Merge pull request #16458 from hrydgard/desktop-friendly-msaa

Implement MSAA support for desktop GPUs in Vulkan

Common/Data/Convert/SmallDataConvert.h

@@ -21,7 +21,7 @@ extern const float one_over_255_x4[4];
 extern const float exactly_255_x4[4];
 
 // Utilities useful for filling in std140-layout uniform buffers, and similar.
-// NEON intrinsics: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0491f/BABDCGGF.html
+// NEON intrinsics: https://developer.arm.com/documentation/den0018/a/NEON-Intrinsics?lang=en
 
 // LSBs in f[0], etc.
 inline void Uint8x4ToFloat4(float f[4], uint32_t u) {

Common/GPU/D3D11/thin3d_d3d11.cpp

@@ -271,6 +271,7 @@ D3D11DrawContext::D3D11DrawContext(ID3D11Device *device, ID3D11DeviceContext *de
 	caps_.fragmentShaderDepthWriteSupported = true;
 	caps_.fragmentShaderStencilWriteSupported = false;
 	caps_.blendMinMaxSupported = true;
+	caps_.multiSampleLevelsMask = 1;   // More could be supported with some work.
 
 	D3D11_FEATURE_DATA_D3D11_OPTIONS options{};
 	HRESULT result = device_->CheckFeatureSupport(D3D11_FEATURE_D3D11_OPTIONS, &options, sizeof(options));
@@ -309,6 +310,8 @@ D3D11DrawContext::D3D11DrawContext(ID3D11Device *device, ID3D11DeviceContext *de
 		dxgiDevice->Release();
 	}
 
+	caps_.isTilingGPU = false;
+
 	// Temp texture for read-back of small images. Custom textures are created on demand for larger ones.
 	// TODO: Should really benchmark if this extra complexity has any benefit.
 	D3D11_TEXTURE2D_DESC packDesc{};

Common/GPU/D3D9/thin3d_d3d9.cpp

@@ -763,6 +763,8 @@ D3D9Context::D3D9Context(IDirect3D9 *d3d, IDirect3D9Ex *d3dEx, int adapterId, ID
 	caps_.fragmentShaderDepthWriteSupported = true;
 	caps_.fragmentShaderStencilWriteSupported = false;
 	caps_.blendMinMaxSupported = true;
+	caps_.isTilingGPU = false;
+	caps_.multiSampleLevelsMask = 1;  // More could be supported with some work.
 
 	if ((caps.RasterCaps & D3DPRASTERCAPS_ANISOTROPY) != 0 && caps.MaxAnisotropy > 1) {
 		caps_.anisoSupported = true;

Common/GPU/OpenGL/thin3d_gl.cpp

@@ -557,6 +557,7 @@ OpenGLContext::OpenGLContext() {
 	caps_.framebufferStencilBlitSupported = caps_.framebufferBlitSupported;
 	caps_.depthClampSupported = gl_extensions.ARB_depth_clamp || gl_extensions.EXT_depth_clamp;
 	caps_.blendMinMaxSupported = gl_extensions.EXT_blend_minmax;
+	caps_.multiSampleLevelsMask = 1;  // More could be supported with some work.
 
 	if (gl_extensions.IsGLES) {
 		caps_.clipDistanceSupported = gl_extensions.EXT_clip_cull_distance || gl_extensions.APPLE_clip_distance;
@@ -601,6 +602,10 @@ OpenGLContext::OpenGLContext() {
 		caps_.vendor = GPUVendor::VENDOR_UNKNOWN;
 		break;
 	}
+
+	// Very rough heuristic!
+	caps_.isTilingGPU = gl_extensions.IsGLES && caps_.vendor != GPUVendor::VENDOR_NVIDIA && caps_.vendor != GPUVendor::VENDOR_INTEL;
+
 	for (int i = 0; i < GLRenderManager::MAX_INFLIGHT_FRAMES; i++) {
 		frameData_[i].push = renderManager_.CreatePushBuffer(i, GL_ARRAY_BUFFER, 64 * 1024);
 	}

Common/GPU/Vulkan/VulkanBarrier.cpp

@@ -9,4 +9,5 @@ void VulkanBarrier::Flush(VkCommandBuffer cmd) {
 	imageBarriers_.clear();
 	srcStageMask_ = 0;
 	dstStageMask_ = 0;
+	dependencyFlags_ = 0;
 }

Common/GPU/Vulkan/VulkanContext.cpp

@@ -57,7 +57,7 @@ std::string VulkanVendorString(uint32_t vendorId) {
 	case VULKAN_VENDOR_ARM: return "ARM";
 	case VULKAN_VENDOR_QUALCOMM: return "Qualcomm";
 	case VULKAN_VENDOR_IMGTEC: return "Imagination";
-
+	case VULKAN_VENDOR_APPLE: return "Apple";
 	default:
 		return StringFromFormat("%08x", vendorId);
 	}
@@ -253,15 +253,21 @@ VkResult VulkanContext::CreateInstance(const CreateInfo &info) {
 			VkPhysicalDeviceProperties2 props2{VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2};
 			VkPhysicalDevicePushDescriptorPropertiesKHR pushProps{VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR};
 			VkPhysicalDeviceExternalMemoryHostPropertiesEXT extHostMemProps{VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT};
+			VkPhysicalDeviceDepthStencilResolveProperties depthStencilResolveProps{VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES};
+
 			props2.pNext = &pushProps;
 			pushProps.pNext = &extHostMemProps;
+			extHostMemProps.pNext = &depthStencilResolveProps;
 			vkGetPhysicalDeviceProperties2KHR(physical_devices_[i], &props2);
 			// Don't want bad pointers sitting around.
 			props2.pNext = nullptr;
 			pushProps.pNext = nullptr;
+			extHostMemProps.pNext = nullptr;
+			depthStencilResolveProps.pNext = nullptr;
 			physicalDeviceProperties_[i].properties = props2.properties;
 			physicalDeviceProperties_[i].pushDescriptorProperties = pushProps;
 			physicalDeviceProperties_[i].externalMemoryHostProperties = extHostMemProps;
+			physicalDeviceProperties_[i].depthStencilResolve = depthStencilResolveProps;
 		}
 	} else {
 		for (uint32_t i = 0; i < gpu_count; i++) {
@@ -329,7 +335,7 @@ bool VulkanContext::MemoryTypeFromProperties(uint32_t typeBits, VkFlags requirem
 	for (uint32_t i = 0; i < 32; i++) {
 		if ((typeBits & 1) == 1) {
 			// Type is available, does it match user properties?
-			if ((memory_properties.memoryTypes[i].propertyFlags & requirements_mask) == requirements_mask) {
+			if ((memory_properties_.memoryTypes[i].propertyFlags & requirements_mask) == requirements_mask) {
 				*typeIndex = i;
 				return true;
 			}
@@ -569,17 +575,17 @@ void VulkanContext::ChooseDevice(int physical_device) {
 	}
 
