _.diag(SPV_ERROR_INVALID_DATA, inst)
              << _.VkErrorID(4781)
              << "Expected 32-bit int type for Base operand: "

third_party/spirv-tools/source/val/validate_decorations.cpp

@@ -1916,7 +1916,7 @@ spv_result_t CheckComponentDecoration(ValidationState_t& vstate,
     if (!vstate.IsIntScalarOrVectorType(type_id) &&
         !vstate.IsFloatScalarOrVectorType(type_id)) {
       return vstate.diag(SPV_ERROR_INVALID_ID, &inst)
-             << vstate.VkErrorID(4924)
+             << vstate.VkErrorID(10583)
              << "Component decoration specified for type "
              << vstate.getIdName(type_id) << " that is not a scalar or vector";
     }
@@ -2083,6 +2083,204 @@ spv_result_t CheckDecorationsFromDecoration(ValidationState_t& vstate) {
   return SPV_SUCCESS;
 }
 
+bool AllowsLayout(ValidationState_t& vstate, const spv::StorageClass sc) {
+  switch (sc) {
+    case spv::StorageClass::StorageBuffer:
+    case spv::StorageClass::Uniform:
+    case spv::StorageClass::PhysicalStorageBuffer:
+    case spv::StorageClass::PushConstant:
+      // Always explicitly laid out.
+      return true;
+    case spv::StorageClass::UniformConstant:
+      return false;
+    case spv::StorageClass::Workgroup:
+      return vstate.HasCapability(
+          spv::Capability::WorkgroupMemoryExplicitLayoutKHR);
+    case spv::StorageClass::Function:
+    case spv::StorageClass::Private:
+      return vstate.version() <= SPV_SPIRV_VERSION_WORD(1, 4);
+    case spv::StorageClass::Input:
+    case spv::StorageClass::Output:
+      // Block is used generally and mesh shaders use Offset.
+      return true;
+    default:
+      // TODO: Some storage classes in ray tracing use explicit layout
+      // decorations, but it is not well documented which. For now treat other
+      // storage classes as allowed to be laid out. See Vulkan internal issue
+      // 4192.
+      return true;
+  }
+}
+
+bool UsesExplicitLayout(ValidationState_t& vstate, uint32_t type_id,
+                        std::unordered_map<uint32_t, bool>& cache) {
+  if (type_id == 0) {
+    return false;
+  }
+
+  if (cache.count(type_id)) {
+    return cache[type_id];
+  }
+
+  bool res = false;
+  const auto type_inst = vstate.FindDef(type_id);
+  if (type_inst->opcode() == spv::Op::OpTypeStruct ||
+      type_inst->opcode() == spv::Op::OpTypeArray ||
+      type_inst->opcode() == spv::Op::OpTypeRuntimeArray ||
+      type_inst->opcode() == spv::Op::OpTypePointer ||
+      type_inst->opcode() == spv::Op::OpTypeUntypedPointerKHR) {
+    const auto& id_decs = vstate.id_decorations();
+    const auto iter = id_decs.find(type_id);
+    if (iter != id_decs.end()) {
+      bool allowLayoutDecorations = false;
+      if (type_inst->opcode() == spv::Op::OpTypePointer) {
+        const auto sc = type_inst->GetOperandAs<spv::StorageClass>(1);
+        allowLayoutDecorations = AllowsLayout(vstate, sc);
+      }
+      if (!allowLayoutDecorations) {
+        res = std::any_of(
+            iter->second.begin(), iter->second.end(), [](const Decoration& d) {
+              return d.dec_type() == spv::Decoration::Block ||
+                     d.dec_type() == spv::Decoration::BufferBlock ||
+                     d.dec_type() == spv::Decoration::Offset ||
+                     d.dec_type() == spv::Decoration::ArrayStride ||
+                     d.dec_type() == spv::Decoration::MatrixStride;
+            });
+      }
+    }
+
+    if (!res) {
+      switch (type_inst->opcode()) {
+        case spv::Op::OpTypeStruct:
+          for (uint32_t i = 1; !res && i < type_inst->operands().size(); i++) {
+            res = UsesExplicitLayout(
+                vstate, type_inst->GetOperandAs<uint32_t>(i), cache);
+          }
+          break;
+        case spv::Op::OpTypeArray:
+        case spv::Op::OpTypeRuntimeArray:
+          res = UsesExplicitLayout(vstate, type_inst->GetOperandAs<uint32_t>(1),
+                                   cache);
+          break;
+        case spv::Op::OpTypePointer: {
+          const auto sc = type_inst->GetOperandAs<spv::StorageClass>(1);
+          if (!AllowsLayout(vstate, sc)) {
+            res = UsesExplicitLayout(
+                vstate, type_inst->GetOperandAs<uint32_t>(2), cache);
+          }
+        }
+        default:
+          break;
+      }
+    }
+  }
+
+  cache[type_id] = res;
+  return res;
+}
+
+spv_result_t CheckInvalidVulkanExplicitLayout(ValidationState_t& vstate) {
+  if (!spvIsVulkanEnv(vstate.context()->target_env)) {
+    return SPV_SUCCESS;
+  }
+
+  std::unordered_map<uint32_t, bool> cache;
+  for (const auto& inst : vstate.ordered_instructions()) {
+    const auto type_id = inst.type_id();
+    const auto type_inst = vstate.FindDef(type_id);
+    uint32_t fail_id = 0;
+    // Variables are the main place to check for improper decorations, but some
+    // untyped pointer instructions must also be checked since those types may
+    // never be instantiated by a variable. Unlike verifying a valid layout,
+    // physical storage buffer does not need checked here since it is always
+    // explicitly laid out.
+    switch (inst.opcode()) {
+      case spv::Op::OpVariable:
+      case spv::Op::OpUntypedVariableKHR: {
+        const auto sc = inst.GetOperandAs<spv::StorageClass>(2);
+        auto check_id = type_id;
+        if (inst.opcode() == spv::Op::OpUntypedVariableKHR) {
+          if (inst.operands().size() > 3) {
+            check_id = inst.GetOperandAs<uint32_t>(3);
+          }
+        }
+        if (!AllowsLayout(vstate, sc) &&
+            UsesExplicitLayout(vstate, check_id, cache)) {
+          fail_id = check_id;
+        }
+        break;
+      }
+      case spv::Op::OpUntypedAccessChainKHR:
+      case spv::Op::OpUntypedInBoundsAccessChainKHR:
+      case spv::Op::OpUntypedPtrAccessChainKHR:
+      case spv::Op::OpUntypedInBoundsPtrAccessChainKHR: {
+        // Check both the base type and return type. The return type may have an
+        // invalid array stride.
+        const auto sc = type_inst->GetOperandAs<spv::StorageClass>(1);
+        const auto base_type_id = inst.GetOperandAs<uint32_t>(2);
+        if (!AllowsLayout(vstate, sc)) {
+          if (UsesExplicitLayout(vstate, base_type_id, cache)) {
+            fail_id = base_type_id;
+          } else if (UsesExplicitLayout(vstate, type_id, cache)) {
+            fail_id = type_id;
+          }
+        }
+        break;
+      }
+      case spv::Op::OpUntypedArrayLengthKHR: {
+        // Check the data type.
+        const auto ptr_ty_id =
+            vstate.FindDef(inst.GetOperandAs<uint32_t>(3))->type_id();
+        const auto ptr_ty = vstate.FindDef(ptr_ty_id);
+        const auto sc = ptr_ty->GetOperandAs<spv::StorageClass>(1);
+        const auto base_type_id = inst.GetOperandAs<uint32_t>(2);
+        if (!AllowsLayout(vstate, sc) &&
+            UsesExplicitLayout(vstate, base_type_id, cache)) {
+          fail_id = base_type_id;
+        }
+        break;
+      }
+      case spv::Op::OpLoad: {
+        const auto ptr_id = inst.GetOperandAs<uint32_t>(2);
+        const auto ptr_type = vstate.FindDef(vstate.FindDef(ptr_id)->type_id());
+        if (ptr_type->opcode() == spv::Op::OpTypeUntypedPointerKHR) {
+          // For untyped pointers check the return type for an invalid layout.
+          const auto sc = ptr_type->GetOperandAs<spv::StorageClass>(1);
+          if (!AllowsLayout(vstate, sc) &&
+              UsesExplicitLayout(vstate, type_id, cache)) {
+            fail_id = type_id;
+          }
+        }
+        break;
+      }
+      case spv::Op::OpStore: {
+        const auto ptr_id = inst.GetOperandAs<uint32_t>(1);
+        const auto ptr_type = vstate.FindDef(vstate.FindDef(ptr_id)->type_id());
+        if (ptr_type->opcode() == spv::Op::OpTypeUntypedPointerKHR) {
+          // For untyped pointers, check the type of the data operand for an
+          // invalid layout.
+          const auto sc = ptr_type->GetOperandAs<spv::StorageClass>(1);
+          const auto data_type_id = vstate.GetOperandTypeId(&inst, 2);
+          if (!AllowsLayout(vstate, sc) &&
+              UsesExplicitLayout(vstate, data_type_id, cache)) {
+            fail_id = inst.GetOperandAs<uint32_t>(2);
+          }
+        }
+        break;
+      }
+      default:
+        break;
+    }
+    if (fail_id != 0) {
+      return vstate.diag(SPV_ERROR_INVALID_ID, &inst)
+             << "Invalid explicit layout decorations on type for operand "
+             << vstate.getIdName(fail_id);
+    }
+  }
+
+  return SPV_SUCCESS;
+}
+
 }  // namespace
 
