Merge branch 'master' into vk-bestpractices-arm-lod-clamping

.github/ISSUE_TEMPLATE/bug_report.md

@@ -12,7 +12,7 @@ assignees: ''
 
 Platform / OS / Hardware: 
 
-Github hash: 
+Flycast version: 
 
 Hardware: 
 

core/hw/pvr/ta_vtx.cpp

@@ -125,7 +125,6 @@ class BaseTAParser
 
 	static void reset()
 	{
-		tileclip_val = 0;
 		memset(FaceBaseColor, 0xff, sizeof(FaceBaseColor));
 		memset(FaceOffsColor, 0xff, sizeof(FaceOffsColor));
 		memset(FaceBaseColor1, 0xff, sizeof(FaceBaseColor1));
@@ -398,7 +397,7 @@ case num : {\
 
 				//32B
 			case ParamType_User_Tile_Clip:
-				SetTileClip(data->data_32[3] & 63, data->data_32[4] & 31, data->data_32[5] & 63, data->data_32[6] & 31);
+				setClipRect(data->data_32[3] & 63, data->data_32[4] & 31, data->data_32[5] & 63, data->data_32[6] & 31);
 				data += SZ32;
 				break;
 
@@ -414,7 +413,7 @@ case num : {\
 				//PolyType :32B/64B
 			case ParamType_Polygon_or_Modifier_Volume:
 				{
-					TileClipMode(data->pcw.User_Clip);
+					setClipMode(data->pcw.User_Clip);
 					//Yep , C++ IS lame & limited
 					#include "ta_const_df.h"
 					if (CurrentList == ListType_None && !startList(data->pcw.ListType))
@@ -466,7 +465,7 @@ case num : {\
 				//32B
 				//Sets Sprite info , and switches to ta_sprite_data function
 			case ParamType_Sprite:
-				TileClipMode(data->pcw.User_Clip);
+				setClipMode(data->pcw.User_Clip);
 				if (CurrentList != ListType_None || startList(data->pcw.ListType))
 				{
 					VertexDataFP = ta_sprite_data;
@@ -502,23 +501,25 @@ case num : {\
 	{
 		TaCmd = ta_main;
 		BaseTAParser::reset();
+		setClipRect(0, 0, 39, 14);
+		setClipMode(0);
 	}
 
 private:
-	static void SetTileClip(u32 xmin,u32 ymin,u32 xmax,u32 ymax)
+	static void setClipRect(u32 xmin, u32 ymin, u32 xmax, u32 ymax)
 	{
-		u32 rv=tileclip_val & 0xF0000000;
-		rv|=xmin; //6 bits
-		rv|=xmax<<6; //6 bits
-		rv|=ymin<<12; //5 bits
-		rv|=ymax<<17; //5 bits
-		tileclip_val=rv;
+		u32 rv = tileclip_val & 0xF0000000;
+		rv |= xmin; 		// 6 bits
+		rv |= xmax << 6;	// 6 bits
+		rv |= ymin << 12;	// 5 bits
+		rv |= ymax << 17;	// 5 bits
+		tileclip_val = rv;
 	}
 
-	static void TileClipMode(u32 mode)
+	static void setClipMode(u32 mode)
 	{
 		//Group_En bit seems ignored, thanks p1pkin
-		tileclip_val=(tileclip_val&(~0xF0000000)) | (mode<<28);
+		tileclip_val = (tileclip_val & ~0xF0000000) | (mode << 28);
 	}
 
 	//Polys  -- update code on sprites if that gets updated too --

core/rend/vulkan/quad.cpp

@@ -23,6 +23,19 @@
 
 #include <memory>
 
+namespace {
+	size_t hashQuadVertex(const QuadVertex& vertex)
+	{
+		size_t seed = 0;
+		seed ^= std::hash<float>{}(vertex.x) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+		seed ^= std::hash<float>{}(vertex.y) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+		seed ^= std::hash<float>{}(vertex.z) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+		seed ^= std::hash<float>{}(vertex.u) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+		seed ^= std::hash<float>{}(vertex.v) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+		return seed;
+	}
+}
+
 static VulkanContext *GetContext()
 {
 	return VulkanContext::Instance();
@@ -169,7 +182,7 @@ void QuadDrawer::Init(QuadPipeline *pipeline)
 		descSet.reset();
 }
 
