vk: Add VK_EXT_provoking_vertex optimization

core/rend/vulkan/drawer.cpp

@@ -404,7 +404,15 @@ bool Drawer::Draw(const Texture *fogTexture, const Texture *paletteTexture)
 	static const float scopeColor[4] = { 0.75f, 0.75f, 0.75f, 1.0f };
 	CommandBufferDebugScope _(cmdBuffer, "Draw", scopeColor);
 
-	setFirstProvokingVertex(pvrrc);
+	if (VulkanContext::Instance()->hasProvokingVertex())
+	{
+		// Pipelines are using VK_EXT_provoking_vertex, no need to
+		// re-order vertices
+	}
+	else
+	{
+		setFirstProvokingVertex(pvrrc);
+	}
 
 	// Upload vertex and index buffers
 	VertexShaderUniforms vtxUniforms;

core/rend/vulkan/oit/oit_drawer.cpp

@@ -341,7 +341,15 @@ bool OITDrawer::Draw(const Texture *fogTexture, const Texture *paletteTexture)
 	bool firstFrameAfterInit = oitBuffers->isFirstFrameAfterInit();
 	oitBuffers->OnNewFrame(cmdBuffer);
 
-	setFirstProvokingVertex(pvrrc);
+	if (VulkanContext::Instance()->hasProvokingVertex())
+	{
+		// Pipelines are using VK_EXT_provoking_vertex, no need to
+		// re-order vertices
+	}
+	else
+	{
+		setFirstProvokingVertex(pvrrc);
+	}
 
 	// Upload vertex and index buffers
 	UploadMainBuffer(vtxUniforms, fragUniforms);

core/rend/vulkan/oit/oit_pipeline.cpp

@@ -49,6 +49,17 @@ void OITPipelineManager::CreatePipeline(u32 listType, bool autosort, const PolyP
 	  0.0f,                                         // depthBiasSlopeFactor
 	  1.0f                                          // lineWidth
 	);
+
+	// Dreamcast uses the last vertex as the provoking vertex, but Vulkan uses the first.
+	// Utilize VK_EXT_provoking_vertex when available to set the provoking vertex to be the
+	// last vertex
+	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provokingVertexInfo{};
+	if (GetContext()->hasProvokingVertex())
+	{
+		provokingVertexInfo.provokingVertexMode = vk::ProvokingVertexModeEXT::eLastVertex;
+		pipelineRasterizationStateCreateInfo.pNext = &provokingVertexInfo;
+	}
+
 	vk::PipelineMultisampleStateCreateInfo pipelineMultisampleStateCreateInfo;
 
 	// Depth and stencil
@@ -211,6 +222,17 @@ void OITPipelineManager::CreateFinalPipeline(bool dithering)
 	  0.0f,                                         // depthBiasSlopeFactor
 	  1.0f                                          // lineWidth
 	);
+
+	// Dreamcast uses the last vertex as the provoking vertex, but Vulkan uses the first.
+	// Utilize VK_EXT_provoking_vertex when available to set the provoking vertex to be the
+	// last vertex
+	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provokingVertexInfo{};
+	if (GetContext()->hasProvokingVertex())
+	{
+		provokingVertexInfo.provokingVertexMode = vk::ProvokingVertexModeEXT::eLastVertex;
+		pipelineRasterizationStateCreateInfo.pNext = &provokingVertexInfo;
+	}
+
 	vk::PipelineMultisampleStateCreateInfo pipelineMultisampleStateCreateInfo;
 