 spv_result_t ValidateDecorations(ValidationState_t& vstate) {
@@ -2094,6 +2292,7 @@ spv_result_t ValidateDecorations(ValidationState_t& vstate) {
   if (auto error = CheckVulkanMemoryModelDeprecatedDecorations(vstate))
     return error;
   if (auto error = CheckDecorationsFromDecoration(vstate)) return error;
+  if (auto error = CheckInvalidVulkanExplicitLayout(vstate)) return error;
   return SPV_SUCCESS;
 }
 

third_party/spirv-tools/source/val/validate_function.cpp

@@ -258,7 +258,8 @@ spv_result_t ValidateFunctionCall(ValidationState_t& _,
               _.HasCapability(spv::Capability::VariablePointers) &&
               sc == spv::StorageClass::Workgroup;
           const bool uc_ptr = sc == spv::StorageClass::UniformConstant;
-          if (!ssbo_vptr && !wg_vptr && !uc_ptr) {
+          if (!_.options()->before_hlsl_legalization && !ssbo_vptr &&
+              !wg_vptr && !uc_ptr) {
             return _.diag(SPV_ERROR_INVALID_ID, inst)
                    << "Pointer operand " << _.getIdName(argument_id)
                    << " must be a memory object declaration";

third_party/spirv-tools/source/val/validate_interfaces.cpp

@@ -48,6 +48,29 @@ bool is_interface_variable(const Instruction* inst, bool is_spv_1_4) {
   }
 }
 
