Acceleration Structure Asset Conversion

include/nbl/asset/IAccelerationStructure.h

@@ -60,7 +60,7 @@ class IBottomLevelAccelerationStructure : public AccelerationStructure
 			// Provided by VK_NV_displacement_micromap
 			ALLOW_DISPLACEMENT_MICROMAP_UPDATE_BIT = 0x1u<<9u,
 			// Provided by VK_KHR_ray_tracing_position_fetch
-			ALLOW_DATA_ACCESS_KHR = 0x1u<<11u,
+			ALLOW_DATA_ACCESS = 0x1u<<11u,
 		};
 		static inline bool validBuildFlags(const core::bitflag<BUILD_FLAGS> flags)
 		{
@@ -154,8 +154,6 @@ class ITopLevelAccelerationStructure : public AccelerationStructure
 			PREFER_FAST_TRACE_BIT = 0x1u<<2u,
 			PREFER_FAST_BUILD_BIT = 0x1u<<3u,
 			LOW_MEMORY_BIT = 0x1u<<4u,
-			// Synthetic flag we use to indicate `VkAccelerationStructureGeometryInstancesDataKHR::arrayOfPointers`
-			INSTANCE_DATA_IS_POINTERS_TYPE_ENCODED_LSB = 0x1u<<5u, // this flag really shouldn't be settable outside of `video::IGPU`
 			// Provided by VK_NV_ray_tracing_motion_blur, but we always override and deduce from creation flag because of
 			// https://registry.khronos.org/vulkan/specs/1.3-extensions/html/vkspec.html#VUID-VkAccelerationStructureBuildGeometryInfoKHR-dstAccelerationStructure-04927
 			//MOTION_BIT = 0x1u<<5u,
@@ -237,7 +235,7 @@ class ITopLevelAccelerationStructure : public AccelerationStructure
 			static_assert(alignof(Instance<blas_ref_t>)==8ull);
 		};
 
-		// enum for distinguishing unions of Instance Types when there is no `INSTANCE_DATA_IS_POINTERS_TYPE_ENCODED_LSB` in build flags
+		// enum for distinguishing unions of Instance Types when using a polymorphic instance
 		enum class INSTANCE_TYPE : uint32_t
 		{
 			// StaticInstance

include/nbl/asset/ICPUAccelerationStructure.h

@@ -311,8 +311,6 @@ class ICPUTopLevelAccelerationStructure final : public ITopLevelAccelerationStru
 			if(!isMutable())
 				return;
 			m_buildFlags = buildFlags;
-			// we always clear this flag as we always store instances as polymorphic for ICPUTopLevelAccelerationStructure
-			m_buildFlags &= ~BUILD_FLAGS::INSTANCE_DATA_IS_POINTERS_TYPE_ENCODED_LSB;
 		}
 
 		//
@@ -361,7 +359,7 @@ class ICPUTopLevelAccelerationStructure final : public ITopLevelAccelerationStru
 		inline bool usesMotion() const override
 		{
 			for (const auto& instance : *m_instances)
-			if (instance.getType()!=INSTANCE_TYPE::STATIC)
+			if (instance.getType()!=INSTANCE_TYPE::STATIC || instance.getBase().blas && instance.getBase().blas->usesMotion())
 				return true;
 			return false;
 		}

include/nbl/video/IDeviceMemoryAllocator.h

@@ -9,9 +9,12 @@
 namespace nbl::video
 {
 
-class IDeviceMemoryAllocator
+class NBL_API2 IDeviceMemoryAllocator
 {
 	public:
+		// right now we only support this interface handing out memory for one device or group
+		virtual ILogicalDevice* getDeviceForAllocations() const = 0;
+
 		struct SAllocateInfo
 		{
 			size_t size : 54 = 0ull;
@@ -21,6 +24,21 @@ class IDeviceMemoryAllocator
 			// size_t opaqueCaptureAddress = 0u; Note that this mechanism is intended only to support capture/replay tools, and is not recommended for use in other applications.
 		};
 