-void QuadDrawer::Draw(vk::CommandBuffer commandBuffer, vk::ImageView imageView, QuadVertex vertices[], bool nearestFilter, const float *color)
+void QuadDrawer::Draw(vk::CommandBuffer commandBuffer, vk::ImageView imageView, QuadVertex vertices[4], bool nearestFilter, const float *color)
 {
 	VulkanContext *context = GetContext();
 	auto &descSet = descriptorSets[context->GetCurrentImageIndex()];
@@ -198,3 +211,13 @@ void QuadDrawer::Draw(vk::CommandBuffer commandBuffer, vk::ImageView imageView,
 	commandBuffer.pushConstants(pipeline->GetPipelineLayout(), vk::ShaderStageFlagBits::eFragment, 0, sizeof(float) * 4, color);
 	buffer->Draw(commandBuffer);
 }
+
+size_t QuadBuffer::hashQuadVertexData(QuadVertex vertices[4]) const
+{
+	size_t seed = 0;
+	seed ^= hashQuadVertex(vertices[0]) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+	seed ^= hashQuadVertex(vertices[1]) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+	seed ^= hashQuadVertex(vertices[2]) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+	seed ^= hashQuadVertex(vertices[3]) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
+	return seed;
+}

core/rend/vulkan/quad.h

@@ -50,22 +50,37 @@ class QuadBuffer
 		commandBuffer.draw(4, 1, 0, 0);
 	}
 
-	void Update(QuadVertex vertices[] = nullptr)
+	void Update(QuadVertex vertices[4] = nullptr)
 	{
 		if (vertices == nullptr)
 		{
-			static QuadVertex defaultVtx[] {
+			static QuadVertex defaultVtx[4]
+			{
 				{ -1.f, -1.f, 0.f, 0.f, 0.f },
 				{  1.f, -1.f, 0.f, 1.f, 0.f },
 				{ -1.f,  1.f, 0.f, 0.f, 1.f },
 				{  1.f,  1.f, 0.f, 1.f, 1.f },
 			};
 			vertices = defaultVtx;
 		};
+
+		const size_t quadVerticesHash = hashQuadVertexData(vertices);;
+		if (quadVerticesHash == lastUploadHash)
+		{
+			// data already uploaded
+			return;
+		}
+
+		// new data to upload, update hash
+		lastUploadHash = quadVerticesHash;
 		buffer->upload(sizeof(QuadVertex) * 4, vertices);
+
 	}
 private:
 	std::unique_ptr<BufferData> buffer;
+
+	size_t hashQuadVertexData(QuadVertex vertices[4]) const;
+	size_t lastUploadHash = 0;
 };
 
 class QuadPipeline
@@ -119,7 +134,7 @@ class QuadDrawer
 	QuadDrawer& operator=(const QuadDrawer &) = delete;
 
 	void Init(QuadPipeline *pipeline);
-	void Draw(vk::CommandBuffer commandBuffer, vk::ImageView imageView, QuadVertex vertices[] = nullptr, bool nearestFilter = false, const float *color = nullptr);
+	void Draw(vk::CommandBuffer commandBuffer, vk::ImageView imageView, QuadVertex vertices[4] = nullptr, bool nearestFilter = false, const float *color = nullptr);
 private:
 	QuadPipeline *pipeline = nullptr;
 	std::unique_ptr<QuadBuffer> buffer;

core/rend/vulkan/texture.h

@@ -125,6 +125,11 @@ class SamplerManager
 		const vk::SamplerAddressMode vRepeat = tsp.ClampV ? vk::SamplerAddressMode::eClampToEdge
 				: tsp.FlipV ? vk::SamplerAddressMode::eMirroredRepeat : vk::SamplerAddressMode::eRepeat;
 