+// Special validation for varibles that are between shader stages
+spv_result_t ValidateInputOutputInterfaceVariables(ValidationState_t& _,
+                                                   const Instruction* var) {
+  auto var_pointer = _.FindDef(var->GetOperandAs<uint32_t>(0));
+  uint32_t pointer_id = var_pointer->GetOperandAs<uint32_t>(2);
+
+  const auto isPhysicalStorageBuffer = [](const Instruction* insn) {
+    return insn->opcode() == spv::Op::OpTypePointer &&
+           insn->GetOperandAs<spv::StorageClass>(1) ==
+               spv::StorageClass::PhysicalStorageBuffer;
+  };
+
+  if (_.ContainsType(pointer_id, isPhysicalStorageBuffer)) {
+    return _.diag(SPV_ERROR_INVALID_ID, var)
+           << _.VkErrorID(9557) << "Input/Output interface variable id <"
+           << var->id()
+           << "> contains a PhysicalStorageBuffer pointer, which is not "
+              "allowed. If you want to interface shader stages with a "
+              "PhysicalStorageBuffer, cast to a uint64 or uvec2 instead.";
+  }
+  return SPV_SUCCESS;
+}
+
 // Checks that \c var is listed as an interface in all the entry points that use
 // it.
 spv_result_t check_interface_variable(ValidationState_t& _,
@@ -107,6 +130,12 @@ spv_result_t check_interface_variable(ValidationState_t& _,
     }
   }
 
+  if (var->GetOperandAs<spv::StorageClass>(2) == spv::StorageClass::Input ||
+      var->GetOperandAs<spv::StorageClass>(2) == spv::StorageClass::Output) {
+    if (auto error = ValidateInputOutputInterfaceVariables(_, var))
+      return error;
+  }
+
   return SPV_SUCCESS;
 }
 
@@ -135,17 +164,22 @@ spv_result_t NumConsumedLocations(ValidationState_t& _, const Instruction* type,
       }
       break;
     case spv::Op::OpTypeMatrix:
-      // Matrices consume locations equal to the underlying vector type for
-      // each column.
-      NumConsumedLocations(_, _.FindDef(type->GetOperandAs<uint32_t>(1)),
-                           num_locations);
+      // Matrices consume locations equivalent to arrays of 4-component vectors.
+      if (_.ContainsSizedIntOrFloatType(type->id(), spv::Op::OpTypeInt, 64) ||
+          _.ContainsSizedIntOrFloatType(type->id(), spv::Op::OpTypeFloat, 64)) {
+        *num_locations = 2;
+      } else {
+        *num_locations = 1;
+      }
       *num_locations *= type->GetOperandAs<uint32_t>(2);
       break;
     case spv::Op::OpTypeArray: {
       // Arrays consume locations equal to the underlying type times the number
       // of elements in the vector.
-      NumConsumedLocations(_, _.FindDef(type->GetOperandAs<uint32_t>(1)),
-                           num_locations);
+      if (auto error = NumConsumedLocations(
+              _, _.FindDef(type->GetOperandAs<uint32_t>(1)), num_locations)) {
+        return error;
+      }
       bool is_int = false;
       bool is_const = false;
       uint32_t value = 0;
@@ -215,10 +249,31 @@ uint32_t NumConsumedComponents(ValidationState_t& _, const Instruction* type) {
           NumConsumedComponents(_, _.FindDef(type->GetOperandAs<uint32_t>(1)));
       num_components *= type->GetOperandAs<uint32_t>(2);
       break;
-    case spv::Op::OpTypeArray:
-      // Skip the array.
-      return NumConsumedComponents(_,
-                                   _.FindDef(type->GetOperandAs<uint32_t>(1)));
+    case spv::Op::OpTypeMatrix:
+      // Matrices consume all components of the location.
+      // Round up to next multiple of 4.
+      num_components =
+          NumConsumedComponents(_, _.FindDef(type->GetOperandAs<uint32_t>(1)));
+      num_components *= type->GetOperandAs<uint32_t>(2);
+      num_components = ((num_components + 3) / 4) * 4;
+      break;
+    case spv::Op::OpTypeArray: {
+      // Arrays consume all components of the location.
+      // Round up to next multiple of 4.
+      num_components =
+          NumConsumedComponents(_, _.FindDef(type->GetOperandAs<uint32_t>(1)));
+
+      bool is_int = false;
+      bool is_const = false;
+      uint32_t value = 0;
+      // Attempt to evaluate the number of array elements.
+      std::tie(is_int, is_const, value) =
+          _.EvalInt32IfConst(type->GetOperandAs<uint32_t>(2));
+      if (is_int && is_const) num_components *= value;
+
+      num_components = ((num_components + 3) / 4) * 4;
+      return num_components;
+    }
     case spv::Op::OpTypePointer:
       if (_.addressing_model() ==
               spv::AddressingModel::PhysicalStorageBuffer64 &&
@@ -301,9 +356,10 @@ spv_result_t GetLocationsForVariable(
     }
   }
 
