Vulkan Backend

Vulkan Backend

Release boundary: This page documents an experimental implementation outside the blocked and uncertified stable-v1 product scope (Windows 11 x64/MSVC v143 with D3D11 or Windows NullRHI and C++ modules). It is not a supported or release-certified backend.

SparkEngine has a Vulkan RHI implementation for Windows, Linux, and MoltenVK-based macOS development work. It is experimental and outside stable-v1, not a primary cross-platform supported renderer.

Overview

• Namespace: Spark::RHI::Vulkan
• Files: Graphics/RHI/Vulkan/VulkanDevice.{h,cpp} (~550 LOC header), VulkanCommandList.cpp, VulkanDescriptorCache.{h,cpp}, VulkanFormatHelpers.cpp.
• Guard: #ifdef SPARK_VULKAN_SUPPORT (set by CMake when ENABLE_VULKAN=ON).
• Linkage: libvulkan.so.1 / vulkan-1.dll via volk loader.
• API level: Vulkan 1.4 — dynamic rendering, push descriptors, timeline semaphores, synchronization2.

Architecture

Class	Purpose	RHI interface
VulkanDevice	Instance + physical device + logical device + queues	IRHIDevice
VulkanBuffer	VkBuffer + VkDeviceMemory, optional persistent mapping	IRHIBuffer
VulkanTexture	VkImage + VkImageView, tracks current layout	IRHITexture
VulkanShader	VkShaderModule compiled from SPIR-V	IRHIShader
VulkanSampler	VkSampler	IRHISampler
VulkanPipelineState	VkPipeline + VkPipelineLayout	IRHIPipelineState
VulkanSwapChain	VkSurfaceKHR + VkSwapchainKHR	IRHISwapChain
VulkanCommandList	Per-frame primary VkCommandBuffer	IRHICommandList
VulkanDescriptorCache	LRU-cached VkDescriptorSet layouts	internal

Queue families

Queue selection picks the best-fit families from the platform:

struct QueueFamilyIndices
{
    std::optional<uint32_t> graphicsFamily;   // graphics + present
    std::optional<uint32_t> computeFamily;    // async compute (optional)
    std::optional<uint32_t> transferFamily;   // DMA copies (optional)
    std::optional<uint32_t> presentFamily;
    bool IsComplete() const;
};

Separate compute and transfer queues are used when the adapter exposes them; otherwise the graphics queue handles all three.

Features

• Vulkan 1.4 dynamic rendering — no VkRenderPass objects; pipelines bind color / depth formats inline via VK_KHR_dynamic_rendering (core in 1.3+).
• Push descriptors — frequent CBV/SRV bindings skip descriptor-set allocation via VK_KHR_push_descriptor.
• Timeline semaphores — unified CPU/GPU synchronization primitive (VK_KHR_timeline_semaphore). Used for async compute and frame pacing.
• Synchronization2 — coarse-grained VkPipelineStageFlags2 and image-barrier builders.
• VMA / explicit memory — currently plain vkAllocateMemory with a per-heap budget cache; the VMA port is on the Tier-2 roadmap.
• Descriptor cache — VulkanDescriptorCache memoizes VkDescriptorSetLayout so render passes don't rebuild them every frame.
• Validation layer — Debug builds automatically enable VK_LAYER_KHRONOS_validation; the VK_EXT_debug_utils callback routes messages through Spark::Logger.
• Lavapipe development route — Linux development tests may use Mesa Lavapipe with explicitly configured software-ICD and display settings. This is not a generic fallback or a support/certification claim. See Cross-Compilation: Wine Testing.

Capability detection

auto& caps = device->GetCapabilities();
caps.rayTracingSupport         // VK_KHR_ray_tracing_pipeline
caps.meshShaderSupport         // VK_EXT_mesh_shader
caps.bindlessResourceSupport   // VK_EXT_descriptor_indexing + update-after-bind
caps.conservativeRasterSupport // VK_EXT_conservative_rasterization

Factory registration

// RHIFactory.cpp
case GraphicsBackend::Vulkan:
    device = std::make_unique<Vulkan::VulkanDevice>();
    break;

On Linux, RHIFactory::GetRecommendedBackend() can select Vulkan from the compiled and available candidates. Runtime availability and failure handling are owned by RHIBridge, not by the factory selector.

Known issues / scope

• MoltenVK macOS development path — no certified surface or packaged-runtime proof exists; the native Metal backend is also partial and blocked.
• Mesh shaders — plumbed through VK_EXT_mesh_shader, requires recent Mesa (RADV) or NVIDIA drivers.
• DXR equivalent — implemented via VK_KHR_ray_tracing_pipeline but currently flagged experimental. See DXR Raytracing for the D3D12 sibling.
• Release evidence — full production-pass execution, GPU readback, shader-toolchain, and supported-driver evidence remain incomplete.

Related

• RHI Abstraction Layer — shared interface.
• D3D12 Backend — Windows modern alternative.
• OpenGL Backend — experimental development alternative when compiled.
• Cross-Compilation: Wine Testing — running D3D11 builds under Wine + DXVK (Vulkan).