+		struct SAllocation
+		{
+			static constexpr size_t InvalidMemoryOffset = 0xdeadbeefBadC0ffeull;
+			bool isValid() const
+			{
+				return memory && (offset!=InvalidMemoryOffset);
+			}
+
+			core::smart_refctd_ptr<IDeviceMemoryAllocation> memory = nullptr;
+			size_t offset = InvalidMemoryOffset;
+		};
+
+		virtual SAllocation allocate(const SAllocateInfo& info) = 0;
+
+
 		//! IMemoryTypeIterator extracts memoryType indices from memoryTypeBits in arbitrary order
 		//! which is used to give priority to memoryTypes in try-allocate usages where allocations may fail with some memoryTypes
 		//! IMemoryTypeIterator will construct SAllocateInfo from object's memory requirements, allocateFlags and dedication using operator()
@@ -85,20 +103,6 @@ class IDeviceMemoryAllocator
 
 				uint32_t currentIndex = 0u;
 		};
-
-
-		struct SAllocation
-		{
-			static constexpr size_t InvalidMemoryOffset = 0xdeadbeefBadC0ffeull;
-			bool isValid() const
-			{
-				return memory && (offset!=InvalidMemoryOffset);
-			}
-
-			core::smart_refctd_ptr<IDeviceMemoryAllocation> memory = nullptr;
-			size_t offset = InvalidMemoryOffset;
-		};
-		virtual SAllocation allocate(const SAllocateInfo& info) = 0;
 
 		template<class memory_type_iterator_t=DefaultMemoryTypeIterator>
 		inline SAllocation allocate(

include/nbl/video/IDeviceMemoryBacked.h

@@ -26,14 +26,16 @@ class IDeviceMemoryBacked : public IBackendObject
 {
     public:
         //!
-        struct SCachedCreationParams
+        constexpr static inline uint8_t MaxQueueFamilies = 7;
+        struct SCreationParams
         {
             // A Pre-Destroy-Step is called out just before a `vkDestory` or `glDelete`, this is only useful for "imported" resources
             std::unique_ptr<ICleanup> preDestroyCleanup = nullptr;
             // A Post-Destroy-Step is called in this class' destructor, this is only useful for "imported" resources
             std::unique_ptr<ICleanup> postDestroyCleanup = nullptr;
             // If more than one, then we're doing concurrent resource sharing
             uint8_t queueFamilyIndexCount = 0u;
+            const uint32_t* queueFamilyIndices = nullptr;
             // Thus the destructor will skip the call to `vkDestroy` or `glDelete` on the handle, this is only useful for "imported" objects
             bool skipHandleDestroy = false;
 
@@ -43,7 +45,8 @@ class IDeviceMemoryBacked : public IBackendObject
                 return queueFamilyIndexCount>1u;
             }
         };
-        inline const SCachedCreationParams& getCachedCreationParams() const {return m_cachedCreationParams;}
+        // TODO: change name later on, but right now too much code to refactor
+        inline const SCreationParams& getCachedCreationParams() const {return m_cachedCreationParams;}
 
         //! We need to know to cast to `IGPUBuffer` or `IGPUImage`
         enum E_OBJECT_TYPE : bool
@@ -87,15 +90,14 @@ class IDeviceMemoryBacked : public IBackendObject
         //! Returns the allocation which is bound to the resource
         virtual SMemoryBinding getBoundMemory() const = 0;
 