-  // Vulkan 14.1.3: Tessellation control and mesh per-vertex outputs and
-  // tessellation control, evaluation and geometry per-vertex inputs have a
-  // layer of arraying that is not included in interface matching.
+  // Vulkan 15.1.3 (Interface Matching): Tessellation control and mesh
+  // per-vertex outputs and tessellation control, evaluation and geometry
+  // per-vertex inputs have a layer of arraying that is not included in
+  // interface matching.
   bool is_arrayed = false;
   switch (entry_point->GetOperandAs<spv::ExecutionModel>(0)) {
     case spv::ExecutionModel::TessellationControl:
@@ -357,51 +413,33 @@ spv_result_t GetLocationsForVariable(
 
   const std::string storage_class = is_output ? "output" : "input";
   if (has_location) {
-    auto sub_type = type;
-    bool is_int = false;
-    bool is_const = false;
-    uint32_t array_size = 1;
-    // If the variable is still arrayed, mark the locations/components per
-    // index.
-    if (type->opcode() == spv::Op::OpTypeArray) {
-      // Determine the array size if possible and get the element type.
-      std::tie(is_int, is_const, array_size) =
-          _.EvalInt32IfConst(type->GetOperandAs<uint32_t>(2));
-      if (!is_int || !is_const) array_size = 1;
-      auto sub_type_id = type->GetOperandAs<uint32_t>(1);
-      sub_type = _.FindDef(sub_type_id);
+    uint32_t num_locations = 0;
+    if (auto error = NumConsumedLocations(_, type, &num_locations))
+      return error;
+    uint32_t num_components = NumConsumedComponents(_, type);
+
+    uint32_t start = location * 4;
+    uint32_t end = (location + num_locations) * 4;
+    if (num_components % 4 != 0) {
+      start += component;
+      end = start + num_components;
     }
 
-    uint32_t num_locations = 0;
-    if (auto error = NumConsumedLocations(_, sub_type, &num_locations))
-      return error;
-    uint32_t num_components = NumConsumedComponents(_, sub_type);
+    if (kMaxLocations <= start) {
+      // Too many locations, give up.
+      return SPV_SUCCESS;
+    }
 
-    for (uint32_t array_idx = 0; array_idx < array_size; ++array_idx) {
-      uint32_t array_location = location + (num_locations * array_idx);
-      uint32_t start = array_location * 4;
-      if (kMaxLocations <= start) {
-        // Too many locations, give up.
-        break;
-      }
+    auto locs = locations;
+    if (has_index && index == 1) locs = output_index1_locations;
 
-      uint32_t end = (array_location + num_locations) * 4;
-      if (num_components != 0) {
-        start += component;
-        end = array_location * 4 + component + num_components;
-      }
-
-      auto locs = locations;
-      if (has_index && index == 1) locs = output_index1_locations;
-
-      for (uint32_t i = start; i < end; ++i) {
-        if (!locs->insert(i).second) {
-          return _.diag(SPV_ERROR_INVALID_DATA, entry_point)
-                 << (is_output ? _.VkErrorID(8722) : _.VkErrorID(8721))
-                 << "Entry-point has conflicting " << storage_class
-                 << " location assignment at location " << i / 4
-                 << ", component " << i % 4;
-        }
+    for (uint32_t i = start; i < end; ++i) {
+      if (!locs->insert(i).second) {
+        return _.diag(SPV_ERROR_INVALID_DATA, entry_point)
+               << (is_output ? _.VkErrorID(8722) : _.VkErrorID(8721))
+               << "Entry-point has conflicting " << storage_class
+               << " location assignment at location " << i / 4 << ", component "
+               << i % 4;
       }
     }
   } else {
@@ -460,38 +498,19 @@ spv_result_t GetLocationsForVariable(
         continue;
       }
 
-      if (member->opcode() == spv::Op::OpTypeArray && num_components >= 1 &&
-          num_components < 4) {
-        // When an array has an element that takes less than a location in
-        // size, calculate the used locations in a strided manner.
-        for (uint32_t l = location; l < num_locations + location; ++l) {
-          for (uint32_t c = component; c < component + num_components; ++c) {
-            uint32_t check = 4 * l + c;
-            if (!locations->insert(check).second) {
-              return _.diag(SPV_ERROR_INVALID_DATA, entry_point)
-                     << (is_output ? _.VkErrorID(8722) : _.VkErrorID(8721))
-                     << "Entry-point has conflicting " << storage_class
-                     << " location assignment at location " << l
-                     << ", component " << c;
-            }
-          }
-        }
-      } else {
-        // TODO: There is a hole here is the member is an array of 3- or
-        // 4-element vectors of 64-bit types.
-        uint32_t end = (location + num_locations) * 4;
-        if (num_components != 0) {
-          start += component;
-          end = location * 4 + component + num_components;
-        }
-        for (uint32_t l = start; l < end; ++l) {
-          if (!locations->insert(l).second) {
-            return _.diag(SPV_ERROR_INVALID_DATA, entry_point)
-                   << (is_output ? _.VkErrorID(8722) : _.VkErrorID(8721))
-                   << "Entry-point has conflicting " << storage_class
-                   << " location assignment at location " << l / 4
-                   << ", component " << l % 4;
-          }
+      uint32_t end = (location + num_locations) * 4;
+      if (num_components % 4 != 0) {
+        start += component;
+        end = location * 4 + component + num_components;
+      }
+
+      for (uint32_t l = start; l < end; ++l) {
+        if (!locations->insert(l).second) {
+          return _.diag(SPV_ERROR_INVALID_DATA, entry_point)
+                 << (is_output ? _.VkErrorID(8722) : _.VkErrorID(8721))
+                 << "Entry-point has conflicting " << storage_class
+                 << " location assignment at location " << l / 4
+                 << ", component " << l % 4;
         }
       }
     }