 	// Depth and stencil
@@ -295,6 +317,17 @@ void OITPipelineManager::CreateClearPipeline()
 	  0.0f,                                         // depthBiasSlopeFactor
 	  1.0f                                          // lineWidth
 	);
+
+	// Dreamcast uses the last vertex as the provoking vertex, but Vulkan uses the first.
+	// Utilize VK_EXT_provoking_vertex when available to set the provoking vertex to be the
+	// last vertex
+	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provokingVertexInfo{};
+	if (GetContext()->hasProvokingVertex())
+	{
+		provokingVertexInfo.provokingVertexMode = vk::ProvokingVertexModeEXT::eLastVertex;
+		pipelineRasterizationStateCreateInfo.pNext = &provokingVertexInfo;
+	}
+
 	vk::PipelineMultisampleStateCreateInfo pipelineMultisampleStateCreateInfo;
 
 	// Depth and stencil
@@ -377,6 +410,17 @@ void OITPipelineManager::CreateModVolPipeline(ModVolMode mode, int cullMode, boo
 	  0.0f,                                         // depthBiasSlopeFactor
 	  1.0f                                          // lineWidth
 	);
+
+	// Dreamcast uses the last vertex as the provoking vertex, but Vulkan uses the first.
+	// Utilize VK_EXT_provoking_vertex when available to set the provoking vertex to be the
+	// last vertex
+	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provokingVertexInfo{};
+	if (GetContext()->hasProvokingVertex())
+	{
+		provokingVertexInfo.provokingVertexMode = vk::ProvokingVertexModeEXT::eLastVertex;
+		pipelineRasterizationStateCreateInfo.pNext = &provokingVertexInfo;
+	}
+
 	vk::PipelineMultisampleStateCreateInfo pipelineMultisampleStateCreateInfo;
 
 	// Depth and stencil
@@ -489,6 +533,17 @@ void OITPipelineManager::CreateTrModVolPipeline(ModVolMode mode, int cullMode, b
 	  0.0f,                                         // depthBiasSlopeFactor
 	  1.0f                                          // lineWidth
 	);
+
+	// Dreamcast uses the last vertex as the provoking vertex, but Vulkan uses the first.
+	// Utilize VK_EXT_provoking_vertex when available to set the provoking vertex to be the
+	// last vertex
+	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provokingVertexInfo{};
+	if (GetContext()->hasProvokingVertex())
+	{
+		provokingVertexInfo.provokingVertexMode = vk::ProvokingVertexModeEXT::eLastVertex;
+		pipelineRasterizationStateCreateInfo.pNext = &provokingVertexInfo;
+	}
+
 	vk::PipelineMultisampleStateCreateInfo pipelineMultisampleStateCreateInfo;
 
 	// Depth and stencil

core/rend/vulkan/pipeline.cpp

@@ -67,6 +67,17 @@ void PipelineManager::CreateModVolPipeline(ModVolMode mode, int cullMode, bool n
 	  0.0f,                                         // depthBiasSlopeFactor
 	  1.0f                                          // lineWidth
 	);
+
+	// Dreamcast uses the last vertex as the provoking vertex, but Vulkan uses the first.
+	// Utilize VK_EXT_provoking_vertex when available to set the provoking vertex to be the
+	// last vertex
+	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provokingVertexInfo{};
+	if (GetContext()->hasProvokingVertex())
+	{
+		provokingVertexInfo.provokingVertexMode = vk::ProvokingVertexModeEXT::eLastVertex;
+		pipelineRasterizationStateCreateInfo.pNext = &provokingVertexInfo;
+	}
+
 	vk::PipelineMultisampleStateCreateInfo pipelineMultisampleStateCreateInfo;
 
 	// Depth and stencil
@@ -189,6 +200,17 @@ void PipelineManager::CreateDepthPassPipeline(int cullMode, bool naomi2)
 	  0.0f,                                         // depthBiasSlopeFactor
 	  1.0f                                          // lineWidth
 	);
+
+	// Dreamcast uses the last vertex as the provoking vertex, but Vulkan uses the first.
+	// Utilize VK_EXT_provoking_vertex when available to set the provoking vertex to be the
+	// last vertex
+	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provokingVertexInfo{};
+	if (GetContext()->hasProvokingVertex())
+	{
+		provokingVertexInfo.provokingVertexMode = vk::ProvokingVertexModeEXT::eLastVertex;
+		pipelineRasterizationStateCreateInfo.pNext = &provokingVertexInfo;
+	}
+
 	vk::PipelineMultisampleStateCreateInfo pipelineMultisampleStateCreateInfo;
 