+		// The W-axis is unused for 2D textures
+		// Try to keep all three of the wrapping-modes the same by just repeating vRepeat for wRepeat
+		// BestPractices-Arm-vkCreateSampler-different-wrapping-modes
+		const vk::SamplerAddressMode wRepeat = vRepeat;
+
 		const bool anisotropicFiltering = config::AnisotropicFiltering > 1 && VulkanContext::Instance()->SupportsSamplerAnisotropy()
 				&& filter == vk::Filter::eLinear && !punchThrough;
 #ifndef __APPLE__
@@ -137,7 +142,7 @@ class SamplerManager
 		return samplers.emplace(
 					std::make_pair(samplerHash, VulkanContext::Instance()->GetDevice().createSamplerUnique(
 						vk::SamplerCreateInfo(vk::SamplerCreateFlags(), filter, filter,
-							mipmapMode, uRepeat, vRepeat, vk::SamplerAddressMode::eClampToEdge, mipLodBias,
+							mipmapMode, uRepeat, vRepeat, wRepeat, mipLodBias,
 							anisotropicFiltering, std::min((float)config::AnisotropicFiltering, VulkanContext::Instance()->GetMaxSamplerAnisotropy()),
 							false, vk::CompareOp::eNever,
 							0.0f, vk::LodClampNone, vk::BorderColor::eFloatOpaqueBlack)))).first->second.get();

core/rend/vulkan/vk_context_lr.cpp

@@ -329,7 +329,7 @@ void VulkanContext::PresentFrame(vk::Image image, vk::ImageView imageView, const
 	getVideoShift(shiftX, shiftY);
 
 	beginFrame(extent);
-	QuadVertex vtx[] {
+	QuadVertex vtx[4] {
 		{ -1, -1, 0, 0, 0 },
 		{  1, -1, 0, 1, 0 },
 		{ -1,  1, 0, 0, 1 },

core/rend/vulkan/vulkan_context.cpp

@@ -407,49 +407,47 @@ bool VulkanContext::InitDevice()
 		else
 			DEBUG_LOG(RENDERER, "Using distinct Graphics and Present queue families");
 
-		// Enable VK_KHR_dedicated_allocation if available
-		bool getMemReq2Supported = false;
-		dedicatedAllocationSupported = false;
-		std::vector<const char *> deviceExtensions = { VK_KHR_SWAPCHAIN_EXTENSION_NAME };
-		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))
-			{
-				deviceExtensions.push_back(VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME);
-				getMemReq2Supported = true;
-			}
-			else if (!strcmp(property.extensionName, VK_KHR_DEDICATED_ALLOCATION_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
+		{
+			if (supportedExtensions.count(extensionName.data()))
 			{
-				deviceExtensions.push_back(VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME);
-				dedicatedAllocationSupported = true;
+				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
+		const bool getMemReq2Supported = tryAddDeviceExtension(VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME);
+		dedicatedAllocationSupported = tryAddDeviceExtension(VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME);
+		dedicatedAllocationSupported &= getMemReq2Supported;
+
 #ifdef VK_ENABLE_BETA_EXTENSIONS
-			else if (!strcmp(property.extensionName, VK_KHR_PORTABILITY_SUBSET_EXTENSION_NAME))
-				deviceExtensions.push_back(VK_KHR_PORTABILITY_SUBSET_EXTENSION_NAME);
+		tryAddDeviceExtension(VK_KHR_PORTABILITY_SUBSET_EXTENSION_NAME);
 #endif
 #ifdef VK_USE_PLATFORM_METAL_EXT
-			else if (!strcmp(property.extensionName, VK_EXT_METAL_OBJECTS_EXTENSION_NAME))
-				deviceExtensions.push_back(VK_EXT_METAL_OBJECTS_EXTENSION_NAME);
+		tryAddDeviceExtension(VK_EXT_METAL_OBJECTS_EXTENSION_NAME);
 #endif
 #ifdef VK_DEBUG
-			else if (!strcmp(property.extensionName, VK_EXT_DEBUG_MARKER_EXTENSION_NAME))
-			{
-				NOTICE_LOG(RENDERER, "Debug extension %s available", property.extensionName.data());
-				deviceExtensions.push_back(VK_EXT_DEBUG_MARKER_EXTENSION_NAME);
-			}
-			else if(!strcmp(property.extensionName, VK_EXT_DEBUG_REPORT_EXTENSION_NAME))
-			{
-				NOTICE_LOG(RENDERER, "Debug extension %s available", property.extensionName.data());
-				deviceExtensions.push_back(VK_EXT_DEBUG_REPORT_EXTENSION_NAME);
-			}
-			else if (!strcmp(property.extensionName, VK_EXT_DEBUG_UTILS_EXTENSION_NAME))
-			{
-				NOTICE_LOG(RENDERER, "Debug extension %s available", property.extensionName.data());
-				deviceExtensions.push_back(VK_EXT_DEBUG_UTILS_EXTENSION_NAME);
-			}
+		tryAddDeviceExtension(VK_EXT_DEBUG_MARKER_EXTENSION_NAME);
 #endif
-		}
-		dedicatedAllocationSupported &= getMemReq2Supported;
 