-        //! For constructor parameter only
-        struct SCreationParams : SCachedCreationParams
-        {
-            const uint32_t* queueFamilyIndices = nullptr;
-        };
-
     protected:
         inline IDeviceMemoryBacked(core::smart_refctd_ptr<const ILogicalDevice>&& originDevice, SCreationParams&& creationParams, const SDeviceMemoryRequirements& reqs)
-            : IBackendObject(std::move(originDevice)), m_cachedCreationParams(std::move(creationParams)), m_cachedMemoryReqs(reqs) {}
+            : IBackendObject(std::move(originDevice)), m_cachedCreationParams(std::move(creationParams)), m_cachedMemoryReqs(reqs)
+        {
+            std::fill_n(m_queueFamilies,MaxQueueFamilies,~0u);
+            std::copy_n(m_cachedCreationParams.queueFamilyIndices,m_cachedCreationParams.queueFamilyIndexCount,m_queueFamilies);
+            m_cachedCreationParams.queueFamilyIndices = m_queueFamilies;
+        }
         inline virtual ~IDeviceMemoryBacked()
         {
             assert(!m_cachedCreationParams.preDestroyCleanup); // derived class should have already cleared this out
@@ -109,8 +111,9 @@ class IDeviceMemoryBacked : public IBackendObject
 
 
         //! members
-        SCachedCreationParams m_cachedCreationParams;
+        SCreationParams m_cachedCreationParams;
         SDeviceMemoryRequirements m_cachedMemoryReqs;
+        uint32_t m_queueFamilies[MaxQueueFamilies];
 };
 
 template<typename T>

include/nbl/video/IGPUAccelerationStructure.h

@@ -411,7 +411,7 @@ class IGPUTopLevelAccelerationStructure : public asset::ITopLevelAccelerationStr
 						if (buildRangeInfo.instanceCount>dstAS->getMaxInstanceCount())
 							return false;
 
-						const bool arrayOfPointers = buildFlags.hasFlags(BUILD_FLAGS::INSTANCE_DATA_IS_POINTERS_TYPE_ENCODED_LSB);
+						const bool arrayOfPointers = instanceDataTypeEncodedInPointersLSB;
 						constexpr bool HostBuild = std::is_same_v<BufferType,asset::ICPUBuffer>;
 						// I'm not gonna do the `std::conditional_t<HostBuild,,>` to get the correct Instance struct type as they're the same size essentially
 						const size_t instanceSize = arrayOfPointers ? sizeof(void*):(
@@ -467,11 +467,13 @@ class IGPUTopLevelAccelerationStructure : public asset::ITopLevelAccelerationStr
 
 
 				core::bitflag<BUILD_FLAGS> buildFlags = BUILD_FLAGS::PREFER_FAST_BUILD_BIT;
+				// What we use to indicate `VkAccelerationStructureGeometryInstancesDataKHR::arrayOfPointers`
+				uint8_t instanceDataTypeEncodedInPointersLSB : 1 = false;
 				const IGPUTopLevelAccelerationStructure* srcAS = nullptr;
 				IGPUTopLevelAccelerationStructure* dstAS = nullptr;
-				// depending on the presence certain bits in `buildFlags` this buffer will be filled with:
+				// depending on value of certain build info members this buffer will be filled with:
 				// - addresses to `StaticInstance`, `MatrixMotionInstance`, `SRTMotionInstance` packed in upper 60 bits 
-				//   and struct type in lower 4 bits if and only if `buildFlags.hasFlags(INSTANCE_DATA_IS_POINTERS_TYPE_ENCODED_LSB)`, otherwise:
+				//   and struct type in lower 4 bits if and only if `instanceDataTypeEncodedInPointersLSB`, otherwise:
 				//	+ an array of `PolymorphicInstance` if our `SCreationParams::flags.hasFlags(MOTION_BIT)`, otherwise
 				//	+ an array of `StaticInstance`
 				asset::SBufferBinding<const BufferType> instanceData = {};
@@ -482,7 +484,6 @@ class IGPUTopLevelAccelerationStructure : public asset::ITopLevelAccelerationStr
 		using HostBuildInfo = BuildInfo<asset::ICPUBuffer>;
 