 	// Depth and stencil
@@ -295,6 +317,17 @@ void PipelineManager::CreatePipeline(u32 listType, bool sortTriangles, const Pol
 	  0.0f,                                         // depthBiasSlopeFactor
 	  1.0f                                          // lineWidth
 	);
+
+	// Dreamcast uses the last vertex as the provoking vertex, but Vulkan uses the first.
+	// Utilize VK_EXT_provoking_vertex when available to set the provoking vertex to be the
+	// last vertex
+	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provokingVertexInfo{};
+	if (GetContext()->hasProvokingVertex())
+	{
+		provokingVertexInfo.provokingVertexMode = vk::ProvokingVertexModeEXT::eLastVertex;
+		pipelineRasterizationStateCreateInfo.pNext = &provokingVertexInfo;
+	}
+
 	vk::PipelineMultisampleStateCreateInfo pipelineMultisampleStateCreateInfo;
 
 	// Depth and stencil

core/rend/vulkan/vk_context_lr.cpp

@@ -71,6 +71,9 @@ bool VkCreateDevice(retro_vulkan_context* context, VkInstance instance, VkPhysic
 			physicalDevice = vkinstance.enumeratePhysicalDevices().front();
 	}
 	context->gpu = (VkPhysicalDevice)physicalDevice;
+
+	const vk::PhysicalDeviceProperties physicalDeviceProperties = physicalDevice.getProperties();
+
 	std::vector<vk::QueueFamilyProperties> queueFamilyProperties = physicalDevice.getQueueFamilyProperties();
 
 	// get the first index into queueFamilyProperties which supports graphics and compute
@@ -124,54 +127,119 @@ bool VkCreateDevice(retro_vulkan_context* context, VkInstance instance, VkPhysic
 			DEBUG_LOG(RENDERER, "Using distinct Graphics and Present queue families");
 	}
 
-	vk::PhysicalDeviceFeatures supportedFeatures;
-	physicalDevice.getFeatures(&supportedFeatures);
-	VulkanContext::Instance()->fragmentStoresAndAtomics = supportedFeatures.fragmentStoresAndAtomics;
-	VulkanContext::Instance()->samplerAnisotropy = supportedFeatures.samplerAnisotropy;
 