third_party/spirv-tools/source/val/validate_memory.cpp

@@ -380,6 +380,16 @@ spv_result_t CheckMemoryAccess(ValidationState_t& _, const Instruction* inst,
              << _.VkErrorID(4708)
              << "Memory accesses with PhysicalStorageBuffer must use Aligned.";
     }
+  } else {
+    // even if there are other masks, the Aligned operand will be next
+    const uint32_t aligned_value = inst->GetOperandAs<uint32_t>(index + 1);
+    const bool is_power_of_two =
+        aligned_value && !(aligned_value & (aligned_value - 1));
+    if (!is_power_of_two) {
+      return _.diag(SPV_ERROR_INVALID_ID, inst)
+             << "Memory accesses Aligned operand value " << aligned_value
+             << " is not a power of two.";
+    }
   }
 
   return SPV_SUCCESS;
@@ -2781,17 +2791,42 @@ spv_result_t ValidatePtrComparison(ValidationState_t& _,
 
   const auto op1 = _.FindDef(inst->GetOperandAs<uint32_t>(2u));
   const auto op2 = _.FindDef(inst->GetOperandAs<uint32_t>(3u));
-  if (!op1 || !op2 || op1->type_id() != op2->type_id()) {
-    return _.diag(SPV_ERROR_INVALID_ID, inst)
-           << "The types of Operand 1 and Operand 2 must match";
-  }
   const auto op1_type = _.FindDef(op1->type_id());
+  const auto op2_type = _.FindDef(op2->type_id());
   if (!op1_type || (op1_type->opcode() != spv::Op::OpTypePointer &&
                     op1_type->opcode() != spv::Op::OpTypeUntypedPointerKHR)) {
     return _.diag(SPV_ERROR_INVALID_ID, inst)
            << "Operand type must be a pointer";
   }
 
+  if (!op2_type || (op2_type->opcode() != spv::Op::OpTypePointer &&
+                    op2_type->opcode() != spv::Op::OpTypeUntypedPointerKHR)) {
+    return _.diag(SPV_ERROR_INVALID_ID, inst)
+           << "Operand type must be a pointer";
+  }
+
+  if (inst->opcode() == spv::Op::OpPtrDiff) {
+    if (op1->type_id() != op2->type_id()) {
+      return _.diag(SPV_ERROR_INVALID_ID, inst)
+             << "The types of Operand 1 and Operand 2 must match";
+    }
+  } else {
+    const auto either_untyped =
+        op1_type->opcode() == spv::Op::OpTypeUntypedPointerKHR ||
+        op2_type->opcode() == spv::Op::OpTypeUntypedPointerKHR;
+    if (either_untyped) {
+      const auto sc1 = op1_type->GetOperandAs<spv::StorageClass>(1);
+      const auto sc2 = op2_type->GetOperandAs<spv::StorageClass>(1);
+      if (sc1 != sc2) {
+        return _.diag(SPV_ERROR_INVALID_ID, inst)
+               << "Pointer storage classes must match";
+      }
+    } else if (op1->type_id() != op2->type_id()) {
+      return _.diag(SPV_ERROR_INVALID_ID, inst)
+             << "The types of Operand 1 and Operand 2 must match";
+    }
+  }
+
   spv::StorageClass sc = op1_type->GetOperandAs<spv::StorageClass>(1u);
   if (_.addressing_model() == spv::AddressingModel::Logical) {
     if (sc != spv::StorageClass::Workgroup &&

third_party/spirv-tools/source/val/validation_state.cpp

@@ -2415,8 +2415,6 @@ std::string ValidationState_t::VkErrorID(uint32_t id,
       return VUID_WRAP(VUID-StandaloneSpirv-Component-04922);
     case 4923:
       return VUID_WRAP(VUID-StandaloneSpirv-Component-04923);
-    case 4924:
-      return VUID_WRAP(VUID-StandaloneSpirv-Component-04924);
     case 6201:
       return VUID_WRAP(VUID-StandaloneSpirv-Flat-06201);
     case 6202:
@@ -2525,6 +2523,8 @@ std::string ValidationState_t::VkErrorID(uint32_t id,
       return VUID_WRAP(VUID-StandaloneSpirv-OpEntryPoint-08722);
     case 8973:
       return VUID_WRAP(VUID-StandaloneSpirv-Pointer-08973);
+    case 9557:
+    return VUID_WRAP(VUID-StandaloneSpirv-Input-09557);
     case 9638:
       return VUID_WRAP(VUID-StandaloneSpirv-OpTypeImage-09638);
     case 9658:
@@ -2534,6 +2534,8 @@ std::string ValidationState_t::VkErrorID(uint32_t id,
     case 10213:
       // This use to be a standalone, but maintenance8 will set allow_offset_texture_operand now
       return VUID_WRAP(VUID-RuntimeSpirv-Offset-10213);
+    case 10583:
+      return VUID_WRAP(VUID-StandaloneSpirv-Component-10583);
     default:
       return "";  // unknown id
   }

third_party/spirv-tools/test/CMakeLists.txt

@@ -116,6 +116,7 @@ set(TEST_SOURCES
   libspirv_macros_test.cpp
   named_id_test.cpp
   name_mapper_test.cpp
+  op_unknown_test.cpp
   opcode_make_test.cpp
   opcode_require_capabilities_test.cpp
   opcode_split_test.cpp