 		// create a UniqueDevice
 		float queuePriority = 1.0f;
@@ -460,7 +458,7 @@ bool VulkanContext::InitDevice()
 		if (samplerAnisotropy)
 			features.samplerAnisotropy = true;
 		device = physicalDevice.createDeviceUnique(vk::DeviceCreateInfo(vk::DeviceCreateFlags(), deviceQueueCreateInfo,
-				nullptr, deviceExtensions, &features));
+				nullptr, enabledExtensions, &features));
 
 #if VULKAN_HPP_DISPATCH_LOADER_DYNAMIC == 1
 		VULKAN_HPP_DEFAULT_DISPATCHER.init(*device);
@@ -924,7 +922,7 @@ void VulkanContext::Present() noexcept
 
 void VulkanContext::DrawFrame(vk::ImageView imageView, const vk::Extent2D& extent, float aspectRatio)
 {
-	QuadVertex vtx[] {
+	QuadVertex vtx[4] {
 		{ -1, -1, 0, 0, 0 },
 		{  1, -1, 0, 1, 0 },
 		{ -1,  1, 0, 0, 1 },
@@ -1314,7 +1312,7 @@ bool VulkanContext::GetLastFrame(std::vector<u8>& data, int& width, int& height)
 	pipeline.BindPipeline(*commandBuffer);
 
 	// Draw
-	QuadVertex vtx[] {
+	QuadVertex vtx[4] {
 		{ -1, -1, 0, 0, 0 },
 		{  1, -1, 0, 1, 0 },
 		{ -1,  1, 0, 0, 1 },

core/rend/vulkan/vulkan_driver.h

@@ -99,7 +99,7 @@ class VulkanDriver final : public ImGuiDriver
 								vk::SamplerMipmapMode::eNearest,
 								vk::SamplerAddressMode::eClampToBorder,
 								vk::SamplerAddressMode::eClampToBorder,
-								vk::SamplerAddressMode::eClampToEdge, 0.0f, false,
+								vk::SamplerAddressMode::eClampToBorder, 0.0f, false,
 								0.f, false, vk::CompareOp::eNever, 0.0f, vk::LodClampNone,
 								vk::BorderColor::eFloatTransparentBlack));
 			}
@@ -115,7 +115,7 @@ class VulkanDriver final : public ImGuiDriver
 								vk::SamplerMipmapMode::eLinear,
 								vk::SamplerAddressMode::eClampToBorder,
 								vk::SamplerAddressMode::eClampToBorder,
-								vk::SamplerAddressMode::eClampToEdge, 0.0f, false,
+								vk::SamplerAddressMode::eClampToBorder, 0.0f, false,
 								0.f, false, vk::CompareOp::eNever, 0.0f, vk::LodClampNone,
 								vk::BorderColor::eFloatTransparentBlack));
 			}