-	// Enable VK_KHR_dedicated_allocation if available
-	bool getMemReq2Supported = false;
-	VulkanContext::Instance()->dedicatedAllocationSupported = false;
-	std::vector<const char *> deviceExtensions = { VK_KHR_SWAPCHAIN_EXTENSION_NAME };
-	for (unsigned i = 0; i < num_required_device_extensions; i++)
-		deviceExtensions.push_back(required_device_extensions[i]);
-	for (const auto& property : physicalDevice.enumerateDeviceExtensionProperties())
+	std::set<std::string> supportedExtensions;
+
+	const auto deviceExtensionProperties = physicalDevice.enumerateDeviceExtensionProperties();
+	for (const auto& property : deviceExtensionProperties)
 	{
-		if (!strcmp(property.extensionName, VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME))
+		supportedExtensions.insert(property.extensionName);
+	}
+
+	std::vector<const char*> enabledExtensions;
+
+	const auto tryAddDeviceExtension = [&supportedExtensions = std::as_const(supportedExtensions), &enabledExtensions]
+	(std::string_view extensionName) -> bool
 		{
-			deviceExtensions.push_back(VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME);
-			getMemReq2Supported = true;
-		}
-		else if (!strcmp(property.extensionName, VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME))
+			if (supportedExtensions.count(extensionName.data()))
+			{
+				enabledExtensions.push_back(extensionName.data());
+				NOTICE_LOG(RENDERER, "Device extension enabled: %s", extensionName.data());
+				return true;
+			}
+			NOTICE_LOG(RENDERER, "Device extension unavailable: %s", extensionName.data());
+			return false;
+		};
+
+	// Required swapchain extension
+	tryAddDeviceExtension(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
+
+	// Enable VK_KHR_dedicated_allocation if available
+	if (physicalDeviceProperties.apiVersion >= VK_API_VERSION_1_1)
+	{
+		// Core in Vulkan 1.1
+		VulkanContext::Instance()->dedicatedAllocationSupported = true;
+	}
+	else
+	{
+		const bool getMemReq2Supported = tryAddDeviceExtension(VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME);
+		if (getMemReq2Supported)
 		{
-			deviceExtensions.push_back(VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME);
-			VulkanContext::Instance()->dedicatedAllocationSupported = true;
+			VulkanContext::Instance()->dedicatedAllocationSupported = tryAddDeviceExtension(VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME);
 		}
 	}
-	VulkanContext::Instance()->dedicatedAllocationSupported &= getMemReq2Supported;
+
+	// Check for VK_KHR_get_physical_device_properties2
+	// Core as of Vulkan 1.1
+	const bool getPhysicalDeviceProperties2Supported =
+		(physicalDeviceProperties.apiVersion >= VK_API_VERSION_1_1)
+		? true : tryAddDeviceExtension(VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME);
+
+	if (getPhysicalDeviceProperties2Supported)
+	{
+		// Enable VK_EXT_provoking_vertex if available
+		VulkanContext::Instance()->provokingVertexSupported = tryAddDeviceExtension(VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME);
+	}
+
+	// Get device features
+
+	vk::PhysicalDeviceFeatures2 featuresChain{};
+	vk::PhysicalDeviceFeatures& features = featuresChain.features;
+
+	vk::PhysicalDeviceProvokingVertexFeaturesEXT provokingVertexFeatures{};
+	if (VulkanContext::Instance()->provokingVertexSupported)
+	{
+		featuresChain.pNext = &provokingVertexFeatures;
+	}
+
+	// Get the physical device's features
+	if (getPhysicalDeviceProperties2Supported && featuresChain.pNext)
+	{
+		physicalDevice.getFeatures2(&featuresChain);
+	}
+	else
+	{
+		physicalDevice.getFeatures(&features);
+	}
+
+	if (VulkanContext::Instance()->provokingVertexSupported)
+	{
+		VulkanContext::Instance()->provokingVertexSupported &= provokingVertexFeatures.provokingVertexLast;
+	}
+
+	VulkanContext::Instance()->samplerAnisotropy = features.samplerAnisotropy;
+	VulkanContext::Instance()->fragmentStoresAndAtomics = features.fragmentStoresAndAtomics;
 