third_party/spirv-tools/test/diff/diff_files/diff_test_files_autogen.cmake

@@ -30,6 +30,8 @@ list(APPEND DIFF_TEST_FILES
 "diff_files/different_decorations_vertex_autogen.cpp"
 "diff_files/different_function_parameter_count_autogen.cpp"
 "diff_files/extra_if_block_autogen.cpp"
+"diff_files/function_group_by_full_type_autogen.cpp"
+"diff_files/function_group_by_mapped_id_autogen.cpp"
 "diff_files/index_signedness_autogen.cpp"
 "diff_files/int_vs_uint_constants_autogen.cpp"
 "diff_files/large_functions_large_diffs_autogen.cpp"
@@ -44,5 +46,6 @@ list(APPEND DIFF_TEST_FILES
 "diff_files/spec_constant_composite_autogen.cpp"
 "diff_files/spec_constant_op_autogen.cpp"
 "diff_files/spec_constant_specid_autogen.cpp"
+"diff_files/string_in_ext_inst_autogen.cpp"
 "diff_files/unrelated_shaders_autogen.cpp"
 )

third_party/spirv-tools/test/diff/diff_files/function_group_by_full_type_autogen.cpp

@@ -0,0 +1,194 @@
+// GENERATED FILE - DO NOT EDIT.
+// Generated by generate_tests.py
+//
+// Copyright (c) 2022 Google LLC.
+//
+// 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 "../diff_test_utils.h"
+#include "gtest/gtest.h"
+
+namespace spvtools {
+namespace diff {
+namespace {
+
+// Use functions' `OpFunctionType` to group before grouping by return type only.
+constexpr char kSrc[] = R"(                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %bool = OpTypeBool
+          %int = OpTypeInt 32 1
+        %int_0 = OpConstantNull %int
+
+       %fn_int = OpTypeFunction %bool %int
+   %fn_int_int = OpTypeFunction %bool %int %int
+
+            %f = OpFunction %bool None %fn_int
+          %fp1 = OpFunctionParameter %int
+   %f_prologue = OpLabel
+           %fr = OpIEqual %bool %fp1 %int_0
+                 OpReturnValue %fr
+                 OpFunctionEnd
+
+            %g = OpFunction %bool None %fn_int_int
+          %gp1 = OpFunctionParameter %int
+          %gp2 = OpFunctionParameter %int
+   %g_prologue = OpLabel
+           %gr = OpIEqual %bool %gp1 %int_0
+                 OpReturnValue %gr
+                 OpFunctionEnd)";
+constexpr char kDst[] =
+    R"(;; Use functions' `OpFunctionType` to group before grouping by return type only.
+                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %bool = OpTypeBool
+          %int = OpTypeInt 32 1
+        %int_0 = OpConstantNull %int
+
+       %fn_int = OpTypeFunction %bool %int
+   %fn_int_int = OpTypeFunction %bool %int %int
+
+            %g = OpFunction %bool None %fn_int_int
+          %gp1 = OpFunctionParameter %int
+          %gp2 = OpFunctionParameter %int
+   %g_prologue = OpLabel
+           %gr = OpIEqual %bool %gp1 %int_0
+                 OpReturnValue %gr
+                 OpFunctionEnd
+
+            %f = OpFunction %bool None %fn_int
+          %fp1 = OpFunctionParameter %int
+   %f_prologue = OpLabel
+           %fr = OpIEqual %bool %fp1 %int_0
+                 OpReturnValue %fr
+                 OpFunctionEnd
+)";
+
+TEST(DiffTest, FunctionGroupByFullType) {
+  constexpr char kDiff[] = R"( ; SPIR-V
+ ; Version: 1.6
+ ; Generator: Khronos SPIR-V Tools Assembler; 0
+ ; Bound: 15
+ ; Schema: 0
+ OpCapability Shader
+ OpCapability Linkage
+ OpMemoryModel Logical GLSL450
+ %1 = OpTypeBool
+ %2 = OpTypeInt 32 1
+ %3 = OpConstantNull %2
+ %4 = OpTypeFunction %1 %2
+ %5 = OpTypeFunction %1 %2 %2
+ %6 = OpFunction %1 None %4
+ %7 = OpFunctionParameter %2
+ %8 = OpLabel
+ %9 = OpIEqual %1 %7 %3
+ OpReturnValue %9
+ OpFunctionEnd
+ %10 = OpFunction %1 None %5
+ %11 = OpFunctionParameter %2
+ %12 = OpFunctionParameter %2
+ %13 = OpLabel
+ %14 = OpIEqual %1 %11 %3
+ OpReturnValue %14
+ OpFunctionEnd
+)";
+  Options options;
+  DoStringDiffTest(kSrc, kDst, kDiff, options);
+}
+
+TEST(DiffTest, FunctionGroupByFullTypeNoDebug) {
+  constexpr char kSrcNoDebug[] = R"(                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %bool = OpTypeBool
+          %int = OpTypeInt 32 1
+        %int_0 = OpConstantNull %int
+
+       %fn_int = OpTypeFunction %bool %int
+   %fn_int_int = OpTypeFunction %bool %int %int
+
+            %f = OpFunction %bool None %fn_int
+          %fp1 = OpFunctionParameter %int
+   %f_prologue = OpLabel
+           %fr = OpIEqual %bool %fp1 %int_0
+                 OpReturnValue %fr
+                 OpFunctionEnd
+
+            %g = OpFunction %bool None %fn_int_int
+          %gp1 = OpFunctionParameter %int
+          %gp2 = OpFunctionParameter %int
+   %g_prologue = OpLabel
+           %gr = OpIEqual %bool %gp1 %int_0
+                 OpReturnValue %gr
+                 OpFunctionEnd
+)";
+  constexpr char kDstNoDebug[] = R"(                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %bool = OpTypeBool
+          %int = OpTypeInt 32 1
+        %int_0 = OpConstantNull %int
+
+       %fn_int = OpTypeFunction %bool %int
+   %fn_int_int = OpTypeFunction %bool %int %int
+
+            %g = OpFunction %bool None %fn_int_int
+          %gp1 = OpFunctionParameter %int
+          %gp2 = OpFunctionParameter %int
+   %g_prologue = OpLabel
+           %gr = OpIEqual %bool %gp1 %int_0
+                 OpReturnValue %gr
+                 OpFunctionEnd
+
+            %f = OpFunction %bool None %fn_int
+          %fp1 = OpFunctionParameter %int
+   %f_prologue = OpLabel
+           %fr = OpIEqual %bool %fp1 %int_0
+                 OpReturnValue %fr
+                 OpFunctionEnd
+)";
+  constexpr char kDiff[] = R"( ; SPIR-V
+ ; Version: 1.6
+ ; Generator: Khronos SPIR-V Tools Assembler; 0
+ ; Bound: 15
+ ; Schema: 0
+ OpCapability Shader
+ OpCapability Linkage
+ OpMemoryModel Logical GLSL450
+ %1 = OpTypeBool
+ %2 = OpTypeInt 32 1
+ %3 = OpConstantNull %2
+ %4 = OpTypeFunction %1 %2
+ %5 = OpTypeFunction %1 %2 %2
+ %6 = OpFunction %1 None %4
+ %7 = OpFunctionParameter %2
+ %8 = OpLabel
+ %9 = OpIEqual %1 %7 %3
+ OpReturnValue %9
+ OpFunctionEnd
+ %10 = OpFunction %1 None %5
+ %11 = OpFunctionParameter %2
+ %12 = OpFunctionParameter %2
+ %13 = OpLabel
+ %14 = OpIEqual %1 %11 %3
+ OpReturnValue %14
+ OpFunctionEnd
+)";
+  Options options;
+  DoStringDiffTest(kSrcNoDebug, kDstNoDebug, kDiff, options);
+}
+
+}  // namespace
+}  // namespace diff
+}  // namespace spvtools