 		//! BEWARE, OUR RESOURCE LIFETIME TRACKING DOES NOT WORK ACROSS TLAS->BLAS boundaries with these types of BLAS references!
-		// TODO: Investigate `EXT_private_data` to be able to go ` -> IGPUBottomLevelAccelerationStructure` on Host Builds
 		using DeviceInstance = Instance<IGPUBottomLevelAccelerationStructure::device_op_ref_t>;
 		using HostInstance = Instance<IGPUBottomLevelAccelerationStructure::host_op_ref_t>;
 		static_assert(sizeof(DeviceInstance)==sizeof(HostInstance));

include/nbl/video/ILogicalDevice.h

@@ -26,7 +26,9 @@ class IPhysicalDevice;
 class NBL_API2 ILogicalDevice : public core::IReferenceCounted, public IDeviceMemoryAllocator
 {
     public:
-        constexpr static inline uint8_t MaxQueueFamilies = 7;
+        inline ILogicalDevice* getDeviceForAllocations() const override {return const_cast<ILogicalDevice*>(this);}
+
+        constexpr static inline uint8_t MaxQueueFamilies = IDeviceMemoryBacked::MaxQueueFamilies;
         struct SQueueCreationParams
         {
             constexpr static inline uint8_t MaxQueuesInFamily = 15;
@@ -331,6 +333,11 @@ class NBL_API2 ILogicalDevice : public core::IReferenceCounted, public IDeviceMe
                 m_logger.log("Failed to create Buffer, size %d larger than Device %p's limit (%u)!",system::ILogger::ELL_ERROR,creationParams.size,this,maxSize);
                 return nullptr;
             }
+            if (creationParams.queueFamilyIndexCount>MaxQueueFamilies)
+            {
+                m_logger.log("Failed to create Buffer, queue family count %d for concurrent sharing larger than our max %d!",system::ILogger::ELL_ERROR,creationParams.queueFamilyIndexCount,MaxQueueFamilies);
+                return nullptr;
+            }
             return createBuffer_impl(std::move(creationParams));
         }
         // Create a BufferView, to a shader; a fake 1D-like texture with no interpolation (@see ICPUBufferView)
@@ -343,7 +350,12 @@ class NBL_API2 ILogicalDevice : public core::IReferenceCounted, public IDeviceMe
                 m_logger.log("Failed to create Image, invalid creation parameters!",system::ILogger::ELL_ERROR);
                 return nullptr;
             }
-            // TODO: @Cyprian validation of creationParams against the device's limits (sample counts, etc.) see vkCreateImage
+            if (creationParams.queueFamilyIndexCount>MaxQueueFamilies)
+            {
+                m_logger.log("Failed to create Image, queue family count %d for concurrent sharing larger than our max %d!",system::ILogger::ELL_ERROR,creationParams.queueFamilyIndexCount,MaxQueueFamilies);
+                return nullptr;
+            }
+            // TODO: validation of creationParams against the device's limits (sample counts, etc.) see vkCreateImage docs
             return createImage_impl(std::move(creationParams));
         }
         // Create an ImageView that can actually be used by shaders (@see ICPUImageView)

include/nbl/video/alloc/CAsyncSingleBufferSubAllocator.h

@@ -157,7 +157,7 @@ class CAsyncSingleBufferSubAllocator
         template<typename... Args>
         inline size_type multi_allocate(uint32_t count, Args&&... args) noexcept
         {
-            return multi_alloc(decltype(deferredFrees)::default_wait(),count,std::forward<Args>(args)...);
+            return multi_allocate(TimelineEventHandlerBase::default_wait(),count,std::forward<Args>(args)...);
         }
         //! attempt to allocate, if fail (presumably because of fragmentation), then keep trying till timeout is reached
         template<class Clock=typename std::chrono::steady_clock, typename... Args>

include/nbl/video/asset_traits.h

@@ -196,7 +196,7 @@ struct asset_traits<asset::ICPUBottomLevelAccelerationStructure>
 	// we don't need to descend during DFS into other assets
 	constexpr static inline bool HasChildren = true;
 	// the video type
-	using video_t = IGPUImageView;
+	using video_t = IGPUBottomLevelAccelerationStructure;
 	// lookup type
 	using lookup_t = const video_t*;
 };