 	// create a Device
 	float queuePriority = 1.0f;
-	vk::DeviceQueueCreateInfo deviceQueueCreateInfos[] = {
+	vk::DeviceQueueCreateInfo deviceQueueCreateInfo[] = {
 			vk::DeviceQueueCreateInfo(vk::DeviceQueueCreateFlags(), context->queue_family_index, 1, &queuePriority),
 			vk::DeviceQueueCreateInfo(vk::DeviceQueueCreateFlags(), context->presentation_queue_family_index, 1, &queuePriority),
 	};
-	vk::PhysicalDeviceFeatures features(*required_features);
-	if (VulkanContext::Instance()->fragmentStoresAndAtomics)
-		features.fragmentStoresAndAtomics = true;
-	if (VulkanContext::Instance()->samplerAnisotropy)
-		features.samplerAnisotropy = true;
-	vk::Device device = physicalDevice.createDevice(vk::DeviceCreateInfo(vk::DeviceCreateFlags(),
-			context->queue_family_index == context->presentation_queue_family_index ? 1 : 2, deviceQueueCreateInfos,
-					num_required_device_layers, required_device_layers, deviceExtensions.size(), &deviceExtensions[0], &features));
-	context->device = (VkDevice)device;
+
+
+	vk::Device newDevice{};
+	if (getPhysicalDeviceProperties2Supported)
+	{
+		vk::DeviceCreateInfo deviceCreateInfo(vk::DeviceCreateFlags(), deviceQueueCreateInfo,
+			nullptr, enabledExtensions);
+		deviceCreateInfo.pNext = &featuresChain;
+		newDevice = physicalDevice.createDevice(deviceCreateInfo);
+	}
+	else
+	{
+		newDevice = physicalDevice.createDevice(vk::DeviceCreateInfo(vk::DeviceCreateFlags(), deviceQueueCreateInfo,
+			nullptr, enabledExtensions, &features));
+	}
+
+	context->device = (VkDevice)newDevice;
 #if VULKAN_HPP_DISPATCH_LOADER_DYNAMIC == 1
 	VULKAN_HPP_DEFAULT_DISPATCHER.init(context->device);
 #endif
 
 	// Queues
-	context->queue = (VkQueue)device.getQueue(context->queue_family_index, 0);
-	context->presentation_queue = (VkQueue)device.getQueue(context->presentation_queue_family_index, 0);
+	context->queue = (VkQueue)newDevice.getQueue(context->queue_family_index, 0);
+	context->presentation_queue = (VkQueue)newDevice.getQueue(context->presentation_queue_family_index, 0);
 
 	return true;
 }

core/rend/vulkan/vk_context_lr.h

@@ -92,6 +92,7 @@ class VulkanContext : public GraphicsContext, public FlightManager
 	bool SupportsSamplerAnisotropy() const { return samplerAnisotropy; }
 	bool SupportsDedicatedAllocation() const { return dedicatedAllocationSupported; }
 	bool hasPerPixel() override { return fragmentStoresAndAtomics; }
+	bool hasProvokingVertex() { return provokingVertexSupported; }
 	const VMAllocator& GetAllocator() const { return allocator; }
 	vk::DeviceSize GetMaxMemoryAllocationSize() const { return maxMemoryAllocationSize; }
 	f32 GetMaxSamplerAnisotropy() const { return samplerAnisotropy ? maxSamplerAnisotropy : 1.f; }
@@ -129,10 +130,11 @@ class VulkanContext : public GraphicsContext, public FlightManager
 	bool optimalTilingSupported1555 = false;
 	bool optimalTilingSupported4444 = false;
 public:
+	bool fragmentStoresAndAtomics = false;
 	bool samplerAnisotropy = false;
 	f32 maxSamplerAnisotropy = 0.f;
 	bool dedicatedAllocationSupported = false;
-	bool fragmentStoresAndAtomics = false;
+	bool provokingVertexSupported = false;
 private:
 	u32 vendorID = 0;
 

core/rend/vulkan/vmallocator.cpp

@@ -55,6 +55,8 @@ void VMAllocator::Init(vk::PhysicalDevice physicalDevice, vk::Device device, vk:
 		allocatorInfo.flags |= VMA_ALLOCATOR_CREATE_KHR_DEDICATED_ALLOCATION_BIT;
 
 	allocatorInfo.physicalDevice = (VkPhysicalDevice)physicalDevice;
+	// Top-out at vulkan 1.1
+	allocatorInfo.vulkanApiVersion = (physicalDevice.getProperties().apiVersion >= VK_API_VERSION_1_1) ? VK_API_VERSION_1_1 : VK_API_VERSION_1_0;
 	allocatorInfo.device = (VkDevice)device;
 	allocatorInfo.instance = (VkInstance)instance;
 #if !defined(NDEBUG) || defined(DEBUGFAST)