third_party/spirv-tools/test/diff/diff_files/function_group_by_full_type_dst.spvasm

@@ -0,0 +1,25 @@
+;; Use functions' `OpFunctionType` to group before grouping by return type only.
+                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %bool = OpTypeBool
+          %int = OpTypeInt 32 1
+        %int_0 = OpConstantNull %int
+
+       %fn_int = OpTypeFunction %bool %int
+   %fn_int_int = OpTypeFunction %bool %int %int
+
+            %g = OpFunction %bool None %fn_int_int
+          %gp1 = OpFunctionParameter %int
+          %gp2 = OpFunctionParameter %int
+   %g_prologue = OpLabel
+           %gr = OpIEqual %bool %gp1 %int_0
+                 OpReturnValue %gr
+                 OpFunctionEnd
+
+            %f = OpFunction %bool None %fn_int
+          %fp1 = OpFunctionParameter %int
+   %f_prologue = OpLabel
+           %fr = OpIEqual %bool %fp1 %int_0
+                 OpReturnValue %fr
+                 OpFunctionEnd

third_party/spirv-tools/test/diff/diff_files/function_group_by_full_type_src.spvasm

@@ -0,0 +1,25 @@
+;; Use functions' `OpFunctionType` to group before grouping by return type only.
+                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %bool = OpTypeBool
+          %int = OpTypeInt 32 1
+        %int_0 = OpConstantNull %int
+
+       %fn_int = OpTypeFunction %bool %int
+   %fn_int_int = OpTypeFunction %bool %int %int
+
+            %f = OpFunction %bool None %fn_int
+          %fp1 = OpFunctionParameter %int
+   %f_prologue = OpLabel
+           %fr = OpIEqual %bool %fp1 %int_0
+                 OpReturnValue %fr
+                 OpFunctionEnd
+
+            %g = OpFunction %bool None %fn_int_int
+          %gp1 = OpFunctionParameter %int
+          %gp2 = OpFunctionParameter %int
+   %g_prologue = OpLabel
+           %gr = OpIEqual %bool %gp1 %int_0
+                 OpReturnValue %gr
+                 OpFunctionEnd

third_party/spirv-tools/test/diff/diff_files/function_group_by_mapped_id_autogen.cpp