 	// This is as good a place as any to do this.
-	vkGetPhysicalDeviceMemoryProperties(physical_devices_[physical_device_], &memory_properties);
-	INFO_LOG(G3D, "Memory Types (%d):", memory_properties.memoryTypeCount);
-	for (int i = 0; i < (int)memory_properties.memoryTypeCount; i++) {
+	vkGetPhysicalDeviceMemoryProperties(physical_devices_[physical_device_], &memory_properties_);
+	INFO_LOG(G3D, "Memory Types (%d):", memory_properties_.memoryTypeCount);
+	for (int i = 0; i < (int)memory_properties_.memoryTypeCount; i++) {
 		// Don't bother printing dummy memory types.
-		if (!memory_properties.memoryTypes[i].propertyFlags)
+		if (!memory_properties_.memoryTypes[i].propertyFlags)
 			continue;
-		INFO_LOG(G3D, "  %d: Heap %d; Flags: %s%s%s%s  ", i, memory_properties.memoryTypes[i].heapIndex,
-			(memory_properties.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) ? "DEVICE_LOCAL " : "",
-			(memory_properties.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) ? "HOST_VISIBLE " : "",
-			(memory_properties.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_HOST_CACHED_BIT) ? "HOST_CACHED " : "",
-			(memory_properties.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) ? "HOST_COHERENT " : "");
+		INFO_LOG(G3D, "  %d: Heap %d; Flags: %s%s%s%s  ", i, memory_properties_.memoryTypes[i].heapIndex,
+			(memory_properties_.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT) ? "DEVICE_LOCAL " : "",
+			(memory_properties_.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) ? "HOST_VISIBLE " : "",
+			(memory_properties_.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_HOST_CACHED_BIT) ? "HOST_CACHED " : "",
+			(memory_properties_.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) ? "HOST_COHERENT " : "");
 	}
 
 	// Optional features
@@ -606,6 +612,7 @@ void VulkanContext::ChooseDevice(int physical_device) {
 	deviceFeatures_.enabled.standard.shaderClipDistance = deviceFeatures_.available.standard.shaderClipDistance;
 	deviceFeatures_.enabled.standard.shaderCullDistance = deviceFeatures_.available.standard.shaderCullDistance;
 	deviceFeatures_.enabled.standard.geometryShader = deviceFeatures_.available.standard.geometryShader;
+	deviceFeatures_.enabled.standard.sampleRateShading = deviceFeatures_.available.standard.sampleRateShading;
 
 	deviceFeatures_.enabled.multiview = { VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES };
 	deviceFeatures_.enabled.multiview.multiview = deviceFeatures_.available.multiview.multiview;

Common/GPU/Vulkan/VulkanContext.h

@@ -35,6 +35,7 @@ enum {
 	VULKAN_VENDOR_ARM = 0x000013B5,  // Mali
 	VULKAN_VENDOR_QUALCOMM = 0x00005143,
 	VULKAN_VENDOR_IMGTEC = 0x00001010,  // PowerVR
+	VULKAN_VENDOR_APPLE = 0x0000106b,  // Apple through MoltenVK
 };
 
 VK_DEFINE_HANDLE(VmaAllocator);
@@ -253,6 +254,7 @@ class VulkanContext {
 		VkPhysicalDeviceProperties properties;
 		VkPhysicalDevicePushDescriptorPropertiesKHR pushDescriptorProperties;
 		VkPhysicalDeviceExternalMemoryHostPropertiesEXT externalMemoryHostProperties;
+		VkPhysicalDeviceDepthStencilResolveProperties depthStencilResolve;
 	};
 
 	struct AllPhysicalDeviceFeatures {
@@ -283,6 +285,10 @@ class VulkanContext {
 		return device_extensions_enabled_;
 	}
 
+	const VkPhysicalDeviceMemoryProperties &GetMemoryProperties() const {
+		return memory_properties_;
+	}
+
 	struct PhysicalDeviceFeatures {
 		AllPhysicalDeviceFeatures available{};
 		AllPhysicalDeviceFeatures enabled{};
@@ -401,7 +407,8 @@ class VulkanContext {
 	uint32_t graphics_queue_family_index_ = -1;
 	std::vector<PhysicalDeviceProps> physicalDeviceProperties_;
 	std::vector<VkQueueFamilyProperties> queueFamilyProperties_;
-	VkPhysicalDeviceMemoryProperties memory_properties{};
+
+	VkPhysicalDeviceMemoryProperties memory_properties_{};
 
 	// Custom collection of things that are good to know
 	VulkanPhysicalDeviceInfo deviceInfo_{};