@@ -0,0 +1,181 @@
+// GENERATED FILE - DO NOT EDIT.
+// Generated by generate_tests.py
+//
+// Copyright (c) 2022 Google LLC.
+//
+// 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 "../diff_test_utils.h"
+#include "gtest/gtest.h"
+
+namespace spvtools {
+namespace diff {
+namespace {
+
+// Don't forget to map between source and destination ids when grouping
+// functions by return type.
+constexpr char kSrc[] = R"(                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %void = OpTypeVoid               
+            %1 = OpTypeInt 32 0
+            %2 = OpTypeInt 32 1
+
+      %fn_uint = OpTypeFunction %1
+       %fn_int = OpTypeFunction %2
+
+      %uint_42 = OpConstant %1 42
+     %int_1729 = OpConstant %2 1729
+
+            %f = OpFunction %1 None %fn_uint
+   %f_prologue = OpLabel
+                 OpReturnValue %uint_42
+                 OpFunctionEnd
+
+            %g = OpFunction %2 None %fn_int
+   %g_prologue = OpLabel
+                 OpReturnValue %int_1729
+                 OpFunctionEnd)";
+constexpr char kDst[] =
+    R"(;; Don't forget to map between source and destination ids when grouping functions by return type.
+                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %void = OpTypeVoid               
+            %2 = OpTypeInt 32 0
+            %1 = OpTypeInt 32 1
+
+      %fn_uint = OpTypeFunction %2
+       %fn_int = OpTypeFunction %1
+
+      %uint_42 = OpConstant %2 42
+     %int_1729 = OpConstant %1 1729
+
+            %f = OpFunction %2 None %fn_uint
+   %f_prologue = OpLabel
+                 OpReturnValue %uint_42
+                 OpFunctionEnd
+
+            %g = OpFunction %1 None %fn_int
+   %g_prologue = OpLabel
+                 OpReturnValue %int_1729
+                 OpFunctionEnd
+)";
+
+TEST(DiffTest, FunctionGroupByMappedId) {
+  constexpr char kDiff[] = R"( ; SPIR-V
+ ; Version: 1.6
+ ; Generator: Khronos SPIR-V Tools Assembler; 0
+ ; Bound: 12
+ ; Schema: 0
+ OpCapability Shader
+ OpCapability Linkage
+ OpMemoryModel Logical GLSL450
+ %3 = OpTypeVoid
+ %1 = OpTypeInt 32 0
+ %2 = OpTypeInt 32 1
+ %4 = OpTypeFunction %1
+ %5 = OpTypeFunction %2
+ %6 = OpConstant %1 42
+ %7 = OpConstant %2 1729
+ %8 = OpFunction %1 None %4
+ %9 = OpLabel
+ OpReturnValue %6
+ OpFunctionEnd
+ %10 = OpFunction %2 None %5
+ %11 = OpLabel
+ OpReturnValue %7
+ OpFunctionEnd
+)";
+  Options options;
+  DoStringDiffTest(kSrc, kDst, kDiff, options);
+}
+
+TEST(DiffTest, FunctionGroupByMappedIdNoDebug) {
+  constexpr char kSrcNoDebug[] = R"(                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %void = OpTypeVoid               
+            %1 = OpTypeInt 32 0
+            %2 = OpTypeInt 32 1
+
+      %fn_uint = OpTypeFunction %1
+       %fn_int = OpTypeFunction %2
+
+      %uint_42 = OpConstant %1 42
+     %int_1729 = OpConstant %2 1729
+
+            %f = OpFunction %1 None %fn_uint
+   %f_prologue = OpLabel
+                 OpReturnValue %uint_42
+                 OpFunctionEnd
+
+            %g = OpFunction %2 None %fn_int
+   %g_prologue = OpLabel
+                 OpReturnValue %int_1729
+                 OpFunctionEnd
+)";
+  constexpr char kDstNoDebug[] = R"(                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %void = OpTypeVoid               
+            %2 = OpTypeInt 32 0
+            %1 = OpTypeInt 32 1
+
+      %fn_uint = OpTypeFunction %2
+       %fn_int = OpTypeFunction %1
+
+      %uint_42 = OpConstant %2 42
+     %int_1729 = OpConstant %1 1729
+
+            %f = OpFunction %2 None %fn_uint
+   %f_prologue = OpLabel
+                 OpReturnValue %uint_42
+                 OpFunctionEnd
+
+            %g = OpFunction %1 None %fn_int
+   %g_prologue = OpLabel
+                 OpReturnValue %int_1729
+                 OpFunctionEnd
+)";
+  constexpr char kDiff[] = R"( ; SPIR-V
+ ; Version: 1.6
+ ; Generator: Khronos SPIR-V Tools Assembler; 0
+ ; Bound: 12
+ ; Schema: 0
+ OpCapability Shader
+ OpCapability Linkage
+ OpMemoryModel Logical GLSL450
+ %3 = OpTypeVoid
+ %1 = OpTypeInt 32 0
+ %2 = OpTypeInt 32 1
+ %4 = OpTypeFunction %1
+ %5 = OpTypeFunction %2
+ %6 = OpConstant %1 42
+ %7 = OpConstant %2 1729
+ %8 = OpFunction %1 None %4
+ %9 = OpLabel
+ OpReturnValue %6
+ OpFunctionEnd
+ %10 = OpFunction %2 None %5
+ %11 = OpLabel
+ OpReturnValue %7
+ OpFunctionEnd
+)";
+  Options options;
+  DoStringDiffTest(kSrcNoDebug, kDstNoDebug, kDiff, options);
+}
+
+}  // namespace
+}  // namespace diff
+}  // namespace spvtools

third_party/spirv-tools/test/diff/diff_files/function_group_by_mapped_id_dst.spvasm

@@ -0,0 +1,23 @@
+;; Don't forget to map between source and destination ids when grouping functions by return type.
+                 OpCapability Shader
+                 OpCapability Linkage
+                 OpMemoryModel Logical GLSL450
+         %void = OpTypeVoid               
+            %2 = OpTypeInt 32 0
+            %1 = OpTypeInt 32 1
+
+      %fn_uint = OpTypeFunction %2
+       %fn_int = OpTypeFunction %1
+
+      %uint_42 = OpConstant %2 42
+     %int_1729 = OpConstant %1 1729
+
+            %f = OpFunction %2 None %fn_uint
+   %f_prologue = OpLabel
+                 OpReturnValue %uint_42
+                 OpFunctionEnd
+
+            %g = OpFunction %1 None %fn_int
+   %g_prologue = OpLabel
+                 OpReturnValue %int_1729
+                 OpFunctionEnd